MEASURING PRECIPITATION FROM SPACE
ADVANCES IN GLOBAL CHANGE RESEARCH VOLUME 28
Editor-in-Chief Martin Beniston, Uni...
17 downloads
661 Views
14MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
MEASURING PRECIPITATION FROM SPACE
ADVANCES IN GLOBAL CHANGE RESEARCH VOLUME 28
Editor-in-Chief Martin Beniston, University of Geneva, Switzerland
Editorial Advisory Board B. Allen-Diaz, Department ESPM-Ecosystem Sciences, University of California, Berkeley, CA, U.S.A. R.S. Bradley, Department of Geosciences, University of Massachusetts, Amherst, MA, U.S.A. W. Cramer, Department of Global Change and Natural Systems, Potsdam Institute for Climate Impact Research, Potsdam, Germany. H.F. Diaz, Climate Diagnostics Center, Oceanic and Atmospheric Research, NOAA, Boulder, CO, U.S.A. S. Erkman, Institute for Communication and Analysis of Science and Technology – ICAST, Geneva, Switzerland. R. García Herrera, Facultad de Físicas, Universidad Complutense, Madrid, Spain M. Lal, Centre for Atmospheric Sciences, Indian Institute of Technology, New Delhi, India. U. Luterbacher, The Graduate Institute of International Studies, University of Geneva, Geneva, Switzerland. I. Noble, CRC for Greenhouse Accounting and Research School of Biological Sciences, Australian National University, Canberra, Australia. L. Tessier, Institut Mediterranéen d’Ecologie et Paléoécologie, Marseille, France. F. Toth, International Institute for Applied Systems Analysis, Laxenburg, Austria. M.M. Verstraete, Institute for Environment and Sustainability, EC Joint Research Centre, Ispra (VA), Italy.
The titles published in this series are listed at the end of this volume.
MEASURING PRECIPITATION FROM SPACE EURAINSAT and the Future
edited by
V. Levizzani P. Bauer and
F. Joseph Turk
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN-13 978-1-4020-5834-9 (HB) ISBN-13 978-1-4020-5835-6 (e-book)
Published by Springer, P.O. Box 17,3300 AA Dordrecht, The Netherlands. www.springer.com
Some of the chapters are created by an officer or employee of US Government as part of his/her official duties and are therefore protected by copyright laws as mentioned under Title 17, Section 10-5 of the United States Copyright Law Printed on acid-free paper
All Rights Reserved © 2007 Springer No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
CONTENTS
Contributors Acknowledgments Preface
SECTION 1: CLIMATE MONITORING....………………………………....... 1
European Commission Research for Global Climate Change Studies: Towards Improved Water Observations and Forecasting Capability….….. M. Schouppe and A. Ghazi
xi xxiii xxv
1
3
2
Is Man Actively Changing the Environment?.............................................. D. Rosenfeld
7
3
The Global Precipitation Climatology Project………………………….… A. Gruber, B. Rudolf, M. M. Morrissey, T. Kurino, J. Janowiak, R. Ferraro, R. Francis, A. Chang, and R. F. Adler
25
4
Oceanic Precipitation Variability and the North Atlantic Oscillation…….. P. A. Arkin, H. M. Cullen, and P. Xie
37
5
Global Satellite Datasets: Data Availability for Scientists and Operational Users………………………………..………………….... G. A. Vicente, and the GES DAAC Hydrology Data Support Team
49
SECTION 2: CLOUD STUDIES IN SUPPORT OF SATELLITE RAINFALL MEASUREMENTS……………………………………………….
59
6
61
Cloud Top Microphysics as a Tool for Precipitation Measurements……... D. Rosenfeld
vi
Contents
7
The Retrieval of Cloud Top Properties Using VIS-IR Channels…..……... E. Cattani, S. Melani, V. Levizzani, and M. J. Costa
8
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events Over Africa and Portugal……………………………….... M. J. Costa, E. Cattani, V. Levizzani, and A. M. Silva
79
97
9
3D Effects in Microwave Radiative Transport Inside Precipitating Clouds: Modeling and Applications……………………………...……….. 113 A. Battaglia, F. Prodi, F. Porcù, and D.- B. Shin
10
Cloud Microphysical Properties from Remote Sensing of Lightning within the Mediterranean……..……………….………….…. 127 C. Adamo, R. Solomon, C. M. Medaglia, S. Dietrich, and A. Mugnai
11
The Worth of Long-Range Lightning Observations on Overland Satellite Rainfall Estimation………………………….………………….... 135 E. N. Anagnostou and T. G. Chronis
12
Neural Network tools for Satellite Rainfall Estimation….…….……….… 149 F. J. Tapiador, C. Kidd, V. Levizzani, and F. S. Marzano
SECTION 3: RAINFALL ALGORITHMS………………………………........
163
13
Passive Microwave Precipitation Measurements at Mid- and High Latitudes…………………………………………………………………... 165 R. Bennartz
14
The Goddard Profiling Algorithm (GPROF): Description and Current Applications……….……………….…………………….…... 179 W. S. Olson, S. Yang, J. E. Stout, and M. Grecu
15
Past, Present and Future of Microwave Operational Rainfall Algorithms…...…………………………………………………………..... 189 R. R. Ferraro
16
Space-Borne Radar Algorithms.................................................................... 199 T. Iguchi
17
Rain Type Classification Algorithm…………………….……………….... 213 J. Awaka, T. Iguchi, and K. Okamoto
Contents
vii
18
Dual-Wavelength Radar Algorithm……………………………….....…… K. Nakamura and T. Iguchi
225
19
A Next-generation Microwave Rainfall Retrieval Algorithm for use by TRMM and GPM…………………………………………….... 235 C. Kummerow, H. Masunaga, and P. Bauer
SECTION 4: BLENDED TECHNIQUES………………………………....…..
253
20
The University of Birmingham Global Rainfall Algorithms….………….. C. Kidd, F. J. Tapiador, V. Sanderson, and D. Kniveton
255
21
Multivariate Probability Matching for Microwave Infrared Combined Rainfall Algorithm (MICRA)…………………………………. 269 F. S. Marzano, D. Cimini, and F. J. Turk
22
Toward Improvements in Short-time Scale Satellite-Derived Precipitation Estimates using Blended Satellite Techniques…….……….. F. J. Turk and A. V. Mehta
281
23
Global Rainfall Analyses at Monthly and 3-h Time Scales…….………… 291 G. J. Huffman, R. F. Adler, S. Curtis, D. T. Bolvin, and E. J. Nelkin
24
CPC MORPHING Technique (CMORPH)……………………………..… 307 R. J. Joyce, J. E. Janowiak, P. Xie, and P. A. Arkin
25
CMAP: The CPC Merged Analysis of Precipitation…………………….... 319 P. Xie, P. A. Arkin, and J. E. Janowiak
26
Rainfall Estimation Using a Cloud Patch Classification Map……………. K.-L. Hsu, Y. Hong, and S. Sorooshian
329
COLOUR SECTION …………………………………………………... CP 1–CP 16 SECTION 5: VALIDATING SATELLITE RAINFALL MEASUREMENTS……………………………………………………..……
343
27
Methods for Verifying Satellite Precipitation Estimates…..…………….... 345 E. E. Ebert
28
Assessment of Satellite Rain Retrieval Error Propagation in the Prediction of Land Surface Hydrologic Variables………….……… 357 E. N. Anagnostou
viii
Contents
29
EURAINSAT Algorithm Validation and Intercomparison Exercise.…….. 369 M. Kästner
30
Ground Validation for the Global Precipitation Climatology Project.......... 381 M. M. Morrissey and S. Greene
31
Validation of Rainfall Algorithms at the NOAA Climate Prediction Center………………………………………………………...... 393 J. Janowiak
32
Ground Networks: Are We Doing the Right Thing?................................... W. F. Krajewski
403
SECTION 6: MODELING PRECIPITATION PROCESSES AND DATA ASSIMILATION FOR NWP…………………………………….
419
33
Aerosol Impact on Precipitation from Convective Clouds……..………… A. Khain, D. Rosenfeld, and A. Pokrovsky
421
34
The Wisconsin Dynamic/Microphysical Model (WISCDYMM) and the use of it to Interpret Satellite-Observed Storm Dynamics….…….. 435 P. K. Wang
35
The European Centre for Medium-Range Weather Forecasts Global Rainfall Data Assimilation Experimentation…...………..…...…… 447 P. Bauer, P. Lopez, E. Moreau, F. Chevallier, A. Benedetti, and M. Bonazzola
36
Rainfall Assimilation into Limited Area Models…………...…….………. 459 A. Buzzi and S. Davolio
37
Implementing an Operational Chain: The Florence LaMMA Laboratory.................................................................................................... 471 A. Ortolani, A. Antonini, G. Giuliani, S. Melani, F. Meneguzzo, G. Messeri, A. Orlandi, and M. Pasqui
SECTION 7: APPLICATIONS TO MONITORING WEATHER EVENTS………………………………………………………………………... 483 38
Satellite Precipitation Algorithms for Extreme Precipitation Events…....... 485 R. A. Scofield and R. J. Kuligowski
Contents
ix
39
Application of a Blended MW-IR Rainfall Algorithm to the Mediterranean…………………………………………………….… 497 F. Torricella, V. Levizzani, and F. J. Turk
40
Retrieving Precipitation with GOES, Meteosat, and Terra/MSG at the Tropics and Mid-latitudes…………..………………………………. 509 C. Reudenbach, T. Nauss, and J. Bendix
41
Model and Satellite Analysis of the November 9–10, 2001 Algeria Flood……………………………………………………………… 521 C. M. Medaglia, S. Pinori, C. Adamo, S. Dietrich, S. Di Michele, F. Fierli, A. Mugnai, E. A. Smith, and G. J. Tripoli
42
Modeling Microphysical Signatures of Extreme Events in the Western Mediterranean to Provide a Basis for Diagnosing Precipitation from Space..………………………………………................ 535 G. J. Tripoli, C. M. Medaglia, G. Panegrossi, S. Dietrich, A. Mugnai, and E. A. Smith
43
Online Visualization and Analysis: A New Avenue to use Satellite Data for Weather, Climate, and Interdisciplinary Research and Applications…………………………………………………….…….. 549 Z. Liu, H. Rui, W. L. Teng, L. S. Chiu, G. Leptoukh, and G. A. Vicente
SECTION 8: THE PRESENT AND FUTURE OF SATELLITE PLATFORMS…………………………………………………………………… 559 44
The Space-Based Component of the World Weather Watch’s Global Observing System (GOS)…………………………………………. 561 D. E. Hinsman and J. F. W. Purdom
45
The Meteosat and EPS/Metop Satellite Series……………………………. 571 J. Schmetz, D. Klaes, A. Ratier, and R. Stuhlmann
46
The Evolution of the NOAA Satellite Platforms………………………….. 587 W. P. Menzel
47
Japan’s Role in the Present and Future Satellite Observation for Global Water Cycle Research…………………………………………. 601 R. Oki and Y. Furuhama
x
Contents
48
International Global Precipitation Measurement (GPM) Program and Mission: An Overview..………………………………….………….… 611 E. A. Smith, G. Asrar, Y. Furuhama, A. Ginati, A. Mugnai, K. Nakamura, R. F. Adler, M.-D. Chou, M. Desbois, J. F. Durning, J. K. Entin, F. Einaudi, R. R. Ferraro, R. Guzzi, P. R. Houser, P. H. Hwang, T. Iguchi, P. Joe, R. Kakar, J. A. Kaye, M. Kojima, C. Kummerow, K.-S. Kuo, D. P. Lettenmaier, V. Levizzani, N. Lu, A. V. Mehta, C. Morales, P. Morel, T. Nakazawa, S. P. Neeck, K. Okamoto, R. Oki, G. Raju, J. M. Shepherd, J. Simpson, B.-J. Sohn, E. F. Stocker, W.-K. Tao, J. Testud, G. J. Tripoli, E. F. Wood, S. Yang, and W. Zhang
49
Snowfall Measurements by Proposed European GPM Mission…………... 655 A. Mugnai, S. Di Michele, E. A. Smith, F. Baordo, P. Bauer, B. Bizzarri, P. Joe, C. Kidd, F. S. Marzano, A. Tassa, J. Testud, and G. J. Tripoli
50
Observing Rain by Millimetre–Submillimetre Wave Sounding from Geostationary Orbit………………………………………………….. 675 B. Bizzarri, A. J. Gasiewski, and D. H. Staelin
51
The CGMS/WMO Virtual Laboratory for Education and Training in Satellite Matters.………………………………………………………... 693 J. F. W. Purdom and D. E. Hinsman
52
The International Precipitation Working Group: A Bridge Towards Operational Applications…………..……………………………. 705 V. Levizzani and A. Gruber
List of Acronyms.………………………………………………………………. 713 List of Symbols and Functions…………………………………………….…… 721
CONTRIBUTORS
CLAUDIA ADAMO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy ROBERT F. ADLER – NASA/Goddard Space Flight Center (GSFC), Laboratory for Atmospheres, Code 613.1, Greenbelt, MD 20771, USA EMMANOUIL N. ANAGNOSTOU – Department of Civil and Environmental Engineering, University of Connecticut, 261 Glenbrook Rd., Storrs, CT 06269-2037, USA, and Marie Curie Visiting Scientist, Hellenic Center of Marine Research (HCMR), Anavissos, Greece ANDREA ANTONINI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy PHILLIP A. ARKIN – Cooperative Institute for Climate Studies (CICS) and Earth System Science Interdisciplinary Center (ESSIC), 2207 Computer and Space Science Building, University of Maryland, College Park, MD 20742-2465, USA GHASSEM R. ASRAR – Science Division, NASA Headquarters, Washington, DC 20546, USA JUN AWAKA – Department of Information Science, Hokkaido Tokai University, Minami-ku, Minami-sawa 5-1-1-1, Sapporo 005-8601, Japan FABRIZIO BAORDO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy ALESSANDRO BATTAGLIA – Meteorologisches Institut, Universität Bonn, Auf dem Hügel 20, D-53121 Bonn, Germany
xi
xii
Contributors
PETER BAUER – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom JÖRG BENDIX – Fachbereich Geographie, Laboratory for Climatology and Remote Sensing (LCRS), Philipps-Universität Marburg, Deutschhausstrasse 10, D-35032 Marburg, Germany ANGELA BENEDETTI – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom RALF BENNARTZ – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA BIZZARRO BIZZARRI – c/o Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy DAVID T. BOLVIN – Sci. Sys. & Appl., Inc., NASA/Goddard Space Flight Center (GSFC), Code 613.1, Greenbelt, MD 20771, USA MARINE BONAZZOLA – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom ANDREA BUZZI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy ELSA CATTANI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy ALBERT CHANG† – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA FRÉDÉRIC CHEVALLIER – Laboratoire des Sciences du Climat et l’Environnement (LSCE), CEA-CNRS-UVSQ, Bât. 701, Orme des Merisiers, F-91191 Gif-sur-Yvette CEDEX, France LONG CHIU – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive (DAAC), Greenbelt,
†
deceased
Contributors
xiii
MD 20771, USA, and George Mason University, Fairfax, VA 22030, USA MING-DAH CHOU – Department of Atmospheric Sciences, National Taiwan University, Taipei, Taiwan THEMISTOCLIS G. CHRONIS – NASA/Marshall Space Flight Center (MSFC), Huntsville, AL 35812, USA DOMENICO CIMINI – CETEMPS, Universita’ degli Studi dell’Aquila, via Vetoio, I-67010 Coppito, L’Aquila, Italy MARIA JOÃO COSTA – Centro de Geofísica de Évora, CGE-UE, Universidade de Évora, Rua Romão Ramalho 59, PT-7000-671 Évora, Portugal HEIDI M. CULLEN – The Weather Channel, 300 Interstate North Parkway, Atlanta, GA 30339, USA SCOTT CURTIS – Department of Geography, East Carolina University, East Fifth St., Greenville, NC 27858-4353, USA SILVIO DAVOLIO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy MICHEL DESBOIS – CNRS, Laboratoire de Météorologie Dynamique (LMD), École Polytechnique, 91128 Palaiseau Cedex, France STEFANO DIETRICH – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy SABATINO DI MICHELE – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom JOHN F. DURNING – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA ELIZABETH E. EBERT – Bureau of Meteorology Research Centre (BMRC), GPO Box 1289, Melbourne, Victoria 3001, Australia FRANCO EINAUDI – NASA Goddard Space Flight Center (GSFC), Earth-Sun Exploration Division, Code 610, Greenbelt, MD 20771, USA
xiv
Contributors
JARED K. ENTIN – Science Division, NASA Headquarters, Washington, DC 20546, USA RALPH R. FERRARO – Satellite Climate Studies Branch, Cooperative Research Programs (CoRP), NOAA/NESDIS/STAR, and Cooperative Institute for Climate Studies (CICS), University of Maryland, 2207 Computer and Space Sciences Building, College Park, MD 20742, USA FEDERICO FIERLI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy RICHARD FRANCIS – Met Office, FitzRoy Road, Exeter, Devon, EX1 3PB, United Kingdom YOJI FURUHAMA – Japan Aerospace Exploration Agency (JAXA), Office of Space Applications, Tsukuba Space Center, 2-1-1, Sengen, Tsukuba City, Ibaraki 305-8505, Japan ALBIN J. GASIEWSKI – NOAA Environmental Technology Laboratory (ETL), 325 Broadway R/ET1, Boulder, CO 80305-3328, USA ANVER GHAZI† – European Commission, Research Directorate General, Directorate I: Environment, Unit I2: Global Change, B-1049 Brussels, Belgium AMNON GINATI – European Space Agency (ESA), European Space Research & Technology Centre (ESTEC), Keplerlaan 1, Postbus 299, 2200 AG Noordwijk, The Netherlands GRAZIANO GIULIANI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy MIRCEA GRECU – Goddard Earth Sciences and Technology Center, NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA J. SCOTT GREENE – College of Atmospheric and Geographic Sciences, University of Oklahoma, 100 East Boyd St., SEC Suite 710, Norman, OK 73019, USA ARNOLD GRUBER – Cooperative Institute for Climate Studies (CICS), Earth System Science Interdisciplinary Center (ESSIC), University of Maryland, 2207 Computer and Space Sciences Building, College Park, MD 20742, USA
†
deceased
Contributors
xv
RODOLFO GUZZI – Agenzia Spaziale Italiana (ASI), Viale Liegi 26, I-00198 Roma, Italy DONALD E. HINSMAN – World Meteorological Organization (WMO), WMO Space Programme, 7bis, Avenue de la Paix, Case postale No. 2300, CH-1211 Geneva 2, Switzerland YANG HONG – Center for Hydrometeorology and Remote Sensing (CHRS), Department of Civil and Environmental Engineering, University of California Irvine, E-4130 Engineering Gateway, Irvine, CA 92697-2175, USA PAUL R. HOUSER – George Mason University & Center for Research on Environment and Water (CREW), 4041 Powder Mill Road, Suite 302; Calverton, MD 20705-3106, USA KUO-LIN HSU – Center for Hydrometeorology and Remote Sensing (CHRS), Department of Civil and Environmental Engineering, University of California Irvine, E-4130 Engineering Gateway, Irvine, CA 92697-2175, USA GEORGE J. HUFFMAN – Sci. Sys. & Appl., Inc., NASA/Goddard Space Flight Center (GSFC), Code 613.1, Greenbelt, MD 20771, USA PAUL H. HWANG – NASA/Goddard Space Flight Center (GSFC), Global Precipitation Measurement (GPM) Project, Code 422, Greenbelt, MD 20771, USA TOSHIO IGUCHI – National Institute of Information and Communications Technology (NiCT), Applied Research and Standards Department, 4-2-1 Nukui Kita-machi, Koganei-shi, Tokyo 184-8795, Japan JOHN E. JANOWIAK – NOAA/NWS/NCEP/Climate Prediction Center (CPC), 5200 Auth Road, Camp Springs, MD 20746, USA PAUL JOE – Meteorological Service of Canada, 4905 Dufferin St., Downsview, Ontario, M3H 5T4, Canada ROBERT J. JOYCE – NOAA/NWS/NCEP/Climate Prediction Center (CPC), 5200 Auth Road, Camp Springs, MD 20746, USA MARTINA KÄSTNER – Deutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre (IPA), Oberpfaffenhofen, D-82234 Wessling, Germany RAMESH KAKAR – Science Division, NASA Headquarters, Washington, DC 20546, USA
xvi
Contributors
JACK A. KAYE – Science Division, NASA Headquarters, Washington, DC 20546, USA ALEXANDER KHAIN – Program of Atmospheric Sciences, Institute of Earth Sciences, The Hebrew University of Jerusalem, Jerusalem 91904, Israel CHRIS KIDD – School of Geography, Earth and Environmental Sciences, The University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom DIETER KLAES – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany DOMINIC R. KNIVETON – Department of Geography, University of Sussex, Falmer, Brighton, BN1 9SJ, United Kingdom MASAHIRO KOJIMA – Japan Aerospace Exploration Agency (JAXA), Office of Space Applications, Tsukuba Space Center, 2-1-1, Sengen, Tsukuba City, Ibaraki 305-8505, Japan WITOLD F. KRAJEWSKI – C. Maxwell Stanley Hydraulics Laboratory, The University of Iowa, Iowa City, IA 52242-1585, USA ROBERT J. KULIGOWSKI – NOAA/NESDIS/Center for Satellite Applications and Research (STAR), E/RA2, World Weather Building, 5200 Auth Rd., Camp Springs, MD 20746-4304, USA CHRISTIAN KUMMEROW – Department of Atmospheric Science, Colorado State University, Ft. Collins, CO 80523, USA KWO-SEN KUO – Coelum Inc., NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA TOSHIYUKI KURINO – Japan Meteorological Agency (JMA), 1-3-4 Otemachi, Chiyoda-ku, Tokyo 100-8122, Japan GREGORY G. LEPTOUKH – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive (DAAC), Greenbelt, MD 20771, USA, and George Mason University, 4400 University Drive, Fairfax, VA 22030, USA DENNIS P. LETTENMAIER – Department of Hydrology, Wilson Ceramic Lab., University of Washington, Box 352700, Seattle, WA 98195-2700, USA
Contributors
xvii
VINCENZO LEVIZZANI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy ZHONG LIU – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive, Greenbelt, MD 20771, USA, and George Mason University, 4400 University Drive, Fairfax, VA 22030, USA PHILIPPE LOPEZ – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom NAIMENG LU – National Satellite Meteorological Center (NSMC), China Meteorological Administration (CMA), Beijing 100081, China FRANK S. MARZANO – Centre of Excellence CETEMPS, University of L’Aquila, Italy, and Department of Electronic Engineering, Università “La Sapienza”, via Eudossiana 18, I-00184 Roma, Italy HIROIKO MASUNAGA – Department of Atmospheric Science, Colorado State University, Ft. Collins, CO 80523, USA CARLO M. MEDAGLIA – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy AMITA V. MEHTA – Goddard Earth Sciences and Technology Center, NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA SAMANTHA MELANI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy FRANCESCO MENEGUZZO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Biometeorologia (IBIMET), Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy W. PAUL MENZEL – NOAA/NESDIS, Center for Satellite Applications and Research, Cooperative Institute for Meteorological Satellite Studies (CIMSS) and Department of Atmospheric and Oceanic Sciences, University of WisconsinMadison, 1225 W. Dayton Street, Madison, WI 53706, USA GIANNI MESSERI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy
xviii
Contributors
CARLOS A. MORALES RODRIGUEZ – Instituto de Astronomia, Geofísica e Ciências Atmosféricas, Departamento de Ciências Atmosféricas, Rua do Matão, 1226 - Cidade Universitária, CEP: 05508-090, São Paulo, SP, Brazil EMMANUEL MOREAU – NOVIMET SA, Bât. Mermoz, 10-12 Avenue de l’Europe, 78140 Vélizy, France PIERRE MOREL – Université de Paris, Paris, France MARK M. MORRISSEY – College of Atmospheric and Geographic Sciences, University of Oklahoma, 100 East Boyd St., SEC Suite 710, Norman, OK 73019, USA ALBERTO MUGNAI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy KENJI NAKAMURA – Hydrospheric Atmospheric Research Center (HYARC), Nagoya University, Furocho, Chikusaku, Nagoya 464-8601, Japan TETSUO NAKAZAWA – Meteorological Research Institute (MRI), Japan Meteorological Agency (JMA), 1-1 Nagamine, Tsukuba-city, Ibaraki 305-0052, Japan THOMAS NAUSS – Fachbereich Geographie, Laboratory for Climatology and Remote Sensing (LCRS), Philipps-Universität Marburg, Deutschhausstrasse 10, D-35032 Marburg, Germany STEVEN P. NEECK – Science Division, NASA Headquarters, Washington, DC 20546, USA ERIC J. NELKIN – Sci. Sys. & Appl., Inc., NASA/Goddard Space Flight Center (GSFC), Code 613.1, Greenbelt, MD 20771, USA KEN’ICHI OKAMOTO – Osaka Prefecture University, Sakai, Osaka 599-8531, Japan RIKO OKI – Japan Aerospace Exploration Agency (JAXA), Earth Observation Research and application Center (EORC), 1-8-10 Harumi Chuo-ku, Tokyo 104-6023, Japan WILLIAM S. OLSON – Joint Center for Earth Systems Technology (JCET), University of Maryland Baltimore County (UMBC), Suite 320, 5523 Research Park Drive, Baltimore, MD 21228, USA
Contributors
xix
ANDREA ORLANDI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy ALBERTO ORTOLANI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy GIULIA PANEGROSSI – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA MASSIMILIANO PASQUI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy SABRINA PINORI – SERCO-DATAMAT Consortium, c/o ESA/ESRIN, via. G. Galilei, I-00044 Frascati, Italy ALEXANDER POKROVSKY – Program of Atmospheric Sciences, Institute of Earth Sciences, The Hebrew University of Jerusalem, Jerusalem 91904, Israel FEDERICO PORCÙ – Dip. di Fisica, Università di Ferrara, Edificio C, via Saragat 1, I-44100 Ferrara, Italy FRANCO PRODI – Dip. di Fisica, Università di Ferrara, Edificio C, via Saragat 1, I-44100 Ferrara, Italy, and Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy JAMES F. W. PURDOM – Cooperative Institute for Research in the Atmosphere (CIRA), Colorado State University, Ft. Collins, CO 80523-1375, USA GARUDACHAR RAJU – Indian Space Research Organization (ISRO), New BEL Road, Bangalore 560 094, India ALAIN RATIER – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany CRISTOPH REUDENBACH – Fachbereich Geographie, Laboratory for Climatology and Remote Sensing (LCRS), Philipps-Universität Marburg, Deutschhausstrasse 10, D-35032 Marburg, Germany DANIEL ROSENFELD – Program of Atmospheric Sciences, Institute of Earth Sciences, The Hebrew University of Jerusalem, Jerusalem 91904, Israel
xx
Contributors
BRUNO RUDOLF – Deutscher Wetterdienst (DWD), Postfach 10 04 65, D-63004 Offenbach am Main, Germany HUALAN RUI – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive, Greenbelt, MD 20771, USA, and SSAI, Lanham, Maryland, USA VICTORIA L. SANDERSON – School of Geography, Earth and Environmental Sciences, The University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom JOHANNES SCHMETZ – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany MICHEL SCHOUPPE – European Commission, Information Society and Media Directorate-General, DG INFSO - G5 - ICT for the Environment, Beaulieu 31 03/20, B-1049 Brussels, Belgium RODERICK A. SCOFIELD† – NOAA/NESDIS, World Weather Building, 5200 Auth Rd., Camp Springs, MD 20746-4304, USA J. MARSHALL SHEPHERD – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA ANA MARIA GUEDES DE ALMEIDA E SILVA – Centro de Geofísica de Évora, CGE-UE, Universidade de Évora, Rua Romão Ramalho 59, PT-7000-671 Évora, Portugal JOANNE SIMPSON – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA ERIC A. SMITH – NASA/Goddard Space Flight Center (GSFC), Laboratory for Atmospheres/Code 613.1, Greenbelt, MD 20771, USA BIUNG-JU SOHN - School of Earth and Environmental Sciences, Seoul National University, NS80, Seoul, 151-747, Korea ROBERT SOLOMON – United States Department of Agriculture, Forest Service, Pacific Wildland Fire Sciences Lab., 400 N. 34th Street, Suite 201, Seattle, WA 98103, USA
†
deceased
Contributors
xxi
SOROOSH SOROOSHIAN – Center for Hydrometeorology and Remote Sensing (CHRS), Department of Civil and Environmental Engineering, University of California Irvine, E-4130 Engineering Gateway, Irvine, CA 92697-2175, USA DAVID H. STAELIN – Research Laboratory of Electronics (RLE), Department of Electrical Engineering and Computer Science, and Engineering Systems Division, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA ERIC F. STOCKER – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA JOHN E. STOUT – George Mason University, 4400 University Drive, Fairfax, VA 22030, USA ROLF STUHLMANN – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany WEI-KUO TAO – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA FRANCISCO J. TAPIADOR – Instituto de Ciencias Ambientales (ICAM), Universidad de Castilla-La Mancha, Av. Carlos III s/n, 45071 Toledo, Spain ALESSANDRA TASSA – SERCO-DATAMAT Consortium, c/o ESA/ESRIN, via. G. Galilei, I-00044 Frascati, Italy WILLIAM TENG – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive (DAAC), Greenbelt, MD 20771, USA, and SSAI, Lanham, Maryland, USA JACQUES TESTUD – NOVIMET SA, Bât. Mermoz, 10-12 Avenue de l’Europe, 78140 Vélizy, France FRANCESCA TORRICELLA – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy GREGORY J. TRIPOLI – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA
xxii
Contributors
F. JOSEPH TURK – Naval Research Laboratory (NRL), Marine Meteorology Division, 7 Grace Hopper Avenue, Monterey, CA 93943-5502, USA GILBERTO A. VICENTE – NOAA/NESDIS/SSD – Products Implementation Branch (PIB), WWB, Room 510, E/SP22, 5200 Auth Road, Camp Springs, MD 20746, USA PAO K. WANG – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA ERIC F. WOOD – Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ 08544, USA PINPING XIE – NOAA/NWS/NCEP/Climate Prediction Center (CPC), 5200 Auth Road, Camp Springs, MD 20746, USA SONG YANG – George Mason University, and Goddard Earth Sciences and Technology Center, NASA/Goddard Space Flight Center (GSFC), Code 912.1, Greenbelt, MD 20771, USA WENJIAN ZHANG – Department of Observation & Telecommunications, China Meteorological Administration (CMA), Beijing 100081, China
ACKNOWLEDGMENTS
The idea to write this book came out at the end of the Project EURAINSAT – European satellite rainfall analysis and monitoring at the geostationary scale funded by the European Commission between 2001 and 2003. Several of us participated one way or another in the project, and a major outcome was that it contributed significantly to narrow the distance between the European research groups and the rest of the world. Many of the projects were encouraged by the launch of the first satellite mission dedicated to measure rainfall, the Tropical Rainfall Measuring Mission (TRMM). The outstanding success of TRMM has laid the foundation for a new international collaboration between scientific institutions and space agencies that are working towards another milestone, the Global Precipitation Measurement (GPM) mission. The book is the result of this renewed spirit of cooperation that will have to further increase as we are facing the challenges of the years to come. The community also felt that it was time to put together the amount of knowledge, technology and vision that is available in the field for the benefit of students, professionals and decision makers. More than 20 years have passed since the last book on this subject was printed. Many satellite missions have been launched in between while, at the same time, scientists have made substantial progresses towards transforming satellite rainfall “estimates” into real “measurements” and to produce operational rainfall products readily available for a wide field of applications ranging from climate research and numerical weather prediction to hydrology and agriculture. This book represents a significant effort and each one of the authors did not spare herself or himself in providing top class material, most of the time contributing results and ideas that were worth publishing in peer reviewed journals. The result is a book, which not only photographs the state of the art of the discipline, but also projects it into the future. Several people and organizations need to be warmly thanked and it would be difficult to do it one by one. However, we feel it is appropriate to say at least a few “thank you’s” that will enclose almost all the others.
xxiii
Acknowledgments
xxiv
First, The European Commission (EC) has to be acknowledged for its vision in financing a small group of European scientists on the occasion of the EURAINSAT project and for allowing them to reach out to the rest of the world. Dr. Anver Ghazi, former Head of EC Unit global Change, was a man of vision and unfortunately he has left us while putting together this book. It is an honour for us to have his Preface as one of the last contributions towards building the European Research Area. Two other great friends and colleagues have left us in the meantime, but we feel that they are still with us through their pages in this book: Dr. Rod Scofield and Dr. Al Chang. Thank you, friends. Each one of the authors deserves gratitude for the dedication and the patience with us during the very long time it took to finally print this volume. Our Institutions, CNR, ECMWF and NRL, were supportive in terms of allocating significant portions of our time to the project thus testifying the importance of these pages for an entire community. Two major international organizations were also very much behind us: the International Precipitation Working Group (IPWG) and its major sponsor, the World Meteorological Organization (WMO). They gave us the opportunity to work together and transform our efforts into a global strategy for the future. Finally, our families were deprived of many hours and are part of the project through their understanding and their moral and practical support. Without them this book would have never been printed.
Bologna, Reading, and Monterey, 13 October 2006 Vincenzo Levizzani, Peter Bauer and F. Joseph Turk
PREFACE
Since the last two decades, awareness by the international community of the threats hanging over the planet has increased significantly. Progress in sciences and technologies made it possible to improve our inventory of the state of the environment; evidence was given that rapid changes are occurring across the globe. At present, the complex dynamics of planet Earth and its multidimensional and interrelated processes are at the heart of current scientific investigations. In this context, it is primordial for Europe to further strengthen its capacity to understand, detect, and forecast global change. Detailed process studies and modeling relies permanently on the availability of systematic observations of atmospheric, terrestrial, and oceanic parameters including those of climate. Such a comprehensive observing capability includes large panoply of instruments on various platforms such as ground stations, ships, buoys, floats, ocean profilers, unmanned autonomous underwater and airborne vehicles, balloons, aircraft and satellites. Among these platforms, it is worth noting the unsurpassed coverage brought by Low Earth Orbiting (LEO) and Geostationary (GEO) satellites. Since about three decades, satellites are transmitting homogeneous measurements of many variables characterizing environmental processes and changes. As part of its 5th Framework program for European research and technological development (1998–2002), the European Commission has been supporting a portfolio of 26 research projects in the field of generic Earth observation technologies for the environment. The related 240 research and user organizations have cooperated and delivered innovative processing, modeling, and integration techniques. They have demonstrated elements of future monitoring systems and assessed the strengths and limitations of current space missions. Among these 26 European projects, EURAINSAT has successfully explored the real-time exploitation of data from both LEO and GEO satellites for rainfall estimation and subsequent assimilation into numerical weather prediction models. EURAINSAT has not only promoted the use of SEVIRI data from the recently declared operational METEOSAT-8 satellite,
xxv
xxvi
Preface
it has also prepared the scientific ground for the multilateral Global Precipitation Measurement (GPM) constellation to be launched in the period 2007-2008. The project, which was presented on the occasion of the Flood Media Event organized by the European Commission on 13 October 2003, builds upon several European research projects addressing the issue of precipitation such as EUROTRMM, MUSIC, and MEFFE. The project also opens new collaboration perspectives at European and global levels. It is thus a privilege to introduce this book presenting state of the art in the field of measuring precipitation from space. Anver Ghazi† Head of Unit, Global Change European Commission, Research Directorate General, Brussels, Belgium
†
Hyderabad, 7 March 1940 – Köln, 25 July 2005
Section 1 Climate Monitoring
1 EUROPEAN COMMISSION RESEARCH FOR GLOBAL CLIMATE CHANGE STUDIES: TOWARDS IMPROVED WATER OBSERVATIONS AND FORECASTING CAPABILITY Michel Schouppe and Anver Ghazi† European Commission, Research Directorate General, Brussels, Belgium
Abstract
Improved forecasting at the local, regional and global scales of the interrelated Earth system processes requires, as a prerequisite, a comprehensive global observing strategy susceptible to support the progressive establishment of a global science of integration. This paper concentrates on climate change research and observations, including the key parameters of the global water cycle. Reference is made to the related European research program in the field of “Global Change and Ecosystems” (thematic sub-priority 1.6.3. of the 6th research Framework Programme of the European Union, 2002–2006).
Keywords
Earth system, global change, water, floods, global observations, forecasting, European research, 6th Framework Programme
1
INTRODUCTION
Looking at our global environment, water is often mentioned to be the resource challenge of this century. Water must satisfy the need of all forms of life, starting with the rising water demand by a world’s population that has tripled during the 20th century. Sustainable management of water resources has to consider the increasing imbalance between the geographical and seasonal demand for, and availability of, water. It has to cope with pollution and waste of water, also where it is in abundant supply. The European Union (EU) is pursuing a common water policy since a number of years. The Water Directive of the European Parliament and of the European Council establishes †
Hyderabad, 7 March 1940 – Köln, 25 July 2005
3 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 3–6. © 2007 Springer.
4
M. Schouppe and A. Ghazi
a framework for the Community action in the field of water policy. This EU Directive calls notably for integration and coordination of river basin management across national borders. It also calls for an improved management of the hydrological extremes of floods and droughts. Water management and more generally sustainable development rely heavily on the improved understanding of the multidimensional mechanisms of the water cycle and their interactions with other climate-related processes. This paper concentrates on climate change research and observations, including the key parameters of the global water cycle.
2
WATER CYCLE AND PRECIPITATION
The global water cycle includes many components, mainly atmospheric water vapor, cloud cover, precipitation, surface and subsurface runoff, soil moisture, groundwater, oceans, snow, glaciers, and ice sheets. These components – that illustrate water in its various phases – are affected by a number of physical, chemical, biological, and human-induced processes that play a key role in the Earth’s climate. Precipitation is unanimously recognized as one of the most central variables of the global water cycle, mainly because of its direct significance for the availability of water for drinking and the agriculture, but also because of its impacts on, for instance, runoff over land, soil moisture, floods, stream flow, ocean salinity or atmospheric circulation through associated latent heating. To date, precipitation data are recorded from various remote and in situ instruments (mainly rain gauges, rain radars, and microwave, visible or infrared sensors) based on various platforms (mainly ground-based, airborne or space-borne). However, the fact that water cycle dynamics occur at a wide variety of spatial and time-scales makes them very challenging to observe, understand, and predict. Precipitation is characterized by various regimes. Moreover, determination of precipitation inputs and regimes requires ideal observation of the vertical hydrometeor structure, especially with respect to droplet size, shape, and temperature. Existing instruments and new emerging sensing technologies have each their own advantages but also limitations with respect to spatial sampling and resolution, spatial coverage, temporal coverage, cost of purchase and operation, calibration, accuracy, and consistency of the retrievals, etc. To date, the quantification of adequate precipitation products is hampered by a lack of long-term, stable and high quality observational data and a lack of integration among the various available observation data sources. The observation challenge is a prerequisite for our understanding of the water cycle and its interactions with the climate system at the local, regional, and global scales. Moreover, further research is also required in order to reduce current limitations of the climate models in their ability to simulate aspects of the water cycle such as precipitation amounts and frequency on seasonal
European Commission Research for Global Climate Change Studies
5
and longer time-scales, not only at very large grid scales, but also at lower scales such as catchment scales. At weather time-scales, new optimized nearreal-time assimilation schemes and increased lead-time of weather forecasts open the perspective for improved flood warning. Better forecasts also rely on stronger interdiscipline linkages. Meteorologists and hydrologists, for example, could further expand intelligent coupling of numerical weather forecast models with runoff models.
3
TESTS AND APPLICATIONS
Nowadays, more and more evidence is accumulated that weather and climate are characterized by chaotic aspects that could be inherited from a nonlinear behavior of the internal dynamics of the Earth system. The scientific community is step-by-step investigating the occurrence and amplitude of climate change with respect to changes in the forcings and feedbacks patterns of the Earth system. There is a consensus to categorize water vapor among the largest forcing agents and to group the processes involving water in its three phases among the most important feedback mechanisms that amplify or damp climate perturbations. This is notably true for the water vapor-cloudradiation feedback. Scientists who have studied long-term climatological data confirm increases in the frequency of extreme events – such as extreme temperatures or exceptional intensity of precipitations – decreases in sea level rise and seasonal and perennial snow and ice. In its Third Assessment Report 2001 the Intergovernmental Panel on Climate Change (IPCC) forecasts warmer climates, owing to more frequent and more intense hydrological extremes. Observed climate trends and projected scenarios raise the urgency to address several key scientific questions. To what extent is the global water cycle intensifying? What is the level of interactions between variations in the water cycle and other biogeophysical cycles? More particularly, is the observed increase in extreme events linked to climate change? Is this the result of natural variability or is it caused by human pressure on the environment, or both? One should consider the numerous interconnected factors at play. Climate change is likely to have an incremental effect. Human pressure (for instance, on the atmospheric composition) interferes with the natural flows of energy. More directly, human activities influence the hydrological cycle at various scale, notably through changes in land use, land cover including deforestation, irrigation and drainage, extended pavement, etc.
4
RESEARCH STRATEGY
Quantitative answers to the above questions require a comprehensive and integrated research strategy that builds upon ongoing disciplinary research on parts of the Earth system, but progressively evolves towards addressing interrelations between the dynamics of the Earth system as a whole. Such an
6
M. Schouppe and A. Ghazi
interdisciplinary Earth science vision is challenging and likely to take time to materialize. However, the process of establishing its foundation has already started, notably on both sides of the Atlantic. Such a science of integration is to be articulated over a comprehensive strategy that encompasses the expansion of observing capabilities, the exploitation of increasing computing capabilities, the optimization of performing assimilation schemes, the further development of numerical models in order to deliver both a deeper understanding of the Earth processes and improved forecasting and prediction capabilities.
5
EUROPEAN FRAMEWORK FOR RESEARCH
In the general frame of the 6th Framework Programme for EU research (FP6), “Global Change and Ecosystems” (sub-Priority 1.6.3) has been identified among the main thematic priorities for European research during the period 2002–2006 (http://www.cordis.lu). This “Global Change and Ecosystems” sub-priority comprises a horizontal activity focused on global observations and operational forecasting that directly contributes to the comprehensive strategy described above. The aim is to make systematic observations of primordial parameters, including those of climate, the sea, land and atmosphere, so as to improve forecasting, consolidate long-term observations for the modeling, establish common European database and help Europe to contribute in an integrated way to global observing systems such as the Global Climate Observing System (GCOS), the Global Ocean Observing System (GOOS) or the Global Terrestrial Observing System (GTOS). Focus is on Earth system-integrated observations and international cooperation. Activities to be supported could bring new research elements to the ongoing “Global Monitoring for the Environment and Security” initiative (GMES). Obviously rainfall products are essential to flood warning and mitigation systems. European research is also very active in this area which is included in the “Global Change and Ecosystems” sub-priority under “mechanisms of desertification and natural disasters.” Supported activities aim, in this case, at identifying the links to climate change, assessing and mapping risks while considering the environmental and socioeconomic consequences of floods, storms, and extreme events.
2 IS MAN ACTIVELY CHANGING THE ENVIRONMENT? Daniel Rosenfeld Hebrew University of Jerusalem, Institute of Earth Sciences, Jerusalem, Israel
1
PRECIPITATION AS A CENTERPIECE IN CLIMATE CHANGE
Water is the lifeblood of our livelihood on Earth. The habitability of Earth has been determined mainly by water availability. The deserts are sparsely populated not because they are too hot, but because they lack water. The availability of water to the hottest desert on Earth, e.g., the Nile River has resulted in one of the most densely populated regions on Earth. Temperaturedriven inhabitable areas are due to too low temperatures, and not due to excessively high temperatures. Therefore, our main concern with respect to climate variability and change, at least as far as our living conditions are concerned, has to be its manifestation with respect to water resources, which is determined to a large extent by precipitation. The role of precipitation goes far beyond merely replenishing our water resources. Although only radiative forcing has been considered until now in changing the climate, it is only one of the two major components that energize the climate system. Because the air is transparent to sunlight at most of the solar wavelengths, most of the solar radiation is delivered to the atmosphere indirectly by surface heating. However, most (~77%) of the solar radiation reaching the surface is consumed to evaporate water and the energy is transported in the atmosphere as latent heat within the water vapor. Only when and where the reverse process occurs, i.e., precipitation back to the surface, the released latent heat of condensation causes a net warming of the air, decreases its density and so creates pressure gradients that propel the global circulation of the atmosphere. In terms of energy budget, 47% of the atmospheric radiative loss is balanced by surface sensible (11%) and latent (36%) heat 7 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 7–24. © 2007 Springer.
8
D. Rosenfeld
fluxes. Therefore, the roles of radiative and latent heating in the climate system are of similar magnitudes. These facts have been well known for a long time. The need for a better documentation of the way the latent heating of the atmosphere propels the global circulation has been the main motivation for launching the Tropical Rainfall Measuring Mission (TRMM) satellite in late 1997. So, why have we not considered until now “latent heat forcing” in relation to climate change, in parallel with the “radiative heating forcing”? Latent heat forcing is defined here as the change in the atmospheric heating due to man-made induced changes in precipitation. The term sounds unfamiliar, because we rarely considered man-made impacts on the precipitation, and most that we have done so far with respect to precipitation was driven by radiative considerations, mainly along the following lines: the added greenhouse gases act to accelerate the hydrological cycle, whereas the radiative impacts of the added aerosols slow down the hydrological cycle and redistribute the precipitation. Under such circumstance the changes in latent heating became response and not forcing. We did not consider latent heat forcing because we had little appreciation of man-made induced changes in precipitation by cloud-aerosol interactions and land use changes. Recent observations, mainly from TRMM, have shown that the same microphysical changes that incur the “Cloudmediated aerosol radiative forcing” mainly in shallow clouds, can suppress precipitation from deeper clouds, and change the nature of the precipitation in the deepest tropical convective clouds. Separation of the response and forcing components is a challenge due to the many feedback processes. However, this does not detract from the importance of recognizing the major role of the latent heat forcing. The latent heat forcing is not likely to change the mean global temperature, but not less importantly, it would change the precipitation and climate patterns, storm tracks, etc., which can have profound regional and global impacts that likely occur now. The climate change terminology has to adjust, because in its present from it relates only to radiative effects, and so limits us conceptually from considering other energy forms. The latent heat forcing is a meaningful term if one can show that anthropogenic activity has a substantial impact on precipitation. In the subsequent sections we will address this possibility.
2
MECHANISMS FOR ANTHROPOGENIC IMPACTS ON PRECIPITATION
2.1 Greenhouse gases radiative effects on precipitation The increased GHG reduces thermal radiation from the surface, and increases instead the convective fluxes. Assuming a negligible change in the ratio between latent and sensible heat fluxes, more evaporation would occur and be balanced by the same amount of additional precipitation. This leads towards a
Is Man Actively Changing the Environment ?
9
warmer and wetter climate. The global average precipitation has increased during the last century by about 2.4 mm per decade (Dai et al. 1997, Fig. 10), in line with the nearly 3.0 mm per decade obtained for a 1% per year CO2 increase, as simulated in a coupled ocean–atmospheric GCM study (Russell et al. 1995). Spatially, the precipitation increases during the last century occurred over most of the middle and high-latitude land areas, but not over the low latitude continents (Dai et al. 1997).
2.2 Aerosol radiative effects Aerosols reflect and absorb solar radiation, thereby blocking surface heating and evaporation. Because precipitation must balance evaporation on a global scale, added aerosols to the atmosphere has to slow down the hydrological cycle (Ramanathan et al. 2001) and reduce the precipitation amount. Therefore, estimating the changes in the global energy budget can provide us indirectly the changes in global precipitation. It is much easier to evaluate the global energy budget than measuring global precipitation at the same confidence level. However, based on energy budget considerations alone we cannot predict the spatial and temporal distribution of precipitation, which matters practically much more than the globally averaged amount. An energy flux of 1 W m–2, when used for evaporating water from the Earth surface, evaporates 12.6 mm per year. Because, on a global average, 77% of the surface heating is consumed for evaporation, a net loss of 1 W m–2 of surface heating would result in a loss of evaporation and subsequent precipitation of about 10 mm per year. This would be the precipitation loss due to backscatter of short wave solar radiation to space by white aerosols, such as sulfates, which are mostly anthropogenic over the land areas. This leads towards a cooler and drier climate. The precipitation losses do not end here. In the case of absorbing aerosols, such as carbonaceous particles from smoke and vehicles, the lower troposphere is heated, but without mediation by the surface and the associated surface evaporation. This results in decreasing evaporation and hence precipitation, and leads towards a warmer and drier climate. Indeed, a decreasing trend of solar radiation and surface evaporation (Roderick and Farquhar 2002) has been observed in recent decades (Gilgen et al. 1998; Stanhill and Cohen 2001), while warming was recorded at the same time. This can occur only if a smaller fraction of the surface heating is consumed by evaporation.
2.3 Land use effects The decreasing evaporation fraction (known as the Bowen ratio) can occur not only due to the increase in absorbing aerosols, but also due to changes in land use. Extensive deforestation and cultivation induces decreases in evaporation and in the Bowen ratio. It also induces increases in the surface albedo, which
D. Rosenfeld
10
causes greater reflection of solar energy back to space, and further reduction in the surface heating and evaporation. The anthropogenic changes in the global energy budget of the surface are given in Table 1 (Wild et al. 2004). Table 1. Estimated changes in energy fluxes over global land surfaces for the period 1960–1990 (energy gain for the surface is signed positive). (After Wild et al. 2004.)
(a) Change in absorbed short wave radiation (b) Change in downward long wave radiation (a) + (b) = Surface radiative forcing (c) Change in upward long wave radiation (d) Change in net radiation (a) + (b) + (c) (e) Change in ground heat flux (f ) Change in ice melt
–6 to –9 W m–2 +2 to +3 W m–2 –3 to –7 W m–2 –2 W m–2 –5 to –9 Wm–2 –0.01 Wm–2 –0.2 Wm–2
(1, 2, 3) (4, 5, 6) (4) (7) (7)
1. Gilgen et al. (1998); 2. Liepert (2002); 3. Stanhill and Cohen (2001); 4. Wild et al. (2004); 5. Wild et al. (1997); 6. Garratt et al. (1999); 7. Ohmura (2004).
This amounts to changes in turbulent (latent and sensible heat) fluxes of +5 to +9 W m–2, which means loss of latent and sensible heat flux from the surface to the atmosphere by this amount. The increase in surface sensible heating and drying has the same effect on clouds and precipitation as increasing the amount of pollution aerosols, as described in the next section. This is so because it induces stronger thermals that become stronger updrafts at the cloud base, which produce greater vapour super-saturation and hence a larger fraction of the aerosols nucleate into a greater number concentration of cloud droplets (Williams et al. 2002; Williams and Stanfill 2003).
3
THE AEROSOL CONTROL ON CLOUD PRECIPITATION EFFICIENCY
Clouds precipitate when they survive sufficiently long for the water and/or ice particles to grow to sizes that are large enough to fall to Earth surface. This can happen in two ways. At temperatures above freezing, droplets grow by attracting water vapor through diffusion, until they reach an “effective radius” (re) of about 14 µm, which is related to the ratio of the total volume divided by the total surface area of all the drops in a given cloud volume. After this point, the droplets continue to grow by colliding and coalescing with other water droplets. Eventually, when the drops are bigger than about 0.2 mm in diameter, they fall through the cloud and reach the Earth’s surface as rain. This precipitation process is, however, highly sensitive to the size of the initial cloud droplets. Those with diameters less than about 25 µm are so small that they
Is Man Actively Changing the Environment ?
11
float in air and have a low probability of growing into raindrops by colliding and coalescing with other droplets. Droplets larger than about 30 µm, on the other hand, coalesce much faster (see in Fig. 1).
Figure 1. Large cloud droplets are collected by a falling raindrop, but small cloud droplets follow more closely the airflow streamlines and bypass the falling raindrop, thereby slowing down its further growth.
The other way in which a cloud can grow particles to precipitation size is through ice processes, which operate also when droplet coalescence processes are absent. Ice particles are first formed either when water droplets freeze as they ascend to heights where the temperatures decrease below 0oC, or when ice crystals are nucleated on aerosol particles called ice nuclei. The ice particles collect unfrozen drops faster than water drops of the same mass (Pinsky et al. 1998), and also evaporate more slowly. This process leads often to the growth of ice hydrometeors that fall to Earth, melting to form rain if they reach temperatures above 0oC. If the falling ice particles are large, they may not melt at all before reaching the ground; this is hail. Snowflakes are aggregates of ice crystals, but snow can also be enriched by collecting cloud drops. Therefore, snow that falls from clouds with smaller cloud drops is also suppressed (Borys et al. 2003). Pollution affects these precipitation processes because all cloud droplets – whether formed through the water or ice route – must initially form around an existing aerosol particle, known as a cloud condensation nucleus. But the number of these nuclei depends on the purity of the air. Clean air has relatively few cloud condensation nuclei per unit volume, which means that only about 100 cloud droplets are formed in every cm3 of air. Polluted air, in contrast, has about 1,000 or more cloud condensation nuclei per cm3, mainly in the form of
12
D. Rosenfeld
additional smoke and aerosol particles. Since the total amount of water in polluted and unpolluted clouds at a particular height is about the same, the water in dirty clouds is distributed over a much larger number of droplets, which then must be smaller. In other words, there are lots of very small droplets in polluted clouds, which cannot easily grow into larger drops of precipitation size during the lifetime of the cloud (Fig. 2). Pollution therefore hinders rainfall.
Figure 2. Clean clouds have fewer water droplets per unit volume than polluted clouds, although the size of the droplets in clean clouds increases quickly with height above the base of the cloud. Polluted clouds have a larger number of smaller drops, and their size increases only slowly with height. The small size of the drops in the polluted clouds slows their conversion into rainfall. (From Rosenfeld and Woodley 2001.)
Pollution can also affect the growth of droplets that form through the ice phase. The reason is that polluted clouds have lots of tiny droplets, which freeze more slowly at sub-zero temperatures than the larger drops found in clean clouds. Droplets that are smaller than 30 µm tend to remain in a supercooled liquid state until about –25oC and even down to a chilly –38oC if the cloud contains very small droplets in vigorously ascending air currents (Rosenfeld and Woodley 2000). These supercooled liquid droplets float in the air flowing around the falling ice precipitation particles, and therefore manage to avoid being captured. Cloud model simulations have shown that cleaning up the air in this case would have caused freezing at much higher temperatures and more than double the rainfall amount from this cloud (Khain et al. 2001). The ice particles, in other words, fail to collect enough water from the small
Is Man Actively Changing the Environment ?
13
cloud droplets to grow to precipitation size. Once again, pollution slows down rainfall. Cloud-physics measurements have been usually made from an aircraft equipped with suitable instruments. However, measurements characterizing the processes over large areas and short periods of time are impractical. A satellite-based approach was developed to meet the need for covering large areas (Section 2 of this book). This approach triggered a new era in cloud physics and led to appreciation of the extent by which man-made effects on precipitation occur. Not much was known before this new era about the impact of aerosols from anthropogenic aerosols on precipitation. For example, it was assumed initially that industrial and urban pollution inhibited precipitation (Gunn and Phillips 1957). However, later reports of enhanced rainfall downwind of paper mills (Eagen et al. 1974) and over major urban areas (Braham 1981) suggested that giant CCN caused enhancement of the precipitation (Johnson 1982). However, attempts to correlate the urban-enhanced rainfall to the air pollution sources failed to show any relationship (Gatz 1979). The most plausible explanation for the urban rain enhancement invokes the heat-island effect and increased friction, both of which would tend to increase the surface convergence, resulting in more cloud growth and rainfall over and downwind of the urban areas. On the other hand, the suggestion published in Nature (Cerveny and Balling 1998) that air pollution might enhance precipitation on the large scale in the northeastern USA and Canada and the speculative explanations for this effect would appear to confuse the issue. The truth is that rather little was known definitively about this subject.
4
AEROSOLS FROM URBAN AND INDUSTRIAL AIR POLLUTION
Space-borne measurements of ship tracks in marine stratocumulus provided the first evidence that effluents from ship stacks change cloud microstructure such that their water is redistributed into a larger number of smaller droplets (Coakley et al. 1987). Extrapolation of these observations to clouds that are sufficiently thick for precipitation (i.e., at least 2 km from base to top) would mean that the effluents have the potential to suppress precipitation. Application of the imaging scheme of Rosenfeld and Lensky (1998) to the Advanced Very High Resolution Radiometer (AVHRR) onboard the National Oceanic and Atmospheric Administration (NOAA) orbiting weather satellites have now revealed numerous “ship-track” like features in clouds over land, emanating from major urban and industrial pollution sources. Because the tracks originate evidently from pollution sources, they are named “pollution tracks” by Rosenfeld and Woodley (2003), who identified them as a frequently occurring global phenomenon. Rosenfeld (2000) analyzed the precipitation effects for such a case for pollution tracks in Australia. Satellite retrievals of cloud
14
D. Rosenfeld
microstructure clearly revealed plumes, embedded in extensive cloudy areas, in which the clouds had reduced particle sizes. These plumes originated from major urban areas and industrial facilities such as coal-fired power plants. The satellite retrievals in the polluted and unpolluted regions showed little cloud drop coalescence (as inferred by the methodology of Rosenfeld and Lensky, 1998) in the polluted region and strong coalescence in the pristine clouds. In addition, the Precipitation Radar (PR) onboard the Tropical Rainfall Measuring Mission (TRMM) satellite revealed that the plumes of polluted clouds were devoid of precipitation, whereas the ambient clouds had precipitation intensities exceeding 10 mm h–1. Although producing no precipitation, the clouds in the plumes were as thick as the adjacent precipitating clouds and had no shortage of water. In addition, the PR detected a “bright band” signature, which is indicative of melting snow, in the adjacent precipitating unpolluted clouds, showing further that the pollution also suppresses the processes leading to the growth of ice particles to precipitation size.
Figure 3. Air pollution decreases the drop sizes of convective clouds over the British Isles. This NOAA-AVHRR image from 18 April 1995, 1337 UT was analyzed by the scheme of Rosenfeld and Lensky (1998), showing convective rain clouds with large drops (re > 20 µm, well exceeding the 14 µm precipitation threshold) in the north-westerly flow from the Atlantic Ocean. The clouds interact with the air pollution over the populated land areas and become composed of small drops (re < 10 µm, too small for precipitating) that appear in yellow shades. Note that the sharp distinction of the clouds around the latitude of Glasgow. Northern Scotland is sparsely populated and hence the clouds remain pristine with large drops, as indicated by the red shades (see also color plate 2).
Is Man Actively Changing the Environment ?
15
Quantitative satellite assessment of rainfall downwind of large cities over the USA has shown that warm-season convective rainfall has increased relative to the rural areas (Shepherd et al. 2002; Shepherd and Burian 2003). This has been attributed mainly to the effect of the urban heat island that triggered thunderstorms over the city, although an added component of pollution aerosols effect on invigoration of the thunderstorms is also possible (Williams et al. 2002; Andreae et al. 2004). The invigoration occurs due to the suppression of early onset of precipitation. This has two effects: (1) the water is not lost to rain, but rather it is carried to the heights where it freezes and releases the latent heat of freezing, which adds buoyancy to the cloud volume; (2) the delay of the onset of rainfall also delays the onset of downdraft and the respective dissipation of the convective cloud cell. Givati and Rosenfeld (2004) have quantified the suppression of precipitation on a regional scale due to urban and industrial air pollution such as caused by the pollution tracks in Australia (Rosenfeld 2000), in clouds that are expected to be especially vulnerable (see Fig. 3 for an example of polluted clouds over the British Isles). Such are the orographic shallow precipitation clouds that form over topographical barriers downwind of major coastal areas. For example, the typical rainfall situation in California is from clouds that move inland with the pristine air from the Pacific Ocean. The air mass becomes polluted during its passage over the densely inhabited and industrial coastal areas. This polluted air ascends the hills to the east and forms shallow and short-living orographic clouds, which are responsible for most of the enhanced precipitation over the hills with respect to the upwind lowland. Hilltop measurements in the clouds have shown how the added pollution slows down the accretion of the cloud drops on ice hydrometeors, by about a factor of two (Borys et al. 2003). This should reduce the orographic enhancement factor of precipitation. Indeed, rain gauge analyses of century-long time series showed that the rainfall over the hills was decreased by up to 25% relative to the upwind lowland with major urban areas (Rosenfeld and Givati 2004), in both California and Israel (see Fig. 4). These precipitation losses occur in regions where water is in high demand and great shortage. For example, Israel already has started building seawater desalination plants, at a cost of more than 50 US cents per m3. This puts the economical price on the lost precipitation in Israel. The amount of lost water is estimated by more than 4 × 109 m3 of rainfall volume per year in the Sierra Nevada section of central California alone. At least 1/3 of this amount is loss of exploitable hydrological water.
D. Rosenfeld
16
Figure 4. Topographic cross section showing the effects of urban air pollution on precipitation as the clouds move from west to east from the coast to the Sierra Nevada Mountains and to the eastern slopes. The boxes show the amount of the annual precipitation (mm per year) in each topographic location and the numbers above them show the loss or gain of precipitation (mm per year) at each site. Maritime air (zone 1) is polluted over coastal urban areas (zones 2, 3) – no decrease in precipitation occurs. The polluted air rises over mountains downwind and forms new polluted clouds (zone 4) – decreases of 15%–20% (losses of 220 mm per year) in the ratio between the western slopes to the coastal and plain areas. The clouds reach to the high mountains (zone 5). All precipitation is snow – slight decrease of 5%–7% (loss of 65 mm per year) in the ratio between the summits to the plain areas. The clouds move to the high eastern slopes of the range (zone 6) – increase of 14% (gain of 66 mm per year) in the ratio between the eastern slopes to the plain. (Givati and Rosenfeld 2004, © American Meteorological Society.)
5
SMOKE AEROSOLS FROM BURNING VEGETATION
Suspicions that smoke from burning vegetation suppresses precipitation arose already more than 30 years ago, based on laboratory experiments (Gunn and Phillips 1957) and on observations that precipitation was reduced downwind of seasonal agricultural burning of sugar cane fields in Australia (Warner 1968). Vegetation burning emits large concentrations of small Cloud Condensation Nuclei (CCN) (Hobbs and Radke 1969; Kaufman and Fraser 1997), which modify the cloud drop size distribution (DSD) so that the same amount of water is redistributed over a larger number of smaller drops. The coalescence efficiency of cloud droplets into raindrops is greatly reduced when the radius of the largest cloud droplets is smaller than about 25 µm (Mason and Jonas 1974), which is equivalent to an effective radius (re) threshold of 14 µm (Rosenfeld and Gutman 1994). Effective radius is the cloud droplet size distribution parameter, which is observable by satellites. It has already been observed by satellite that re decreased below the precipitation threshold of re = 14 µm in clouds infected by smoke from burning vegetation in the Amazon (Kaufman and Fraser 1997) and Indonesia (Rosenfeld and Lensky 1998). Satellite observations of the Tropical-Rainfall-Measuring-Mission (Rosenfeld
Is Man Actively Changing the Environment ?
17
1999) show that warm rain processes in convective tropical clouds that ingest smoke from forest fires are practically shut off. The tops of the smokeinfected clouds (see Fig. 5 for a picture of the tops of smoky clouds in the Amazon) has to start developing ice, i.e., grow to altitudes colder than about – 10oC, for the clouds to start precipitating. In contrast, adjacent tropical clouds in the cleaner air precipitate most of their water before ever freezing. Similar TRMM observations were documented in the Amazon (Rosenfeld and Woodley 2003). The very deep clouds that developed there in the smoke, however, had stronger precipitation radar reflectivities and more lightning. Andreae et al. (2004) validated the satellite inferences by in situ measurements of CCN, cloud DSD and precipitation in smoky and smoke-free clouds over the Amazon.
Figure 5. Smoke rising within convective clouds and detrains from their tops. This picture was taken over the Amazon in clouds that were measured with cloud physics aircraft, which documented the small cloud drops and lack of precipitation forming processes by drop coalescence (Andreae et al. 2004).
It was observed by in situ (Andreae et al. 2004) and satellite (Rosenfeld and Lensky 1998) that clouds have to reach greater heights for onset of precipitation in the more polluted and smoky conditions. The lack of early precipitation allows updrafts to accelerate and transport cloud water in deep convection to the high and supercooled regions, where it can release additional latent heat of freezing, which it would not have delivered in the clean case of early rainout.
18
D. Rosenfeld
The added water is available for production of intense ice precipitation, hail and lightning, creating more violent convective storms. Confining the heating to the lower troposphere in the case of cleaner air leads to more confined tropical response and less energy propagation to higher latitudes. Higher latitude propagation is more efficient when the level of maximum heating associated with the release of latent heating in the convective updrafts is seen in the middle to upper troposphere, typical of thunderstorms. Shallow convection, or even cumulus congestus have a maximum heating level in the lower troposphere and thus affect only the regional climate. A shift to more thunderstorms has the effect of generating planetary scale upper level waves that affect the global climate. For example, a perturbation in convection in tropical South America may affect the weather in Europe and Asia in the time scale of intraseasonal oscillations (Grimm and Silva Dias 1995). Quantitative assessment of the smoke microphysical effect on a regional basis is still lacking. Indications of the magnitude of the effect can be obtained from a radar study of rainfall in Thailand, under conditions of suppressed and intense cloud drop coalescence (Rosenfeld and Woodley 2003, Fig. 6). This is relevant, because it has been already shown that cloud drop coalescence in the tropics is dominated by the aerosols, and the variability in the coalescence in the studied clouds in Thailand appeared to be related to the amount of smoke from agricultural burning. The radar estimates of rain cell properties were partitioned using in situ observations of the presence or absence of detectable raindrops on the windshield of the aircraft as it penetrated the updrafts of growing convective towers, 200–600 m below their tops at about the –8oC level (about 6.5 km amsl). Cells observed to contain detectable raindrops during these aircraft penetrations were found to have smaller first-echo depths than cells without observed raindrops when growing through the aircraft penetration level. This faster formation of raindrops is attributed to a rapid onset of coalescence in the convective cells. Convective cells exhibiting a rapid onset of coalescence (Category 2 clouds with detectable raindrops when growing through the aircraft penetration level) produced over a factor of two more rainfall than cells in which the onset of coalescence was slower (Category 1 clouds with no detectable raindrops when growing through the aircraft penetration level). This was true also for convective cloud systems over a fixed-size area covering about 2,000 km2 (Table 1 of Rosenfeld and Woodley 2003).
Is Man Actively Changing the Environment ?
19
10
7
10
6
10
5
10
4
10
3
10
2
3
Rvol [m ]
Strong Coalescence Activity W eak Coalescence Activity
3
5
7 9 11 13 15 17 Hm ax [km]
Figure 6. The mean rain volumes of convective rain cells as a function of maximum precipitation echo top height for cells growing on days with weak coalescence activity (Category 1) and on days with strong coalescence activity (Category 2). The data are plotted at the center point of 2 km intervals of maximum echo top height (Hmax). For the purpose of showing the trend on a logarithmic scale, a value of 102 m3 was used instead of the zero value of RVOL for the 3 km maximum echo top height interval for cells growing on days with weak coalescence activity (Category 1). (After Rosenfeld and Woodley 2003, © American Meteorological Society.)
Figure 6 quantifies the increase of rain production of convective cells with maximum echo top height in both coalescence categories, but the increase is greater for clouds exhibiting a rapid onset of coalescence. The data are plotted at the center point of 2 km intervals of maximum echo top height, Hmax. The difference in rainfall production is the largest for the shallowest clouds, but even the deepest convective cells, in which mixed-phase precipitation forming processes are dominant, the rain-volume from clouds with suppressed coalescence was significantly smaller than the rainfall from clouds with active coalescence. For example, deep tropical cumulonimbus cells (all cells with Hmax >10 km) that exhibit a rapid onset of coalescence produce twice as much more rain volume than cells with slower coalescence activity with a P-value of 0.002. Most important, lifecycle analysis of convective cloud systems in which the convective cells reside showed that the results for the cell scale are preserved on the scale of cloud systems. These radar estimates are likely the lower bound on the actual rainfall differences, because rainfall that occurs in clouds with suppressed coalescence has larger drops than raindrops of the same rain intensity that fall from clouds with active coalescence (Rosenfeld and Ulbrich 2003). This means that the
D. Rosenfeld
20
radar estimated rainfall from the clouds with active coalescence is relatively underestimated with respect to the rainfall from clouds with suppressed coalescence, and hence the estimated effect of coalescence on rainfall is just the lower bound.
6
REVERSING POLLUTION INDUCED SUPPRESSION OF PRECIPITATION
The coin of demonstrated sensitivity of precipitation forming processes to aerosols has two sides. We have already seen how smoke and air pollution aerosols produce large concentration of small CCN that induce small cloud droplets, which are slow to coalesce and form precipitation. In contrast, large soluble aerosols can produce large cloud droplets that are faster to coalesce and form precipitation. Enhancing precipitation by seeding clouds with large hygroscopic particles and ice nuclei is not a new idea. However, this practice did not enjoy much credibility, mainly due to the large natural variability in the conditions compared to the magnitude to the effects of advertent cloud seeding (Silverman 2001, 2003). Natural hygroscopic cloud seeding occurs on a grand scale by salt aerosols from evaporated sea spray. This was shown to be an important mechanism that restores the precipitation from clouds that form in polluted air that flows to the Indian Ocean from south Asia (Rosenfeld et al. 2002), and so cleansing the air from the particulate air pollution. This process leads to the conversion of the polluted continental air mass to a pristine maritime air mass. Therefore, it appears that the oceans are the “green lungs” of the atmosphere, to a large extent because they are salty. The effect occurs not only over oceans, but also over salty land surface areas, such as the desiccated bed of the Aral Sea (Rudich et al. 2002). Due to the ocean salt aerosols, the detrimental impacts of air pollution on precipitation are limited mainly to the land areas, most strongly in the densely populated areas, where people rely on the precipitation for their livelihood. The net effect can take the form of redistribution of precipitation from land to the ocean. More quantitative assessment of that awaits global circulation models that can take these effects into account. Desert dust has mixed effects, because it can contain large particles with varying amounts of soluble materials (Rosenfeld et al. 2001). In any case, desert dust can enhance precipitation due to its strong ice nucleating ability (Rosenfeld and Farbstein 1992; Rosenfeld et al. 2001). This provides us with hope that we will be able to affect the precipitation favorably by both controlling emission and deliberate release of aerosols that incur the desired effects. Spaceborne measurements of the natural, inadvertent and advertent effects on precipitation are an emerging technology with large but yet little exploited potential. The METEOSAT Second Generation (MSG) satellite that was commissioned in early 2004, has opened a new era in which continuous
Is Man Actively Changing the Environment ?
21
(scan every 15 minutes) microphysical satellite retrieval is possible both during day and night times (see Section 2 in this book).
7
IMPACTS ON THE GENERAL CIRCULATION OF THE ATMOSPHERE
The evidence so far shows that anthropogenic aerosols and changes of land surface properties do affect substantially the precipitation and hence the latent heat release. Having defined “latent heat forcing” as the difference between natural and anthropogenically modified latent heat release, and in view of the substantial anthropogenic effects on precipitation, the latent heat appears to be a major component in the energetics of climate change, on a par with the greenhouse gases forcing. If this happens, these changes already must be with us and may possibly explain some of the oddities of recent climatic events. These changes are additional to those induced by the increased greenhouse gases. A global observation system from satellites that measures aerosols, clouds composition and precipitation, is the best way to document quantitatively these effects. General circulation models (GCM) do not contain yet the necessary processes to fully address this issue quantitatively. The best that could be done until now is sensitivity studies, which show considerable sensitivity of the climate system to the impact of anthropogenic aerosols on the precipitation (Nober et al. 2003). Because the hydrological cycle and the anthropogenic effects on it are such a major constituents of the climate system, proper observations and simulations of them are a necessary direction in future development, if we want to have any hope in proper understanding of man-made impacts on our environment and water resources. These new insights no doubt will be featured in future debates on climate change and the impact of pollution on the environment with water resources at the top of the list. Finally, in recognition of the recent findings the 14th Council of World Meteorological Organization, Geneva, May 2003, accepted the following resolution: “Congress noted with concern the new additional evidence, also presented at the 8th WMO Scientific Conference on Weather Modification, that was pointing to an apparent substantial reduction of the rainfall efficiency of clouds by plumes of smoke caused by biomass burning (agricultural practices, forest fires, cooking and heating) and industrial processes. Congress also noted the evidence that such non-raining clouds could regain their raining ability once they moved over oceans or large bodies of water (such as the Aral Sea) because sea-salt was then mixed into the clouds and overrode the detrimental effect of the smoke particles. Therefore, Congress recommended CAS to establish an ad hoc Group on Biomass Burning and Smoke Plumes in general, charge it to prepare a summary report for information of the Members,
D. Rosenfeld
22
addressing relevant issues such as (1) the climatology of smoke and weather active aerosol (Cloud Condensation Nuclei or CCN) plumes, (2) the in situ and remote measurement of CCN and cloud droplet concentrations, (3) strategies to reduce biomass burning and hence the density of smoke plumes, and (4) the seeding procedures and evaluation methods to re-establish raining ability of clouds affected by smoke plumes, and CAS to report to Fifteenth Congress.”
8
REFERENCES
Andreae, M. O., D. Rosenfeld, P. Artaxo, A. A. Costa, G. P. Frank, K. M. Longo, and M. A. F. Silva-Dias, 2004: Smoking rain clouds over the Amazon. Science, 303, 1337–1342. Borys, R. D., D. H. Lowenthal, S. A. Cohn, and W. O. J. Brown, 2003: Mountain and radar measurements of anthropogenic aerosol effects on snow growth and snowfall rate. Geophys. Res. Lett., 30, 1538, doi: 10.1029 /2002GL016855. Braham, R. R., Jr., 1981: Summary of urban effects on clouds and rain. Meteorological Monographs, Boston, 18, 141–152. Cerveny, R. S. and R. C. Balling. Jr, 1998: Weekly cycles of air pollutants, precipitation and tropical in the coastal NW Atlantic region. Nature, 394, 561–563. Coakley, J. A., R. L. Bernstein, and P. R. Durkee, 1987: Effects of ship-stack effluents on cloud reflectivity. Science, 237, 1020–1022. Dai, A., I. Y. Fung, and A. D. Delgenio, 1997: Surface observed global land precipitation variations during 1900–88. J. Climate, 10, 2943–2962. Eagen, R. C., P. V. Hobbs, and L. F. Radke, 1974: Particle emissions from a large Kraft paper mill and their effects on the microstructure of warm clouds. J. Appl. Meteor., 13, 535–552. Garratt, J. R., D. M. O’Brian, M. R. Dix, J. M. Murphy, G. L. Stephens, and M. Wild, 1999: Surface radiation fluxes in transient climate simulations. Global Planet. Change, 20, 33–55. Gatz, D. F., 1979: Investigation of pollutant source strength rainfall relationships at St. Louis. J. Appl. Meteor., 18, 1245–1251. Gilgen, H., M. Wild, and A. Ohmura, 1998: Means and trends of short wave irradiance at the surface estimated from Global Energy Balance Archive Data. J. Climate, 11, 2042–2061. Givati, A. and D. Rosenfeld. Quantifying precipitation suppression due to air pollution. J. Appl. Meteor., 43, 1038–1056. Grimm, A. M. and P. L. Silva Dias, 1995: Analysis of tropical extratropical interactions with influence functions of a barotropic model. J. Atmos. Sci., 52, 3538–3555. Gunn, R. and B. B. Phillips, 1957: An experimental investigation of the effect of air pollution on the initiation of rain. J. Meteor., 14, 272–280. Johnson, D. B., 1982: Role of giant and ultragiant aerosol particles in warm rain initiation. J. Atmos. Sci., 39, 448–460. Kaufman, Y. J. and R. S. Fraser, 1997: The effect of smoke particles on clouds and climate forcing. Science, 277, 1636–1638. Khain, A. P., D. Rosenfeld, and A. Pokrovsky, 2001: Simulating convective clouds with sustained supercooled liquid water down to –37.5oC using a spectral microphysics model. Geophys. Res. Lett., 28, 3887–3890. Liepert, B. G., 2002: Observed reductions of surface solar radiation at sites in the United States and worldwide from 1961 to 1990. Geophys. Res. Lett., 29, art. no. 1421. Nober, F., H.-F. Graf, and D. Rosenfeld, 2003: Sensitivity of the global circulation to the suppression of precipitation by anthropogenic aerosols. Global Planet. Change, 37, 57–80. Ohmura, A., 2004: Cryosphere during the twentieth century. Geophys. Monogr., 150, 239–257. Pinsky, M. B., A. P. Khain, D. Rosenfeld, and A. Pokrovsky, 1998: Comparison of collision velocity differences of drops and graupel particles in a very turbulent cloud. Atmos. Res., 49, 99–113.
Is Man Actively Changing the Environment ?
23
Ramanathan, V., P. J. Crutzen, J. T. Kiehl, and D. Rosenfeld, 2001: Aerosols, climate and the hydrological cycle. Science, 294, 2119–2124. Roderick, M. L. and G. D. Farquhar, 2002: The cause of decreased pan evaporation over the past 50 years. Science, 298, 1410–1411. Rosenfeld, D., 1999: TRMM Observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 3105–3108. Rosenfeld, D., 2000: Suppression of rain and snow by urban and industrial air pollution. Science, 287, 1793–1796. Rosenfeld, D. and H. Farbstein, 1992: Possible influence of desert dust on seedability of clouds in Israel. J. Appl. Meteor., 31, 722–731. Rosenfeld, D. and I. M. Lensky, 1998: Spaceborne sensed insights into precipitation formation processes in continental and maritime clouds. Bull. Amer. Meteor. Soc., 79, 2457–2476. Rosenfeld, D. and C. W. Ulbrich, 2003: Cloud microphysical properties, processes, and rainfall estimation opportunities. Chapter 10 of “Radar and Atmospheric Science: A Collection of Essays in Honor of David Atlas”. R. M. Wakimoto and R. Srivastava, Eds., AMS, Meteorological Monographs, 52, 237–258. Rosenfeld, D. and W. L. Woodley, 2000: Convective clouds with sustained highly supercooled liquid water down to –37.5oC. Nature, 405, 440–442. Rosenfeld, D. and W. L. Woodley, 2001: Pollution and Clouds. Physics World, Institute of Physics Publishing LTD, Dirac House, Temple Back, Bristol BS1 6BE, UK, 33–37. Rosenfeld, D. and W. L. Woodley, 2003: Closing the 50-year circle: From cloud seeding to space and back to climate change through precipitation physics. Chapter 6 of “Cloud Systems, Hurricanes, and the Tropical Rainfall Measuring Mission (TRMM)” W.-K. Tao and R. F. Adler, Eds., 234 pp., AMS, Meteorological Monographs 51, 59–80. Rosenfeld, D., Y. Rudich, and R. Lahav, 2001: Desert dust suppressing precipitation – A possible desertification feedback loop. Proc. Natl. Acad. Sci.USA, 98, 5975–5980. Rosenfeld, D., R. Lahav, A. P. Khain, and M. Pinsky, 2002: The role of sea-spray in cleansing air pollution over ocean via cloud processes. Science, 297, 1667–1670. Rudich Y., D. Rosenfeld, and O. Khersonsky, 2002: Treating clouds with a grain of salt. Geophys. Res. Lett., 29, doi:10.1029/2002GL016055, 2002. Russell, G. L., J. R. Miller, and D. Rind, 1995: A coupled atmosphere – ocean model for transient climate change studies. Atmos. – Ocean, 33, 687–730. Silverman, B. A., 2001: A critical assessment of glaciogenic seeding of convective clouds for rainfall enhancement. Bull. Amer. Meteor. Soc., 82, 903–923. Silverman, B. A., 2003: A critical assessment of hygroscopic seeding of convective clouds for rainfall enhancement, Bull. Amer. Meteor. Soc., 84, 1219–1230. Shepherd, J. M., H. Pierce, and A. J. Negri, 2002: Rainfall modification by major urban areas: Observations from spaceborne rain radar on the TRMM satellite. J. Appl. Meteor., 41, 689–701. Shepherd, J. M. and S. J. Burian, 2003: Detection of urban-induced rainfall anomalies in a major coastal city. Earth Interactions, 4, 1–17. Stanhill, G. and S. Cohen, 2001: Global dimming: a review of the evidence for a widespread and significant reduction in global radiation with discussion of its probable causes and possible agricultural consequences. Agricultural and Forest Meteorology, 107, 255–278. Wild, M., A. Ohmura, and U. Cubasch, 1997: GCM simulated surface energy fluxes in climate change experiments. J. Climate, 10, 3093–3110. Wild, M., A. Ohmura, H. Gilgen, and D. Rosenfeld, 2004: On the consistency of trends in radiation and temperature records and implications for the global hydrological cycle. Geophys. Res. Lett., 31, L11201, doi:10.1029/2003GL019188.. Williams, E. and S. Stanfill, 2002: The physical origin of the land-ocean contrast in lightning activity. C. R. Physique, 3, 1–16.
24
D. Rosenfeld
Williams, E., D. Rosenfeld, N. Madden, J. Gerlach, N. Gears, L. Atkinson, N. Dunnemann, G. Frostrom, M. Antonio, B. Biazon, R. Camargo, H. Franca, A. Gomes, M. Lima, R. Machado, S. Manhaes, L. Nachtigall, H. Piva, W. Quintiliano, L. Machado, P. Artaxo, G. Roberts, N. Renno, R. Blakeslee, J. Bailey, D. Boccippio, A. Betts, D. Wolff, B. Roy, J. Halverson, T. Rickenbach, J. Fuentes, and E. Avelino, 2002: Contrasting convective regimes over the Amazon: Implications for cloud electrification. J. Geophys. Res., 107, D20, 8082, doi:10.1029/2001JD000380.
3 THE GLOBAL PRECIPITATION CLIMATOLOGY PROJECT Arnold Gruber1, Bruno Rudolf 2, Mark M. Morrissey3, Toshiyuki Kurino4, John Janowiak5, Ralph Ferraro1, Richard Francis6, Albert Chang7†, and Robert F. Adler7 1
National Oceanic and Atmospheric Administration, National Environmental Satellite and Data Information Service, Camp Springs, MD, USA (current affiliation: Cooperative Institute for Climate Studies, Earth System Science Interdisciplinary Center, University of Maryland, College Park, MD, USA) 2 Deutscher Wetterdienst, Offenbach, Germany 3 University of Oklahoma, School of Meteorology, Norman, OK, USA 4 Japan Meteorological Agency, Tokyo, Japan 5 National Oceanic and Atmospheric Administration , National Weather Service,Climate Prediction Center, Camp Springs, MD, USA 6 EUMETSAT, Darmstadt, Germany (current affiliation: Met Office, Exeter, United Kingdom) 7 National Aeronautics and Space Administration, Goddard Space Flight Center, Greenbelt, MD, USA
1
INTRODUCTION
The global precipitation climatology project was formed by the World Climate Research Program in 1986 (WCRP 1986) to exploit the capabilities of satellites, combined with gauges to provide the best available estimates of global precipitation. The objectives were to: • Improve understanding of seasonal to interannual and longer-term variability of the global hydrological cycle. • Determine the atmospheric heating needed for climate prediction models. • Provide an observational data set for model validation and initialization and other hydrological applications.
†
Deceased 26 May 2004
25 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 25–35. © 2007 Springer.
A. Gruber et al.
26
Figure 1. Organization structure of the GPCP.
Initially the goal was to provide 10 years (1987–1996) monthly mean fields of precipitation on a 2.5 degree × 2.5 degree latitude/longitude grid. However, as the GPCP evolved it was possible to extend the time period back to 1979 (Adler et al. 2003) and produce data sets at pentad, 2.5 degree × 2.5 degree scales (Xie et al. 2003 ) and global daily estimates of precipitation at 1 degree × 1 degree latitude/longitude scales (Huffman et al. 2001 ). Examples of these data sets will be shown later.
2
ORGANIZATION
Since first described by Arkin and Xie (1994) the organization of the GPCP has been modified to the current structure shown in Fig. 1. The geostationary satellite operators of Japan (GMS), Europe (METEOSAT), and the USA (GOES) collect geostationary infrared satellite data for the estimation of precipitation using the GPI (Arkin and Meisner 1987). The USA also provides NOAA polar orbiting data for areas where there is no geostationary coverage, typically over the Indian Ocean area. These data, covering the latitudes 40N–40S are combined at the Geostationary Satellite Precipitation Data Center located at the NOAA National Weather Service. Microwave estimates are provided by two centers. One is located at NASA/GSFC, USA, and provides oceanic estimates of rainfall based on the emission characteristics of precipitation at low frequencies (19 GHz) (Wilheit et al. 1991) and the other is obtained over land using the scattering characteristic of raining clouds (Ferraro and Marks 1995). The latter is
The Global Precipitation Climatology Project
27
located at NOAA/NESDIS in the USA. Gauge data are collected and analyzed at the Global Precipitation Climatology Center (GPCC) located at the Deutscher Wetterdienst (DWD), Germany. These estimates of rain (MW, IR, and gauges) are sent to the GPCP Merge Development Center where they are merged and global precipitation maps are prepared. The basic merging procedure including error estimates is described by Huffman et al. (1997). For the latest version of the monthly mean products rain estimates from the TOVS sounder are utilized in high latitude precipitation estimates where the infrared and microwave estimates are either not available or unreliable. The Surface Reference Data Center located at the University of Oklahoma has the responsibility for collecting and maintaining high quality reference data and for performing validation of the rain products. In the following sections a brief review of the data sets and some of the applications of the data will be provided.
3
DATA SETS
The GPCP produces three global rainfall data sets; monthly mean on 2.5 × 2.5 degree latitude/longitude grid, pentad mean also on a 2.5 × 2.5 degree grid and a daily rainfall map on 1 × 1 degree latitude/longitude grid. These data sets share much data in common but there are differences in the input data. One important factor is that there are different satellite sources of data with different lengths of time especially for the monthly and pentad means. The input data to the various products are summarized in Table 1. More details including the procedures for the product development are in the references cited in the text below, which briefly summarizes each product. The monthly mean is the second version of this data set and differs from the first version in that it is extended back in time to 1979 and has complete global coverage (Version 1 began in 1987 and suffered extensive gaps pole ward of 60 N, S latitude). The monthly mean is a blend of satellite infrared and microwave estimates of rainfall and gauges. The analysis procedure is to use stepwise bias corrections; i.e., infrared estimates are adjusted to microwave estimates (presumed less biased) and then satellite estimates adjusted to gauges. The final blending uses inverse error weighting. Version 1 is described more completely in Huffman et al. (1997) and Version 2 is described by Adler et al. (2003). The annual average rainfall from Version 2 for the period 1979–1998 is shown in Fig. 2. The global average rainfall is 2.61 mm per day. The average over land is 1.96 mm per day and the value over the oceans is 2.85 mm per day. The largest amounts of rain occur over the tropical oceans approaching 10 mm per day in the equatorial western Pacific. It is interesting to note that the precipitation in the NH mid-latitude storm tracks exhibit values that are as high as some of the tropical areas, especially over land.
A. Gruber et al.
28 Table 1. Summary of data sources for GPCP products.
GPCC Gauge Data GHCN +CAMS Gauge Data SSM/I Data Geostationary IR MSU TOVS Polar IR * OPI GTS gauge data
Monthly Mean V2 1986–present
Pentad
Daily
07/1987–11/1987, 01/1988–present 1986–present 1979–present
07/1987–11/1987, 01/1988–present 1997–present
1979–1986 07/1987–11/1987, 01/1988–present 1986–present 07/1987–11/1987, 01/1988–present 1986–present 01/1979–06/1987, 12/1987
1997–present 1997–present 1979–present
* Used to fill gaps in geostationary IR OPI – Outgoing longwave radiation precipitation Index GHCN – Global Historical Climate Network CAMS – Climate Assessment Monitoring System (Climate Prediction Center, NOAA) SSM/I – Special Sensor Microwave Imager MSU – Microwave Sounding Unit TOVS – TIROS Operational Vertical Sounder GTS – Global telecommunication System GPCC – Global precipitation Climatology Centre.
Figure 2. GPCP annual mean precipitation 1979–1998 based on Version 2.
The Global Precipitation Climatology Project
29
The Pentad data set also is a blend of satellite estimates and gauges and extends from 1979 and as the Version 2 is continuing. The analysis procedure is somewhat different than used in the monthly mean product. Satellite rain estimates are combined by maximum likelihood estimates, and then bias is removed by solving a Poisson equation with gauges as boundary conditions. The pentad data sum to the GPCP Version 2 monthly means. The pentad data development is described in Xie et al. (2003). The daily global precipitation product is based only satellite estimates. The methodology of using both microwave and infrared is somewhat different than in Version 2 and is fully described by Huffman et al. (2001). The main reason for satellite only is that daily gauges are not uniform in their definition of a day so that it is not practical to obtain a global analysis from gauges that will be homogeneous. However, the influence of the gauges is felt since the daily estimates are constrained to sum to the monthly totals from Version 2. The advantages of the higher temporal resolution of the daily and the pentad data sets is that important variability in precipitation such as the 30–60 day Madden–Julian Oscillations can be detected and monitored. And the higher spatial resolution of the daily data are more useful for hydrological and water management applications. An example of what is gained from the higher spatial temporal resolution is shown in Fig. 3 which shows time longitude sections for the latitude band 5N-5S for the period January 1997–October
Figure 3. Time longitude sections at 5N-5S, January 1997–October 1998; monthly, pentad and daily data (see also color plate 1).
A. Gruber et al.
30
1998, which encompasses the 97/98 El Niño. Monthly, pentad and daily sections are shown. Note that the El Niño is well depicted in all the sections but the higher resolution data can depict Madden–Julian Oscillations and daily disturbances.
4
VALIDATION
An important aspect of the work done by the GPCP is validation of the data. Figure 4 shows some the validation of the monthly mean GPCP rain estimates against data from a mesoscale gauge precipitation network over Oklahoma, for a specific grid box. These gauges were not used in the gauge analysis that was merged with the satellite estimates of rain. The upper part of the figure shows a time series from January 1994 through 1998. The red curve is the multisatellite (MW and IR) estimates of rain, the green curve is the satellite and gauge merged estimates and the blue line is the reference gauge data.
Figure 4. Validation of GPCP monthly mean precipitation. Top –time series, bottom – scatter plot. See text for explanations.
The Global Precipitation Climatology Project
31
This comparison shows the importance of the gauges in the blending procedure in determining the correct magnitude of the rainfall. The lower part of the figure is a scatter plot between of the GPCP rain against the reference data. There is essentially no bias and the correlation between the two data sets is 0.93. More information about validation can be obtained from the Surface Reference Data web page: http://srdc.evac.ou.edu/.
5
APPLICATIONS
As indicated in the introduction some of the goals were to produce a data set that can be used to understand the seasonal and longer variability of precipitation and use these data for model validation. Some examples of the application of these data will be shown in this section. Seasonal variability is shown in the time longitude sections of precipitation anomalies of Fig. 5 for the tropical zone 5N-5S for the period July 1987 through 1998. This is a particularly good area to look at seasonal variability since it is influenced by the quasi-periodic El Niño. The right hand side of the figure shows anomalies of rain which is contrasted to anomalies in sea surface temperature shown on the left-hand side. The points to notice are the good agreement between the anomalies especially in depicting El Niño and La Niña conditions. Noteworthy are the dominant El Niño of 1997/98 and the weaker and extended El Niño conditions of 1991–1995.
Figure 5. Time latitude sections of sea surface temperature anomalies (left) and monthly mean precipitation anomalies (right).
32
A. Gruber et al.
With regard to looking for trends in the data one has to be cautious since the record is relatively short about 23 years and composed of estimates from different satellites and times and different algorithms. When we look at the time series of globally averaged precipitation, shown in Fig. 6, we see a weak negative trend; however, in view of the variability of the time series and the use of various satellite inputs over the period this weak negative trend is not believable. However, when we look at trends on a regional basis we can find areas where the trend is more significant (Fig. 7). For example the area in the western pacific shows a region of large negative and positive trends. These trends are consistent with analysis of atoll data performed by Morrissey and Graham (1996). Another application of the data is validation of model outputs. In the example below we show difference maps between GPCP and NCEP/NCAR Reanalysis data (Janowiak et al. 1998). Figure 8 shows that over the ocean heavy rain areas (ITCZ, SPCZ, storm tracks) the reanalysis underestimates rain and in the tropical dry zones it overestimates rain. There are also overestimates of rain over tropical land areas of NE South America extending into the Caribbean and SE. Comparisons such as this can provide invaluable information for diagnosing deficiencies in the model physics and parameterization schemes.
Figure 6. Time series of globally averaged monthly mean precipitation. Red line is 12-month running mean, blue line is 5-month running mean. Trend is mm per day per month.
The Global Precipitation Climatology Project
33
Figure 7. Regional trends – mm per day per month. Note values are multiplied by 100.
Figure 8. Mean annual difference between NCEP/NCAR reanalysis data and the GPCP product.
6
CONCLUDING REMARKS AND FUTURE OUTLOOK
The GPCP has been successful in developing and producing rainfall data sets that are now useful and will continue to do so. We have exceeded our original mandate which was to produce a 10-year monthly mean data set on a 2.5 degree × 2.5 degree grid. In fact we are producing monthly mean and pentad
A. Gruber et al.
34
data set from 1979 onward on a 2.5 degree grid and a daily data from 1997 onward on a 1 degree × 1 degree grid. Nevertheless there are still challenges for the GPCP. Among the them are: • The determination of absolute values of rain over the open ocean. This is difficult to achieve because there is generally no reference data available. However, the use of the TRMM precipitation radar data should prove to be of value for this problem. • The determination of solid precipitation rates is still largely unknown, although the water equivalent of solid precipitation is included over land through the gauges. Recently some work has been done that suggests that high frequency (150, 183 GHz) passive microwave data available from AMSU may help in identifying solid precipitation. • Accurate estimates of precipitation in regions of complex terrain. This is a challenge for both in situ and remote-sensing measurements.
Acknowledgments: The NOAA Office of Global Programs is to be acknowledged for their support. The statements contained within the manuscript are not the opinions of the funding agencies or the US government, but reflect the authors’ opinions.
7
REFERENCES
Adler, R. F., G. J. Huffman, A. Chang, R. Ferraro, P.-P. Xie, J. Janowiak, B. Rudolf, U. Schneider, S. Curtis, D. Bolvin, A. Gruber, J. Susskind, and P. A. Arkin, 2003: The Version 2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 1147–1167. Arkin, P. A. and B. N. Meisner, 1987: the relationship between large scale convective rainfall and cold cloud over the western hemisphere during 1982–1984. Mon. Wea. Rev., 115, 41–74. Arkin, P. A. and P.-P. Xie, 1994: The Global Precipitation Climatology Project: First Algorithm Intercomparison Project. Bull. Amer. Meteor. Soc., 75, 401–419. Ferraro, R. and G. Marks, 1995: The development of SSM/I rain-rate retrieval algorithms using ground based radar measurements. J. Atmos. Oceanic Technol., 12, 755–770. Huffman, G. J., R. F. Adler, P. A. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The global precipitation climatology project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78, No.1, 5–20. Huffman, G. J., R. F. Adler, M. M. Morrissey, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multi-satellite observations. J. Hydrometeor., 2, 36–50. Janowiak, J. E., A. Gruber, C. R. Kondragunta, R. E. Livezey, and G. J. Huffman, 1998: A comparison of the NCEP/NCAR reanalysis precipitation and the GPCP rain gauge-satellite combined data set with observational error considerations. J. Climatol., 11, 2960–2979. Morrissey, M. and N. E. Graham, 1996: Recent trends in rain gauge precipitation measurements from the Tropical Pacific: evidence for an enhanced hydrological cycle. Bull. Amer. Meteor. Soc., 77, 1206–1219.
The Global Precipitation Climatology Project
35
WCRP, 1986: Report of the workshop on global large scale precipitation data sets for the World Climate Research Programme. WCP-111, WMO/TD-No. 94, 45 pp. Wilheit, T. J., A. T. C. Chang, and L. S. Chiu, 1991: Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution functions. J. Atmos. Oceanic Technol., 8, 118–136. Xie, P.-P., J. E. Janowiak, P. A. Arkin, R. F. Adler, A. Gruber, R. Ferraro, G. J. Huffman, and S. Curtis, 2003: GPCP pentad precipitation analyses: An experimental data set based on gauge observations and satellite estimates. J. Climate, 16, 2197–2214.
4 OCEANIC PRECIPITATION VARIABILITY AND THE NORTH ATLANTIC OSCILLATION Phillip A. Arkin1, Heidi M. Cullen2, and Pingping Xie3 1
ESSIC, University of Maryland, College Park, MD, USA The Weather Channel, Atlanta, GA, USA 3 Climate Prediction Center, NOAA/NWS/NCEP, Camp Springs, MD, USA 2
1
BACKGROUND
The North Atlantic Oscillation (NAO) was originally discovered by Sir Gilbert Walker (Walker and Bliss 1932). Walker found that the strengths of the semipermanent low and high pressure systems in the North Atlantic Ocean were inversely correlated during Northern Hemisphere winter. Periods when both systems were stronger than average, resulting in an intensified pressure gradient and stronger than average westerly winds, alternate with periods when both are weaker than average and the oceanic westerlies are relatively weak. Changes from strong to weak gradient conditions are characterized by a “seesaw” of atmospheric mass from south to north in the North Atlantic Ocean. The NAO is the dominant coherent mode of climate variability in Boreal winter in the region (van Loon and Rogers 1978), with a strong signature throughout the troposphere. Variability associated with the NAO has been shown to control the location and intensity of the storm track in the North Atlantic and Western Europe (Hurrell 1995), and exerts a strong influence on temperature and precipitation anomalies in eastern North America, Europe and the Mediterranean region (Cullen et al. 2002). Indices representing the variability of the NAO have been computed from a variety of atmospheric data. The original and simplest such index is based on the normalized difference between surface pressure at stations that represent the variations of the Azores High and the Icelandic Low (van Loon and Rogers 1978). More recently, in an attempt to capture more effectively
37 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 37–47. © 2007 Springer.
P. A. Arkin et al.
38
the large-scale variability of the full atmospheric phenomenon, indices based on empirical orthogonal function analysis of surface pressure or midtropospheric geopotential height have been developed and published (Barnston and Livezey 1987). All indices of the NAO are formulated so that high values represent periods when the pressure gradient in the North Atlantic, and hence the westerlies, are stronger than average. Spectral analysis of NAO indices indicates that, while variability is found on all time scales longer than a few days, significant peaks are found at periods of 2–3 and 7–10 years, along with an increasing trend over the last 50 years (Hurrell 1995). Precipitation anomalies associated with the NAO have been described by Walsh and Portis (1999) and Cullen et al. (2002). However, precipitation variability over the Atlantic Ocean associated with the NAO and the associated atmospheric circulation features have not been well described, principally due to the lack of reliable oceanic precipitation data sets. The recent availability of global precipitation data sets may have reduced this impediment, and the remainder of this paper is an attempt to use these new data sets to provide a thorough description of the climate variability associated with the NAO. We begin with a description of the data used in the study, followed by a presentation of the regional manifestations of the NAO in precipitation, storminess and circulation during both the winter and summer seasons. In section 4 we investigate the tropical variations in precipitation associated with the NAO, and we finish in section 5 with conclusions and a discussion of further work required.
2
DATA
We utilize three primary data sets in this study. Precipitation will be obtained from the CPC Merged Analysis of Precipitation (CMAP; Xie and Arkin 1996, 1997), while analyses of the large-scale atmospheric circulation are derived from the NCAR/NCEP reanalysis (Kalnay et al. 1996). These will be supplemented by an analysis of the frequency of cyclonic storminess derived from surface pressure data. CMAP provides monthly maps of precipitation averaged over 2.5° × 2.5° areas for the period January 1979–December 2002 (regularly updated) based on a variety of estimates derived from satellite observations and rain-gauge data. The CMAP algorithm uses a linear combination of all available satellite estimates, with weights over land derived from a comparison to a gauge-only analysis. Over ocean, the weights are determined from retrospective analysis comparing the different satellite-derived estimates to atoll rain-gauge observations. These weights are extrapolated into higher latitudes in a seasonally and latitudinally varying pattern based on the characteristics of each product. In regions where no usable satellite-based estimate is available,
Oceanic Precipitation Variability and the North Atlantic Oscillation
39
a version of CMAP using precipitation forecasts from the NCEP/NCAR Reanalysis (Kalnay et al. 1996) is created. The model-derived precipitation amounts are treated as another estimate, with weights derived in a manner similar to the others. This permits the creation of two versions of CMAP: one that is spatially complete and includes information that is partially model-dependent and another that is purely observation-based but with gaps in coverage. In both cases, the intermediate product is combined with a gauge-based analysis to minimize bias. The combination is done in such a manner as to preserve the gradients in the satellite combination while using the absolute value of the gauge analysis where sufficient gauges are available. In this study we will use the version of CMAP that includes model-derived information as our primary source, because we expect that the model forecasts are sufficiently skillful in the high latitudes of the North Atlantic Ocean to be useful. The influence of the model forecasts on our results will be determined by comparing them to a parallel analysis using the observation-only CMAP. Circulation data used in this study will be derived from the NCEP/NCAR reanalysis (Kalnay et al. 1996). This reanalysis utilized a modern data assimilation/forecast system together with as complete as possible an observational database. Its results have been used in more than 1,000 published studies (Kalnay, personal communication, 2002). We have examined winds and geopotential heights at 1,000 and 500 hpa, and will use the 500 hpa geopotential height variability in this paper to describe the NAO signal in atmospheric circulation. Since it has been shown that the NAO is the principal control on the winter storm tracks in the region (Hurrell 1995), we will utilize an additional data set based on an automated tracking of cyclonic circulation centers in surface pressure fields. This analysis was produced by Chandler and Jonas (personal communication, Center for Climate Systems Research at Columbia University and NASA/GISS) and consists of gridded maps of the number of cyclonic centers passing through each region during each month from January 1962 to December 1998. In this study we use the results for the 20year period from January 1979 to December 1998.
3
REGIONAL MANIFESTATIONS
The manifestations of the NAO have been most extensively studied in the region of the North Atlantic Ocean and the adjacent portions of North America and Europe, and in the Northern Hemisphere winter season. In the following sections, we will investigate the precipitation, storminess and circulation anomalies that characterize the NAO during the Boreal winter (December–March) and Boreal summer (June–September) seasons.
40
P. A. Arkin et al.
3.1 Boreal winter In the long-term winter (DJFM) mean, the fields of precipitation, storm frequency and height (Fig. 1) exhibit a consistent relationship over the North Atlantic, with collocated axes of maximum storminess and precipitation associated with the strongest gradients in geopotential height. A bifurcation appears in the eastern Atlantic, with the strongest axis continuing over northern Europe and a secondary maximum extending over the Mediterranean. The highest values of precipitation are located to the southwest of the most frequent storminess. Monthly means were subtracted to obtain anomalies, and correlations calculated against the NAO index published on the Climate Prediction Center website. The patterns obtained (Fig. 2) are consistent with previous work but exhibit significant new details over the ocean. Precipitation exhibits an alternating pattern of correlation with the NAO, with positive values over the highest latitudes, negative between 30°–40° N, positive between the equator and 20° N, and negative further south. These correlations are strongest over the ocean, but extend into the continents.
Figure 1. Mean December–March CMAP precipitation (mm per day. top left), 500 hpa heights (dm, top right) and storm frequency (storms per month in each 2º × 2º area, below left). Contours are 0.2, 0.5, 1, 2, 4, 6, 8, 12 and 16 for precipitation, the same with an additional one at 0.1 for storminess, and at 10 dm interval for height, with the southernmost contour at 580 dm.
Oceanic Precipitation Variability and the North Atlantic Oscillation
41
Figure 2. Correlation of CMAP precipitation (top left), 500 hpa heights (top right) and storm frequency (below left) with NAO index during December–March. Contours for precipitation and storminess are at ± 0.15, 0.3, 0.4, 0.5 and 0.6; for height at ± 0.25, 0.4, 0.5, 0.6, 0.7 and 0.8.
The 500 hpa height correlations are less complex, with strong negative correlations in the North Atlantic, positive correlations to the south, and an additional center of negative correlation further south. This pattern is to be expected, considering that this NAO index is based on empirical orthogonal function analysis of 700 hpa height anomalies. The frequency of storms exhibits a correlation pattern similar to that of precipitation in the North Atlantic, but with essentially no sign of the area of positive correlations south of 25°N. Composites of precipitation, storminess frequency, and height computed based on quartiles of the NAO index using monthly anomalies confirm the implications of the correlation plots: high index periods are accompanied by increased precipitation and storminess in the higher latitudes, with an anomalous height gradient somewhat further south than the maxima in precipitation and storm frequency (see Fig. 3). Low index periods exhibit decreases in precipitation and storminess in the northern latitudes, with increases between 30°–40° N extending from the Atlantic eastward as far as the Middle East. The westerly anomalous wind maxima at 1,000 and 500 hpa are located south of the greatest increases in precipitation and storminess.
42
P. A. Arkin et al.
Figure 3. Composite CMAP precipitation (mm per day, left) and storm frequency (storms per month, right) anomalies based on quartiles of NAO index during December–March. Plots are difference (high minus low) between high and low index months. Contours are at ± 0.2, 0.5, 1, 2, 4 and 8.
While the consistency of these results with those of earlier studies is encouraging, we attempted to explicitly address their robustness. Since the NAO exhibits relatively little serial correlation on the monthly time-scale (one month lag correlation of 0.16), correlations with an absolute value of 0.3 or greater are highly significantly different from zero. As one can see from Fig. 2, the areas of significance in geopotential height are large; while the values of the precipitation correlations are slightly lower, they are still significant over substantial areas and the pattern is similar. The negative correlation between precipitation and NAO near the equator in the Atlantic is significant as well. Assessing the robustness of composite patterns is more challenging; explicit statistical tests are not well suited as many of the underlying assumptions are violated. Here we have used Monte Carlo testing to determine the relationship between composites chosen based upon an NAO
Figure 4. Composite CMAP precipitation anomalies during the highest (left) and lowest (right) quartiles of December – March months based upon the NAO index normalized by the standard deviation of 1000 randomly selected composites. Contours are at ±1, 1.5, 2 and 3 standard deviations.
Oceanic Precipitation Variability and the North Atlantic Oscillation
43
index and those chosen based on random time series. A set of 1,000 composites of CMAP precipitation for the months December–March based on random series was created, using the same methodology as described above. The standard deviation of the sets of high and low composites was calculated and used to normalize the composites based upon the NAO index. The results are shown in Fig. 4, and indicate that at least the northern triad of composite anomalies is reliably distinct from noise. Further, these patterns are very similar, although reversed, during periods of high and low NAO index.
3.2 Boreal summer The long-term summer mean circulation, precipitation, and storm frequency patterns are displaced northward and weakened relative to those of winter. The subtropical features in geopotential height are stronger relative to the higher latitude features. The oceanic precipitation maximum in the middle latitudes shows evidence of strong tropical influence, while the maximum in storm frequency is displaced well northward and is quite weak over Europe.
Figure 5. Correlation of CMAP precipitation (top left), 500 hpa heights (top right) and storm frequency (below left) with NAO index during June – September. Contours as in Fig. 2 except for height, which here has the same contours as precipitation and storminess.
44
P. A. Arkin et al.
Correlations between the NAO index and precipitation, storm frequency and 500 hpa geopotential height (Fig. 5) show that, while the NAO continues to influence the circulation during summer, its manifestations are markedly weaker than in winter. The areas of correlation with absolute values >0.3 are smaller, and the multicomponent pattern observed during winter is not present. In geopotential height, particularly at 1,000 hpa (not shown), the correlation maxima are more fragmented than during winter, while in precipitation and storm frequency they are more restricted to northern latitudes. Precipitation correlations are strongest over Northern Europe, while storminess correlations extend more uniformly from eastern North America across the Atlantic and into Europe. One feature, the area of positive correlation between precipitation and the NAO in the western subtropical Atlantic, raises the question of whether some connection exists between the NAO and tropical storm activity. Some aspects of the changes in geopotential height correlation between winter and summer are notable. At 500 hpa, both the middle and the southern band of correlations exhibit two clear centers during the summer (Fig. 5), while during winter (Fig. 2) only the southernmost band exhibits dual centers. At 1,000 hpa (not shown), the singular dipole seen in winter is replaced in summer with dual centers in the southern band of positive correlations. Note that this implies that no simple index based on one pair of stations can capture the full variability associated with the NAO. In addition, the centers during the summer are displaced eastward relative to the winter. Composites of precipitation, storm frequency and geopotential height based on the NAO index for the months from June to September (not shown) largely confirm the results suggested by the correlations shown in Fig. 5. The storminess and precipitation features are displaced well northward relative to the winter, and the southern features found in precipitation during winter are not seen in summer. The composite anomalies in precipitation and geopotential height are stronger over the European continent during summer, rather than being strongest over the ocean as observed during winter. The storm frequency composite anomaly is more uniform over land and ocean than the other composites. The positive correlation between NAO index and precipitation in the western subtropical Atlantic is borne out by the composite precipitation anomaly, but is not supported by the other composites. The Monte Carlo calculations of precipitation composites give results supportive of these findings: the composite precipitation anomalies over the European continent and in northern latitudes are robust, with substantial areas exceeding 1.5 standard deviations. These features are approximately symmetric between low and high index periods. In the western tropical Atlantic, interestingly, an area of positive anomaly in precipitation that exceeds 1.5 standard deviations is found during high index months without any indication of a corresponding negative anomaly during low index months.
Oceanic Precipitation Variability and the North Atlantic Oscillation
4
45
GLOBAL ASSOCIATIONS
Hoerling et al. (2001) have shown that the increasing trend in the NAO index during the period from 1950 to 2000 is associated in atmospheric models forced by observed sea surface temperature variability with a trend in tropical precipitation over the Indian and Pacific oceans. Here we examine our shorter period of record to determine whether periods of high and low NAO index exhibit a systematic relationship to tropical precipitation variability. The correlations between the NAO and CMAP precipitation over the globe exceed 0.3 in absolute value in few areas outside the tropical and Northern Hemisphere Atlantic Ocean (not shown). The area of negative correlations hinted at in the southernmost latitudes in Fig. 2 (December– March) extends to about 20° S centered on the east coast of South America. The location and shape hints at a relationship to the South Atlantic Convergence Zone, and areas of negative correlation <0.3 are also seen in the region of the South Pacific Convergence Zone and in the convergence zone east of Southern Africa. An area of positive correlation exceeding 0.3 is found in the eastern Indian Ocean and western Maritime Continent. The composite anomalies in these regions are consistent with the correlation patterns and are substantial. However, these are in areas of large natural variability, and so when normalized by the standard deviation of the random composites only a few areas exceed 1 (Fig. 6). Here we see that the anomalies across the Maritime Continent are substantial during periods of high NAO index, and not at all during low index periods, while the convergence zone anomalies are less systematic. No such robust features are seen during the JJAS season.
Figure 6. Composite CMAP precipitation anomalies during the highest (left) and lowest (right) quartiles of December–MarchFigure 2. Correlation of CMAP precipitation (top left), 500 hpa heights (top right) and storm frequency (below left) with NAO index during December–March. Contours for precipitation and storminess are at ± 0.15, 0.3, 0.4, 0.5 and 0.6; for height at ± 0.25, 0.4, 0.5, 0.6, 0.7 and 0.8.
P. A. Arkin et al.
46
5
CONCLUSIONS AND POINTS FOR FURTHER STUDY
We have shown that the NAO has a robust signal in precipitation, storminess and atmospheric circulation in the Atlantic Ocean in Boreal winter, with anomalies extending to about 20° S. A Boreal summer signal is present, but is weaker and restricted to higher northern latitudes except for a positive correlation between NAO index and CMAP precipitation in the western subtropical Atlantic in the Northern Hemisphere. The relationship between the NAO and near-surface and mid-tropospheric geopotential height extends well into the tropics, but not to the Southern Hemisphere. Some evidence supporting a relationship between positive NAO index and positive anomalies in precipitation in the western Pacific and eastern Indian oceans is seen. A number of questions are raised by the results presented here: Are the results dependent on the time scale examined? Here we used monthly values, but the NAO has significant variability on scales down to a few days. Repeating these analyses using available pentad data should improve our understanding of the phenomenon. How sensitive are the results to the precipitation data set used? While the coherence with geopotential height leads us to believe that the results are robust, we will check the consistency of details by repeating the analyses using both the observation-only CMAP and the GPCP precipitation data. Is the tropical signal found here consistent with the results of Hurrell et al. (2001)? Are tropical precipitation variations (presumably associated with sea surface temperature anomalies) somehow influencing the NAO? Clearly further investigation is required. What role is sea surface temperature in the Atlantic Ocean playing? Is there evidence of local predictability associated with memory in the ocean?
6
REFERENCES
Barnston, A. G. and R. E. Livezey, 1987: Classification, seasonality and persistence of low frequency atmospheric circulation patterns. Mon. Wea. Rev., 115, 1083–1126. Cullen, H., A. Kaplan, P. A. Arkin, and P. B. deMenocal, 2002: Impact of the North Atlantic Oscillation on Middle Eastern climate and streamflow. Climatic Change, 55, 315–338. Hoerling, M. P., J. W. Hurrell, and T. Xu, 2001: Tropical origins for recent North Atlantic climate change. Science, 292, 90–92. Hurrell, J. W., 1995: Decadal trends in the North Atlantic Oscillation: Regional temperatures and precipitation. Science, 269, 676–679. Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Woollen, Y. Zhu, A. Leetmaa, B. Reynolds, M. Chelliah, W. Ebisuzaki, W. Higgins, J. Janowiak, K. C. Mo, C. Ropelewski, J. Wang, R. Jenne, and D. Joseph, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471. van Loon, H. and J. C. Rogers, 1978: The seesaw in winter temperatures between Greenland and Northern Europe. Part I: General description. Mon. Wea. Rev., 106, 296–310. Walker, G. T. and E. W. Bliss, 1932: World Weather V. Mem. Roy. Meteor. Soc., 4, 53–84.
Oceanic Precipitation Variability and the North Atlantic Oscillation
47
Walsh, J. E. and D. H. Portis, 1999: Variations of precipitation and evaporation over the North Atlantic Ocean. 1958–1997. J. Geophys. Res., 104, 16613–16631. Xie, P. and P. A. Arkin, 1996: Analyses of global monthly precipitation using gauge observations, satellite estimates and numerical model predictions. J. Climate, 9, 840–858. Xie, P. and P. A. Arkin, 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539–2558.
5 GLOBAL SATELLITE DATASETS: DATA AVAILABILITY FOR SCIENTISTS AND OPERATIONAL USERS Gilberto A. Vicente1 and the GES DAAC Hydrology Data Support Team NASA/Goddard Earth Sciences Distributed Active Archive Center, Greenbelt, MD, USA Members of the GES DAAC Hydrology Data Support Team include John Bonk2, Long S. Chiu1, Pat Hrubiak2, Zhong Liu1, Hualan Rui2 and William L. Teng2 Past members include S. Ahmad, L. Li, A. K. Sharma, G. Serafino 1 Center for Earth Observing and Space Research, George Mason University, Fairfax, VA, USA 2 Science Systems and Applications, Inc., Lanham, MD, USA Abstract
This article describes the global rainfall data and products collected, generated, archived and distributed by the Distributed Active Archive Center (DAAC) located in the National Aeronautics and Space Administration (NASA) Goddard Space Flight Center (GSFC). We focus on the 6-year-old Tropical Rainfall Measuring Mission (TRMM) satellite rainfall data and product generation and distribution by the DAAC, providing detailed information on the availability and access of other data sets and satellite products relevant for the development and validation of rainfall estimation algorithms and use. We also discuss many of the issues regarding to the rainfall data access, product generation and distribution, validation, documentation, development of standards, and intense interactions with a broad range of users.
Keywords
Rainfall estimation, TRMM, NASA, DAAC, satellite data, satellite products
1
INTRODUCTION
The availability of satellite data and products to both the operational and the scientific community (including those in universities, government, and other research organizations) operates in different ways and has distinct requirements. The role of a data/product distribution center is to provide valuable new data, such as the lower resolution data used in many operational applications, as well as higher resolution data required by the scientific community.
49 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 49–58. © United States Government 2007.
G. A. Vicente et al.
50
This creates the need of a close working relationship with the data providers in the designing and delivery of products, assists in the preparation and completeness of documentation, and provides continuity services across missions. The data/product distribution needs to maintain detailed level inventory and work with the science data user providing expertise in using data, help users produce special data products and support education and public outreach activities. It is also important that the satellite data/product distribution centers help in the definition of standards to insure that data are preserved in a usable form, facilitate the use of data by scientists who were not involved with the data collection. This article describes the rainfall data/product access and distribution activities in the Distributed Active Archive Center (DAAC) located in the National Aeronautics and Space Administration (NASA) Goddard Space Flight Center (GSFC). We concentrate on the 6-year-old Tropical Rainfall Measuring Mission (TRMM) satellite rainfall data and product generation and distribution by the DAAC. The TRMM satellite data and products have been used for a variety of scientific and operational applications through many means of distribution. It has led the DAAC to deal with many of the issues regarding to the data access, product generation and distribution, validation, documentation, development of standards, and intense interactions with a broad range of users.
2
A BRIEF HISTORY OF THE DAAC’S
Beginning in 1990, NASA established the Earth Observing System Distribution Information System (EOSDIS) Distributed Active Archive Centers (DAACs). These centers were selected on the basis of existing experience in data and information systems, resident science expertise in the use of the data to be held by the DAAC, and the willingness and ability of the host institution to make a long-term commitment to supporting the DAAC function for NASA. DAACs were individually oriented to specific Earth science disciplines, consistent with the science expertise resident at each DAAC’s host institution. The plan included to get the data management and user services infrastructure, to begin immediately to improve the quality of data and information services to science users and to prototype the new kinds of capabilities that would be required to support the Earth Observing System (EOS) program. This plan was conducted under the banner of EOSDIS Version 0. In 1990 a goal was established of having the DAAC’s and the supporting system operational by July of 1994. A major goal of the Version 0 effort was to build active and effective working relationships between the Earth Sciences Distribution Information Systems (ESDIS) Project and the DAACs. NASA recognized that organizational work had to be completed, with functioning DAACs working actively with the ESDIS Project and with each other, before implementation
Global Satellite Datasets
51
of the system needed at the DAACs to support Terra (formerly AM-1) era requirements. Otherwise organizational and cross-organization coordination and cooperation would have to be sorted out in the midst of the implementation and early operation of a new and of necessarily complex operational system. Version 0 became operational in August 1994 (one month later than the July 1994 goal established in 1991) when the last element of the system was put in place, the system level catalog capability. The DAAC archive, catalog, and distribution systems are the operational elements of Version 0, providing at least a baseline level of operational data management and user services, building on heritage systems at each DAAC. Emphasis was placed on user involvement with the emerging DAACs, and each DAAC was required to establish a User Working Group (whose membership was approved by NASA Headquarters) which would meet periodically (twice a year) to review DAAC plans and progress and provide guidance on science sensitive areas, such as prioritization of data sets to be held by the DAAC, services to be provided, etc. The science community sets priorities for data collected, archived, and distributed by the DAACs and for the services provided. A User Working Group, with international memberships of independent research scientists, oversees each DAAC’s management, meeting regularly to ensure that data products address user needs. The DAAC’s maintain close ties with science user communities and with remote-sensing instrument teams in order to provide useful products. They also collaborate with each other to provide standard services and documentation, access to common searching tools, and to preserve data for future generations of scientists. Figure 1 shows the archived versus the distributed data set volume distributed by the Goddard DAAC from its implementation in 1994 to the year 2003.
Figure 1. Cumulative statistics of all GDAAC satellite data and products as of August 2003.
52
3
G. A. Vicente et al.
THE NASA GSFC DAAC AND THE TRMM
The TRMM is a joint mission of NASA and the National Space Development Agency (NASDA) of Japan to monitor and study tropical and subtropical rainfall systems. TRMM has been acquiring data from shortly after its launch on November 28, 1997 to the present. All TRMM standard products are processed by the TRMM Science Data and Information System (TSDIS) and archived and distributed by the Goddard DAAC (GDAAC). In addition to the standard products the GDAAC generates and/or maintains a set of derived TRMM products (e.g., satellite coincidence subsets, parameter subsets, resampled gridded subsets, GIS-compatible files) to facilitate use of TRMM data by the general public. TRMM data are reprocessed with improved science algorithms approximately once per year, currently at version 5. The GDAAC stores archive and distribution information on TRMM standard and derived products in a database, as well as global ancillary data. In order to better understand the data usage patterns and requirements of TRMM users, statistics are routinely derived from the database for the entire TRMM data set or for specific groups of data products. For example, the total cumulative archive and distribution data volumes (files) for TRMM satellite standard products (uncompressed and as of December 2003) are 94.95 TB (5.15 × 106 files) and 112.13 TB (6.88 × 106 files), respectively. The Utilization Rate (UR) for the total, defined as the ratio of the distribution to archive total number of files of these satellite products is 1.34 (not including anonymous ftp distribution), (Fig. 2). Overall, the UR has increased steadily as TRMM progressed, and the trend is continuing. As measured by the UR, the most frequently requested satellite orbital data products are TMI brightness temperature, and PR and TMI rain profiles, with UR in the range of 10 and above. Most of the satellite gridded data products have a UR above 10, with a few above 20. Because some of the gridded products can also be accessed via anonymous ftp, the statistics of which are not included here, their UR is actually even higher. These statistics not only help the GDAAC to better serve its TRMM users, but also are useful inputs to the design of future satellite data support systems, such as those of the TRMM follow-on mission, the Global Precipitation Measurement (GPM) mission.
4
THE GDAAC TRMM DATA AND PRODUCTS
TRMM satellite data for each orbit are stored on board and transmitted to the ground via the NASA Tracking and Data Relay Satellite System (TDRSS). The TRMM Science Data and Information System (TSDIS) process the TRMM science data into standard products. These products are transferred
Global Satellite Datasets
53
Figure 2. Cumulative statistics of all TRMM satellite data archived and distributed as of December 2003. The Utilization ratio is the ratio of the number of requested files to that archived.
to the Goddard DAAC for archival and distribution and can be ordered via the TRMM Data Search and Order Web Interface. In addition to the standard products, the GDAAC also provides, as part of its value-added TRMM support to facilitate analysis and processing by users, certain special products, such as subsets, real-time multi-satellite precipitation data sets and ancillary data. Included as standard products are surface-based observations or rainfall from rain gauges and ground radars, which are used to calibrate and validate the satellite measurements. The TRMM data and products are classified in three distinct levels of complexities: Level 1 products are the Visible/Infrared Scanner (VIRS) calibrated radiances, the TRMM Microwave Imager (TMI) brightness temperatures, and the Precipitation Radar (PR) return power and reflectivity measurements. Level 2 products are derived geophysical parameters (e.g., rain rate and latent heat) at the same resolution and location as those of the level 1 data (orbital data products). Level 3 products are space-time averaged parameters (Wharton and Myers 1997), analyzed products or those produced from merging measurements from TRMM and other sources (gridded products). These include GOES Precipitation Index (GPI) and Special Sensor Microwave Imager (SSM/I) derived products.
G. A. Vicente et al.
54
4.1 TRMM orbital data products: Available from 12/20/1997 Orbital data products are level 1 time referenced instrument data at full resolution, radiometrically corrected with calibration coefficients, located by position on the Earth, and converted to a scale line or image format. They are also level 2 derived geophysical parameters at the same resolution and location of the level 1 data. The data and products are still in “strips” that trace the path of the satellite’s orbit. Table 1. TRMM orbital data products. Data Product
Description
Visible and Infrared Calibrated VIRS (0.63, 1.6, 3.75, 10.8, and 12 µm) radiances at 2.2 Radiance (VIRS) km resolution over a 720 km swath Microwave Brightness Temperature (TMI) Data Product
Calibrated TMI (10.65, 19.35, 21, 37, and 85.5 GHz) brightness temperatures at 5–45 km resolution over a 760 km swath Description
Radar Power (PR)
Calibrated PR (13.8 GHz) power at 4 km horizontal, and 250 m vertical, resolutions over a 220 km swath
Radar Reflectivity (PR)
Calibrated PR (13.8 GHz) reflectivity at 4 km horizontal, and 250 m vertical, resolutions over a 220 km swath
TMI Hydrometeor (cloud liquid water, precipitable water, cloud ice, Hydrometeor Profile precipitable ice) profiles in 14 layers at 5 km horizontal resolution, (TMI) along with latent heat and surface rain, over a 760 km swath Radar Surface Cross- PR (13.8 GHz) normalized surface cross-section at 4 km horizontal Section (PR) resolution and path attenuation (in case of rain), over a 220 km swath Radar Rain Characteristics (PR)
Rain type; storm, freezing, and bright band heights; from PR (13.8 GHz) at 4 km horizontal resolution over a 220 km swath
Radar Rainfall Rate PR (13.8 GHz) rain rate, reflectivity, and attenuation profiles, at 4 km and Profile (PR) horizontal, and 250 m vertical, resolutions, over a 220 km swath Combined Rainfall Profile (PR, TMI)
Combined PR/TMI rain rate and path-integrated attenuation at 4 km horizontal, and 250 m vertical, resolutions, over a 220 km swath
VIRS Raw Data
Reconstructed, unprocessed VIRS (0.6 3, 1.6, 3.75, 10.8, and 12 km) data
TMI Raw Data (TMI)
Reconstructed, unprocessed TMI (10.65, 19.35, 21, 37, and 85.5 GHz) data
Global Satellite Datasets
55
4.2 TRMM, SSM/I, GPI and other gridded data products These are geophysical parameters that have been spatially and/or temporally resampled from levels 1 and 2 data – fully processed products on a regular geographic grid and consistent time slices. Table 2. TRMM gridded data products. Data Product
Description
Begin Date
End Date
Rain rate, conditional rain rate, rain Monthly 5° × 5° frequency, and freezing height for a latitude 1998-01-01 Oceanic Rainfall band from 40° N to 40° S, from TMI
Current*
Monthly 1° × 1° Global rain rate from SSM/I SSM/I Rain Data Product
Begin Date
End Date
Total and conditional rain rate, radar reflectivity, path-integrated attenuation at 2, 4, 6, 10, 15 km for convective and stratiform 1997-12-01 rain; storm, freezing, and bright band heights, and snow-ice layer depth for a latitude band from 40° N to 40° S, from PR
Current*
Rain rate probability distribution at surface, 2 Monthly 5° × 5° km, and 4 km for a latitude band from 40° N 1997-12-01 Surface Rain Total to 40° S, from PR
Current*
Rain rate, cloud liquid water, rain water, Monthly 5° × 5° cloud ice, grauples at 14 levels for a latitude 1997-12-01 Combined Rainfall band from 40° N to 40° S, from PR and TMI
Current*
Daily 1° × 1° TRMM and Other- Calibrated geosynchronous IR rain rate using 1998-01-01 GPI Calibration TRMM estimates Rainfall
Current*
Monthly 1° × 1° Merged rain rate from TRMM, geosynchro1998-01-01 TRMM and Other nous IR, SSM/I, rain gauges Sources Rainfall
Current*
Monthly 5° × 5° and .5° × .5° Spaceborne Radar Rainfall
Description
1998-01-01 2001-08-01
* At the time of writing.
4.3 TRMM ancillary data sets Ancillary data sets provide additional information at the user disposal to facilitate in the TRMM algorithm development, intercomparison and ground validation studies. These data and products have been collected and stored in the GDAAC and are publicly available via FTP or web access.
G. A. Vicente et al.
56 Table 3. TRMM ancillary data products. Data Product
Horizontal Resolution
Description
1.0° × 1.0°
Spatial Coverage
Daily GPI
NOAA’s GPI IR Rain Data
Hourly GMS 5
Images of Visible and Infrared Spin Scan Radiometer (VISSR). Contains three IR 4.0 km and one visible channel.
Eastern Hemisphere
Monthly GPCC
Global Precipitation Climatology Centre (GPCC) rain gauge analysis for the Global 1.0° × 1.0° Precipitation Climatology Project GPCP)
Global
30 minutes GOES 8/10
GOES radiance data. Contains four IR and 4.0 km one visible channels
Western Hemisphere
Monthly GPCP
GPCP Version 2 Combined Precipitation 2.5° × 2.5° Data Set
Global
3 per day METEOSAT 7
Images from the visible, infrared and water 2.5–5.0 km vapor channels
Europe, the Middle East, Africa, and Atlantic Ocean
IR brightness temperature data, merged Hourly Merged from all available geostationary satellites 4.0 km Global IR (GOES 8/10, METEOSAT – 7/5 & GMS).
Global
Global
0.5° × 0.5° Global Hourly GPROF Gridded Orbit-by-orbit Precipitation Data Near-real 6.0 (SSM/I) Sets 0.25° × 0.25° time Global Monthly GSSTF
Goddard Satellite-based Surface Turbulent Fluxes
Version 1
2.0° × 2.5°
Version 2
1.0° × 1.0°
Global
Monthly GSSRB
Goddard Satellite-Retrieved Surface Radiation Budget
Daily TOVS
Contains temperature/humidity profiles, cloud cover information, and surface 1.0° × 1.0° parameters, derived from NOAA-11 and NOAA-14
Global
Daily AVHRR
NDVI product, and atmospherically corrected channel radiance
8 km
Global
6-hourly NCEP
NCEP 4-time daily analyses
1.0° × 1.0°
Global
0.5° × 0.5°
Global
Monthly CAMS Climate Analysis Monitoring System ETOPO5
5
0.5° × 0.5°
The Earth Topography Five Minute System 5 min
40N – 40S 90E – 170W
Global
TRMM DATA ACCESS AND SERVICES
TRMM standard products can be ordered from the GDAAC via the Web Hierarchical Ordering Mechanism (WHOM) (Rui et al. 1999) a data search and order web interface. The hierarchical organization of data into tables allows quick navigation to the data of interest, easily identification of temporal and spatial coverage via a calendar and color-coded mapping tools. The ordering is facilitated through a “shopping cart”. The WHOM system also
Global Satellite Datasets
57
allows the request of on-demand and on-the-fly subsettings by customizable pages by data set and product and access to data pool through direct FTP downloading. Whenever the FTP option is not viable for large data sets the GDAAC Hydrology Data Support Team distributes them in tapes. In addition to the standard and gridded products available via the WHOM, the GDAAC also provides the TRMM Online Visualization and Analysis System (TOVAS) (Liu et al. 2002), a user friendly web-based interface for visualization and analysis of gridded rainfall products and other precipitation data. It is applicable to a variety of researches and applications, such as climate study and monitoring, weather events study and monitoring, agricultural crop monitoring, rainfall algorithm study, and data products comparison. Currently there are five precipitation products available through the TOVAS. Three are TRMM products and the other two are monthly precipitation data and monthly precipitation data produced at the Global Precipitation Climatology Center.
6
RELEVANT ADDRESSES FOR DATA ACCESS
At the time of writing, these are relevant GDDAC web site addresses and FTP sites for satellite data and products access: Web sites: daac.gsfc.nasa.gov lake.nascom.nasa.gov/data/dataset/TRMM/01_Data_Products/06_Ancillary/ lake.nascom.nasa.gov/data/dataset/TRMM/01_Data_Products/04_Subset daac.gsfc.nasa.gov/CAMPAIGN_DOCS/hydrology/TRMM_analysis.html FTP sites: lake.nascom.nasa.gov/data/TRMM/Ground_Instruments/csi/ lake.nascom.nasa.gov/data/TRMM/Ancillary/ lake.nascom.nasa.gov/data/TRMM/Orbital/ lake.nascom.nasa.gov/data/TRMM/Gridded/ lake.nascom.nasa.gov/data/TRMM/Geo_Regions/ daac.gsfc.nasa.gov/data/hydrology/precip/gpcp/gpcp_v2_combined/aeolus.n ascom.nasa.gov/pub/merged/ larry.gsfc.nasa.gov/pub/ncep_data/
7
CONCLUSION
The access, archiving and distribution of global satellite data for research and operational use require the attention and work of a full time dedicated team. Since 1994 the NASA Goddard DAAC has been performing this task by providing free access to its satellite data/product archives, tools for data manipulation and one-to-one assistance to users. This article describes the TRMM data archiving and distribution process at the Goddard DAAC, as
G. A. Vicente et al.
58
well as the global ancillary precipitation related data and products need for algorithm development and ground validation. The Goddard DAAC is also responsible for the archiving and free distribution of other satellite data from the Earth Observing System (EOS) program such as the TERRA/AQUA Moderate Resolution Imager Spectroradiomenter (MODIS), AQUA Atmospheric Infrared Sounder (AIRS), Sea-viewing Wide Field-of-view Sensor (SeaWiFS) and the Solar Radiation and Climate Experiment (SORCE).
8
REFERENCES
Liu, Z., L. Chiu, W. Teng, and G. Serafino, 2001: A Simple online analysis tool for visualization of TRMM and other precipitation data sets. Science Data Processing Workshop, Greenbelt, MD, USA. Rui, H., B. Teng, A. K. Sharma, and L. Chiu, 1999: NASA/Goddard DAAC hierarchical search and order system for TRMM data: a web-based approach. 22nd IUGG General Assembly, Birmingham, UK. Wharton, S. W. and M. F. Myers, 1997: MTPE EOS Data Products Handbook, Volume 1, TRMM and AM-1. (Available from Code 902, NASA/Goddard Space Flight Center, Greenbelt, MD 20771, USA.)
Section 2 Cloud Studies in Support of Satellite Rainfall Measurements
6 CLOUD TOP MICROPHYSICS AS A TOOL FOR PRECIPITATION MEASUREMENTS Daniel Rosenfeld Hebrew University of Jerusalem, Institute of Earth Sciences, Jerusalem, Israel
1
INTRODUCTION
Rainfall measurements from space are based on the interpretation of the electromagnetic radiation that is scattered and emitted from clouds, precipitation and the underlying surface, and is monitored by the satellite instruments at the various wavebands. The interaction of the radiation with the cloud and precipitation particles strongly depends on their composition and size distribution, as described by the Mie theory. Therefore, variability in the cloud microstructure and precipitation properties for clouds having the same macroscopic properties and rain intensity would result in substantial changes in the satellite measured radiation that comes from the rain cloud and hence would cause large variability in the inferred rain intensity for a given actual rainfall. The variability in precipitation composition and size distribution affects mainly the direct rainfall measurements. Direct measurements use the portion of the electromagnetic spectrum that interacts strongly with the precipitation particles while weakly interacting with the small cloud particles. The passive microwave band is used for direct measurements of precipitation because precipitation particle size lies within this waveband and radiation interacts most strongly with particles that are of similar wavelength. The strength of the interaction for smaller particles decreases with the 6th power of the particle size (Rayleigh scattering). Using microwave takes us closest to the desirable situation of “visible” precipitation within “transparent” clouds. An overview of the use of the electromagnetic spectrum for cloud and precipitation analysis is given, among others, by Karlsson (1997), Kidder and Vonder Haar (1995), Levizzani et al. (2001), Petty (1995).
61 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 61–77. © 2007 Springer.
62
D. Rosenfeld
Indirect measurements are defined as such that infer the precipitation by observing other cloud properties. Potential rain clouds are sufficiently optically thick to be opaque in the visible and IR wavebands. Therefore, the radiation that reaches the satellite sensors typically comes from the cloud droplets and ice particles near the cloud tops, with little or no contribution from the actual precipitation particles. The indirect measurements relate the precipitation mainly to the cloud top temperatures and its type, i.e., convective or stratiform. However, precipitation in clouds of a given depth can vary greatly in clouds of different composition, i.e., cloud particle size distribution and phase – ice or water. These differences can determine not only the precipitation properties, but also the difference between precipitating or rainless clouds that have otherwise the same cloud top temperatures and dimensions (Rosenfeld 1999, 2000). Given this state of affairs, it is clear that recognizing cloud microstructure can potentially be used for improving the accuracy of rainfall measurements from space with direct measurement methods and even more so for the indirect methods. In this section we will review the ways cloud microstructure affect the precipitation, and the ways by which this can be utilized for enhancing the accuracy of precipitation measurements from space. The simplest indirect rainfall measurement is the GOES precipitation index (GPI) that estimates rainfall from the product of the mean fractional coverage of cloud colder than 235 K in a 2.5° × 2.5° box, the length of the averaging period in hours and a constant rainrate of 3 mm h –1 (Arkin and Meisner 1987). This method works very well in the tropical convective clouds, but fails in the extra-tropical rain cloud systems. Atlas and Bell (1992) suggested that the fundamental reason for the excellent correlation between the area of cloud tops <235 K and the surface precipitation on the large scale in the tropics is because the cold cloud tops are in fact the anvils of the cumulonimbus (Cb) clouds. In the deep tropics, with little dynamic complicating factors, the area time integral of the anvils area is proportional to the amount of air that is advected to the upper troposphere by the Cb clouds throughout their lifecycle. The requirement for inclusion of complete lifecycle of the convective systems explains why the GPI correlates so highly with the integral surface rainfall over large areas, but has little skill over small domains. The robust GPI relation between cold cloud top temperature area and rainfall amount on the surface were competitive with the direct rainfall measurement methods for convective clouds over the tropical oceans (Ebert et al. 1996). However, over land this was no longer the case. McCollum et al. (2000) showed that the both the GPI and passive microwave rainfall data overestimated the rainfall by a factor of 2 over Africa, while having no significant bias over South America. They suggested that this is due to systematic microphysical changes in the cloud properties between the two continents, such that the smaller drops that dominate the clouds over Africa affect the passive microwave signal and the time-area extent of the cold
Cloud Top Microphysics as a Tool for Precipitation Measurements
63
cloud tops in ways that appears as greater rainfall than the true amount. How that can happen will be the subject of the next sections.
2
RELATIONS BETWEEN CLOUD DROP SIZE AND PRECIPITATION FORMING PROCESSES
Clouds form by condensation of vapor on aerosols that serve as cloud drop condensation nuclei (CCN). The cloud droplets continue growing by condensation of vapor. Satellite measurements can detect re, the cloud drop effective radius (re = /, where r is the radius of the cloud droplets in the measurement volume) using methods that were pioneered by Arking and Childs (1985) and Nakajima and King (1990). The probability of collision and coalescence of drops with re < 12 µm is very small, to the extent that raindrops cannot form by this mechanism within the lifetime of clouds. This probability increases fast with droplet size, so that in clouds with re > ~14 µm coalescence of cloud droplets into raindrops leads to fast formation of rainfall (Rosenfeld and Gutman 1994; Gerber 1996). When cloud drops are too small for creating raindrops by coalescence, precipitation can still be formed by ice processes. This requires that the cloud will develop to heights above the zero isotherm level. The small cloud droplets can remain super-cooled (i.e., at a liquid state but colder than 0°C) typically up to –15°C to –25°C, and in extreme cases all the way to the homogeneous freezing isotherm of –38°C (Rosenfeld and Woodley 2000). Ice precipitation develops in such clouds when ice crystals form and grow at the expense of the cloud drops due to the smaller vapor pressure on ice than on water. When the crystals grow they get rimed by the super-cooled cloud droplets, and so become graupel particles of several mm in diameter. When graupel continues growing beyond about 1 cm they can become hailstones. Larger water drops freeze faster at higher temperatures. Therefore, clouds with large droplets that are fast to coalesce into raindrops also produce ice precipitation particles, typically graupel, at relatively high temperature of –5°C to –10°C. Therefore, the re of the droplets near cloud top can be used to infer also the presence of ice precipitation from clouds with cold tops, using the same threshold of re > 14 µm as for “warm rain” clouds. In mature or stratiform clouds with slow vertical air motions there is sufficient time for the cloud drops to freeze and become ice crystals that aggregate into snow flakes. The super-cooled water at the upper and hence colder portions of such clouds is typically completely converted into ice crystals and precipitation particles, so that the clouds are said to be glaciated. Ice crystals that form by heterogeneous freezing in a super-cooled cloud grow on expense of the cloud drops, so that in such glaciated cloud each ice crystal contains a water amount that was previously distributed in many drops. Therefore, the ice crystals are much larger than the cloud droplets from which they were formed, and hence possess much larger re than that of the
64
D. Rosenfeld
source water cloud. These ice crystals aggregate with time into snow flakes. When they fall through super-cooled clouds the snow flakes continue growing by accreting the cloud droplets. Distinction must be made here between this situation and ice clouds that form by homogeneous freezing of cloud droplets near or above the –38°C isotherm. Such homogeneously frozen drops retain their mass and become similarly small and numerous ice crystals, that have no efficient mechanism to aggregate into precipitation particles at such cold temperatures and small sizes. This situation can be detected by the existence of ice clouds with small re at temperatures <–38°C. Such clouds would be poor precipitators, because only a small fraction of the cloud water is converted into rainfall. On the other hand, clouds that form large ice particles at high temperatures indicate high precipitation efficiency. This principle can be used for assigning relative rainfall amounts for clouds having the same physical dimensions.
3
INFERRING PRECIPITATION FORMING PROCESSES FROM SATELLITE RETRIEVED T-RE RELATIONS
The sensors on board the recent satellites have a family of spectral bands in the solar and terrestrial portion of the radiation spectrum. For example, the geostationary Meteosat Second Generation (MSG) satellite has 12 spectral bands, from which cloud composition can be retrieved. These channels, 2.1 and 3.8 µm in particular, make it possible to measure parameters such as thermodynamic phase and re in addition to visible reflectance and the thermal emission temperature. Much more information about the cloud microstructure and precipitation forming processes in convective clouds can be obtained from analyses of complete cloud clusters, residing in areas containing thousands of satellite pixels. The underlying assumption is that the microphysical evolution of a convective cloud can be represented by composition of the instantaneous values of the tops of convective clouds at different heights. This is based on the knowledge that cloud droplets form mainly at the base of convective clouds, and grow with increasing height or decreasing T. The form of the dependence of re on T contains vital information about the cloud and precipitation processes, as described below. The T–re relations are obtained from an ensemble of clouds having tops covering a large range of T. Usually many pairs of T–re for each 1°C interval are observed in a region containing a convective cloud cluster. The points with smaller re for a given T are typically associated with the younger cloud elements, whereas the larger re for the same T are associated with the more mature cloud elements, in which the droplet growth had more time to progress by coalescence, and ice particles had more time to develop. Therefore, it is useful to plot not only the
Cloud Top Microphysics as a Tool for Precipitation Measurements
65
median value of the T–re relation, but also, say, the 15th and 85th percentiles, for representing the younger and more mature cloud elements within the measurement region. Based on the shapes of the T–re relations, Rosenfeld and Lensky (1998) defined the following five microphysical zones in convective clouds (see Fig. 1): Diffusional droplet growth zone: Very slow growth of cloud droplets with depth above cloud base, indicated by small –dre/dT. Droplet coalescence growth zone: Large increase of the droplet growth with height, as depicted by large –dre/dT at T warmer than freezing temperatures, indicating rapid cloud-droplet growth with depth above cloud base. Such rapid growth can occur there only by drop coalescence. Rainout zone: A zone where re remains stable between 20 and 25 µm, probably determined by the maximum drop size that can be sustained by rising air near cloud top, where the larger drops are precipitated to lower elevations and may eventually fall out as rain from the cloud base. This zone is so named, because droplet growth by coalescence is balanced by precipitation of the largest drops from cloud top. Therefore, the clouds seem to be raining out much of their water while growing. The radius of the drops that actually rain out from cloud tops is much larger than the indicated re of 20–25 µm, being at the upper end of the drop size distribution there. Mixed phase zone: A zone of large indicated droplet growth with height, occurring at T < 0°C, due to coalescence as well as to mixed phase precipitation formation processes. Therefore, the mixed phase and the coalescence zones are ambiguous at 0 < T < –38°C. The conditions for determining the mixed phase zone within this range are specified in Rosenfeld and Lensky (1998). Glaciated zone: A nearly stable zone of re having a value greater than that of the rainout zone or the mixed phase zone at T < 0°C. These zones are idealizations. Not all clouds conform to this idealized picture. The transition between the coalescence and mixed phase zones, which are not separated by a rainout zone, cannot be determined, and are therefore set arbitrarily to –6°C in accordance with aircraft observations. The height of the glaciation zone can be overestimated in the cases of highly maritime clouds that grow through a deep rainout zone, because the scarcity of water in the super-cooled portions of the clouds causes small ice particles, which sometimes can be mistaken for a mixed phase cloud. Addition of more spectral bands can help in separating the water from the ice, irrespective of the particle size. On the other hand, in vigorous clouds with active coalescence the height of the glaciation zone can be underestimated, because the high amounts of large ice hydrometeors dominate the radiative properties of the clouds, even when they coexist with cloud super-cooled water.
D. Rosenfeld
66
General -40
-30
-30 Glaciated Mixed Phase Rainout Coalescence
0 10
o
o
-10
Mixed Phase Rainout
10
20 5
-10 0
Diffusional growth 0
Glaciated
-20 T [ C]
-20 T [ C]
Maritime
-40
Coalescence
20
10 15 20 25 30 35 r [µ m] eff
0
-40
-30
-30 Glaciated
0
Coalescence
10
Diffusional growth
20
Glaciated Mixed Phase
-20 o
T [ C]
o
T [ C]
Mixed Phase
-10
10 15 20 25 30 35 r [µ m] eff
Continental - extreme
Continental - moderate -40
-20
5
-10 0
Diffusional growth
10 20
0
5
10 15 20 25 30 35 r [µ m] eff
0
5
10 15 20 25 30 35 r [µ m] eff
Figure 1. The classification scheme of convective clouds into microphysical zones, according to the shape of the T–re relations. Note that in extremely continental clouds re at cloud base is very small, the coalescence zone vanishes, mixed phase zone starts at T < –15oC, and the glaciation can occur in the most extreme situation at the height of homogeneous freezing temperature of –38oC. In contrast, maritime clouds start with large re at their base, crossing the precipitation threshold of 14 µm a short distance above the base. The deep rainout zone is indicative of fully developed warm rain processes in the maritime clouds. The large droplets freeze at relatively high temperatures, resulting in a shallow mixed-phase zone and a glaciation temperature reached near –10oC. (From Rosenfeld and Woodley 2003, courtesy of Amer. Meteor. Soc.)
All these microphysical zones are defined only for convective cloud elements. Multilayer clouds start with small re at the base of each cloud layer. This can be used for distinguishing stratified from convective clouds by their microstructure. Typically, a convective cloud has a larger re than a layer cloud at the same height, because the convective cloud is deeper and contains more water in the form of larger drops. Additional vertical information can be obtained by using channels that penetrate to different depth below cloud tops (Rosenfeld et al. 2004). Chang
Cloud Top Microphysics as a Tool for Precipitation Measurements
67
and Li (2003) pioneered this concept using 1.6, 2.1 and 3.7 µm channels of MODIS, retrieving the vertical profiles near the tops of stratiform clouds. In addition to the microphysical zones, it also can be determined that convective clouds start precipitating at re > 14 µm (Rosenfeld and Gutman 1994; Gerber 1996). This can be used quantitatively for improving the accuracy of rainfall measurements from space, as demonstrated by Lensky and Rosenfeld (1997) for the NOAA/AVHRR. This principle was applied to an operational rainfall product (Ba and Gruber 2001).
4
THE DEPENDENCE OF RAINFALL REMOTE SENSING ON HYDROMETEOR SIZE DISTRIBUTIONS
The previous sections provided a physical basis for indirect measurements of precipitation based on the retrieved cloud top composition and temperature, using the visible and IR wavebands. Direct measurements of precipitation use the microwave frequencies that interact directly with the precipitation size particles (diameter >0.1 mm). Direct measurements are divided into active and passive. Active microwave measurements require a radar instrument that transmits pulses of radiation and receives the back scattered echoes. The echo intensity is converted into precipitation intensity according to the radar equation. The backscatter occurs mainly in the Rayleigh regime, where the intensity of the scattered radiation is proportional to the 6th power of the particle size. This highly nonlinear relation causes a serious problem of nonuniqueness between the echo and precipitation intensities. Small concentrations of large hydrometeors can produce the same reflectivity factor (Z, [mm6 m–3]) as much larger concentrations of smaller hydrometeors that form much greater equivalent rain intensity (R, [mm h–1]). This nonuniqueness in the Z–R relations has been historically the weakest point in radar rainfall measurements from both surface and space-borne platforms. Rainfall measurements from space with passive microwave rely on both thermal emission and back scatter. The thermal emission does not depend so strongly on the particle size, but because of that it can’t distinguish between cloud water and precipitation. The thermal emission can be used mainly above the cold background of the oceans, which appear cold due to the small microwave emissivity of flat water surfaces. Most of the signal from deep convection comes from the backscatter of the upwelling thermal radiation back downward. This signal strongly depends on the particle size, as in the case for the radar. The larger the hydrometeors the more energy is backscattered to the surface and the lower the satellite measured brightness temperature becomes. It can be also viewed as larger particles backscatter more strongly the 3 K background of the outer space, but it is inaccurate in a
D. Rosenfeld
68
strict physical sense. Here again the same rain intensity that is associated with larger hydrometeors would be interpreted as a stronger passive microwave signal and heavier rain. Ice hydrometeors are colder and have smaller emissivity than raindrops, and hence they would create lower brightness temperature for the same precipitation intensity, and more so when they reside higher in the cloud and at lower temperatures. Therefore, the correct interpretation into rainfall of both active and passive microwave measurements depends strongly on the relations between the hydrometeor size distributions, types and the rain intensity. It is essential to obtain information about the hydrometeor sizes for achieving reasonable accuracy of the precipitation intensities. This can be achieved by space borne radar measurements with multiple wavelength radar, as planned for the Global Precipitation Measurement (GPM; Smith et al. 2007) mission. The already available space borne radar onboard the Tropical Rainfall Measuring Mission (TRMM) satellite has only a single wavelength, and hence requires external independent information about the Z–R relations. Passive microwave measurements are conducted in several wavelengths simultaneously, but have very limited capability to resolve the particle size, especially in deep convective clouds.
5
CLOUD MICROSTRUCTURE AND Z–R RELATIONSHIPS
In the previous section we have seen that independent external information on hydrometeors type and sizes is essential for improving the accuracy of both radar and passive microwave precipitation measurements. Such information can be obtained from the inferred cloud microstructure and precipitation forming processes as obtained from the T–re relations. We will review here the precipitation evolution in microphysically maritime and continental clouds. Microphysically maritime clouds are composed of low concentration of large cloud drops that coalesce readily into warm rain. In contrast, microphysically continental clouds are composed of small drops that form precipitation mainly by ice processes.
5.1 Microphysically maritime clouds: Evolution of warm rain In a hypothetical rising cloud column with active coalescence, the initial dominant process would be widening of the cloud drop size distribution into large concentrations of drizzle drops; the drizzle continues to coalesce with other drizzle and cloud drops into raindrops, which will continue to grow asymptotically to the equilibrium raindrop-size distribution (RDSD), with the median volume drop diameter D0e = 1.76 mm (Hu and Srivastava 1995). Therefore, during the growth phase of the precipitation particles the rain rate R increases with D0, median volume drop diameter, and this would increase
Cloud Top Microphysics as a Tool for Precipitation Measurements
69
D0 for a given R. Ideally, for rainfall with drops that fall from cloud top while growing, R would increase with the fall distance from the cloud top, mainly by growth of the falling drops due to accretion and coalescence, and to a lesser extent by addition of new small raindrops, until the raindrops grow sufficiently large for breakup to become significant. Shallow orographic clouds can present conditions such as some distance below the tops of convective clouds. Therefore, similar evolution of R can be observed on a mountain slope, such as documented by Fujiwara (1965). Different values of R near cloud top or in shallow orographic clouds can come mainly from changing NT, the total concentration of raindrops, because the drop size is bounded by the limited vertical fall distance along which they can grow. This would cause orographic precipitation to have small drops and for R to depend mainly on NT, and more so with shallower clouds and stronger orographic ascent, because the stronger rising component supplies more water for the production of many small raindrops not too far below cloud top, which are manifested as a larger R.
5.2 Microphysically continental clouds: Evolution of cold rain Microphysically “continental” clouds are characterized by narrow cloud drop size distributions, and therefore by having little drop coalescence and warm rain. Most raindrops originate from melting of ice hydrometeors that are typically graupel or hail in the convective elements, and snowflakes in the mature or stratiform clouds. Graupel and hail particles grow without breakup while falling through the super-cooled portion of the cloud, and continue to grow by accretion in the warm part of the cloud, where they melt. Large melting hailstones shed the excess melt-water in the form of a RDSD about which little is known. The shedding stops when the melting particles approach the size of the largest stable raindrops, which are later subject to further breakup due to collisions with other raindrops. In fact, new raindrop formation is limited only to the breakup of preexisting larger precipitation particles. Therefore, we should expect that in such clouds there would be, for a given R, a relative dearth of small drops and excess of large drops compared to microphysically “maritime” clouds with active cloud drop coalescence. Deep continental convective clouds would therefore initiate the precipitation by forming large drops that with maturing approach DSDe from above. This is in contrast with the approach from below for maturing maritime RDSD. Recent satellite studies (Rosenfeld and Lensky 1998) have shown that microphysically maritime clouds are associated typically with a “rainout” zone, i.e., the fast conversion of cloud water to precipitation cause the convective elements to lose water to precipitation while growing. This leaves less water carried upward to the super-cooled zone, so that weaker ice precipitation can develop aloft. Williams et al. (2002) have recognized this
D. Rosenfeld
70
as a potential cause to the much greater occurrence of lightning in continental compared to maritime clouds. Williams et al. (2002) noted that frequent lightning occurred also in very clean air during high atmospheric instability, probably because the strong updraft leaves little time to the formation of warm rain, and carries the large raindrops that manage to form up to the super-cooled levels of the clouds, where they freeze and participate in the cloud electrification processes (Atlas and Williams 2003). This difference between continental and maritime clouds means that mostly warm rain would fall even from the very deep maritime convection, which reaches well above the freezing level, whereas precipitation from continental clouds would originate mainly in ice processes. Therefore, the expected difference in RDSD between microphysically maritime and continental clouds is expected to exist also for the deepest convective clouds that extend well into the sub-freezing temperatures.
1000 Florida Cont Florida Mar LBA Cont LBA M ar India Cont India Mar Kw aj M ar
-3
N [m m m ]
100
10
1
0.1
0.01 0
1
2
3
4
5
6
D [m m ] Figure 2. Disdrometer measured RDSDs of continental (solid lines) and maritime (dashed lines) rainfall, as microphysically classified by VIRS overpass. The RDSD is averaged for the rainfall during ± 18 h of the overpass time, and the concentrations are scaled to 1 mm h-1. The disdrometers are in Florida (Teflun B), Amazon (LBA), India (Madras) and Kwajalein. (From Rosenfeld and Ulbrich 2003, courtesy of Amer. Meteor. Soc.)
Cloud Top Microphysics as a Tool for Precipitation Measurements
71
5.3 Quantifying the role of cloud top re on RDSD The ultimate test for the role of cloud microstructure is comparing the RDSD of clouds at the same location, but at different times, when they possess maritime or continental microstructure. Rosenfeld and Ulbrich (2003) did exactly that. They used the VIRS (Visible and Infrared sensor) onboard TRMM (Tropical Rainfall Measuring Mission) satellite to retrieve the microstructure of rain clouds over disdrometer sites. The clouds were classified into continental, intermediate and maritime, using the methodology of Rosenfeld and Lensky (1998). The RDSDs from the continental and maritime classes during the overpass time + 18 h were lumped together and plotted in Fig. 2. Indeed, the continental and maritime RDSDs are well separated in Fig. 2, with the continental clouds producing greater concentrations of large drops and smaller concentrations of small drops. A comparison between the directly measured disdrometer rainfall and the calculated accumulation by applying the TRMM Z-R relations (Iguchi et al. 2000) to the disdrometer measured Z resulted in a relative overestimate by more than a factor of two of the rainfall from the microphysically continental clouds compared to the maritime clouds. The evidence shows that it is mainly the cloud microstructure that is responsible for the large systematic difference in the RDSD and Z–R relations between maritime and continental clouds. There are several possible causes for these differences, all working in the same direction:
5.4 Extent of coalescence The cloud drop coalescence in highly maritime clouds is so fast that rainfall is developed low in the growing convective elements and precipitates while the clouds are still growing. The large concentrations of raindrops that form low in the cloud typically fall before they have the time to grow and reach equilibrium RDSD, thereby creating the rainout zone (Rosenfeld and Lensky 1998) less than 2 km above cloud base height. Therefore, D0 remains much smaller than D0e, as was shown in Fig. 6b of Rosenfeld and Ulbrich (2003). In microphysically continental clouds with suppressed coalescence the cloud has to grow into large depth before start precipitating, by either warm or cold processes. The raindrops that fall through the lower part of the cloud grow by accretion of small cloud drops, so that they tend to breakup much less than drops that grow mainly by collisions with other raindrops, as is the case for maritime clouds. This process allows D0 to exceed D0e in the growing stages of the precipitation, and later approach it from above when the raindrop collisions become more frequent with the intensification of the rainfall.
D. Rosenfeld
72
5.5 Warm versus cold precipitation processes The rainout of the maritime clouds (Rosenfeld and Lensky 1988) depletes the cloud water before reaching the super-cooled levels (Zipser and LeMone 1980; Black and Hallett 1986), so that mixed phase precipitation would be much less developed in the maritime clouds compared to the continental. This is manifested in the smaller reflectivity aloft in the maritime clouds (Zipser and Lutz 1994), which is a manifestation of the smaller hydrometeors that form there (Zipser 1994). In contrast, the suppressed coalescence in continental clouds leaves most of the cloud water available for growth of ice hydrometeors aloft, typically in the form of graupel and hail. These ice hydrometeors can grow indefinitely without breakup, until they fall into the warm part of the cloud and melt. The melted hydrometeors continue to grow by accretion of cloud droplets, until they exceed the size of spontaneous breakup or collide with other raindrops. Therefore, convective rainfall that originates as ice hydrometeors would have D0 > D0e, and would approach D0e from above with maturing of the RDSD.
5.6 Strength of the updrafts Updrafts are typically stronger in more continental clouds, and therefore contribute to more microphysically continental clouds and less warm rain processes, as discussed already above. In addition, stronger updrafts allow drops with greater minimal size to fall through them. In addition, stronger updrafts leave less time for forming of warm rain and rainout, and advect more cloud water to the super-cooled zone. Therefore, due to the reasons already discussed in (a) and (b), the stronger updrafts are likely to lead to precipitation with greater D0 and smaller R for the same Z.
5.7 Evaporation More continental environments have typically higher cloud base and lower relative humidity at the sub-cloud layer. Evaporation depletes preferentially the smaller raindrops and works to increase D0.
6
RELATIONS BETWEEN PRECIPITATION MEASUREMENT BIASES AND CLOUD MICROSTRUCTURE
Now we can return and try explaining the large discrepancies between satellite measurements and rain-gauge estimates that were found over central Africa, while in the Amazon regions the rain gauges coincide closely with satellite estimates (McCollum et al. 2000). This is consistent with the in situ microphysical observations showing that clouds in the Amazon are microphysically maritime, similar to equatorial pacific clouds (Stith et al. 2002), except for periods when they are polluted by smoke from forest fires (Andreae
Cloud Top Microphysics as a Tool for Precipitation Measurements
73
et al. 2004). McCollum et al. (2002) have also shown that remote sensing of rainfall measurements by both passive microwave (SSM/I) and surface radar measurements have relative overestimates when moving from the east coast of the USA to central USA by 25–30% (see Fig. 3). McCollum et al. (2002) suggested that this bias is caused by the greater continentality of the rain clouds in central USA. When they used multispectral algorithm that takes into account cloud top microstructure the systematic bias somewhat decreased.
Figure 3. Spatial distribution of area-averaged multiplicative bias for the SSM/I with respect to the estimates of the bias-adjusted hourly digital precipitation radar rainfall estimates on a national grid. (From McCollum et al. 2002, courtesy of Amer. Meteor. Soc.)
Additional indication of the precipitation forming processes is the lightning activity. Clouds are electrified when graupel collides with ice crystals in a super-cooled water cloud. Therefore lightning is a manifestation of intense ice precipitation forming processes. Tropical maritime clouds have between one and two orders of magnitude less lightning for the same amount of rainfall of continental clouds (Petersen and Rutledge 1998). Satellite rainfall estimates in the Amazon regions and central Africa are comparable in magnitude, while there is much more lightning activity over central Africa with much less rain gauge measured rainfall. We postulate that rainfall regime over the Amazon is less microphysically continental than that over central Africa, and hence having smaller hydrometeors and larger extent of cold anvils for the same rainfall amounts. This suggestion is further supported by findings of Petersen and Rutledge (2001). Greater continentality is characterized by larger amounts of cloud water carried up to the upper portions of the cloud, where it freezes and forms large ice hydrometeors, and the released latent heat of freezing invigorates the updrafts and loft the large ice particles to great heights (Andreae et al. 2004). The large ice particles aloft produce smaller passive microwave brightness temperatures that are interpreted as greater rain intensities, by a
74
D. Rosenfeld
factor of 2–3 compared to the maritime clouds. The large raindrops that form when these ice particles melt equally cause radar overestimates of rain intensities by a similar factor of 2–3 compared to the maritime clouds. As shown earlier in this section and elsewhere (Rosenfeld and Lensky 1998; Andreae et al. 2004), the continentality of the clouds can be quantified independently of the radar by satellite retrieved T–re relationships.
Figure 4. MSG image from 20 May 2003 1342 UTC, over central Africa at a 1200 × 1200 km2 rectangle between 1–12N and 15–26E. The area shows the transition between the relatively microphysically maritime clouds over the forested area (dark surface) and microphysically continental clouds over the dry lands of the Sahel to the north (bright surface). The T–re relations of the continental clouds (1) show much smaller re for a given T compared to the maritime clouds (2). The median re of the maritime clouds (the yellow line) saturates near T = –20°C, indicating glaciation at that temperature. The small median re at area 1 even above the –40°C isotherm indicates homogeneous glaciation of the cloud water and hence low precipitation efficiency. The color scheme is red for the visible, green for 3.9 µm reflectance component, and blue for temperature. For full description and interpretation of the color table is given in Rosenfeld and Lensky (1998). The T–re lines represent percentiles of re for a given T in 10% steps for each line, between 5–95%. The median is between the yellow and green lines. (see also color plate 3)
Cloud Top Microphysics as a Tool for Precipitation Measurements
75
Figure 5. The measurement bias of the TRMM precipitation radar (PR, middle panel) and TRMM passive microwave (TMI, lower panel) in relation to the continentality of the rain clouds, as given by the mean 30 dBZ echo top height in precipitation features with TMI signal of ice scattering. Note the large overestimate where large ice hydrometeors exist high in the clouds. (Presented by S. Nesbitt at the TRMM Hawaii Scientific Conference, Honolulu, HI, 22–26 July 2002.)
7
CONCLUSIONS
Differences in clouds microstructure can explain systematic biases of up to a factor of 3 in passive MW and radar direct rainfall measurements. The cloud microstructure can be obtained by T–re relations that are obtained from the operational NOAA orbital satellites. The MSG, which was commissioned in early 2004, is the first of a new generation of geostationary satellites that have sufficient resolution for providing useful T–re relations of convective clouds (e.g., Fig. 4). Combining the cloud’s microphysical continentality from T–re analyses such as shown in Fig. 4 with the radar and passive microwave measurements has the potential of eliminating much of the measurement biases shown in Fig. 5. Night time capabilities for microphysical measurements are also emerging (Lensky and Rosenfeld 2003a). Indirect rainfall measurements can be also substantially improved using the information about cloud top composition. Only the first steps have been done so far in this direction during daylight (Lensky and Rosenfeld 1997; Ba and Gruber 2001) and night (Lensky and Rosenfeld 2003b). The implication for future missions is that rainfall measuring satellite should include both microwave and VIS/IR sensors, and the rain estimation should use this added information, without which systematic bias errors greater than a factor of 2 are difficult to avoid.
D. Rosenfeld
76
8
REFERENCES
Andreae, M. O., D. Rosenfeld, P. Artaxo, A. A. Costa, G. P. Frank, K. M. Longo, and M. A. F. Silva-Dias, 2004: Smoking rain clouds over the Amazon. Science, 303, 1337–1342. Arking, A. and J. D. Childs, 1985: Retrieval of cloud cover parameters from multispectral satellite images. J. Climate Appl. Meteor., 24, 322–333. Arkin, P. A. and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the western hemisphere during 1982–84. Mon. Wea. Rev., 115, 51–74. Atlas, D. and T. Bell, 1992: The relation of radar to cloud area-time integrals and implications for rain measurements from space. Mon. Wea. Rev., 120, 1997–2008. Atlas, D. and C. R. Williams, 2003: An anatomy of a continental tropical convective storm. J. Atmos. Sci., 60, 3–15. Ba, M. and A. Gruber, 2001: GOES Multispectral Rainfall Algorithm (GMSRA). J. Appl. Meteor., 40, 1500–1514. Black, R. A. and J. Hallett, 1986: Observations of the distribution of ice in hurricanes. J. Atmos. Sci., 43, 802–822. Chang, F. L. and Z. Li, 2003: Retrieving vertical profiles of water-cloud droplet effective radius: Algorithm modification and preliminary application. J. Geophys. Res., 108 (D24), 4763, doi:10.1029/2003JD003906. Fujiwara, M., 1965: Raindrop-size distribution from individual storms. J. Atmos. Sci., 22, 585–591. Gerber, H., 1996: Microphysics of marine stratocumulus clouds with two drizzle modes. J. Atmos. Sci., 53, 1649–1662. Hu, Z. and R. Srivastava, 1995: Evolution of the raindrop-size distribution by coalescence, breakup and evaporation: Theory and observations. J. Atmos. Sci., 52, 1761–1783. Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain-profiling algorithm for the TRMM Precipitation Radar. J. Appl. Meteor.,: 39, 2038–2052. Ebert, E. E., M. J. Manton, P. A. Arkin, R. J. Allam, G. E. Holpin, and A. Gruber, 1996: Results from the GPCP algorithm intercomparison program. Bull. Amer. Meteor. Soc., 77, 2875–2887. Karlsson, K.-G., 1997: An introduction to remote sensing in meteorology. SMHI, Norrköping, 315 pp. Kidder, S. Q. and T. H. Vonder Haar, 1995: Satellite meteorology. An Introduction. Academic Press, San Diego, 466 pp. Lensky, I. M. and D. Rosenfeld, 1997: Estimation of precipitation area and rain intensity based on the microphisical properties retrieved from NOAA AVHRR data. J. Appl. Meteor., 36, 234–242. Lensky, I. M. and D. Rosenfeld, 2003: Satellite-based insights into precipitation formation processes in continental and maritime convective clouds at night-time. J. Appl. Meteor., 42, 1227–1233. Lensky, I. M. and D. Rosenfeld, 2003: A night rain delineation algorithm for infrared satellite data. J. Appl. Meteor., 42, 1218–1226. Levizzani, V., J. Schmetz, H. J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2001: Precipitation estimations from geostationary orbit and prospects for METEOSAT Second Generation. Meteorol. Appl., 8, 23–41. McCollum, J. A., A. Gruber, and M. Ba, 2000: Comparison of monthly mean satellite estimates of precipitation with gauges over Africa. J. Appl. Meteor., 39, 666–679. McCollum, J. A., W. F. Krajewski, R. R. Ferraro, and M. B. Ba, 2002: Evaluation of biases of satellite rainfall estimation algorithms over the continental United States. J. Appl. Meteor., 41, 1065–1080. Nakajima, T. and M. D. King, 1990: Determination of the optical thickness and effective particle radius of clouds from reflected solar radiation measurements. Part I: Theory. J. Atmos. Sci., 47, 1878–1893.
Cloud Top Microphysics as a Tool for Precipitation Measurements
77
Petersen, W. A. and S. A. Rutledge, 1998: On the relationship between cloud-to-ground lightning and convective rainfall. J. Geophys. Res., 103, 14025–14040. Petersen, W. A. and S. A. Rutledge, 2001: Regional variability in tropical convection: observations from TRMM. J. Climate, 14, 3566–3586. Petty, G. W., 1995: The status of satellite-based rainfall estimation over land. Remote Sens. Environ., 51, 125–137. Rosenfeld, D., 1999: TRMM observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 3105–3108. Rosenfeld, D., 2000: Suppression of rain and snow by urban and industrial air pollution. Science, 287, 1793–1796. Rosenfeld, D. and A. Gagin, 1989: Factors governing the total rainfall yield of continental convective clouds. J. Appl. Meteor., 28, 1015–1030. Rosenfeld, D. and G. Gutman, 1994: Retrieving microphysical properties near the tops of potential rain clouds by multispectral analysis of AVHRR data. Atmos. Res., 34, 259–283. Rosenfeld, D. and I. M. Lensky, 1998: Space borne based insights into precipitation formation processes in continental and maritime convective clouds. Bull. Amer. Meteor. Soc., 74, 2457–2476. Rosenfeld, D. and C. W. Ulbrich, 2003: Cloud microphysical properties, processes, and rainfall estimation opportunities. Chapter 10 of “Radar and Atmospheric Science: A Collection of Essays in Honor of David Atlas”. Edited by Roger M. Wakimoto and Ramesh Srivastava. Meteorological Monographs, 52, 237–258, AMS. Rosenfeld, D. and W. L. Woodley, 2003: Closing the 50-year circle: From cloud seeding to space and back to climate change through precipitation physics. Chapter 6 of “Cloud Systems, Hurricanes, and the Tropical Rainfall Measuring Mission (TRMM)” edited by Drs. Wei-Kuo Tao and Robert Adler, 234pp., pp. 59–80, Meteorological Monographs, 51, AMS. Rosenfeld, D., E. Cattani, S. Melani, and V. Levizzani, 2004. Considerations on daylight operation of 1.6 µm vs 3.7 µm channel on NOAA and METOP Satellites. Bull. Amer. Meteor. Soc., 85, 873–880. Smith, E. A., G. Asrar, Y. Furuhama, A. Ginati, C. Kummerow, V. Levizzani, A. Mugnai, K. Nakamura, R. F. Adler, V. Casse, M. Cleave, M. Debois, J. Durning, J. Entin, P. Houser, T. Iguchi, R. Kakar, J. Kaye, M. Kojima, D. Lettenmaier, M. Luther, A. Mehta, P. Morel, T. Nakazawa, S. Neeck, K. Okamoto, R. Oki, G. Raju, M. Shepherd, E. Stocker, J. Testud, and E. Wood, 2007: International Global Precipitation Measurement (GPM) program and mission: an overview. In: Measuring precipitation from space. EURAINSAT and the future. V. Levizzani, P. Bauer and J. F. Turk, eds., Springer, 611–654. Stith, J. L., J. E. Dye, A. Bansemer, and A. J. Heymsfield, 2002: Microphysical observations of tropical clouds. J. Appl. Meteor., 41, 97–117. Williams, E., D. Rosenfeld, N. Madden, J. Gerlach, N. Gears, L. Atkinson, N. Dunnemann, G. Frostrom, M. Antonio, B. Biazon, R. Camargo, H. Franca, A. Gomes, M. Lima, R. Machado, S. Manhaes, L. Nachtigall, H. Piva, W. Quintiliano, L. Machado, P. Artaxo, G. Roberts, N. Renno, R. Blakeslee, J. Bailey, D. Boccippio, A. Betts, D. Wolff, B. Roy, J. Halverson, T. Rickenbach, J. Fuentes, and E. Avelino, 2002: Contrasting convective regimes over the Amazon: Implications for cloud electrification. J. Geophys. Res., 107, D20, 8082, doi:10.1029/2001JD000380. Zipser, E. J., 1994: Deep cumulonimbus cloud systems in the tropics with and without lightning. Mon. Wea. Rev., 122, 1837–1851. Zipser, E. J. and M. A. LeMone, 1980: Cumulonimbus vertical velocity events in GATE. Part II: Synthesis and model core structure. J. Atmos. Sci., 37, 2458–2469. Zipser, E. J. and K. Lutz, 1994: The vertical profile of radar reflectivity of convective cells: a strong indicator of storm intensity and lightning probability. Mon Wea. Rev., 122, 1751–1759.
7 THE RETRIEVAL OF CLOUD TOP PROPERTIES USING VIS-IR CHANNELS Elsa Cattani1, Samantha Melani2, Vincenzo Levizzani1, and Maria João Costa3 1
Institute of Atmospheric Sciences and Climate, ISAC-CNR, Bologna, Italy Institute of BioMeteorology, IBIMET-CNR, La.M.M.A. (Laboratory for Meteorology and Environmental Modelling), Florence, Italy 3 Department of Physics and Evora Geophysics Centre, University of Evora, Evora, Portugal 2
1
INTRODUCTION
The remote sensing of the cloud optical and microphysical properties from solar reflection and emission measurements is exceedingly important for an improved understanding of the Earth’s climate system, since clouds are a strong modulator of the shortwave and longwave components of the Earth’s radiation budget (King et al. 1992). Cloud microphysical characterization using multispectral techniques in the visible (VIS), near-infrared (NIR), and infrared (IR) range is an efficient proxy for the detection of precipitating systems. Recent studies have demonstrated that cloud properties can be effectively exploited to identify the genesis and evolution of the cloud mass. Rosenfeld and Lensky (1998) examined the evolution of the effective radius (Re) of convective cloud particles versus cloud top temperature (Tc) to infer information about the efficiency of the precipitation forming processes. They associated five different microphysical stages in the temporal evolution of a cloud with peculiar trends of Tc as a function of Re: diffusional droplet growth, coalescent droplet growth, rainout zone, mixed-phase precipitation, and glaciation. This kind of analysis provides a potential tool to improve satellite rainfall estimation techniques (Rosenfeld and Gutman 1994; Rosenfeld 2007) based on microwave data, especially in case of rain from clouds without large ice particles, whose detection is more difficult over land due to the high surface emissivity. 79 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 79–95. © 2007 Springer.
E. Cattani et al.
80
Sensors on board polar and geostationary platforms have recently entered a new era. The new generation of satellite sensors is characterized by a greater number of spectral channels in the VIS, NIR and IR, a better spatial resolution, and an increased data availability. Noteworthy among these satellites is the MODerate resolution Imaging Spectroradiometer (MODIS) on board the Earth Observing System (EOS) satellites (Barnes et al. 1998) equipped with 36 wavebands whose spatial resolution ranges from 250 m to 1 km. On the geostationary side it is worth mentioning the Spinnig Enhanced Visible and InfraRed Imager (SEVIRI) (Schmetz et al. 2002), the main instrument on board the new European geostationary platform Meteosat Second Generation (MSG) with 12 spectral channels and a spatial resolution of 4.8 km at the sub-satellite point, due to an oversampling factor of 1.6 and a 3 km sampling distance, for all channels except the High Resolution Visible (HRV) channel (1.67 km spatial resolution, with a sampling distance of 1 km at the nadir and an oversampling factor of 1.67). These improved observational capabilities are ideal to enhance our understanding of cloud microphysics through cloud top retrievals of hydrometeor radiative properties. It is unfeasible to discriminate among the wealth of cloud physical properties using data out of a single narrow portion of the electromagnetic spectrum. Moreover, an appropriate spatial resolution is needed to limit the occurrence of partially cloud covered pixels. Finally, the SEVIRI 15 min repeat cycle opens unprecedented scenarios to analyze the temporal evolution of cloud systems. In this work radiative transfer simulations are documented for reviewing the physical concepts behind the retrieval of cloud parameters from satellite sensor data. Radiance data in the VIS, NIR, and IR channels were simulated in the presence of water and ice clouds to estimate the sensitivity of the spectral radiances to the effective radius, cloud optical thickness, cloud top temperature/height, and thermodynamic phase. Uncertainties associated with satellite measurements were considered, and the influence of the solar and viewing geometry, the surface radiative properties, and the atmospheric water vapor amount on the cloud parameter retrieval were investigated.
2
THE RADIATIVE TRANSFER MODEL AND ITS SET UP
Simulated radiances were produced via the Signal Simulator for Cloud Retrieval (SSCR) radiative transfer model (RTM) (Nakajima and Tanaka 1986, 1988; Stamnes et al. 1988). SSCR is a 1D, plane parallel model conceived to compute radiances, transmittances, and plane and spherical albedoes in presence of water and ice clouds at VIS, NIR and IR wavelengths. The RTM can simulate radiance data as measured by a satellite sensor using the response functions of the various channels. The radiative transfer calculations are based on a combined discrete-ordinate/matrix-operator method, with the
The Retrieval of Cloud Top Properties Using VIS-IR Channels
81
delta-M approach for the representation of the phase function and the corrections for the single scattering radiation. The atmosphere is vertically divided in several homogenous sublayers and limited at the ground either by a Lambertian surface, characterized by a user-specified nonspectral albedo, or by an ocean surface, defined by a built-in albedo model that takes into account wind speed. The atmospheric models, which include vertical profiles of temperature, pressure and 28 gaseous species, are deduced from the LOWTRAN7 radiative transfer model (Kneizys et al. 1988). Only the water vapor profile can be modified by the user. The gas absorption is computed by means of a three term k-distribution. SSCR represents clouds as vertically homogeneous layers defined by the following input parameters: thermodynamic phase, Re, size distribution type (choice between three different types of functions, i.e., power law, log-normal, and modified gamma), optical thickness at 0.5 µm (τ), and top and bottom heights. The vertical superimposition of different cloud layers is not allowed and Mie theory is applied to cloud particle scattering considering only spherical ice particles. SSCR was used to analyze the sensitivity of the radiance data measured at various satellite sensor channels to the cloud optical and microphysical properties and to explore the possibility of using these spectral data for cloud property retrieval. The simulated data refer to satellite sensor channels centered at 0.6, 1.6, 3.7, 11 and 12 µm. The response functions of MODIS have been used to carry out the computations, but analogous results can be obtained using the response functions of other sensors characterized by similar spectral characteristics as SEVIRI’s. The simulations were performed with the following model setup: • •
• • • •
mid-latitude summer atmospheric profile; variations in the illumination and viewing conditions were simulated using 3 different solar and satellite zenith angle values (θ0 and θ, respectively), 5°, 30° and 70°, and one relative azimuth angle value (Φ), 80°; the surface contribution to the total signal was accounted for different surface types, i.e., sea, vegetation, snow, and desert soil; the cloud optical thickness at 0.5 µm ranged from 1 to 200, to evaluate the radiative behavior of a wide set of cloudy scenarios; a log-normal size distribution was assumed for cloud particles, with Re in the range [1, 40] µm for water clouds and up to 200 µm for ice clouds; the cloud bottom and top heights were fixed at 0.5 and 1.5 km, respectively, for water clouds, and 8.5 and 10 km for ice clouds.
Finally, the results of the sensitivity study were analyzed taking into account the radiometric performances of the MODIS and SEVIRI sensors.
E. Cattani et al.
82
3
SENSITIVITY ANALYSIS
The last few years have witnessed the development of numerous algorithms that exploit multispectral satellite measurements for a global monitoring of cloud microphysical and optical properties, such as particle size and optical thickness, and cloud macrophysical properties, i.e., the cloud height and thermodynamic phase. Most of these methods rely on the synergetic use of various spectral channels and on the combination of the information content of the different channels. Substantial efforts were devoted to retrieve the optical thickness and the effective particle size because they are crucial elements for the accurate determination of cloud radiative bulk properties (single-scattering albedo, phase function, asymmetry parameter, etc.) (King et al. 1992; Yang and Baum 2003), that very often are parameterized in the RTMs as functions of R e (e.g., Key and Schweiger 1998). Among the most widespread techniques for the retrieval of τ and Re, the VIS/NIR bispectral technique dwells on the different dependencies of the reflected solar radiation in the VIS (0.6, 0.8 µm) and NIR (1.6, 2.1 and 3.7 µm) channels on cloud optical thickness and particle size. In the VIS channels the scattering of incident radiation by cloud particles is conservative and thus the single-scattering albedo is about 1 and does not depend on cloud particle size. At NIR wavelengths the single-scattering albedo is less than unity and is affected by cloud particle size. Nakajima and King (1990) and Nakajima and Nakajima (1995) applied the technique to data from the Advanced Very High Resolution Radiometer (AVHRR). Other examples can be found in Arking and Childs (1985) and Han et al. (1994). NIR channels (in particular those centered at 1.6 and 3.7 µm) are often used in conjunction with the thermal IR (11 and 12 µm) channels for the thermodynamic phase detection. An accurate determination of the cloud particle phase is a fundamental prerequisite for the retrieval of Re and τ. Cloud phase is considered an a priori information needed by many Re and τ retrieval methods since the optical properties of liquid and solid particles are distinct. A wrong cloud phase attribution may thus result in a wrong Re and τ retrieval. Several methods for cloud phase determination were developed such as that of King et al. (1992) who use the ratio between the 1.6 and 0.6 µm reflectances. Clouds of different phase but with similar particle size and optical thickness are characterized by similar VIS reflectances and distinct NIR reflectances, due different absorption effectiveness of water with respect to ice (see Fig. 1). For this reason water clouds are expected to exhibit larger values of the reflectance ratio than ice clouds. The brightness temperature differences (BTD) between 11 and 12 µm have been analyzed together with the BTDs between 3.7 and 11 µm and 3.7 µm reflectance, as shown by Key and Intrieri (2000) who proposed an algorithm for the day and
The Retrieval of Cloud Top Properties Using VIS-IR Channels
83
Figure 1. Imaginary part of the refractive index for water and ice with superimposed relevant VIS/NIR/IR channels.
nighttime determination of cloud phase. Much earlier Inoue (1987) used the split window technique for cloud type classification. The analysis of twodimensional histograms of BTDs as a function of the brightness temperature at 11 µm with a selection of appropriate thresholds allowed to identify various cloud types, as cirrus, dense cirrus, cumulonimbus and cumulus. Cloud top height or cloud top temperature are macrophysical cloud characteristics that are very important for the cloud phase detection and cloud type classification. The simplest method to infer them is to use IR radiance data or the equivalent brightness temperature (at 11 µm), eventually corrected for thermal radiation emitted by the surface, and then to derive the cloud top as the height where the brightness temperature matches the temperature profile (Nakajima and Nakajima 1995). A more complex method is the CO2 slicing, that exploits radiance data around the CO2 absorption band at 15 µm. This method proved to be especially effective for detecting thin cirrus clouds that are often missed by simple IR and VIS approaches (Zhang and Menzel 2002).
3.1 VIS channel In this section the results of the sensitivity analysis of 0.6 µm simulated reflectances are presented. The dependence of the reflectances on cloud optical thickness and the influence of surface reflection on the relation between the reflectances and τ will be discussed.
E. Cattani et al.
84
(a)
(b)
Figure 2. VIS reflectances as a function of cloud optical thickness at 0.5 µm for water (a) and ice (b) clouds. Reflectances were computed for solar (θ0) and satellite (θ) zenith angles of 30°, a relative azimuth angle (Φ) of 80°, and a sea surface characterized by albedo ωs = 0.05. The different curves refer to various Re values, varying between 3 and 39 µm (water) and 3 and 200 µm (ice).
In Fig. 2 VIS reflectances are plotted as a function of cloud optical thickness at 0.5 µm for water (Fig. 2a) and ice (Fig. 2b) clouds. Reflectances were computed for θ0 = θ = 30° and Φ = 80°. The simulations were carried out for a sea surface characterized by albedo ωs = 0.05 in order to avoid interference from surface reflection. In each graph the different curves refer to various Re values, ranging from 3 to 39 µm for water clouds and from 3 to 200 µm for ice clouds. VIS reflectances show a marked dependence on the optical thickness, but a very scarce sensitivity to Re. A slight increase of the reflectance with decreasing Re is evident from Fig. 2, but this makes it only possible to distinguish small cloud particles with Re < 6 µm from considerably larger particles with Re > 40 µm. This implies that it is not possible to exploit this dependence for a retrieval of the effective radius. No capabilities to distinguish cloud phase can be attributed to the channel: from an analysis of Fig. 2, water and ice clouds with the same Re and τ values exhibit quite similar reflectances, due to the fact that at VIS wavelengths the scattering is conservative. Considering the radiometric performances of MODIS and SEVIRI and the sensitivity of the VIS reflectances to τ, an estimation of the errors on the retrieved τ values was done. For MODIS it was possible to obtain an estimate of the radiometric error (standard deviation) out of the values of the Uncertainty Index disseminated with Level 1B data sets. The percent uncertainties derived from the index in case of VIS channels are directly applicable to the reflectance data. Using several Uncertainty Indexes for a number of MODIS granules an average percent uncertainty value of ± 2.5% was found. For SEVIRI the official EUMETSAT short-term radiometric error reported by Schmetz et al. (2002), 0.27 at 5.3 W m–2·sr–1·µm–1, was used. From the propagation of errors, neglecting the error associated to the
The Retrieval of Cloud Top Properties Using VIS-IR Channels
85
solar irradiance, and assuming a constant percent error on radiance, the percent error on the reflectance was determined to be equal at ±5%. In Fig. 3 each τ value used in the simulations is displayed with its uncertainty interval (shaded areas) whose extremes were computed as the τ values that correspond to ρ ± ∆ρ, where ρ is a reflectance value at the fixed geometry, Re and τ, and ∆ρ is the error associated to the reflectance. Figure 3 refers to a water cloud with an effective radius of 12 µm and the solar and satellite geometry is the same as in Fig. 2. The τ uncertainty intervals are computed for ∆ρ/ρ = 2.5%·(light gray shaded area) and 5%·(dark gray shaded area). Note that from Fig. 3 the retrieval of cloud optical thickness is not reliable for τ values greater than ≈30–40. For τ = 30 a percent error of about ±16% is found, in case of ∆ρ/ρ = 5%, whereas for ∆ρ/ρ = 2.5% and τ = 40 the percent error is of ±10%. After that the decreasing sensitivity of the reflectance data and the increasing ∆ρ damage the τ retrieval capability. Surface reflection can substantially influence the top of the atmosphere signal and hence the relation between reflectance and τ. In Fig. 4 the VIS reflectance vs τ curves for the same cloudy scenario of Fig. 2 refer to different surface albedo (ωs) values that are taken from the ωs data set of the RTM Streamer (Key and Schweiger 1998). The curve with ωs = 0 is also plotted as a reference.
Figure 3. Uncertainties in the retrieved τ depending on measurement (reflectance) errors for a water cloud with Re = 12 µm. The solar and satellite geometries are the same as in Fig. 2. The light gray shaded area represent the τ uncertainty for a percent error on VIS reflectance of ±2.5%, whereas the dark gray shaded area is relative a percent error on VIS reflectance of ±5%.
E. Cattani et al.
86
reflectance at 0.6 µm
1.0 0.8 0.6
Re = 12 µm water cloud
0.4
ωs = 0 ωs = 0.05 sea ωs = 0.14 veg. ωs = 0.19 sand ωs = 0.83 snow
0.2 0.0 1
10
100
cloud optical thickness at 0.5 µm Figure 4. Simulated reflectances at 0.6 µm as a function of τ and surface albedo ωs under the same conditions as in Fig. 2. The effective particle radius is fixed at Re = 12 µm.
Surface reflection largely contributes to the satellite signal in particular for optical thickness lower than 10–20, having as a consequence an overestimation of the cloud optical thickness. Moreover, highly reflecting surfaces, like snow in Fig. 4, can completely cancel the dependence of the VIS reflectance on τ.
3.2 NIR channels Data in the NIR channels, in particular at 3.7 µm, are widely used for the retrieval of the cloud effective radius. Radiances at these wavelengths depend almost exclusively on Re, especially in the case of thick clouds. The thermodynamic phase is another cloud parameter that can modulate the NIR signal, due to the different absorption properties of water and ice at these wavelengths (see Fig. 1). However, several phenomena may intervene by modifying the relation between the radiance values and Re or the cloud phase, i.e., the surface reflection, the water vapor absorption and, for the waveband at 3.7 µm, the thermal emission. Radiative transfer simulations at 1.6 and 3.7 µm were carried out in order to exemplify the phenomena previously summarized. The behavior of the reflectances as a function of Re for water clouds is shown in Fig. 5a and b for the 1.6 and 3.7 µm channels, respectively. Similarly in Fig. 6a and b are plotted the reflectances at 1.6 and 3.7 µm for ice clouds.
The Retrieval of Cloud Top Properties Using VIS-IR Channels
87
(a)
(b) Figure 5. Reflectances at 1.6 (a) and 3.7 µm (b) for water clouds as a function of the cloud effective radius. The different curves refer to various VIS optical thickness values.
The various curves refer to different cloud optical thickness values at 0.5 µm. Reflectances shown in Fig. 5 and 6 represent only the cloud signal without any other contributions, so they were computed in dry atmospheric conditions and for a surface with ωs = 0. Moreover, the thermal emitted component at 3.7 µm was omitted. The reflectances at both NIR channels are very sensitive to the cloud particle effective radius: in particular the reflectances decrease with increasing Re, due to the increase of the absorption by cloud particles, which is more efficient at 3.7 µm than at 1.6 µm for both ice and water clouds. A slight dependence of the NIR reflectances on τ can be noted, which is
E. Cattani et al.
88
a)
b) Figure 6. Same as in Fig.5 but for ice clouds.
drastically reduced in case of Re > 3–4 µm and thicker clouds (τ > 20 at 1.6 µm and τ > 10 at 3.7 µm). These conditions are necessary for the simultaneous retrieval of τ and Re, exploiting the combined use of NIR and VIS reflectances. Only in this case the sensitivity of the VIS and NIR reflectances to τ and Re is nearly orthogonal. This means that τ and Re can be determined nearly independently and thus measurement errors in one channel have little impact on the cloud property determined primarily by the other channel (Nakajima and King 1990). Additionally, for Re < 3–4 µm and thin clouds multiple solutions of τ and Re are possible, as stated by Nakajima and King (1990) and Nakajima and Nakajima (1985). By comparing Fig. 5a with Fig. 6a and Fig. 5b with Fig. 6b, the sensitivity of the NIR reflectances to the cloud thermodynamic phase can be
The Retrieval of Cloud Top Properties Using VIS-IR Channels
89
understood, being due to the different absorption properties of water and ice clouds in both spectral channels (see also Fig. 1). As in the previous section for τ, the uncertainty in the retrieved Re depending on the reflectance errors was evaluated for both 1.6 µm and 3.7 µm channels. The percent errors on reflectance at 1.6 µm for MODIS and SEVIRI were determined in the same way as for the VIS channel: from the Uncertainty Index values in percent error of about ±5% was found for MODIS, whereas a percent error of ±10% was computed from the short-term radiometric error for SEVIRI. The determination of the percent error on reflectance for the 3.7 µm channel is more complex because it is very difficult to weigh the error due to the extraction of the thermal emitted component from the total signal. This error has to be combined with the one from the total signal to obtain the uncertainty in the solar reflected component. Therefore, for this channel the uncertainty in the retrieved Re was evaluated for increasing values of the percent error on the reflectance, from 5 to 30%, in order to fix a maximum limit in the reflectance error beyond which the Re retrieval is surely unreliable. Figure 7 shows the uncertainty in the retrieved
a)
b)
c)
d)
Figure 7. Uncertainty in the retrieved Re as a function of the reflectance errors for water and ice clouds at 1.6 µm (a and b) and 3.7 µm (c and d). The different gray shaded areas are relative to different reflectance error values.
90
E. Cattani et al.
effective radius as a function of the percent error on reflectances at 1.6 µm (Fig. 7a and b) and 3.7 µm (Fig. 7c and d) for water and ice clouds. The uncertainty intervals associated to each Re value are represented by shaded areas (different gray shades are relative to various percent errors on the reflectances) and were computed using the NIR reflectances displayed previously in Figs. 5 and 6 for τ = 200, in order to eliminate the dependence of the reflectance data on cloud optical thickness. From Fig. 7 it comes out that the effective radius retrieval is not reliable in case of Re < 4 µm for both spectral channels and cloud phases. These Re values are characterized by percent errors of up to 50% and uncertainty intervals that overlap, showing the very scarce sensitivity of the reflectance to such small Re values. For water clouds at 1.6 µm (Fig. 7a) the Re retrieval proves reliable in case of a percent error on reflectances of 5%. In this case the errors on the effective radii are of about 10%, especially for Re > 10 µm. While increasing the error on the reflectance data up to 10% a deterioration in the capability to retrieve Re arises with Re errors >20% thus preventing to appreciate the Re variations. At 3.7 µm (Fig. 7c) the errors on the retrieved Re are <10% for percent errors on the reflectance data up to 15%. For reflectance errors ≥20% (results not shown in Fig. 7c) the sensitivity to the effective radius is definitively compromised. The performances of the 1.6 µm channel in case of ice clouds (Fig. 7b) and for a reflectance error of 5% are even better than those for the water clouds, showing Re errors much
Figure 8. Different contributions to the top of the atmosphere signal at 3.7 µm for a water cloud.
The Retrieval of Cloud Top Properties Using VIS-IR Channels
91
lower than 10% (~5–6%). This is due to the lower reflectance values for ice clouds than for water clouds, which induce smaller absolute errors on reflectances (∆ρ). For Re > 70 µm the uncertainty intervals associated to each Re value overlap because of the weak dependence of the reflectance on the effective radius (see Fig. 6a), and the retrieval loses its reliability. The Re retrieval is completely compromised by a reflectance error of 10%. The Re retrieval for ice clouds at 3.7 µm (Fig. 6d) is effective if ∆ρ/ρ ≤ 15%. With increasing ∆ρ the uncertainties on the retrieved Re values are such that it is not possible to distinguish different Re values. Moreover, the retrieval is not efficient for cloud particles with R e > 45–50 µm, due to the very low reflectance values and the almost complete absence of sensitivity of the reflectance on Re (see Fig. 6b). NIR radiance data can be affected by different spurious signals that can modify the sensitivity to the cloud parameters. Figure 8 shows the effects of water vapor absorption, surface reflection and thermal emission on the cloud signal at 3.7 µm and for water clouds with τ = 5. A high reflecting surface (ωs = 0.370 at 3.7 µm) increases the total signal, particularly in case of small Re values where the cloud absorption is weaker. The water vapor above cloud top is responsible for the absorption of radiation coming out of the cloud top. This phenomenon is more effective the lower the cloud top because of the greater amount of water vapor, and the greater θ and θ0. 3.7 µm radiances are more affected by the water vapor absorption than the 1.6 µm ones. The thermal emission, which is present only at 3.7 µm, produces an additive signal that is scarcely dependent on Re, as can be noted from Fig. 8 by comparing the curves with empty triangles and squares. For this reason several retrieval methods subtract this component from the total signal and use only the solar reflected component (Nakajima and Nakajima 1995; King et al. 1997).
3.3 IR channels Brightness temperatures (BT) at 11 and 12 µm were simulated for ice and water clouds in order to determine their sensitivity to the principal cloud parameters, Re, τ and the thermodynamic phase, and to evaluate the influence, which the water vapor can exert. In Fig. 9 the BTDs between 11 and 12 µm are plotted as a function of the brightness temperatures at 11 µm (BT11) for a water cloud. Two values of the effective radius, 5 µm (circles) and 15 µm (triangles), and 18 τ values ranging from 0 to 200 have been considered. Cloud top and bottom heights were fixed at 1.5 and 0.5 km, respectively. The behavior of the BTD and BT11 was evaluated (1) in absence of water vapor below and above the cloud layer (solid line), (2) with water vapor below the cloud layer (dashed line), and (3) with water vapor below and above the cloud (dotted line). For these simulations the standard water vapor
92
E. Cattani et al.
Figure 9. Brightness temperature differences as a function of the brightness temperature at 11 µm for a water cloud. The different symbols refer to R evalues used in the simulations, 5 µm (circles) and 15 µm (triangles). The simulations were carried out for dry atmospheric conditions (solid line), in presence of water vapor below the cloud layer (dashed line) and with water vapor above and below the cloud (dotted line). The different τ values used are specified in the figure.
amounts relative to the Midlatitude Summer profile were used. The surface was considered as a black body with a temperature of 294.20 K. Figure 10 is the same as Fig. 9 but for an ice cloud with top and bottom heights at 10 and 8.5 km, respectively. The brightness temperatures simulated in dry atmospheric conditions for both water and ice clouds are very sensitive to cloud optical thickness variations: BT11 decreases with increasing τ, ranging from the value of surface temperature for τ = 0 to the cloud top temperature for high τ values, as the cloud absorbs increasing amount of surface emitted radiation. For τ > 20 the cloud is sufficiently thick to be considered a black body and thus BT11 does not vary anymore. Less marked is the sensitivity of BT11 to Re: the greater the effective radius values the lower the BT11, due to the greater absorption of ice or water cloud particles. As for the BTD their dependence on τ can not be used in a retrieval of cloud optical thickness, due to the fact that the same BTD value can correspond to very different τ values, whereas the high sensitivity to Re could be exploited for the retrieval. Note that cloudy scenarios with 1 < BTD < 5 K can be equally associated to water or ice clouds (Giraud et al. 2001) as can be seen from Figs. 9 and 10. Therefore
The Retrieval of Cloud Top Properties Using VIS-IR Channels
93
Figure 10. Same as in Fig. 9 but for an ice cloud.
it could be very difficult to give a correct interpretation of BTD signatures without phase information. The introduction of water vapor below cloud bottom affects BT11 and BTD only in case of thin cloud layers (τ < 2). Water vapor behaves as an absorber of the surface emitted radiation and hence BT11 decreases with respect to the values assumed in dry conditions, whereas BTD increases. In general this phenomenon can be neglected for water clouds as the majority of them have optical thickness greater than 2. Only water clouds are affected by the presence of water vapor above cloud top because the water vapor profiles have their maximum at low altitudes. The effect can be noted at every τ value since in this case the water vapor absorbs the radiation coming out of the cloud top.
4
CONCLUSIONS
Radiative transfer simulations were used to evaluate the sensitivity of radiance data at VIS, NIR and IR satellite sensor channels to the principal cloud parameters, and to explore the possibility of exploiting it in retrieval procedures. Reflectance data at 0.6 µm proved to be the most suitable for the retrieval of the cloud optical thickness. The retrieval is feasible for τ values up to 30–40, with relative errors not greater than 10% in case of errors on the reflectance ∆ρ/ρ ≤ 5%. The surface reflection and the slight dependence of the reflectance on Re have to be taken into account for a correct τ retrieval.
E. Cattani et al.
94
The Re dependence of the VIS reflectance can introduce significant errors if not accounted for during the τ retrieval. For this reason VIS channels are often used in conjunction with NIR channels in iterative retrieval procedures, where the Re estimates from NIR wavebands are exploited as first guess for the τ retrieval and vice versa, until convergence is achieved. The Re retrieval can be carried out using radiances at 1.6 or 3.7 µm, according to the different τ values, cloud phase, surface reflection and water vapor amount. 3.7 µm radiances can accurately provide Re for much thinner clouds (τ > 10) than 1.6 µm radiances (τ > 20). Actually these τ values together with Re ≥ 3–4 µm are requested to have reflectances that don’t depend on the cloud optical thickness and to prevent τ and Re multiple solutions. In case of water clouds the Re retrieval is more reliable at 3.7 µm, with errors on the retrieved Re lower than 10% for reflectance errors ≤ 15%, than at 1.6 µm, where Re errors of about 10% can be obtained only for reflectance errors not greater than 5% and Re > 10 µm. The performances of the 1.6 µm channel improve in case of ice clouds, with Re error values ≤ 10% for ∆ρ/ρ = 5%, whereas those of the 3.7 µm reflectance data are quite similar to the previous ones obtained for water clouds. Moreover the 1.6 µm reflectances are sensitive to much greater Re values than the reflectances at 3.7 µm are (up to ~70 µm for 1.6 µm and Re = 45–50 µm for 3.7 µm). The surface reflection can bring about an underestimation of the retrieved Re value. Generally greater τ values are needed at 1.6 µm with respect to 3.7 µm to neglect this phenomenon. This is due to the fact that water and ice cloud particles absorb much more radiation at 3.7 µm than at 1.6 µm, and numerous surface types are characterized by greater albedo values at 1.6 µm than at 3.7 µm. Also the presence of water vapor above the cloud top has to be taken into account, as it can significantly modify the top of the atmosphere signal, especially at 3.7 µm, for low clouds and zenith views. BTD vs BT11 diagrams proved to be useful for the determination of τ, the cloud top temperature and Re. By examining the diagrams relative to large enough areas, containing clear and cloudy satellite sensor pixel with variable τ values, it is possible to extract some information about these cloud parameter from the interpretation of the typical ‘arch’ signature of the BTD vs BT11 curves. Also at these wavelengths the water vapor absorption represent a spurious effect that has to be taken into account, particularly in case of low clouds.
5
REFERENCES
Arking, A. and J. D. Childs, 1985: Retrieval of cloud cover parameters from multispectral satellite images. J. Climate Appl. Meteor., 24, 322–333. Barnes, W. L., T. S. Pagano, and V. V. Salomonson, 1998: Prelaunch characteristics of the MODerate Resolution Imaging Spectroradiometer (MODIS) on EOS-AM1. IEEE Trans. Geosci. Remote Sens., 36, 1088–1100.
The Retrieval of Cloud Top Properties Using VIS-IR Channels
95
Giraud,V., O. Thouron, J. Riedi, and P. Goloub, 2001: Analysis of direct comparison of cloud top temperature nad infrared split window signature against independent retrievals of cloud thermodynamic phase. Gephys. Res. Lett., 28, 983–986. Han, Q., W. B. Rossow, and A. A. Lacis, 1994: Near-global survey of effective droplet radii in liquid water clouds using ISCCP data. J. Climate, 7, 465–497. Inoue, T., 1987: A cloud classification with NOAA 7 split-window measurements. J. Geophys. Res., 92, 3991–4000. Key, J. R. and A. J. Schweiger, 1998: Tools for atmospheric radiative transfer: Streamer and Fluxnet. Computers and Geosciences, 24, 443–451. Key, J. R. and J. M. Intrieri, 2000: Cloud particle determination with the AVHRR. J. Appl. Meteor., 39, 1797–1804. King, M. D., Y. J. Kaufman, W. P. Menzel, and D. Tanré, 1992: Remote sensing of cloud, aerosol, and water vapor properties form the MODerate Resolution Imaging Spectrometer (MODIS). IEEE Trans. Geosci. Remote Sens., 30, 2–27. King, M. D., S.-C. Tsay, S. E. Platnick, M. Wang, and K.-N. Liou, 1997: Cloud retrieval Algorithms for MODIS: optical thickness, effective radius, and thermodynamic phase. ATBD-MOD-05. Kneizys, F. X., E. P. Shettle, L. W. Abreu, J. H. Chetwynd, G. P. Anderson, W. O. Gallery, J. E. A. Selby, and S. A. Clough, 1988: Users Guide to LOWTRAN 7. AFGL-TR-88-0177. Nakajima, T. and M. D. King, 1990: Determination of the optical thickness and effective particle radius of clouds from reflected solar radiation measurements. Part I: Theory. J. Atmos. Sci., 47, 1878–1893. Nakajima, T. Y. and T. Nakajima, 1995: Wide-area determination of cloud microphysical properties from NOAA AVHRR measurements for FIRE and ASTEX regions. J. Atmos. Sci., 52, 4043–4059. Nakajima, T. and M. Tanaka, 1986: Matrix formulation for the transfer of solar radiation in a plane-parallel scattering atmosphere. J. Quant. Spectrosc. Radiat. Transfer, 35, 13–21. Nakajima, T. and M. Tanaka, 1988: Algorithms for radiative intensity calculations in moderately thick atmospheres using a truncation approximation. J. Quant. Spectrosc. Radiat. Transfer, 40, 51–69. Rosenfeld, D., 2007: Cloud top microphysics as a tool for precipitation measurements. In: Measuring precipitation from space – EURAINSAT and the future. V. Levizzani, P. Bauer, and F. J. Turk, eds, Springer, 61–78. Rosenfeld, D. and G. Gutman, 1994: Retrieving microphysical properties near the top of potential rain clouds by multispectral analysis of AVHRR data. Atmos. Res., 34, 259–283. Rosenfeld, D. and I. M. Lensky, 1998: Satellite-based insights into precipitation formation processes in continental and maritime convective clouds. Bull. Amer. Meteor. Soc., 79, 2457–2476. Schmetz, J., P. Pili, S. Tjemkes, D. Just, J. Kerkmann, S. Rota, and A. Ratier, 2002: An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc., 83, 977–992. Stamnes, K., S.-C. Tsay, W. Wiscombe, and K. Jayaweera, 1988: Numerically stable algorithm for discrete-ordinate-method radiative transfer in multiple scattering and emitting layered media. Appl. Opt., 27, 2502–2509. Yang, P. and B. A. Baum, 2003: Satellite remote sensing/Cloud properties. Vol. 5, Encyclopedia of Atmospheric Sciences, ed. J. R. Holton, J .A. Curry and J. A. Pyle, Academic Press, 1956–1965. Zhang, H. and W. P. Menzel, 2002: Improvement in thin cirrus retrievals using an emissivityadjusted CO2 slicing algorithm. J. Geophys. Res., 107, 4327–4340.
8 CLOUD MICROPHYSICAL PROPERTIES RETRIEVAL DURING INTENSE BIOMASS BURNING EVENTS OVER AFRICA AND PORTUGAL Maria João Costa1, Elsa Cattani2, Vincenzo Levizzani2, and Ana Maria Silva1 1
Department of Physics and Evora Geophysics Centre, University of Evora, Evora, Portugal Institute of Atmospheric Sciences and Climate, ISAC-CNR, Bologna, Italy
2
1
INTRODUCTION
Clouds are a major driving force behind climate mechanisms. They strongly modulate the energy balance of the Earth through absorption and scattering of solar radiation and absorption and emission of terrestrial radiation, and on the other hand, clouds and precipitation are the regulating factors of the hydrologic cycle. Although the importance of clouds is widely recognised, their impact is associated with great uncertainties due to the complexity and space-time variation of the cloud phenomena, therefore the global monitoring of their optical and microphysical properties retrieved from multispectral satellite sensor data becomes a main task/necessity. A great number of studies were conducted on the possible modification of cloud properties through the interaction with atmospheric aerosol particles, as this may lead to important changes of the Earth’s climate. On the one hand, aerosol particles acting as cloud condensation nuclei may be responsible for direct modifications of the cloud properties (Albrecht 1989; Bréon et al. 2002; Kawamoto and Nakajima 2003) – first indirect effect. It consists of a decrease of droplet size due to an increase in droplet concentration (assuming a constant liquid water content), when a cloud is polluted with anthropogenic aerosol particles serving as additional cloud condensation nuclei (Twomey 1974). On the other hand, due to the decrease of cloud particle size, aerosol particles may indirectly interfere with cloud lifetime and precipitation efficiency, producing the second indirect effect, 97 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 97–111. © 2007 Springer.
98
M. J. Costa et al.
which is characterized by the diminution of the efficiency of precipitation forming processes, tending to increase the liquid water content, cloud lifetime and thickness (Rosenfeld 1999, 2000). Both effects may interfere with the planetary albedo. The biomass burning aerosol is considered one of the main responsible of this kind of cloud properties modification. In addition, this type of aerosols contains organic compounds and black carbon, the latter being a strong absorber of solar radiation that can greatly impact cloud formation and evaporation (Ackerman et al. 2000). Many regions of the Earth are characterized by the emission of biomass burning aerosol connected to the agricultural practises and forest fires, but the quantification of the effects of the aerosol-cloud interaction is far from being completely achieved. Satellite measurements provide indispensable data for effective global observations of clouds properties. However, satellite retrievals are in general less accurate than in situ observations, which are an essential tool for comparisons with satellite-derived parameters, constituting the quality control of global satellite products. The development of the present methodology is motivated by the existence of a new generation of geostationary (GEO) satellite measurements such as those from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) flying on Meteosat Second Generation (MSG) (Schmetz et al. 2002). This innovative sensor opens new perspectives with respect to past and present GEO systems since it provides the necessary additional spectral measurements, supplied until now exclusively by low Earth orbit (LEO) satellite sensors. The doubled sampling frequency and improved spatial resolution prompts for global monitoring of cloud properties and eventually advances in cloud-aerosol interaction studies, facilitating the task of comparing the derived cloud properties with in situ measurements as well. Data from MSG-1 SEVIRI was not yet distributed on a regular basis at the moment the present study was carried out and therefore measurements from the MODerate resolution Imaging Spectroradiometer (MODIS) (Barnes et al. 1998) onboard Terra and Aqua LEO satellites were used, which present comparable spectral channels in the visible (VIS), thermal infrared (IR), and near IR (NIR). The methodology for the characterization of the cloud properties is based on satellite multi-spectral measurements used in combination with radiative transfer calculations to retrieve the cloud optical thickness (COT), particle effective radius (ER) and cloud top temperature (CTT). The retrieval procedure is applied to strong aerosol events from intense biomass burning aerosol transports that occurred in Southern Africa and Portugal in summer 2000 and 2003, respectively, in order to investigate possible alterations of the cloud properties. Comparisons between the retrieved parameters and MODIS cloud products have been carried out. Moreover, for the case study over Southern Africa comparisons with in situ measurements of the Southern African Fire-Atmosphere Research Initiative 2000 (SAFARI 2000)
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events
99
(http://mercury.ornl.gov/safari2k/) field campaign are presented, in the attempt to improve the understanding of the interactions between clouds and aerosol.
2
METHODOLOGY
The first step of the methodology consists of the cloud detection over the area selected for the study, as well as the particle phase determination (liquid water or ice), assuming that clouds at one time are made of either liquid water or ice particles, hence no mixed phase clouds are considered in the study. The pixel classification procedure relies on a bi-spectral technique that uses MODIS measurements in the VIS and IR channels, centred at 0.65 and 11 µm respectively. The satellite measurements are initially classified in terms of the underlying surface (land or water) using a land-sea mask. Subsequently, the histograms of the VIS radiance measurements and IR brightness temperature values are analysed to determine threshold values that define the limits between clear sky, water clouds and ice clouds. Such threshold classification is done at the pixel level and when the pixel is cloudy, four possible cases are distinguished: water clouds over the ocean, ice clouds over the ocean, water clouds over land, and ice clouds over land (see Table 2). The VIS, NIR (centred at 3.75 µm) and IR radiance measurements corresponding to the pixels classified in the four categories are used to retrieve COT, ER and CTT using the algorithm proposed by Nakajima and Nakajima (1995) and Kawamoto et al. (2001). The four categories are treated separately because relevant differences in cloud and surface characterization must be taken into account. The algorithm relies on the comparison between the modelled cloud radiances in the three spectral bands and the corresponding satellite radiance measurements, deprived of the undesirable components, such as the solar radiation reflected by the surface and the thermal radiation emitted from the cloud layer and the surface, in order to obtain only the cloud signal. These corrections are based on the use of LookUp Tables (LUTs) calculated using the radiative transfer code (RTC) RSTAR (Nakajima and Tanaka 1986, 1988). The LUTs contain the radiative quantities necessary for the cloud properties retrieval, namely the cloud reflected radiances and spherical albedo in the VIS and NIR, the transmission in the VIS, NIR and IR and the reflection and atmospheric emitted radiation in the NIR and IR spectral bands. The RTC calculations are done taking into account the MODIS spectral response functions for each of the three spectral channels. The LUTs are built for a grid of selected values of the COT, ER, CTT, equivalent water vapour above the cloud (ewvu), equivalent water vapour of the cloud layer (ewvc), solar zenith (θ0), satellite zenith (θ ) and relative azimuth (Φ) angles, as shown in Table 1. The cloud is characterised by a lognormal size distribution and mean values of surface temperature and
M. J. Costa et al.
100
reflectance, as well as standard atmospheric vertical profiles (McClatchey et al. 1971), are provided according to the actual conditions (see Table 3). Besides the COT, ER and CTT, the cloud top height (CTH), and pressure (CTP) are also retrieved, from the CTT values by linear interpolation of the selected atmospheric vertical profile. The code was modified to yield also the cloud type as output. In its original version, the cloud type or the cloud geometrical thickness had to be initially defined. In this modified version, the cloud type is established from the COT and CTP values, using the ISCCP cloud classification (Rossow and Schiffer 1999). Table 1. Grid of values used to build the LookUp Tables.
Grid parameter COT ER (µm) CTT (K) ewvu (g m–2) ewvc (g m–2) θ0 (º) θ (º) φ (º)
Grid values Cloud phase Liquid water
Ice
1, 2, 4, 6, 9, 14, 20, 30, 50, 70
0.1, 0.5, 1, 2, 4, 8, 16, 32, 48, 64 5, 10, 20, 40, 60, 80, 100, 120, 140, 160, 2, 4, 6, 9, 12, 15, 20, 25, 30, 35, 40 180 250, 260, 270, 280 220, 230, 240, 250, 260, 270 50, 5000, 10000, 20000, 30000, 40000, 50000 50, 5000, 10000, 20000, 30000, 40000, 50000 0, 5, 10, 20, 30, 35, 40, 45, 50, 55, 60, 65, 70 0, 5, 10, 20, 30, 35, 40, 45, 50, 55, 60 0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160, 170, 180
Comparisons between the retrieval results and time – space collocated MODIS official cloud product data sets (King et al. 1998) of COT, ER and CTT were carried out. The idea was to validate the present retrievals against a state of the art retrieval algorithm, in particular as regards to the setup of the algorithm and the selection and classification of cloudy pixel. In situ measurements taken during the intensive SAFARI 2000 campaign in Southern Africa (Swap et al. 2003), conducted from 13 August to 25 September 2000, were used as a further source of comparison data for the African case study. The datasets used are the cloud effective radius measurements made by the UK Met Office C-130 aircraft (Keil and Haywood 2003), and the cloud and aerosol layers bottom and top heights as a part of the cloud and aerosol measurements from the ER-2 Cloud Physics Lidar (CPL) (McGill et al. 2003).
3
CASE STUDIES
The methodology described in the previous section is applied to two selected episodes where marine stratocumulus and convective clouds are observed in the presence of intense biomass burning events, one in the southern hemisphere and the other in the northern hemisphere.
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events
101
Figure 1. Absorbing aerosol particles detected by the Earth Probe TOMS. The grayish spots locate the absorbing aerosol. The white boxes delimit the location of the MODIS granules used for the cloud top properties retrieval. Images available online at http://toms.gsfc. nasa.gov/ aerosols/aerosols.html.
The first case study concerns the intense fire season that occurred in Southern Africa during summer 2000 with a peak in late August and early September. The region is subject to one of the most frequent occurrences of biomass burning in the world. The heaviest burning was in western Zambia, southern Angola, northern Namibia, and northern Botswana. Yet, the smoke from these fires may be transported over substantial distances during several days or even weeks being detected some thousands of kilometres away from the fire sources. The present analysis focused on the African coasts between Southern Angola and Northern Namibia. This oceanic area is normally characterized by the presence of semipermanent stratiform clouds, so it provides a good opportunity to increase the understanding of cloud microphysics and cloud-aerosol interactions. The aerosol index maps from the Total Ozone Mapping Spectrometer (TOMS) onboard the Earth Probe (Torres et al. 1998) presented in Fig. 1 for the 5, 11 and 13 September 2000 reveal the presence of absorbing aerosol particles as smoke. The greyish spots over the dominating background colour delimit the areas where the absorbing particles are present. The white boxes delimit the approximate location of the MODIS-Terra granules used in each of the days. 5 September 2000 is representative of background conditions since the aerosol event can be barely noted on the upper right part of the white box, which corresponds to a cloud-free area. On 11 September aerosol and background conditions co-exist since the aerosol event does not concern the lower part of the granule, but only the upper part, where clouds are also detected. A vast aerosol mass extends over the ocean in a cloudy area on 13 September. The second case study focuses on a smoke transport event originated from the numerous uncontrolled fires burning across continental Portugal during August 2003, which turned out to be the worst fire season that Portugal faced in the last 23 years if not ever. The fires continued to spread during several days taking advantage of the hot, windy and dry conditions all over
M. J. Costa et al.
102
Figure 2. Same as Fig. 1 but for the Portuguese fires.
the country. It is estimated that an area of about 5.6% of the entire Portuguese forest area burnt until 20 August 2003. Nevertheless, the fires continued at least until mid September, contributing to aggravate the scenario. The TOMS images presented in Fig. 2 illustrate the regions with absorbing aerosol particles, coming from the fires burning across Portugal and Spain. Clouds are detected on both days in the areas affected by the absorbing aerosol plumes. The white boxes indicate once again the geographical location of the MODIS-Terra (4 August) and MODIS-Aqua (5 August) granules used for the retrieval.
4
RESULTS OF THE RETRIEVAL PROCEDURE
The threshold determined from the bi-spectral analysis in terms of the brightness temperature (TB) and of the VIS radiance (IVIS), are summarised in Table 2. In Table 3, the vertical profiles, as well as the surface reflectance and temperature used, are indicated. One of the apparent effects out of the cloud-aerosol interaction is the reduction of the cloud particle size due to the increase of cloud condensation nuclei that induce a redistribution of the cloud water content. The analysis was therefore concentrated on the retrieved ER values of water clouds to detect any cloud modification due to the aerosol effect. Figure 3 shows the frequency histograms of the ER (for the water clouds over the ocean) corresponding to the three days analysed. The classification of the MODIS pixels for 5 September 2000 allowed for distinguishing water clouds located mainly over the ocean, but not in the area where the absorbing aerosols are detected (Fig. 1a). Therefore, we are probably in this case dealing with clouds in the absence of any absorbing aerosol from the biomass fires. The frequency histogram (Fig. 3a) presents a peak of the ER around 11 µm with a low spread of the values. In the 11 September case, the water clouds detected by the classification cover most of the oceanic area of the granule, however only the upper part of the granule corresponds to the coexistence of the absorbing aerosols and water clouds, as shown in Fig. 1b. The histogram of the 11 September shows a bimodal distribution with a peak of the ER towards 14 µm and a smaller peak around 6 µm. This second peak can be connected to the presence of
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events 05/09/2000 1015UTC
11/09/2000 0940UTC
(a)
103
3/09/2000 0925UTC
(b)
(c)
Figure 3. Histograms of the ER obtained for the African biomass burning event. They refer to water clouds detected over ocean. Table 2. Threshold values obtained from the bi-spectral cloud classification.
African fires Iberian Peninsula fires
Ice clouds
Water clouds Over ocean Over land
Case study
260K40Wm-2sr-1µm-1 260K40Wm-2sr-1µm-1
260K
TB ≤ 260K
260K
TB ≤ 260K
Table 3. Vertical profile, surface reflectance and temperature values used.
Case study
Vertical profile
African fires Iberian Peninsula fires
Tropical Midlatitude summer
Ocean VIS Surface reflectance temperature
Land VIS Surface reflectance temperature
0.05
288K
0.10
305K
0.05
285K
0.12
305K
biomass burning aerosol on the upper part of the granule, where the pixels with lower particle sizes were located. The mean ER value and the respective standard deviation were calculated only for the area where, according to TOMS information, the absorbing aerosol particles were present, obtaining ER = 7.7 ± 3.2 µm and CTH = 5.2 ± 0.9 km. This ER value is clearly lower than the one considering the whole granule (ER = 12.7 ± 6.1 µm) and with a lower standard deviation. Additional information from vertically resolved measurements indicate that the smoke plume in the surrounding cloud-free area (approximate location: 22.5° S and 12.2° E) was detected between 2 and 5 km of height (Kaufman et al. 2003). Since the retrieved average CTH is quite close to these measurements, it could well be that the cloud suffered the influence of the aerosol particles, reflecting in lower values of the ER. However, the retrieved CTH values may have been overestimated due to the vertically separated cloud and aerosol layers, which are currently not taken into account by the method.
M. J. Costa et al.
104 04/08/2003 1130UTC
05/08/2003 135UTC
(a)
(b)
Figure 4. Same as Fig. 3 but for the Iberia forest fire event.
On 13 September 2000, absorbing aerosol particles (see Fig. 1c) are suspended all over the oceanic area covered by the MODIS granule where also water clouds are distinguished by the pixel classification technique. The ER values for 13 September 2000 are spread over a larger interval and peak around 11 µm. As it appears from the TOMS map (Fig. 1c), the absorbing aerosol is located over the ocean and spatially coexists with the clouds in the area. However, no trend could be distinguished in this case. Figure 4 shows the frequency histograms of the ER (for the water clouds over the ocean) corresponding to the results of the Portuguese case study. According to our pixel classification technique the MODIS granule area were mainly covered by water cloud. Note from the TOMS maps presented in Fig. 2 that the absorbing aerosol particles are present on both days (4 and 5 August 2003) even though they do not cover the whole area swept by the granules. The mean ER value found for the 4 August 2003 case is quite small with a relatively low dispersion of the values. In fact, about 95% of the examined pixels present ER < 10 µm (Fig. 4a). On 5 August, the average ER and the spread of the values are higher, as evident from the broad form of the histogram in Fig. 5b, that shows two higher peaks at lower ER values well differentiating from the rest of the distribution. In this case, two water cloud systems were identified by the classification. In particular these cloudy pixels, characterised by lower ER values, refer to the cloud system located near the Portuguese coast, where the absorbing aerosol was detected (Fig. 2b). The mean ER value and standard deviation calculated for the area where TOMS indicated the presence of the absorbing aerosol, result in lower mean ER values and associated standard deviation (ER = 7.9 ± 3.1 µm) with respect to that obtained considering all cloudy pixels of the granule (ER = 11.8 ± 6.7 µm). The exceptionally low effective radius values found in the water clouds near the Portuguese coast could be connected to the interaction of clouds with the smoke coming from the fires burning across the country.
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events
(a)
(b)
105
(c)
Figure 5. Comparisons between the official MODIS cloud products and the cloud parameters retrieved from the present methodology (ER, COT and CTT), for the African case study, 13 September 2000. Relative differences are shown with respect to the MODIS products, for pixels over ocean classified as covered by water clouds.
On the other hand, once more it can simply be the result of aerosol and cloud layers at different atmospheric levels not taken into account by the RTC calculations. Additional information to understand the cause of the low ER retrieved values could be obtained from atmospheric aerosol profiles of the regions, which unfortunately are not available.
5
COMPARISONS WITH MODIS CLOUD PRODUCTS AND IN SITU MEASUREMENTS
The results of the retrieval (ER, COT and CTT) are compared with the corresponding MODIS cloud products for both case studies. The results of the comparisons for the African case study are shown in Fig. 5. The frequency of occurrence is plotted as a function of the relative differences between the cloud parameters of the MODIS official products and our retrieved cloud parameters. 13 September was selected as representative of a typical scenario containing biomass burning aerosols and clouds, but similar results are obtained for the other days (not shown). The results reported in these graphs are relative to the MODIS pixels over ocean classified as covered by water clouds. The agreement between the MODIS official product parameters and the ones retrieved according to the present methodology is encouraging since for all three cloud parameters the frequency of occurrence distributions are sharp and peaked near zero. Only the COT histogram is peaked near the 15% relative difference, pointing out a slight underestimation of the COT by the present retrieval method. However, more than 50% of cases show relative differences lower than 15%. The same analysis was conducted for the case study over Portugal, whose results are reported in Fig. 6 for 5 August 2003, confirming once again an agreement between the two retrieval procedures. In this case, the histogram of ER differences is the broadest, but 50% of the
M. J. Costa et al.
106
(a)
(b)
(c)
Figure 6. Same as Fig. 5 for the case study of the Portuguese fires, 5 August 2003.
examined cases show differences lower than 10%. It is conceivable that a better agreement can be obtained by refining the selection technique of the cloud covered pixels and the cloud thermodynamic phase determination. Further comparisons were carried out using data sets of the SAFARI 2000 measurements campaign, for the Southern Africa case study. Results of the comparisons between the retrieved ER values and the corresponding measurements made by the Met Office C-130 aircraft for 2, 7, 11, 14 and 18 September are displayed in Fig. 7 (different symbols refer to the different days). Average values of the aircraft-measured ER are plotted against the satellite-retrieved ER. Due to the different spatial resolution of the data each aircraft measurement was referred to the nearest MODIS pixel and then, considering all the aircraft measurements attributed to the same satellite pixel, the average value of the aircraft effective radius was computed. The satellite retrieved ER values are in general larger than the ones measured by the C-130, even if the differences are moderate. For example, on 2 September most of the retrieved ER values fall within the range of 10–15 µm and the measured ones are in the range 5–10 µm. A good agreement between the two data sets is found on 7 September although very few coincident satellite-aircraft data were available on this day. The reasons for the above described findings could be twofold: on one hand, aircraft measurements may refer to different altitudes in the cloud from bottom to top, whereas the satellite retrieved ER belongs only to the uppermost cloud layers. On the other hand, the typical scenario observed during these days showed an elevated biomass aerosol layer and low-level stratiform clouds located below it (Keil and Haywood 2003; Haywood et al. 2003). The present retrieval algorithm does not consider the presence of an aerosol layer in the atmosphere. These two reasons could partially justify the differences between measured and retrieved ER in Fig. 7. This remains to be verified by means of radiative transfer simulations of the near infrared radiance data (usually employed for the ER retrieval, as in the present work), where aerosol load is properly described with respect to the scattering properties.
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events
107
Figure 7. Scatter plot of the comparisons between the retrieved ER values and the corresponding measurements made by the UK Met Office C-130 aircraft on 2, 7, 11, 14 and 18 September.
This constitutes ongoing research work of the group and is in the way to be evaluated. On 18 September the aerosol load was greatly diminished off the coast of Namibia and Angola, so the ER values retrieved during this day can be considered characteristic of clouds, which did not undergo the aerosol influence. Note from the scatter plot in Fig. 7 that this day was characterized by the largest ER for both the data sets with respect to the previous days, e.g., 2 or 7 September. This can probably be considered as a signal of the aerosol influence on cloud microphysical properties in the other days (2, 7, 11 and 14 September), when the ER values were lower. For 13 September one more comparison was done between the satellite retrieved CTH and the one measured by the Cloud Physics Lidar (CPL) onboard the NASA ER-2 aircraft. In Fig. 8 the vertical lines identify the bottom and top heights of the various layers retrieved by the CPL as a function of flight time. The CTH retrieved from the present algorithm is also plotted. The CPL data are always attributed to the retrieved CTH of the nearest MODIS pixel. Due to the availability of only one MODIS granule
108
M. J. Costa et al.
data per day for this case study, the time coincidence between the aircraft measurements and the retrieval is not every time optimal, but always within the 2–3 h time frame. Data from a GEO satellite sensor like SEVIRI with a frequency of measurement of 15 min would greatly improve the comparisons and this is a goal for the future. Note that a large portion of the flight was characterized by the presence of a thick aerosol layer located above the cloud top. The agreement between the CPL and MODIS CTH is good enough when the aerosol layer is absent, whereas there is a clear overestimation of the cloud top height computed by the retrieval algorithm in the other case. Again, this confirms that the presence of an absorbing aerosol layer above the cloud damages the quality of the cloud top parameter retrieval from satellite sensors with the present methodology, as already anticipated from the previous comparisons. The incorporation of the aerosol layer in this methodology is still being investigated.
Figure 8. Comparisons between the retrieved CTH and that measured by the Cloud Physics Lidar (CPL) onboard the NASA ER-2 aircraft, for the case study of Southern Africa, 13 September 2000.
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events
6
109
CONCLUSION
A methodology to retrieve the cloud properties specifically COT, ER (near cloud top) and CTT, is presented. A fundamental part of the method relies on a well-known algorithm (Nakajima and Nakajima 1995; Kawamoto et al. 2001). The methodology was adapted to the retrieval of cloud properties from MODIS data and to the near future application to MSG-SEVIRI measurements. Foreseen future applications could be the monitoring (GEO temporal and spatial scales) of cloud top microphysical/optical properties. The case studies revealed in some conditions a substantial decrease of the cloud effective radius when aerosol particles were present over the same area. However, it is not clear from the present results whether this effect is due to cloud-aerosol interactions or to the existence of an atmospheric aerosol layer separated from the cloud layer. This latter case can induce considerable errors while performing the satellite retrievals if not adequately modelled. Vertical profiling of the atmosphere can be very useful in these studies. The results obtained from the present methodology were compared against retrievals of the official MODIS cloud products and datasets of ER and CTH from aircraft measurements taken during the SAFARI 2000 campaign, in order to evaluate the performances of the present methodology and the possible effects of the aerosol-cloud interactions. The comparisons have pointed out an acceptable agreement with MODIS cloud parameters, whereas more complicated is the interpretation of the comparisons with the SAFARI 2000 data. It will be necessary to undertake an investigation of other case studies similar to those examined in this work together with a sensitivity analysis of the variations of the microphysical cloud properties in presence of biomass burning aerosol layers. These are necessary steps towards a better understanding of the aerosol-cloud interactions. Moreover, similar studies using MSG-SEVIRI data are planned, taking advantage of the high image frequency and considering the atmospheric aerosol layer. Acknowledgements: MODIS data were acquired as part of the NASA’s Earth Science Enterprise. The algorithms were developed by the MODIS Science Teams. The data were processed by the MODIS Adaptive Processing System (MODAPS) and Goddard Distributed Active Archive Center (DAAC), which also archives and distributes them. Aircraft cloud effective radius measurements of the UK Met Office C-130 aircraft are available for download at ftp://daac.ornl.gov/data/safari2k/atmospheric/ uk_met_c-130/ from the Oak Ridge National Laboratory (ORNL) Distributed Active Archive Center (DAAC), Oak Ridge, Tennessee, USA. Data of the ER-2 Cloud Physics Lidar (CPL) are available online at http://virl.gsfc.nasa.gov/cpl/ safari2000_pass.html from the NASA Goddard Space Flight Center, Greenbelt, Maryland, USA. The authors are grateful
M. J. Costa et al.
110
to the OpenCLASTR project (http://www.ccsr.u-tokyo.ac.jp/~clastr/) for making available the RSTAR (system for transfer of atmospheric radiation) and CAPCOM (Comprehensive Analysis Program for Cloud Optical Measurement) packages for use in this research. This study was funded by EURAINSAT (http://www.isac.cnr.it/~eurainsat/), a shared-cost project (contract EVG1-2000-00030) co-funded by the Research Directorate General of the European Commission within the research and technological development activities of a generic nature of the Environment and Sustainable Development subprogram (5th Framework Programme). Partial support from the Italian National Group for the Prevention from Hydrogeological Disasters (GNDCI) is acknowledged. One of the authors (MJC) was financially supported by the Subprograma Ciência e Tecnologia do 2o Quadro Comunitário de Apoio. Funding was also provided by the Portuguese Foundation for Science & Technology under grant POCTI/CTA/42917/ 2001.
7
REFERENCES
Ackerman, A. S., O. B. Toon, D. E. Stevens, A. J. Heymsfield, V. Ramanathan, and E. J. Welton, 2000: Reduction of tropical cloudiness by soot. Science, 288, 1042–1047. Albrecht, A. A., 1989: Aerosols, cloud microphysics, and fractional cloudiness. Science, 245, 1227–1230. Barnes, W. L., T. S. Pagano, and V. V. Salomonson, 1998: Prelaunch characterstics of the MODerate Resolution Imaging Spectroradiometer (MODIS) on EOS-AM1. IEEE Trans. Geosci. Remote Sens., 36, 1088–1100. Bréon, F.-M., D. Tanré, and S. Generoso, 2002: Aerosol effect on cloud droplet size monitored from satellite. Science, 295, pp 834–838. Kaufman, Y. J., J. M. Haywood, P. V. Hobbs, W. Hart, R. Kleidman, and B. Schmid, 2003: Remote sensing of vertical distributions of smoke aerosol off the coast of Africa. Geophys. Res. Lett., 30, 1831–1834. Kawamoto, K. and T. Nakajima, 2003: Seasonal variation of cloud particle size from AVHRR remote sensing. Geophys. Res. Lett., 30, 1810–1813. Kawamoto, K., T. Nakajima, and T. Y. Nakajima, 2001: A global determination of cloud microphysics with AVHRR remote sensing. J. Climate, 14, 2054–2068. Keil, A. and J. M. Haywood, 2003: Solar radiative forcing by biomass burning aerosol particles during SAFARI 2000: A case study based on measured aerosol and cloud properties. J. Geophys. Res., 108, doi: 10.1029/2002JD002315. King, M. D., S.-C Tsay, S. Platnick, M. Wang, and K. N. Liou, 1998: Cloud retrieval algorithms for MODIS: optical thickness, effective particle radius, and thermodynamic phase. ATBD Reference Number: ATBD-MOD-05. McClatchey, R. A., W. Fenn, J. E. A. Selby, F. E. Volz, and J. S. Garing, 1971: Optical properties of the Atmosphere, AFCRL-TR- 71-0279, Enviro. Research papers, No 354, Air Force Cambridge Research Laboratories, Hanscom AFB, Bedford, MA, 85 pp. McGill, M. J., D. L. Hlavka, W. D. Hart, E. J. Welton, and J. R. Campbell, 2003: Airborne lidar measurements of aerosol optical properties during SAFARI 2000. J. Geophys. Res., doi: 10.1029/2002JD002370. Nakajima, T. Y. and T. Nakajima, 1995: Wide-area determination of cloud microphysical properties from NOAA AVHRR measurements for FIRE and ASTEX regions. J. Atmos. Sci., 52, 4043–4059.
Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events
111
Nakajima, T. and M. Tanaka, 1986: Matrix formulation for the transfer of solar radiation in a plane-parallel scattering atmosphere. J. Quant. Spectrosc. Radiat. Transfer, 35, 13–21. Nakajima, T. and M. Tanaka, 1988: Algorithms for radiative intensity calculations in moderately thick atmospheres using a truncation approximation. J. Quant. Spectrosc. Radiat. Transfer , 40, 51–69. Rosenfeld, D., 1999: TRMM observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 3105–3108. Rosenfeld, D., 2000: Suppression of rain and snow by urban and industrial air pollution. Science, 287, 1793–1796. Rossow, W. B. and R. A. Schiffer, 1999: Advances in understanding clouds from ISCCP. Bull. Amer. Meteor. Soc., 80, 2261–2287. Schmetz, J., P. Pili, S. Tjemkes, D. Just, J. Kerkmann, S. Rota, and A. Ratier, 2002: An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc., 83, 977–992. Swap, R. J., H. J. Annegarn, J. T. Suttles, M. D. King, S. Platnick, J. L. Privette, and R. J. Scholes, 2003: Africa burning: A thematic analysis of the Southern African Regional Science Initiative (SAFARI 2000). J. Geophys. Res., 108, doi: 10.1029/2003JD003747. Torres, O., P. K. Bhartia, J. R. Herman, Z. Ahmad, and J. Gleason, 1998: Derivation of aerosol properties from satellite measurements of backscattered ultraviolet radiation: theoretical basis. J. Geophys. Res., 103, 17099–17110. Twomey, S., 1974: Pollution and the planetary albedo. Atmos. Res., 8, 1251–1256.
9 3D EFFECTS IN MICROWAVE RADIATIVE TRANSPORT INSIDE PRECIPITATING CLOUDS: MODELING AND APPLICATIONS Alessandro Battaglia1∗, Franco Prodi1,2, Federico Porcù1, and Dong-Bin Shin3 1
University of Ferrara, via Paradiso 12, Ferrara, Italy ISAC-CNR, via Gobetti 101, Bologna, Italy 3 School of Computational Sciences, George Mason University, Fairfax, VA, USA 2
Abstract
New rainfall techniques urge microwave physically based retrievals to produce an error estimate, to be more precise at the instantaneous level and to provide a correct geolocation of the raining areas. Within 3Dstructured clouds the coupling between horizontal and vertical inhomogeneity introduces additional uncertainties to instantaneous estimates because of the azimuthal dependence of the radiation field. In fact conical scanning microwave radiometers looking at the same point at the ground from different positions may measure quite different brightness temperatures. For a radiometric scene over mirror surfaces, this 3D radiative effect, averaged over a TMI-like footprint, can be fully described by 1D slant path models, except for strongly scattering highly developed raining cells at 85 GHz. A simulation radiative transfer study applied to Goddard Cumulus Ensemble Cloud Resolving Models, shows that the fore/aft view configuration, now available for some new generation sensors, may help in capturing features like tilted or different stage raining cells, emission peaks, asymmetric ice decks, especially in convective regions. This definitely leads to a better insight in the slant path cloud properties and to an improvement of Bayesian technique driven rain retrievals.
Keywords
Microwave radiative transfer, 3D effects, slant path
∗
presently at Institute of Meteorology, University of Bonn, Bonn, Germany
113 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 113–125. © 2007 Springer.
A. Battaglia et al.
114
1
INTRODUCTION
Microwave radiative transfer computations are increasing their relevance as the emphasis on rainfall “physically” based satellite remote-sensing retrievals (Shin and Kummerow 2003, and reference therein) increases. In fact, these approaches use statistical methods to match observed TBs with an appropriate TBs database, precomputed by forward radiative transfer (RT, hereafter) computations applied to properly selected datasets of atmospheric profiles. In this sense the word “physically” implies the attempt to model the dynamic, the microphysics and the radiative properties of the precipitating system under observation. This approach is double-edged: its strength resides in the possibility of evaluating how each single effects impacts the non linear TB– RR relationship and, thus of tempting an error estimate of the RR product; conversely its weakness resides in the chance that one of this complex modeling can be badly described, thus spoiling all the retrieval. Until recently the major part of MW rainfall products has been intended for climatological purposes; at this level temporal and spatial averaging washes out many uncertainties typical of the retrieval procedure, primarily all these effects connected to the 3D structure of clouds. To our knowledge all rain retrieval algorithms from multi-spectral microwave measurements have presumed the existence of plane parallel clouds, that allows fast RT forward computations. Only in a second moment many authors have investigated biases and random errors which are introduced when an inherently 3D problem is treated with a 1D solution (Kummerow 1998), but the most attention has been paid to 3D effects that induces a bias in the retrieval (the beam filling effect overall). In view of the next generation available sensor packages (MSG and those following in the GPM concept overall), a renewed impulse to instantaneous rainfall estimates has been done by blending techniques where quantitative MW algorithms train the geostationary rapid update cycle. In fact, over ocean (where we concentrate our analysis), thanks to their slicing capabilities, multispectral passive microwave techniques perform superiorly for instantaneous applications respect to VIS/IR techniques which generally infer precipitation only from cloud top information. While for temporally accumulated products large temporal sampling errors overcome spatial sampling effects, at the instantaneous level the coarse spatial resolution of PMW sensors is a critical issue. Therefore in order to achieve a real improvement in a correct location of the raining pixels (Bauer et al. 1998), in the discrimination rain/no rain, and in the estimate of the rain rate with a physically based error (bias + root mean square error), the interplay between the characteristics of the PMW observing system (like spatial resolution, viewing geometry, instrument frequency and polarization) and of the observed cloud system (actual geometry of the cloud, that is vertical and horizontal development and patch structures, vertical and horizontal inhomogeneity of scattering parameters) has to be deepened.
3D Effects in Microwave Radiative Transport Inside Precipitating Clouds
115
In this paper, after briefly reviewing the state of the art of some well assessed 3D issues, we will focus our attention on some other important 3D effects; in particular we will try to assess when, for a conical scanning radiometer, a 3D description is really necessary and which kind of radiative transfer approaches are best suited for different cases. Finally, as a step towards a better understanding of the 3D structure of clouds, we try to investigate and exploit the potentialities of the fore/after viewing radiometer configuration (already realized in Windsat and planned for EGPM radiometers).
2
3D CLOUD STRUCTURE: OBSERVATIONS AND MODELING
The cloud structure inhomogeneity readily observed by in situ and remote measurements (e.g., reflectance imagery, radar profiles) plays a key role in the evaluation of 3D effects in cloud analysis and precipitation retrieval. Convective systems usually present small scale variability of the horizontal structure, which extend in the vertical for several kilometers: studies over tropical regions show that only 5% of the convective updrafts have diameters larger than 4.3 km (Jorgensen and LeMone 1989), which is the smaller footprint available for passive microwave spaceborne sensors (TMI). Other studies found that the median convective cell size for summer storms near the mid-Atlantic coast of the USA is only 1.9 km (Goldhirsch and Musiani 1986). Moreover, for convective mesoscale-organized systems the horizontal asymmetry of the structure is also impacting the representativeness of a slant path view. Cloud systems as squall lines of the leading-line/trailing-stratiform type, common over the tropics and at mid-latitudes as well, has a leading line of contiguous convective cells, separated at scales of the cell size order, and a trailing stratiform cloud shield extending backwards of the leading line. The slant path observation of this kind of systems strongly depends on the viewing direction, and, since they can extend for few hundred kilometers, different parts of the system can be seen with different azimuthal angles. Usually stratified precipitating systems show a more symmetric and shallow 2D structure, that should be less sensitive to slant path observation. The precipitation structure in a typical mid-latitude frontal system, shows different small scale, elongated features (rain bands) of enhanced precipitation, embedded in the general frontal stratified precipitation pattern. Even deep convection can develop within stratified structures, especially if forced by orography, breaking the horizontal homogeneity at a spatial scale related to the topographic features. In these cases a spatial horizontal gradient of 12 h cumulated precipitation reaches 120 mm km–1 (Steiner et al. 2001). Slant path observation of inhomogeneous cloud system over complex terrain is even more difficult, given the variability of the orientation and emissivity of the background surface.
A. Battaglia et al.
116
Cloud resolving models have given great strides in generating fine-scale cloud and storm properties by providing vertical structure of the atmosphere and of the hydrometeors. Their role is therefore crucial in the modeling of the non linear relationship between hydrometeor concentration and surface RR and in the computation of scattering parameters that are needed to simulate the upwelling TBs. Here in the following, as test examples, we use some simulations from the Goddard Cumulus Ensemble Model (Tao and Simpson 1993), of a tropical squall line and of two mid-Atlantic fronts (warm and cold). At this stage, their resolution (2 km or 4 km in the horizontal and 0.5 km in the vertical) is the highest scale at which we can evaluate 3D radiative effects.
3
RT COMPUTATIONS
Many authors faced the problem of 3D RT computations for raining scenes: Roberti et al. (1994) reviewed and provided a clear intercomparison between different techniques. Compared to other ones (e.g., the VDOM, Haferman et al. 1997), the Monte Carlo methods, in its different versions (backward, Roberti et al. 1994; forward–backward, Liu et al. 1996; forward, Roberti and Kummerow 1999; Battaglia and Kummerow 2000), seems the more suitable to investigate MW RT through complicated 3D structures. In exchange for the simplicity, a fairly substantial computational penalty (reducible with biasing techniques, see Roberti and Kummerow 2000) is paid, but certainly manageable for off-line applications. However, in many cases the 3D structure of clouds does not preclude the use of 1D radiative transfer approaches. In fact the plane parallel assumption does not require homogeneity at distances arbitrarily far from the FOV of the sensing instrument (Platnick 2001). For instance for pure absorbing atmosphere and Fresnel-like surfaces, the radiation sensed by PMW radiometers originates just in the FOV projected slant tube. In these cases 1D approximations work very well, with the simple expedient of taking in account geometric effects in case of off-nadir looking radiometers (Liu et al. 1996). In presence of scattering and/or diffusive surfaces, radiation sensed at the satellite may not be originated in the slant tube of observation. In this case it occurs an horizontal displacement of radiation in the direction perpendicular to the viewing direction; this horizontal displacement can then be used to assess the scales over which horizontal inhomogeneities are important in remote-sensing problems. Thanks to its intrinsic tracing procedure, the Monte Carlo techniques makes it possible to record the statistics about the number of scattering events undergone by radiation, the source of the signal (thus the evaluation of the weighting functions), the fraction of radiation coming from outside the observing slant FOV, etc. This motivates why MC methods are believed to be the most adaptive tool to investigate 3D MW RT aspects.
3D Effects in Microwave Radiative Transport Inside Precipitating Clouds
117
In the following analysis we will adopt three different RT solutions: a backward Montecarlo, slightly modified from Roberti’s code, and the Eddington approximation (Kummerow 1993) in its 1D plane parallel (PP) and slant path (SP) version (similarly to that already done by Bauer et al. 1998, Roberti and Kummerow 1999, Liu et al. 1996). In these two approximations the structure is horizontally homogeneous while the vertical profile is reconstructed by using the vertical profile above the observation point at the ground for the PP, and the slant profile defined by the ray traced from the sensors downward to the surface and then mirror-reflected upward for the SP. Note that all calculated TBs are ascribed to the position at the ground the sensors is looking at.
4
3D EFFECTS AT MICROWAVE REGIME
The TBs sensed by PMW sensors are the result of the interplay between the intrinsic 3D structure of the clouds and the radiometer properties. Many different aspects of this interconnection are usually denoted 3D issues. For instance the mismatching between the coarse instrument FOV and the fine-structured cloud horizontal inhomogeneities leads to the beam filling problem. In case the radiometer FOV is not uniformly filled, the nonlinearity between retrieved physical parameters (like RR, SST, LWC, IWC) and measured TBs introduces both bias (for instance, an underestimation of the RR in case of rainfall homogeneous assumption across the FOV) and random errors. The way for correcting and estimating these errors has been faced by many authors and can be tackled in different ways (Kummerow 1998, and reference therein). Here in the following we avoid the beam filling complication (by assuming that the coverage of the rain column within the FOV is known from some other independent measurements) and, by focusing on RT calculations performed at the maximum available resolution (in our case the resolution of the CRM, 2 × 2 km2 or 4 × 4 km2), we single out those 3D radiative effects that are not reproducible with 1D approximations. We will briefly categorize them as geometric and diffusion effects.
4.1 Geometric effects At emission channels, upwelling radiation measured by PM radiometers with conical scan geometry (viewing angle Θv) results from absorption/emission processes along the whole slant beam. Therefore at low frequency channels the 3D RT problem reduces to a 2D one. As shown in literature (Weinman and Davies 1978; Liu et al. 1996; Roberti et al. 1994; Bauer et al. 1998) leakages coming directly from the warm side of the cloud or scattered by the Fresnel-like cold surface can be accounted for by 1D SP approximation but
118
A. Battaglia et al.
not by 1D PP modeling.1 For a fore looking radiometer observing a cloud system with height Hc the geometric effects broaden the radiometric signal of the cloud in a wider region 2Hc tan Θv before and Hc tan Θv after the physical boundaries of the cloud.
4.2 Diffusion effects At scattering channels, due to the redirectioning of radiation by diffusion events, the measured signal can originate from outside the geometrical slant beam, so that at this level the radiometer is receiving “photons” from a 3D region of the cloud, less defined than the 2D slant tube.
Figure 1. Average number of scattering events at three frequencies for different values of the EWP (left panel) and mean lateral displacement ∆s vs. average number of scattering at 85.5H GHz.
To exemplify, we show in Fig. 1, for the three different CRM simulations, the average number of scattering events (computed with the MC code at TMI footprint) as a function of the slant EWP (equivalent water path) that is the sum of the LWP and the IWP. The number of scattering events increases with the EWP and with the frequency (at 19.3 GHz the scattering becomes negligible) but it is almost independent of the cloud system under observation. To assess the importance of the 3D-diffusion effect we have computed ∆s⊥, the absolute horizontal displacement of radiation in the direction perpendicular to the viewing direction between the emission and the sensor position. In the right panel of Fig. 1 this quantity is shown as a function of the number of scattering events. In this case the relation strongly depends on the cloud system. For the same number of scattering events (thus the same 1
Small differences still remain because the intersection between the 3D cloud structure with the slant beam does not result in homogeneous plane parallel layers: in fact the SP approximation does not perfectly accomodate cloud edges!
3D Effects in Microwave Radiative Transport Inside Precipitating Clouds
119
EWP), the TCOF22, that simulates a squall line system with a very high spatial variability of scattering properties nearby the convective raining cells, have a much greater ∆s⊥ than the cold and warm front case studies. In fact these two last raining systems do have a lower vertical development (freezing level height is nearly half the value of the squall line) and therefore with the same slant EWP they have a much higher hydrometeor content, hence larger extinction coefficients; this does not allow radiation to travel as much as for the squall line. Note that when the ∆s⊥ becomes comparable with the horizontal resolution of the radiometer the 3D diffusion effects starts being relevant and the SP-1D being quite bad an approximation. For TMI-like radiometers, for the simulations of our database, this is true only at the 85.5 GHz. As already noticed by Kummerow (1998) and consistently with earlier findings by Roberti et al. (1994), the 1D modeling, compared to the 3D correct one, introduces a rms more than a bias error. In fact in 1D model approximations, radiation remains trapped by construction in the slant tube:
Figure 2. Average ∆s ⊥ as a function of the ∆T B computed between MC and 1D SP approximation at TMI resolution for the H-85.5 GHz. The underlying gray image gives the probability density for these two variables for the analyzed database.
A. Battaglia et al.
120
no contribution from outside the tube is allowed and abrupt not physical variations may happen for contiguous pixels. 3D MC radiation field can be considered a sort of redistribution of the 1D radiation pattern. The computed ∆TB = TB ( MC ) − TB (1D − SP ) shows consistent differences (>10 K) at TMI-like resolution only at the highest frequency (85.5 GHz). Areas with positive ∆TB are closely followed by areas with negative ∆TB, thus confirming the redistribution conjecture. Figure 2 shows the difference ∆TBs for the whole database at 85.5 GHz, and its strict correlation to the ∆s⊥. Obviously the presence of consistent ∆s⊥ (comparable with the instrument FOV) has to couple with an horizontal inhomogeneity of scattering properties to give appreciable differences between 1D and 3D models. Otherwise if these two conditions are not simultaneously satisfied (for instance in a homogeneous thick ice deck or in presence of a strong absorbing inhomogeneous water vapor field) no real 3D effects are really detectable. As a prospective for better instantaneous RR retrievals, the 1DSP (not the 1D-PP) model partially accommodates for the geometrical problems due to surface scattering and oblique viewing angle but still cannot take into account diffusion effects due to cloud scattering. Therefore it reproduces TB quite accurately for emission dominated frequencies (10 and 19 GHz, even at CRM resolution) but not at higher frequencies. If convolved to TMI resolution the result remains acceptable for 37.0 GHz but not for 85.5 GHz. Therefore, this last frequency (and, a fortiori, higher MW frequencies) has to be treated more carefully by performing dedicated 3D RT simulations especially with high EWP and in coincidence with highly vertically developed and strongly horizontally inhomogeneous raining systems.
5
FORE/AFT RADIOMETER
From above considerations, we can deduce that TMI-like radiometers sense primarily the 2D slant-tube structure of the cloud. Recently launched (US Navy Coriolis) and future planned spacecrafts (EGPM drone) host or are going to host conically scanning space-borne MW radiometers able to provide fore/aft views of the swath, a novelty with respect to imagers like SSM/I and TMI. In this configuration, after a revisit time of about 5 min (for typical LEO satellite), the radiometer is potentially looking at the same2 scene from the back instead of from the front side. Besides well established
2
It can be assumed that the potential precipitating systems under observation have not significantly changed during the revisit time, at least over current PMW resolution.
3D Effects in Microwave Radiative Transport Inside Precipitating Clouds
121
benefits,3 it has to be evaluated whether or not the use of additional information from backward direction (hence a double number of TBs) can improve the current MW algorithm in term of rain/no rain discrimination, geolocation of raining pixels (like in Bauer et al. 1998), retrieval of rain intensity and hydrometeor profile. This goal is here pursued by a simulation study with fore/aft TB s computed from the former CRM database on a TMI-like radiometer. Because of different optical paths in 3D-structured clouds, different TBs are measured when observing the same pixel at the ground from supplementary directions as shown in Fig. 3. This allows to reveal features and information that cannot be picked up by using one view only. For instance, in Fig. 3, the comparative study of fore and aft minima patterns reveals that, sometimes, fore minima (labeled by Bi ) are followed by an aft minimum (labeled by Ai) with the same low temperature. The horizontal distance between correspondent minima, through geometric considerations, can be used for a stereographic reconstruction of the altitude of the high concentration ice area causing the TB depression [i.e., around 11 km and 9 km for minima (A1,B1) and (A2,B2), 30 and 24 km horizontally separated, respectively]. Other times, due to occultation of emission or scattering areas by close different altitude clouds or tilted systems (Hong et al. 2000), different pattern in the structure of fore/after TBs are found. In Fig. 3 while
Figure 3. Cross section for the extinction coefficient and the simulated fore/after TB profiles at 85.5H GHz.
3
Primarily the potential to reduce radiometric noise (particularly relevant for
polarimetry) by a factor 2 through remapping to a common grid and averaging, and to check the consistency of the wind retrievals.
122
A. Battaglia et al.
Figure 4. Mean RR in the fore/after TB plane at 19.3H GHz (left panels) and 85.5 H GHz (right panels). Top (bottom) panels correspond to the GCE TCOF22 (MIDACF) simulation (see also color plate 5).
the fore radiometer detects only one minimum (B3) the after radiometer detects three minima (A3, A4 and A5). This indicates the presence of three side by side convective cells, characterized by different altitude. The first cell (located around y = 175 km) develops higher in the troposphere and obscures the other two, when seen by a forward radiometer. Similarly the side emission peak, originated by very warm emission from the side of the cloud, whether present in one view is not present in the other (in Fig. 3 P1 has no corresponding maximum). A systematic analysis of the simulated fore/aft TB database shows that the differences in TB s are higher and spread out more around the convective cells for higher frequencies (in Fig. 4 compare right with left panels) and for taller systems (compare top with bottom panels); the higher the RR and the frequency the lower the correlation between fore/aft TBs. TBs have been grouped in classes 2 K-wide and for each pixel mean rain rate RR and standard deviation σ(RR) have been computed for the different TMIfrequency. The error structure for the fore/after view is substantially better than the error structure of a pure fore configuration, especially at high
3D Effects in Microwave Radiative Transport Inside Precipitating Clouds
123
frequency. To have a very rough index of the improvement that the fore/aft configuration could provide versus the single and double looking TMIradiometer, the total std for the RR is computed for each simulation and is shown in Table 1. At 10 GHz the fore/aft configuration adds practically no information, while the improvement seems to increase at higher frequencies. Note that at 19 and 37 GHz, the two channels more correlated to RR, the best results of the double viewing are obtained for the TCOF22 system. Table 1. Quantification of the total stde for the RR computed for different TMI frequency. Here we consider horizontally polarized channel because they are more sensitive due to the lower surface emissivity.
freq [GHz] 10.7H (1 // 2TB) 19.3H (1 // 2TB) 37.0H (1 // 2TB) 85.5H (1 // 2TB)
σ(RR) {1 // 2TB} [mm h-1] TCOF22 MIDACF 0.35 // 0.35 0.38 // 0.37 0.82 // 0.73 0.85 // 0.81 2.20 // 1.25 3.21 // 2.22 4.30 // 3.30 5.04 // 3.65
MIDAWF 0.22 // 0.22 0.27 // 0.27 0.62 // 0.44 1.99 // 0.87
5.1 Retrieval results To evaluate whether the observations at two different viewing angles really improve the performance of rainfall estimation we have performed synthetic retrievals based on a simple Bayesian approach (Shin and Kummerow 2003). Figure 5 shows the scatter plots of the true and retrieved rain rates for the three different CRM simulations and the two TMI-like radiometers. The first radiometer is assumed to have only a forward view ( φ = 90º) and the other one have double views (forward and backward, φ = 90º + 270º). Rain retrieval statistics such as rain bias, RMS statistics and correlation are also represented for each experiment. A similar result (not shown) has been found for integrated water content. The sensor design having forward and backward scans turns out to perform better than the sensor with single view for two CRM simulations, the Tropical squall line and the mid-Atlantic cold front. On the other hand no improvement at all is found for the mid-Atlantic warm front simulation. For this simulation rain rates are typically lower than for the other two: the synthetic retrieval with only one viewing angle is already pretty good. Therefore rainfall retrieval improvements with fore/aft viewing angles seem to be more relevant in high developed 3D-structured clouds with high rain rates. For these structures the presence of ice cores, uncorrelated with rainfall underneath, can mislead one viewing angle only radiometer, while in a less extent a fore/aft viewing radiometer.
124
A. Battaglia et al.
Figure 5. Scatter plot of true and retrieved rain rates from synthetic retrievals for the two radiometers ( φ = 90 forward, and φ = 90+270 forward and backward viewing) and the three CRM simulations.
3D Effects in Microwave Radiative Transport Inside Precipitating Clouds
6
125
CONCLUSIONS
The fore/aft view capability of new PMW sensors enhances the reliability of a 3D cloud observation, and as a consequence, improves the retrievals of rain-rate and path-integrated quantities and allows the identification of cloud features that, otherwise, can lead to errors in rain/no rain discrimination and rain estimation. To test the significance of the fore/aft viewing angle concept, experimental data from air-borne radiometers with along track scanning radiometers and from space-borne radiometers with fore/after view (like Windsat) have to be analyzed and compared with independent observations. This will finally lead to applications in retrievals for drone GPM satellites and airborne high resolution scanning radiometers.
7
REFERENCES
Battaglia, A. and C. Kummerow, 2000: Forward Montecarlo computations of polarized microwave radiation. Proc. 13th Int. Conf. on Clouds and Precipitation, Reno Area, Nevada, USA, 256–259. Bauer, P., L. Schanz, and L. Roberti, 1998: Correction of three dimensional effects for passive microwave remote sensing of convective clouds. J. Appl. Meteor., 37, 1619–1632. Goldhirsh, J. and B. Musiani, 1986: Rain cell size characteristics derived from radar observations at Wallops Island, Virginia. IEEE. Trans. Geosci. Remote Sens., GE-24, 947–954. Haferman, J. L., T. F. Smith, and W. F. Krajewski, 1997: A multi-dimensional discrete ordinates method for polarized radiative transfer. I. Validation for randomly oriented axisymmetric particles. J. Quant. Spectr. Radiat. Transfer, 58, 379–398. Hong, Y., J. L. Haferman, W. S. Olson, and C. D. Kummerow, 2000: Microwave brightness temperatures from tilted convective systems. J. Appl. Meteor., 39, 983–998. Jorgensen, D. P. and M. A. LeMone, 1989: Vertical velocity characteristics of oceanic convection. J. Atmos. Sci., 46, 621–640. Kummerow, C., 1993: On the accuracy of the Eddington approximation for radiative transfer in the microwave frequencies. J. Geophys. Res., 98, 2757–2765. Kummerow, C., 1998: Beamfilling errors in passive microwave rainfall retrievals. J. Appl. Meteor., 37, 356–370. Liu, Q., C. Simmer, and E. Ruprecht, 1996: Three-dimensional radiative transfer effects of clouds in the microwave spectral range. J. Geophys. Res., 101(D2), 4289–4298. Platnick, S., 2001: Approximations for horizontal photon transport in cloud remote sensing problems. J. Quant. Spectr. Radiat. Transfer, 68, 75–99. Roberti, L. and C. Kummerow, 1999: Monte Carlo calculations of polarized microwave radiation emerging from cloud structures. J. Geophys. Res., 104, 2093–2104. Roberti, L., J. Haferman, and C. Kummerow, 1994: Microwave radiative transfer through horizontally inhomogeneous precipitating clouds. J. Geophys. Res., 99, 16, 707–716. Shin, D. B. and C. D. Kummerow, 2003: Parametric rainfall algorithms for passive microwave radiometers. J. Appl. Meteor., 42, 1480–1496. Steiner, M., J. A. Smith, M. L. Baeck, Y. Zhang, and R. A. Houze, Jr., 2001: Space-time variability of heavy orographic rainfall. Prepr. 30th Int. Conf. on Radar Meteorology, Munich, 19–24 July, American Meteorological Society, 527–529. Tao, W. K. and J. Simpson, 1993: Goddard cumulus ensemble model. Part I: Model description. Terrest. Atmos. Oceanic Sci., 4, 35–72. Weinman, J. A. and R. Davies, 1978: Thermal microwave radiances from horizontally finite clouds of hydrometeors. J. Geophys. Res., 83, 3099–3107.
10 CLOUD MICROPHYSICAL PROPERTIES FROM REMOTE SENSING OF LIGHTNING WITHIN THE MEDITERRANEAN Claudia Adamo1, Robert Solomon2, Carlo M. Medaglia1, Stefano Dietrich1, and Alberto Mugnai1 1
ISAC-CNR, via Gobetti 101, Bologna, Italy USDA Forest Service, Pacific Wildland Fire Sciences Laboratory, Seatlle, WA, USA
2
Abstract
The joint use of concurrent data from the Precipitation Radar (PR) and Lightning Imaging Sensor (LIS) instruments, on the Tropical Rainfall Measuring Mission (TRMM) satellite, provides a unique means to investigate storm characteristics and to make assessments about the relationship between convection and electrification. We discuss some results derived from the observation of the vertical structure of precipitating clouds in the southern Mediterranean during a 5-month period. In this study, we find there is a strong, differentiable relationship between convective and stratiform systems that produce lightning.
Keywords
Cloud microphysics, remote sensing, lightning
1
INTRODUCTION
Criteria for characterizing cloud types based on lightning flash rate can be used to augment the identification of precipitation systems via VIS-IR geostationary satellites. Geostationary satellites, such as METEOSAT, allow rapid measurements of the space-time development of cloud systems and precipitation but are limited in their ability to distinguish precipitation types as they measure the upward directed radiation from the top of cloud systems. Microwave measurements are able to give a much better picture of the microphysical structure with the clouds although such platforms do not currently provide continuous coverage of precipitation systems, as do geostationary satellites.
127 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 127–134. © 2007 Springer.
128
C. Adamo et al.
However, an important consideration neglected by many cloud microphysical retrieval algorithms is the choice between utilizing those based on stratiform, or lower rain rate, systems and convective, or high rain rate, systems (especially for very severe storms). Climatologically based algorithms tend to be optimized for the lower precipitation systems. For this, a lightning based criteria has the potential to better discriminate between stratiform and convective cloud regimes. Thunderstorm electrification/lightning within clouds has a strong microphysical origin. Using this link between electrification and the occurrence of lightning, we present a preliminary study about the remote sensing of electrified clouds. The dominate source of in-cloud charge is the result of colliding ice particles. Rimed ice particles rise and grow within the convective updraft. Collisions between these rimed particles and with smaller ice particles rising in the updraft result in the larger particles taking one charge while the smaller ice takes the opposite. The sign and amount of charge is a strong function of the liquid water content, particle size and temperature (e.g., Saunders et al. 1994). The region in which the charge transfer takes place is delineated by approximately the –5°C and –25°C isotherms and is referred to as the charging zone. Larger particles such as hail and graupel remain suspended in the updraft and fall out when their terminal velocity exceeds the updraft while the lighter ice particles are lofted to the upper regions of the cloud establishing an electric field within the cloud. Based on observations (Marshall et al. 1995a, b), if the in-cloud electric field is greater than a threshold (approximately 100–250 kV m–1) lightning is initiated. Modeling results illustrate the relationships between lightning and microphysical parameters such as the maximum updraft velocity at the charging zone boundary and the peak liquid water flux into the charging zone (Solomon and Baker 1998) and with the ice crystal concentration in the charging zone (Schroeder 2000). These studies show the strong dependence of lightning flash rate on ice content, updraft velocity and liquid water. In the next section we briefly introduce the two TRMM sensors which are used in this study. The subsequent sections contain some preliminary results, our comments and conclusions.
2
REMOTE SENSING OF THUNDERSTORMS
The TRMM satellite has five instruments, each designed to observe different cloud properties. In particular we will make use of data from the Precipitation Radar (PR) and from the Lightning Imaging Sensor (LIS).
Cloud Microphysical Properties from Remote Sensing of Lightning
129
2.1 Lightning Imaging Sensor LIS is an optical sensor that detects lightning by looking for small changes in light intensity and is capable of observing lightning during daylight hours with a detection efficiency of greater than 90% (Christian et al. 1999) while viewing a particular location for approximately 90 s with a spatial resolution of 3–6 km. The full diurnal cycle is covered every 80 days at each location by LIS. We are interested in observing storms over the Mediterranean basin, but as the TRMM platform can only observe regions as far north as 37° N we focus on the southern part of the Mediterranean sea, from latitude 27° N to 37° N and longitude 4° E to 24° E. The individual variations in light intensity that are recorded by LIS are called events. If several events occur during the same 2 ms time frame and are also adjacent to each other, the events are bunched together to form a group. Groups that are temporally and spatially close to each other are considered to constitute a lightning flash. So, each lightning flash can be composed of many groups and up dozens (if not hundreds) of individual events.
2.2 Precipitation Radar The PR is the first space borne, active, range gated radar having a swath width of 215 km and a resolution of 4.3 km at the nadir and a vertical resolution of 250 m. The PR can measure the rain from the ground to an altitude of 20 km with a sensitivity greater than 0.5 mm h–1. For the purpose of this study, the LIS and PR data (specifically for the PR those provided from the 2A25 GSFC-DAAC TRMM data product) are combined by co-locating individual events and flashes with the corresponding PR attenuation corrected reflectivity profile for data between May and September 2000. That is, the location of each event and flash is mapped to the gridded PR data to determine the radar reflectivity and rainfall rate where that particular event or flash occurred. Convective and stratiform rainfall events are determined from the 2A25 data product.
3
RESULTS AND COMMENTS
Table 1 gives the total number of events and flashes associated with convective, stratiform and other (undetermined/no precipitation) precipitation regimes during the 5-month period studied. The vast majority of both events and flashes are associated with convective precipitation. However, the 30,055 events fall into only 761 unique PR profiles. Similarly, for the events associated with stratiform precipitation, 896 unique PR profiles are defined by the 6,459 events. There is a similar trend with the lightning flash obtained profiles. This is not completely unexpected as convective cells, in general, tend to be much more
C. Adamo et al.
130
electrically active than stratiform clouds. Figure 1 collects the attenuation corrected reflectivity and rain rates for the PR grid points with convective (upper panel) and stratiform events (lower panel). Table 1. Number of events and flashes.
Cloud type
Events [#]
Flash [#]
Convective Stratiform Other
30,055 6,459 5,957
984 134 115
Percent of total Event Flash 71 80 15 11 14 9
Figure 1. Averaged PR reflectivity (dBZ) (solid line – left panels) and standard deviation (dotted line) of all profiles in which convective (upper panel) or stratiform (lower panel) lightning events occurred. Right panels show the average rain rate (mm h–1) for the same set of PR profiles for convective (upper) and stratiform (lower pair) lightning events and standard deviations (dotted lines).
The power averaged reflectivity is given by the solid line (left panels), standard deviation by the dotted line and average rain rate (right panels) in mm h–1, solid line, and standard deviation, dotted line, for the different regimes. There is a marked difference between the convective and stratiform profiles. The reflectivity is substantially higher throughout most of the vertical extent and the rainfall rate is much higher, especially near the surface, for those PR profiles which contain a convective event.
Cloud Microphysical Properties from Remote Sensing of Lightning
131
However, the standard deviation, especially for the rain rate profiles, is quite large and in essence, prevents one from determining rain-rate information from lightning alone. Figure 2 is the same as Fig. 1 but looking at those profiles that contain at least one flash. The same trends are observed with these profiles as well.
Figure 2. Same as Fig. 1 but with PR profiles containing at least one lightning flash.
The results show that the reflectivity profiles for convective and stratiform clouds for those coincident with at least one lightning flash and/or event are quite different in both the maximum reflectivity, structure and vertical extent of the reflectivity illustrating two different microphysical regimes. The precipitation is also, expectedly so, quite different. In Fig. 3, we show four panels in which the profiles have been divided for different event rates. They show changes in the cloud structure for different values of event rates. With increasing event rate, the maximum reflectivity tends to increase (somewhat in magnitude but also in the vertical extent of higher reflectivity) for both convective and stratiform clouds while the two remain to be quite distinct from each other, providing a useful means of quantifying the cloud structure for each. Similarly, comparing the number of events with the maximum reflectivity we see that convective clouds are associated with a greater maximum reflectivity than those in stratiform
132
C. Adamo et al.
Figure 3. The average profiles of reflectivity have been divided into different event rates (number of events per minute per square kilometer [min–1 km–2)].
clouds. Again, looking at flash rate (Fig. 4) we see a very pronounced change in the vertical structure of the clouds. However, due to the large standard deviations associated with the profiles, we can only state that they are statistically different in the lower levels of the clouds. Some of our inability to differentiate properties within the upper levels of the clouds may be due to the PR’s 17 dBZ measurement threshold. Regarding the precipitation, we are unable to say anything with certainty; although, qualitatively they are strikingly different. The average profiles of reflectivity obtained from an PR-LIS study on the Mediterranean Sea for data from May to September 2000, selecting the days in which there were events with lightning, showed increasing event rates, the maximum reflectivity tended to increase (somewhat in magnitude but also in the vertical extent of higher reflectivity) for both convective and stratiform clouds while the two remain to be quite distinct from each other, providing a useful means of quantifying the cloud structure for each. Similarly, comparing the number of events of the analyzed storm with the maximum reflectivity, we saw that convective clouds are associated with a greater maximum reflectivity than those in stratiform clouds.
Cloud Microphysical Properties from Remote Sensing of Lightning
133
Figure 4. The average profiles of reflectivity have been divided for flash rates. Solid and dashed line are for convective and stratiform events respectively. Less than 1[min km2]–1 in the left panel, and greater than 1 [min km2]–1 in the right panel.
4
CONCLUSIONS
There is a strong relationship between microphysical properties and lightning in both convective and stratiform clouds. Although, it is well known that there are differences between convective and stratiform regimes, there is a lack of direction when choosing between microphysical profiles using microwave retrieval algorithms. The preliminary results presented here, however, illustrate that lightning may be used to help in this regard by distinguishing between different microphysical characteristics. 1. Flash and/or event rates within convective and stratiform clouds can give a measure of the microphysical structure of electrified clouds. However, due to large values of standard deviation it is only possible to state with some certainty, mass properties in the lower levels of the lightning producing clouds. 2. Although LIS allows for the detection of intra-cloud and cloud-toground lightning, its limited temporal resolution, i.e., it only observers a specific region for approximately 90 s, does not permit an analysis of storm evolution which may prove vital for effective storm characterization. This topic is in its infancy, mainly due to the lack of quantitative correlation studies based upon a reliable contemporary data set. However, the advantage of the satellite perspective is enormous; on one side there is the capability of registering intra-cloud (IC) strokes. In such a context the satellite perspective has the advantage to provide more information. Also, the inclusion of ground based lightning detection can provide a constant coverage of electrical/microphysical evolution. This in turn can be used to help create a database of microphysical properties via numerical modeling to be used for a multisensor retrieval algorithm (Dietrich et al. 2000).
C. Adamo et al.
134
Acknowledgments: This study has been funded by the Italian National Group for Prevention from Hydro-Geological Disasters (GNDCI), by the Italian Space Agency (ASI), and within the frame of EURAINSAT, a shared-cost project (contract EVG1-2000-00030) co-funded by the Research DG of the European Commission within the RTD activities of a generic nature from the Environment and Sustainable Development sub-program (5th Framework Program). This work is part C. Adamo PhD Thesis (Adamo, 2004).
5
REFERENCES
Adamo, C., 2004: On the use of lightning measurements for the microphysical analysis and characterization of intense precipitation events over the Mediterranean area. PhD thesis, University of Ferrara, Ferrara, Italy. Christian, H. J., R. J. Blakeslee, S. J. Goodman, D. A. Mach, M. F. Stewart, D. E. Buechler, W. J. Koshak, J. M. Hall, W. L. Boeck, K. T. Driscoll, and D. J. Boccippio, 1997: The Lightning Imaging Sensor. Proc. Int. Conf. on Atmospheric Electricity , 46, 749–760. Dietrich, S., C. Adamo, A. Mugnai, and R. Bechini: A physical profile-based multisensor precipitation retrieval. Proc. 2000 EUMETSAT Meteorological Satellite Data Users’ Conf., Bologna, 29 May–2 June, 2000, EUM P 29, 443–449. Marshall, T. C., W. D. Rust, and M. Stolzenburg, 1995a: Electric structure and updraft speeds in thunderstorms over the Southern Great Plains. J. Geophys. Res., 100, 1001–1015. Marshall, T. C., M. McCarthy, and W. D. Rust, 1995b: Electric field magnitudes and lightning initiation in thunderstorms. J. Geophys. Res., 100, 7097–7103. Saunders, C., 1994: Thunderstorm electrification laboratory experiments and charging mechanisms. J. Geophys. Res., 99, 10773–10784. Schroeder, V., 2000: How does lightning initiate and what controls lightning frequency? PhD thesis, University of Washington, Seattle, WA, USA. Solomon, R. and M. B. Baker, 1998: Lightning frequency and type in convective storms. J. Geophys. Res., 101, 14983–14998.
11 THE WORTH OF LONG-RANGE LIGHTNING OBSERVATIONS ON OVERLAND SATELLITE RAINFALL ESTIMATION Emmanouil N. Anagnostou and Themis G. Chronis Civil and Environmental Engineering, University of Connecticut, Storrs, CT, USA
1
INTRODUCTION
Lightning is a physical phenomenon that was excessively deified and feared for centuries. Even in the progressive minds of the ancient Greek society lightning was considered as the inescapable wrath of Zeus, seeking to punish mankind. It was in the dawn of the 1750s when Benjamin Franklin, the American philosopher and inventor, brought the lightning phenomenon to human proportions by relating it with atmospheric electricity. Years later, in the beginning of the 20th century C.T.R. Wilson (inventor of the Cloud Chamber) first used electric field measurements to estimate the structure of thunderstorm charges involved in lightning discharges (Krider 1996). Besides studies related to thunderstorm evolution, other aspects of lightning were addressed. For example, the propagation of electromagnetic waves excited by lightning strokes in the Very Low Frequency band (VLF or sferics) stimulated the interest of many researchers such as Bremmer (1949), Budden (1951) and Schumann (1954). Along the same lines, Pierce (1977) described the physics of sferics based upon work conducted after World War II. Lightning has strong relevance to numerous scientific fields ranging from atmospheric physics/chemistry to water and energy cycle. In the 1990s, NASA prioritized particular interest on lightning research giving it a hydrological applications dimension (Goodman 1988). This effort was further supported by the launch of the Tropical Rainfall Measuring Mission (TRMM) satellite (Simpson et al. 1988) – the first of its kind to carry on the same platform precipitation radar (PR), a multifrequency radiometer, and a Lightning Imaging Sensor (LIS) (Christian et al. 1999). 135 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 135–148. © 2007 Springer.
136
E. N. Anagnostou and T. G. Chronis
In this paper we will discuss the connection of lightning with convective precipitation processes, and present the advancements that lightning data can offer on the precipitation estimation. In the subsequent paragraphs we will discuss the physical properties of lightning formation, and the microphysics of electrified convective systems. We will highlight the current state of the art on continuous lightning monitoring from ground-based radio receivers and illustrate the physical consistency of lightning with passive microwave observations. We will conclude by describing current techniques on combining satellite and lightning data aiming at the improvement of highfrequency precipitation estimation over large regions.
2
PHYSICS OF ELECTRIFIED CLOUDS
Lightning is an electrical manifestation of thermodynamic and mechanical work performed by storm updrafts. Updrafts determine the supply, growth and transport of water condensate to the upper regions of storms, and directly control the dynamics of charge separation that lead to lightning. Lightning is related to the cloud microphysics with the presence of separated (positive and negative) electrical charges inside a thundercloud (Williams et al. 1989). A regular lightning charge occurs when the potential built-up inside a cloud has exceeded the breakdown threshold of the surrounding air, which value varies around 200 kV m–1 as a function of atmospheric humidity. The fact that clouds can develop electrical fields is far from surprising. Inside the clouds, the density of suspended material is increased to a significantly higher level than the surrounding environment so eventually any dielectric material available can receive charge trough common turboelectric effects (piezoelectric, thermoelectric, etc.). This is also stated as the convective hypothesis, according to which there is a discontinuity of electrical conductivity between the dry surrounding air and the saturated cloud, where free moving ions are captured (Ziegler et al. 1991). Laboratory studies have identified originally uncharged ice particles that acquired substantial electrical charge without the aid of an external electrical force (Takahasi et al. 1999). One of the key mechanisms proposed is the so-called charge transfer and its thermodynamic implications during collision between vapor-grown ice particles and hail at the presence of super-cooled droplets. Regarding the polarity of the charge, this is dependent on the temperature and liquid water content. Solomon and Baker (1998) portrayed an interesting representation of the charge that hail receives during these collisions and its dependence on liquid water content (hail is charged positively or negatively depending upon the environment in which it grows). A nonconductive charge separation has also been proposed based on a thermoelectric process. In this case the charge is transferred while for instance the outer surface of a hail droplet melts while the interior is still frozen. According to Simpson and Scarse (1937) who conducted pioneering research on thunderstorm
The Worth of Long-range Lightning Observations on Overland Satellite Rainfall
137
electrification by examining numerous in-cloud electric field profiles, a typical thundercloud has three distinctive charged regions: an upper positive charged region at about –30° C, a middle negative at about –10° C and a mixed positive and negative at around 0° C isotherm. There are two major categories of electrical discharges: a lightning type that discharges via two opposite charges inside the same or different cloud, is the so-called Intra Cloud (IC) strikes, a group that represents almost 80% of the worldwide lightning activity. Conversely discharges that are channeled from a point inside the cloud to the ground are called cloud-toground (CG) and represent the minority of such electrical activity yet they can release energy, several orders of magnitude higher that the IC discharges (up to 1015 joules). A typical CG would be channeled from a negatively charged lower cloud part to a positive ground. In the same manner, depending upon the structure and physical location of the charges inside the cloud, one may well encounter the formation of a positive lightning strike (positive cloud base to negative earth). A general consideration regarding the ratio of IC to CG lightning strikes, as Boccipio et al. (2001) state, shows a seasonal variation across the continental USA, yet having an average value of 0.9. There are other lightning categories that fall outside the scope of this paper; nevertheless, the interested reader can browse through other impressive expressions of atmospheric electricity as the “elves”, sprites, “blue-jets”, etc. (Lyons et al. 2003).
3
LONG-RANGE LIGHTNING DETECTION
Lightning over larger areas (continental to global scale) can be also detected on a continuous basis from ground radio receivers operating at the Very Low Frequency (VLF else known as atmospherics or sferics) (Chronis and Anagnostou 2003; Anagnostou et al. 2002; Lee 1986a, b). Lee (1986a, b) was first to show that sferics receiver networks could achieve large regional coverage due to the relatively low attenuation of the lightning-excited electromagnetic signal. He developed such a system with receivers deployed originally in the UK, Cyprus and Gibraltar. The University of Connecticut and National Observatory of Athens recently deployed a second-generation sferics network (named Zeus) covering Europe and African continents, and to a lesser accuracy West Asia and the Atlantic and Indian oceans. Zeus system, built by Resolution Displays Inc., makes use of advanced computing technology, signal processing algorithms, GPS and satellite communications networking to improve the state of the art in receiver design at long-range frequencies. Details about the system and its long-range locating error characteristics are provided in Chronis and Anagnostou (2003) and in Zeus web page (http://sifnos.engr.uconn.edu). The authors have demonstrated a mean locating error of about 15 km within the network periphery, while at
138
E. N. Anagnostou and T. G. Chronis
very long ranges (>5,000 km) the locating error exhibited a larger variability with a mean at around 150 km.
4
PHYSICAL CONSISTENCY OF LIGHTNING AND MICROWAVE OBSERVATIONS
A number of studies have targeted the relation between microwave properties of precipitating particles and electrification. The majority of these investigations have resulted in empirical relationships between lightning intensity (e.g., flash rate) and convective parameters (such as updraft velocity, precipitation, ice water content). Toracinta et al. (2002) for example used lightning frequency information to classify reflectivity profiles from TRMM PR. Along the same lines, Cecil and Zipser (1999) studied the radiometric and electrification properties of well-formed hurricane eye walls while Chang et al. (2001) combined synoptic meteorological observations with passive microwave and lightning observations to study the 1998 Groundhog Day storm over Florida.
Figure 1. Rays of Tb85, CG flash rate and rainfall rate along a cross section of an MCS in Oklahoma.
The Worth of Long-range Lightning Observations on Overland Satellite Rainfall
139
As already stressed out, lightning formation is a strong indication of convective activity and the presence of ice particles. In this section we will draw a parallel between these physical properties and microwave observations of rainfall. In particular, we will use a number of independent observations to show the correlation between lightning activity and convective structure. Our analysis is based on four months of data (May through August 2002) originating from a large area in the US Southern Planes [1030 W to 930 W and 330 N to 370 N]. The data include (a) high frequency (85 and 37 GHz) passive microwave brightness temperatures from three Special Sensor Microwave/Imager (SSM/I) sensors onboard F13, F14, and F15 defense meteorological satellites, (b) CG lightning location obtained from the U.S. National Lightning Detection Network (NLDN), and (c) hourly rain-rate fields from stage III (i.e., rain gauge calibrated) WSR-88D precipitation products. The SSM/I overpass represents a time window of ~3 minutes, while the corresponding flash density corresponds to ±½-hourly accumulations of CG flashes around the SSM/I overpass time, and the hourly WSR88D rain rates are selected from the hour that contains the SSM/I overpass time. The common spatial scale of the collocated data is that of the 85 GHz frequency having the coarser resolution (15 km × 17 km). Initially the WSR-88D rain rate fields were available at 4-km resolution, while NLDN lightning data are associated with a location error below 1 km in the above area (Cummins et al. 1998).
Figure 2. Scatter plot of mean CG flash rate for different categories of 85 GHz (left panel) and 37 GHz (right panel) brightness temperatures.
The collocated data were classified into two main categories, pixels having (or neighboring to pixels with) nonzero flash density (electrified pixels) and those associated with zero flash density (nonelectrified). We make use of this distinction in order to examine the passive microwave scattering properties of electrified versus the nonelectrified areas in a precipitation
140
E. N. Anagnostou and T. G. Chronis
regime. The data analysis resulted in about 30,000 (9,000) nonelectrified (electrified) SSM/I pixels associated with precipitation. Our analysis based on the above data exhibits a tight relation between flash rate and the depression in the passive microwave signal at 85 GHz (Tb85), which is the channel mostly sensitive to ice scattering (Mohr et al. 1999). Figure 1 presents a characteristic example of this relation. It displays corresponding plots of Tb85, CG flash rate, and WSR-88D rain rate profiles taken across a Mesoscale Convective System in Oklahoma. The figure shows a very good spatial correlation between Tb85 and CG lightning frequency patterns, while there is a minor shift (<15 km) between radar rainfall patterns and those of Tb85 and CG flash rates. This is likely related to spatial displacements between precipitation ice processes concentrated in regions with strong updrafts and surface precipitation beneath. Figure 2 (left panel) highlights a mean nonlinear (nearly exponential) relationship of flash rate to Tb85. In Fig. 3 (left panel) we show that Tb85 values from electrified pixels have a tendency to “shift” towards lower temperatures, comparatively to Tb85 values from rainy pixels that reside outside the electrified areas.
Figure 3. Relative frequency histograms of Tb85 and Tb37 in cases of electrified and nonelectrified passive microwave observations in rainfall.
A mixture of emission and scattering is depicted by the lower wavelength of 37 GHz (Tb37), which is directly influenced by the presence of super-cooled hydrometeors. The strong relationship exhibited between flash rate and Tb85 is not apparent in the case of Tb37 channel (Fig. 2, right panel). It is shown that Tb37 has low sensitivity to the intensity of electrification. An explanation for this is that the existence of super-cooled cloud water does not require a convective regime, something compelling at the presence of graupel and intense lightning activity. According to Smith et al. (1992) there may be counteraction between the 85 and 37 GHz signatures since emission by super-cooled water droplets may increase Tb85 (smoothing out abrupt brightness temperature changes). Compensation of this warming effect can be provided by the existence of a thicker ice layer above the super-cooled
The Worth of Long-range Lightning Observations on Overland Satellite Rainfall
141
water layer, fact argued here to be the dominant case of intense electrification. To support our argument we use the polarization corrected temperature difference of 37GHz versus 85GHz (Tb37-Tb85), which quantifies the warming effect mentioned above. In Fig. 4 we plot the probability distribution of Tb37-Tb85 for electrified and nonelectrified pixels. The histogram of nonelectrified pixels shows a mean around zero, expressing an equal impact by emission and ice scattering, which is a typical characteristic of the stratiform mixed phase. On the other hand, in electrified/convective areas, the difference shows an apparent shift to positive values, fact explained by the faster decrease of the Tb85 due to scattering compared to the Tb37. Finally, in Fig. 5 we show the relationship of mean rain rate to Tb85 (Tb37) for electrified and nonelectrified pixels. We note differentiation in the Tb85-rain rate relationships of electrified and nonelectrified rain areas, which demonstrate the significance of lightning information in passive microwave rain estimation. First, the relationship of electrified pixels gives almost twice as much rain the nonelectrified relationship does. Second, in the nonelectrified pixels the relationship “breaks apart” as Tb85 reduces to colder than 200 K temperatures, while in the electrified pixels, it sustains the monotonic character throughout the whole range of brightness temperatures. These distinct differences in the Tb85-rain relationships between electrified and nonelectrified clouds indicate that lightning information can have a positive impact on the overland passive microwave rain estimation, which is primarily based on Tb85 observations. Having established the quantitative connection of lightning to convection and precipitating ice processes, we will now demonstrate the potential improvements in quantitative precipitation estimation coming from combining lightning information with the lesser definitive but quasi-continuous satellite infrared (IR) observations.
Figure 4. Frequency histogram of the Tb37-Tb85 difference for electrified and nonelectrified passive microwave observations in rainfall.
142
5
E. N. Anagnostou and T. G. Chronis
HIGH-RESOLUTION LARGE-REGIONAL RAINFALL ESTIMATION FROM COMBINING SATELLITE IR AND LIGHTNING DATA
Geostationary Infrared (IR) imagery has been a useful tool to rainfall monitoring due to its advantage of providing global coverage and high frequency (15–30 min) of observations. There has been significant research on developing rain retrieval algorithms from IR data. Among the first algorithms is the development of a simple threshold technique named GOES Precipitation Index (GPI) (Arkin and Meisner 1987). The technique implemented an IR threshold to delineate areas of mean rainfall value, based on which Arkin and Xie (1994) produced precipitation products over tropical and subtropical regions. Further improvement of geosynchronous satellite precipitation retrievals through regional calibration by the use of microwave observations (which have a more direct physical link to precipitation rate) is currently approached in a number of studies with the most recent including Hsu et al. (1999), Todd et al. (2001) and Huffman et al. (2003). Other studies have tried to calibrate an IR algorithm, on the basis of highresolution passive microwave retrievals, in deriving convective and stratiform precipitation, but with moderate success (Negri et al. 2002; Anagnostou et al. 1999). A generality to be drawn from these studies is the weakness of satellite IR measurements to sufficiently characterize the convective precipitation variability. The incorporation of regional lightning information from long-range network data would offer the additional information needed to bring convective rain area and rate estimation to accuracy levels suitable for studying the water cycle at high spatial-temporal frequency.
Figure 5. Scatter plot of mean rain rate for different categories of Tb85 and Tb37 in electrified and nonelectrified passive microwave rainfall observations.
Relatively few studies have related lightning information with precipitating regimes observed by geostationary satellites. Grandt (1992) studied
The Worth of Long-range Lightning Observations on Overland Satellite Rainfall
143
thunderstorm evolution by combining Meteosat images and sferics location retrievals, during different seasons over Africa and Europe and classified convective clouds using indexes related to cloud-top brightness temperatures. Grecu et al. (2000) used a combination of lightning and IR brightness temperature observations to retrieve rainfall. In this study it is shown that the aforesaid combination could reduce by 15% the error variance of rain volume estimates compared to an IR only scheme for clouds associated with lightning. The authors also demonstrated that lightning information could improve the overall rainfall estimation accuracy compared to an IR only approach. The above study although targeted the significance of lightning information in rainfall estimation its result is limited by sample size and regional extent (15 days over Southern USA) as well as an increased uncertainty in the overland passive microwave rainfall fields used as reference for calibration/validation. The most comprehensive studies on the use of lightning data jointly with IR as a mean for advancing our capabilities on continuous precipitation monitoring over large regions are those of Morales and Anagnostou (2003) (hereafter MA03) and Chronis et al. (2004) (hereafter CH04). Below we describe the main findings from those studies. MA03 developed a technique to retrieve rainfall rates by combining satellite IR observations with data from a long-range lightning network of sferics receivers in US East Coast and the Caribbean (see Fig. 1 of MA03). The study used TRMM Precipitation Radar (PR) rainfall products to calibrate and evaluate the combined IR/lightning rain retrieval. The hypothesis made in formulating the methodology is that precipitation from electrified clouds has distinct ice microphysical properties compared to nonelectrified clouds; consequently, developing different rain retrieval parameterizations for clouds with and without lightning would contribute to improving precipitation estimation, while a better evaluation of convective rainfall may be achieved by combining flash rate and IR brightness temperature variables compared to using sole IR data. The authors developed parameterizations for the estimation of total rain area and the convective rain area portion that are distinct for lightning and lightning free precipitating clouds. The parameterizations were developed using bulk variables evaluated on the basis of cloud systems delineated by the 258 K isotherm in the IR array. Those variables are the cloud area defined by the 258 K isotherm, the area contained within the isotherm of the most frequent IR temperature in the cloud system, and the lightning area in the case of electrified clouds. The rainfall rate relationships were developed in a probabilistic way by matching the cumulative distributions of IR brightness temperatures and lightning rates with the corresponding PR precipitation estimates. These rain relations were evaluated separately for convective/stratiform rain types, lightning/lightning-free clouds, and land/ocean surfaces exhibiting distinct functional forms as shown in MA03.
144
E. N. Anagnostou and T. G. Chronis
Three months (December 1997 to February 1998) of coincident data over a large region (i.e., 125W-45W in longitude and 40N-10S in latitude) were used to evaluate the algorithm parameters and assess its performance. Additional validation was performed based on rain gauge observations from a network in Florida. As shown in MA03 (see Fig. 7 of MA03) the estimated rain areas are well determined when compared against rain areas observed by TRMM-PR. Overall, the technique underestimates both lightning (21%) and lightning-free (29%) rain areas. The standard error and correlation coefficients for lightning (lightning-free) areas are 36% (43%) and 0.97 (0.96), respectively. Retrievals in lightning clouds are shown to have lower error (higher correlation) on the definition of convective and stratiform rain areas compared to the lightning-free clouds, this effect is more pronounced in the estimation of convective precipitation (we note a correlation increase of 0.15). Comparisons with a validation rain gauge network in Florida revealed that the combined technique is able to represent the observed precipitation distribution at scales ranging from 100 km2 to 4 degree2 with low overall bias (6%) at 100 km2, which increases at coarse resolution (~35% at 4 degree2). The authors investigated the significance of lightning information on rainfall estimation accuracy. In that respect, the same technique was set to run without lightning information: namely, all clouds were set as nonelectrified. Comparisons with TRMM-PR showed that lightning information improves rainfall estimation accuracy when compared to the nonlightning scenario. In rain area determination there was an overall bias reduction of 31%. In rain volume the lightning information was shown to introduce bias reduction of 10%. In rain gauge comparisons, the bias reduction from incorporating lightning data was more pronounced. It ranged from 80% to about 38% (9%) for the 0.1 degree and 1 degree (2 degree) spatial scale, respectively. The increase in correlation coefficient was 0.13 for the hourly gauge data. For more details on the above findings the interested reader is referred to MA03. A subsequent study by CH04 came to confirm and strengthen the findings by MA03. The authors developed a technique (named OMVRIOS) somewhat different in terms of algorithmic structure, but based on the same principles. Two main differences are the use of overland passive microwave retrievals from SSM/I observations as reference rainfall dataset and the simplifications introduced in the algorithm by reducing the number of parameterizations. OMVRIOS technique was demonstrated on a different geographic regime, the European continent, based on Zeus sferics network. The sferics data in this study exhibited better location error and detection efficiency characteristics compared to the US network data. This was mainly due to improvements in the new system (lower noise floor at each receiver location, improved signal frequency and noise reduction, improvements in the locating algorithm by incorporating a more accurate model for
The Worth of Long-range Lightning Observations on Overland Satellite Rainfall
145
the sferics signal velocity) and the larger number and better configuration of receivers deployed in Europe. The basic commonalities of the two techniques is the use of bulk cloud cluster variables, and that lightning information is used to first differentiate between electrified versus nonelectrified cloud cluster parameterizations and second as a variable for the electrified clouds in the estimation of convective rain area and rain rate. Below we discuss some of the key aspects of OMVRIOS, while more details can be found in CH04. The technique was assessed and compared against two existing microwave-calibrated IR retrievals (the PERSIANN of Hsu et al. 1997 and VAR of Huffman et al. 2003) as well as the same technique but without the lightning information (OMVRIOS no-lightning) on the basis of independent six-hourly rain accumulation measurements from ECMWF gauge network spread across Europe (see Fig. 1 of CH04). The error statistics of OMVRIOS estimates against gauges exhibited scale dependence: the bias decreased from 30% at 0.1-degree resolution to ~5% at 5 degree resolution. The root mean square difference decreased from 93% (0.1 degree) to 78% (5 degrees), while the corresponding correlation increased from 0.78 to 0.95. The retrieval would detect about 50% of the total rainfall, and over 80% (95%) of rainfall accumulations exceeding 20 mm (25 mm) in a 6-hourly interval. Its overall false rain detection rate was 10%, which was shown to drop to zero for rain accumulations exceeding 8 mm in a 6-hourly interval. A main conclusion derived from the comparison of the technique against IRonly rain retrievals was that lightning information helps reduce IR retrieval uncertainty. In terms of rain detection the combined lightning/IR retrieval was shown to have significantly higher critical success index scores (in the order of 40%–100% increase) compared to the IR-only techniques. In terms of rainfall rates it was shown to have lower (25%–40%) root mean square differences compared to the other IR retrievals, and a nearly 0.3 increase in correlation with 6-hourly rain gauges. Improvements are also apparent in the cumulative distributions of 6-hourly rainfall accumulations derived from the different techniques and rain gauge rainfall measurements. As shown in Fig. 12 of CH04, OMVRIOS exhibits a much closest agreement with rain gauges compared to the other techniques. The demonstrated positive impact of continuous long-range lightning measurements on high-frequency precipitation estimation from satellites will help advance water and energy cycle understanding at both regional and climate scale. With the increasing availability of regional lightning data over the major convective chimneys of earth (e.g., Africa, South America, South Asia) algorithms like OMVRIOS can now provide improved rainfall fields (in terms of both resolution and accuracy) to what is currently available from sole IR techniques. This would advance our ability to study in more detail convective system dynamics and microphysics, as well as the role of surface conditions (e.g., soil moisture) on the water cycle variability.
E. N. Anagnostou and T. G. Chronis
146
6
CLOSING REMARKS
The main focus of this paper has been on the use of lightning information as a mean of advancing precipitation estimation from satellite observations. The key aspect of lightning is its tight physical connection to convection in a cloud, which is one of the weak points of satellite IR observations. It was shown that lightning information could help improve both overland passive microwave and IR based precipitation estimation. In terms of an overland passive microwave retrieval the advancement comes from the demonstrated microphysical differences of lightning versus lightning-free clouds resulting in distinct Tb85-rain rate relationships. In terms of an IR rain retrieval the improvement is more pronounced as it was demonstrated in the associated studies. The approach of incorporating lightning data in an IR retrieval is through classification of cloud systems in electrified versus nonelectrified, and consequently using the flash rate and IR temperature variables jointly in estimating the convective rain area and rain rate. Validation studies in U.S. and Europe have shown that the combined IR/lightning approach is superior to other techniques that are based solely on IR observations. Research is now underway to expand those combined algorithms in other continental convective regimes such as Africa, South Asia and the Amazon basin.
7
REFERENCES
Anagnostou, E. N., T. Chronis, and D. P. Lalas, 2002: New receiver network advances longrange lightning monitoring. EOS Feature Article, AGU, 83(50), 589, 594–595. Anagnostou, E. N., A. J. Negri, and R. F. Adler, 1999: A satellite infrared technique for diurnal rainfall variability studies. J. Geophys. Res., 104 (D24), 31477–31488. Arkin, P. A. and B. N. Meisner, 1987: The relationship between larger scale convective rainfall and cold cloud over the Western Hemisphere during 1982–84. Mon. Wea. Rev., 115, 51–74. Arkin, P. A. and P. Xie, 1994: The global Precipitation Climate Project: First Algorithm Intercomparison Project. Bull. Amer. Meteor. Soc., 77, 2875–2887. Boccippio, D. J., K. L. Cummins, H. J. Christian, and S. J. Goodman, 2001: Combined satellite- and surface-based estimation of the intracloud–cloud-to-ground lightning ratio over the continental United States. Mon. Wea. Rev., 129, 108–122. Bremmer, H., 1949: Terrestrial radio waves. Elsevier Press, New York. Budden, K. G., 1951: The propagation of a radio-atmospheric. Phil. Mag., Ser 7, 42, 1–19. Cecil, D. J. and E. J. Zipser, 1999: Relationships between tropical cyclone intensity and satellite-based indicators of inner core convection: 85-GHz ice-scattering signature and lightning. Mon. Wea. Rev., 127, 103–123. Chang, D.-E., J. A. Weinman, C. A. Morales, and W. S. Olson, 2001: The effect of spaceborne microwave and ground-based continuous lightning measurements on forecasts of the 1998 Groundhog Day Storm. Mon. Wea. Rev., 129, 1809–1833. Christian, H. J., R. J. Blakeslee, S. J. Goodman, D. A. Mach, M. F. Stewart, D. E. Buechler, W. J. Koshak, J. M. Hall, W. L. Boeck, K. T. Driscoll, and D. J. Boccippio, 1999: The Lightning Imaging Sensor. Proc. 11th Int. Conf. on Atmos. Electricity, Guntersville, Alabama, 746–749.
The Worth of Long-range Lightning Observations on Overland Satellite Rainfall
147
Chronis, T., E. N., Anagnostou, and T. Dinku, 2004: High-frequency estimation of thunderstorms via satellite infrared and a long-range lightning network in Europe. Quart. J. Roy. Meteor. Soc., 130, 1555–1574. Chronis, T. G. and E. N. Anagnostou, 2003: Error analysis for a long-range lightning monitoring network of ground-based receivers in Europe. J. Geophys. Res., 108 (D24), 4779, doi:10.1029/2003JD003776. Cummins, K., M. Murphy, E. Bardo, W. Hiscox, R. Pyle, and A. Pifer, 1998: A combined TOA/MDF technology upgrade of the U.S. National Lightning Detection Network. J. Geophys. Res., 103, 9035–9044. Goodman, S. J., D. E. Buechler, and P. J. Meyer, 1988: Convective tendency images derived from a combination of lightning and satellite data. Wea. Forecasting, 3, 173–188. Grandt, C., 1992: Thunderstorm monitoring in south Africa and Europe by means of Very Low Frequency sferics. J. Geophys. Res., 97 (D16), 18215–18226. Grecu, M. and E. N. Anagnostou, 2001: Overland precipitation estimation from TRMM passive microwave observations. J. Appl. Meteor., 40, 1367–80. Grecu, M., E. N. Anagnostou, and R. F. Adler, 2000: Assessment of the use of lightning information in satellite infrared rainfall estimation. J. Hydrometeor., 1, 211–221. Hsu, K., X. Gao, S. Sorooshian, and H. V. Gupta, 1997: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36, 1176– 1190. Hsu, K., H. Gupta, X. Gao, and S. Sorooshian, 1999: Estimation of physical variables from multichannel remotely sensed imagery using a neural network: application to rainfall estimation. Water Resources Res., 35, 1605–1618. Huffman, G. J., R. F. Adler, E. F. Stocker, D. T. Bolvin, and E. J. Nelkin, 2003: Analysis of TRMM 3-hourly multi-satellite precipitation estimates computed in both real and postreal time. Proc. 12th Conf. Satellite Meteorology and Oceanography, 9–13 Feb. 2003, Long Beach, CA. Krider, E. P., 1996: 75 years of research on the physics of a lightning discharge, Chapter 11 in Historical Essays on Meteorology, 1919–1995: The Diamond Anniversary History Volume of the American Meteorological Society, AMS, Boston, MA, June. Lee, A. C. L., 1986a: An experimental study of the remote location of lightning flashes using a VLF arrival time difference technique. Quart. J. Roy. Meteor. Soc., 112, 203–229. Lee, A. C. L., 1986b: An operational system for remote location of lightning flashes using a VLF arrival time difference technique. J. Atmos. Oceanic Technol., 3, 630–642. Lyons, W. A., T. E. Nelson, E. R. Williams, S. A. Cummer, and M. A. Stanley, 2003: Characteristics of sprite-producing positive cloud-to-ground lightning during the 19 July 2000 STEPS mesoscale convective systems. Mon. Wea. Rev., 131, 2417–2427. Mohr, K. I., R. Toracinta, E. J. Zipser, and R. E. Orville, 1996: A comparison of WSR-88D reflectivities, SSM/I brightness temperatures, and lightning for mesoscale convective systems in Texas. Part II. SSM/I brightness temperatures and lightning. J. Appl. Meteor., 35, 919–931. Morales, C. and E. N. Anagnostou, 2003: Extending the capabilities of high-frequency rainfall estimation from geostationary-based satellite infrared via a network of long-range lightning observations. J. Hydrometeor., 4, 141–159. Negri, A. J., L. Xu, and R. F. Adler, 2002: A TRMM-calibrated infrared rainfall algorithm applied over Brazil. J. Geophys. Res., 107 (D20), 8048, doi:10.1029/2000JD000265. Simpson, G. C. and F. J. Scrase, 1937: The distribution of electricity in thunderstorms. Proc. Roy. Soc. A, 161, 309–52. Simpson, S., R. F. Adler, and R. G. North, 1988: A proposed tropical rainfall measuring mission. Bull. Amer. Meteor. Soc., 69, 278–295. Smith, E. A., H. J. Cooper, X. Xiang, A. Mugnai, and G. J. Tripoli, 1992: Foundations for statistical-physical precipitation retrieval from passive microwave satellite measurements.
148
E. N. Anagnostou and T. G. Chronis
Part I: Brightness-temperature properties of a time-dependent cloud-radiation model. J. Appl. Meteor., 31, 506–531. Solomon, R. and M. Baker, 1998: Lightning flash rate and type in convective storms, J. Geophys. Res., 103 (D12), 14041–14057. Takahashi, T., T. Takuya, and S. Yasuo, 1999: Charges on graupel and snow crystals and the electrical structure of winter thunderstorms. J. Atmos. Sci., 56, 1561–1578. Todd, M. C., C. Kidd, D. Kniveton, and T. J. Bellerby, 2000: A combined satellite infrared and passive microwave technique for estimation of small scale rainfall. J. Atmos. Oceanic Technol., 18, 742–754. Toracinta, E. R., D. J. Cecil, and E. J. Zipser, 2002: Radar, passive microwave, and lightning characteristics of precipitating systems in the Tropics. Mon. Wea. Rev., 130, 802–824. Williams, E. R., M. E. Weber, and R. E. Orville, 1989: The relationship between lightning type and convective state of thunderclouds. J. Geophys. Res., 94 (D11), 13213–13220. Ziegler, C. D., J. D. McGorman, and P. Ray, 1991: A model evaluation of noninductive graupel-ice charging in the early electrification of a mountain thunderstorm. J. Geophys. Res., 96 (D7), 12833–12855.
12 NEURAL NETWORK TOOLS FOR SATELLITE RAINFALL ESTIMATION Francisco J. Tapiador1∗, Chris Kidd1, Vincenzo Levizzani2, and Frank S. Marzano3 1
School of Geography, Earth and Environmental Sciences. The University of Birmingham, UK 2 Consiglio Nazionale delle Ricerche, Institute of Atmospheric Sciences and Climate, Bologna, Italy 3 Department of Electrical Engineering, University of L’Aquila, L’Aquila, Italy
Abstract
This paper shows some examples of how Neural Networks (NNs) have been used for satellite rainfall estimation (SRE). We start with a simple case that illustrates how NNs can be used, presenting a simulation of the well-known auto-estimator procedure. Secondly, we present an overview of a direct approach for sensor fusion using NNs. Next section is devoted to an explanation of the role played by NN in indirect procedures such as the cloud motion winds rain estimation method. The last part of the paper demonstrates that NNs can be used to simulate complex, physically based algorithms with no loss of performances and in a fraction of the time required by those algorithms.
Keywords
Neural Networks, Satellite Rainfall Estimation
1
INTRODUCTION
A Neural Network (NN) is a variational method of constrained optimization: given an objective function and a set of constraints, a NN algorithm provides the best (nonlinear) function for the input/output relationship, or alternatively the most likely pattern from a single set of inputs. The mathematics behind
∗
Presently at Institute of Environmental Sciences, University of Castilla-La Mancha, Toledo, Spain
149 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 149–161. © 2007 Springer.
F. J. Tapiador et al.
150
NNs have long been established and theorems and results have been provided in the last decades, demonstrating valuable properties such as the NNs ability to map any function satisfying the Dirichlet’s conditions (Kolmogorov 1963). Moreover, most of the classification statistical procedures can be taken as particular cases of NNs (Haykin 1999) and NNs nonparametric approach presents many advantages over other statistical procedures (Sarle 1994). Hornik et al. (1989) have also proved that MLP NN can approximate any measurable function up to an arbitrary degree or accuracy: as a semiparametric regression estimator NN can model a nonlinear function in a finite number of steps. The most common form of NNs is the multilayer perceptron (MLP). While other researchers have successfully used other NN paradigms such as Hopfield networks or Self-Organizing Maps (SOMs) (Hsu et al. 2002), our presentation will be limited to MLP NNs examples. A MLP consists in at least two layers of neurons linked by synapses with an assigned weight. Details can be found in Tapiador et al. (2004a, b). The key idea is to use an iterative procedure minimizing a given cost function between inputs and outputs. This process called training is performed through dynamically modifying the weights of the net with a sample of representative inputs and their expected outputs. This is iteratively done until the net is able to model the link between them: a NN learns comparing its own outcomes with the outputs provided, adjusting the weight given to the synapses in order to minimize the differences between output and model estimation. Once the learning process has finished, the net can deal not only with the original training examples, but also with new inputs not seen before which would be correctly recognized (generalization capability). In general, NNs are appropriate when either we have not a good description of the physical processes involved, the physics of the problem is not deterministic but statistically described or when there is a proper physically based algorithm but it is slow. In this paper we present some NN examples for satellite rainfall estimation (SRE) covering these three aspects.
2
A SIMPLE EXAMPLE
We will illustrate how NNs work with a simple example. Let us suppose that the relationship between infrared (IR) top cloud temperature (Tb) and rainfall rate (RR) were described by the auto-estimator relationship (Vicente et al. 1998):
(
RR = 1.1183 ⋅ 10 11 ⋅ exp − 3.6382 ⋅ 10 −2 ⋅ Tb1.2
)
If we were unaware of this relationship but we had the IR as inputs and the rain rates as outputs, we could try to model the function linking the inputs
Neural Network tools for Satellite Rainfall Estimation
151
and outputs. The simplest NN to accomplish this would be represented by Fig. 1.
Figure 1. A simple NN aiming to simulate the auto-estimator (Vicente et al. 1998).
By training this NN with a sufficient number of cases, we obtain that: 1
RR = ∑ F (W ⋅ Tb + B )
(1)
i =1
where F(·) is the sigmoid function and B and W are 1×1 matrices (because we only have one synapse [W] and a neuron [B]). Their values are: F (•) =
1 1 + exp(− •)
B = [1.54] W = [− 15.48]
(2)
This gives: RR =
160 ⎛ ⎞ 1 + exp⎜⎜15.48 Tb − 1.5439 ⎟⎟ ⎝ ⎠
(3)
The fit of this model with the actual data is shown in Fig. 2. As can be seen the fit is not perfect. From a mathematical point of view, this NN represents the least-squares fit for the curve. However, we have used the simplest possible NN: a NN with only input and output units can only
F. J. Tapiador et al.
152
provide a simple curve fitting and not a real nonlinear modelling of the problem.
Figure 2. Relationship between data (dots) and the simple NN model (line) in Fig. 1.
If we now include just two additional neurons in a so-called hidden (or intermediate) layer (Fig. 3) and we train again the NN using the same data as before, we obtain two new matrices defining the NN:
⎡1.48 ⎤ B = ⎢⎢0.93⎥⎥ ⎢⎣7.26⎥⎦
⎡ 29.6 ⎤ ⎢− 13.9⎥ ⎥ W =⎢ ⎢ − 4.1 ⎥ ⎢ ⎥ ⎣ 9.8 ⎦
(4)
Neural Network tools for Satellite Rainfall Estimation
Figure 3. A simple NN that perfectly simulates the auto-estimator.
Therefore, we obtain the curve shown in Fig. 4.
Figure 4. Relationship between data (dots) and the simple NN model (line) in Fig. 3.
153
F. J. Tapiador et al.
154
As can be seen, the agreement is almost perfect (R2 = 0.9998). Using just four neurons, a NN is capable to simulate the auto-estimator example. However, real problems of SRE involve such number of uncertainties (beam-filling effect, limited time collocation of the measures, different view angles, etc.) that the relationship between measures becomes very complex. Typical NNs used in direct fusion approaches consist in hundreds of neurons in several layers.
3
DIRECT FUSION
Several works have shown how NNs can be used into a direct fusion framework for SRE (Sorooshian et al. 2000). IR sensors can provide large coverage and good spatial and temporal resolutions but are limited due to the indirect relationship between top cloud temperature and rainfall. On the other hand, passive microwave (PMW) sensors can provide a more direct estimate since radiation at these wavelengths are more affected by raindrops but have less spatial and temporal resolutions. The aim of fused approaches is to merge both kind of sensors to improve the temporal and spatial resolution and the directness of the estimate. A complete example of this approach can be seen in Tapiador et al. (2004a).
Figure 5. NN vs MW total rain estimates at 0.1° resolution (left) and aggregated at 2.5° (right). The improvement due to the resampling is noticeable.
The basic idea is to use the NN to relate the IR data as input with a reference rainfall such as PMW-derived estimates, rain gauge or radar data. The hypothesis is that, for co-registered images, there is a function between both measures. The relationship between both datasets is nonlinear and quite indirect, but some success has been reported on relating the IR data with
Neural Network tools for Satellite Rainfall Estimation
155
surface rainfall at the long term: statistical methods such as GPI (Arkin et al. 1979) have proven that the hypothesis can be assumed. The aim of using NNs for the process is to be able to mathematically describe the process using the available data and then generalize for new measures. Figure 5 shows a result of comparing a NN estimate with an independent PMW total rainfall estimate. While at the 0.1° resolution the correlation is 0.76, spatial aggregation aiming to minimize the uncertainties increases the correlation up to 0.98 for a 2.5° resampling. This effect has been studied by Turk et al. (2002).
4
INDIRECT FUSION
There is also a growing use of NNs as part of more complex algorithms. NNs are then used for modeling a particular part of the process, for instance in a cloud classification scheme previous to the rain estimate (Hsu et al. 2002). We will illustrate this indirect approach with another example. A method currently in development is the use of NNs as a part of a cloud motion winds diffusion scheme. The rationale is as follows; the cloud motion as seen by the satellite can be modelled in terms of fluid dynamics. Thus, the NavierStokes equations express the dynamics of the fluid: ρ
⎛ ∂ 2u ∂ 2u ∂ 2u ⎞ ∂p du =− + ν ⎜ 2 + 2 + 2 ⎟ + ρg x ⎜ ∂x dt ∂x ∂y ∂z ⎟⎠ ⎝
ρ
⎛ ∂ 2v ∂ 2v ∂ 2v ⎞ ∂p dv =− + ν ⎜ 2 + 2 + 2 ⎟ + ρg y ⎜ ∂x dt ∂y ∂y ∂z ⎟⎠ ⎝
ρ
⎛ ∂2w ∂2w ∂2w ⎞ ∂p dw =− + ν ⎜ 2 + 2 + 2 ⎟ + ρg z ⎜ ∂x dt ∂z ∂y ∂z ⎟⎠ ⎝
(6)
Where (u,v,w) are the components in x,y,z of the velocity, p is the pressure, ρ is the fluid density, ν is the viscosity and gx,y,z the gravity vector components. Using the nabla operator to condense the notation G G G dv ρ = −∇p + ν∇ 2 v + ρg dt
(7)
If we now simplify the model as a bidimensional discrete case the variations of the pressure fields in the x and y dimensions from t0 to t1 are equivalent to an affine transformation (Birkhoff and Mac Lane 1996). So Pt1 = ( APt0 + B)
(8)
F. J. Tapiador et al.
156
Where P is the pressure matrix, and A and B are the affine transformation matrices. The velocity for the unit of time is then: G v = AP + B − P = ( A − I )P + B
(9)
Where I is the singular matrix. By taking derivatives in (9): G G dv = ( A − I )v dt
(10)
is obtained from the left side of the equation. From the right side we have that: G ∇ 2 v = ∇ 2 [( A − I )P + B ] = 0
(11)
Substituting now equations 10 and 11 into 8, and assuming that the gravity is negligible: G ∇p = − ρ ( A − I )v
(12)
Equation 11 shows that the divergence of the pressure in a cloud patch is a linear combination of the velocity components. Thus, the analysis of the bidimensional image provided by the satellite is equivalent (with the required simplifications) to the motion of the cloud movement from the point of view of fluid dynamics. This is important since we can then model the problem of the cloud movement as equivalent to the flow of the brightness temperature as seen by the satellite, which is what we measure from, for example the Meteosat Second Generation satellite. Therefore, we can model as follows. First, we consider that the brightness temperature of a cloud patch remains constant after a short period of time, in our case 15 min: T ( x , y , t ) = T ( x + δ x , y + δ y , t + δt )
(13)
Expanding the rhs and gathering the terms of the δ increments above the second in ε: T ( x , y , t ) = T ( x, y , t ) + δx
∂T ∂T ∂T + δy + δt +ε ∂x ∂y ∂t
(14)
Neural Network tools for Satellite Rainfall Estimation
157
Applying the chain rule: δx ∂T δy ∂T ∂T + + + O(δt ) = 0 δt ∂x δt ∂y ∂t
(15)
If the elapsed time is negligible: ∂T dx ∂T dy ∂T + + =0 ∂x dt ∂y dt ∂t
(16)
Simplifying the notation by naming the components of the velocity as u and v and the partial derivatives of the brightness temperature in x and y by Tx and Ty we have: T x u + T y v + Tt = 0
(17)
Equation (17) is a conservation law that states that the image irradiance of a given area remains constant during the motion process. However, using only equation (17) the velocity cannot be locally determined (Horn and Schunck 1980). We need additional constraints based upon what we observe about the system. Since cloud patches usually move quite homogeneously – we do not expect that the MSG image after 15 min is a random version of the current, but a smoothly displaced one – we can assume that some kind of quantity is conserved during the process. Horn and Schunck (1980) proposed as a simple functional to be minimised the sum of the squares of the Laplacians of the x and y components of the movement. Including (17) in the functional to ensure that the conservation of irradiance is satisfied, we obtain this functional: E=
⎡ ⎛ ∂ 2 u ∂ 2 u ∂ 2 v ∂ 2 v ⎞⎤ ⎢α T x u + T y v + Tt + ⎜⎜ 2 + 2 + 2 + 2 ⎟⎟⎥dxdy ∂y ∂x ∂y ⎠⎥⎦ ⎢⎣ ⎝ ∂x
∑∑ (
)
(18)
Where α is a proportionality factor that gives the relative weight of the two constraints. Using the method of Lagrange multipliers to minimize (18), we obtain that:
(
)
(
)
∂u = ∇ 2 u − αT x T x u + T y v + Tt ∂t ∂v = ∇ 2 v − αT y T x u + T y v + Tt ∂t
(19)
158
F. J. Tapiador et al.
If we assume that the clouds do not move independently of the surrounding atmosphere, we can simply solve (19) using iterative procedures to obtain the components of the velocity at each point. This will give us a smooth field, which will represent the movement of the air masses. If we were only interested in the movement of the clouds, we could introduce additional terms to preserve the flow discontinuities. In terms of the PMW cloud motion-driven precipitation procedure we propose, the differences were found small enough not to be considered in this preliminary work and the air masses movement approach was chosen. Nevertheless, the possibility of distinguishing between the actual clouds and the rest of the atmosphere will be exploited in further refinements of the current method.
Figure 6. A sample of the diffusion method compared with an IR-based procedure and the actual radar data. (see also color plate 4)
Neural Network tools for Satellite Rainfall Estimation
159
NNs are then used to relate the cloud motion with the rainfall systems motion. Assuming there is a complex, nonlinear relationship between cloud motion and rain systems motion, a NN can be trained to model this link when PMW or other rain estimates are available. After this training, the NN is able to provide an estimate of the rain system movement using just the IR information. The operational process is as follows: when a PMW rainfall estimate becomes available at time T, we start calculating the air mass movement using the IR data. Then, we drive the precipitation as derived by the PMW from T to the next future IR image at T + 15 min. This simulates the movement of the PMW estimate from T to T + 15. While no additional PMW information is available, we continue calculating the air mass movement and are driven the PMW estimate from time T. The reason to do so instead of directly moving the previous driven-estimate is to exploit the existence of sub-pixel movements that the algorithm provides. These subpixel displacements will be truncated if a running drive will be performed, resulting in a delay of the actual rainfall estimates. For a 3-h lag the displacement could be as large as 30 km. Experiments using both approaches were performed obtaining noticeable differences. This method is still being improved and validated. Figure 6 shows two snapshots of the process. The rain system movement appears as reasonable and preliminary analyses show an improvement albeit small on the actual quantitative rainfall estimates.
5
SIMULATION OF COMPLEX ALGORITHMS
The most successful application of NNs in satellite rainfall estimation is the simulation of complex, physically based algorithms. These algorithms are usually time-consuming and slow methods, which either impedes their operational use or limit their use to massive computer systems availability. This topic has extensively been exposed for radiative transfer calculations and geophysical variables by Göttsche and Olesen (2002), Krasnopolsky and Chevallier (2003), and Krasnopolsky and Schiller (2003). A specific example for SRE of this NN usage follows. Briefly, the idea is to simulate the TRMM 2A12 algorithm (Kummerow et al. 1996) using a NN in order to speed up the calculations. By training the NN with the same inputs as the algorithm, it is possible to obtain virtually the same results (Fig. 7) in O(104) of the time required by the algorithm. Additional advantages of NNs usage are the ability to directly calculate the Jacobian and include new inputs with minimal modifications.
F. J. Tapiador et al.
160
Figure 7. A comparison between G-PROF 2A12 profiles and the NN simulation using the same data. Errors are negligible.
6
CONCLUSIONS
NNs are perfectly suitable to simulate complex algorithms, and thus can be used to speed up physically based models. However, all the current operational rainfall estimation algorithms are not model-based but statistically based and here the success of using NNs has been proven as limited since direct approaches for sensor fusion lack of a direct relationship between cloud top temperature and surface rainfall. Nonetheless, NNs can still provide comparable performances to other methods. Another possible application of NNs in SRE is use them as a part of other algorithms to model a complex process previous to the actual rainfall estimate.
Neural Network tools for Satellite Rainfall Estimation
7
161
REFERENCES
Arkin, P. A., 1979: The relationship between fractional coverage of high cloud and rainfall accumulations during GATE over the B-scale array. Mon. Wea. Rev., 107, 1382–1387. Birkhoff, G. and S. Mac Lane, S., 1996: Affine geometry. In A Survey of Modern Algebra, 5th edn. New York: Macmillan, pp. 268–275. Göttsche, F.-M. and F. S. Olesen, 2002: Evolution of neural networks for radiative transfer calculations in the terrestrial infrared. Remote Sens. Environ., 80, 157–164. Haykin, S. S., 1999: Neural Networks: A Comprehensive Foundation, 2nd dn. Upper Saddle River, NJ: Prentice-Hall Horn, B. K. P. and B. G. Schunck, 1981: Determining optical flow. Artificial Intelligence, 17, 185–203. Hornik, K., M. Stinchcombe, and H. White, 1989: Multilayer neural networks are universal approximators. Neural Networks, 2, 359–366. Hsu, K., S. Sorooshian, Y. Hong, and X. Gao, 2002: Cloud classification and rainfall estimation using GOES imagery. Int. Conf. Quantitative Precipitation Forecasting, Reading, UK, 2–6 September. Kolmogorov, A. N., 1963: On the representation of continuous functions of several variables by superposition of continuous functions of one variable and addition. Amer. Mathematical Soc. Trans., 28, 55–99. Krasnopolsky, V. M. and F. Chevallier, 2003: Some neural network applications in environmental sciences. Part II: Advancing computational efficiency of environmental numerical models. Neural Networks, 16, 335–348. Krasnopolsky, V. M. and H. Schiller, 2003: Some neural network applications in environmental sciences. Part I: Forward and inverse problems in geophysical remote measurements. Neural Networks, 16, 321–334. Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. and Remote Sensing, 34, 1213–1232. Sarle, W. S., 1994: Neural networks and statistical models. Proc. 19th Annual SAS Users Group International Conference. Cary, NC, SAS Institute, 1538–1550. Sorooshian, S., K. L. Hsu, X. Gao, H. V. Gupta, B. Imam, and D. Braithwaite, 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 2035–2046. Tapiador, F. J., C. Kidd, V. Levizzani, and F. S. Marzano, 2004: A neural networks-based fusion technique to estimate half hourly rainfall estimates at 0.1. resolution from satellite passive microwave and infrared data. J. Appl. Meteor., 43, 576–594. Tapiador, F. J., C. Kidd, K. L. Hsu, and F. S. Marzano, 2004: Neural networks in satellite rainfall estimation. Meteorol. Appl., 11, 1–9. Turk, F. J., E. E. Ebert, H.-J. Oh, and B. J. Sohn, 2002: Validation and applications of a realtime global precipitation analysis. Proc. IGARSS, 24–28 June 2002, Toronto, Canada. Vicente, G. A., R. A. Scofield, and W. P. Menzel, 1998: The operational GOES infrared rainfall estimation technique. Bull. Amer. Meteor. Soc., 79, 1883–1898.
Section 3 Rainfall Algorithms
13 PASSIVE MICROWAVE PRECIPITATION MEASUREMENTS AT MID- AND HIGH LATITUDES Ralf Bennartz Dept. of Atmospheric and Oceanic Sciences, University of Wisconsin–Madison, Madison, WI, USA
Abstract
This paper focuses on current passive microwave retrieval problems associated with mid- and high latitudes. A short discussion of state-of-theart techniques for precipitation identification and retrieval methods for mid- and higher latitudes is given. A new algorithm to discriminate between frozen and liquid precipitation on the ground is proposed and the first validation results for this algorithm are shown. This algorithm is based on a simple empirical relation between the height of the freezing level and the likelihood for snowfall to occur at the ground. It is validated against synoptic surface observations. Using two years of radar data obtained over the Baltic area, the impact of different sampling densities for current and future operational satellites is briefly discussed. It is found that the summertime convective maximum over land is currently not properly sampled.
1 INTRODUCTION Since the launch of the first Special Sensor Microwave/Imager (SSM/I; Hollinger et al. 1987) in 1987, passive microwave remote sensing of precipitation has become a standard instrument in various areas of atmospheric research. Furthermore, with the launch of the Tropical Rainfall Measuring Mission (TRMM; Kummerow et al. 1998) in 1997, tremendous progress in mapping and understanding tropical precipitation has been achieved. The Advanced Scanning Microwave Radiometer (AMSR-E; Kawanishi et al. 2003) onboard Aqua and other sensors such as the Advanced Microwave Sounding Unit (AMSU; Goodrum et al. 1999) provide a higher spatial resolution or additional channels at higher frequencies which are deemed advantageous not only for the retrieval of temperature and humidity profiles, but especially for the identification and retrieval of 165 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 165-177 © 2007 Springer.
166
R. Bennartz
precipitation. Forthcoming satellite missions such as the Global Precipitation Measurement (GPM; Smith et al. 1994) are currently in their definition phase and will allow improved monitoring of global precipitation within the next five to ten years. Compared to tropical rainfall, remote sensing of mid- and high latitude precipitation is especially challenging because of several factors that affect the retrieval. Firstly, the precipitation intensity, especially at higher latitudes, is usually much lighter than in the tropics and many of the rainfall events that contribute significantly to the total annual rain amount are at the low end of the possible range of detection for passive microwave remote sensing. Secondly, a large and highly variable fraction of the total annual precipitation falls as snow. The lack of liquid precipitation limits possible passive microwave retrievals to the use of the scattering signal at higher frequencies which is only indirectly linked to surface precipitation (see Bennartz et al. 2002; Bennartz and Michelson 2003; and references therein). Also the optical properties of frozen precipitation are more variable and less well known than those of liquid precipitation (see e.g., Skofronick-Jackson and Wang 2000; Petty 2001a, b; Bennartz and Petty 2001; SkofronickJackson et al. 2002, 2003). This results in larger uncertainties if the precipitation intensity is derived for frozen precipitation than for cases with a significant amount of liquid precipitation. At the same time, the strong interannual variability of precipitation at mid- and high latitudes leads to strong variations in the diurnal cycle of precipitation, especially over land. While in wintertime the diurnal cycle is weak, summertime precipitation is largely, and sometimes entirely, driven by convective precipitation. Temporal undersampling might therefore not only lead to an overall bias in precipitation estimates, but also to a false representation of the annual cycle of precipitation. Thirdly, associated with the occurrence of frozen precipitation, the surface emissivity might change on time scales of minutes over land surfaces. Over water surfaces the variable coverage with sea ice also strongly affects surface emissivity. The use of classical window channels for precipitation retrieval is strongly hampered by these factors. The current paper is subdivided into three parts. In the first part (section 2) we give a short overview on current methods used for passive microwave remote sensing of precipitation at higher latitudes. This overview is not intended to be a complete literature review on the broad variety of existing algorithms. The intention is rather to exemplarily discuss current venues of research. The second part of this paper introduces a novel method that uses passive microwave observations to discriminate between snowfall and rain at the surface. In the third part we briefly address the aforementioned sampling issues associated with the current set of operational weather satellites.
Passive Microwave Precipitation Measurements at Mid- and High Latitudes
167
2 OVERVIEW OF CURRENT RETRIEVAL ALGORITHMS FOR AMSR AND AMSU 2.1
Precipitation identification and retrieval
Currently, the AMSR-E onboard Aqua and the SSM/I on various DMSP satellites are the passive microwave sensors that provide global coverage and are most well suited for precipitation retrieval. AMSR’s operational precipitation retrieval algorithms (Shin and Kummerow 2003; Shibata et al. 2003; Wilheit et al. 2003) have a strong TRMM heritage and are currently in the process of being validated in the extratropical regions. Research on the development and validation of mid- and high-latitude precipitation retrievals has become increasingly important in the last several years. NASA and ESA currently support studies to extend the knowledge gained mainly from TRMM to higher latitudes. These studies include preparation studies for GPM, for its European counterpart EGPM, as well as development and validation studies for the AMSR-E onboard Aqua. ESA’s EGPM is currently planned to be equipped with a sounding-type instrument covering the oxygen absorption band around 60 GHz, as well as the oxygen line at 118 GHz. Modeling studies and some initial aircraft measurements indicate that differential scattering between channels with similar weighting functions at 60 GHz and 118 GHz might be a promising new technique to derive precipitation information globally, with little or no sensitivity to surface emissivity variations (Bauer and Mugnai 2004; Bauer and Moreau 2005). With respect to the operational satellites NOAA-15, 16, and 17, NOAA produces a set of global standard precipitation products for the AMSU (Bennartz et al. 2002). Other algorithms were derived by Weng et al. (2003), McCollum and Ferraro (2003), Chen and Staelin (2003), and Staelin and Chen (2000). The use of the scattering signal at typical passive microwave window frequencies (e.g., 89 and/or 150 GHz) alone allows for identification of precipitating areas, but the assignment of instantaneous precipitation rates at the ground is typically associated with large uncertainties. Some of the reasons for these uncertainties are outlined below. Recent studies in support of now-casting therefore attempt to separate different rainfall intensity classes instead of rain rates at the ground (Weng et al. 2003). The data is then presented in the form of false color images, where the color code gives a visual impression of how strongly it is raining. While this approach is not suitable for climate research or data assimilation purposes, it provides forecasters with a product that is easy to interpret.
R. Bennartz
168
2.2
Discrimination between stratiform and convective precipitation
Several recent studies have also pointed out that the discrimination between convective and stratiform rainfall is of utmost importance to derive meaningful rain rates from the scattering signal. The relation between surface rain rate and scattering-based precipitation estimates is strongly dependent on the type of precipitation event (Bennartz and Michelson 2003; and references therein). These differences are mainly caused by variations in ice particle size distribution, shapes, and type (Skofronick-Jackson and Wang 2000; Petty 2001a; b; Bennartz and Petty 2001; Skofronick-Jackson et al. 2002, 2003). Many state-of-the-art rain retrieval algorithms account for the differences between stratiform and convective precipitation either explicitly (McCollum and Ferraro 2003) or implicitly (Staelin and Chen 2000; Chen and Staelin 2003). Implicit methods usually use a wide range of different frequencies around 54 and 183 GHz and apply a joint statistical retrieval technique (Staelin and Chen 2000; Chen and Staelin 2003 use a neural network) that was developed for a variety of different conditions. The underlying assumption is that the different nature of convective and stratiform precipitation is resembled in the observed spectral signature and can thus be accounted for in the multichannel inversion. Explicit methods classify the type of precipitation and assign different rain rate to brightness temperature relations depending on the classification. The obvious advantage of the explicit classification methods is that an additional parameter (type of precipitation) is derived that helps validate the algorithms.
3 DETECTION OF SNOWFALL AT THE GROUND The discrimination of frozen from liquid precipitation at the ground has been identified as a crucial issue, especially in high- and mid-latitude areas where snow coverage varies widely on annual and interannual time scales. From a passive microwave standpoint the determination of the phase of falling precipitation at the ground is an ill-posed problem, since passive measurements are only sensitive to column integrated quantities. For instance, if the precipitation melts just a few meters above the surface, it will not have any measurable “liquid” signature. Passive microwave instruments are therefore not directly sensitive to the phase of falling precipitation at any given level. Indirectly, the phase of precipitation at the surface can be determined with some accuracy by assessing the height of the freezing level. If the freezing level is known, the only additional information needed is what distance it takes for the falling precipitation to be melted. A reasonable value for this is in the order of 500–600 m for snowfall (this number will be
Passive Microwave Precipitation Measurements at Mid- and High Latitudes
169
substantiated in Section 3.2. In weakly precipitating systems there are basically two ways to assess the height of the freezing level: (1) If the precipitation is not too intense, it is possible to derive the temperature profile, e.g., from AMSU sounding channels. This can either be done explicitly or implicitly. Recent work (Kongoli et al. 2003) using the AMSU 54 GHz and 183 GHz sounding channels shows that the discrimination of frozen from liquid precipitation can be achieved empirically from satellite measurements. (2) Over ocean it is also possible to infer the freezing level with some accuracy from the water vapor column amount (WVCA) inferred in or near the precipitating area (Petty 1994). Using this method, two pieces of information, in principle, are needed to derive the phase of falling precipitation at the ground: precipitation identification and the WVCA for each pixel. While over ocean areas this method allows to directly infer the height of the freezing level from satellite data alone, over land areas the high surface emissivity does not permit deriving WVCA and the method would have to rely on auxiliary data. In this paper we show some initial results applying this second technique to AMSR data in the northern hemisphere.
3.1
Algorithm description and estimated accuracy
We use the algorithm described in Petty (1994) to infer the freezing level from the WVCA in precipitating areas. Precipitation is identified using the scattering index defined in Petty (1994) which has been readjusted to AMSR-E (Petty, personal communication). We assume observations with scattering index at 89 GHz larger than 5 K to likely be precipitating. For those pixels, Fig. 1 shows the relation between freezing level and WVCA as published in (Petty 1994) and also gives a range of variability due to the above uncertainties. Assuming that if the freezing level falls below 0.5 km the precipitation will be completely or partially frozen, we can expect that a WVCA below 8 kg m–2 will correspond to frozen precipitation and a WVCA above 13 kg m–2 will correspond to liquid precipitation. In the transition range the estimate will be more uncertain and at 11 kg m–2 there is a 50% chance that the precipitation is frozen. The accuracy of this method depends on several factors, namely (1) the accuracy of the WVCA, (2) the accuracy of the relation between WVCA and the freezing level, and (3) the accuracy to which the falling distance of frozen precipitation is known. While each of these parameters is difficult to isolate, the functional behavior of the relation allows for the derivation of some information on the order of magnitude of the expected error. Note that an integral validation of the method against independent data will be performed in Section 3.2.
R. Bennartz
170
Figure 1. Freezing level as a function of water vapor content in precipitating areas following Petty (1994). The dashed curves give the standard deviation resulting from a 2 kg m–2 error in the water vapor column amount. The two horizontal lines are at 0.5 km and zero freezing level.
Figure 2 shows a sample AMSR overpass and some auxiliary data. The overpass captures a warm front in the southern part of the Baltic Sea, which is associated with a wide rainband that can be seen in the radar data (Fig. 2a). The variation in WVCA across the front, as well as the increase of the freezing level, can be clearly identified in the AMSR-E imagery. The aforementioned range of WVCA between 8 and 13 kg m–2 translates in Fig. 2 to a spatial distance of roughly 120 km. We might therefore conclude that the delineation of frozen from liquid precipitation with this method is associated with a geographic uncertainty of about one degree. Figure 2 represents a smoothly varying warm front. For a cold front the transition between warm, moist regions and cold dry regions is usually much smoother, so a smaller error is expected for those cases.
3.2
Validation with surface observations
In order to validate the above algorithm, we have collocated AMSR-E observations with coastal weather observer reports for the northern hemisphere between 50 degrees and 80 degrees north for the period October 2002 until February 2003. Valid collocated satellite and surface observations had
Passive Microwave Precipitation Measurements at Mid- and High Latitudes
171
to be spatially separated by less than 50 km and temporally by less than 1 h. The AMSR-E observation had to be completely over ocean (this information is provided in the AMSR-E level 2 ocean suite) and the aforementioned scattering index had to indicate precipitation from the AMSR-E data. The weather station had to be no higher than 50 m above sea level. (a)
(b)
(c)
(d)
Figure 2. (a) Radar reflectivity derived from operational weather radar composites 0145 UTC 22 October 2002, (b) AMSR-E water vapor column amount, (c) AMSR-E derived freezing level height, (d) 0000 UTC synoptic weather chart.
From this data set we extracted all collocated observations where the weather observer unambiguously reported either snowfall or rainfall. This led to a data set of 449 collocated snowfall observations and 741 collocated rainfall observations. Figure 3 (upper panel) shows the occurrence of different freezing level heights for the rain and the snow data set. The data is binned in 100 m intervals. For freezing levels between roughly zero and 1,500 m, mixed
172
R. Bennartz
observations are reported. Only rainfall is reported above 1,500 m, while very few observations report rain below zero meters. The lower panel of Fig. 3 gives the likelihood that a precipitation event seen by AMSR-E is associated with snowfall (the ratio between the snowfall cases and the total number of cases in each bin multiplied by 100). The smooth curve is a cumulative gaussian fit through the data.
Figure 3. The upper panel shows the frequency of occurrence of freezing levels (binned in 100 m intervals) for AMSR-E observations close to coastal synoptic observation sites, that either reported snowfall (449 cases) or rainfall (741 cases) at the ground. The lower panel shows the relative likelihood of frozen precipitation derived from the above histograms and a cumulative gaussian curve fitted to the data (smooth line). For details see text.
For a freezing level height of 600 m the likelihood that the precipitation falls as rain is exactly 50%. For freezing levels between roughly zero and 1,200 m the likelihood of frozen precipitation decreases smoothly to zero. Note that in reality for those freezing level heights a mixture of frozen and liquid precipitation at the ground is likely to occur and that a hard separation between frozen and liquid precipitation does not agree with our everyday experience of surface precipitation.
Passive Microwave Precipitation Measurements at Mid- and High Latitudes
3.3
173
Application examples
Figure 4 provides an application example for the above method. Panels 4a and 4c show the fraction of snowfall events over the North Atlantic for October 2002 and January 2003, respectively. Panels 4b and 4d give the total number of precipitating pixels per box as identified by a scattering index larger than 5 K. In January, the precipitation north of about 70 degrees falls entirely as snow, whereas in October, especially to the west of Greenland, the transition pattern is more longitudinal. Hudson Bay and large parts of the eastern Atlantic close to Greenland are ice-covered in January and hence no information about precipitation can be inferred. (a)
(b)
(c)
(d)
Figure 4. (a) Fraction of precipitation events with snowfall at the surface over the northern Atlantic for October 2002, (b) total number of precipitation events found per 0.5 × 0.5 degree box, (c) same as (a) but for January 2003 (d) same as (b) but for January 2003. (see also color plate 8)
174
R. Bennartz
4 SAMPLING Precipitation is highly variable in time and exhibits a strong diurnal cycle, especially in the summer, over land. Figure 5 shows the contribution of isolated convective precipitation to the total precipitation in the Baltic area as a function of daytime and month for the years 2001 and 2002. These data were generated using an automated classification algorithm based on BALTEX Radar Data Centre (Michelson et al. 2000) data for these two years. The algorithm and its accuracy are described in detail in Walther and Bennartz (2006). As can be seen from Fig. 5, isolated convective precipitation contributes up to 90% of the total precipitation in summer time. The maximum of convective precipitation occurs in the afternoon between 3 pm and 5 pm local time. In winter time, precipitation is governed either by frontal systems or by cold air outbreaks over warm water surfaces that trigger intensive convection. Both types of events do not exhibit a strong diurnal cycle. In Fig. 5, we provide the approximate central overpass times of NOAA15 and NOAA-16, and AQUA over the Baltic area. The convective maximum at around 4 pm local time is not captured by any of these satellites.
Figure 5. Fraction of rainfall caused by isolated convective events over the Baltic Sea for the years 2001 and 2002 as function of local time and month. Overlaid are central overpass times for NOAA-15 (black), NOAA-16 (dark gray), and AQUA over the Baltic area.
Passive Microwave Precipitation Measurements at Mid- and High Latitudes
175
As a result of the above considerations, the current coverage of high latitude areas with operational satellites equipped with passive microwave remotesensing instruments does not allow to fully resolve the diurnal cycle of precipitation at latitudes between 50 and 70 degrees. Additionally, the use of visible and near infrared data from geostationary satellites to account for these current shortcomings is hampered by the satellites’ oblique incidence angle at higher latitudes.
5 CONCLUSIONS Within the next couple of years passive microwave measurements of precipitation will become increasingly utilized in areas such as climate research or data assimilation. The new generation of passive microwave radiometers (such as AMSR) allow screening for and retrieving precipitation globally at a high spatial resolution. However, stable operational algorithms that work at high latitudes are still in the process of being developed and validated. In this paper we show that for those sensors the determination of precipitation phase at the surface from passive microwave measurements alone is possible within reasonable error bars. The simple method we present here is not restricted to AMSR-E. It can be used for any passive microwave sensor that allows the derivation of water vapor column amount and information about the presence of rain. This is basically possible from all current passive microwave imagers (AMSU, AMSR, TMI, SSM/I, SSMIS, etc). Since the water vapor column amount can not be inferred from passive microwave observations over land, the use of this technique is hampered over land surface. However, if the water vapor column amount can be obtained from other sources (e.g., numerical weather prediction models) with a reasonable accuracy, the method can be applied over land surfaces too, given that the passive microwave observations can identify precipitating areas. Over land, a higher temporal sampling is crucial to resolve the strong diurnal cycle of precipitation especially in the summer months. Forthcoming satellite missions such as GPM and its European counterpart EGPM will allow to fully resolve the diurnal cycle of precipitation. In this context EGPM will especially be of crucial importance. With its planned equator crossing time around 3 pm to 4 pm, it will exactly capture the afternoon maximum of summer precipitation. Additionally, the combination of a sensitive nadir-pointing precipitation radar with passive microwave sensors on EGPM will help address many of the aforementioned issues associated with the relation between the scattering signal and the surface rain rate. Acknowledgements: I would like to thank Andi Walter (Free University of Berlin) and Mark Kulie (University of Wisconsin) for their help. I would like
176
R. Bennartz
to express my gratitude to Grant Petty (University of Wisconsin) who has provided his algorithms to calculate scattering and polarization indices on AMSR data. Daniel Michelson (Swedish Meteorological and Hydro-logical Institute) has provided the radar data. This research was funded by NASA via grant NAG 5-11106 and in part by ESA-ESTEC Contract No. 17272/03/NL/GS.
6 REFERENCES Bauer, P. and A. Mugnai, 2004: Precipitation Profile retrieval using temperature sounding microwave observations. J. Geophys. Res., 108(D23), doi:10.1029/2003JD003572. Bauer, P., E. Moreau, and S. Di Michele, 2005: Hydrometeor retrieval accuracy from microwave window and sounding channel observations. J. Appl. Meteor., 44, 1016–1032. Bennartz, R., A. Thoss, A. Dybbroe, and D. B. Michelson, 2002: Precipitation analysis using the Advanced Microwave Sounding Unit in support of nowcasting applications. Meteorol. Appl., 9, 177–189. Bennartz, R. and D. B. Michelson, 2003: Correlation of precipitation estimates from spaceborne passive microwave sensors and weather radar imagery for BALTEX PIDCAP. Int. J. Remote Sens., 24, 723–739. Bennartz, R. and G. W. Petty, 2001: The sensitivity of microwave remote sensing observations of precipitation to ice particle size distributions. J. Appl. Meteor., 40, 345– 364. Chen, F. W. and D. H. Staelin, 2003: AIRS/AMSU/HSB precipitation estimates. IEEE Trans. Geosci. Remote Sens., 41, 410–417. Goodrum, G., K. B. Kidwell and W. Winston (Eds.), 1999: NOAA KLM User's Guide, Published by NOAA. Hollinger, J. R., R. Lo, G. Poe, R. Savage, and J. Pierce, 1987: Special Sensor Microwave/Imager User’s Guide. NRL Tech. Rep., Naval Research Laboratory, Washington, DC, 177 pp. Kawanishi, T., T. Sezai, Y. Ito, K. Imaoka, T. Takeshima, Y. Ishido, A. Shibata, M. Miura, H. Inahata, and. R. W. Spencer, 2003: The Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), NASDA's contribution to the EOS for global energy and water cycle studies. IEEE Trans. Geosci. Remote Sens., 41, 184–194. Kongoli, C., P. Pellegrino, and R. R. Ferraro, 2003: A new snowfall detection algorithm over land using measurements from the Advanced Microwave Sounding Unit (AMSU). Geophys. Res. Lett., 30, Art. No. 1756. Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Tropical Rainfall Measuring Mission (TRMM) Sensor Package. J. Atmos. Oceanic Technol., 15, 809–817. McCollum, J. R. and R. R. Ferraro, 2003: Next generation of NOAA/NESDIS TMI, SSM/I, and AMSR-E microwave land rainfall algorithms. J. Geophys. Res., 108(D8): Art. No. 8382. Petty, G. W., 2001a: Physical and microwave radiative properties of precipitating clouds. Part I: Principal component analysis of observed multichannel microwave radiances in tropical stratiform rainfall. J. Appl. Meteor., 40, 2105–2114. Petty, G. W., 2001b: Physical and microwave radiative properties of precipitating clouds. Part II: A parametric 1D rain-cloud model for use in microwave radiative transfer simulations. J. Appl. Meteor., 40, 2115–2129. Shibata, A., K. Imaoka, and T. Koike, 2003: AMSR/AMSR-E level 2 and 3 algorithm developments and data validation plans of NASDA. IEEE Trans. Geosci. Remote Sens., 41, 195–203.
Passive Microwave Precipitation Measurements at Mid- and High Latitudes
177
Shin, D. B. and C. Kummerow, 2003: Parametric rainfall retrieval algorithms for passive microwave radiometers. J. Appl. Meteor., 42, 1480–1496. Skofronick-Jackson, G. M., and J. R. Wang, 2000: The estimation of hydrometeor profiles from wideband microwave observations. J. Appl. Meteor., 39, 1645–1656. Skofronick-Jackson, G. M., A. J. Gasiewski A. J., and J. R. Wang, 2002: Influence of microphysical cloud parameterizations on microwave brightness temperatures. IEEE Trans. Geosci. Remote Sens., 40, 187–196. Skofronick-Jackson, G. M., J. R. Wang, and G. M. Heymsfield, 2003: Combined radiometerradar microphysical profile estimations with emphasis on high-frequency brightness temperature observations. J. Appl. Meteor., 42, 476–487. Smith, E. A. and co-authors (2004): Global Precipitation Measurement (GPM) Mission: Past progress and future prospects. GPM Report Series # 6. Available gpm.gsfc.nasa.gov. Staelin, D. H. and F. W. Chen, 2000: Precipitation observations near 54 and 183 GHz using the NOAA-15 satellite. IEEE Trans. Geosci. Remote Sens., 38, 2322–2332. Weng, F. Z., L. M. Zhao, and R. R. Ferraro, 2003: Advanced microwave sounding unit cloud and precipitation algorithms. Radio. Sci., 38, Art. No. 8068. Wilheit, T., C. D. Kummerow, and R. Ferraro, 2003: Rainfall algorithms for AMSR-E. IEEE Trans. Geosci. Remote Sens., 41, 204–214. Petty, G. W., 1994: Physical retrievals of over-ocean rain rate from multichannel microwave imagery, Part 2: Algorithm implementation. Meteorol. Atmos. Phys., 54, 101–121. Michelson, D., T. Andersson, J. Koistinen, C. Collier, J. Riedl, J. Szturc, U. Gjertsen, A. Nielsen, and S. Overgaard, 2000: BALTEX Radar DATA Centre Products and their Methodologies. SMHI Report, RMK 90, SMHI, SE-60176 Norrköping, Sweden. Walther, A. and R. Bennartz, 2006: Radar-based precipitation type analysis in the Baltic area. Tellus, 58, 331–343.
14 THE GODDARD PROFILING ALGORITHM (GPROF): DESCRIPTION AND CURRENT APPLICATIONS William S. Olson1, Song Yang2, John E. Stout2, and Mircea Grecu3 1
Joint Center for Earth Systems Technology, University of Maryland, Baltimore, MD, USA George Mason University, Fairfax, VA, USA 3 Goddard Earth Sciences and Technology Center and NASA/Goddard Space Flight Center, Greenbelt, MD, USA 2
1 INTRODUCTION The use of Bayesian estimation methods in passive microwave radiometry follows from a recognition that the total information content of radiometer observations is insufficient to determine a “unique” estimate of surface rain rate or precipitation vertical profile. In other words, for a given set of multifrequency microwave observations at a given location, there exist several precipitation profiles that are radiatively consistent with the observations, and so iterative methods for seeking a unique solution (Smith et al. 1994) would not necessarily return a better estimate. Also, being non-iterative, Bayesian methods are relatively computationally efficient, since iterative forward radiance calculations are not required. Here, the Goddard Profiling Algorithm (GPROF) is described, and applications of the most recent implementation (Version 6) of the algorithm are presented and critiqued.
2 ALGORITHM DESCRIPTION GPROF is based upon a Bayesian technique originally described in Kummerow et al. (1996) with an extension to latent heating estimation by Olson et al. (1999). A summary of more recent developments in the algorithm can be found in Kummerow et al. (2001). In the algorithm, cloudresolving model simulations, coupled to a radiative transfer code, are used to 179 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 179–188. © United States Government 2007.
W. S. Olson et al.
180
generate a large supporting database of simulated precipitation/latent heating vertical profiles and corresponding upwelling microwave radiances. Given a set of observed multichannel microwave radiances from a particular sensor, the entire database of simulated radiances is scanned; the “retrieved” profile is composited from those profiles in the database that correspond to simulated radiances consistent with the observed radiances. Formally, a GPROF estimate of profile parameters, Ê[x], is given by:
ˆ [ x] = ∑ x E k k
{ (
exp −0.5 I ( x k ) − I s o
) (S T
I
+ OI )
−1
( I s ( x ) − I o ) + C} k
Mˆ
(1)
where the model profile vector xk contains all parameters, including the surface rain rate, convective rain rate, liquid/ice-phase precipitation and latent heating profiles, corresponding to the simulated radiance indices, IS(xk). The radiance indices, constructed from radiances at the different radiometer channel frequencies/polarizations, are the normalized polarization and scattering indices defined by Petty (1994). IO is a vector of sensor observed radiance indices, similarly defined. SI and OI are error covariance matrices of the simulated and observed microwave radiance indices, respectively. Additional information regarding the observed profile, such as estimates of the area fractions of convective and stratiform rain within the nominal satellite
Figure 1. Schematic of the GPROF algorithm.
The Goddard Profiling Algorithm (GPROF): Description and Current Applications
181
footprint (14 × 14 km2 for TMI) and the freezing level, is included in the constraint term, C. The summation in (1) is over all simulated profiles/radiance indices in the supporting cloud-radiative model database. ˆ is a normalization factor. M The GPROF algorithm is shown schematically in Fig. 1. Simulated radiometer footprints and calculated radiance indices of the algorithm’s database are shown at right, while the observed footprint and radiance indices are at left. Observed radiance indices are compared to each set of simulated radiance indices in the database – the precipitation/latent heating parameters of those simulations that are more radiatively consistent with the observations (e.g., model #1 in the figure) contribute strongly to the GPROF estimate of the parameters, while those simulations that are less consistent radiatively contribute much less. Therefore, the algorithm performs a kind of radiative filtering of the database. Since multichannel passive microwave observations contain limited information regarding precipitation and related parameters, there are, in fact, a distribution of these parameters that are consistent with any set of observations at a given footprint location. Equation (1) gives the mean of this distribution, but it is also possible to calculate the variance of the distribution for a single estimated parameter using:
{(
)}
σˆ 2 [x] = Eˆ x − Eˆ [x]
2
(2)
which yields a measure of the uncertainty in the estimate of x due to the limited information content of the observations. The uncertainty represented by (2) would exist even if the cloud-radiative model simulations in the GPROF supporting database and the radiometer observations were perfect, and so additional uncertainties in GPROF estimates due to modeling or observational errors may occur. However, since true validation of precipitationrelated quantities using independent observations is difficult, (2) at least provides a lower bound on the error of GPROF estimates- a basic “building block” for estimates of the random error in derived products. Algorithmderived estimates of random error for two case studies will be presented in GPROF-Applications, below.
3 EVALUATION OF RAIN RATE ESTIMATES In collaboration with several TRMM investigators, independent estimates of rain rate, convective proportion, and latent heating were collected and compared to GPROF algorithm estimates based upon TRMM Microwave Imager (TMI) radiance observations. Although independent evaluation of surface rain rate estimates is important in its own right, surface rain rate can also be viewed as a proxy for vertically integrated latent heating, and the
182
W. S. Olson et al.
convective proportion is an indicator of heating profile shape. Ground validation radar estimates of rain rate and convective proportion coincident with TRMM overpasses of significant rain events were collected from the Melbourne, FL and Kwajalein (Marshall Islands) sites over an entire year (1998). Surface rainfall rate estimates from the Version 5 TMI, PR, and TMI-PR algorithms and ground-based radar measurements from Melbourne, FL and Kwajalein, Marshall Islands showed varying degrees of agreement. The Version 6 TMI estimates generally showed greater consistency with ground validation radar rain observations. The analysis yielded a correlation of 0.83 between Version 6 TMI estimates and Kwajalein ground radar rain rates at 50 km resolution. The correlation between Version 6 TMI and PR estimates at Kwajalein at the same resolution was 0.84.
Figure 2. Scatterplot histogram of Version 6 PR and TMI instantaneous, 60 km resolution rain rate estimates over ocean for the month of July 2001.
Since comparisons to ground validation radars can bias error statistics toward local conditions, TMI rain rate and convective proportion estimates were also compared to PR estimates over the entire TRMM observing domain. Shown in Fig. 2 is a comparison of all coincident Version 6 PR and TMI rain estimates over ocean for the month of July 2001. Estimated rain rates greater than a few tenths of a mm h–1 are strongly correlated; the low bias of TMI estimates at very low rain rates does not contribute appreciably to the total rain of the distribution. Error modeling of TMI rain estimates suggests that 70–90% of the random difference between TMI and PR
The Goddard Profiling Algorithm (GPROF): Description and Current Applications
183
instantaneous rain estimates at half-degree resolution can be explained by random errors in the TMI estimates; the remainder is due to errors in the PR estimates and differences in the spatial sampling of rain by the two instruments within half-degree boxes.
4 GPROF APPLICATIONS 4.1 Instantaneous precipitation estimates Shown in Fig. 3 and 4 are applications of GPROF Version 6 to TMI observations of two precipitation systems. Hurricane Bret developed in the Gulf of Mexico in mid-August, 1999, and attained hurricane strength on 21 August; see Lawrence et al. (2001). The GPROF estimates of precipitation in Fig. 3 are derived from TMI observations of Bret on 21 August at 2245 UTC, as the storm moved northward towards the coast of Texas, USA. Estimated surface rain rates are most intense (>10 mm h–1) in the nearly circular eyewall of Bret, with somewhat less intense rains in bands to the northeast of the eye. Note that the resolution of instantaneous rain estimates from GPROF is 14 km, and so rains are possibly more intense at subfootprint resolution. The eyewall and innermost rain band are surrounded by less intense, stratiform rains (<5 mm h–1). Errors in estimates of instantaneous rain rates, given by (2), are shown in the top-right panel of Fig. 3. Note that the errors in estimates of lighter rains can be equal to or greater than 100%; while for the most intense rains, errors are ~60%. These errors are characteristic of satellite passive microwave estimates of rain rate, although errors in rain rate estimates generally increase with increasing rain rate, percentage errors tend to decrease with rain rate. These errors are substantially reduced by space- and timeaveraging (see Bauer et al. 2002). Total precipitation water content (the combined water content of all precipitating hydrometeors) and Q1–QR estimates along the transect A-B are shown in the lower panel of Fig. 3. Note that the deepest precipitation structures and greatest heating rates along the transect are associated with the eyewall of Bret. The heating vertical structure is “convective” in the eyewall, in the sense that heating rates are positive through most of the troposphere with a maximum at midlevels. Rain rates and heating rates are greater in the left branch of the eyewall, relative to the right branch in the transect, which is consistent with the higher-resolution PR-derived distribution of rain intensity (not shown). Although the rain rates are a relative minimum in the eye, the TMI does not detect a “clear” eye due to the limited horizontal resolution of the instrument and its oblique viewing angle (52.8° from vertical). The quasi-stratiform rainband to the northeast of the eye exhibits maximum water contents just
184
W. S. Olson et al.
Figure 3. GPROF estimates of surface rain rate (upper left panel), surface rain rate error (upper right panel), and cross sections of total precipitation and Q1–QR along the A–B transect (lower panel) from TMI observations of Hurricane Bret on 21 August 1999. In the lower panel, total precipitation (rain, graupel, and snow) water contents are shaded and heating rates are contoured in white at levels of –3, –1, 1, 3, 6, 9, and 12 K h–1.
above the freezing level (estimated to be ~4.7 km), with relatively weak heating aloft and evaporative cooling below the freezing level. The outer rainband near “B” in the transect has convective structure but is much shallower in the vertical than the eyewall. Although GPROF applications to observations of tropical precipitation systems have been a primary focus of research in the past, the recent expansion of the algorithm’s cloud resolving model database to include midlatitude simulations has made extra-tropical applications possible. The GPROF estimates shown in Fig. 4 are derived from two TRMM overpasses of a baroclinic system on 17 January 2000 near 09 UTC. The cold-frontal rainband is primarily contained in the lower swath, with a frontal occlusion and post-frontal precipitation contained in the upper swath. Maximum rain rates are seen in the upper portion of the cold frontal band. Note that the vertical precipitation and heating structures along the A–B transect are much shallower than those retrieved from the Bret data. Maximum water contents (0.6 g m–3) and heating rates (3 K h–1) are consistent with weaker updrafts in the extra-tropical system.
The Goddard Profiling Algorithm (GPROF): Description and Current Applications
185
Figure 4. Same as Fig. 3, but for TMI observations of an extra-tropical cyclone over the North Atlantic on 17 January 2000. The two swath sections depicted represent observations that were separated by ~90 min of time, causing a slight discontinuity in the GPROF estimates near the inter-swath boundary.
Finally, note that the instantaneous heating estimates shown in Figs. 3 and 4 are sometimes subject to large random errors, given the indirect inference of heating from TMI precipitation signatures. These figures are presented to indicate the qualitative plausibility of GPROF-derived heating structures, even though the individual heating rate estimates may contain large quantitative errors. Random errors in GPROF heating estimates are reduced significantly with space- and time-averaging.
4.2 Large-scale precipitation and latent heating GPROF estimates of precipitation and latent heating based upon TMI data can be aggregated to produce estimates of their large-scale distributions in the Tropics and Sub-tropics. Shown in Fig. 5 are GPROF estimates aggregated in 2.5° × 2.5° latitude/ longitude boxes over the month of January 2000. Note that the distributions in Fig. 5 are not “smooth” in part because the temporal sampling by TMI of a given 2.5° box is limited (about 1 day – 1 near the equator). Temporal sampling by microwave radiometers is expected to improve in the future, as additional satellite radiometers will be launched as part of the follow-on mission to TRMM. Nevertheless, the main features of
186
W. S. Olson et al.
global rain distributions are captured by TMI observations, including the Inter-Tropical Convergence Zone (ITCZ), the South Pacific Convergence Zone, and rains along the southern portion of the mid-latitude storm tracks in the Northern Hemisphere. Rains in the Tropics are predominantly convective; however, regions of peak rainfall along the ITCZ exhibit a relative minimum of convection, indicating the contribution of organized mesoscale convective systems to the total rainfall. Latent heating basically follows the pattern of surface rain rate. Heating is generally a maximum near 7 km altitude where the rainfall is most intense- another signature of organized precipitation systems. In regions where less organized convection dominates the spectrum of precipitation systems, the heating maximum occurs at lower altitudes.
5 CONCLUDING REMARKS This brief survey is intended to illustrate some of the strengths and weaknesses of precipitation and latent heating estimation using a Bayesian method (GPROF). It should be noted that any remote-sensing method is limited by the information content of the input observations. GPROF was designed to exploit not only the multifrequency sensing capability of microwave radiometers, but also information drawn from the horizontal distributions of observed radiances. In the future, the identification of precipitation system type (isolated convection, short-lived convective lines, squall lines, tropical cyclones, extra-tropical cyclones) from radiometer observations may lead to more specific estimates from the algorithm. Latent heating estimates require contextual information and should benefit from the identification of system type. Another important area of study is the construction of the algorithm’s supporting database. First, there are potential biases in the cloud-radiative model calculations; in particular the simulation of precipitation-sized ice aloft has been studied (Tao et al. 2003). To the extent that biases in the cloud-radiative model simulations can be quantified and the physical mechanisms understood, they should be corrected. In addition, the general representativeness of the database must be examined. That is, the population of simulated precipitation structures in the database should reflect the natural population of these structures. High-resolution Precipitation Radar (PR) observations can be used as a guide to determine the natural population of precipitation structures. Missing structures should be included in the algorithm’s database and artificial structures eliminated. Since the a priori distribution of precipitation structures in the database has a significant impact on precipitation and heating estimates (Shin and Kummerow 2004), due to the limited information content of the radiometer observations, proper database construction is critical.
The Goddard Profiling Algorithm (GPROF): Description and Current Applications
187
Figure 5. Mean surface rain rates, convective rain proportions, and latent heating rates at 7 and 3 km altitude, derived from TMI observations from January 2000, using the GPROF algorithm (see also color plate 9).
The authors are currently conducting the “validation phase” of the Version 6 GPROF algorithm. Rain rate estimates from the PR and ground-based radar provide reference estimates. Latent heating vertical profiles from the algorithm are being compared to rawinsonde budget estimates, such as those from SCSMEX (Johnson and Ciesielski 2002), while heating profiles inferred
188
W. S. Olson et al.
from dual-Doppler radar observations in combination with radiosondederived thermal structure provide another reference.
6 REFERENCES Bauer, P., J.-F. Mahfouf, W. S. Olson, F. S. Marzano, S. Di Michele, A. Tassa, and A. Mugnai, 2002: Error analysis of TMI rainfall estimates over ocean for variational data assimilation. Quart. J. Roy. Meteor. Soc., 128, 2129–2144. Johnson, R. H. and P. E. Ciesielski, 2002: Characteristics of the 1998 summer monsoon onset over the northern South China Sea. J. Meteor. Soc. Japan, 80, 561–578. Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 1213–1232. Kummerow, C., Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, D.-B. Shin, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. Lawrence, M. B., L. A. Avila, J. L. Beven, J. L. Franklin, J. L. Guiney, and R. J. Pasch, 2001: Atlantic hurricane season of 1999. Mon. Wea. Rev., 129, 3057–3084. Olson, W. S., C. D. Kummerow, Y. Hong, and W.-K. Tao, 1999: Atmospheric latent heating distributions in the Tropics derived from passive microwave radiometer measurements. J. Appl. Meteor., 38, 633–664. Petty, G. W., 1994: Physical retrievals of over-ocean rain rate from multichannel microwave imagers. Part I: Theoretical characteristics of normalized polarization and scattering indices. Meteor. Atmos. Phys., 54, 79–99. Shin, D.-B. and C. Kummerow, 2004: Parametric rainfall retrieval algorithms for passive microwave radiometers. J. Appl. Meteor., 42, 1480–1496. Smith, E. A., C. Kummerow, and A. Mugnai, 1994: The emergence of inversion-type profile algorithms for estimation of precipitation from satellite passive microwave measurements. Remote Sens. Reviews, 11, 211–242. Tao, W.-K., C.-L. Shie, J. Simpson, S. Braun, R. H. Johnson, and P. E. Ciesielski, 2003: Convective systems over the South China Sea: Cloud-resolving model simulations. J. Atmos. Sci., 60, 2929–2956.
15 PAST, PRESENT AND FUTURE OF MICROWAVE OPERATIONAL RAINFALL ALGORITHMS Ralph R. Ferraro NOAA/National Environmental Satellite, Data and Information Service, College Park, MD, USA
1 INTRODUCTION Scientists throughout the worldwide remote-sensing community have been involved in the development of rain rate retrieval algorithms for use on operational passive microwave (MW) satellite sensors for nearly 25 years. The early motivation was in the preparation for the Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave/Imager (SSM/I), which was first launched in June 1987. Work with early research satellites flown by the National Aeronautics and Space Administration (NASA) such as the Nimbus-5 & 6 Electronically Scanning Microwave Radiometer (ESMR) and the Nimbus-7 Scanning Multichannel Microwave Radiometer (SMMR) led to the development of research algorithms that were then formulated for use in an operational environment (Hollinger 1990). Within the USA, as confidence grew with the ability of the MW sensors to accurately map rain areas and offer a qualitative assessment of the rain intensity, various components of the National Oceanic and Atmospheric Administration (NOAA) began to request these products to support their operations and fulfill several of NOAA’s missions, including advanced short term weather forecasts and warnings, and seasonal to interannual climate monitoring. The understanding of the rainfall signatures at MW frequencies has increased over the last 25 years (i.e., compare the early work of Wilheit et al. 1977 to recent work, Wilheit et al. 2003). These techniques primarily rely on the emission signal of rain drops over the oceans at frequencies at or below 37 GHz and the scattering signal of ice particles in the precipitation layer 189 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 189–198. © United States Government 2007.
190
R. R. Ferraro
over land at frequencies at or above 85 GHz. Perhaps just as dramatic has been the change in computer processing capability, data storage, and data communications. This has led to a vastly different way in which operational satellite products can be generated. No longer are agencies such as NOAA’s National Environmental Satellite, Data and Information Service (NESDIS) or the US Navy’s Fleet Numerical Meteorology and Oceanography Center (FNMOC) forced to share computer resources on mainframe computers for all of their product systems. Now, satellite derived products like precipitation can be generated on dedicated workstations and data delivered to customers through fast data communications. And no longer do data have to be stored on large magnetic data tapes and be manually retrieved by a tape librarian. Although not an operational mission, NASA’s Tropical Rainfall Measurement Mission (TRMM) perhaps serves as the best paradigm for the current state of the art in routine rainfall retrieval from MW sensors through its TRMM Science Data and Information System (TSDIS). It is the purpose of this paper to briefly describe the evolution of operational MW derived rainfall retrieval algorithms, provide examples of their application at NOAA, and present some initial thoughts as to what the future has to offer.
2 SSM/I OPERATIONAL ALGORITHM The first SSM/I instrument was launched on 19 June 1987 aboard the DMSP F-8 satellite. Other instruments have successfully operated on board the F-10 (December 1990), F-11 (November 1991), F-13 (March 1995), F-14 (April 1997) and F-15 (December 1999) satellites, the latter three of which are currently operational. The operational DMSP SSM/I algorithm developed at NOAA/NESDIS (henceforth denoted as the NESDIS algorithm) was based on the groundbreaking study of Grody et al. (1991), who utilized 19, 22 and 85 GHz measurements to detect the decrease in observed brightness temperatures caused by scattering from millimeter-sized ice particles associated with precipitation. They then applied this scheme to map global rainfall, while at the same time, remove false signatures caused by other radiometrically similar features such as desert sand, snow cover and sea-ice. This algorithm was further improved upon by Ferraro (1997) who introduced an emission component over ocean to expand the detection of oceanic rainfall and introduce empirically derived relationships to convert these signals into instantaneous rain rates. The NESDIS algorithm was adopted by FNMOC in the mid-1990s (Colton and Poe 1994) and has been running operationally for the last 10 years. During the 1990s, the requirement for an operational algorithm was vastly different than an algorithm that was being used in a R&D mode. Hence, this explains the rather simple nature in the NESDIS algorithm, which has been virtually unchanged for the past decade. Despite this characteristic, the
Past, Present and Future of Microwave Operational Rainfall Algorithms
191
rainfall product is widely used within the international community for a variety of applications ranging from short-term rainfall potential assessment (Fig. 1 and see Ferraro et al. 1999, 2002) to global, monthly rainfall monitoring (Fig. 2). Additionally, the Global Precipitation Climatology Project (GPCP) (Huffman et al. 1997) uses the land component of this algorithm (Ferraro 1997).
Figure 1. 24-h rainfall potential (inches) (bottom) derived from SSM/I instantaneous rain rates (top) on 1510 UTC, 18 September 2003 for Hurricane Isabel (see also color plate 7).
The evolution of the NESDIS algorithm did not come strictly from researchers at NOAA; it benefited substantially by the intercomparison programs sponsored by the GPCP and the NASA WetNet project (i.e., see
192
R. R. Ferraro
Ebert et al. 1996). These intercomparisons, although originally viewed as a “sort of competition” by its participants, actually have led to a rather unique bond between the algorithm developers worldwide and have benefited the science immensely! Additionally, NOAA’s and FNMOC’s willingness to distribute the products to fulfill user needs led to an important userdeveloper feedback process, which ultimately factored into significant algorithm improvements. For example, several changes to the screening for false signatures (that vary by region and season), improvement to stratiform rain over ocean (via an emission component to the algorithm) and a method to perform coastline retrievals (by expanding the land component of the algorithm to adjacent coastal waters) all have come about by participating in these intercomparisons and distributing the products to users. The lessons learned from this process have been invaluable!
Figure 2. December 2003 SSM/I derived rainfall anomaly based on a 1987–2002 base period using the algorithm of Ferraro (1997).
3 AMSU OPERATIONAL ALGORITHM The first Advanced Microwave Sounding Unit (AMSU) was placed into operation on the NOAA-15 satellite on May 1998. Subsequent AMSUs are operational on board the NOAA-16 (September 2000) and NOAA-17 (June 2002) satellites. This three-satellite constellation offers global observations approximately every 4 h. The heritage of the SSM/I algorithm was used to develop the original AMSU operational algorithm (Grody et al. 2001), even though there are significant differences between the two sensors (Table 1). Recent utilization of the high frequency measurements at 150 and 183 GHz has led to a substantially improved retrieval scheme that has been operational since August 2000 (Weng et al. 2003). By designing a brand new
Past, Present and Future of Microwave Operational Rainfall Algorithms
193
product generation system modelled after TSDIS, NESDIS was able to generate operational AMSU derived rain rates shortly after the launch of NOAA–15 through the Microwave Surface and Precipitation Products System (MSPPS) (Ferraro et al. 2002b). Shown in Figs. 3 and 4 are AMSU derived product composites at daily and monthly time scales, respectively, which highlight several of the derived parameters, including precipitation. Additionally, multiple satellite products that utilize all three NOAA AMSUs are being generated and have proven to be superior due to the improved temporal sampling. Figure 5 presents the validation of the combined satellite product for a case study over Australia as described by Ebert (2007). Table 1. Comparison of AMSU and SSM/I sensors.
Characteristic Primary Window Channels Polarization Scan Geometry
AMSU 23.8, 31.4, 50.3, 89, 150 GHz Mixed Cross Track: 0–48 degrees
FOV properties
Vary with view angle Fixed with frequency ~2200 km
Swath Width
SSM/I 19.4, 22.2, 37, 85.5 GHz V and H Conical: Fixed 45 degrees Fixed across scan Vary with frequency ~1400 km
Figure 3. NOAA-15 AMSU derived hydrological products for 22 January 2004.
194
R. R. Ferraro
Figure 4. NOAA-16 AMSU derived monthly product composite for December 2003.
Figure 5. Comparison of monthly derived rainfall from AMSU (all three satellites) (left) to gauge analysis (right) for December 2003 with accompanying statistics.
4 CONSOLIDATION OF ALGORITHMS FOR TRMM, AMSR AND SSM/I The current era of MW rain retrievals is such that even so-called research missions are becoming an integral part of satellite data used by operational agencies. These data are critical to fill voids in the current operational satellite series. For example, TRMM measurements and products are used extensively at NOAA and FNMOC in their monitoring of tropical cyclones.
Past, Present and Future of Microwave Operational Rainfall Algorithms
195
The multitude of satellite data sources does not come without its problems. Perhaps at the forefront are differences between the retrieval algorithms. Bias errors in the retrieved rain rates that vary between each satellite are troublesome for many applications. The only way to try to overcome this problem is to develop a consensus algorithm applicable to several sensors (of varying configurations) and based on the same physical principles. The first attempt to achieve such an algorithm has come about through recent improvements to the Goddard Profiling Algorithm (GPROF) (Kummerow et al. 2001). Motivated by close coordination within the NASA Earth Observing System (EOS) Aqua satellite (launched in May 2002) Advanced Microwave Scanning Radiometer (AMSR) science team, the land portion of the NESDIS algorithm was incorporated within GPROF by matching available precipitation rates/profiles (based on cloud model and radiative transfer model simulations) to similar radiance vectors from the NESDIS technique. Due to the flexible nature of GPROF, TRMM scientists have successfully developed the ocean portion of the algorithm to be applicable to multiple sensors. This version of GPROF, V5, has been utilized for SSM/I, TMI and AMSR (Wilheit et al. 2003). Continued improvements that include a convective/stratiform series of profiles used in the retrievals over land have been incorporated in GPROF V6, which will be implemented at TSDIS and for AMSR in March 2004 (McCollum and Ferraro 2003). Results indicate very little bias between the GPROF rain rates for all three of these sensors.
5 FUTURE AND DISCUSSION What does the future have to offer? First, there is a “convergence” of satellite missions on many levels. Beginning in late 2005, NOAA and EUMETSAT will combine their polar satellite programs, where EUMETSAT will contribute the METOP satellite series as one of the three operational polar orbiting satellites. It will contain a sensor package similar to AMSU. In addition, NOAA is consolidating the DMSP and NOAA polar programs into the National Polar-orbiting Operational Environmental Satellite System (NPOESS), ready for launch around 2008. NPOESS will contribute the other two operational satellites to the three satellite constellation. It will contain the most advanced MW radiometer flown to date: the (CMIS). METOP will continue to operate as one satellite in this three-satellite constellation. Also, NOAA and NASDA have established cooperative agreements to share satellite resources. Finally, there will the NASA sponsored Global Precipitation Measurement (GPM) program (Smith et al. 2007), that will utilize all available polar orbiting satellites (both research and operational), supplemented with a core satellite that will contain a dual-polarization radar along with an advanced microwave radiometer, to achieve 3-h global coverage of precipitation. GPM, although a research mission, will serve as
196
R. R. Ferraro
the prototype for an international operational precipitation-monitoring mission. Since the ultimate goal of precipitation retrieval from space is to meet user requirements, whether on a synoptic or climate scale, there needs to be a continued effort to drive the sensor and algorithm development by such requirements and not be platform specific (e.g., GOES and POES). If successful, GPM will help define these requirements and the future operational precipitation mission will likely require a geostationary microwave sensor as part of the constellation (i.e., see Bizzarri et al. 2007). However, retrieval algorithms should take advantage of the full spectral range of measurements (e.g., visible, infrared and MW) and the IPWG should be a leader in pursuing such algorithm development. Several important improvements with retrieval algorithms are anticipated. There will be a consolidation of algorithms into a general framework that offers portability to different satellite platforms and will instil international collaboration. As previously described, the GPROF approach has served as a prototype for a single retrieval approach for TMI, SSM/I and AMSR. It is envisioned that the GPM mission and the IPWG will further guide the development of a consensus retrieval approach. What are truly needed are parametric retrieval algorithms: those that require specific physical quantities (i.e., atmospheric temperature profile and moisture profiles, cloud parameters, land surface cover, etc.) as input into the retrieval, then, use all available data sources (i.e., satellite-based, ground-based, and ancillary information) to best obtain that information. This approach is attractive since it can accommodate multiple sources of data; yet, use the most advanced physical retrieval schemes. As an example, Fig. 6 presents a schematic of such retrieval for over land. There also needs to be an expansion of the MW retrieval algorithms to incorporate the full range of available MW frequencies in the 6–183 GHz range. These are needed in order to retrieve light stratiform rain and falling snow over mid and high latitudes. For example, NOAA has recently implemented a falling snow detection algorithm over land for use with AMSU (Kongoli et al. 2003, Fig. 7). Lastly, it is anticipated that the land retrievals will be greatly improved through the use of sophisticated emissivity models now being developed and tested. Acknowledgements: I wish to acknowledge the NOAA Office of Global Programs for their support. The statements contained within the manuscript are not the opinions of the funding agency or the US government, but reflect the author’s opinions.
Past, Present and Future of Microwave Operational Rainfall Algorithms
197
Figure 6. Example of parametric rain rate retrieval over land. Inputs come from satellite, numerical models, climate data sets and ground based measurements.
Figure 7. NOAA-16 AMSU precipitation type and rainfall rate for 10 December 2003. The algorithm is based upon Kongoli et al. (2003) and utilizes AMSU measurements at 54, 89, 150 and 183 GHz.
6 REFERENCES Bizzarri, B., A. Gasiewski, and D. Staelin, 2007: Observing rain by millimetre-submillimetre wave sounding from geostationary orbit. In: Measuring Precipitation from Space – EURAINSAT and the Future, V. Levizzani, P. Bauer and F. J. Turk, eds, Springer, 675– 692. Colton, M. and G. Poe, 1994: Shared processing program, Defense Meteorological Satellite Program, Special Sensor Microwave/Imager Algorithm Symposium, 8–10 June 1993. Bull. Amer. Meteor. Soc., 75, 1663–1669.
198
R. R. Ferraro
Ebert, E., 2007: Methods for verifying satellite precipitation estimates. In: Measuring Precipitation from Space – EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 345–356. Ebert, E. E., M. J. Manton, P. A. Arkin, R. E. Allam, and A. Gruber, 1996: Results from the GPCP algorithm Intercomparison programme. Bull. Amer. Meteor. Soc., 77, 2875–2887. Ferraro, R. R., 1997: SSM/I derived global rainfall estimates for climatological applications. J. Geophys. Res., 102, 16, 715–735. Ferraro, R. R., N. C. Grody, and J. A. Kogut, 1986: Classification of geophysical parameters using passive microwave satellite measurements. IEEE Trans. Geosci. Rem. Sensing, 24, 1008–1013. Ferraro, R. R., S. J. Kusselson, and M. Colton, 1999: An introduction to passive microwave remote sensing and its applications to meteorological analysis and forecasting. National Weather Digest, 22, 11–23. Ferraro, R. R., P. Pellegrino, S. Kusselson, M. Turk, and S. Kidder, 2002a: Validation of SSM/I and AMSU derived tropical rainfall potential (TRaP) during the 2001 Atlantic Hurricane Season. NOAA Technical Report NESDIS 105, U.S. Department of Commerce, NOAA, 47 pp. Ferraro, R., F. Weng, N. Grody, I. Guch, C. Dean, C. Kongoli, H. Meng, P. Pellegrino, and L. Zhao, 2002b: NOAA satellite-derived hydrological products prove their worth. EOS, Trans. of Amer. Geophys. Union, 83, 429–437. Grody, N. C., 1991: Classification of snow cover and precipitation using the Special Sensor Microwave Imager. J. Geophys. Res., 96, 7423–7435. Grody, N., J. Zhao, R. Ferraro, F. Weng, and R. Boers, 2001: Determination of precipitable water and cloud liquid water from the NOAA 15 AMSU., J. Geophys. Res., 106, 2943– 2953. Hollinger, J., 1990: SSM/I instrument evaluation. IEEE Trans. Geo. Rem. Sens., 28, 781–790. Huffman, G. J., R. F. Adler, P. Arkin, A. Chang, R. R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78, 5–20. Kongoli, C., P. Pellegrino, and R. Ferraro, 2003: A new snowfall detection algorithm over land using measurements from the AMSU. Geophys. Res. Lett., 30, 1756–1759. Kummerow, C., Y. Hong, W. Olson, S. Yang, R. Adler, J. McCollum, R. Ferraro, G. Petty, and T. Wilheit, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. McCollum, J. R. and R. R. Ferraro, 2003: The next generation of NOAA/NESDIS SSM/I, TMI and AMSR-E microwave land rainfall algorithms. J. Geophys. Res., 108, 8382– 8404. Smith, E. A., G. Asrar, Y. Furuhama, A. Ginati, A. Mugnai, K. Nakamura, R. Adler, M.-D. Chou, M. Desbois, J. Durning, J. Entin, F. Einaudi, R. Ferraro, R. Guzzi, P. Houser, P. Hwang, T. Iguchi, P. Joe, R. Kakar, J. Kaye, M. Kojima, C. Kummerow, K.-S. Kuo, D. Lettenmaier, V. Levizzani, N. Lu, A. Mehta, C. Morales, P. Morel, T. Nakazawa, S. Neeck, K. Okamoto, R. Oki, G. Raju, J. M. Shepherd, J. Simpson, B.-J. Sohn, E. F. Stocker, W.-K. Tao, J. Testud, G. Tripoli, E. Wood, S. Yang, and W. Zhang, 2007: International Global Precipitation Measurement (GPM) program and mission: An overview. In: Measuring Precipitation from Space – EURAINSAT and the Future, V. Levizzani, P. Bauer and F. J. Turk, eds., Springer, 611–654. Weng, F., L. Zhao, G. Poe, R. Ferraro, X. Li, and N. Grody, 2003: AMSU cloud and precipitation algorithms. Radio Sci., 38, 8068–8079. Wilheit, T. T., A. T. C. Chang, M. S. V. Rao, E. B. Rodgers, and J. S. Theon, 1977: A satellite technique for quantitatively mapping rainfall rates over the ocean. J. Appl. Meteor., 16, 551– 560. Wiheit, T., C. Kummerow, and R. Ferraro, 2003: Rainfall algorithms for AMSR-E. IEEE Trans. Geosci. Rem. Sens., 41, 204–214.
16 SPACE-BORNE RADAR ALGORITHMS Toshio Iguchi National Institute of Information and Communications Technology, Japan
1 INTRODUCTION This article describes the principles and issues in rain profiling algorithms for a space-borne radar, in particular in the standard algorithm 2A25 for the Tropical Rainfall Measuring Mission’s Precipitation Radar (TRMM PR). The PR onboard the TRMM satellite is the first space-borne weather radar in the history so that the development of the algorithm was a new challenge although similar algorithms had been proposed and tested with airborne radar data. Some of the notable peculiarities of space-borne precipitation radars in comparison with ground-based radars are the observation geometry (huge surface echo, low horizontal resolutions, high vertical resolution), use of an attenuating wavelength, moving platform (no coherence in echoes from different pulses), and operation environment (well-controlled temperature, no attenuation by wet radome, limited size, and power). These differences must be taken into account in the algorithm development. At the same time, the algorithm has to be able to handle a variety of rain systems with different characteristics in the world. In addition to these peculiarities, space-borne radar algorithms share the same difficulties that ground-based radar algorithms suffer from. Some of the common factors in rain retrieval are the infamous fluctuation of Z-R relationship, attenuation in the case of shortwavelength radar, identification of particles’ phase state and inhomogeneity of rain distribution within the scattering volume, although the significance of each factor may differ substantially depending on the application.
2 PRELIMINARIES The TRMM satellite is a non-sun-synchronous satellite orbiting at about 350 km above the surface (or 407 km after the boost of orbit in August 2001) and 199 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSTAT and the Future, 199–212. © 2007 Springer.
T. Iguchi
200
goes around the earth about 16 times a day. It carries five earth observing sensors. They are the Precipitation Radar (PR), TRMM Microwave Imager (TMI), Visible/Infrared Scanner (VIRS), Cloud and the Earth’s Radiant Energy System (CERES), and Lightning Imaging Sensor (LIS). Of these five, the first three are mainly used for rain studies. The swath width of the TMI and VIRS are about 760 km and 720 km, respectively, whereas that of the PR is only 215 km. While the TMI scans its antenna conically and observes the microwave radiation with an incidence angle of about 50° at the surface of the earth, the PR and VIRS scan almost linearly in a plane perpendicular to the flight direction. The TMI and VIRS are passive sensors, while the PR is an active sensor. The PR transmits pulsed electromagnetic waves at 13.8 GHz and measures their echoes. The most important feature of the PR is that it can measure the vertical structure of rain. The vertical resolution of the PR is 250 m and the horizontal resolution at nadir is about 4.3 km. The PR observes the swath of 215 km with 49 beams every 0.6 s. Each beam width is about 0.7 degree. The major characteristics of the PR are listed in Table 1. Table 1. Characteristics of the TRMM Precipitation Radar.
Radar type Antenna type Beam scanning Frequency Polarization TX/RX pulse width RX band width Pulse repetition frequency Data rate Mass Sensitivity # of input samples Range resolution Horizontal resolution Swath width Observational range
Pulse radar 128-elemen. WG slot array Active phased array 13.796, 13.802 GHz Horizontal 1.57/1.67 µs 0.6 MHz 2776 Hz 93.5 kbps 460 kg <0.5 mm h–1 64 (fading noise ≈ 0.7 dB) 250 m 4.3 km (5 km*) at nadir 215 km (250 km*) Surface to 15 km (minimum)
* After the boost of orbit from 350 km to 407 km in August 2001.
Since version 5 of the 2A25 algorithm is described in some detail in Iguchi et al. (2000) the basic idea and remaining issues are emphasized in this article. We assume that we are given fairly accurate radar signals to begin with. In fact, the absolute calibration of the PR is believed to be accurate within ±1 dB and its overall performance is extremely stable. This good
Space-Borne Radar Algorithms
201
stability comes from the space environment in which the operation temperature is controlled very well. Figure 1 shows the standard deviation of the monthly averaged normalized sea surface radar cross section over three years as a function of the incidence angle. The standard deviation is less than 0.14 dB at any incidence angle and its minimum is about 0.05 dB, implying that the long-term variation of the radar sensitivity is less than 0.05 dB, because there is no reason for assuming a correlation between the variations of the radar sensitivity and the natural variations of sea surface. The angle bin dependence is mostly considered to be due to the natural variation of sea surface conditions. Consequently, we do not need to worry about the long-term variation of the sensitivity.
Figure 1. Standard deviation of monthly averaged normalized sea surface radar cross section as a function of incidence angle.
The 2A25 algorithm retrieves the precipitation profiles in two steps. It estimates the true effective reflectivity factor (Ze) from the measured vertical profiles of reflectivity factor (Zm) first. It then converts the estimated effective reflectivity factor (Ze) into the rainfall rate (R). The step to estimate Ze from Zm corresponds to the attenuation correction. Since almost all attenuation at the Ku-band originates in rain itself, the profile of Zm contains some information about the attenuation. However, the relationship between Ze and the specific attenuation due to precipitation, kP, depends on the type of precipitation particles (e.g., snow or rain), their drop size distribution (DSD) and their temperature. This dependence is crucial in attenuation correction when the attenuation becomes large. The step to convert Ze to R also depends on the same factors.
T. Iguchi
202
3 RELATION BETWEEN ATTENUATION AND OBSERVABLES Let us denote the attenuation factor to range r by A(r), and the two-way pathintegrated attenuation (PIA) expressed in dB to range r by PIA(r). A(r) and PIA(r) are related by
A(r ) = exp(− 0.1ln(10) PIA(r ) )
(1)
The measured apparent radar reflectivity factor Zm(r) is related to the true effective radar reflectivity factor Ze(r) by the following equations:
Z m (r ) = Z mt (r ) + δ Z m (r )
(2)
where δ Z m (r ) is the measurement error in Zm(r), and Zmt(r) is defined by
Z mt ( r ) = Z e (r ) exp(−0.1ln(10) PIA(r ))
(3)
PIA(r) consists of several different contributions. r
(
)
PIA(r ) = 2∫ k P ( s ) + kCLW ( s ) + kWV ( s ) + kO2 ( s ) ds 0
(4)
The factor 2 on the right side of (4) corresponds to the two-way attenuation, and kP, kCLW, kWV, and kO2, are the specific attenuations due to precipitation, cloud liquid water, water vapor, and molecular oxygen, respectively. At the Ku-band, we only need to take these four sources of attenuation into account. Up to version 5 of 2A25 the attenuation due to cloud liquid water, water vapor and molecular oxygen has been totally ignored because their contributions were considered to be negligibly small. In version 6, the effect of kCLW, kWV and kO2 are included to remove small biases. Since none of cloud liquid water, water vapor or molecular oxygen appears in radar echoes, we need to adopt a model to provide their spatial distributions. The details of the model used in version 6 will be given elsewhere. This kind of use of external data other than the information in a radar echo itself is inevitable. Other external data used in 2A25 are the satellite altitude and attitude data, topography data, and climatological surface temperature data. These external data are used to determine the surface-cluttered ranges and to estimate the temperature profile and freezing height. Experience shows that we can approximate the kP-Ze and R-Ze relations by power law relations:
Space-Borne Radar Algorithms
203
k P = αZ eβ
(5)
and
R = αZ eb
The dependence of these relations on the hydrometeor type, DSD and temperature is translated into the dependence of coefficients α, β, a and b on them. We assume that we can ignore kCLW, kWV, and kO2 so that r
PIA(r ) = PIAP (r ) = 2 ∫ k P ( s )ds 0
(6)
If the power law relation of kP-Ze in (4) is exact and if β is independent of range r, we can solve the integral equation for Ze(r) that is obtained by substituting (5) into (6) and then (6) into (3). If we define St(r) by r
St (r ) = ∫ α ( s ) Z mtβ ( s )ds 0
(7)
then the solution becomes
Z e (r ) =
Z mt (r ) [1 − qSt (r )] 1 / β
(8)
where q = 0.2β ln(10). This is equivalent to the Hitschfeld-Bordan solution. PIAP(r) can then be expressed in terms of St(r) as
2 PIAP (r ) = − ln (1 − qSt (r ) ) q
(9)
In reality, however, we never have Zmt(r) or exact values of α and β at each range. We can obtain only Zm(r) which is different from Zmt (r) by δ Zm(r). Instead of α and β that are unknown to us, we need to use model values α0 and β0 that are inferred from some circumstantial information. We define q 0 by q0 = 0.2 β0 ln(10) and Sm(r) by r
S m (r ) = ∫ α 0 ( s ) Z mβ0 ( s )ds 0
(10)
and we use
PIAPm (r ) = −
2 ln (1 − ε q0 S m (r ) ) q0
(11)
204
T. Iguchi
to estimate PIAP(r) from data. ε is an adjustable parameter that accounts for the deviation of α0 from α. If the model represents the true profile of α well, ε must be close to unity. As (11) indicates, however, a small deviation of ε q0 Sm(r) from q St(r) may create a large difference in the estimate of PIAP(r) when q St(r) is close to 1, i.e., when the attenuation is large. The difficulty in attenuation correction originates here. There is fortunately an extra piece of information in the space-borne radar echo that can be used to prevent the instability. The resulting information is the surface echo behind the rain echo. The apparent decrease of the surface echo gives an estimate of the total path-attenuation through rain that can be used as a constraint in attenuation correction. The attenuation correction method that uses the information of surface echo in this way is called the surface reference technique (SRT).
4 FACTORS THAT AFFECT THE SPECIFIC ATTENUATION Before we describe the way how the surface information is used in 2A25, we briefly mention the major factors that varies α and β. The first factor is the phase state of hydrometeors. If it is ice, absorption becomes very small in comparison with liquid water, and the attenuation is caused only by scattering. As a result, kP for ice phase particles is smaller than that for liquid phase particles if the equivalent DSDs are the same. Therefore, it is very important to estimate the actual freezing height which may differ from the 0ºC isotherm height substantially, especially in a convective system. The second factor is the DSD. Since the dependence of the backscattering cross section of a single particle on the particle size differs from that of the extinction cross section, if the variation of DSD has more than one degree of freedom, the kP-Ze relation cannot be expressed by a constant pair of α and β for a given particle type. The DSD information is important not only in rain but also in a solid precipitation region. In a solid precipitation region, densities and shapes of particles are known to affect the kP-Ze relation as well. The third factor is the temperature of precipitation particles because the refractive index that affects the backscattering and extinction cross sections changes with temperature, especially when particles are liquid. The fourth factor that modifies the coefficients is the inhomogeneity of rain distribution within the radar resolution cell. In fact, the radar measures the backscattered echo power averaged over the resolution cell. Since both kP-Ze and R-Ze relations are nonlinear. the same relations are not applicable to the averaged quantities such as k P , R and Z e , even if the relations are β exact and without any bias locally. For example, k P = αZ eβ ≠ α Z e .
Space-Borne Radar Algorithms
205
This kind of problems that originate from the inhomogeneity of hydrometers distribution within the footprint of observation is often called the nonuniform beam filling (NUBF) problems.
5 DIFFICULTY OF NUBF PROBLEMS In the case of attenuation, the NUBF problem is more complicated than that in the Ze-R conversion because attenuation is a cumulative effect. The averaged power of electromagnetic waves reaching at a certain range depends not on the local inhomogeneity of rain distribution but on its distribution in the whole propagation path. The resulting nonuniform pattern of attenuation at a given range must be multiplied by the lateral distribution pattern of nonuniform Ze(r) to obtain Zmt(r). If we denote the average of a given quantity within the radar beam at a constant radar range by angular brackets ⋅ , A(r ) and Z e (r ) are not enough to define Z mt (r ) , because the correlation between the nonuniform patterns of A(r) and Ze(r) must be taken into account. In other words,
Z mt (r ) = Z e (r ) A(r ) ≠ Z e (r ) A(r )
(12)
Note that A(r ) in (12) cannot be calculated with k P (r ) either because r A(r ) = exp⎛⎜ − 0.2 ln(10) ∫ k P ( s )ds ⎞⎟ 0 ⎝ ⎠ r ≠ exp⎛⎜ − 0.2 ln(10) ∫ k P ( s ) ds ⎞⎟ 0 ⎝ ⎠
(13)
In other words, the averaged attenuation factor A to range r is different from the attenuation factor calculated by using the specific attenuation averaged at each range. Therefore, we have three problems to solve. The first one is to relate k P (r ) to Z e (r ) . This problem is a local problem in the sense that the relation only depends on the nonuniformity at a range. The second problem is to relate k P (r ) with A(r ) . In other words, we need to find how to estimate from k P (r ) and non-uniformity information the effective specific attenuation keff(r) that gives A(r ) by the following equation:
206 r A(r ) = exp⎛⎜ − 0.2 ln(10) ∫ keff ( s )ds ⎞⎟ 0 ⎝ ⎠
T. Iguchi
(14)
The third problem is to relate Z mt (r ) to Z e (r ) and A(r ) . The latter two problems are not local, but the correlation of inhomogeneity along the range direction must be considered. An attempt to solve the second problem is treated in Iguchi (2001), but the problem still remains open. Since NUBF is still an open issue, no correction for this effect is included in version 6 of 2A25. We also ignore it in the rest of this article.
6 ATTENUATION CORRECTION AND DSD ESTIMATION 2A25 uses (8) to calculate the attenuation-corrected Ze(r). Therefore, the problem is to find an appropriate combination of α and β that gives the correct effective specific attenuation keff. As mentioned above, both α and β may depend on range, DSD, inhomogeneity, etc. Since we ignore the attenuation due to cloud water, water vapor, and oxygen and the effect of inhomogeneity in this article, keff and kP are the same. To use (8), we assume that β0 is independent of range. All variations in the kP-Ze relation are included in the variation of α as
k P (r ) = ε ( DSD)α 0 (r ) Z eβ 0 (r )
(15)
The variation of α0(r) with r includes the changes due to phase-state difference, temperature difference, and storm type difference. The variation of α due to the changes in DSD is separated into the adjustable factor ε(DSD). Since we can adjust only one parameter in each beam, ε(DSD) is independent of range. According to (9), the PIA at surface range rs due to precipitation can be expressed as
2 PIAP (rs ) = − ln(1 − ς t ) q
(16)
where ς t is defined by
ς t = q St (rs )
(17)
Space-Borne Radar Algorithms
207
In practice, we cannot calculate ς t from the radar echo, firstly because the rain echo near the surface is cluttered by the surface echo, secondly because we do not know the exact values of α and β, and thirdly because we can use only Zm(r) instead of Zmt(r). Let us denote by rb the upper limit of range at which the radar echo is still free from the surface clutter. We define ς t′ by
ς t′ = q St (rb )
(18)
Once we assume certain conditions for the rain profile between rb and rs such as the slope of Ze there and the continuity at rb, we can establish the relationship between ς t and ς t′ so that we can calculate ς t from ς t′ . As an estimator of ς t′ , we use ες m′ where ς m′ is defined by
ς m′ = q0 S m (rb )
(19)
In other words, Zmt is replaced with the measured Zm, and α and β are replaced with α0 and β0, respectively. Here, α and β are the true values of the coefficients in the kP-Ze relation, and α0 and β0 are their model values. ε is an adjustable parameter that represents the uncertainty in q0 Sm(rb). We estimate the value of ε so that ες m′ becomes an unbiased estimate of ς t′ . ε must therefore satisfy the following equation:
ς t′ = ες m′
or
rb
rb
0
0
q ∫ αZ mβ ( s )ds = εq0 ∫ α 0 ( s ) Z mβ0 ( s )ds
(20)
Although ε may include uncertainties in all three kinds of quantities α0, β0, and Zm(r) in q0 Sm(rb), the major uncertainty is considered to come from the difference between α0 and α , because the variation of β is generally small and because the contribution of random error δ Z m (r ) in the integral Sm(rb) is relatively small due to the fact that δ Z m (r ) at different range bins r = ri and r = rj where i ≠ j are statistically independent and do not add up as much as Z mβ0 (r ) themselves. Therefore, the deviation of ε from unity indicates the deviation of α from α0. If the vertical profile of the phase state and temperature that is assumed to determine the profile of α0(r) is close to the true profile, the deviation implied by ε can be attributed to the deviation of the true DSD from the model DSD, and ε can be used as a DSD index. This interpretation is assumed when we write (15).
T. Iguchi
208
From DSD data collected with ground-based instruments at several different locations in the world, we can choose a pair of α0 and β0 together with the uncertainty in ε. In this sense, we can rewrite (20) in the form of an observation equation as in (21) below and interpret ln ε as the error term in the observation variable ln ς m′ for ln ς t′ .
ln ς m′ = ln ς t′ − ln ε ( DSD)
(21)
As described earlier, ς t′ is related to the attenuation to the surface PIAs ≡ PIA(rs) through (16) and the relation between ς t′ and ς t . On the other hand, the apparent decrease of the surface echo ∆σ m0 also gives and independent estimate of PIAs.
∆σ m0 = PIAs + δ ∆σ o
m
(22)
Here ∆σ m0 is defined as the difference between the reference value of the 0 normalized radar cross section of the surface σ ref and the measured 0 apparent normalized radar cross section σ m . In version 6 the bias in σ ref caused by the attenuation due to cloud liquid water and water vapour is taken into account. For the sake of simplicity of discussion, we ignore such attenuation here. Thus, we have two independent observables ς m′ and ∆σ m0 that are related to PIAs. If δ ∆σ o in (22) is zero, ∆σ m0 will determine the value of ε uniquely. m This corresponds to the α-adjustment method described in Iguchi and Meneghini (1994). In actual data, however, we cannot ignore the error term in (22) so that the problem becomes probabilistic. The problem is to find the a posteriori probability distribution of ε for given ς m′ and ∆σ m0 . Since PIAs can be calculated for given ς m′ and ε by using (21) and the assumed Ze profile between rb and rs, this probability p (ε | ς m′ , ∆σ m0 ) can be rewritten as follows by using the Bayes theorem
p (ε | ς m′ , ∆σ m0 ) ∝ p (∆σ m0 | PIAs (ς m′ , ε )) p (ε )
(23)
Space-Borne Radar Algorithms
209
2A25 assumes that the probability density function (pdf) of ∆σ m0 for given PIAs, i.e., p (∆σ m0 | PIAs ) , follows the normal distribution with mean equal to PIAs, or equivalently p (δ ∆σ 0 ) follows the normal distribution with mean m 0. The variance of the pdf is calculated as the sum of the variance in the 0 that is given by 2A21 and the variance in each reference value σ ref measurement of σ m0 due to fading noise, which corresponds to 0.7 dB. The pdf of p(ε) is assumed to be lognormal with mean 1.0. The standard deviation of this pdf was about 0.25 in version 5, but has been increased to 0.4 for stratiform rain and 0.3 for convective rain in version 6 based on the analysis of disdrometer data at several locations over the world. Note that when ∆σ m0 is large, the likelihood of PIAs has a peak at a large value of PIAs. When PIAs is large, as (16) shows, 1 − ς t becomes nearly zero irrespectively of the uncertainty of PIAs as long as it remains relatively small. Therefore, the distribution of ς t for a large ∆σ m0 becomes very narrow and hence makes the distribution of ε very narrow as well. On the other hand, when ∆σ m0 is small, p (ε ) rather than p (∆σ m0 | PIAs , ∆σ m0 ) in (23) effectively determines the final width of p (ε | ς m′ , ∆σ m0 ) . In this case, the surface reference does not work as a constraint on ε. This is the way how the surface reference technique should be used. Once p (ε | ς m′ , ∆σ m0 ) is obtained, Ze(r) can be calculated as
Z e (r ) = ∫
Z m (r ) p(ε |ς m′ , ∆σ m0 )dε 1/ β [1 − ε q0 S m (r )]
(24)
In version 6, Zm(r) in this equation is replaced by Z m′ (r ) that is corrected for attenuation by cloud liquid water, water vapor, and oxygen.
(
)
r Z m′ ( r ) = Z m (r ) exp⎛⎜ 0.2 ln(10) ∫ kCLW ( s ) + kWV ( s ) + kO2 ( s ) ds ⎞⎟ 0 ⎝ ⎠
(25)
The rainfall rate is also calculated in a similar way.
R(r ) = ∫ a(ε ) Z eb (ε ) (r ) p (ε | ς m′ , ∆σ m0 ) dε
(26)
T. Iguchi
210
Note that in version 5, instead of taking the expectations with respect to ε, the most probable value of ε, was used to calculate Ze(r) and R(r). If we denote by εm the most probable value of ε that maximizes p (ε | ς m′ , ∆σ m0 ) , then Ze(r) and R(r) were calculated by using this εm as
Z e (r ) =
Z m (r ) [1 − ε m q0 S m (r )]1 / β
(27)
and
R(r ) = a (ε m ) Z eb (ε m ) (r )
(28)
7 FLOW OF 2A25 ALGORITHM Figure 2 shows a simplified flow chart of 2A25. From the radar echo, we calculate the measured radar reflectivity factors Zm(ri) at ri, (i =1,…,n) and the normalized surface cross section σ m0 . The latter is compared with the reference value σ ref to obtain an estimate ∆σ m0 of the path-integrated attenuation to the surface PIA(rs). From the absolute values and profile characteristics of Zm(ri), we classify the rain into three types, i.e., stratiform, convective, and others. According to the rain type and climatological surface temperature, the initial profiles of a0, b0, α0 and β0 are defined. With these α0 and β0 we calculate ς m′ and the pdf of p (ε | ς m′ , ∆σ m0 ) . By using this pdf itself (in version 6) or the most probable estimate of εm (in version 5), we calculate the attenuation corrected Ze(ri). This process corresponds to finding the best DSD parameter ε that matches the attenuation estimated from the Ze(ri) profile with the attenuation estimated from the surface echo ( σ m0 ). The matching is not carried out in a deterministic way, but in a probabilistic way. Ze(ri) are then converted into R(ri) by the formula R = aZ eb where a and b are modified from εa0 and b0 according to the pdf of ε. This adjustment makes the R-Ze relation with coefficients a and b consistent with the final kP-Ze relation with α0 and β0 provided that the deviation of ε from unity is caused by the deviation of DSD from the model.
Space-Borne Radar Algorithms
211
Figure 2. Simplified flow chart of rain profiling algorithm 2A25.
8 SUMMARY AND DISCUSSION The basic idea of rain profiling algorithm for the TRMM Precipitation Radar is described. The process of estimating rainfall rates from radar echoes can be divided into two steps. The first step that corresponds to the attenuation correction is the conversion of the measured radar reflectivity factor Zm(r) into the effective radar reflectivity factor Ze(r), and the second step is the conversion of Ze(r) into the rain profile R(r). Both steps need certain information about the electromagnetic properties of precipitation particles as well as their DSD. The inhomogeneous distribution of particles also affects both conversion steps. This article presents how the surface echo can be used to extract the DSD information that gives an important constraint in both attenuation correction and Ze-R conversion. In this step, we have to assume the type of precipitation particles at each range and their initial DSD. If we incorrectly assume the phase of precipitation particles, the estimated DSD parameters may be biased. A dual-frequency precipitation radar is planned on the Global Precipitation Measurement’s core satellite. By using the radar reflectivities measured simultaneously from the same scattering volume at two wavelengths, we should be able to extract better information about the attenuation by precipitation that can be used to identify the phase state of particles. With two frequencies, the SRT can be used at a much wider range of rainfall rate.
212
T. Iguchi
Under certain conditions, we can obtain a constraint on the DSD without the surface echo. The difference of radar reflectivitities between the two frequencies will also provide an important piece of information about the DSD too. These factors all contribute to improve the accuracy of rain estimation. Nevertheless, the dual-frequency algorithm is not error free. The attenuation caused by water vapor and cloud liquid water will not be negligible at the higher frequency of 35.5 GHz. The nonuniform beam filling problem remains. In fact, since the dual frequency algorithm relies more on the attenuation information in general, the significance of NUBF problem is lager. In these sense, we need to think carefully what kind of information is available to narrow down the uncertainties of these factors that cannot be measured directly by radar. Acknowledgment: This work is partly supported by Japan Aerospace Exploration Agency (JAXA).
9 REFERENCES Iguchi, T. and R. Meneghini, 1994: Intercomparison of single frequency methods for retrieving a vertical rain profile from airborne or space-borne radar data. J. Atmos. Oceanic Technol., 11, 1507–1516. Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain profiling algorithm for the TRMM Precipitation Radar. J. Appl. Meteor., 39, 2038–2052. Iguchi, T., 2001: Rain profiling algorithm for TRMM and GPM Precipitation Radar. Proceedings Of the 5th International Symposium on Hydrological Applications of Weather Radar–Radar Hydrology, November 19–22, Kyoto, 545–550.
17 RAIN TYPE CLASSIFICATION ALGORITHM Jun Awaka1, Toshio Iguchi2, and Ken’ichi Okamoto3 1
Hokkaido Tokai University, Sapporo 005-8601, Japan Communications Research Laboratory, Koganei, Tokyo 184-8795, Japan 3 Osaka Prefecture University, Sakai, Osaka 599-8531, Japan 2
1 INTRODUCTION Retrieval of precipitation rate from a radar requires a knowledge of rain type, on which drop size distribution (DSD) depends (Battan 1973; Meneghini and Kozu 1990). In the Tropical Rainfall Measuring Mission (TRMM) (Kummerow et al. 2000), a standard Precipitation Radar (PR) algorithm, called 2A23, classifies rain into three main categories: stratiform, convective, and other. Since the TRMM PR observes three-dimensional rain fields, rain type can be obtained by examining both vertical and horizontal patterns of precipitation echo. Amitai (1999) has proposed a rain type classification scheme which examines storm height and horizontal gradient of radar reflectivity factor, Z, at several heights. The TRMM PR algorithm 2A23 uses two independent methods for rain type classification: one is a vertical profile method (Awaka et al. 1997), and the other a horizontal pattern method, which is based on the University of Washington convective/stratiform separation method (Steiner et al. 1995), which has been developed for ground based radar data. Both methods classify rain into stratiform, convective, and other. Results of the two methods are combined, and 2A23 outputs a unified rain type. This paper describes in some detail the most recent version (V6) of 2A23, data of which will be released to the public in 2004.
213 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINAST and the Future, 213–224. © 2007 Springer.
J. Awaka et al.
214
2 TRMM PR ALGORITHM 2A23 2.1 Outline of 2A23 Table 1 shows some parameters of the TRMM PR, which are relevant to 2A23. For simplicity, the table shows the case of the satellite altitude being 350 km. (In August 2001, the altitude of the TRMM satellite was changed from 350 km to 400 km for extending the mission life by reducing the atmospheric drag.) The TRMM PR operates at the frequency of 13.8 GHz, and has a range resolution of 250 m (Kozu et al. 2001). When the satellite altitude is 350 km, the size of the antenna footprint is about 4.3 km at nadir. One scan of data consists of 49 angle bins (i.e., antenna beam directions) of data. Table 2 summarizes the main objectives of 2A23 (version 6). Detection of bright band (BB), which is also called as melting layer in the literature (e.g., Fabry and Zawadzki 1995; Klaassen 1988), plays a key role in the V-method. Determination of the width of BB is newly introduced in version 6 of 2A23. Detection of shallow non-isolated is also introduced in addition to the detection of shallow isolated which already exists in earlier versions. Table 1. Some parameters of TRMM PR. Satellite altitude Frequency Range resolution Horizontal resolution Observable range Scan angles Swath width
350 km (before boost) * 13.8 GHz 250 m (normal sample) 4.3 km (at nadir) 15 km above mean sea level –17º to +17º (49 angle bins) 215 km
(*Altitude was changed to 400 km in August 2001) Table 2. Main objectives of 2A23 (V6). Detection of bright band (BB). – when BB is detected, the height, the strength, and the width (i.e., thickness) of BB are determined. Classification of rain type into the following three categories: – stratiform, convective, and other. Detection of shallow isolated and shallow non-isolated. Output of Rain/No-rain flag. Computation of the estimated height of freezing level. Output of the height of storm top.
Figure 1 shows a flow chart of 2A23. Inputs such as measured reflectivity factor, Zm, rain/no-rain information (called minimum echo flag), and range of storm top are fed in from a level-1 TRMM PR algorithm called 1C21. In the V-method, detection of BB is carried out first. When BB is detected, the
Rain Type Classification Algorithm
215
width of BB is determined. Then rain type is classified into stratiform, convective and other. Detection of shallow isolated and shallow non-isolated is also made in the V-method. Independently in the H-method, rain type is classified into stratiform, convective and other. Based on the combination of rain types by the two methods, a unified rain type is determined for each angle bin of data.
Figure 1. Flow chart of 2A23 (V6).
2.2 V-method 2.2.1.
Detection of BB
Detection of BB is carried out in several stages: a main idea is to detect clear BB peaks in one scan of data consisting of 49 angle bins of data, and when clear BB peaks are detected, then the algorithm goes on to detect less clear BB peaks. Detection of a clear BB peaks is carried out (i) by a peak detection using a spatial filter (Awaka et al. 1998) and (ii) by examining the slope of Z m in the upper part of BB peak, the latter approach is added in version 6 of 2A23. The spatial filter has the following simple form: ⎛ − 1 − 1 − 1⎞ ⎜ ⎟ ⎜ 2 2 2⎟ ⎜ − 1 − 1 − 1⎟ ⎝ ⎠
(1)
where the row indicates the direction of antenna scan angle and the column the range direction. The above spatial filter examines the second derivative of Zm with respective to the range from the satellite. Since adjacent three antenna scans of data are applied to the spatial filter, a large output of the
J. Awaka et al.
216
filter is expected to occur at the location of BB peak, which usually extends somewhat uniformly in the horizontal direction. The spatial filter selects BB candidates, each of which satisfies the following conditions, (a) (b) (c) (d)
output of the spatial filter exceeds a given threshold the value of Zm decreases appreciably above the detected peak the height of the peak must appear almost at the same height the peak must be within a BB window region
where the BB window is defined by referring to an estimated height of the freezing level, Hfreeze [km], which is computed by: Hfreeze = (Ts - 273.13)/6.0
(2)
and where Ts [K] is a climatological temperature at mean sea level; the lapse rate of the atmospheric temperature is assumed to be 6.0 [K km–1]. In version 6 of 2A23, the BB window is defined in such a way that the height of BB (HBB) should satisfy the following inequalities: Hfreeze - 2.5 < HBB < Hfreeze + 2.5 [km]
(3)
HBB > 0, HBB < 6.5 [km]
(4)
The spatial filter can miss BB, for example, when the BB peak smears near antenna scan edges. To rescue this problem, a steepness of the slope of Zm in the upper part of BB peak is examined. Roughly speaking, the slope of Zm is defined to be large when: 10 log10 Zm[i][j+2] - 10 log10 Zm[i][j] > 5.0
(5)
and 10 log10 Zm[i][k+1] - 10 log10 Zm[i][k] > 2.0 (with k=j, j+1)
(6)
where Zm[i][j] stands for Zm at the i-th angle bin and the j-th range bin, with the index j to increase in the downward direction. A peak associated with a large slope of Zm is then obtained. When thus detected peak having a large slope of Zm satisfies several conditions, such as conditions (c), (d) mentioned earlier for the peaks detected by the spatial filter, for example, the peak becomes a candidate for the clear BB peak. The clear BB peaks are eventually obtained by applying a median filter to the height of the BB candidates, that is, peaks which do not belong to BB are filtered out.
Rain Type Classification Algorithm
217
When clear BB peaks are detected in one scan of data, less clear BB peaks are then detected by imposing weaker conditions on the range profile of Zm. For example, when a small peak exists adjacent to a clear BB peak and the height of the small peak is close to the averaged height of the clear BB peaks, the small peak is also a BB peak. When BB is detected, the strength of BB and the width of BB are then determined. The strength of BB is simply defined as the maximum value of Zm in the BB peak. The width of BB in the nadir direction is defined as the difference between the upper boundary of BB and the lower boundary of the BB, where the definition of the lower boundary of BB is very close to that by Fabry and Zawadzki (1995), and the definition of the upper boundary is the one which is somewhere in between the definition by Fabry and Zawadzki (1995) and that by Klaassen (1988). In directions other than nadir, the effect of a smeared BB peak is empirically corrected for. Since this topic is not relevant to the rain type classification, we will not discuss it any further.
2.2.2.
Rain types by the V-method
The V-method classifies rain into three categories: stratiform, convective, and other. The meaning of each category is as follows: 1. When BB is detected, rain type is stratiform. 2. When BB is not detected, but if the radar echo is strong, rain type is convective. 3. Other type is defined as not-stratiform and not-convective. In the above, the rain type is determined for each angle bin, which means that we assume that the rain type does not change along a radar beam. A quantitative statement for strength of the radar echo in the above item (2) is as follows: Zmmax[i] > 39 [dBZ],
(7)
where Zmmax[i] stands for the maximum value of Zm[i][j] along the range (whose bin number is denoted by j): The maximum value of Zm for a given angle bin number i is searched by varying j in a clutter free region ranging from the storm top down to the clutter free bottom, the latter of which is defined as the range closest to the surface where the effect of the surface echo can be neglected. It should be noted that the other type of rain by the V-method is defined as not convective and not-stratiform. This means that the other type by the V-method consists of the following cases: 1. Cloud. 2. Actually stratiform, but BB detection fails.
J. Awaka et al.
218
3. Ambiguous because radar echo is not strong enough to be convective and BB does not exist. 4. Simply noise.
2.2.3.
Shallow isolated and shallow non-isolated
The V-method also detects shallow rain, which consists of shallow isolated and shallow non-isolated. (In previous versions of 2A23, only shallow isolated is detected.)
Figure 2. Shallow condition.
Figure 2 illustrates a shallow condition adopted in 2A23. When the height of storm top, Hstorm, is much lower than the estimated height of freezing level, Hfreeze, it is defined as shallow rain. There are two levels of confidence for the shallow rain; “possible” and “certain”. Over ocean, Hfreeze-Hstorm>1.0 [km] means shallow rain “possible”, and Hfreeze-Hstorm>1.5 [km] means shallow rain “certain”. Over land, however, the judgment is always shallow rain “possible” no matter how low is the height Hstorm because Hfreeze may not be trustworthy over land. When the region of shallow rain is isolated from the other non-shallow rain certain areas, this shallow rain is called as shallow isolated. Shallow nonisolated is defined as the shallow rain which is not shallow isolated.
2.3 H-method The H-method also classifies rain into three categories: stratiform, convective, and other, but with the definitions of these being different from those by the V-method. The H-method is based on the University of Washington convective/stratiform separation method (Steiner et al. 1995), which examines the horizontal pattern of Zm at a given height; this original horizontal pattern method is applicable to the data with a 2 km horizontal resolution. Since the horizontal resolution of the TRMM PR is about 4.3 km, the original horizontal pattern method is not readily applicable to the TRMM PR data. Besides, the TRMM PR observes precipitation not only over low flat
Rain Type Classification Algorithm
219
areas but also over high mountain areas; in the latter case, examining the horizontal pattern of Zm at a given height would become impossible sometimes. To enable the horizontal pattern method applicable to the TRMM PR data, the following modifications are made: 1. Instead of examining a horizontal pattern of Zm at a given height, a horizontal pattern of Zmmax(in R) is examined; here Zmmax(in R) is the maximum of Zm in the rain region, R, which is defined as the clutter free region below Hfreeze (minus 1 km margin). The quantity Zmmax(in R) is obtained for each angle bin. 2. Parameters are changed so that they are suitable to the TRMM data with a 4.3 km horizontal resolution. Choice of parameters was made before the launch of the TRMM satellite using a ground radar data in such a way that the result with a simulated 4.3 km horizontal resolution data produces almost the same result as that with a 2 km horizontal resolution data. 3. Third category of other type is also introduced to handle the noise case. In the H-method, detection of convective rain is made first. If Zmmax(in R) exceeds 40 dBZ, or Zmmax(in R) stands out against the background area, this pixel is regarded as a convective center. (Here, we use the term “pixel” for identifying the location of Zmmax(in R) in a horizontal plane.) Rain type for a convective center is convective, and rain type for the (four) pixels nearest to the convective center is also convective. If rain type is not convective and if the rain echo is certain to exist, rain type is stratiform. Rain type by the H-method is other if the radar echoes below Hfreeze (minus 1 km margin) at a given angle bin are very weak so that the echo there may possibly be noise. This means that the other type by the H-method consists of (i) cloud only case and (ii) noise only case.
2.4 Unification of rain types Since the algorithm 2A23 uses two independent methods for classifying rain types, it would not be friendly to the users if 2A23 outputs the rain types by the two methods separately. To make the result user-friendly, 2A23 outputs the unified rain type. In version 6 of 2A23, the unified rain type is expressed by a 3-digit number (while in the previous versions, the unified rain type is expressed by a 2-digit number): the first digit indicates the main category of the type (1: stratiform, 2: convective, 3: other), and the remaining second and third digits distinguish the sub-category. Let the rain type for the ith angle bin be rainType[i]. Then the main category of the unified rain type is obtained by rainType[i]/100. The main category of unified rain type would satisfy the need of most users.
220
J. Awaka et al.
It should be mentioned that the unification of rain types is made in such a way that the rain type by the V-method and that by the H-method can be reconstructed from the unified rain type by using a suitable look-up table. One of the most surprising findings obtained immediately after the launch of the TRMM satellite is on ubiquitous shallow isolated rain. Shallow isolated rain is observed in a dot-like manner, which implies that shallow isolated rain may not be stratiform which is characterized by a rather continuous wide rain area. But the strength of the radar echo from shallow isolated rain is usually weak. Hence, regarding the shallow isolated rain as convective may contradict with our usual notion that the convective rain should be strong. In version 5, shallow isolated rain is classified as stratiform, convective, or other depending on the strength of radar echo, and the most of shallow isolated rain is classified as stratiform because of weak radar echo.
Figure 3. Angle bin (i.e., antenna scan angle) dependence of the counts of three main rain types (left panel), and that of detected BB (right panel).
In version 6 of 2A23, all the shallow isolated rain is classified as convective because shallow isolated rain has convective characteristics (Schumacher and Houze 2003), though the radar echo of shallow isolated rain is usually weak. Note that shallow rain consists of shallow isolated and shallow non-isolated rain (see Section 2.2.2). Most of the shallow non-isolated rain is classified as stratiform because its radar echo is usually weak.
3 SOME STATISTICAL RESULTS Figure 3 shows angle bin dependence of the count of the following: stratiform, convective, and other types on the left panel, and detected BB on the right panel. The angle bin number 25 in the abscissa corresponds to the nadir direction, and the angle bin numbers 1 and 49 to the scan edges, where the
Rain Type Classification Algorithm
221
antenna scan angle, i.e., the off nadir angle, is about ±17°. The figure is based on one month statistics of data in February 1998; the total data for both over land and ocean are used. The left panel of Fig. 3 indicates that among the three main categories of rain type, the most populous is stratiform, and the least is other. The count of “other” type in version 6 of 2A23 becomes very small when compared with that in version 5 (not shown), because of the changes in the parameters for characterizing “other”. In version 6, “other” means ice cloud only or possibly just noise. Figure 3 indicates that the count of stratiform and that of convective depend on the angle bin, but the count of other is almost independent of the angle bin. The count of detected BB shows a large dependence on the angle bin. In what follows, the reasons for these angle bin dependencies are examined by turns. First on the stratiform: in Fig. 3, the count of stratiform is rather constant near nadir, but the count decreases near scan edges. This occurs because the stratiform precipitation includes a large number of shallow non-isolated; when the storm top is low, the rain echo may be masked by a smeared surface clutter near scan edges. (The surface echo is a clutter to the observation of precipitation.) Since the TRMM PR has a rather small range resolution, ∆R, which is 250 m, and a large size of footprint of about 4.3 km (see Table 1), a sharp surface echo at nadir smears to a large extent near a scan edge as illustrated in Fig. 4.
Figure 4. Shapes of surface echo, at nadir (left) and near scan edge (right).
In Fig. 3, the count of the detected BB shows a sharp decrease near scan edges. This occurs because the shape of BB smears near scan edges due to a mechanism similar to the one illustrated in Fig. 4, and because it becomes very difficult to detect the smeared BB. Figure 5 shows the angle bin dependence of the counts of stratiform under the following conditions: Hstorm > 2, 3, 4, and 5 [km] (thin lines). The figure also shows the angle bin dependence of the unconditional count of stratiform (thick line), which is identical to that of stratiform in Fig. 3. The thin curves in Fig. 5 have almost the same shape with parallel shifts in the vertical
222
J. Awaka et al.
Figure 5. Unconditional stratiform count (thick), and the stratiform counts under the condition that the height of the storm top is greater than 2 km, 3 km, 4 km, and 5 km (thin).
direction. The shape of the thin curve for Hstorm > 2 [km] near scan edges is almost the same as that of the thick curve near scan edges, which indicates that the smearing of the surface echo extends up to about 2 km at scan edges. A closer look at the thin curves in the figure indicates that the count of stratiform slightly decreases near scan edges, which implies that a sensitivity of the TRMM PR may be slightly lower near scan edges. Let us move on to a discussion on the count of convective, which exhibits a dependency on the angle bin as shown in Fig. 3. If the majority of convective precipitation has a tall storm top and a strong precipitation rate, we would expect that the count of convective may be almost independent of the angle bin, because the high storm top is free from the effect of surface clutter, and the strong precipitation rate produce a large radar echo which is not affected by a sensitivity of the TRMM PR. In version 6 of 2A23, however, all the shallow isolated is classified as convective (see Section 2.2.3). Figure 6 separately shows the angle bin dependency of the count of shallow isolated (thick line) and that dependency of the count of convective which excludes the shallow isolated (thin lines). The count of shallow isolated shows angle bin dependence because of the masking effect by the surface clutter. The count of convective which excludes shallow isolated is almost independent of the angle bin, which is a characteristics of tall and strong precipitation. Finally, the count of other type in Fig. 3 is almost constant over the angle bin. This is what we expect because the other type in version 6 of 2A23 consists of ice cloud and noise as described in the early part of this section. The ice cloud may be observed uniformly over the entire angle bins because
Rain Type Classification Algorithm
223
Figure 6. Count of shallow isolated (SI) shows a large dependence on angle bin, i.e., on antenna scan angle (thick line). When the SI count is subtracted from, the remaining convective count becomes almost constant over the angle bin (thin line).
of its high storm top, which is too high to be masked by the surface clutter, and the noise may also appear uniformly over the entire angle bins because of its randomness; thus making the count of other type almost constant against the angle bin. It should be noted in Fig. 3 that at the very edges of angle bins, i.e., at the angle bin number being 1 and 49, the count of stratiform increases a little and the count of convective decreases a little. This phenomenon occurs because the applicability of the H-method is not guaranteed at the very edges of angle bins. (The stand-out condition for the convective precipitation by the H-method requires an averaged Zm in the background area, which cannot be well defined at the very edges of angle bins because Zm is available in about one half of the background area only; the other half of the background area is in the outside of the PR swath and the data there is missing).
4 CONCLUDING REMARKS Though the rain type classification algorithm 2A23 works fine, there are several things that we should be aware of. Among other things, it should be noted that all the shallow isolated is classified as convective in version 6 of 2A23; the strength of most shallow isolated is weak, which is not the characteristics of the ordinary convective precipitation. Detection of BB, which plays a key role in the rain type classification by the V-method, should be improved. A substantial improvement in the detection of BB is anticipated in the future Global Precipitation Measurement (GPM) project, because for GPM a dual frequency radar is planned to be used (Kobayashi and Iguchi 2003); at least the discrimination of a true BB peak from a false BB peak, arising due to a large attenuation effect in the strong convective precipitation, would become relatively easy.
224
J. Awaka et al.
Acknowledgments: Development of 2A23 has been sponsored by Japan Aerospace Exploration Agency, JAXA (former National Space Development Agency, NASDA).
5 REFERENCES Amitai, E., 1999: Relationships between Radar properties at high elevations and surface rain rate: Potential use for spaceborne rainfall Measurements. J. Appl. Meteor., 38, 321–333. Awaka, J., T. Iguchi, and K. Okamoto, 1998: Early results on rain type classification by the Tropical Rainfall Measuring Mission (TRMM) precipitation radar. Proc. 8th URSI Commission Final Open Symp., Aveiro, Portugal, 143–146. Battan, L. J., 1973: Radar Observation of the Atmosphere, University of Chicago Press, Chicago, IL. Fabry, F. and I. Zawadzki, 1995: Long-term radar observations of the melting layer of precipitation and their interpretation. J. Atmos. Sci., 52, 838–851. Klaassen, W., 1988: Radar observations and simulation of the Melting layer of precipitation. J. Atmos. Sci., 45, 3741–3753. Kobayashi, S. and T. Iguchi, 2003: Variable pulse repetition frequency for the Global Precipitation Measurement Project (GPM). IEEE Trans. Geosci. Remote Sens., 41, 1714–1718. Kozu, T., T. Kawanishi, H. Kuroiwa, M. Kojima, K. Oikawa, H. Kumagai, K. Okamoto, M. Okumura, H. Nakatsuka, and K. Nishikawa, 2001: Development of precipitation radar onboard the Tropical Rainfall Measuring Mission (TRMM) satellite. IEEE Trans. Geosci. Remote Sens., 39, 102–116. Kummerow, C., J. Simpson, O. Thiele, W. Barnes, A. T. C. Chang, E. Stocker, R. F. Adler, A. Hou, R. Kakar, F. Wentz, P. Ashcroft, T. Kozu, Y. Hong, K. Okamoto, T. Iguchi, H. Kuroiwa, E. Im, Z. Haddad, G. Huffman, B. Ferrier, W. S. Olson, E. Zipser, E. A. Smith, T. T. Wilheit, G. North, T. Krishnamurti, and K. Nakamura, 2000: The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. J. Appl. Meteor., 39, 1965–1982. Meneghini, R. and T. Kozu, 1990: Spaceborne Weather Radar, Artech House, Boston·London. Schumacher, C. and R. A. Houze, Jr., 2003: The TRMM Precipitation Radar’s view of shallow, isolated rain. J. Appl. Meteor., 42, 1519–1524. Steiner, M., R. A. Houze, Jr., and S. Yuter, 1995: Climatological characterization of threedimensional storm structure from operational radar and rain gauge data. J. Appl. Meteor., 34, 1978–2007.
18 DUAL-WAVELENGTH RADAR ALGORITHM Kenji Nakamura1 and Toshio Iguchi2 1
Hydrospheric Atmospheric Research Center, Nagoya University, Japan National Institute of Information and Communications Technology, Japan
2
1 INTRODUCTION The dual-wavelength precipitation radar (DPR) is one of the key instruments for the GPM core satellite. Precipitation observation using the DPR along with nearly simultaneous observation by a microwave radiometer is essential for accomplishing the GPM mission. The 3-hourly global rain mapping will be achieved by the constellation satellites with microwave radiometers, and the microwave radiometer rain estimate will be improved or tuned by the information provided by the DPR. TRMM achievements have clearly shown that the comparison of the results from radar with microwave radiometer data is very effective to improve the rain retrievals. To achieve the GPM’s global precipitation mapping, the radar onboard the core satellite should have: (a) high sensitivity to detect weak rain and snow, (b) capability to discriminate solid precipitation from liquid one, and (c) better accuracy of rain retrieval than the TRMM PR. To meet the above requirements, the DPR has been designed, and the rain retrieval algorithms are under development. Here, the basic types of DPR algorithms and the generic rain profile retrieval will be described. It should be emphasized that the DPR algorithms are far from matured, and still open to many new ideas. In addition, groundbased or airborne radar experiments are required to validate each algorithm.
2 BASIC SPECIFICATION OF THE DPR For the study of the DPR algorithms, the performance of the DPR should first be specified. The specifications include the sensitivity, the swath, the range resolution, etc. At least the same performance of the TRMM
225 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINAST and the Future, 225–234. © 2007 Springer.
226
K. Nakamura and T. Iguchi
precipitation radar (PR) was required for the lower frequency radar (14-GHz Ku-band radar), which makes the DPR observation a natural extension of the TRMM PR observation. For the higher frequency radar, a 35-GHz channel was selected. Radio waves in this band suffer from much more rain attenuation than the Ku-band radio waves and the rain echoes will show a very different measured profiles from those in the Ku band. The direct combination of the measured profiles at the two bands makes possible the accurate precipitation profile retrievals. One of the big decisions was the nominal range resolution of the Ka-band radar (the higher frequency part of the DPR) to be 500 m, though the TRMM PR and the 14-GHz radar of the DPR has 250-m range resolution. The main reason for it is to attain the sensitivity of nearly 12 dBZ. For the global precipitation observation, significant sensitivity improvement is required for snow observation and for detection of very light rain or highly rainattenuated rain signatures. Sacrificing the range resolution by a factor of two will improve the minimum detectable signal by 6 dB. Though very shallow rain/snow may be missed by the degraded range resolution, such rain/snow is hard to be detected even with 250-m range resolution due to range smearing and ground clutter. The swath of the Ka-band radar was another issue and the current solution is to scan the radar beams across the swath of about a half of the TRMM PR swath, that is, about 100 km. The minimum requirement for the DPR swath was that the largest footprint (a size of about 40 km) of the microwave radiometer onboard the same core satellite should be covered. Another requirement was that the swath should be wide enough to cover the core area of a rain system. TRMM PR data analysis shows that (in tropical area of 120–150E and 10–20N, August 1998) 95% of the core area of precipitation with more than 35 dBZ has a linear dimension of less than 100 km. The current major parameters of the DPR are: Frequency: 13.60 and 35.55 GHz (dual-wavelength). Range resolution: 250 m (14 GHz), 250/500 m (35 GHz). Swath: about 220 km (14 GHz), about 100 km (35 GHz). Horizontal resolution: about 5 km at nadir. Sensitivity: nearly 12 dBZ for 35 GHz. For the DPR algorithms, the beam matching at the two channels is crucial. The DPR is carefully designed to realize sufficient beam matching. This issue is related to not only the sensor design but also calibration techniques for the beam patterns.
3 TYPES OF THE DPR ALGORITHMS The DPR algorithms may fall into several types according to the combination of the sensor data depending on the swath where the algorithms are applied. The algorithm for the narrowest swath may be for the swath of the
Dual-Wavelength Radar Algorithm
227
Ka-band radar, where both Ka- and Ku-band radar data will be available. The second one may be for the Ku-band radar swath, and the third may be for the microwave radiometer swath.
3.1
For the Ka/Ku narrow swath
The first one is the combination of the Ka-band and Ku-band radar data. This combination can be applied in the narrowest swath of the DPR, since both data must be available. This is the primary algorithm of the DPR. Here several algorithms are conceivable. The first one is a Ku-band single wavelength algorithm which is the same as the TRMM PR algorithm (Iguchi et al. 2000) utilizing the Hitschfeld-Bordan solution with iterations and the so-called surface reference technique (SRT) (Meneghini et al. 2000). The second one may be the Ka-band single wavelength method. Though the Ka-band radar suffers from strong attenuation, it may well be used for weak rain cases. In addition, the SRT should work well because of strong attenuation. Thus, an algorithm like the TRMM PR algorithm may be applicable to the Ka-band radar data. The third is the Ka/Ku combined algorithm. This algorithm is a new one and is expected much to improve the precipitation profiling. The fourth may be the combined algorithm using Ka/Ku radar data and microwave radiometer data. The attenuation caused by components other than precipitation particles becomes crucial in the Ka-band. Gaseous and cloud attenuation may be estimated from the microwave radiometer data even though the pixel sizes are different. Algorithms for a dual-frequency radar have been investigated using airborne radars, and are still being developed (Meneghini et al. 1992, 1997; Kozu et al. 1991). The algorithms may fall into several categories. One is a simple dual-wavelength radar algorithm that utilizes the difference of the rain attenuation between two radar reflectivity profiles. The algorithm in this category takes advantage of the fact that the rain rate is less sensitive to the variation of the raindrop size distribution (DSD) than the radar reflectivity. This algorithm is applicable even to two profiles over a limited range interval to retrieve rain rate. One crucial disadvantage is that this technique suffers from the error when the scattering deviates from Rayleigh scattering (in other words, Mie scattering effect). The second is more sophisticated using a two-parameter DSD and it retrieves two parameters at each range bin using two radar reflectivities at Ka and Ku-band radio waves (Mardiana et al. 2004) (see next section). This technique uses a coupled pair of the first order differential equations, and needs initial values. The SRT may be used for the determination of the initial values. Another interesting extension from the TRMM PR is the application of the SRT to the DPR. The use of SRT in the DPR algorithm is expected to improve the accuracy of rain attenuation correction over the ocean, and the
K. Nakamura and T. Iguchi
228
combination of surface echoes at two channels may make the SRT applicable over land. As another application, the SRT has its own beam filling problem (Nakamura 1991), which in turn means that the SRT for the DPR may be used to correct the beam filling errors.
3.2
For the Ku/Ka wide swath
The Ku-band radar is currently designed to have a more than two times wider swath than the Ka-band radar. Inside the Ka-band radar swath, precise precipitation profile measurement can be done. The extension of the precise profile and/or the raindrop size distribution to the wider swath is the core of the algorithm. This part, however, has not yet been developed.
3.3
For the radiometer swath
The combination of PR and radiometer data has been applied in a TRMM algorithm. The basic idea is to use the PR fine resolution data to correct the beam filling bias in the microwave radiometer rain retrieval. One crucial issue in the passive microwave rain retrieval is the height of the liquid precipitation layer (Masunaga et al. 2002; Ikai and Nakamura 2003). The error in the estimate of the liquid precipitation layer directly affects the accuracy of rain retrieval in the so-called emission mode where the column integrated absorption is primarily used. The precipitation profile retrieved from the DPR has a potential to improve the instantaneous rain retrieval using the microwave radiometer. However, these algorithms which use the DPR data to improve directly the microwave radiometer rain retrieval is yet to be well developed.
4 FORMULATION OF THE DPR COMBINED ALGORITHM In this section, we describe a formulation of the DPR profiling algorithm. We derive a coupled pair of differential equations for two parameters in the gamma distribution model for the DSD in this example. However, the formulation can be easily generalized to other types of model. In particular, almost exactly the same formulation can be used if one of the two parameters is proportional to the scale factor (N0 in this example) because we do not use any special property of the gamma distribution when we define Ib(D0) and It(D0) below. In a gamma distribution model, the number density N(D,r) of drops whose diameter is between D and D+dD at range r is given by
N ( D, r )dD = N 0 (r ) D μ exp[−( μ + 3.67) D / D0 (r )]dD
(1)
Dual-Wavelength Radar Algorithm
229
where D0 is the mass-weighted median diameter of the DSD, and μ the shape parameter. The measured radar reflectivity factor Zm(r) is given as the true effective radar reflectivity factor Ze(r) multiplied by the attenuation factor: r
Z m (r ) = Z e (r ) exp[−0.2 ln(10) ∫ k ( s )ds ]
(2)
0
Here, k is the specific attenuation expressed in dB km–1 and is defined in terms of the DSD parameters and the extinction cross section σ t ( D, λ ) of a drop with diameter D for electromagnetic waves with wavelength λ
k (r ) = ck N 0 (r ) I t ( D0 (r ), μ , λ ) = ck N 0 (r ) ∫ σ t ( D, λ ) D μ exp[−( μ + 3.67) D / D0 (r )]dD
(3)
ck = log10 e if It is in m2, N in m-3, and r in m. However, if It is measured in mm2 and r is in km, ck = 10 −3 log10 e . Similarly, the effective radar reflectivity factor Ze(r) is defined as
Z e (r ) = N 0 (r ) I b ( D0 (r ), μ , λ )
(4)
where
I b ( D0 (r ), μ , λ ) = cZ (λ ) ∫ σ b ( D, λ ) D μ exp[−( μ + 3.67) D / D0 (r )]dD
(5)
and 2
cZ (λ ) = λ4 /(π 5 K )
(6)
We assume that the shape factor μ does not change with range. In what follows, we are not going to indicate the dependence of functions Ib and It on μ and λ explicitly. Suppose we have measurements of Zm(r) at two frequencies, f1 and f2. If we use suffixes 1 and 2 for corresponding variables, equation (1) becomes r
Z m1 (r ) = N 0 (r ) I b1 ( D0 (r )) exp[−a ∫ N 0 ( s ) I t1D0 ( s )ds ] 0
(7)
K. Nakamura and T. Iguchi
230 r
Z m 2 (r ) = N 0 (r ) I b 2 ( D0 (r )) exp[−a ∫ N 0 ( s ) I t 2 D0 ( s )ds ]
(8)
0
Here a = 0.2ck ln(10) . Taking the logarithms of these equations and differentiating them with respect to r, we obtain
d d ln N 0 d ln( I b1 ( D0 )) dD0 (r ) ln[ Z m1 (r )] = + − dr dr dD0 dr
(9)
a exp(ln N 0 ) I t1 ( D0 (r )) and
d d ln N 0 d ln( I b 2 ( D0 )) dD0 (r ) ln[Z m 2 (r )] = + − dr dr dD0 dr
(10)
a exp(ln N 0 ) I t 2 ( D0 (r )) Rearrangement of these equations gives
dD0 (r ) 1 = dr b1 − b2 ⎡ d ln(Z m1 ) d ln(Z m 2 ) ⎤ ×⎢ − + aN 0{I t1 ( D0 ) − I t 2 ( D0 )}⎥ dr dr ⎣ ⎦
(11)
and
1 dN 0 (r ) 1 = N 0 dr b1 − b2 d ln( Z m 2 ) ⎡ d ln( Z m1 ) ⎤ × ⎢b2 − b1 + aN 0{b2 I t1 ( D0 ) − b1 I t 2 ( D0 )}⎥ dr dr ⎣ ⎦
(12)
where
d ln( I b1 ( D0 )) dD0 d b2 = ln( I b 2 ( D0 )) dD0
b1 =
(13) (14)
Dual-Wavelength Radar Algorithm
231
These are the equations which D0(r) and N0(r) must satisfy. These equations are numerically solvable and give a set of solutions. Since only the derivatives of ln Z m1 and ln Z m 2 appear in equations (4) and (5), constant offsets in ln Z m1 and ln Z m 2 are irrelevant in possible solutions. Among the possible solutions we need to choose the one that most closely reconstructs the measured profiles, Zm1 and Zm2. This is the case when the radar is well calibrated and when there is no uncertainty of attenuation other than the attenuation by rain itself. If we need to solve the equations in the interval that lies in the middle of the entire observation range with unknown attenuation to the boundary of the interval, we need to know the attenuation to a range that is within the interval. Solving the equations backward for a given pair of the path-integrated attenuations to the surface is such a case. When b1=b2 the right-hand sides of equations (4) and (5) become singular. This situation happens at the point where the quantity Ib1/Ib2 takes its maximum and also at the point D0=0.0. The latter case corresponds to the Rayleigh scattering case in which the particle dimensions are much smaller than the wavelength. In this case, since Ib1=Ib2, the combination of equations (2) and (3) gives
ln Z m1 (r ) − ln Z m 2 (r ) = A1 (r ) − A2 (r )
(15)
where rs
A1 (rs ) = −a ∫ N 0 ( s ) I t1 ( D0 ( s ))ds
(16)
0
rs
A2 (rs ) = −a ∫ N 0 ( s ) I t 2 ( D0 ( s ))ds
(17)
0
This is the case when the difference between the radar reflectivity factors measured at two different frequencies can be attributed solely to the attenuation difference. Note that the non-Rayleigh characteristics appear firstly in It and then Ib as the diameter D0 increases. Therefore, this formulation may be applicable to the combination of some relatively long wavelengths. Note that in this case we cannot separate the effect of N0 from D0 on the attenuation difference. In other words, we can estimate only a single parameter in the DSD model. Therefore, we need to assume some functional relationship between N0 and D0. Nevertheless, in the Ka-band, the rainfall rate R is nearly proportional to the attenuation and its dependence on the DSD parameters is rather small. Therefore, without retrieving N0 and D0
232
K. Nakamura and T. Iguchi
separately, we can still infer the rain distribution (this corresponds to the simple combined algorithm).
5 REMARKS ON THE DPR COMBINED ALGORITHM The objective of the DPR combined algorithm is to retrieve precipitation profiles. The radars are assumed to detect only precipitating particles, since the expected sensitivities are 17 dBZ and 12 dBZ for the Ku and Ka-band radar, respectively. This sensitivity is too poor to detect non-precipitation particles such as cloud particles. Though the precipitation profile consists of liquid particles (rain), solid particles (ice, graupel, etc.) and a mixture of them, in the previous section we restricted our problem only to liquid particles, since the profiling of solid and/or mixed phase particle using the DPR has only a few investigations (Liao et al. 1997). For the liquid particle profiling, the target is to retrieve the raindrop size distribution. Since only two profiles will be obtained from the DPR measurement, a two parameter DSD may be appropriate. We have not only the two profiles but also the surface signatures. As well developed as the SRT in TRMM PR algorithm, the surface signatures can be used to correct the rain attenuation and/or calibrate the radar constants. The surface signatures can be used as additional constraints in the rain estimation or be used as an initial value to solve the equations for profiling. A method similar to the SRT is the use of the mirror image of rain echo over ocean. Since the ocean surface is a good reflector of microwaves, the rain echo over ocean has the mirror images. The ratio of the intensity of the mirror image to the direct echo includes the effect of path attenuation and surface conditions (Meneghini and Atlas 1986; Li and Nakamura 2002). However, a simple use of mirror images in the TRMM PR data has not contributed to the improvement of rain retrieval yet. The mirror image observed by the DPR may help understanding the characteristics of the mirror image and for the exploration of its use. The accuracy of the estimation of the rain profile depends on the range over which the finite differences are applied, since the radar signature fluctuates due to incoherent scattering. For the conventional Z-method, the fluctuation is not a major error source, but for the DPR profiling technique, it becomes a major error source, since the fluctuation is not negligible after taking finite differences. The practical DRP algorithm should combine the low range resolution profiling algorithm and a high resolution singlewavelength profiling algorithm. The Ka-band radar data are not available in strong rain due to severe attenuation. In this case, only a single wavelength algorithm can be applied but may be with surface signatures at two wavelengths. An algorithm that realizes a smooth transition from a dual-frequency mode to a singlefrequency mode should be developed.
Dual-Wavelength Radar Algorithm
233
Another important issue is the beam filling and beam mismatching effect. Due to the finite differences, any error may be emphasized and potentially become a significant error source.
6 CONCLUSIONS The DPR algorithms consist of several categories. The algorithms in each category still need further development. Among them, the Ka/Ku-band combined algorithm is very important and also very interesting, and is one of the keys for the success of the GPM. The basic idea of the combined algorithm is to determine two parameters of the raindrop size distribution at each range bin using the two profiles of measured radar reflectivity from the DPR. The two parameter retrieval seems feasible at least for the liquid precipitation region. After the development of the rain retrieval for the TRMM precipitation radar, radar rain retrieval showed a big progress. Before the TRMM PR, radar rain estimates are thought to be rather qualitative instead of quantitative. The rain retrieval using the ground-based radars are operationally calibrated by available rain gauge networks in Japan. There is a common phrase “distribution is by radar and accuracy is by rain gauges.” The TRMM PR changed this situation at least for the space-borne radar. Before the TRMM launch, attenuated radio waves at 14 GHz adopted in the PR were thought to make the rain retrieval complicated. After the launch, however, the attenuating frequency was found to have a potential to deduce DSD variations by incorporating the SRT. The combined algorithms in the DPR will fully utilize the attenuating characteristics of the radar radio waves. The DSD variation is a big issue even though the combination of the SRT gives us some hints. One of the biggest expectations to the DPR is the global mapping of the DSD variation. There are, however, other precipitations as solid precipitation and mixed phase precipitations. In addition, gaseous and cloud attenuation should also be taken into consideration. Much more studies with simulations and field tests are essential.
7 REFERENCES Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain-profiling algorithm for the TRMM precipitation radar. J. Appl. Meteor., 39, 2038–2052. Ikai, J. and K. Nakamura, 2003: Comparison of rain rates over the ocean derived from TRMM microwave imager and precipitation radar. J. Atmos. Oceanic Technol., 20, 1709– 1726. Kozu, T., K. Nakamura, R. Meneghini, and W. C. Boncyk, 1991: Dual-parameter radar rainfall measurement form space: A test results from an aircraft experiment. IEEE Trans. Geosci. Remote Sens., 29, 690–703.
234
K. Nakamura and T. Iguchi
Li, J. and K. Nakamura, 2002: Characteristics of the mirror image of precipitation observed by the TRMM precipitation radar. J. Atmos. Oceanic Technol., 19, 145–158. Liao, L., R. Meneghini, T. Iguchi, and A. Detwiler, 1997: Estimation of snow parameters from dual-wavelength airborne radar. Prepr. 28th Conf. Radar Meteor., Am. Meteor. Soc., 510–511. Mardiana, R., T. Iguchi, and N. Takahashi, 2004: A dual-frequency rain profiling method without the use of surface reference technique. IEEE Trans. Geosci. Remote Sens., 42, 2214–2225. Masunaga, H., T. Iguchi, R. Oki, and M. Kachi, 2002: Comparison of rainfall products derived from TRMM microwave imager and precipitation radar. J. Appl. Meteor., 41, 849–862. Meneghini, R. and D. Atlas, 1986: Simultaneous ocean cross-section and rainfall measurements from space a nadir-looking radar. J. Atmos. Oceanic Technol., 3, 400–413. Meneghini, R., T. Iguchi, T. Kozu, L. Liao, K. Okamoto, J. A. Jones, and J. Kwiatkowski, 2000: Use of the surface reference technique for path attenuation estimates from TRMM precipitation radar. J. Appl. Meteor., 39, 2053–2070. Meneghini, R., T. Kozu, H. Kumagai, and W. C. Boncyk, 1992: A study of rain estimation methods from space using dual-wavelength radar measurements at near nadir incidence over ocean. J. Atmos. Oceanic Technol., 9, 364–382. Meneghini, R., H. Kumagai, J. R. Wang, T. Iguchi, and T. Kozu, 1997: Microphysical retrievals over stratiform rain using measurement from an airborne dual-wavelength radar-radiometer. IEEE Trans. Geosci. Remote Sens., 35, 487–506. Nakamura, K., 1991: Biases of rain retrieval algorithms for spaceborne radar caused by nonuniformity of rain. J. Atmos. Oceanic Technol., 8, 363–373.
19 A NEXT-GENERATION MICROWAVE RAINFALL RETRIEVAL ALGORITHM FOR USE BY TRMM AND GPM Christian Kummerow1, Hirohiko Masunaga1, and Peter Bauer2 1
Department of Atmospheric Science, Colorado State University, Fort Collins, CO, USA European Centre for Medium-Range Weather Forecasts, Reading, UK
2
1 INTRODUCTION Passive microwave rainfall algorithms have evolved steadily from those designed for the early Electronically Scanning Microwave Radiometer (ESMR), through the Scanning Multichannel Microwave Radiometer (SMMR) on Nimbus-7, and the Special Sensor Microwave Imager (SSM/I) instruments flying on the Defense Meteorological Satellite Program (DMSP). A number of algorithms fitting roughly three classes have emerged. These are (a) the “emission type” algorithms (e.g., Wilheit et al. 1991; Berg and Chase 1992; Chang et al. 1999) that use low-frequency channels to detect the increased radiances due to rain over radiometrically cold oceans; (b) the “scattering” algorithms (Spencer et al. 1983; Grody 1991; Ferraro and Marks 1995) that correlate rainfall to radiance depressions caused by ice scattering present in many precipitating clouds; and (c) the “multichannel inversion” type algorithms (Olson 1989; Mugnai et al. 1993; Kummerow and Giglio 1994; Smith et al. 1994; Petty 1994; Bauer et al. 2001; Kummerow et al. 2001) that seek to invert the entire radiance vector simultaneously. Among these, the Wilheit et al. (1991) and Kummerow et al. (2001) algorithms are used operationally for the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) as well as the Advanced Microwave Scanning Radiometer (AMSR-E) flying on Aqua, while the Wilheit et al. (1991) and Ferraro and Marks (1995) algorithms are used with SSM/I in the Global Precipitation Climatology Project (GPCP) over ocean and land, respectively. The Bauer et al. (2001) algorithm is used at ECMWF 235 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 235–252. © 2007 Springer.
236
C. Kummerow et al.
for rain assimilation experiments. In each case, algorithms have been optimized for the corresponding satellite sensor. Algorithm intercomparison efforts, initially aimed at identifying the “best” algorithms have not been able to make much headway, as each algorithm appears to have strengths and weaknesses related to specific applications, while none appears to be universally better than the others. The next advance in global precipitation monitoring, the Global Precipitation Measurement (GPM) mission, is providing new impetus towards a common algorithm framework. The GPM concept consists of a core satellite, with a dual-frequency precipitation radar (DPR), and a multichannel microwave imager (GMI). This component is similar in concept to the TRMM design but with improved radar capabilities and an orbit that will cover between 65–70° of latitude. In addition, the GPM concept uses a constellation of operational and dedicated radiometers to produce global, three hourly rainfall products required by many applications. The fact that radiometers for the GPM constellation are not fully specified and will evolve throughout the mission based on contributions from a number of different space agencies immediately imposes a number of highlevel requirements upon any algorithm designed for these sensors. Of utmost importance is the need for a transparent, parametric algorithm that insures uniform rainfall products across all sensors. The requirement for transparency is clear. A mission of GPM’s scope should not rely on a single black box operated by any one individual. Instead, it requires an open architecture that will allow the international community to participate in the algorithm development, its refinement, and its error characterization. The requirement for a parametric algorithm is also self-evident. Since GPM is being designed as an ongoing cooperative concept among many agencies, algorithms cannot be designed for specific radiometers with defined frequencies, viewing geometry, spatial resolutions or noise characteristics. The algorithm should be applicable to any sensor. Such a requirement leads naturally to a generalized framework that avoids the need for specific frequencies for their application. Finally, the algorithm should be robust in such a way that differences between sensors can be confidently interpreted as physical differences between observed scenes rather than artefacts of the algorithm. Together with the above requirements, algorithms designed for the future should also be able to fully characterize uncertainties at any space and time scale being considered by the users. This ranges from instantaneous estimates needed for many hydrologic and weather forecasting applications to large space and time averages required for climate model verification and climate trend monitoring. While such a requirement is also perhaps selfevident, such a complete error characterization does not currently exist and is undoubtedly the greatest challenge facing the community.
Microwave Rainfall Retrieval Algorithm
237
2 THE ALGORITHM Rainfall retrieval algorithms are not fully constrained. Instead, a priori information must be supplied to help constrain the estimated 3-dimensional (3D) properties of precipitating clouds. The requirement that the GPM algorithm be adaptable to any satellite sensor and that it produces realistic uncertainty estimates for global application reduces the large set of previous algorithms to those that involve physical forward/inverse modeling where the statistical properties of the a priori information and the models can be formulated in a consistent way. In a physical framework, the optimum estimate of a state vector (precipitation profile), x, must be obtained using an observation vector (brightness temperatures), y, plus additional a priori information. Due to errors in modeling and observation (error covariance R), the relation between state and observation is usually described by probability density functions (pdf’s)This can be formalized with Bayes’ theorem (e.g., Rodgers 2000):
P(x | y ) =
P ( y | x) P ( x) P(y )
(1)
P(x|y) is the posteriori probability of x when y is observed. P(y|x) is the probability of making observation y when x is present, while P(x) and P(y) are the a priori probabilities of x and y, respectively. The latter may come from global statistics of state and observations. The determination of P(y|x) requires a model that translates between state and observation space. This model may also be used to compute P(y) if P(x) is assumed to fully describe the a priori distribution of x. Examples of the application of the above principle are the ‘Bayesian’ rainfall retrieval schemes that found rather wide distribution in recent years (Evans et al. 1995; Kummerow et al. 1996; Olson et al. 1996; Haddad et al. 1997; Marzano et al. 1999; Bauer et al. 2001, Kummerow et al. 2001; Viltard et al. 2004). One particular problem associated with these rainfall retrievals is that the model that connects states and observations, i.e., y = F(x) + ε (where ε is the modeling error), is generally nonlinear. This immediately implies that the inversion of this relation is state dependent, and the inversion must be formulated differently depending on whether (a) a first guess of the actual state, xb, and its error covariance, B, is known and Gaussian with respect to the true state or (b) only a pdf of state x is known from which the pdf of y can be calculated. If (a) applies, Eq. (1) can be transformed to:
1 ⎧ 1 ⎫ T T P (x | y ) = exp⎨− [y − F(x)] R −1 [y − F(x)] − [x − x b ] B −1 [x − x b ]⎬ (2) 2 ⎩ 2 ⎭
C. Kummerow et al.
238
The probability of P(x|y) is maximized when the first derivative of Eq. (2) vanishes. This can be solved numerically by iterative procedures, and represents the “variational” retrievals. If Eq. (2) applies, it is more appropriate to seek the expected value of x. From practical considerations, the expected value is often expressed as (Olson et al. 1996):
E ( x) =
∑x i
i
{
}
exp − 0.5[y − F (x i )] R −1 [y − F(x i )]
{
T
}
∑ exp − 0.5[y − F(x i )] R −1 [y − F(x i )] T
(3)
i
even though the formulation in Eq. (3) makes the assumption that P(x) and P(y) are well known. This will be called “Bayesian” method even though both approaches are based on Bayes’ theorem. Both solutions employ a forward model that often consists of combined cloud resolving and radiative transfer models, the latter involving clear-sky atmospheric and surface models. In the variational framework, these models have to be directly inverted because the difference between model-calculated and observed brightness temperatures must be translated into increments to the initial physical state. Here, adjoint models have recently been developed for rainfall retrieval purposes (Moreau et al. 2003). The first-guess state and its error characteristics, however, are difficult to obtain for precipitation and this method is only possible in a well-constrained large-scale model (Moreau et al. 2003, 2004). The main advantage of such a method is its global applicability and its flexibility with respect to any input data, while its disadvantage is the requirement of a well-defined first guess and the computational cost. In the Bayesian method, the biggest challenge is the definition of the a priori database, P(x), because it is not well known for precipitating clouds. Historically, Bayesian schemes used precipitation profiles derived from a set of existing cloud-resolving model (CRM) simulations to construct the a priori database of potential precipitation structures that might be seen by a radiometer. The CRMs provide a physically consistent set of full 3D hydrometeor and latent heating profiles. They also provide a simple method to use understood physical processes to constrain an inversion. The biggest disadvantage of the Bayesian algorithm is its lack of general applicability because only a few CRM simulations are available (and useful) to construct a valid P(x). A first-guess constraint may be possible to help constrain P(x) in the future. Inherent to both methods is the impact of the dynamical and microphysical formulations in the forward model that often dominate the uncertainties of the radiative transfer modeling.
Microwave Rainfall Retrieval Algorithm
239
3 THE PARAMETRIC ALGORITHM FRAMEWORK Most likely, future algorithms will be composed of elements present in both approaches, i.e., pdf-type estimators with static databases and variational elements where reliable first-guess information is available. The Bayesian, as well as variational techniques have the added advantage that they are intrinsically parametric and uncertainties in the a posteriori rainfall can be computed in a straightforward fashion. Nonetheless, the approaches are not without pitfalls. Variational approaches are contingent upon a good first guess, which is often difficult to make in the case of precipitation systems. Current Bayesian schemes, in addition to relying on incomplete CRM simulations, use procedures such as rainfall screening, freezing-level estimates and convective/stratiform classification in order to improve the retrieval performance. Incomplete databases require procedures to solve the problem when none of the simulated profiles are close to the observed brightness temperature vector. More subtle, but perhaps more important to both schemes, are the errors in the a priori database itself. Errors in the CRM simulations will cause radiometers with different channel combinations to retrieve different rainfall amounts. This potential aliasing is simple to illustrate with a hypothetical CRM that consistently produces too much ice in the simulation—thus creating simulations with large Tb depressions at high frequency for even modest rainfall rates. A sensor with only low frequency channels (e.g., 10–37 GHz) may not be susceptible to this problem and will retrieve approximately the correct rainfall (all else being correct). A sensor with only high frequency channels (e.g., 85 and 150 GHz), on the other hand, will match large Tb depressions to relatively modest rain cases with large Tb depressions found in the CRM simulation. This will cause a consistent underestimation and a different result than that obtained from the first sensor. Conceptually, a parametric algorithm must therefore address two distinct issues. It must avoid any channel specific procedures in the algorithm, and it must create an a priori database that is consistent with all the brightness temperatures that may be observed by individual radiometers. Avoiding channel specific procedures will be seen to be a relatively straightforward task in both the Bayesian and variational frameworks. Building a representative a priori database with a verifiable error model is a far more challenging task. The error model, in particular, is very difficult to construct because of the role of CRMs in the forward computations. While they are useful in the sense that they add physical constraints to the clouds that may be retrieved, they are extremely difficult to verify quantitatively since they are constructed to simulate physically consistent scenes rather than the details
240
C. Kummerow et al.
of any one observed cloud realization. A more complete and verifiable a priori database appears thus to be crucial for any algorithm in the GPM era. Following is a description of a prototype parametric algorithm using both variational as well as Bayesian methods. While the various aspects of the algorithm are being developed separately and for different sensors, they are presented here as parts of the same conceptual algorithm for illustration purposes. In this conceptual algorithm, the a priori database is constructed from a combination of TRMM precipitation radar (PR), TMI and CRM information when the TMI footprint contains rainfall as determined from the PR. When the radiometer footprint does not contain rainfall, a variational technique is used with the radiometer data to obtain the clear air parameters. Together, these two components lead to a consistent 3D distribution of geophysical parameters that are fully representative of the observed scenes and fully consistent with the observation vector. Benefits of this more representative database are discussed in Viltard et al. (2004). The rainfall retrieval itself follows the Bayesian formalisms cited earlier.
3.1 The non-raining simulations Over oceans, passive microwave radiances depend upon column-integrated water vapor, cloud liquid water, sea surface temperature, and surface wind speed. These geophysical parameters can be retrieved simultaneously from the TMI itself for which the TRMM PR shows no rainfall. Techniques such as those described in the literature (e.g., Wentz 1997) do exactly this. Unfortunately, that technique has some shortcomings with respect to the current objectives. It only works over ocean, and it seeks consistency only among the channels used in the particular inversion. An alternative algorithm for clear-sky applications makes use of the previously introduced variational approach. Due to the many free parameters, in particular over land surfaces, the physical framework has to be kept very simple and only a few bulk parameters may be retrieved. As an example, we chose a set of four or six free parameters for ocean and land, respectively. Over ocean, these are surface skin temperature, near-surface wind speed, water vapor path, and cloud liquid water path, while over land these are surface skin temperature, effective water coverage, vegetation coverage, surface roughness, water vapor path, and cloud liquid water path. The surface skin temperature also determines the effective atmospheric temperature by assuming a constant lapse rate. The effective atmospheric temperature has to be understood as the temperature of the lower atmosphere where most of the water vapor is present, while the effective water coverage over land summarizes the true coverage with open water and soil moisture. The effect of soil moisture and open water on land surface emissivity is very similar. Atmospheric absorption was calculated according to Liebe et al. (1992), sea-surface emissivity with the model of Ellison et al.
Microwave Rainfall Retrieval Algorithm
241
(2003) and land surface emissivity with the model of Bauer and Grody (1995). The atmosphere consists of seven layers with constant depths.
Figure 1. Example of variational retrieval of surface skin temperature (a), effective water coverage (b), effective vegetation coverage (c) and water vapor path (d) using SSM/I data on November 1, 2003. Scales at the bottom refer to units [K], [], [], [kg m–2], respectively.
Figure 2. TB-departures (observation minus simulation) using first-guess (a, b) and after variational retrieval (c, d) at 19.35 GHz (a, c) and 85.5 GHz(b, d) and horizontal polarization. Scales at the bottom refer to units [K].
First-order climatological values were assumed for the above parameters to initialize the minimization of some SSM/I overpasses over South America and the Southern Caribbean on November 1, 2003. Figure 1 shows the resulting retrievals for the surface skin temperature, effective water coverage, vegetation coverage and water vapor path, respectively. The fields represent reasonable distributions showing the Amazon River basin in both Fig. 1b and 1c as well as the orography-dependent surface temperature distribution. The water vapor fields over ocean reach very low values in the presence of clouds, which indicate a possible aliasing effect between water vapor and liquid water absorption. Nonetheless, these results indicate that a variational retrieval is feasible for clear-sky applications providing background fields
242
C. Kummerow et al.
for the hydrometeor retrievals in the presence of rain. More realistic first-guess values as well as error covariances may be obtained from climatological fields produced by global model analyses. Potential refinements in the physical models used in the inversion may also lead to further improvements. Figure 2 illustrates the brightness temperature departures before and after the retrieval at 19.35 and 85.5 GHz. The departures are quite large in areas with strong water vapor gradients and in the presence of clouds over sea with values above 20 K. This indicates that the first-guess values chosen for this example are only appropriate for illustration rather than operational application; however, the minimization performs well and reduces the departures to within instrument error limits. Clouds and light precipitation may be present in those areas where large departures remain after the retrieval (in particular at 85.5 GHz). The problem with precipitation is artificial, as the real database would be constructed from TMI data for which rain/no rain information is available from the PR.
3.2 The raining scene Figure 3 illustrates the overall flow of the algorithm described here to derive precipitation profiles consistent with both radar and radiometer measurements. The non-raining parameter retrievals, indicated by blue-colored items in Fig. 4, were introduced in the previous section. In this section, the rain-profiling scheme using PR, TMI, and CRM information is outlined. The PR identifies pixels with radar echoes significantly above the noise threshold as “rain certain.” The weakest detectable signal by PR corresponds roughly to 0.5 mm h–1 in rain rate. The GPM 35-GHz radar is currently planned to have a threshold of approximately 12 dBZ, which corresponds to roughly 0.2 mm h–1. If PR detects a rain signal, the rain profile that best fits the PR reflectivity profile is selected from a set of precomputed CRM simulations. The reflectivity of the cloud-model profiles was obtained by computing single particle backscattering and extinction properties based upon Mie theory and assuming a gamma drop size distribution (DSD) with a given median volume diameter (D0) and µ = 3. As an example, the initially assumed DSD model may be taken, which was constructed to be consistent with the Z-R relations assumed in the TRMM PR operational rain-profiling algorithm (2A25) developed by Iguchi et al. (2000). The particle size distributions of other hydrometeor species are the same as adopted by the CRM except for melting particles. Here, the microwave properties of melting hydrometeors are simplified in such a way that the particle size distributions are linearly transformed under an averaged dielectric function from ice to liquid within a half-kilometer layer below the freezing height. This simplified treatment of melting particles could be
Microwave Rainfall Retrieval Algorithm
243
replaced with a more elaborate microphysical model in the future (Bauer 2001; Olson et al. 2001; Battaglia et al. 2003). The CRMs used in the study were the Goddard Cumulus Ensemble Model (GCE) and the University of Wisconsin Non-hydrostatic Modeling System (UW-NMS), which are the same simulations used in the a priori database used in the Goddard Profiling Algorithm (Kummerow et al. 2001).
Figure 3. Algorithm flowchart. Blue colored items are related to the non-raining (NR) parameter retrieval, yellow to the PR profile matching, and red to the comparison of matched profiles in the Tb space.
The best fit in the PR reflectivity matching is defined as the one having the least root-mean-squared difference between observed (PR-1C21 attenuationuncorrected reflectivity) and computed reflectivity. When the observed PathIntegrated Attenuation (PIA) from the PR is sufficient to provide a robust signal, there is additional DSD information available from the radar. In this case, the best-fit solution with respect to both the reflectivity profile as well as the PIA is sought by searching simulated profiles with several different DSD assumptions. While computationally different from the PR algorithm developed by Iguchi et al. (2000), philosophically this step matches the PR procedure by
244
C. Kummerow et al.
adjusting the retrieved DSD to match both the reflectivity and the PIA when it is deemed robust. Figure 4 shows a snapshot of surface rain rate given by the matched CRM profile and 2A25 surface rain, exhibiting good qualitative agreement. A direct pixel comparison of the observed and reconstructed surface rainfall for this scene shows a bias of 1.5% with a correlation of 0.96. The light grey portion of Fig. 3, which includes an iterative step if PIA is robust, shows the flow of the above procedure. At this point in the retrieval, the raining and non-raining scenes must be merged. As discussed in the previous section, the non-raining retrievals are applied to all TMI footprints in which PR observed no rain. The rain retrieval, however, is applied to PR pixels that generally have higher spatial resolution than the TMI footprints. This difference in resolution can lead to small areas within partially raining TMI footprints for which the clear air retrieval could not be performed but for which PR observes no rain. These areas must be filled by an interpolation scheme before the final step in the raining retrieval can be completed. This interpolation scheme is also used to prescribe the surface conditions under the raining pixels.
Figure 4. Top: Surface rain rate given by CRM profiles that best fit the measured PR profiles. Bottom: 2A25 surface rain rate for the same scene as the top panel.
Figure 5 illustrates this procedure, originally developed by Shin and Kummerow (2003), for the cloud liquid water field over ocean. All clear air fields are treated in a similar manner. Figure 5a represents the TMI retrieval for column water vapor (CWV) in this example. Figure 5b shows the CWV field associated with the raining retrievals, and Fig. 5c the final CWV field in which TMI CWV has been mapped to PR pixel locations and missing values have been interpolated. The slightly lower CWV values in precipitation (relative to the non-raining surroundings) might be an artifact
Microwave Rainfall Retrieval Algorithm
245
of the algorithm still under development. Inspection of the raining profiles, however, indicate they are nearly all saturated. As such, the lower CWV values might be real if one takes into account the lower temperatures associated with evaporation cooling associated with light precipitation. The interpolation over land will introduce greater uncertainties as the effective water coverage, vegetation cover, and surface roughness can vary rapidly. Precipitation profiles obtained from the PR matching technique are assigned to the satellite swath. The geophysical parameters unobservable by PR such as SST, surface wind speed, water vapor, and cloud water are provided from the TMI retrieval in non-raining scenes and interpolated to the raining field of view (FOV). Radiative transfer calculations using the Eddington approximation (Kummerow 1993) are then performed along slant paths that intersect a few neighboring PR pixels to properly take into account the TMI incidence angle of 52.8°. The computed brightness temperatures are convolved with the Gaussian antenna pattern using the 3dB-beam width of each TMI channel. Figure 6 shows the retrieved liquid and ice water contents, along with the computed brightness temperatures along the scan center of the rain feature shown in Fig. 4. The observed and computed brightness temperatures generally exhibit good but not perfect agreement. Figure 5. Simulation steps for a non-raining scene over ocean. (a) Shows the columnar water vapor retrieved from the TMI data for pixels in which PR detected no rain. (b) Column water vapor in raining pixels as determined through the selected cloud resolving model profile. (c) The merged and interpolated final column water vapor field.
The dark grey portion of Fig. 3 summarizes the final procedure. If the computed Tb’s at the lower frequency channels are lower than the observed ones, the assumed drop sizes can be decreased in order to increase the liquid water content determined from PR. As can be seen from the diagram, however, this can only be done for those pixels for which PR is not able to determine its own DSD through the PIA estimate. Variational
246
C. Kummerow et al.
methods that seek to adjust the DSD to simultaneously fit PR and TMI observations are also possible. Such solutions may eventually prove superior to the current approach. They are, however, less transparent. In the current formulation, the final iterative procedure will adjust DSD, but only if the DSD is the one assumed by PR and not when it can be directly observed by the sensor. In addition to any Tb disagreements in the emission channels, Fig. 6 shows occasional discrepancies in the scattering channel (85 GHz), which can be attributed to an uncertainty in the microphysical treatment of ice hydrometeors in CRM. This discrepancy is minimized by interactively updating the ice density in the CRM model. Precipitation profiles consistent with both radar and radiometer are thus obtained by repeating the entire procedure with updated DSD and ice density models.
3.3 The a priori databases Construction of the a priori databases is straightforward once the 3D raining and non-raining parameters derived from TRMM TMI and PR swath overlap data have been determined. Compared to previous efforts that relied solely on the CRMs to provide cloud structures, the current methodology insures that the a priori database is more fully grounded in observations, which would improve the databases’ representativeness of actual rainfall spectra. Figure 6. Top two panels: Vertical cross section at the scan center of precipitation water and ice given by CRM profiles that best fit the measured PR profiles. Bottom four panels: Observed TMI brightness temperatures (solid lines) and computed brightness temperatures (dashed lines) at 10 GHz, 19 GHz, 37 GHz, and 85 GHz (vertical polarization).
Through the database construction process, furthermore, it is possible to relate the derived rainfall profiles to the environmental geophysical parameters controlling rainfall formation such as surface and upper-level
Microwave Rainfall Retrieval Algorithm
247
humidity, wind velocity field, freezing height, and aerosols using other satellite retrievals and/or objective analysis data archives. Since a set of observed brightness temperatures are not always sufficient to single out a proper rainfall profile, those quantities could be used to separate and index the databases to better constrain the retrieval. The resultant a priori databases would not only improve the algorithm performance but also provide climatological insights on the physical processes governing rainfall properties. A priori databases can be constructed for any sensor once the sensor characteristics are defined. One important exception is that the current procedure only refers to microwave window channels. Sounding channels in the 60- and 118-GHz oxygen absorption bands as well as the 183-GHz water vapor absorption channels depend upon details of the temperature or humidity profiles that are not observed directly by TRMM. These are not well represented by the above procedure and would not be well represented by the a priori database.
3.4 The retrieval Once the a priori databases of hydrometeor profiles and clear scenes, as well as their corresponding Tb’s are constructed for each sensor, a Bayesian retrieval methodology can be used to select those profiles that are consistent with the observations. Synthetic retrievals using procedures similar to those described here for a number of radiometer designs are presented in Shin and Kummerow (2003). Synthetic, in this case, meant that satellite brightness temperatures were simulated from the 3D geophysical parameter derived for the a priori database. Results from that work shows very small biases between satellites (<2%) and root mean square retrieval errors that varied with satellite instrument specifications, as one would expect. Radiometers with higher spatial resolutions and more channels tended to outperform less sophisticated sensors. The most important radiometer characteristics needed to reduce random errors appear to be the availability of low frequency channels (sensitive to liquid water emission) with good spatial resolution. Results from a simplified implementation of the algorithm described here are shown in Fig. 7. The simplified scheme uses only the brightness temperature difference (TbV – TbH) at 19 GHz in the retrieval. This use of 19-GHz channels is not necessarily a simplification as a sensor may well exist that only has these channels. The simplification was introduced by directly using the TRMM PR rainfall product plus some arbitrary but constant assumptions to compute Tbs needed in the a priori database. An additional simplification was made to use observed rather than computed brightness temperatures for the rain-free
248
C. Kummerow et al.
scenes. This is equivalent to the procedure described earlier, but only applicable to the TMI instrument as the observed Tb cannot be readily transformed to other sensors. The simple algorithm is used primarily to analyze error characteristics of the more complex retrieval algorithms. As such, it is functionally quite similar to the main algorithm presented here but not parametric.
Figure 7. (a) Retrieved conditional rainfall. (b) The probability of rain. (c) Retrieved conditional rain for probability of rain greater than 50%. (d) Uncertainty of rain [%] (see also color plate 6).
Panel (a) of Fig. 7 shows the retrieved rainfall rate. The entire area is covered by light rain surrounding smaller convective cores. The 100% rain coverage may seem physically unreasonable. What is shown, however, is the conditional rain rate. Panel (b) shows the probability of rain as derived from a lookup table correlating the observed Tb and sea surface temperature to the probability that PR saw rain in that radiometer FOV. The area of light rain can be seen to have a fairly low but nonzero probability of rain and as such may still seem unphysical. The plot is shown to illustrate the probabilistic nature of the Bayesian schemes described here. Because the radar observes small, but nonzero rain probabilities for virtually all observed 19-GHz Tbs,
Microwave Rainfall Retrieval Algorithm
249
the retrieval cannot arbitrarily use a Tb threshold for a rain/no rain discrimination without introducing errors. Panel (c) shows the same results as panel (a), but with about 50% probability of rain threshold. This may be a more conventional approach to showing rain maps. Panel (d) provides the instantaneous uncertainty (expressed in %) of the retrieved rainfall. Areas of light rain have large percentage errors, while moderate rain has significantly more signal. These results match Viltard et al. (2004) who found similar numbers with a different implementation of a Bayesian approach.
4 THE ERROR MODEL There are many sources of error that must be carefully defined if true uncertainties are to be quantified. Both Bayesian and variational retrievals principally imply an error calculation (Olson et al. 1996; Kummerow et al. 1996; Marzano and Bauer 2001; Bauer et al. 2002). Yet, the intrinsic uncertainties computed from these schemes are only one component. The Bayesian scheme has four independent sources of error. These are (1) the uncertainties introduced by imperfect or incomplete data used to construct the a priori database; (2) the uncertainty introduced by the inversion methodology; (3) the uncertainty resulting from potentially unknown changes in regional and seasonal cloud properties; and (4) the uncertainty introduced by any errors in the algorithm formulation. The Bayesian methodology deals only with the second source of error. The variational approach is less susceptible to database errors, particularly if the first-guess field is robust. Variational approach uncertainties, however, can be computed explicitly only for linear inversions. As such, error models are not simple to implement. Validation of rainfall retrievals with independent data has proven equally difficult due to problems associated with the representativeness of any data source. Within the Bayesian framework, and a priori databases generated from PR observations instead of CRMs, the three error sources not directly dealt with by the Bayesian methodology can be quantified. Errors in the construction of the a priori database can be quantified by simply changing assumptions used in the database construction over a reasonable range of values. Because the observations are used to quantify the relative occurrence of various rain realizations, small errors in the database construction are not likely to have significant impacts upon the final result. A major problem encountered when using only CRMs was the completeness and representativeness of the a priori databases. With over six years of TRMM radar data representing roughly 1010 raining pixels, these issues are no longer a source of uncertainty. Any errors introduced by smaller databases used for computational purposes can be explicitly evaluated. The second source of error was that introduced by the inversion. Bayes’ theorem explicitly accounts for these. It does not, however, account for the
250
C. Kummerow et al.
third source of error – regional and seasonal changes in cloud properties. While it seems that these potential biases could be removed by using the appropriate database as determined from the PR data, this is not possible for time periods outside the TRMM era. Conceptually, the simplest method to quantify these errors is to run the retrieval multiple times, changing only the time/space domain for which the database is constructed. Differences between retrievals in this case must be attributed to changes in the rainfall properties that the TRMM radar is able to capture but the radiometer is not. This constitutes a bias error if large space/time domains are considered. A careful analysis using six years of TRMM radar data should be able to quantify the typical magnitude of these errors for use with prior or future radiometer data. The last source of error is due to any shortcomings in the algorithm formulation. Of particular importance are any changes in parameters that are assumed constant in the forward model or specified incorrectly by the CRMs if these are used. The variational retrieval framework provides a test bed for solving this task to the degree of detail that is resolved by global model cloud/precipitation parameterizations and plane-parallel radiative transfer modeling. Ground-based data can alternatively be used.
5 CONCLUDING REMARKS The entire algorithm as described does not currently exist. If it did, we could not speak of the “next generation of microwave rainfall algorithms” as implied by the title. Nonetheless, we have attempted to outline a framework, and give specific examples of work currently underway to achieve the goals set forth by new measurement missions and new user demands. There is no doubt that perhaps some of the details are too vague, while some of the illustrative implementations are too detailed and perhaps not optimal. We do not dispute that but instead would encourage new researchers to enhance, refine or simply amend any of the procedures described here for their optimal use. In speaking about a next generation algorithm, we only provided a specific solution to illustrate that improved methodologies are possible over what is currently done without wishing to imply that the specific solution offered here is the only possible solution. Continued impetus provided by the Global Precipitation Mission related research around the world will undoubtedly help to focus and refine the concepts presented here.
6 REFERENCES Battaglia, A., C. Kummerow, D.-B. Shin, and C. Williams, 2003: Constraining microwave brightness temperatures by radar bright band observations, J. Atmos. Oceanic Technol., 20, 856–871. Bauer, P., 2001: Including a melting layer in microwave radiative transfer simulation for clouds. Atmos. Res., 57, 9–30.
Microwave Rainfall Retrieval Algorithm
251
Bauer, P. and N. C. Grody, 1995: The potential of combining SSM/I and SSM/T2 measurements to improve the identification of snowcover and precipitation. IEEE Trans. Geosci. Remote Sens., 33, 252–261. Bauer, P., P. Amayenc, C. D. Kummerow, and E. A. Smith, 2001: Over ocean rainfall retrieval from multi-sensor data of the Tropical Rainfall Measuring Mission. Part II: algorithm implementation. J. Atmos. Oceanic Technol., 18, 1838–1855. Bauer, P., J.-F. Mahfouf, S. Di Michele, F. S. Marzano, and W. S. Olson, 2002: Errors in TMI rainfall estimates over ocean for variational data assimilation. Quart. J. Roy. Meteor. Soc., 128, 2129–2144. Berg, W. and R. Chase, 1992: Determination of mean rainfall from the special sensor microwave/imager (SSM/I) using a mixed lognormal distribution. J. Atmos. Oceanic Technol., 9, 129–141. Chang, A. T. C., L. S. Chiu, C. Kummerow, and J. Meng, 1999: First results of the TRMM microwave imager (TMI) monthly oceanic rain rate: comparison with SSM/I. Geophys. Res. Lett., 26, 2379–2382. Ellison, W. J., S. J. English, K. Lamkaouchi, A. Balana, E. Obligis, G. Deblonde, T. J. Hewison, P. Bauer, G. Kelly, and L. Eymard, 2003: A comparison of new permittivity data for sea water with AMSU, SSM/I and airborne radiometers observations. J. Geophys. Res., 108, ACL 1-1–1-14. Evans, F., F. J. Turk, T. Wong, and G. Stephens, 1995: A Bayesian approach to microwave precipitation retrieval. J. Appl. Meteor., 34, 260–279. Ferraro, R. R. and G. F. Marks, 1995: The development of SSM/I rain rate retrieval algorithms using ground based radar measurements. J. Atmos. Oceanic Technol., 12, 755– 770. Grody, N. C., 1991: Classification of snow cover and precipitation using the Special Sensor Microwave/Imager (SSM/I). J. Geophys. Res., 96, 7423–7435. Haddad, Z. S., E. A. Smith, C. Kummerow, T. Iguchi, M. R. Farrar, S. L. Durden, M. Alves, and W. S. Olson, 1997. The TRMM ‘day-1’ radar/radiometer combined rain-profiling algorithm. J. Meteorol. Soc. Japan, 75, 799–808. Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain-profiling algorithm for the TRMM precipitation radar. J. Appl. Meteor., 39, 2038–2052. Kummerow, C., 1993: On the accuracy of the Eddington approximation for radiative transfer in the microwave frequencies. J. Geophys. Res., 98, 2757–2765. Kummerow, C. and L. Giglio, 1994: A passive microwave technique for estimating rainfall and vertical structure information from space, Part I: Algorithm description. J. Appl. Meteor., 33, 3–18. Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 1213–1232. Kummerow, C., Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, D.-B. Shin, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. Liebe, H., P. Rosenkranz, and G. Hufford, 1992: Atmospheric 60 GHz oxygen spectrum: New laboratory measurements and line parameters. J. Quant. Spec. Rad. Trans., 48, 629–643. Marzano, F. S. and P. Bauer, 2001: Sensitivity analysis of airborne microwave retrieval of stratiform precipitation to the melting layer parameterization. IEEE Trans. Geosci. Remote Sens., 39, 75–91. Marzano, F. S., A. Mugnai, G. Panegrossi, N. Pierdicca, E. A. Smith, and J. Turk, 1999: Bayesian estimation of precipitating cloud parameters from combined measurements of spaceborne microwave radiometer and radar. IEEE Trans. Geosci. Remote Sens., 37, 596–613
252
C. Kummerow et al.
Moreau, E., P. Bauer, and F. Chevallier, 2003: Variational retrieval of rain profiles from spaceborne passive microwave radiance observations. J. Geophys. Res., 108, ACL 11–1 — 11–18. Moreau, E., P. Lopez, P. Bauer, A. M. Tompkins, M. Janiskovà, and F. Chevallier, 2004: Variational retrieval of temperature and humidity profiles using rain-rates versus microwave brightness temperatures. Quart. J. Roy. Meteor. Soc., 130, 827–852. Mugnai, A., E. A. Smith, and G. J. Tripoli, 1993: Foundation of physical-statistical precipitation retrieval from passive microwave satellite measurements. Part II: Emission source and generalized weighting function properties of a time dependent cloud-radiation model. J. Appl. Meteor., 32, 17–39. Olson, W. S., 1989: Physical retrieval of rainfall rates over the ocean by multispectral radiometry: Application to tropical cyclones. J. Geophys. Res., 94, 2267–2280. Olson, W. S., C. D. Kummerow, G. M. Heymsfield, and L. Giglio, 1996: A method for combined passive-active microwave retrievals of cloud and precipitation profiles. J. Appl. Meteor., 38, 1763–1789. Olson, W. S., C. D. Kummerow, Y. Hong, and W.-K. Tao, 1999: Atmospheric latent heating distributions in the tropics derived from satellite passive microwave radiometer measurements. J. Appl. Meteor., 38, 633–664. Olson, W. S., P. Bauer, C. D. Kummerow, Y. Hong, and W.-K. Tao, 2001: A melting-layer model for passive/active microwave remote sensing applications. Part II: Simulations of TRMM observations. J. Appl. Meteor., 40, 1164–1179. Petty, G. W., 1994: Physical retrievals of over-ocean rain rate from multichannel microwave imagery. Part I: Theoretical characteristics of normalized polarization and scattering indices. Meteor. Atmos. Phys., 54, 79–99. Rodgers, C. D., 2000: Inverse Methods for Atmospheric Sounding: Theory and Practice. World Scientific Publishing, Co., River Edge, NJ. Shin, D.-B. and C. Kummerow, 2003: Parametric rainfall retrieval algorithms for passive microwave radiometers, J. Appl. Meteor., 42, 1480–1496. Smith, E. A., X. Xiang, A. Mugnai, and G. Tripoli, 1994: Design of an inversion-based precipitation profile retrieval algorithm using an explicit cloud model for initial guess microphysics. Meteor. Atmos. Phys., 54, 53–78. Spencer, R. W., D. W. Martin, B. B. Hinton, and J. A. Weinman, 1983: Satellite microwave radiances correlated with radar rain rates over land. Nature, 304, 141–143. Viltard, N., C. Burlaud, and C. Kummerow, 2004: Rain retrieval from TMI brightness temperature measurements using a PR-based database. J. Appl. Meteor. Climatol., 45, 455–466. Wentz, F. J., 1997: A well-calibrated ocean algorithm for Special Sensor Microwave/ Imager. J. Geophys. Res., 102 (C4), 8703–8718. Wilheit, T. T., A. T. C. Chang, and L. S. Chiu, 1991: Retrieval of monthly rainfall indices from microwave radiometric measurement using probability distribution functions. J. Atmos. Oceanic Technol., 8, 118–136.
Section 4 Blended Techniques
20 THE UNIVERSITY OF BIRMINGHAM GLOBAL RAINFALL ALGORITHMS Chris Kidd1, Francisco J. Tapiador1, Victoria Sanderson1, and Dominic Kniveton2 1
The University of Birmingham, Edgbaston, UK The University of Sussex, Falmer, Brighton, UK
2
Abstract
This paper outlines the development of a combined infrared/passive microwave algorithm designed to produce high-resolution short-duration rainfall estimates at a quasi-global scale. Data from infrared observations are calibrated with rainfall estimates generated by a frequency difference algorithm through a cumulative histogram matching approach. The calibration used for the conversion is initially generated at a daily 1 degree × 1 degree scale before being accumulated into a weighted pentad 5 degree × 5 degree database. Examples of the estimates generated from this technique are shown for global and regional case studies: although the statistical results are very similar to those of the GOES Precipitation Index. The delineation of rain area is improved with accurate estimation of the occurrence of rainfall.
Keywords
Global rainfall, cumulative histogram matching
1 INTRODUCTION There is an increasing demand for rainfall information throughout a range of temporal and spatial scales. Conventional observations, from rain gauges or radar are limited, in accuracy and extent, to specific regions of the globe. The use of satellite observations of precipitation on a global basis is therefore crucial in providing a complete picture of the distribution and occurrence of global rainfall. This is particularly true over data sparse regions such as over the oceans or over large unpopulated regions of the landmasses. Large spatial or long temporal scale rainfall information tends to be reliable due to the space/time averaging that takes place. However, shortterm rainfall retrievals are fundamentally more difficult: rainfall is a highly 255 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 255–267. © 2007 Springer.
256
C. Kidd et al.
variable parameter in time and space. This characteristic is compounded by the availability of suitable satellite observations. The use of infrared data provided by geostationary satellites provides good temporal sampling, typically 15–30 min, but since the observations are of the cloud tops there is only an indirect association between the measured cloud top temperature and the rainfall at the surface. Observations using passive microwave measurements have shown a much more direct relationship with the surface rainfall: passive microwave frequencies are opaque to clouds, with the main source of attenuation being the precipitation-sized particles themselves. Unfortunately, the temporal sampling from such sensors is at present very limited, with maybe no more than six observations possible each day. Following intercomparison projects such as the Algorithm Intercomparison Programme (Ebert et al. 1996) and the Precipitation Intercomparison Project (Adler et al. 2001) it was noted that the infrared-based algorithms performed best over climatological scales, whilst for instantaneous estimates the passive microwave based estimates were best. It was therefore a logical progression that a combination of the two methods of observation ought to yield better results than either one alone. Development of algorithms utilising the information from the infrared and passive microwave built upon the success of the GOES Precipitation Index (GPI; see Arkin and Meisner 1987). This was accomplished in two ways. Passive microwave estimates could be used to calibrate monthly largescale rainfall totals, which could then be applied to smaller shorter-scale rainfall retrievals from the infrared alone (e.g., Adler et al. 1993). The other method adjusted the parameters of the GPI algorithm: the GPI assumes that any cloud less than 235 K produces rainfall at 3 mm h–1. These two values could be altered to best match the longer-term rainfall retrievals generated by the passive microwave observations (e.g., Xu et al. 1999). Although these techniques show a good degree of success it is necessary to have a relatively long-term infrared and passive microwave data set in order to achieve a good calibration.
2 ALGORITHM DESIGN The algorithm presented here relies upon the regional calibration of colocated infrared and passive microwave observations and is based upon initial work by Kidd (1999) over the western Pacific. This work found that the calibration thresholds and rain rates varied significantly over relatively short time scales (<<1 month) and therefore this must be reflected in the calibration scheme. It was also recognised that the number of valid (i.e., raining) data points needed to be sufficient to be representative of the raining system being observed. This latter issue imposes a constraint on the time and spatial scale of the calibration data set. Turk et al. (2000) uses an instantaneous calibration domain over a 5° × 5° window: this therefore
The University of Birmingham Global Rainfall Algorithms
257
produces a good temporal calibration, but at the expense of the regional scale. The technique described here produces a daily calibration at a 1° × 1° scale. Both techniques use some of smoothing/averaging of the calibration to ensure spatial and temporal gradation between different calibration regimes. Interestingly, the number of data points available to either technique remains very similar.
2.1
Satellite data sets
Global infrared data, described by Janowiak et al. (2001), are obtained from the NOAA Climate Prediction Center (CPC) in near real-time. This data is a composite from five geostationary satellites (Meteosat-5/7 and GOESE/W/9) covering a global region from 60° N to 60° S. The data are remapped to a common rectilinear grid with a spatial resolution of approximately 4 km and a temporal resolution of 30 min. Inter-satellite calibration and parallax corrections are applied at CPC before distribution. These infrared data are then resampled, by averaging a 3 × 3 region of pixels, to a resolution of 1/10 degree (~11 km) similar to the highest resolution of the Special Sensor Microwave/Imager (SSM/I) sensor. Passive microwave data collected by the SSM/I sensor, onboard the DMSP series of satellites, are obtained from the NASA Marshall Space Flight Center. Three sensors are currently available, namely the F13, F14 and F15, with morning descending and evening ascending orbital paths providing up to six overpasses each day. The data comprises of brightness temperature measurements in seven channels at four frequencies, vertical and horizontal polarizations at 19.35, 37.0 and 85.5 GHz and vertical only at 22.235 GHz.
2.2
Data preparation
Before the rainfall algorithm is applied any ambiguous surface signal needs to be eliminated. Background surfaces covered by snow and ice (an example of snow/ice cover is shown in Fig. 1) are likely to have a similar signature to that of precipitation and therefore must be identified and masked out. This step is critical: most snow/ice algorithms identify certain snow/ice cover, whereas for the combination process all possible snow/ ice must be identified to avoid contamination of the final product. The methods of Ferraro et al. (1996) and Grody and Basist (1996) are applied (for ice and snow respectively) to all the passive microwave overpasses for the retrieval day: a snow/ice background is flagged if the majority of SSM/I observations within a grid cell identify snow/ice. The rainfall algorithm used for the retrieval of rainfall is the frequency difference algorithm based upon that used in the AIP-3 and PIP-3 studies (see Ebert et al. 1996). The algorithm utilises the 19 GHz and 85 GHz channels to identify a scattering signal generated by precipitation. Over water the vertical channel is used while over land, to reduce surface
C. Kidd et al.
258
artefacts, the horizontal channels are used. The algorithm has been calibrated through comparison with surface data from the UK radar network and by the Tropical Rainfall Measurement Mission (TRMM) Precipitation Radar (PR).
Figure 1. Observations of snow or ice from the DMSP SSM/I sensor for 23 January 2004. Dark areas represent 100% of SSM/I pixels identified as snow/ice, while lighter grey areas indicate regions of marginal snow/ice cover.
2.3
Combination strategy
The processing steps are outlined in a flow diagram (see Fig. 2). Once the two primary data sets have been pre-processed, the infrared data is subsampled to 1/10 degree resolution and the passive microwave used to generate the rainfall retrieval, each coincident 1/10 pixel are added to a database. The database contains the number of observations of cloud-top temperatures and rainfall rates for each 1 degree grid cell over a one-day period. Therefore, assuming perfect passive microwave coverage (6 overpasses a day) and infrared observations, a total of 600 points are possible (Fig. 2, step A). The daily databases are then amalgamated to ensure a more continuous and contiguous coverage (Fig. 2, step B). Daily databases for the preceding four days, together with the current day, are accumulated using a time-weighted function. Spatial accumulation is also performed using a distance weighted function over ±2 degrees from the central grid cell. These procedures ensure grid-cell to grid-cell as well as day-to-day continuity in the rainfall estimates. Once the accumulated database has been generated a cumulative histogram matching technique is applied, relating the cold cloud-top temperature pixels with the heaviest rainfall retrievals. This procedure ensures that the statistical distribution of rainfall intensities identified by the passive microwave data is replicated in the infrared-derived rainfall estimates (Fig. 2, step C). The matching technique generates a cloud-top temperature to rain rate look-uptable which can then be used to convert the global infrared data set into passive microwave calibrated infrared rainfall retrievals. The combined
The University of Birmingham Global Rainfall Algorithms
259
rainfall product has an output resolution of 1/10 degree every 30 min, allowing the user to resample to their requirements. By dividing the processing into three distinct phases data supply problems can be reduced: if the passive microwave data is not available for one particular day the calibration proceeds without the data for that day. Although the rainfall information for that particular day will be missing, a product can still be generated from the other four days in the calibration.
Figure 2. Flow diagram of the University of Birmingham IR-SSM/I combined technique.
3 RESULTS 3.1
Global scale estimates
The results of the combined technique applied at the daily scale are shown in Fig. 3, together with the GPI rainfall estimate and that retrieved from the DMSP SSM/I data alone. It can be seen that the GPI (top) tends to estimate the largest extent of rainfall while estimating the lowest rainfall amount: the GPI is designed to be used primarily at the monthly/coarse scale where temporal and spatial averaging makes this less problematic. The passive microwave derived rainfall estimate (middle) tends to be generate the smallest extent of rain area, whilst producing relatively high rainfall amounts. The combined product (bottom) shows that the rain area tends to be more realistic: the rainfall
260
C. Kidd et al.
amounts derived from the combined algorithms are between that identified by the GPI and the SSM/I products. It can also be noted that the combined algorithm provides a greater level of detail than that possible with the GPI technique. The combined algorithm produces quasi-global (60°N to 60°S) rainfall estimates routinely on a 1/10 degree 30-min basis: these estimates may be viewed at http://www.isac.cnr.it/~eurainsat, together with daily GPI and SSM/I-only products.
Figure 3. Comparison of daily rainfall totals derived from the GPI (top), DMSP SSM/I frequency difference algorithm (middle) and, the IR-SSM/I combined algorithm (bottom).
The University of Birmingham Global Rainfall Algorithms
261
Figure 4. Statistics for daily 0.1 degree rainfall estimates for the combined algorithm (PMIRo) compared with those of the UAGPI, GPI and SSM/I estimates. The vertical scale is the number of comparisons made (after Kidd et al. 2003, courtesy of Amer. Meteor. Soc.).
3.2
Regional scale results
Comparison with surface data was performed over Niger for June– September 2001. Statistical comparison between the surface daily gauge data and satellite estimates are shown in Fig. 4 which shows the statistical occurrence of correlation, skill score and ratio for the SSM/I-only, GPI, Universally Adjusted GPI (UAGPI; Xu et al. 1999) and the current combined algorithm (PM-IRo). In terms of the correlation, the infrared and combined techniques generally do equally well, with the SSM/I-only product showing poorer results. This result is replicated in the skill score statistic, where the combined algorithms show slightly better skill than the infraredonly GPI rainfall estimates. Finally, the ratio statistics show that the SSM/I product seriously underestimates the rainfall occurrence (~30%), primarily due to poor temporal sampling, while the GPI greatly overestimates the rainfall occurrence (~200%+). The combined algorithm results in a much
262
C. Kidd et al.
better rainfall identification, although the combined algorithms do show a slight bias of about +10% in the occurrence of rainfall. Real-time inter-comparisons of daily estimates against surface gauge data over Australia are shown in Fig. 5, and can be viewed at http://www.bom.gov.au/bmrc/wefor/staff/eee/SatRainVal/sat_val_aus.html. Inter-comparisons over the USA can be found at http://www.cpc.ncep.noaa.gov/products/janowiak/ us_web.html: a European inter-comparison should be available shortly.
Figure 5. Example of the University of Birmingham merged satellite rainfall estimates over Australia (courtesy of Dr. Beth Ebert, Bureau of Meteorology, Melbourne).
4 DISCUSSION An important finding is that statistically there was little improvement using the combination technique over existing infrared-only techniques, even when these infrared algorithms are used at a finer spatial and temporal scales than they were originally designed for. A possible explanation is that the amount of improvements in any combined algorithm is very much limited by the indirectness of the infrared observations: no matter how good the calibration is, the cloud tops do not truly represent the position, extent and intensity of the rainfall fields. One approach is to dismiss the ability of the infrared to “observe” rainfall altogether and rely upon the passive microwave data alone to provide the quantitative rainfall information. However, as mentioned above, the passive microwave data does not have an adequate level of temporal sampling. Although the infrared data may not be able to provide rainfall information it can provide information on cloud-top motion and development and hence can be used to aid the rainfall retrieval process.
The University of Birmingham Global Rainfall Algorithms
263
This approach has been utilised by the CMORPH technique described by Joyce et al. (2004). This method utilises the retrievals of rainfall from a range of passive microwave sensors including the SSM/I, TMI & AMSU-A/B. The rainfall estimates are generated at a resolution of 8km with wind motion fields generated from 5 degree × 5 degree infrared cloud-top information are used to move the estimates. Although the spatial resolution of the motion fields are somewhat coarse, they do represent the general atmospheric motion in that region. With the advent of new sensors, such as the Meteosat Second Generation (MSG) Spinning Enhanced Visible and Infrared Imager (SEVIRI) instrument, better temporal and spectral resolutions allow new rainfall estimation techniques to be developed (see Levizzani et al. 2000). The technique described below takes advantage of the MSG SEVIRI 4 km spatial resolution and 15min observation cycle to generate fine scale cloud-top motion to advect the passive microwave rainfall estimates between successive overpasses. Passive microwave rainfall estimates have only been derived from the SSM/I in the present study, although other sensors could be included in the technique if required. The initial step is to generate the cloud motion fields from the MSG data using a relatively simple moving correlation matrix (see Fig. 6). Rainfall estimates, generated from the passive microwave observations are then advected along the infrared derived wind fields until the next passive microwave overpass occurs. Ideally, the advected rainfall would be dynamically adjusted so that the start and end values matched those of each overpass, but in the current configuration this is deemed to too time consuming. This technique has been applied to one day of MSG and SSM/I comparison made with surface radar data over the European region. Figure 7 shows the original SSM/I rainfall field for 0745 UTC and the advected rainfall estimate for 0945 UTC: movements in the areas of rainfall can be seen over Eire and over Denmark. The results provide an interesting insight into the differences between the infrared and passive microwave observations of rainfall. Figure 8 shows the correlation of an infrared-only technique (optimised against the radar), the SSM/I-alone retrievals, and the advected rainfall estimates against co-located radar data every 15 min at 5 km resolution. The SSM/I-only retrievals show that correlations with the co-temporal radar data generates values of ~0.4, falling away to a background correlation value of about 0.1– 0.15 about ±45 min either side. Despite the infrared data being optimised against the radar data, correlations of the infrared data rarely exceed 0.2 at any time, yet alone at the times at which they coincide with the radar data. Thus, if statistics instantaneous satellite overpasses are generated, the passive microwave estimates will outperform the infrared, whilst if longer-term estimates are generated the infrared estimates will outperform the passive microwave: the balance depends upon the number of passive microwave observations available.
264
C. Kidd et al.
Figure 6. Wind fields generated from the Meteosat Second Generation SEVIRI sensor for 0745-0945 UTC on 12 May 2003.
Figure 7. Results of the advection technique on 12 May 2003: original passive microwave estimates for 0745 UTC (left) and advected estimates for 0945 UTC (right, values in mm h–1).
The University of Birmingham Global Rainfall Algorithms
265
Figure 8. Time-correlation statistics for rainfall retrievals from the SSM/I, IR and advection techniques on 12 May 2003. Note that the maximum correlations of the SSM/I correspond with the satellite overpass times.
266
C. Kidd et al.
The goal of the advection technique is to prolong the high correlations as long as possible to improve the final rainfall algorithm performance. The time correlation plot for the advected technique is also shown in Fig. 8. As expected the technique does not improve upon the correlations at the time of the SSM/I observations but does slightly improve the correlations in the period after the passive microwave observations. Although the improvement is small, it should be noted that this comparison is made at 15 min, 5 km resolution, mid-latitude rainfall regime with relatively small areas of precipitation.
5 CONCLUSIONS The use of multiple satellite observations to derive rainfall estimates hold the promise of improved results. Although some improvements can be made under certain conditions, the amount of improvement has been rather limited. Some of this can be attributed to the inability to gain any more information on rainfall from the infrared measurements of the cloud tops. Another problem is that the addition of multiple sensors can introduce more noise into the retrievals than if using a single sensor alone. However, it should be noted that the combined techniques are being applied at much finer spatial and temporal scale than previously used. The future development of the GPM mission will improve rainfall estimates substantially by, not only providing more passive microwave observations, but also through a better understanding of the precipitation processes. Acknowledgements: This research has been funded by the EURAINSAT project, a shared-cost project (EVG1-2000-00030) co-funded by the Research DG of the European Commission with the RTD activities of a generic nature of the Environment and Sustainable Development subprogramme (5th Framework Programme). Global infrared data is courtesy of John Janowiak (CPC) and passive microwave data courtesy of the GHCC, NASA/MSFC.
6 REFERENCES Adler, R. F., A. J. Negri, P. R. Keehn, and I. M. Hakkarinen, 1993: Estimation of monthly rainfall over Japan and surrounding waters from a combination of low-orbit microwave and geosynchronous IR data. J. Appl. Meteor., 32, 335–356. Adler, R. F., C. Kidd, G. Petty, M. Morrisey, and M. H.Goodman, 2001: Intercomparison of global precipitation products: The Third Precipitation Intercomparison Project (PIP-3). Bull. Amer. Meteor. Soc., 82, 1377–1396. Arkin, P. A. and B. N. Meisner, 1987: The relationship between large scale convective rainfall and cold cloud over the Western Hemisphere during 1982–84. Mon. Wea. Rev., 115, 51–74.
The University of Birmingham Global Rainfall Algorithms
267
Ebert, E. E., M. J. Manton, P. A. Arkin, R. J. Allam, G. E. Holpin, and A. Gruber, 1996: Results from the GPCP Algorithm Intercomparison Programme. Bull. Amer. Meteor. Soc., 77, 2875–2887. Ferraro, R. R., F. Weng, N. Grody, and A. Basist, 1996: An eight-year (1987–1994) time series of rainfall, clouds, water vapor, snow cover and sea ice derived from SSM/I measurements. Bull. Amer. Meteor. Soc., 77, 891–905. Grody, N. and A. Basist, 1996: Global identification of Snowcover Using SSM/I Measurements. IEEE Trans. Geosci. Remote Sens., 34(1), 237–249. Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2001: A real-time global half-hourly pixelresolution infrared dataset and its application. Bull. Amer. Meteor. Soc., 82, 205–217 Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2003: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at 8 km, ½ hourly resolution. J. Hydrometeor., 5, 487–503. Kidd, C., 1999: Results of an infrared/passive microwave rainfall estimation technique. Proc. Remote Sensing Society, 8–10 September 1999, Cardiff, Wales, 685–689. Kidd, C., D. R. Kniveton, M. C. Todd, and T. J. Bellerby, 2003: Satellite rainfall estimation using combined passive microwave and infrared algorithms. J. Hydrometeor., 4, 1088– 1104. Levizzani, V., P. P.Alberoni, P. Bauer, L. Bottai, A. Buzzi, E. Cattani, M. Cervino, P. Ciotti, M. J. Costa, S. Dietrich, B. Gozzini, A. Khain, C. Kidd, F. S. Marzano, F. Meneguzzo, S. Migliorini, A. Mugnai, F. Porcù, F. Prodi, R. Rizzi, D. Rosenfeld, L. Schanz, E. A. Smith, F. Tampieri, F. Torricella, J. F. Turk, G. A. Vicente, and G. Zipoli, 2000: Use of the MSG SEVIRI Channels in a Combined SSM/I, TRMM, and Geostationary IR Method for Rapid Updates of Rainfall. Proc. 1st MSG RAO Workshop, Bologna. 17–19 May 2000. ESA SP-452, 63–66. Turk, F. J., J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000: Combining SSM/I, TRMM and infrared geostationary satellite data in a near-real time fashion for rapid precipitation updates: advantages and limitations. Proc. 2000 EUMETSAT Meteorological Satellite Data Users’ Conf., Bologna, Italy, 29 May-2 June, 452–459. Xu, L., X., Gao, S. Sorooshian, P. A. Arkin, and B. Imam, 1999: A microwave infrared threshold technique to improve the GOES Precipitation Index. J. Appl. Meteor., 38, 569– 579.
21 MULTIVARIATE PROBABILITY MATCHING FOR MICROWAVE INFRARED COMBINED RAINFALL ALGORITHM (MICRA) Frank S. Marzano1, Domenico Cimini1, and F. Joseph Turk2 1
Centre of Excellence CETEMPS – University of L’Aquila, L’Aquila, Italy Naval Research Laboratory, Monterey, CA, USA
2
Abstract
The proposed Microwave Infrared Combined Rainfall Algorithm (MICRA) is a statistical integration method using the satellite microwavebased rain rate estimates, assumed to be accurate enough, to calibrate space-borne infrared measurements on sufficiently limited subregions and time windows. The proposed methodology is focused on new statistical technique, namely the multivariate probability matching (MPM), aimed at employing both average and texture information as well as multispectral data. The MPM is formulated and analyzed in terms of relative detection and estimation accuracy. Rainfall retrieval is pursued at the space-time scale of typical geostationary observations, that is at a spatial resolution of few kilometers and a repetition period of few tens of minutes. In order to demonstrate the potentiality of MICRA, case studies are also discussed.
Keywords
Rainfall, microwave radiometry, infrared radiometry, satellite sensors, inversion algorithms
1 INTRODUCTION The accurate retrieval of surface rain rate (R) from space-borne remotesensing systems on a global scale with high temporal and spatial resolutions is one of the major goals of current scientific research (Smith et al. 1998; Levizzani et al. 2001; Marzano et al. 2002). Satellite-based methodologies can offer several advantages with respect to ground-based techniques. The latter, such as those using rain gauges and radars, generally suffer from spatial coverage problems as they provide incomplete coverage on a global scale, particularly over the oceans where such instruments are sparse or not existing. 269 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 269–279. © 2007 Springer.
270
F. S. Marzano et al.
The problem of using satellite remote-sensing data appears to be fairly complicated since presently there is not a single space-borne platform which can carry all the suitable instruments to observe all the above mentioned properties to the rainfall product. From a meteorological point of view, visible (VIS) and infrared (IR) radiometers can give information on cloud top layers (e.g., Richards et al. 1981; Adler et al. 1988; Ba and Gruber 2001). On the other hand, microwave (MW) radiometers can detect cloud structure and rain rate (e.g., Panegrossi et al. 1998; Marzano et al. 1999; Pulvirenti et al. 2002). Regarding platforms, Geosynchronous Earth Orbit (GEO) satellites can ensure a coverage with a high temporal sampling (order of half an hour), while Low Earth Orbit (LEO) satellites have the advantage to carry microwave sensors without loosing too much in spatial resolution (order of kilometer to tens of kilometers). The major drawback of LEO’s is the low temporal sampling, only twice a day in a given place at mid latitude. Therefore, LEO-MW and GEO-IR radiometry are clearly complementary for monitoring the Earth’s atmosphere and a highly variable phenomenon such as precipitation. In order to exploit this synergy, one can resort to approaches to combine IR measurements and MW-based estimates on a statistical basis (Adler et al. 1993; Kummerow and Giglio 1995; Levizzani et al. 1996; Vicente et al. 1998; Turk et al. 1999; Xu et al. 1999; Todd et al. 2000; Marzano et al. 2001; Miller et al. 2001). By properly choosing a space-time resolution, the ergodicity of the rain process and satellite observations can be invoked. The inversion algorithm, that is retrieving rain rate from IR data, can be then derived by using statistical regression or probability matching of the involved variables (or corresponding statistical moments). Even though less physical, the statistical matching exhibits several peculiar features that can be easily exploited for an operational global-scale approach. Indeed, artificial neural network can be conveniently applied to the same problem dealing with empirically trained algorithms showing comparable performances (e.g., Hsu et al. 1997; Bellerby et al. 2000). In this work, a systematic analysis of statistical integration methods is carried out in order to use MW-based rain-rate estimates to calibrate IR measurements. Here we limit our interest to the use of microwave radiometric data, derived from the Special Sensor Microwave Imager (SSM/I) aboard the Defense Meteorological Special Program (DMSP) satellites coupled with data from the Visible Infrared Spinning Scan Radiometer (VISSR) aboard MeteoSat satellites, even though the method could be easily generalized to different data types. The proposed MW-IR Combined Rainfall Algorithm (MICRA) is based on the statistical integration of collocated GEO-IR and LEO-MW data, accomplished on a global scale using an ensemble of subregions, partially overlapped. This time-space segmentation permits also to analyze stationarity and homogeneity statistical properties of the precipitation random process.
Multivariate Probability Matching for MICRA
271
A new technique, named multivariate probability matching (MPM), is investigated in terms of relative estimate accuracy, algorithm parameter sensitivity, cloud classification impact and computing efficiency. Similar techniques, available in literature, are also considered for comparison. As an application, a case study is finally discussed in order to demonstrate the potentiality of mapping rainfall using the proposed statistical integration technique.
2 COMBINED RAINFALL ALGORITHM The general idea behind the MICRA considered statistical integration technique is to combine the appealing spatial and temporal sampling of IR sensors, mounted on geostationary platforms, with the higher accuracy of passive MW methods for of rain-rate retrieval. The statistical integration techniques are applied within a procedure which is supposed to run continuously on global scale. This procedure is based on a background process and a foreground process. The block diagram of MICRA is schematically sketched in Fig. 1.
Background process (BP)
TB
MW Inversion of Rain (INV)
TIR
SpaceSpace-Time Collocation (COL)
R
PrePre-processing (PRE)
- Classification - Temporal memory - Thresholding
Foreground process (FP)
IRIR-R Statistical Integration (INT) TIR
IR Retrieval of Rain (RET)
ˆ R
Figure 1. Conceptual block diagram of MICRA.
The background process consists first of estimating the surface rain rate from available LEO-MW measurements by means of either empirical retrieval algorithms or inversion schemes based on parametric cloud radiative models (inversion step). This means that we are considering an estimator F-1 which enables the inversion of a set of TBs at frequency νn and polarization pm, generally spanning from 10 GHz to 150 GHz and over two linear orthogonal
F. S. Marzano et al.
272
polarizations for rainfall applications, to provide a rain-rate product spatially integrated within the nominal area A (e.g., Ferraro 1997; Smith et al. 1998). Notice that the field-of-views of satellite TBs are frequency dependent (ranging from 60 km down to a few kilometers) and the estimation accuracy over land may be largely worse than over ocean. On the other hand, as already said, the estimator can include any type of rainfall measurement both from space and from ground. The second step of the background process pursues the combination of LEO-MW sensor data with data coming from GEO-IR sensor in space and time on a global scale (collocation step). The first step of the background process is to locate temporally the GEO-IR data within the past few tens of minutes of the LEO-MW data time and to remap into the geographic coordinates both GEO-IR and LEO-MW available observations. Note that, since spatial resolution of MW data is generally worse than IR ones, a MW field-of-view of nominal area A generally includes more than one IR pixel. For DMSP-SSM/I products, for instance, the nominal resolution of 25 km corresponds at mid-latitudes at about 5 × 5 pixels of MeteoSat-VISSR IR channel. Thus, for a given MW-based rain rate R, attributed to a nominal area A, we can compute several spatial moment of IR brightness temperature TIR: (i) average value Ta within A; (ii) minimum value Tm within A; (iii) standard deviation σT within A. If NA are the IR pixels within the nominal area A, a numerical evaluation of previous quantities is given by: Ta ≅
1 NA
NA
∑ l =1
TIRl ; Tm = Min A [TIR ]; σ T =
NA
∑
(TIRl − Ta )2 (N A − 1)
(1)
l =1
where MinA is the minimum operator within A. The inclusion of Tm (or, equivalently the standard deviation σT), can give information of spatial texture of IR field within the nominal area A. As a result of the background process, a data set is generated, containing the per-pixel rain rate retrieved from LEO-MW data, the co-located GEO-IR brightness temperature and the pixel geolocation. This process is continuously ongoing, since new LEO-MW and GEO-IR data are continuously ingested on a global scale depending on available satellite platforms. A preprocessing stage, consisting in a cloud type classification and windowing, is accomplished after each background process as illustrated in Fig. 1. A foreground process is started to derive the R-TIR inverse relationship once the data set has been updated. The entire globe is divided in subregions S that are equally sized (α degrees × α degrees) and spaced (β degrees). The parameter α is generally chosen larger than β in order to assure a smooth transition between adjacent subregions. The IR retrieval relationships are updated every time a new set of combined data have been added to the data
Multivariate Probability Matching for MICRA
273
set relative to that subregion, and are derived using data archived in a time window of several hours (integration step). As a matter of fact, to assure that only the most recent rain history is captured and to guarantee a statistical significance of the training set, the R-TIR inverse relationship for a given subregion is derived using only the most recent combined data. The last step is represented by the prediction of the surface rain rate from IR measurements in a given subregion by applying the derived R-TIR algorithm (retrieval step).
2.1 Multivariate Probability Matching (MPM) The probability matching criterion was first introduced in radar meteorology to derive rain-rate estimates from radar reflectivity measurements (Calheiros and Zawadzki 1987; Atlas et al. 1990). Its extension to satellite rain measurements is almost straightforward, even though some deductions ought to be derived with caution. The basic idea behind the probability matching technique is to derive the inverse relationship between measured IR data and rain rate using the corresponding histograms of the occurrences of R and the average value Ta. If pR and pTa are the probability density functions (PDF) of R and Ta, respectively, we can translate this concept by the following equality: p R ( R )dR = pTa (Ta )dTa
(2)
There is a theoretical and experimental evidence that the correlation c between R and Ta is basically negative, i.e., c(R,Ta)<0. This indicates that higher rain rates are associated to lower IR brightness temperatures due to the increasing cloud opacity and top height. Moreover, both R and Ta are positive defined. From (2) the Univariate Probability Matching (UPM) can be stated as: Rˆ
Ta 0
Ta 0
~ Ta
R0
Ta
0
0
∫ p R ( R)dR = ~∫ pTa (Ta )dTa = ∫ pTa (Ta )dTa − ∫ pTa (Ta )dTa
(3)
where Ta0 is the threshold value corresponding to the minimum detectable R. In order to extend UPM to the bivariate case, we can consider the minimum value Tm within the nominal area A as the second random variable. Thus we need to consider the joint PDF pTaTm of Ta and Tm so that (2) becomes: p R ( R )dR = pTaTm (Ta , Tm )d Ta dTm
(4)
F. S. Marzano et al.
274
At this stage, it is opportune to set ta=Ta0−Ta. Since it results that c(R,Tm)<0, we can also pose tm = Tm0 − Tm, noting that c(R,tm)>0. Coupled with Ta0, the threshold value Tm0 is defined by: ~ ~ Ta Tm
R0
∫ pR ( R)dR = ∫ ∫ pTaTm (Ta ,Tm )dTa dTm
(5)
Ta 0 Tm0
0
From (4) and (5), the UPM can be generalized to the bivariate or MPM by: ~ ta ~ tm
Rˆ
∫p
R0
R
~ ~ ( R)dR =PR ( R ≤ Rˆ ) = ∫ ∫ ptatm (t a , t m )dt a dt m = Ptatm (t a ≤ ta , t m ≤ tm )
(6)
0 0
where Ptatm is the joint cumulative distribution function (CDF) of ta and tm. Details are given by Marzano et al. (2004).
3 TESTS AND APPLICATIONS The test and the optimization of the proposed integration technique has been carried out considering only the nearly coincident MW and IR passages over a given subregion. Starting from IR Ta’s within each subregion and with a 24-h backward window, we have estimated the rain rate using both MPM and UPM methods. This R estimate has been then compared with the MWderived rain rates, available at the successive MW overpass and used as a reference (“ground truth”). This evaluation strategy has been set up basically to assess the capability of each integration technique to “calibrate” IR measurements in terms of R. To evaluate the algorithms scores, two sets of indexes have been considered. The first set of six parameters reported in Table 1 allows to evaluate the rain detection capability of each method, i.e., to discriminate between the raining and non-raining pixels (Marzano et al. 2004). Table 1. Rain detection indexes.
PODNR= FAR= ISE=
Nn Nr + Nn
Nr Rr + Nr
Rn − Nr Rn + Nr
PODR=
Rr Rn + Rr
Rr Nr − Rr + Nr Nr + Nn Rr CSI= Rr + Rn + Nr
HKI=
Multivariate Probability Matching for MICRA
275
Table 2. Rain estimation indexes.
FSE=
⎛ 1 ⎜ ⎝N
N
∑ i =1
1 N
⎞ ( ε ri ) 2 ⎟ ⎠
2
NBIAS=
N
∑r i =1
FMR= RMW − ε r RMW
1
i
FVR=
1 N 1 N
N
∑
i =1 N
∑
i =1
ε
ri
ri
σ 2 RMW − σ 2ε r σ 2 RMW
In Table 1, the first letter stands for MW-derived R (“observations”), while the second for the one estimated from IR. Thus, Rr is for rain observations correctly classified, Nn is for no-rain observations correctly classified, Rn is for rain observations incorrectly classified and Nr is for no-rain observations incorrectly classified. The second set of parameters, reported in Table 2, allows to evaluate the rain estimation capability of each method. In Table 2, N is the total number of IR-MW observations, R MW the mean value of MW-based rain rates, ε r the mean value (bias) of the rain rates estimate errors, while σ2RMW and σ2εr are the standard deviations of MW-based R’s and of the rain rates estimate errors, respectively.
Figure 2. Cumulated rainfall (mm h–1) on November 1–17, 1999 within Meteosat sector (see also color plate 11).
F. S. Marzano et al.
276
Results in terms of previously mentioned indexes are reported in Table 3, where the optimal value of each index is also indicated. The results are also compared with those obtained from MIRRA technique (Miller et al. 2001). Table 3. Rain detection and estimation indexes for the period November 1–17, 1999. Index PODNR
MPM 0.96
MIRRA 0.92
UPM 0.96
Optimal 1
PODR
0.37
0.35
0.15
1
FAR
0.63
0.86
0.64
0
CSI
0.22
0.10
0.12
1
HKI
0.32
0.12
0.33
1
ISE
0.01
–0.15
0.49
0
NEB
0.10
0.12
0.11
0
FMR
0.89
0.89
0.89
1
FVR
0.84
0.86
0.82
1
FSE
0.77
0.71
0.80
0
Several studies have been utilized to optimize and to evaluate the statistical integration technique. Satellite data have been collected from SSM/I-DMSP and from VISSR-MeteoSat. As an example, we show the data analysis relative to the period November 1–17, 1999. During this period strong cyclonic perturbations have hit the Mediterranean area. The whole field of view of MeteoSat-7 is shown in Fig. 2 in terms of microwave-derived accumulated rainfall within that period. From the analysis in Table 3, it emerges that MPM demonstrates a fairly good ability to collect the raining pixels and a very good capability to collect the non-raining labeled pixels. With respect to estimate accuracy, MPM shows a significant improvement with respect to both UPM and MIRRA techniques.
4 CONCLUSIONS The MICRA procedure is based on the collocation of GEO-IR and LEOMW data, accomplished on a global scale using an ensemble of subregions, partially overlapped. A new techniques, named multivariate probability matching (MPM), has been investigated in terms of relative estimate accuracy, algorithm parameter sensitivity, cloud classification impact and computing efficiency. As an application, some case studies have been discussed, in order to demonstrate the potentiality of monitoring rainfall attenuation and precipitation using the proposed MICRA techniques.
Multivariate Probability Matching for MICRA
277
We have shown that the MPM method shows better performances for rain detection and slight lower scores for rain estimation. In many respects, MPM shows better results than univariate probability matching and linear regression methods. Even though we limited our analysis to DMSP-SSM/I, and MeteoSat-VISSR data, the proposed techniques can be easily generalized to rain-rate estimates, derived from other space-borne sensors and groundbased instruments, as well as to multispectral VIS-IR channels. We note that the MICRA statistical integration techniques can offer some advantages with respect to artificial neural network approaches due to their higher efficiency within the training phase that is essential when designing an adaptive rapidupdating procedure on a global scale. Future work will be devoted to embed these MICRA statistical techniques within current operational frameworks in order to verify the expected improvements and validate them on a long-term and global scale basis by comparing with conventional ground data. Besides, the MW inversion step can be refined by using climatologically tuned empirical algorithms and, for some critical regions, physically based MW algorithm able to predict the precipitation spatial structure. Indeed, further retrieval constraints, derived from available geophysical fields (such local wind flow and orography) and meteorological dynamical condition (such as rain cloud advection), can help to improve the estimate of surface rain rate. The scenario of MW radiometers aboard a multi-satellite constellation raises the issue about the consistency of rain-rate products derived from instruments with different specifications (e.g., channel frequencies, field-ofviews, radiometric accuracies). A simple way to apply the proposed MICRA technique could be the choice of a “reference” radiometer whose rain products can be used to “calibrate” all the others adopting the same probability matching concepts illustrated here. Finally, the recent launch of Meteosat Second Generation (MSG) constitutes a unique opportunity to exploit multivariate (multispectral) approaches to satellite rainfall retrieval. Acknowledgements: This work has been partially supported by the Italian Space Agency (ASI) and by the EURAINSAT project (contract n. EVG1 2000-00030) funded by the European Commission. The first author thanks the Naval Research Laboratory, Monterey, CA for hosting him as a visiting scientist in 1999 to carry out part of the work.
5 REFERENCES Adler, R., A. J. Negri, P. R. Keehn, and I. M. Hakkarinen, 1993: Estimation of monthly rainfall over Japan and surrounding waters from a combination of low-orbit microwave and geosynchonous IR data. J. Appl. Meteor., 32, 335–356. Adler, R. F. and A. J. Negri, 1988: A satellite infrared technique to estimate tropical convective and stratiform rainfall. J. Appl. Meteor., 27, 30–51. Atlas, D., D. Rosenfeld, and D. B. Wolff, 1990: Climatologically tuned reflectivity-rain rate relations and links to area-time integrals. J. Appl. Meteor., 29, 1120–1135.
278
F. S. Marzano et al.
Ba, M. B. and A. Gruber, 2001: GOES multispectral rainfall algorithm. J. Appl. Meteor., 40, 1500–1514. Bellerby, T., M. Todd, D. Kniveton, and C. Kidd, 2000: Rainfall estimation from a combination of TRMM Precipitation Radar and GOES multispectral satellite imagery through the use of an artificial neural network. J. Appl. Meteor., 39, 2115–2128. Calheiros, R. V. and I. I. Zawadzki, 1987: Reflectivity rain-rate relationships for radar hydrology in Brazil. J. Climate Appl. Meteor., 26, 118–132. Ferraro, R. R., 1997: SSM/I derived global rainfall estimates for climatological applications. J. Geophys. Res., 102, 16,715–16,735, . Hsu, K. L., X. Gao, S. Sorooshian, and H. V. Gupta, 1997: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36, 1176–1190. Kidd, C., D. Kniveton, and C. Barrett, 1998: The advantages and disadvantages of statistically derived empirically calibrated passive microwave algorithms for rainfall estimation. J. Atmos. Sci., 55, 1576–1582. Kummerow, C. and L. Giglio, 1995: A method for combining passive microwave and infrared rainfall observations. J. Atmos. Oceanic Technol., 12, 33–45. Levizzani, V., F. Porcù, F. S. Marzano, A. Mugnai, E. A. Smith, and F. Prodi, 1996: Investigating a SSM/I microwave algorithm to calibrate METEOSAT infrared instanttaneous rain-rate estimates. Meteor. Appl., 3, 5–17. Levizzani, V., J. Schmetz, H. J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2001: Precipitation estimations from geostationary orbit and prospects for Meteosat Second Generation. Meteor. Appl., 8, 23–41. Marzano, F. S., A. Mugnai, and F. J. Turk, 2002: Precipitation retrieval from spaceborne microwave radiometers and combined sensors. In Remote Sensing of Atmosphere and Ocean from Space, F. S. Marzano and G. Visconti, eds., Kluwer Academic, Dordrecht, The Netherlands, 107–126. Marzano, F. S., A. Mugnai, G. Panegrossi, N. Pierdicca, E. A. Smith, and J. Turk, 1999: Bayesian estimation of precipitating cloud parameters from combined measurements of spaceborne microwave radiometer and radar. IEEE Trans. Geosci. Remote Sens., 37, 596–613. Marzano, F. S., M. Palmacci, D. Cimini, G. Giuliani, and F. J. Turk, 2004: Multivariate statistical integration of satellite infrared and microwave radiometric measurements for rainfall retrieval at the geostationary scale. IEEE Trans. Geosci. Remote Sens., 42, 1018–1032. Marzano, F. S., F. J. Turk, P. Ciotti, S. Di Michele, and N. Pierdicca, 2001: Potential of combined spaceborne microwave and infrared radiometry for near real-time rainfall attenuation monitoring along earth satellite. Int. J. Satell. Commun., 19, 385–412. Miller, S. W., P. A. Arkin, and R. Joyce, 2001: A combined microwave infrared rain rate algorithm. Int. J. Remote Sens., 22, 3285–3307. Panegrossi, G., S. Dietrich, F. S. Marzano, A. Mugnai, E. A. Smith, X. Xiang, G. J. Tripoli, P. K. Wang, and J. P. V. Poiares Baptista, 1998: Use of cloud model microphysics for passive microwave-based precipitation retrieval: significance of consistency between model and measurement manifolds. J. Atmos. Sci., 55, 1644–1673. Pulvirenti, L., P. Castracane, F. S. Marzano, N. Pierdicca, and G. d’Auria, 2002: A physicalstatistical approach to match satellite passive microwave retrieval to the Mediterranean climatology. IEEE Trans. Geosci. Remote Sens., 40, 2271–2284. Richards, F. and P. Arkin, 1981: On the relationship between satellite-observed cloud cover and precipitation. Mon. Wea. Rev., 109, 1081–1093. Smith, E. A., J. Lamm, R. Adler, J. Alihouse, K. Aonashi, E. Barrett, P. Bauer, W. Berg, A. Chang, R. Ferraro, J. Ferriday, S. Goodman, N. Grody, C. Kidd, C. Kummerow, G. Liu, F. S. Marzano, A. Mugnai, W. Olson, G. Petty, A. Shibata, R. Spencer, F. Wentz,
Multivariate Probability Matching for MICRA
279
T. T. Wilheit, and E. Zipser, 1998: Results of WetNet PIP-2 projects. J. Atmos. Sci., 55, 1483–1536. Todd, M. C., C. Kidd, D. Kniveton, and T. J. Bellerby, 2000: A combined satellite passive infrared and passive microwave technique for estimation of small scale rainfall. J. Atmos. Oceanic Technol., 18, 742–755. Turk, J. F., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000: Meteorological applications of precipitation estimation from combined SSM/I, TRMM, and geostationary satellite data. In: Microwave Radiometry and Remote Sensing of the Environment, P. Pampaloni, ed., VSP Intern. Sci. Publisher, Utrecht, The Netherlands, 353–363. Vicente, G. A., R. A. Scofield, and W. P. Menzel, 1998: The operational GOES infrared rainfall estimation technique. Bull. Amer. Meteor. Soc., 79, 1883–1898.
22 TOWARD IMPROVEMENTS IN SHORT-TIME SCALE SATELLITE-DERIVED PRECIPITATION ESTIMATES USING BLENDED SATELLITE TECHNIQUES F. Joseph Turk1 and Amita V. Mehta2 1
Marine Meteorology Division, Naval Research Laboratory, Monterey, CA, USA National Aeronautics and Space Administration–University of Maryland Baltimore (UMBC) Joint Center for Earth Systems Technology, Greenbelt, MD, USA
2
Abstract
This article describes a strategy for blending the intermittently spaced precipitation estimates from various passive microwave sensors onboard low-Earth orbiting meteorological satellites with the observations from rapid time-update coincident geostationary-based visible and infrared imagers. Validation datasets are presented which demonstrate the performance of the technique at 3-hourly time scales, and its capability to capture regional diurnal cycles of precipitation.
Keywords
Passive microwave, geostationary, visible, infrared, satellite, blended, validation, gauge, NRL, TRMM, AMSR, SSM/I, precipitation, remote sensing, rainfall, rain-rate
1 CURRENT PASSIVE MICROWAVE-BASED SATELLITE CONSTELLATION Space-based passive microwave (PMW)-based radiometers have continued to expand in both number and capability since the deployment of the first Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager (SSMI) onboard the F-8 satellite in 1987. Most recently, the Earth Observing System (EOS) Aqua satellite, launched in 2002, deployed the Advanced Microwave Scanning Radiometer (AMSR-E), with expanded spectral capabilities. The AMSR-E sensor is augmented by (as of February 2004) nine operational (or with near-real-time capabilities) satellites with PMW rainfall-measuring capabilities: the Tropical Rainfall Measuring 281 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 281–290. © United States Government 2007.
282
F. J. Turk and A. V. Mehta
Mission (TRMM) Microwave Radiometer (TMI) and the Ku-band Precipitation Radar (PR), the SSMI onboard F-13/14/15, the National Oceanic and Atmospheric Administration (NOAA) Advanced Microwave Sounding Unit-B (AMSU-B, onboard NOAA-15/16/17), Coriolis/Windsat, and the first SSMI replacement, the SSMI/S, onboard DMSP F-16 (and through F-20, when the SSMI/S sensor will be replaced by the Conically Scanning Microwave Imager and Sounder (CMIS) as the DMSP program folds into the National Polar-orbiting Operational Environmental Satellite System (NPOESS) program (Cunningham et al. 2003).
Figure 1. Adjustments required to instantaneous-based rain estimates obtained from seven different passive microwave sensors (TMI, SSMI on F-13/14/15, AMSU-B on NOAA15/16/17) in order to match the TRMM PR rainfall histogram for all over-ocean pixels. Data are from three months between November 2003 and January 2004, between ±20° latitude. Each graph represents a different 3-h local time observation period; the last graph depicts the overall result for all local times. The solid 1:1 line denotes the TRMM-PR (no adjustment needed, since it is the reference).
Owing to different sensor frequencies, orbit patterns, scanning modes and polarization states, different precipitation-retrieval algorithms (Weng et al. 2002; Wilheit et al. 2003; Kummerow et al. 2001) are applied to different
Improvements in Short-Time Scale Satellite-Derived Precipitation
283
sensors, which yield different retrieved precipitation characteristics. This complicates the merging of the various instantaneous estimates into precipitation accumulations. One way to normalize everything is by analyzing a collection of coincident observations, where one satellite is chosen as the reference. A good satellite to choose as the reference is the TRMM satellite, as its non-sun-synchronous tropical orbit provides a number of precisely aligned space and time (under 1 min) coincident data pixels, as the TMI scan intersects the polar-orbiting sensor swaths. Turk et al. (2003) attempted this with TMI and SSMI observations. A better way to accomplish the merging is by carefully frequency-matching each of the rain-rate histograms of the non-reference satellite sensors to the reference. By subsequently adjusting each sensor’s instantaneous rainfall estimate, this procedure assures that over a sufficiently long period of time, each sensor is contributing equal rainfall statistics to the overall merged rainfall product. However, since each satellite has a different local overpass time, the satellite rainfall estimates should be binned into local time intervals as well. For example, Fig. 1 depicts the adjustments required for seven of the aforementioned satellites (in this example using the TRMM PR as the reference rain-rate), for over-water pixels between ±20° latitude and for 3 months between November 2003 and January 2004, separated into 3-h local time windows. Overall, the sensors are already fairly well-matched for rain-rates under 10 mm h–1, but in order to represent the smaller number of large rain events, the AMSU-B estimates require a large adjustment past 15 mm h–1 and PR past 30 mm h–1 . Since the PR is best capable of capturing the small amount of highest intensity rainfall events (the tail of the histogram), in practice one would most likely select this sensor as the reference. Figure 2 shows the corresponding results for the overland pixels. Since the (Version 5) TMI 2A12 overland rainfall algorithm is nearly the same as the SSMI algorithm, there is little adjustment required between TMI and SSMI. Most noticeable is the abundance of observations between 06–09 and 18–21 local time, and the lack of any non-TRMM observations between 03–06 and 15–18 local time. This often leaves short-time interval (e.g., 3 h) precipitation analyses with a limited number of intermittently spaced observations, and may significantly underrepresent the actual rainfall evolutionary state throughout the time interval. For numerical weather prediction (NWP) and now-casting applications requiring a global precipitation analysis on time scales of 3–6 h, the PMW-based 7-satellite constellation of Fig. 1 has a worst-case revisit time of approximately 6 hours in the tropical latitudes. A blended satellite technique attempts to fill-in the missing space and time observations with rapid-time, wide area coverage, multispectral imaging capabilities available from operational geostationary-based satellites. In the next section we describe one such technique, the NRL blended satellite technique, as well as some validation results gathered from comparisons with dense, rapid-time reporting rain-gauge networks.
284
F. J. Turk and A. V. Mehta
Figure 2. Same as Fig. 1, but for overland pixels only.
2 BLENDING DISJOINT SATELLITE OBSERVATIONS The NRL blended technique is based upon an adaptive, ongoing analysis of an underlying collection of time and space-intersecting pixels from all operational geostationary-orbiting visible and infrared (VIS/IR) imagers and PMW imagers. As new PMW datasets arrive, the PMW-derived rain-rate pixels are paired with their time and space-coincident geostationary IR brightness temperatures (TB) and visible reflectance data, using (currently) a 15-min maximum allowed time offset (denoted by ∆t) between the pixel observation times and a 10-km maximum allowed spatial offset (denoted by ∆d). Prior to this, the geostationary data are averaged to the approximate resolution of the PMW rainfall datasets, parallax-induced geolocation displacements are accounted for. Geostationary data are not used past a maximum 70° viewing zenith angle. Each collocated data increments histograms of the IR TB and the PMW rain-rate in the nearest 2° box, as well as the eight surrounding boxes (this overlap assures a fairly smooth
Improvements in Short-Time Scale Satellite-Derived Precipitation
285
transition in the histogram slopes between neighboring boxes). As soon as a 2° box is refreshed with new LEO data, a probabilistic histogram matching relationship (Calheiros and Zawadzki 1987) is updated using the PMW rainrate and IR TB histograms, and a freshly updated TB-R lookup table is created. To assure that the most timely rain history is maintained, the histograms of these coincident data are accumulated backwards from the current clock time (the “look-back” time) until the PMW coverage of a given 2° box exceeds a 90% coverage threshold. With this threshold, the average age of the data in each 2° box is about 4 h when the 7-satellite constellation is used. For comparison, with a smaller maximum allowed ∆t = 5 min, it takes about 35 h look-back time in order to achieve the 90% Earth coverage for the 7-satellite case. While the smaller time offset is preferred in order to capture as nearly time-coincident data as possible, it comes with the expense of increased age of coincident data observations. This is the reasoning behind the ∆t = 15 minutes time offset in the coincident data alignment procedure. The transfer of this information to the stream of steadily arriving geostationary data is then a relatively simple lookup table procedure, where an inverse-distance weighted average rain-rate is computed from the nine nearest-neighbor lookup table-derived rain-rates surrounding each IR pixel. 850 hPa wind vectors from the Navy Operational Global Atmospheric Prediction System (NOGAPS) forecast model are analyzed against a 2-min resolution topographic database, and a correction is applied in regions of likely orographic effects on both the upslope and downslope (Vicente et al. 2002). The previous 30-min time history of the 11 µm IR brightness temperatures is analyzed for regions of active cloud top temperature growth or decay, and scaling factors are applied to intensify and lighten the overall rain-rate. At 3-hourly time intervals, the rainfall accumulations are updated to produce the final products. This is done by backwards time-integrating the instantaneous PMW and IR-derived estimates, and producing output products at typical intervals, e.g., 3, 6, 12, 24 h, etc. Over the accumulations interval, PMW estimates are fully weighted, whereas the IR-based estimates are given a weight that depends upon their time proximity to a PMW observation, and the pixel latitude (PMW revisit is shorter at higher latitudes, so less IR information needs to be blended in). In essence, the IR-based rainfall estimates are only blended in where they are needed in space and time. Near the equator, the time-proximity threshold is longer than it is at higher latitudes, owing to the longer PMW revisit time. The technique has been run in the operational processing system at NRL since 1998. Future adaptations and formulations should be used to take full advantage of the expanded thermal and solar spectral capabilities offered in the Meteosat Second Generation (MSG) Spinning Enhanced Visible and Infrared Imager (SEVIRI) and future GOES-R Advanced Baseline Imager (ABI) geostationary-based imagers. Already, Marzano et al. (2002) have
286
F. J. Turk and A. V. Mehta
tested multivariate probability matching and nonlinear multiple regression techniques for the PMW-IR blending.
3 VALIDATION Satellite-derived precipitation totals are typically validated in terms of their performance against daily rain-gauge analyses, since a majority of the international gauge networks report 24-h totals at 0000 or 1200 UTC each day. In order to properly validate satellite-derived precipitation estimates at 3-hourly or less interval, the gauge network should be dense enough to capture as much rainfall variability within the validation grid size, and ideally should sample as many instantaneous estimates as possible within the validation temporal averaging size. However, as the time-averaging scale falls under 6 h (the approximate worst case revisit time in the tropics), the NRL blended technique operates under suboptimal conditions, where PMW observations are likely missing over much of the tropics and subtropics.
Figure 3. Satellite vs. gauge (RS vs. RG) scatterplots and correlation coefficient for the all data analyzed during June–August 2000 period over the southern Korean peninsula. To account for rain fallout time, the Automated Weather Station (AWS) rain-gauge data were first averaged for ±10 minutes centered about the precise time of each satellite pixel. Then, the data were both averaged to a spatial size of 1° latitude-longitude box size (left figure) and 3° (right figure), after which various time averages were further applied. From upper left of each side, the time averaging period is set to 1, 2, 3, 6, 12 and 24 h.
Improvements in Short-Time Scale Satellite-Derived Precipitation
287
Figure 4. Rain-gauge site location map for five Florida rain-gauge networks: (1) NASA/Kennedy Space Center high density 30-gauge network (KSC); (2) South Florida Water Management District low density 139-gauge network (SFL); (3) St. John’s River Water Management District low density 69-gauge network (STJ); (4) Southwest Florida Water Management District medium density 575-gauge network (SWF); and (5) Tropical Rainfall Measuring Mission high density 16-gauge network (TRMM).
Oh et al. (2002) used the AWS dense (500+ rain-gauges) network operated by the Korean Meteorological Agency (KMA) over the southern Korean peninsula in order to example several IR-style rainfall estimation techniques. During June–August 2000, the finest-scale accumulations possible from the NRL blended technique (0.1° spatial/1-hourly temporal) were collected. These data were analyzed against the AWS network data after averaging over a variety of space and time scale combinations. In the preprocessing of the AWS data, a ±10 minute average of the rain-gauge data (relative to the satellite pixel time) was first applied, in order to account for rain fallout time. In Fig. 3, we display the scatterplots of the satellite-gauge comparisons after averaging the gauge and satellite data onto common 1° boxes, and then averaging over various time intervals between 1 and 24 h. The correlation for the 24-h totals is 0.67, falling to 0.55 for the 3-h totals, suggesting that the blended technique is retaining a certain amount of information regarding the state of the accumulated rainfall even at a 3-h time scale. While this is encouraging, in the analysis of these data no diurnal separation of the data was retained (e.g., an accumulations interval was treated equally regardless of the local time of day), so it is not possible to say whether the satellite
288
F. J. Turk and A. V. Mehta
technique is faithfully capturing the diurnal rainfall variability over the Korean peninsula during these three months.
Figure 5. Normalized diurnal rain-rate cycles (use left ordinate) over central-south Florida during June 2003 as measured by five rain-gauge networks and NRL operational blended rainfall algorithm. Rain-gauge networks consist of: (1) high density NASA/Kennedy Space Center Operations network (KSC – 20 gauges); (2) low density South Florida Water Management District network (SFL – 139 gauges); (3) low density St. Johns River Water Management District network (STJ – 69 gauges); (4) medium density Southwest Florida Water Management District network (SWF – 575 gauges); and (5) high density TRMM Ground Validation network (16 gauges). For normalized rain-rate diurnal cycles, diagram shows 3-hourly network averages separately (thin solid colored lines), All-gauge average (thick solid red line), and satellite average (thick solid black line). Actual diurnal rain-rate cycles (use right ordinate) for All-gauge and satellite averages are shown in red and black dash lines, respectively.
To examine further if the NRL blended technique is capturing diurnal variations, we examine an additional data analysis, which was carried out over northern Florida using data collected from five separate networks whose locations are depicted in Fig. 4. The rainfall over Florida has a strong diurnal component with an afternoon maximum, modified regionally by the timing of the local sea breeze on the east and west coasts. These gauge data were averaged into 0.25° grid boxes and into 3-h local time accumulations intervals. Figure 5 shows the normalized rain-rate diurnal cycles for each of the five networks, and the total overall normalized and actual rain-rate cycles for the gauge and NRL blended-satellite estimates. Although the satellite technique is biased low relative to the gauge networks (Turk et al. 2003), the normalized satellite-derived rain-rate appears to track the evolution of the rain-rate throughout the day, notably the afternoon maximum near 1500
Improvements in Short-Time Scale Satellite-Derived Precipitation
289
local time. The gauges in southern Florida (SFL) also recorded a nighttime precipitation maximum which pushed the overall normalized gauge values to a second, smaller maximum, which is less apparent in the satellite estimates.
4 CONCLUSIONS We have described an operational method for blending intermittently spaced, PMW-based rainfall observations with rapid update VIS/IR-based observations, which is intended to produce an update of global precipitation accumulations at a 3-hourly refresh rate. While there are several techniques (current and past) that accomplish this through various methodologies, there has been very little validation of the blended satellite techniques at sub-daily time scales. In this article, we have presented example validation results from the NRL blended satellite technique using dense, rapid-time sampling rain-gauge networks in Korea and Florida. The results demonstrate that the technique retains skill in estimating the accumulated precipitation over a 3-h interval, with an highest correlation obtained for daily scale. Using a overall rain-gauge analysis from five separate rain-gauge networks throughout Florida, the 3-hourly average rainfall estimated by the NRL technique was shown to track the diurnal variations observed by gauge network.
Acknowledgements: The first author acknowledges the support from the National Aeronautics and Space Administration Earth Sciences Division under grant NNG04HK11I, and the Oceanographer of the Navy through the program office Space and Naval Warfare Systems Command, PMW-150, under program element 0603207N. We thank Brad Fisher from the NASA TRMM-GV group for making the TRMM-GV gauge rain data, and Ken Romie, Hydrologic Data Section, Southwest Florida Water Management District, for providing the Southwest Florida gauge data.
5 REFERENCES Calheiros, R. V. and I. Zawadzki, 1987: Reflectivity rain-rate relationship for radar hydrology and Brazil. J. Climate Appl. Meteor., 26, 118–132. Cunningham, J. D., F. L. Ricker, and C. S. Nelson, 2003: The National Polar-Orbiting Operational Environmental Satellite System Future U.S. Operational Earth Observation System. Proc. IGARSS 2003, 21–25 July, Toulouse, France, CD-R, I:353–356. Kummerow, C. D., Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, D. B. Shin, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1817.
290
F. J. Turk and A. V. Mehta
Marzano, F. S., M. Palmacci, D. Cimini, and J. F. Turk, 2002: Statistical integration of satellite passive microwave and infrared data for high-temporal sampling and retrieval of rainfall. Proc. IGARSS-2002, CD-R, Toronto, Canada, 24–28 June. Oh, H. J., B. J. Sohn, E. A. Smith, F. J. Turk, A. S. Seo, and H. S. Chung, 2002: Validating infrared-based rainfall retrieval algorithms with 1-minute spatially dense raingauge measurements over the Korean peninsula. Meteor. Atmos. Phys., 81, 273–287. Turk, F. J., E. E. Ebert, H.-J. Oh, B.-J. Sohn, V. Levizzani, E. A. Smith, and R. R. Ferraro, 2003: Validation of an operational global precipitation analysis at short time scales. Prepr. 12th Conf. on Satellite Meteor. and Ocean., CD-R, Long Beach, CA, 9–13 Feb., paper J1.2, 21 pp. Vicente, G. A., J. C. Davenport, and R. A. Scofield, 2002: The role of orographic and parallax correction on real time, high resolution satellite rain rate observation. Int. J. Remote Sensing, 23, 221–230. Weng, F., L. Zhao, R. R. Ferraro, G. Poe, X. Li, and N.C Grody, 2002: Advanced Microwave Sounding Unit (AMSU) cloud and precipitation algorithms. Radio Sci., 38(4), 8068–8079. Wilheit, T., C. D. Kummerow, and R. Ferraro, 2003: Rainfall algorithms for AMSR-E. IEEE Trans. Geosci. Remote Sens., 41, 204–214.
23 GLOBAL RAINFALL ANALYSES AT MONTHLY AND 3-H TIME SCALES George J. Huffman1,2, Robert F. Adler1, Scott Curtis3, David T. Bolvin1,2, and Eric J. Nelkin1,2 1
NASA/GSFC Laboratory for Atmosphere, Greenbelt, MD, USA Science Systems and Applications, Inc., Greenbelt, MD, USA 3 East Carolina University, Greenville, NC, USA 2
1 INTRODUCTION The last two decades of research in estimating precipitation from satellite remote-sensing data has demonstrated tremendous utility and potential for providing the long-term, global coverage that a wide range of research and operational groups need for their work. This paper focuses on combining data from multiple sensors, sometimes including gauge analyses. The work covers a wide range of scales, yet it shares a common set of design choices: • The input data are all converted to precipitation estimates before use, rather than combining sensor parameters, such as radiances. • No model-based estimates are employed, due to strong user comment. • “Best” estimators of precipitation are used to provide the underlying statistics for calibrations. This makes products traceable to a single “standard”, although it opens the possibility that the standard can introduce errors. • The calibrating product’s statistics are used to adjust precipitation datasets possessing the “best” spatial/temporal coverage. • The adjustments are computed for month-long periods for simplicity and stability. • It is a priority to estimate a field of random error separately for each precipitation field.
291 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 291–305. © 2007 Springer.
G. J. Huffman et al.
292
The following sections describe a climatologically oriented monthly product and a much finer scale 3-hourly product that the authors have developed, give some examples of each, and outline the future directions that might be in store for these products or their successors.
2 GPCP VERSION 2 MONTHLY SG COMBINATION METHODS The Global Precipitation Climatology Project (GPCP, an international activity of the Global Energy and Water Cycle Experiment) has the goal of developing and producing long-term, global precipitation analyses at monthly and finer time scales. The authors’ research group developed and now computes the GPCP Version 2 Satellite-Gauge (SG) data set based on a variety of input data sets and techniques provided by other GPCP components (Table 1). Table 1 Summary of input data sets used in the GPCP Version 2 SG. Algorithm
Input data
GPCC gauge analysis CAMS+GHCN gauge analysis Emission-based passive microwave estimates Scattering-based passive microwave estimates AGPI
~6500 surface stations ~6500 surface stations SSM/I on DMSP F08, F11, F13
TOVS-based estimates OPI
Space scale 2.5°
Time scale Monthly
Areal coverage Global land
2.5°
Monthly
Global land
2.5°
Monthly
60° N – 60° S July 1987– ocean present
SSM/I on DMSP F08, F11, F13
2.5°
Monthly
Global
all GEO and LEO IR Tb’s
2.5°
Pentad
40° N– 40° S 1986–1996
1°
3-hour
1°
Monthly
2.5°
Monthly
40° N – 40° S 1997– present Global July 1987– present Global 1979–June 1987
TOVS sounding data LEO-IR
Time coverage 1986– present 1979–1985
July 1987– present
These inputs were selected when the Version 2 SG was designed to provide a reasonable, stable base from the changing mix of quasi-global satellite and rain-gauge information that has been recorded over the period of continuous satellite records related to precipitation, namely 1979 to the present.
Global Rainfall Analyses
293
Highlights of the SG algorithm are summarized below for each major data epoch, drawing on the more-detailed material in the online documentation (ftp://precip.gsfc.nasa.gov/pub/gpcp-v2/doc/V2_doc) and Adler et al. (2003). Although reasonable care has been taken to minimize discontinuities between the data epochs, users need to carefully examine their results for possible statistical inhomogeneities due to known changes in data coverage. Throughout the record, the GPCP SG algorithm applies the Legates (1987) climatological bias correction to all gauge analyses, which raises estimated amounts by 5–300% depending on assumed gauge type, windiness, and snowiness.
2.1 Combination methods – Mid-1987–Present For most of the period of record, beginning in July 1987, but not including December 1987 due to operational considerations, the SG incorporates Special Sensor Microwave/Imager (SSM/I) passive microwave estimates. To avoid possible changes in bias due to shifts in the time-of-day of SSM/I observations, the GPCP SG only uses data from the early morning Defense Meteorological Satellite Program (DMSP) platforms (F08, F11, F13). Despite the relatively high quality ascribed to SSM/I estimates, all current algorithms falter in cold-land, icy-surface, and polar conditions. In contrast, Television-Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS) estimates are considered less reliable at lower latitudes, but seem to provide useful data at higher latitudes. To draw on their perceived strengths, the SSM/I and TOVS estimates are composited as follows: • In the band 40° N-S, the SSM/I estimates are used “as is”. • As needed, mostly in mountain and winter land cases, in the band 40° NS, the TOVS data are adjusted to the zonally averaged bias of the SSM/I data and inserted in SSM/I data voids. • Just outside of the band 40° N-S, the SSM/I and TOVS data are averaged. • Further towards the poles the SSM/I-TOVS average is replaced with biasadjusted TOVS data. The bias adjustment is anchored on the Equator-ward side by the SSM/I-TOVS average and on the polar side by zonal-average monthly climatological rain-gauge analyses. • In the 70°–90° N polar cap, TOVS data are adjusted to the zonal-average bias of the available monthly rain-gauge data. • In the 70°–90° S polar cap, TOVS data are adjusted to the bias of the annual zonal-average climatology of the rain-gauge data. The infrared brightness temperatures (IR Tb’s) are corrected for zenith-angle viewing effects and inter-satellite calibration differences, as they are in all IR products described in this paper, and are converted into precipitation estimates by applying the Adjusted Geostationary Operational Environmental Satellite
294
G. J. Huffman et al.
(GOES) Precipitation Index (AGPI; Adler et al. 1994): A month of approximately time/space matched IR Tb’s and SSM/I rain estimates are collected on a global grid in the latitude band 40° N-S. For each grid box, the GPI rain/no-rain threshold of 235K (Arkin and Meisner 1987) is applied, then all the “raining” subsetted IR pixels are used to compute a single conditional rain rate such that they sum to the total rain in the coincident subset of SSM/I pixels. This procedure is carried out separately for geostationary and low-Earth orbit (GEO and LEO) IR data. With all of the input data now in the form of monthly precipitation estimates on a 2.5° × 2.5° grid, the Multi-Satellite (MS) estimate is computed as: • AGPI estimates where available (40° N-S), • weighted combination of the merged SSM/I-TOVS estimates and the LEO-AGPI elsewhere in 40° N-S, and • composited SSM/I-TOVS data outside of 40° N-S. • Finally, the SG product is produced in two steps: • The MS estimate is adjusted to the large-scale gauge average for each grid box over land. • The gauge-adjusted MS estimate and the gauge analysis are combined in a weighted average, where the weights are the inverse (estimated) error variance of the respective estimates. Throughout the processing, an estimate of the random error is produced for each grid box in each input, intermediate, and output precipitation estimate field following Huffman (1997). The errors are similarly calculated in each of the earlier periods described below.
2.2 Combination methods – Earlier periods The period before the start of SSM/I observations requires more approximate schemes. During the period 1986–June 1987, plus December 1987, Outgoing Longwave Radiation (OLR) Precipitation Index (OPI) data are climatologically calibrated by the 1988–1996 GPCP SG estimates and used in place of the SSM/I-TOVS component (above). The MS field is built from geoAGPI estimates where available (40° N-S) and calibrated OPI estimates elsewhere, then the combination with the gauges proceeds as in the recent era to produce the SG. The GPCP summaries of GEO-IR data are not available during 1979–1985, so the OPI data are used “as is” for the MS estimates, and the gauge combination proceeds as in the recent era to produce the SG.
2.3 Quality summary The GPCP SG validates relatively well against standard and special gauge data sets, in part because the gauge adjustment scheme prevents significant bias (Krajewski et al. 2000; Adler et al. 2003) and in part because the
Global Rainfall Analyses
295
adjustment accounts for uncertainty in the gauge analysis. In general there is a decrease in accuracy as the precipitation becomes light, the environment becomes more polar, and/or the surface becomes icy or frozen. Over oceans there is a general lack of gauge data, so the SG equals the MS, and validation studies are quite limited. Validation against the Pacrain atoll rain-gauge data (Morrissey et al. 1995) in the tropical Pacific Ocean tends to show a low bias of some 16%, which is traceable to the calibration by the Wilheit et al. (1991) estimates used in that region (Adler et al. 2003). For the latitude band 30° N–30° S the time series of estimated area-average precipitation over ocean from GPCP rather closely matches that of Goddard Profiling (GPROF; Kummerow et al. 1996; Olson et al. 1999) estimates made with TRMM Microwave Imager (TMI) data. The Version 5 GPROFTMI estimates exceed the GPCP SG in this region by about 7%, while the new Version 6 results (still in production) appear much closer to the GPCP averages. In regions of complex terrain the microwave estimates sometimes fail to capture orographic enhancements, and this shortcoming is propagated to the AGPI, MS, and SG products. Meanwhile the gauge analysis tends to underestimate the precipitation because relatively few gauges are located at the higher elevations, where the heavier precipitation occurs (Nijssen et al. 2001). This known problem is under study, but no corrective scheme has been developed. Finally, validation of the random error estimation scheme demonstrated that the Huffman (1997) parameterization gives reasonably good estimates over a wide variety of conditions in the United States (Krajewski et al. 2000).
2.4 Other GPCP products The GPCP computes two other shorter-interval products. Both are calibrated such that they approximately sum to the GPCP SG at 2.5° × 2.5° monthly scale. First, the GPCP Pentad product provides estimates each pentad (5-day period) on a globally complete 2.5° × 2.5° latitude/longitude grid for the period 1979–present. The GPCP Pentad analysis calibrates the pentad version of the CPC Merged Analysis of Precipitation product to the GPCP SG (Xie et al. 2003). The second GPCP short-interval product is the OneDegree Daily (1DD) product, which provides estimates each day on a globally complete 1° × 1° lat./long. grid for the period October 1996–present. In broad terms the Version 2 SG’s concept of calibrating the IR by the microwave and recalibrating TOVS estimates at higher latitudes is applied in the 1DD at the daily scale (Huffman et al. 2001). One important departure from the SG is that the rain/no-rain threshold Tb for the IR is set locally based on monthly match-ups between the IR Tb’s and SSM/I passive microwave estimates.
296
G. J. Huffman et al.
3 THE MPA In parallel with the GPCP effort, a monthly 1° × 1° product, 3B43, was developed for TRMM that draws on multiple satellites with calibration from TRMM and features combination with gauge analysis. The TRMMcalibrated AGPI that feeds into the monthly product was released as a separate product (3B42), originally on pentads, but for individual days in Version 5. In all cases the spatial domain was the latitude band covered by the TMI, 40° N–40° S. The next research goal was to develop a relatively fine-scale set of estimates incorporating all routinely available quasi-global remote sensing and in situ precipitation estimators. The concept was named the Multisatellite Precipitation Analysis (MPA), and the initial work was carried out in a real-time framework. The primary merged microwave-infrared product is computed at the 3-hourly, 0.25° × 0.25° latitude/longitude resolution over the latitude band 50° N–50° S. The MPA is computed twice, first as a besteffort monitoring product about 6 hours after real-time (RT), and then as a post-real-time research-quality product about 10–15 days after the end of each month. See ftp://aeolus.nascom.nasa.gov/pub/merged/3B4XRT_doc for details; a refereed publication is forthcoming. The MPA is computed with a wider selection of satellite data sets than the GPCP SG (Table 2). Various satellites come on line at different points during the period of record of the MPA, so the user needs to be cognizant of how the data mix at a particular time might affect a given study. Two additional data sources are passive microwave data from the Advanced Microwave Scanning Radiometer – Earth Observation Satellite (AMSR-E) and the Advanced Microwave Sounding Unit B (AMSU-B). The AMSU-B algorithm (Zhao and Weng 2002; Weng et al. 2003) computes Ice Water Path (IWP) from the 89- and 150-GHz channels, with a surface screening by ancillary data. The IWP–precipitation rate relations are derived from cloud model data, with a maximum precipitation rate of 30 mm h–1. The current algorithm cannot provide information on precipitation systems that lack the ice phase, so the estimates are deficient wherever “warm rain” dominates, such as in the oceanic subtropical highs. A third new data source in the MPA is the CPC merger of the international complement of GEO-IR data into half-hourly 4 × 4 km2-equivalent lat./long. grids (hereafter the “CPC merged IR”; Janowiak et al. 2001). Finally, the post-real-time MPA makes use of precipitation estimates from the TRMM Combined Instrument (TCI) estimates (TRMM product 2B31; Haddad et al. 1997a, b), which employ data from both TMI and the TRMM Precipitation Radar (PR). We now describe processing for the post-real-time MPA, followed by a short summary on how the real-time processing differs.
Global Rainfall Analyses
297
Table 2. Summary of input data sets used in the TRMM MPA. The shaded entries are only used in the post-real-time product. All others appear in both the real-time (perhaps for only part of the record) and the post-real-time. A space scale of “pixel” indicates that the data are accessed at the native resolution of the original pixels, while a time scale of “swath” indicates that the values are instantaneous and observed at times that depend on location and orbit geometry. Algorithm
Input data
GPCC gauge analysis CAMS gauge analysis TRMM Combined Instrument GPROF
~6500 surface stations ~6500 surface stations TRMM PR and TMI
NESDIS High Frequency VAR
Space scale 1°
Time scale Monthly
Areal coverage Global land
0.5°
Monthly
Global land
Pixel
Swath
Global land
TMI
Pixel
Swath
40° N–40°S
SSM/I on DMSP F13, F14, F15 AMSR-E on Aqua AMSU-B on NOAA 15,16,17 all GEO and LEO IR Tb’s all GEO IR Tb’s
Pixel
Swath
70° N–70°S
Pixel
Swath
70° N–70°S
Pixel
Swath
1°
3 hour
4 km
30 min.
Time coverage 1998– present 1998– present 1998– present 1998– present 1998– present
June 2003– present Global January 2000– present 40° N–40°S 1998–6 Feb. 2000 60° N–60°S 7 Feb. 2000– present
3.1 High Quality (HQ) microwave estimates All of the passive microwave data are converted to precipitation estimates and averaged to the 0.25° spatial grid over the time range ±90 min from the nominal 3-hourly observation time. The gridded estimates are adjusted to a “best” estimate using probability matching (Miller 1972). The calibrating data source for the post-real-time MPA is the TCI. Since the TCI only occasionally intersects any of the sensors other than TMI, we compute a TCI–TMI calibration, then apply it to TMI-calibrated values of the other sensors to estimate the TCI-calibrated values. Preliminary work showed that the TMI calibrations of the other sensors’ estimates are adequately represented by climatologically based coefficients representing large zonal bands for ocean and single calibrations for land (for each sensor). The calibrations are computed for a month of match-ups to ensure stability and representativeness, except the TMI–AMSR-E calibration
298
G. J. Huffman et al.
requires 2 months of data to meet these goals. The calibration month in the post-real-time system is a calendar month. In the event of multiple overpasses covering an HQ grid box in a 3-h interval, data from TCI and TCI-adjusted TMI, AMSR-E, and SSM/I are averaged together, while TCI-adjusted AMSU-B estimates are used if no other microwave estimate is available.
3.2 Variable Rain Rate (VAR) IR estimates As shown in Table 2, the post-real-time MPA uses two different IR data sets. Before 7 February 2000, each grid box’s histogram in the 1° × 1° 3-hourly GPCP IR histogram data set is averaged to a single value for the grid box, and plane-fit interpolated to the 0.25° grid. Thereafter, the CPC Merged IR is averaged from its native 4 × 4 km2-equivalent to 0.25° resolution and combined into hourly files as ±30 min from the nominal time. Histograms of time-space matched HQ precipitation rates and IR Tb’s are accumulated for a calendar month, and then probability matched to create spatially varying Tb rain-rate look-up tables that are applied to that month of IR data. There is no precipitation when the 0.25° × 0.25°-average Tb is greater than the local rainno rain threshold value, while colder Tb’s are assigned larger precipitation rates. The highest rain/lowest Tb calibration tends to fluctuate unphysically, so a climatological fitted curve is substituted for the coldest 0.17% of the Tb–precipitation rate curve. Once computed, the HQ-IR calibration coefficients are applied to each 3-h interval IR data set during the calendar month.
3.3 Combined HQ and VAR estimates This product is intended to provide the “best” estimate of precipitation in each grid box at each observation time. Combining data is relatively easy for passive microwave estimates because the sensors are quite similar and GPROF is used for most retrievals, but it is much harder for the HQ and VAR fields. We currently take the HQ estimates wherever they exist, and then populate the remaining grid boxes with VAR estimates. The resulting data fields have a mix of statistical properties, so users for whom homogeneous statistics are important may choose to work with the HQ or VAR estimates alone.
3.4 Rescaling to monthly data The last stage in the post-real-time MPA is to compute the satellite-gauge adjustment and then rescale the individual 3-hourly grids to sum to the monthly value. Experience shows that sub-monthly accumulations of gauge data are not reported with sufficient density to warrant direct inclusion in a
Global Rainfall Analyses
299
global algorithm that provides sub-monthly resolution. This issue was solved in the sub-monthly GPCP data sets by scaling the short-period estimates to sum to a monthly estimate that includes monthly gauge data. Here, we take a similar approach: All of the individual HQ+VAR fields are summed over a calendar month to create a monthly MS field. The MS and gauge fields are combined as in Huffman et al. (1997) to create a post-real-time monthly satellite-gauge combination, which is a TRMM product in its own right (3B43 in Version 6). Then the field of satellite-gauge/MS ratios is computed and used to scale each 3-hourly HQ+VAR field in the month.
3.5 RT algorithm adjustments The real-time and post-real-time systems are designed to be as similar as possible; however, a real-time system cannot reach into the future, so the calibration month is taken as a trailing accumulation of approximately 6 pentads. Through 2004, new coefficients were computed at the end of each pentad, but thereafter the coefficients were recomputed every three hours to better represent heavy-rain outbreaks. A second important difference is the choice of calibrator for the HQ field; the TMI precipitation is used in the real-time because the TCI is only computed after real-time. Finally, there is not sufficient gauge data to drive a MS-gauge combination in the RT system.
3.6 Data set status The real-time MPA estimates are available on a best-effort basis beginning in late January 2002 at ftp://aeolus.nascom.nasa.gov/pub/merged. The postreal-time MPA (TRMM operational product 3B-42), which only provides the final gauge-adjusted HQ+VAR field, is currently in production. Reprocessing starting with the first full month of TRMM (January 1998) commenced in May 2004, while processing of then-current 3B42 data commenced in December 2004.
3.7 Quality summary The MPA is sufficiently new that definitive studies have not yet been completed. Known issues include complex terrain, cold land, and oceanic subtropical highs. Qualitatively, the HQ product shows a relatively large scatter against surface data, and the VAR product is even more uncertain. These results are consistent with previous research on similar fine-scale products (Fisher 2004). The scatter reflects mismatches in time and space between the satellite and surface data, algorithmic problems with the satellite data, and limitations to the information content in the satellite data. Katsanos et al. (2004) compared daily accumulations at 73 gauges around the eastern Mediterranean Sea to daily accumulations for the closest 0.25° MPA grid box. They found reasonable results, with almost unbiased estimates for the
300
G. J. Huffman et al.
low and medium daily precipitation accumulations, especially during the wet period of the year. At higher accumulation values and during the dry period of the year, the satellite data overestimate the rain events compared to the rain gauges. A long-term validation effort is now underway for Australia, the USA, and Western Europe (http://www.isac.cnr.it/~ipwg/validation.html). To date, the MPA tends to perform better in moist convective conditions, and worse in mid-latitude cool season conditions, in common with other satellite-based techniques.
Figure 1. Longitude/time Hovmöller diagram of GPCP Version 2 monthly SG precipitation anomalies along the Equator (averaged 10°N-10°S) for the period 1979–2003. Values range from –6 (black) to +6 (white) mm d-1 in 1 mm d–1 increments.
Global Rainfall Analyses
301
4 EXAMPLES The GPCP SG datasets cover almost 25 years, making it feasible to examine interannual fluctuations, of which the El Niño – Southern Oscillation (ENSO) is the premier example. A positive (negative) ENSO event, known as El Niño (La Niña), is characterized by heavier (lighter) precipitation over the central Pacific Ocean. This is demonstrated in Fig. 1, which shows a longitude/time Hovmöller diagram for monthly rainfall anomalies averaged over the latitude range 10°N-S. Maxima in precipitation anomalies appear every few years in the light-rain areas of the central-to-eastern Pacific (16080°W), most notably in 1982–1983 and 1997–1998. In other years a deficiency of precipitation appears in the central Pacific, including 1988– 1989 and 1998–2000. ENSO events cause broader-scale impacts as well, such as deficits in Amazonia (80–40° W) and the Maritime Continent (120–150° E) and increases in the western Indian Ocean (40–80° E) in El Niño events. We illustrate the effect of ENSO events over land by developing a differential ranked precipitation analysis for the months January 1988– March 2004. First, the normalized precipitation amounts are ranked from least to greatest in each grid box. Low rank values represent abnormally low precipitation (drought), while high values denote excessive precipitation and possible flooding (e.g., Higgins et al. 2000). The precipitation amount attached to a given ranking varies widely across grid boxes, but the rank reliably depicts how rare the amount is from a local perspective. Then the third of months with the highest and lowest NINO3,4 index values are labelled “El Niño” and “La Niña”, respectively, with the rest being “neutral”. Third, for each rank value, the number of occurrences of that rank value in all land grid boxes is accumulated for each of the three ENSO categories and normalized by the total number of grid boxes. Finally, each of the ENSO-event curves is differenced from the neutral curve, showing relative excess or deficit compared to neutral (Fig. 2). Both phases of ENSO show a shift toward droughtier conditions. The shift is more marked in El Niño, while La Niña is characterized by decreases of moderately moist events in favor of a broad range of dry events. El Niño is characterized by dry tropical land areas (Ropelewski and Halpert 1987), while La Niña is characterized by wet tropical land areas (Ropelewski and Halpert 1989). Apparently, drying in subtropical and mid-latitude land areas is important. Turning to the fine-scale estimates, the image shown in Fig. 3 was chosen because it contains four tropical cyclones: Tropical storm (later hurricane) Jeanne over Puerto Rico, hurricane Ivan on the US Gulf Coast, hurricane Javier south of Baja California, and hurricane Isis near 120°W. The relative coverage by HQ and VAR in the final MPA field is shown by coloring the zero-precipitation values from HQ white and those from VAR gray. This real-time field contains passive microwave data from three SSM/I sensors (the nearly north–south swaths) and the TMI. Irregularly shaped gaps in the
302
G. J. Huffman et al.
HQ data over the Andes and Himalaya Mountains denote regions where the GPROF algorithm is not able to make retrievals. Also, this colorization makes it easy to examine the fit between the HQ and VAR patterns. The distribution is fairly typical, ranging from good (east of Japan) to poor (south of Madagascar).
Figure 2. An analysis of GPCP SG precipitation over all land areas in the period 1988–March 2004 showing the differences between the ranked precipitation for El Niño conditions and Neutral conditions (short dash) and for La Niña conditions and Neutral conditions (long dash). The horizontal axis is rank, from driest (-6) to wettest (+6).
Figure 3. Final combined HQ+VAR precipitation field computed with the real-time MPA for 15Z 15 September 2004. Zero values are colored white if computed by the HQ and gray if by VAR; all other precipitation values have the same color value for both sources. Figure courtesy of Harold Pierce, NASA/GSFC Laboratory for Atmospheres and Science Systems and Applications, Inc. (see also color plate 10).
Global Rainfall Analyses
303
Because the instantaneous MPA estimates have a great deal of uncertainty, the authors generally recommend taking averages to increase confidence in the result. Figure 4 displays an example of accumulating the rainfall ascribed to tropical storms that made landfall on the continental U.S. at some point in their track during 2004. For each storm, azimuthal averages of rainfall were computed for each image and the first radius for which the average fell below 1 mm h–1 was taken as the storm’s edge at that time. The resulting accumulation map shows a wide range of values that reflect the details of each storm’s structure and intensity over time. For example, Jeanne has very little precipitation off the eastern tip of Cuba because the storm had decayed significantly as it interacted with Hispaniola, while it regained strength and dropped more rain east of the Bahamas. A digital movie of the successive accumulations during the season is available at http://trmm.gsfc.nasa.gov/ publications_dir/atlantic_2004_tc.html.
Figure 4. Accumulations of rainfall attributed to tropical cyclones that made landfall on the continental U.S. at some point in their track. The storms’ tracks are also shown. Figure courtesy of Harold Pierce, NASA/GSFC Laboratory for Atmospheres and Science Systems and Applications, Inc.
5 FUTURE DEVELOPMENT We plan to make the real-time MPA calibration more consistent with that of the post-real-time MPA both by testing the real-time TRMM PR estimates as
304
G. J. Huffman et al.
the calibration standard for the HQ field, and by testing the use of climatological gauge relationships. A number of procedural questions remain as well, including improving the IR estimation approach, perhaps by incorporating the Convective-Stratiform Technique (Adler and Negri 1988), and improving the merger scheme for the final combination field. The next major development effort is to expand the MPA domain to the full globe by using TOVS precipitation estimates, and successor estimates from Advanced Infrared Sounder (AIRS). On the longer term, the MPA will be upgraded to prepare for the Global Precipitation Measurement core satellite at the end of the decade. The next major challenge in the GPCP is to develop a Version 3 plan that takes advantage of the new short-interval combination schemes. The postreal-time MPA illustrates how Version 3 might incorporate “all” of the available instantaneous satellite data while ensuring consistency with monthly gauge analyses. One issue is the trade-off between providing fine-scale estimates appropriate to the modern data-rich satellite era and ensuring the longest possible, relatively homogeneous climate-oriented precipitation product. Three issues affect both the GPCP and MPA efforts. First, it is necessary to improve analysis schemes to address the complex terrain issue raised earlier for both gauges and satellite. The second overarching issue is that we need practical approaches to estimating error at the finest scales, then aggregating them to larger scales, including months. This work must include choosing what parameter or parameters best distil the multi-scale, partially correlated uncertainty associated with precipitation estimates, as the result of both natural processes and algorithmic uncertainty. Finally, uniform access and navigation routines are currently being developed for the International Satellite Cloud Climatology Program B1 data archive (Schiffer and Rossow 1985), which will provide access to the international constellation of GEO satellites for their entire record, subset to 10 km and 3 h resolution. As well, this sets the stage for future uniform access and navigation of the full GEO satellite archives. In concert with the long-term record of LEO sounder data, the B1 and full data could enable relatively fine-resolution precipitation estimates back to the early 1980s.
6 REFERENCES Adler, R. F., G. J. Huffman, A. Chang, R. Ferraro, P. Xie, J. Janowiak, B. Rudolf, U. Schneider, S. Curtis, D. Bolvin, A. Gruber, J. Susskind, and P. Arkin, 2003: The Version 2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–Present). J. Hydrometeor., 4, 1147–1167. Adler, R. F., G. J. Huffman, and P. R. Keehn, 1994: Global tropical rain estimates from microwave-adjusted geosynchronous IR data. Remote Sens. Rev., 11, 125–152. Adler, R. F. and A. J. Negri. 1988: A satellite infrared technique to estimate tropical convective and stratiform rainfall. J. Appl. Meteor., 27, 30–51. Arkin, P. A. and B. N. Meisner, 1987: The relationship between large-scale convective
Global Rainfall Analyses
305
rainfall and cold cloud over the western hemisphere during 1982–1984. Mon. Wea. Rev., 115, 51–74. Fisher, B., 2004: Climatological validation of TRMM TMI and PR monthly rain products over Oklahoma. J. Appl. Meteor., 43, 519–534. Haddad, Z. S., E. A. Smith, C. D. Kummerow, T. Iguchi, M. R. Farrar, S. L. Durden, M. Alves, and W. S. Olson, 1997a: The TRMM “day-1” radar/radiometer combined rainprofiling algorithm. J. Meteor. Soc. Japan, 75, 799–809. Haddad, Z. S., D. A. Short, S. L. Durden, E. Im, S. Hensley, M. B. Grable, and R. A. Black, 1997b: A new parameterization of the rain drop size distribution. IEEE Trans. Geosci. Remote Sens., 35, 532–539. Higgins, R. W., J. K. E. Schemm, W. Shi, and A. Leetmaa, 2000: Extreme precipitation events in the western United States related to tropical forcing. J. Climate, 13, 793–820. Huffman, G. J., 1997: Estimates of root-mean—square random error for finite samples of estimated precipitation. J. Appl. Meteor., 36, 1191–1201 Huffman, G. J., R. F. Adler, P. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78, 5–20. Huffman, G. J., R. F. Adler, M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 36–50. Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2001: A real-time global half-hourly pixelresolution IR dataset and its applications. Bull. Amer. Meteor. Soc., 82, 205–217. Katsanos, D., K. Lagouvardos, V. Kotroni, and G. J. Huffman, 2004: Statistical evaluation of MPA-RT high-resolution precipitation estimates from satellite platforms over the Central and Eastern Mediterranean. Geophys. Res. Lett., 31, L06116, doi:10.1029/2003 GL019142. Krajewski, W. F., G. J. Ciach, J. R. McCollum, and C. Bacotiu, 2000: Initial validation of the Global Precipitation Climatology Project monthly rainfall over the United States. J. Appl. Meteor., 39, 1071–1086. Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 1213–1232. Legates, D. R., 1987: A Climatology of Global Precipitation. Publications in Climatology, Vol. 40, University of Delaware, 85 pp. Miller, J. R., 1972: A climatological Z-R relationship for convective storms in the Northern Great Plains. 15th Conf. on Radar Meteor., 153–154. Morrissey, M. L., M. A. Shafer, S. E. Postawko, and B. Gibson, 1995: The Pacific rain gage rainfall database. Water Resources Res., 31, 2111–2113. Nijssen, B., G. M. O’Donnell, D. P. Lettenmaier, D. Lohmann, and E. F. Wood, 2001: Predicting the discharge of global rivers. J. Climate, 14, 3307–3323. Olson, W. S., C. D. Kummerow, Y. Hong, and W.-K. Tao, 1999: Atmospheric latent heating distributions in the Tropics derived from satellite passive microwave radiometer measurements. J. Appl. Meteor., 38, 633–664. Ropelewski, C. F. and M. S. Halpert, 1987: Global and regional scale precipitation patterns associated with the El Niño/Southern Oscillation. Mon. Wea. Rev., 115, 1606–1626. Ropelewski, C. F. and M. S. Halpert, 1989: Precipitation patterns associated with the high phase of the Southern Oscillation. J. Climate, 2, 268–284. Schiffer, R. A. and W. B. Rossow. 1985: ISCCP global radiance data set: A new resource for climate research. Bull. Amer. Meteor. Soc., 66, 1498–1505. Xie, P., J. E. Janowiak, P. A. Arkin, R. F. Adler, A. Gruber, R. Ferraro, G. J. Huffman, and S. Curtis, 2003: GPCP pentad precipitation analyses: An experimental data set based on gauge observations and satellite estimates. J. Climate, 16, 2197–2214. Weng, F., L. Zhao, R. Ferraro, G. Poe, X. Li, and N. Grody, 2003: Advanced Microwave Sounding Unit cloud and precipitation algorithms. Radio Sci., 38, 8068–8079.
24 CPC MORPHING TECHNIQUE (CMORPH) Robert J. Joyce1, John E. Janowiak1, Pingping Xie1, and Phillip A. Arkin2 1 2
NOAA/NWS/NCEP, Climate Prediction Center (CPC), Camp Springs, MD, USA Earth System Science Interdisciplinary Center, Univ. of Maryland, College Park, MD, USA
1 INTRODUCTION IR data are available globally nearly everywhere nearly all the time. However, IR channels measure cloud top temperatures and those temperatures do not always correlate well with rainfall. In many instances the cold cloud shield in a precipitating complex may be several times larger than the areal coverage of the actual precipitating region, sometimes with no rainfall directly under the coldest section. In contrast to the IR, relatively low frequency passive microwave (PMW) signals sense the thermal emission of raindrops while higher frequencies sense the scattering of upwelling radiation from the earth to space due to ice particles in the rain layer and tops of convective systems. Although rainfall estimates from PMW instruments are more accurate than those that are derived from IR data, PMW sensors are restricted to low orbit platforms and thus the temporal sampling from them is substantially less compared to geostationary IR data. Given this situation, the natural next step is to combine the data from these disparate sensors to take advantage of the strengths that each has to offer. A number of techniques have been developed in which the IR data are manipulated in a statistical fashion to mimic the behavior of PMW derived precipitation estimates. In these techniques, precipitation estimates are calculated directly from IR data through an empirical relationship between the rain rate and cloud top temperature and are used when PMW data are unavailable. An alternative method of combining these disparate data is proposed, one that uses precipitation estimates derived from low orbiter satellite PMW retrievals exclusively, and whose features are transported via spatial propagation information obtained from geostationary satellite IR data during periods when instantaneous PMW data are not available at a location. 307 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 307–317. © United States Government 2007.
308
R. J. Joyce et al.
This technique is labeled “CMORPH”, short for the Climate Prediction Center (CPC) morphing method.
2 INSTRUMENTS AND DATA 2.1 Infrared The CPC operationally extracts geostationary satellite IR brightness temperature (Tb) information (Janowiak et al. 2001) through the Man-computer Interactive Data Access System (McIDAS; Lazzara et al. 1999). CPC maps each satellite IR image to a rectilinear grid at 0.03635 degrees of latitude and longitude resolution (~4 km at the equator), parallax corrects for geometric mis-navigation of high cloud (Vicente and Scofield 1999), and corrects for cold limb effect of IR retrieval at large zenith angles (Joyce et al. 2001). IR sampling from the METEOSAT and GOES is half-hourly.
2.2 Passive microwave The PMW derived precipitation estimates that are presently used in CMORPH are generated from observations obtained from the NOAA polar orbiting operational meteorological satellites, the US Defense Meteorological Satellite Program (DMSP) satellites, and from the Tropical Rainfall Measuring Mission (TRMM; Simpson et al. 1988) satellite. The PMW instruments aboard these satellites are the Advanced Microwave Sounding Unit (AMSU-B), the Special Sensor Microwave Imager (SSM/I), and the TRMM Microwave Imager (TMI), respectively. The TMI is a nine-channel radiometer that operates at five frequencies that are quite similar to the frequencies of the SSM/I instrument. The TMI offers higher spatial resolution than SSM/I due to the relatively lower TRMM orbit. Surface rainfall derived from the TMI instrument is a product of NASA’s TRMM Science Data and Information System (TSDIS) 2A12 algorithm (Kummerow et al. 1996). The SSM/I sensors aboard the DMSP platforms are operational on the F-13, F-14 and F-15 satellites at the time of this writing. Precipitation estimates used are from the NOAA/NESDIS/ Office of Research Applications SSM/I rainfall algorithm (Ferraro 1997) which utilizes the 85 GHz vertically polarized channel to relate the scattering of upwelling radiation by precipitation-sized ice particles within the rain layer and in the tops of convective clouds to surface precipitation. The AMSU-B instrument is currently operational aboard the NOAA-15, NOAA16 and NOAA-17 polar orbiting satellites. The AMSU-B has five window channels and its cross track swath width (approximate 2200 km) contains 90 FOVs per scan. The NESDIS AMSU-B rainfall algorithm (Weng et al. 2003) performs a physical retrieval of ice water path (IWP) and particle size from the 89 and 150 GHz channels.
CPC Morphing Technique
309
2.3 Rainfall mapping The CPC globally merged IR data analyses are available at half-hourly intervals, so that time resolution was selected to produce spatially complete PMW precipitation analyses. The 0.0727 degree latitude and longitude (8 km at the equator) grid resolution used was determined by considering the spatial resolution of the various input data sources: 4 km (GOES IR), 5 km (Meteosat IR), and the greater than 13 km resolution of the AMSU-B and SSM/I derived precipitation estimates. Also, the grid must be small enough to represent the propagation of rainfall systems in half-hourly increments. The estimates are first mapped to the nearest grid point on global (60o N–60o S) rectilinear grids at 0.0727 degrees of latitude and longitude resolution, separately for each half hour and for each satellite. At locations within the grid where no rainfall estimates are available, an inverse distance squared weighting interpolation of the nearest rainfall estimates is performed to create a spatially complete field. Then for each half hour, satellite rainfall maps are combined by sensor type (TMI, SSM/I, AMSU-B) and saved to separate files. Then half-hourly rainfall from each sensor type is combined. In regions of overlap, TMI is used first, then SSM/I if no estimate from TMI is available, and finally AMSU-B. Each pixel in the half-hourly analyses is tagged with a satellite identification representing the orbiter used to produce the estimate. The NESDIS Satellite Services Division (SSD) daily Interactive Multi-sensor Snow and Ice Mapping System (IMS) product is used as the PMW rainfall snow/sea-ice screening device. Rainfall derived from a past AMSU-B algorithm (Ferraro 2000), differed in many respects from SSM/I and TMI rainfall. Revised AMSU-B rain rate scales were determined dynamically (Joyce et al. 2004) by frequency matching 8-km mapped TMI and SSM/I precipitation estimates with temporally and spatially coincident 8-km mapped AMSU-B estimates from the heaviest to lightest rain rates over the most recent ten-day period. The application of this adjustment to the AMSU-B precipitation estimates resulted in patterns and rain rate distributions resembling co-located SSM/I and TMI rainfall. This normalization procedure is a beneficial tool for reducing sensor dependent systematic bias in composite PMW rainfall maps.
3 METHODOLOGY 3.1 Propagation vector derivation The availability of global half-hourly IR data makes these data attractive to use as a means to propagate PMW derived precipitation, producing spatially and temporally complete global precipitation analyses. Since the IR data provide good measurements of cloud top properties, IR data can be used to
310
R. J. Joyce et al.
detect cloud system movement. A system known as WINDCO was developed to detect and estimate cloud motions from geostationary satellites (Smith and Phillips 1972). The first phase of the WINDCO program used an automated process that selects cloud targets that are either the coldest clouds or near regions where the IR gradient is strong (Herman 1992). Dills and Smith (1992) devised a specialized cloud relative motion tracking technique using geostationary IR and visible data. The purpose for computing cloud system advection vectors (CSAVs) for this project is to propagate PMW derived global rainfall each half hour. This requires total automation, and precludes the use of visible imagery. Figure 1. Hourly precipitation advection vector equivalence in 0.0727 latitude/longitude increments (y-axis) relative to hourly cloud system advection vectors (x-axis) over the USA during 15 May–15 June 2003. Top panel is zonal vector equivalence, bottom panel is meridional.
The direction and speed of cloud tops as detected by satellite IR may not always correlate well with the propagation of the lower precipitating layer of the system. An optimal spatial lag correlation scale would be large enough to include the sharp contrast of the cloud shield edges with the earth’s surface thus helping to focus on the motion of the entire cloud system. However, if the spatial resolution is too large, the resulting CSAV information may miss the variability of the steering currents that provide propagation of cloud system complexes. After various tests it was concluded for this work that spatially lagging overlapping 5o latitude/longitude IR regions centered at 2.5o latitude/ longitude intervals provide a good measure of the movement of entire cloud systems while capturing the bulk of variations in the steering currents. The lag tests are performed on successive IR images using iterative ~8 km pixel shift combinations in both zonal and meridional directions. If only hourly data are available, the same procedure is used except that the CSAV magnitudes are divided in two and are assumed to be the same for both halfhour periods within the hour. More details of deriving CSAVs for use in CMORPH can be found in Joyce et al. (2004).
CPC Morphing Technique
311
Early versions of CMORPH used CSAVs directly to propagate PMWderived precipitation. However, it was soon determined that the west to east and south to north advection rates were too fast in the North Hemisphere mid-latitudes (Fig. 1). To correct this, a speed adjustment procedure was developed by first computing rainfall advection vectors by spatially lagging hourly U.S. NEXRAD Stage II (Klazura 1993) radar rainfall (mapped to the same 8-km grid) in the exact same dimensions and manner CSAVs are computed from IR. The frequency distribution of hourly CSAV and radar rainfall advection rates indicated that north to south rates are quite similar but that west to east CSAV speeds were about twice as fast compared to the radar-derived vectors, and south to north rates were 3–4 times faster (Fig. 1). These systematic differences are consistent with several case studies that show the tendency of IR features to quickly stream to the northeast on the east side of long wave troughs with the actual rainfall also moving in this direction but at a slower rate. The incorporation of this adjustment procedure into the CMORPH processing has resulted in improved propagation of precipitation features. For consistency with the North Hemisphere, the meridional adjustment is applied to vectors of the opposite sign in the South Hemisphere in order to reduce the same long wave trough effect. Further tests have shown that there is scant seasonal dependence in the relationship between the IR-derived and radar derived advection vectors.
3.2 Microwave rainfall propagation and morphing The PMW rainfall propagation process begins by spatially propagating initial fields of 8-km half-hourly instantaneous PMW analysis estimates (t + 0 h) forward in time, by the discrete distance of the corresponding zonal and meridional vectors. Two auxiliary fields that are maintained along with each precipitation estimate are: (1) time stamp (t = 0 for instantaneous) in which the units represent the time, in half-hourly increments, since the scan of the PMW satellite overpass used to define that pixel and (2) satellite identification. All PMW satellite pixels (including those with zero precipitation) within each 2.5o latitude/longitude region are propagated in the same direction and distance to produce the analysis for the next half hour (t + 0.5 h). Finally, if a PMW-derived precipitation estimate from a new scan at “t + 0.5 h” is available at a particular pixel location, then that estimate overwrites the propagated estimate and the associated time stamp for that pixel is set to a value of zero. Otherwise, the time stamp is incremented by a value of “1”. This entire process is repeated each half hour. The propagation process is illustrated graphically in Fig. 2. An initial 0330 GMT time analysis of instantaneous (“t = 0 h”) PMW rainfall (Fig. 2a, leftmost plot) is propagated forward to produce analyses at “t = 0.5” and “t + 1 h” (Fig. 2a) using the IR-derived propagation vectors. This analysis is actually propagated one more time step to “t + 1.5 h” (not shown), but in this .
.
312
R. J. Joyce et al.
case all values are overwritten by precipitation estimates from an updated PMW scan (Figure 2a, rightmost plot) that became available at the “t + 1.5 h” time step (0500 GMT). The continuity of the propagated rainfall clusters in the “t + 0.5 h” and “t + 1.0 h” fields can be appreciated by comparing them with the updated PMW analysis (Fig. 2a, rightmost plot).
Figure 2. Depiction of the propagation and morphing process for a region in the South Pacific. The analyses at 0330 UTC and 0500 UTC are actual passive microwave estimates, i.e., no propagation or morphing has been applied to these data. The 0400 UTC and 0430 UTC are: (a) propagated forward in time, (b) propagated backward in time, and (c) propagated and morphed.
In addition to propagating rainfall estimates forward in time, a completely separate process is invoked in which instantaneous rainfall analyses are spatially propagated backward in time using the same propagation vectors used in the forward propagation, except for reversing the sign of those vectors. The results are stored separately from those computed in the forward propagation process. Thus for the above example, the “t = 1.5 h” updated observed PMW precipitation (Fig. 2b, rightmost plot) is propagated backwards to the “t = 0 h” time frame (Figure 2b, leftmost plot). When all propagated fields have been computed, the “t = 0 h” analysis that contains observed data overwrites the propagated estimates for that time stamp. By propagating the rainfall analyses temporally in both directions, the propagation speed and direction is improved over doing this in a single direction (in time) only.
CPC Morphing Technique
313
To this point only forward/backward propagation of initial/updated PMW derived rainfall patterns, when and where PMW data are not available, has been shown. Changes in the intensity and shape of the rainfall features are accomplished by inversely weighting both forward and backward propagated rainfall by the respective temporal distance from the initial and updated observed analyses. This process is referred here as “morphing”, is representted graphically in Fig. 2c. At each pixel location, the process by which the 0400 UTC (t + 1/2 h) estimate is produced (Fig. 2c, second plot from the left) involves creating a weighted mean as follows. The morphed value: Pmorph (t+1/2 hr) = 0.67 * Pforward (t+1/2 hr) + 0.33 * Pbackward (t+1/2 hr)
(1)
where Pforward is the PMW rainfall estimate forward propagated from initial scan (0330 UTC) and Pbackward is the PMW rainfall estimate backward propagated from updated scan (0500 UTC). Similarly, the CMORPH value for the 0430 GMT analysis is computed as: Pmorph (t+1 h) = 0.33 * Pforward (t+1 h) + 0.67 * Pbackward (t+1 h)
(2)
Each CMORPH estimate’s associated time stamp and satellite identification are extracted from the propagated estimate (forward or backward) with the smallest time stamp. For CMORPH derived from instantaneous PMW information, time stamp = 0.
4 VALIDATION The CMORPH estimates were validated using high-quality rain gauge data over the USA and Australia, and radar data over the USA. The US rain gauge information that was used in this validation is the CPC Realtime Daily Gauge Analysis (Higgins et al. 2000), which is composed of over 7000 stations. The Australian rain gauge analyses (Ebert 2002) were obtained from the Bureau of Meteorology (BOM). The “Stage II” hourly radar composites over the USA were also used to validate the CMORPH estimates. For the validation results that follow, all data sets were gridded to a common 0.25 degree latitude/longitude daily grid.
4.1 United States Comparisons are presented over the USA for the 2 April through 24 December 2003 time period. Time series of US rain gauge analyses comparisons with three indirect measurements of rainfall, statistics generated every 15 days using daily estimates, are illustrated in Fig. 3. The indirect measurements are radar, CMORPH, and “MWCOMB” which is a daily average of all available PMW derived precipitation estimates but without
314
R. J. Joyce et al.
any propagation or morphing of the features. Each data set was interpolated to match the validating gauge analysis resolution, and if a data value at a grid location was missing for any one of the three indirect estimates, it was set to missing for all, to ensure temporal and spatial matching among the data. Radar and CMORPH compare best with the gauge analyses over the USA in correlation and skill, with radar outperforming CMORPH in skill. However, the correlation is quite similar for both, CMORPH has better bias characteristics than radar over the latter half of the period. Perhaps more importantly, CMORPH compares more favorably than MWCOMB with the gauge analyses in all statistics. The difference in correlation with the rain gauge analyses between CMORPH and MWCOMB is statistically significant at the 1% level. This result illustrates that propagation and morphing have positive impacts on the PMW estimates. Figure 3. Time series of selected statistics of US rainfall estimation validation during the 2 April–24 December 2003 period, statistics generated every 15 days using daily estimates. Solid thin line = CMORPH, dashed line = MWCOMB, solid thick line = radar rainfall.
4.2 Australia A comparison during the December 2002 through late January 2004 period, statistics generated every 15 days using daily estimates of CMORPH, MWCOMB, and IR-based GPI (Arkin and Meisner 1987), with Australian rain gauge analyses is displayed in Fig. 4. The CMORPH estimates compare best with the gauge analyses overall during this period and consistently outperforms MWCOMB. The good performance of the GPI is noteworthy during Southern Hemisphere summer, however, correlation and skill of GPI decreases dramatically relative to CMORPH and MWCOMB during all other seasons.
CPC Morphing Technique
315
5 FINAL NOTES A potential shortcoming of the CMORPH method is that when precipitation forms and dissipates over a region between overpasses by PMW instrumentation, it will not be detected. Another way to alleviate this problem is to ingest more PMW information, and future CMORPH plans include incorporating AMSR-E instrument precipitation estimates when they become available. The methods that estimate precipitation from IR can expect only marginal additional improvement in the statistical relationship between IR-derived and PMWderived precipitation estimates because of more PMW data. Meanwhile, as demonstrated in Fig. 5, CMORPH stands to gain substantially with the availability of more passive PMW information. Figure 4. Time series of selected statistics of Australian rainfall estimation validation during the December 2002 through January 2004 period, statistics generated every 15 days using daily estimates. Solid thin line = CMORPH, dashed line = MWCOMB, solid thick line = GPI.
This figure shows the correlation of hourly (averaged up from half-hourly) CMORPH precipitation estimates with hourly radar rainfall over a 2-month period (April–May 2003), as a function of time from the nearest future or past PMW overpass. Note that the correlation with radar jumps from near a value of 0.40 when the most recent information is 2.5 h from PMW overpass (time step 5 on the x-axis) to about 0.55 when PMW information is only 1.5 h from scan time, which represents a 90% improvement in explained variance. The information in Fig. 5 also provides insight concerning conditions when IR-based estimates would be useful additions to the CMORPH technique. Note that correlation of GPI with radar rainfall exceeds that of CMORPH and radar rainfall after time step 6, which means that for this spring case study over the USA, GPI is a better estimate
316
R. J. Joyce et al.
than CMORPH when the nearest past or future PMW pass is more than 3 h away. CPC will examine the relative performance of CMORPH and IRbased estimates to available validation and use that information to devise a strategy on the use of IR-based rainfall in conjunction with the propagation and morphing aspects of CMORPH.
Figure 5. Correlation between hourly radar and CMORPH rainfall estimates (solid line) as a function of half hour periods from nearest future or past microwave satellite overpass (x-axis) for the period April–May 2003. For reference, the dotted line represents the correlation between hourly radar and GPI rainfall estimates over the exact same locations and times that were used in the hourly radar-CMORPH correlation calculations. Shaded plot at bottom of figure shows the number of pairs in the correlations; the scale for that plot is on the right (y-axis label in 1,000s).
Because CMORPH is flexible and can incorporate precipitation information from any algorithm based on information from any instrument, the technique is highly complementary to the proposed Global Precipitation Mission (GPM). CPC looks forward to incorporating precipitation products from GPM into the CMORPH scheme. And while GPM may provide sampling from PMW radiometers every 3 h, CMORPH can add considerable value to GPM precipitation products by melding them with IR data to increase their temporal resolution to 30 min. Finally, although the CMORPH analyses are constructed at spatial resolution of 8 km (at the equator), users are advised to use these data at a spatial resolution of 0.25o latitude/longitude or coarser. The reason for this is because the very high spatial resolution that CMORPH operates on, which is necessary for algorithmic considerations, cannot be justified as an end product since the native resolution of some of the various PMW instrument retrievals is 12 km and larger.
CPC Morphing Technique
317
6 REFERENCES Arkin, P. A. and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982–1984. Mon. Wea. Rev., 115, 51–74. Dills, P. N. and S. B. Smith, 1992: Comparison of profiler and satellite cloud tracked winds. Prepr. 6th Conf. Satellite Meteor. and Ocean., 5–10 January, Atlanta, GA, Amer. Meteor. Soc., 155–158. Ebert, E. E., 2002: Verifying satellite precipitation estimates for weather and hydrological applications. Proc. 1st Int. Precipitation Working Group (IPWG) Workshop, Madrid, Spain, 23–27 September 2002. Ferraro, R. R., 1997: SSM/I derived global rainfall estimates for climatological applications. J. Geophys Res., 102, 16,715–16,735. Ferraro, R. R., F. Weng, N. C. Grody, and L. Zhao, 2000: Precipitation characteristics over land from the NOAA-15 AMSU sensor. Geophys. Res. Lett., 27, 269–2672. Herman, L. D., 1992: Obtaining cloud motion vectors from polar orbiting satellites. Prepr. 6th Conf. Satellite Meteor. and Ocean., 5–10 January, Atlanta, GA, Amer. Meteor. Soc., 110–113. Higgins, R. W., W. Shi, E. Yarosh, and R. Joyce, 2000: Improved United States precipitation quality control system and analysis. http://www.cpc.ncep.noaa.gov/research_papers/ ncep_cpc_atlas/7/index.html Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2001: A real-time global half-hourly pixelresolution IR dataset and its applications. Bull. Amer. Meteor. Soc., 82, 205–217. Joyce, R. J., J. E. Janowiak, and G. J. Huffman, 2001: Latitudinally and seasonally dependent zenith-angle corrections for geostationary satellite IR brightness temperatures. J. Appl. Meteor., 40, 689–703. Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487–503. Klazura, G. E. and D. A. Imy, 1993: A description of the initial set of analysis products available from the NEXRAD WSR-88D system. Bull. Amer. Meteor. Soc., 74, 1293–1312. Kummerow, C., W. S. Olson, and L. Giglio. 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 1213–1232. Lazzara, M. A., J. M. Benson, R. J. Fox, D. J. Laitsch, J. P. Rueden, D. A. Santek, D. M. Wade, T. M. Whittaker, and J. T. Young, 1999: The Man computer Interactive Data Access System: 25 years of interactive processing. Bull. Amer. Meteor. Soc., 80, 271–284. Simpson, J. R., R. F. Adler, and G. R. North, 1988: A proposed Tropical Rainfall Measuring Mission (TRMM) satellite. Bull. Amer. Meteor. Soc., 69, 278–295. Smith, E. and D. Phillips, 1972: Measurements from satellite platforms, Annual Satellite Report No. NASS-11542, 1971–72, SSEC, University of Wisconsin, 1–53. Vicente, G. A., J. C. Davenport, and R. Scofield, 1999: The role of orographic and parallax corrections on real time high resolution satellite rainfall estimation. Proc. 1999 EUMETSAT Meteorological Satellite Data Users’ Conference, 6–10 September 1999, Copenhagen, Denmark, 161–168. Weng, F. W., L. Zhao, R. Ferraro, G. Pre, X. Li, and N. C. Grody, 2003: Advanced Microwave Sounding Unit (AMSU) cloud and precipitation algorithms. Radio Sci., 38(4), 8068–8079.
25 CMAP: THE CPC MERGED ANALYSIS OF PRECIPITATION Pingping Xie1, Phillip A. Arkin2, and John E. Janowiak1 1
Climate Prediction Center, NOAA/NWS/NCEP, Camp Springs, MD, USA ESSIC, University of Maryland, College Park, MD, USA
2
1 INTRODUCTION Thanks to the advent and continuous operation of satellite observations with advanced infrared (IR) and microwave (MW) instruments, significant progress has been made in the last 20 years to quantitatively document global precipitation. Together with gauge observations and precipitation fields produced by numerical models, these satellite estimates provide important information on the spatial distribution and temporal variations of precipitation, especially over the global oceans. Several intercomparisons have been conducted among various individual data sources of precipitation and the results show that all individual data sources present similar distribution patterns of overall structures of global precipitation with differences in smaller scale features and in magnitude. At least three major deficiencies exist in the individual data sources: (1) incomplete global coverage; (2) significant random error, and (3) non-negligible bias (Janowiak 1992; Xie and Arkin 1995; Adler et al. 2001; Ebert and Manton 1998). Several algorithms have been developed in recent years to produce the best possible precipitation analyses by merging these individual data sources. One such algorithm was developed by a research group at NASA/GSFC by combining satellite estimates from SSM/I, IR, and TOVS with gauge observations. The algorithm has been applied successfully to construct monthly precipitation analyses for the Global Precipitation Climatology Project (GPCP) for a 25-year period from 1979 to the present (Huffman et al. 1997; Adler et al. 2003).
319 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 319–328. © United States Government 2007.
320
P. Xie et al.
Another merging algorithm was developed by Xie and Arkin (1996), who take the gauge observations, satellite estimates from IR, OLR, SSM/I, and MSU, and the precipitation fields from the NCEP/NCAR reanalysis as inputs. Using this algorithm, analyses of global monthly, and pentad precipitation, called CPC Merged Analysis of Precipitation (CMAP, Xie and Arkin 1997a, b), have been created for the same period as the GPCP product. In this section, we will provide a brief description of the individual input data sources and the merging algorithm used to define the CMAP merged analyses of global precipitation. We will also illustrate some applications of the CMAP monthly and pentad data sets in examinations of seasonal, interannual and intraseasonal variability of large-scale precipitation.
2 METHODOLOGY The algorithm of Xie and Arkin (1996) is designed to construct global monthly precipitation analyses with complete coverage and improved quality by merging several kinds of individual data sources with different characteristics. The merging of the individual data sources is conducted in two steps. First, to reduce the random error, the satellite estimates and the model outputs are combined linearly through the maximum likelihood estimation method, in which the linear combination coefficients are inversely proportional to the squares of the local random error of the individual data sources. Over the global land areas, the individual random error is defined for each grid box and for each month by comparing the data source with the concurrent gaugebased analysis over the surrounding areas, while over global oceanic areas, it is defined by comparison with the atoll gauge data (Morrissey et al. 1995) over the Tropics and by subjective assumptions regarding the error structures over the extratropics. Since the output of the first step contains bias passed through from the individual inputs, a second step is included to remove it. For that purpose, the gauge-based analysis is combined with the output of the first step. Over land areas, the gauge data and the first-step-output are blended through the method of Reynolds (1988), in which the first-step-output and the gauge data are used to define the relative distribution (or “shape”) and the magnitude of the precipitation fields, respectively. Over the oceans, the bias remaining in the first-step-output is removed by comparison with the atoll gauge data over the Tropics and by subjective assumptions regarding the bias structure over the extratropics. In creating the monthly CMAP, seven individual data sources are used as inputs to the merging process. These are the gauge data (the GPCC gauge-
CMAP: The CPC Merged Analysis of Precipitation
321
based analyses of Schneider (1993), those of Xie et al. (1996) over land, and the atoll gauge observations of Morrissey et al. (1995) over ocean); five sets of satellite precipitation estimates derived from (1) the IR-based GPI (Arkin and Meisner 1987), (2) the SSM/I scattering-based ALG85 (Ferraro 1997), (3) the SSM/I emission-bases algorithm (Wilheit et al. 1991), (4) MSUbased method (Spencer 1993), and (5) the OLR-based OPI (Xie and Arkin 1998); and the precipitation fields produced by the NCEP/NCAR reanalysis (Kalnay et al. 1996). Two versions of the CMAP merged analyses are created. In the first version, all inputs are used to ensure complete spatial coverage (CMAP/A), while only the 6 kinds observation-based fields are utilized in the second version (CMAP/O) for applications in model verifications. The inputs and merging algorithm used to define the pentad CMAP data set are very similar to those for the monthly CMAP (Xie and Arkin 1997b). One of the major differences in defining the pentad analysis is to use the oceanic components of the SSM/I-based estimates of ALG37 (Ferraro 1997) to replace the SSM/I emission-based estimates of Wilheit et al. (1991) which are not available at pentad resolution. The CMAP merged analyses of monthly and pentad precipitation have been constructed on a 2.5o latitude/longitude grid over the globe for the period from 1979 to the present by merging the seven kinds of inputs using the algorithm described above. Figure 1 shows an example of the individual inputs and the merged analysis for January 1994. In general, all of the satellite estimates present similar large-scale patterns, characterized by rain bands associated with the ITCZ, SPCZ over Tropics, and storm tracks over the extratropics. The GPI and the OPI exhibit broader and smoother distributions of raining areas compared to those in the SSM/I-based estimates. Over land, the GPI tends to overestimate precipitation compared to the gauge-based analyses, especially over extratropical land areas. The merged analyses present spatial distribution patterns similar to those in the individual inputs, while their magnitude over land is close to that of the gauge-based analyses.
3 VERIFICATION Cross-validation tests are conducted to examine the quantitative accuracy of the individual inputs and the CMAP merged analysis. First, gauge-based analyses over 10% of randomly selected grid boxes over the global land are withdrawn. The individual satellite estimates and the reanalysis precipitation fields are then merged with the gauge-based analyses over the remaining 90% of the grid boxes. This process is repeated for 10 times so that each of the grid boxes over land is withdrawn once. The merged analyses over the
322
P. Xie et al.
Figure 1. Precipitation (mm/day) for January 1994 as observed in the satellite estimates of GPI, SSM/I scattering (SCT), SSM/I emission (EMS), OPI, MSU, the gauge-based analysis, the NCEP/NCAR reanalysis (REANAL), and the merged analysis (see also color plate 13).
withdrawn grid boxes are finally compared with the gauge-based analysis to access its quantitative accuracy. Presented in Fig. 2 are comparison results over the withdrawn land grid boxes for the monthly individual inputs and merged analyses over tropical land areas for a 3-year period from 1994 to 1996. Both the bias and the random error for the merged analyses have been reduced substantially compared to those for the individual inputs.
CMAP: The CPC Merged Analysis of Precipitation
323
Figure 2. Time series of correlation (top), bias (middle) and relative RMS error (bottom) between the withdraw gauge data and the inputs/merged analysis over tropical land areas (20o S–20o N).
4 APPLICATIONS Spatial and temporal variability of global precipitation observed in the CMAP analyses is examined for the 25-year period from 1979 to 2003. Significant seasonal variations are observed in the CMAP merged analysis (Fig. 3). During the DJF period, the ITCZ is relatively weak over the central and eastern Pacific and the SPCZ is strong. The mid-latitude storm tracks are connected with the ITCZ in the southern hemisphere, while they are more separated in the Northern Hemisphere. During the JJA period, the SPCZ is at its weakest, while the ITCZ is strong over both the eastern and western Pacific, with a slight relative minimum observed over its central part. The CMAP data set is also applied to examine the interannual variability of large-scale precipitation associated with the El Niño – Southern Oscillation (ENSO) phenomena. Shown in Fig. 4 are composite maps of differences in precipitation anomaly associated with the warm and cold ENSO episodes during the 25-year period from 1979 to 2003. Here, a simple approach is adopted to declare a cold/warm ENSO episode if the seasonal mean SST anomaly over the NINO3.4 region exceeds –0.5/0.5oC. During the DJF period, warm ENSO episodes are characterized by more precipitation over the central Pacific, southeastern South America, the extreme northeastern Pacific and adjacent coastal regions of North America, and over a belt extending from the eastern Pacific, across the Gulf of Mexico well into the Atlantic.
324
P. Xie et al.
Figure 3. DJF (top) and JJA (bottom) mean precipitation (mm/day) as obtained from the monthly CMAP for a 25-year period from 1979 to 2003.
Figure 4. DJF (top) and JJA (bottom) mean precipitation differences (mm/day) between warm and cold ENSO episodes for the 25-year period from 1979 to 2003.
CMAP: The CPC Merged Analysis of Precipitation
325
Figure 5. Spatial loading of the first EEOF mode for the bandpassed [20–80 days] components of the pentad CMAP for the DJF periods from 1979 to 2003.
326
P. Xie et al.
Less than normal precipitation, meanwhile, is observed over the western Pacific, the central Pacific away from the Tropics, Amazonia, and South Africa. During the JJA period, there is more precipitation all the way across the Pacific and less precipitation over most of the southeast Asian monsoon area during warm episodes. The storm track over the northwestern Pacific is further north than during the cold episodes, bringing more precipitation over Japan and its adjacent oceanic areas during warm episodes. One of the major components of the global climate, the intraseasonal variability has long been examined using pentad averages of the OLR observed by the NOAA satellites. While the OLR data are thought to be a good index of tropical convection and a reasonable proxy for latent heat release over the Tropics, the CMAP pentad precipitation data set is preferable for its direct and quantitative relation to latent heating. As an illustration of the applications of the pentad CMAP data set in this respect, a bandpass process is first performed for the original pentad CMAP to extract the components of intraseasonal time scales (20–80 days). Extended EOF analysis is then conducted for the bandpassed components to define the dominant structure of large-scale precipitation associated with the evolution of Madden-Julian Oscillation (MJO). Shown in Fig. 5 are the spatial distribution of precipitation variability associated with the leading extended EOF (EEOF) mode for the DJF period. First, a weak positive precipitation anomaly appears over the central and western Indian Ocean, while suppressed precipitation is observed over the maritime continent area and its vicinity (–20 days). As it propagates eastward, the positive precipitation anomaly intensifies and its extension widens (–10 days). It then reaches its maximum at day 0 when the enhanced precipitation is over the Maritime Continent. Upon passing the landmass, the anomaly weakens as it moves towards southeast (day 10–20).
5 SUMMARY Analyses of global monthly and pentad precipitation have been constructed on a 2.5o latitude/longitude grid over the globe for a 25-year period from 1979 to 2003 by merging seven kinds of individual data sources with different characteristics. These include the gauge-based analyses from the GPCC and Xie et al. (1996), estimates inferred from a variety of satellite observations, and the precipitation fields produced by the NCEP/NCAR reanalysis. Called the CPC Merged Analysis of Precipitation (CMAP), the 25-year data sets provide global monthly and pentad precipitation distributions with full coverage and improved quality compared to the individual data sources. The monthly and pentad CMAP data sets are applied to examine the seasonal, interannual, and intraseasonal variations of large-scale precipitation over the globe. The distributions of the seasonal mean precipitation resemble those
CMAP: The CPC Merged Analysis of Precipitation
327
observed in the published long-term means but with major differences over oceans. The interannual variability associated with the ENSO is similar to that found by previous studies based on gauge data but with much more coherent details over the oceans. The evolution patterns of large-scale precipitation anomaly associated with the Madden-Julian Oscillation (MJO) are well depicted by the pentad CMAP data set.
6 REFERENCES Adler, R. F., C. Kidd, G. Petty, M. Morrissey, and H. M. Goodman, 2001: Intercomparison of global precipitation products: The third Precipitation Intercomparison Project (PIP-3). Bull. Amer. Meteor. Soc., 82, 1377–1396. Adler, R. F., G. J. Huffman, A. Chang, R. Ferraro, P. Xie, J. Janowiak, B. Rudolf, U. Schneider, S. Curtis, D. Bolvin, A. Gruber, J. Susskind, and P. Arkin, 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979 - present). J. Hydrometeor., 4, 1147–1167. Arkin, P. A. and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the Western Hemisphere during 1982–84. Mon. Wea. Rev., 115, 51–74. Ebert, E. E. and M. J. Manton, 1998: Performance of satellite rainfall estimation algorithms during TOGA COARE. J. Atmos. Sci., 55, 1537–1557. Ferraro, R., 1997: Special Sensor Microwave Imager derived global rainfall estimates for climatological applications. J. Geophys. Res., 102, 16715–16735. Huffman, G. J., R. F. Adler, P. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 5–20. Janowiak, J. E., 1992: Tropical rainfall: A comparison of satellite derived rainfall estimates with model precipitation forecasts, climatologies, and observations. Mon. Wea. Rev., 120, 448–462. Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Woollen, Y. Zhu, A. Leetmaa, B. Reynolds, M. Chelliah, W. Ebisuzaki, W. Higgins, J. Janowiak, K. C. Mo, C. Ropelewski, J. Wang, R. Jenne, and D. Joseph. 1996: The NCEP/NCAR 40-year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471. Morrissey, M. L., M. A. Shafer, S. E. Postawko, and B. Gibson, 1995: Pacific raingauge data. Water Resour. Res., 31, 2111–2113. Reynolds, R. W., 1988: A real-time global sea surface temperature analysis. J. Climate, 1, 75–86. Schneider, U., 1993: The GPCC quality control system for gauge-measured precipitation data. GEWEX Workshop on Analysis Methods of Precipitation on Global Scale, Rep. WCRP81, WMP/TD-588, Koblenz, Germany, WMO, A5–A9. Spencer, R. W., 1993: Global oceanic precipitation from MSU during 1979–91 and comparisons to other climatologies. J. Climate, 6, 1,301–1,326. Wilheit, T. J., A. T. C. Chang, and L. S. Chiu, 1991: Retrieval of the monthly rainfall indices from microwave radiometric measurements using probability distribution functions. J. Atmos. Oceanic Technol., 8, 118–136. Xie, P. and P. A. Arkin, 1995: An intercomparison of gauge observations and satellite estimates of monthly precipitation. J. Appl. Meteor., 34, 1143–1160. Xie, P. and P. A. Arkin, 1996: Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model prediction. J. Climate, 9, 840–858.
328
P. Xie et al.
Xie, P. and P. A. Arkin, 1997a: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2,539–2,558. Xie, P. and P. A. Arkin, 1997b: Global pentad precipitation analysis based on gauge observations, satellite estimates and numerical model outputs. Amer. Geophy. Union 1997 Fall Meeting, San Francisco, CA, Amer. Geophy. Union. Xie, P. and P. A. Arkin, 1998: Global monthly precipitation estimates from satellite-observed outgoing longwave radiation. J. Climate, 11, 137–164. Xie, P., B. Rudolf, U. Schneider, and P. A. Arkin, 1996: Gauge-based monthly analysis of global land precipitation from 1971–1994. J. Geophy. Res., 101(D14), 19023–19034.
26 RAINFALL ESTIMATION USING A CLOUD PATCH CLASSIFICATION MAP Kuo-Lin Hsu, Yang Hong, and Soroosh Sorooshian Department of Civil and Environmental Engineering, University of California, Irvine, CA, USA
1 INTRODUCTION The development of meteorological satellite systems has been innovating precipitation observations beyond traditional means and enabling frequent observation of precipitation distribution over the remote territories and the broad oceanic regions. In recent years, satellite rainfall estimates from algorithms using geostationary satellite (GOES) sensors and combined GOES and Polar Operational Environmental Satellites (POES) have been rapidly evolving to a certain degree that are suitable as a supplement of the ground in situ observations in providing precipitation information in the hydrological applications. Many precipitation algorithms developed using GOES satellites capable of providing high spatial (4 km) and temporal (15 min) resolution images are considered a unique source for the observation short-term extreme precipitation event (Scofield and Kuligowski 2002). Although the GOES images are frequently used to measure cloud motion at high resolution, the long-wave infrared image does not provide sufficient information to infer the actual rainfall at the ground surface. Experiments show that the corresponding pixel infrared cloud top brightness temperature and surface rainfall rate is not unique. Therefore, from the local pixel based mapping, a fixed temperature and rainfall rate (Tb-R) function is not capable of fitting surface rainfall rates well at all times. To improve the quality of estimates, strategies have been developed using multiple GOES channels, and adjustment or merging rainfall estimates from multiples sources (Adler et al. 1994; Ba and Gruber 2001; Bellerby et al. 2000; Fulton et al. 1998; Hsu et al. 1997; 1999; Huffman et al. 1997; Huffman et al. 2001; Levizzani et al. 2002; Scofield 1987; Sorooshian et al. 2000; Tapiador et al. 2002; Turk et al. 1998; Vicente et al. 1998; Xie and Arkin 1997; Xu et al. 1999). 329 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 329–342. © 2007 Springer.
330
K.-L. Hsu et al.
The cloud-top temperature and rainfall relationship is varied with respect to the cloud types and their environmental conditions. Patch-based approaches view the patch coverage as a unit; precipitation under the patch coverage is assigned based on the cloud types, where heavy rainfalls are assigned to the convective active regions and lighter or no rain is assigned to stratiform and cirrus cloud regions. Figure 1 shows the evolution stage of a convective storm. Higher rainfall intensities are usually found during the towering to mature stages, while lower or no-rain appears during the dissipating stage. From the towering stage to the mature stages, cloud top’s pixel temperatures near the convective core grew colder, while rainfalls intensified. In the dissipating stage, on the other hand, cloud top temperatures were cold, but had very mild or no corresponding rain.
Figure 1. Evolution stages of a convective storm and its rainfall distribution curves.
Figure 2. Single fitting curve models vs. a multiple fitting curve model.
Figure 2a shows the scatterplot of Tb-R relationship from a set of GOES image and radar rainfall. The data points are widely spread, which cannot be fitted by a single function (see Fig. 2b). We propose a cloud patch-based rainfall allocation. Cloud patch is assigned a specific Tb-R curve according
Rainfall Estimation Using a Cloud Patch Classification Map
331
to their coldness, size, and texture features. Figure 2c shows 400 Tb-R curves are assigned to the fitting of the scatter points in Fig. 2a. In this study, a patch-based cloud classification and rainfall estimation algorithm is introduced. This algorithm processes raw satellite image data into the pixel rainfall by series of stages. Details of the algorithm development are described in the following sections.
2 CLOUD PATCH CLASSIFICATION AND RAINFALL ESTIMATION Humans analyze the world surrounding them by observing and interpreting the patterns that occur in time sequence. Through the distinct features of the patterns shown, knowledge is extracted by classification information into a number of categories and subcategories. In the naming of the cloud system, for example, we classify cloud types as convective and stratus clouds based on their visual appearance in the atmosphere. Convective clouds are puffy and vertically piled up, whereas the stratus are flat and layered, sometimes in fibrous forms. Fine classification of cloud systems is mainly based on the cloud system in different altitude levels, as cumulus, altocumulus, and cumulonimbus for the convective clouds, and stratocumulus, altostratus, cirrostratus, and cirrus for the stratus clouds.
Figure 3. A cloud patch classification and rainfall estimation system.
Observation of the mechanisms of the cloud systems suggests that different types of clouds contain different kinds of thermal ascending and cooling structures and form various distributions of water content. Layer clouds have relatively uniform distribution of water content, while convective clouds have highly variable distribution of water content. Based on their thermal structure, water contents and distribution, different clouds types give diversified precipitation distributions over the ground surface. Although cloud systems
332
K.-L. Hsu et al.
are different in nature, such as high and low, layer or convective clouds, they are discernible with the human eye. However, the implementation of an automatic procedure to detect the cloud types and their associated precipitation from a computer model is not an easy task. It requires great effort to process a huge amount of satellite cloud images and build a cloud patch pattern recognition system which eventually can be used as a tool in providing estimation of precipitation amount and distribution of a rain cloud. Using computer image processing and pattern recognition techniques, we developed a patch-based cloud classification and rainfall estimation system based on the satellite infrared images. Illustration of the classification system is depicted in Fig. 3 and further descriptions of each stage are as follows.
2.1 Cloud patch segmentation Segmentation of image data is an important problem for the computer vision, remote sensing, and image analysis. It can be considered a preprocess step before description and recognition of objects. Cloud segmentation is operated through a process that may eventually divide the image into separable patches which are strongly related with cloud systems of the real world contained in the image. A simple approach to separate patch objects from the background image is by applying a constant brightness threshold. The approach separates the gray image into two levels, high and low. It is simple and makes computation easy. Although a single threshold, 253 K for example, seems to work well in the separation of cloud patch from clear sky or no-precipitating regions, within the patch coverage, it still contains several cloud systems existing at different altitudes with various thermal structures and sizes. Without further separation of those cloud systems from a warm threshold, those distinctive cloud systems are mixed together. As a result the precipitation distribution inside the cloud patch cannot be estimated accurately. For a better separation of local cloud systems, a watershed-based (topography) segmentation approach is proposed (Vincent and Soille 1991; Dobrin et al. 1994). The algorithm starts from finding the altitude local minima (Fig. 4a), and then follows to fill the basins from the bottom (see Fig. 4b). The water continues to fill all basins. When two basins would merge from the rising the water level, a reservoir is set to separate them (Fig. 4c). While water level continues to rise, individual basins are formed. The process stops when a designed water table is reached (Fig. 4d). Figure 5 shows the cloud image segmentation using a constant (253 K) threshold and watershed-based separation by gradually increasing threshold temperature from 210 K to 253 K. The source infrared image is listed in Fig. 5a. The constant threshold T253K used to separate pixels under the cloud coverage is listed in Fig. 5b. It shows that several cloud systems are presented in the patch (see Fig. 5a), but a single threshold is not capable of dividing them
Rainfall Estimation Using a Cloud Patch Classification Map
333
into separated patch objects. By applying watershed-based segmentation (Fig. 5c–f), “basins” are filled and separated gradually. Eventually five cloud patches under the threshold temperature of 253 K are identified (see Fig. 5f).
Figure 4. Watershed-based segmentation approach.
Figure 5. Cloud image separation using watershed-based segmentation approach.
2.2 Patch feature extraction After the cloud image is separated into a number of objects (patches), it is required to represent the object scenes with a series of attributes or features. Features that are related to the status of cloud patches in the atmosphere, such as the cloud height (lowest temperature), cloud size and shape, surface textures, surface gradients are extracted for the later patch classification. Basically, those selected patch features are separated into three categories: coldness, geometry, and texture (see Table 1). In addition, all the features listed in Table 1 are extracted from three temperature threshold levels (220 K, 235 K, 253 K). Convective clouds, for example, tend to show puffy and piled up features, whereas stratiform clouds are flat and layered and sometime fibrous in their appearance. Cloud patch features extracted from three separated temperature levels, at 220 K, 235 K, and 253 K demonstrate the existence of the cloud patches at different altitudes in the atmosphere. Figure 6 shows two sets of patch features as extracted from two adjacent cloud patches, denoted as index j and k. After image segmentation, clouds are separated into distinguishable patches. The convective cloud patch j is the mature stage with extensive vertical growth and overshooting top, while convective cloud patch k is in the towering stage, where cloud top brightness temperature is higher than 220 K K and therefore the feature vectors at V220 K are not available (void). Those three
K.-L. Hsu et al.
334
temperature thresholds give a broad classification of cloud patches existing in different altitudes. In addition to the size and texture features, fine classification of cloud patch status and their association with the precipitation amount and distribution in the patch coverage may be investigated. Table 1. Input features extracted from cloud patches. Coldness Features of Cloud Patch 1. Minimum temperature of a cloud patch (Tmin) 2. Mean temperature of a cloud Patch (Tmean) Geometric Features 3. Cloud patch area (AREA) 4. Cloud Patch Shape Index (SI) Texture Features 5. 6.
Standard deviation of cloud patch temperature (STD) Mean value of local (5×5 pixels) standard deviation of cloud temperature ( MSTD 5 x 5 )
7.
Standard deviation of local (5×5 pixels) standard deviation of cloud patch ( STD STD : 5 x 5 )
8.
Gradient of cloud cold top brightness temperature (TOPG)
9.
Gray image texture (Maximum Angular Second Moment)
2.3 Cloud patch classification Classification includes concepts such as categorization, identification, recognition, clustering, and partitioning. Supervised and unsupervised learning techniques are often addressed for this purpose. In this study, we use an unsupervised clustering analysis to classify patch samples into a number of cloud patch groups. Clustering is proceeded based on the similarities of patches measured in the their feature space. The Self-Organizing Feature Map (SOFM) clustering algorithm is used for this purpose (see Fig. 7) (Kohonen 1995; Hsu et al. 1999). The output layer of SOFM is a two-dimensional array of units (or clusters/groups), which are connected to their neighbor units and to the input features. A set of adjustable parameters, called weights, is assigned to the connections between the input features and output units. By sequentially assigning training patterns, the connection weights are adjusted and finally stabilized, i.e., the responses of the output units become ordered. As a result, the similar input features are assigned to the same output unit.
Rainfall Estimation Using a Cloud Patch Classification Map
335
Figure 6. Cloud Patch Feature extraction.
Figure 7. Classification using Self-Organizing Feature Map.
A brief description of the training procedure is summarized below; a detailed description of the training procedure can be found in Kohonen (1995) and Hydkin (1994): 1. Assign random numbers to the weights wij , where i is the index of input feature and j is the index of output units. K K 2. Collect a set of cloud patch samples and normalized as: V ( p) = [ V ( p) , 220 K K K V235 K ( p ) , V253 K ( p ) ], p=1...N, where N is the number of training samples. K 3. Select a normalized patch sample V ( m), m ∈[1, N ] and determine the
output unit that has minimized distance between input and connection weights: K j* = arg min w j − V ( m) j
(1)
K.-L. Hsu et al.
336
4. For all the units cj within a neighborhood of radius r of cj*, perform the weight update with step size (learning rate) α > 0: wij = wij + α (v i ( m) − wij )
(2)
5. Terminate if the wij is converged, or reduce r and α, and go to step 3.
2.4 Patch rainfall estimation Up to the current stage, only a satellite cloud image is required in the image segmentation and cloud patch classification. The next stage is to fit a specific rainfall distribution to each classified cloud patch group. To assign rainfall in the classified patch group, ground radar and low-orbit satellites microwave observations are likely to provide precipitation measurements. For the reliable representation of the classified cloud patch groups and their associated rainfall statistics, samples need to be collected across different seasons and geolocations. After the patch classification is completed through SOFM training. Assume that a given cloud patch sample is assigned to patch group, cj, in the SOFM layer. The concurrent cloud temperatures and precipitation pixels under the cloud patch coverage are sampled out and assigned to the patch group cj. Allocation of pixel rainfall and temperature to the patch groups continues until all the patch samples and rainfall observations are processed. With a sufficient number of samples allocated, a large amount of pixel rainfalls and temperatures are assigned to each of the patch groups in the SOFM layer. The next step was to assign the GOES long-wave infrared temperature (Tb) and hourly radar rainfall rate (R) relationship in each classified patch group. The Probability Matching Method (PMM) (Atlas et al. 1990) was used to match the relationship between the GOES long-wave infrared temperature and hourly radar rainfall rate. A rainfall probability density function (PDF) is calculated for each patch group in the SOFM layer based on the collected Tb and R samples. It is assumed that lower Tb pixels are associated with higher rain rates. With rainfall rates and infrared brightness temperatures consists of a same accumulated probability matched below: 1 − F (Tb* ) =
∫
∞ Tb*
f (Tb ) dTb =
∫
R* 0
f ( R) dR = F ( R * )
(3)
where f (Tb ) and f (R) are the PDFs of brightness temperature and rainfall rate; whereas F (Tb* ) and F ( R * ) are accumulated probability distribution functions of cloud top brightness temperature and radar rainfall, respectively.
Rainfall Estimation Using a Cloud Patch Classification Map
337
In the final step Tb and R data are binned and averaged based on a small temperature increment (∆Tb) in each patch group. The Tb-R relationship is fitted by a nonlinear function below: k
R k = v k 1 + v k 2 ⋅ exp[v k 3 ⋅ (Tb + v k 4 ) v 5 ]
(4)
where R is the rainfall rate (mm h–1), Tb is the cloud top brightness temperature (K), and vk1, vk2, vk3, vk4, and vk5 are parameters with respect to patch group k of SOFM layer parameters [vk1, vk2, vk3, vk4, vk5] are found from an optimization scheme (Duan et al. 1992).
3 EXPERIMENTS One month (June 1999) of GOES and stage IV radar data (NCEP) covering the continental USA were collected and processed to the pixel resolution of 0.04o × 0.04o latitude/longitude scale. The size of the classification groups was a set of 20 × 20 output units in the SOFM layer (see Fig. 8a). The rational for the selection of the classification group size is that the higher the number of units is assigned to the SOFM layer, the finer the classification of cloud patch samples may be obtained. The input feature map presents the contour map of the SOFM weight matrix, wIj . After SOFM is trained, wIj is the sample mean of an input feature I assigned to the cluster j. Here only three features are listed (see Fig. 8b–d). They show the mean sample value of input patch feature (Tmin, AREA235K, and MSTD5x5 at 235K) for the 20 × 20 classification units in the SOFM layer. It is worth noting the clustering process has organized the Tmin (see Fig. 8b) in a manner that the higher temperatures appear in the upper region of the map around 220 K to 240 K from the left to the right hand side, while the lower temperatures appear in the lower region of the map range from 220 K to 200 K, from left-to-right locations. This means a patch sample with Tmin > 230 K will be assigned to the group near the upper-right-hand corner of the SOFM layer. Likewise, a patch sample with Tmin < 210 K will be assigned to the lower-right-hand corner of the SOFM layer. The average pixel rain rate map (see Fig. 8f) also visually shows its relevance to the Tmin feature. In general the rain rate map follows the Tmin feature map, where the lower-half SOFM units appear to contain higher average pixel rain rate. The AREA235K feature is the patch size below the 235 K threshold. The circled regions of Fig. 8c are associated with large cloud patches. The corresponding patch rainfall volume (see Fig. 8e) is relatively higher in those circled spots, especially on the one where Tmin is lower than 205 K.
K.-L. Hsu et al.
338
Figure 8d shows the feature map of the average of local standard deviation of 5 × 5 pixels at 235 K-threshold (MSTD5x5 at 235K). This feature map represents texture variation in a cloud patch. The circled regions are denoted with higher values of MSTD5x5 at 235K , which apparently consist of highly average pixel rain rate (see Fig. 8f). In addition, those high MSTD5x5 at 235K regions are spread from warm to cold cloud patch group (see Fig. 8b). The calibrated Tb-R curves with respect to the 20 × 20 output units of SOFM layer are listed in Fig. 9. With reference to the Tmin shown in Fig. 8b, one could identify those Tb-R curves being associated with patch minimum temperature ranged from 240 K to 200 K. Accompanying with the input feature maps of AREA235K (Fig. 8c) and MSTD5x5 at 235K (Fig. 8d), we may further explain the Tb-R curves and their relevance to the various cloud top temperatures (Tmin), patch sizes (AREA235K), and patch texture variations (MSTD5x5 at 235K).
Figure 8. Input feature maps (see also color plate 12).
In Fig. 9b, several regions, denoted as G0-G6, containing special Tb-R curve groups, are located. Cloud groups on G0 are no-rain warm cloud; cloud groups on G1 and G2 are around the same Tb range, but associated with two different slopes of Tb-R curves. Both G1 and G2 contain Tmin around 230 K and higher, however, a cloud patch classified in G1 region has a lower MSTD5x5 at 235K than G2 region. Cloud groups on G3 and G4 also represent cloud patch groups with Tmin around 215 K and above. G4 group has much higher local texture variations
Rainfall Estimation Using a Cloud Patch Classification Map
339
(MSTD5x5 at 235K ) than G3 group. As a result, the slopes of Tb-R curves are steeper on group G4. Likewise, the slopes of Tb-R curves are steeper on G6 region than on G5 region. Although both regions represent patches Tmin around 200 K, the MSTD5x5 at 235K map (Fig. 8d) reveals that the higher MSTD value, the steeper Tb-R curves.
Figure 9. Cloud patch groups (G0 to G6) contain special Tb-R curves.
Figure 10 shows the lifetime of a convective cloud system and its Tb-R curves. The time period from 1440 UTC to 1900 UTC is the towering to mature stages of the cloud patch, while after 1900 UTC, the cloud patch is in the recession and decay stage. The cloud Tb distribution in the evolution stage of cloud patch is listed in Fig. 10a, while the corresponding sorted Tb-R distribution is displayed in Fig. 10b (the dot plots are the averaged radar rainfall binned at 1 K Tb interval. The line plots are the fitting curves based on the regression functions). They show that the Tb-R curves progressively varied during the evolution stage of a convective storm. One can imagine that the estimation of rainfall rates may not perform well, if only one fixed Tb-R function is assigned to all of the stages. Figure 11 shows hourly rainfall estimates over east New Mexico area during a consecutive 6-h time period, from 00UTC to 05UTC of July 4, 2002. Precipitation observation from NCEP stage IV radar estimates is listed in the upper panel, while the rainfall estimates from cloud patch classification system is listed in the lower panel. The image is processed at 0.04o × 0.04o latitude/longitude scale. It shows that rainfall regions are well matched from both radar and model’s estimates.
K.-L. Hsu et al.
340
Figure 10. The evolution process of a convective cloud and its Tb-R curves.
Figure 11. Rainfall estimates during 6-h period.
Rainfall Estimation Using a Cloud Patch Classification Map
341
4 CONCLUSIONS This paper describes a cloud patch classification approach applied to the surface rainfall estimation. This approach implements image processing and pattern classification techniques to the analysis of long-wave infrared (10.7 µm) cloud images of GOES satellites. In low-level image processing, segmentation using a watershed-based scheme is used to separate cloud patches from their image background. This is followed by image extraction and interpretation where cloud patches are treated as independent objects and are described by object features such as patch coldness, size, shape, and texture. Classification of cloud patch objects is based on an unsupervised clustering scheme. When the patch rainfall is assigned to the classified patch group, the interpretation of the cloud patch property and rainfall relationships are established. Finally, the rainfall distribution of the classified patch group is described by a set of infrared brightness temperature and rainfall rate (Tb-R) functions. Parameters of the nonlinear Tb-R function are calibrated from the spatial and temporal co-located satellite image and radar rainfall map. Multiple seasons of rainfall estimates were generated and evaluated using ground gauge and radar data. Details of the evaluation performance can be found from Hong et al. (2004). For the potential extension of precipitation estimates over the oceans and remote regions, we are exploring using TRMM satellite rainfall measurements in the calibration of model parameters. Evaluation of those results will be discussed in a separate report. Acknowledgements: Support for this research was provided by NASA’s EOS, TRMM, and HyDIS and NSF STC-SAHRA research programs. Data used in this study is provided from NOAA Climate Prediction Center and National Center for Environmental Prediction. It is a pleasure to acknowledge Dan Braithwaite for his processing the data required for this study and Ms. Diane Hohnbaum for her editing of the manuscript.
5 REFERENCES Adler, R. F., G. J. Huffman, and P. R. Keehn, 1994: Global rain estimates from microwave adjusted geosynchronous IR data. Remote Sens. Rev., 11, 125–152. Atlas, D., D. Rosenfeld, and D. B. Wolff, 1990: Climatologically tuned reflectivity-rainrate relationship and links to area-time integrals. J. Appl. Meteor., 29, 1120–1135. Ba, M. B. and A. Gruber, 2001: GOES multiple spectral rainfall algorithm (GMSRA). J. Appl. Meteor., 40, 1500–1541. Bellerby, T., M. Todd, D. Kniveton, and C. Kidd, 2000: Rainfall estimation from a combination of TRMM Precipitation Radar and GOES multispectral satellite imagery through the use of an artificial neural network. J. Appl. Meteor., 39, 2115–2128.
342
K.-L. Hsu et al.
Dobrin, B. P., T. Viero, and M. Gabbouj, 1994: Fast watershed algorithms: analysis and extensions. Nonlinear Image Processing V., 2180, 209–220. Duan, Q., S. Sorooshian, and V. K. Gupta, 1992: Effective and efficient global optimization for conceptual rainfall-runoff model. Water Resource Res., 28(4), 1015–1031. Fulton, R. A, J. P. Breidenbach, D. J. Seo, and D. A. Miller, 1998: The WSD-88D rainfall algorithm. Wea. Forecasting, 13, 377–395. Hong, Y., K. Hsu, X. Gao, and S. Sorooshian, 2004: Precipitation estimation from remotely sensed information using an artificial neural network-cloud classification system. J. Appl. Meteor., 43, 1834–1853. Hsu, K, X. Gao, S. Sorooshian, and H. V. Gupta, 1997: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36, 1176–1190. Hsu, K., H. V. Gupta, X. Gao, and S. Sorooshian, 1999: Estimation of physical variables from multichannel remotely sensed imagery using a neural networks: application to rainfall estimation. Water Resource Res., 35(5), 1605–1618. Huffman, G. J., R. F. Adler, P. A. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolph, and U. Schneider, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 5–20. Huffman, G. J., R. F. Adler, M. M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 36–50. Hykin, S., 1994: Neural Networks: A Comprehensive Foundation. Macmillan College Pub. Kohonen, T., 1995: Self-Organizing Map. Springer-Verlag. Levizzani, V., J. Schmetz, H. J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2000: Precipitation estimations from geostationay orbit and prospects for METEOSAT Second Generation. Meteor. Appl., 8, 23–41. [NCEP] National Center for Environmental Prediction, National Stage IV QPE Product. http://www.emc.ncep.noaa.gov/mmb/ylin/pcpanl/stage4. Scofield, R. A., 1987: The NESDIS operational convective precipitation technique. Mon. Wea. Rev., 115, 1773–1792. Scofield, R. A. and R. J. Kuligowski, 2002: Status and outlook of operational satellite precipitation algorithms for extreme precipitation events. Proc. 1st Workshop Int. Precipitation Working Group; September 23–27 2002; Madrid, Spain, 43–51. Sorooshian, S., K. Hsu, X. Gao, H. V. Gupta, B. Imam, and D. Braithwaite, 2000: An evaluation of PERSIANN system satellite-based estimations of tropical rainfall. Bull. Amer. Meteor. Soc., 81(9), 2035–2046. Tapiador, F. J., C. Kidd, V. Levizzani, and F. S. Marzano, 2002: A neural network PMW/IR combined procedure for short term/small area rainfall estimates. Proc. 1st Workshop Int. Precipitation Working Group, September 23–27 2002, Madrid, Spain, 167–173. Turk, F. J., F. S. Marzano, and E. A. Smith, 1998: Combining geostationary and SSM/I data for rapid rain rate estimation and accumulation. Proc. 9th Conference on Satellite Meteorology and Oceanography, Paris, 462–465. Vicente, G., and R. A. Scofield, 1998: The operational GOES infrared rainfall estimation technique. Bull. Amer. Meteor. Soc., 79, 1883–1898. Vincent, L. and P. Soille, 1991: Watersheds in digital spaces: an efficient algorithm based on immersion simulations.” IEEE Trans. Pattern Ana. Machine Intell., 13(6), 583–598. Xie, P. and P. A. Arkin, 1997: Global precipitation: A 17 years observation based on gauge observations, satellite estimates and numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539–2558,. Xu, L., X. Gao, S. Sorooshian, P. A. Arkin, and B. Imam, 1999: A microwave-infrared threshold technique to improve the GOES Precipitation Index. J. Appl. Meteor., 38, 569–579.
COLOR PLATES
Plate 1. Time longitude sections at 5º N–5º S, January 1997–October 1998; GPCP Version 2 monthly, pentad and daily data. (Figure 3 of Gruber et al., Section 1) Plate 2. Air pollution decreases the drop sizes of convective clouds over the British Isles. This NOAAAVHRR image from 18 April 1995, 1337 UT was analyzed by the scheme of Rosenfeld and Lensky (1998), showing convective rain clouds with large drops (re > 20 µm, well exceeding the 14 µm precipitation threshold) in the northwesterly flow from the Atlantic Ocean. The clouds interact with the air pollution over the populated land areas and become composed of small drops (re < 10 µm, too small for precipitating) that appear in yellow shades. Note that the sharp distinction of the clouds around the latitude of Glasgow. Northern Scotland is sparsely populated and hence the clouds remain pristine with large drops, as indicated by the red shades. (Figure 3 of Rosenfeld, Section 1) Plate 3. MSG image from 20 May 2003 1342 UTC, over central Africa at a 1200 × 1200 km2 rectangle between 1–12N and 15–26E. The area shows the transition between the relatively microphysically maritime clouds over the forested area (dark surface) and microphysically continental clouds over the dry lands of the Sahel to the north (bright surface). The T-re relations of the continental clouds (1) show much smaller re for a given T compared to the maritime clouds (2). The median re of the maritime clouds (the yellow line) saturates near T = –20°C, indicating glaciation at that temperature. The small median re at area 1 even above the –40°C isotherm indicates homogeneous glaciation of the cloud water and hence low precipitation efficiency. The color scheme is red for the visible, green for 3.9 µm reflectance component, and blue for temperature. For full description and interpretation of the color table is given in Rosenfeld and Lensky (1998). The T-re lines represent percentiles of re for a given T in 10% steps for each line, between 5% and 95%. The median is between the yellow and green lines. (Figure 4 of Rosenfeld, Section 2) Plate 4. Sample of the diffusion rainfall estimation method of Tapiador et al. compared with an IR-based procedure and the actual radar data. (Figure 6 of Tapiador et al., Section 2) Plate 5. Mean rainrate (RR) in the fore/after TB plane at 19.3H GHz (left panels) and 85.5H GHz (right panels). Top and bottom panels correspond to the simulations of the Goddard Cumulus Ensemble (GCE) TCOF22 and MIDACF. (Figure 4 of Battaglia et al., Section 2) Plate 6. (a) Retrieved conditional rainfall; (b) probability of rain; (c) retrieved conditional rain for probability of rain greater than 50%; (d) uncertainty of rain [%]. Rain retrieval algorithm of Kummerow et al. (Figure 7 of Kummerow et al., Section 3) Plate 7. 24-h rainfall potential (inches) (right) derived from SSM/I instantaneous rain rates (left) on 1510 UTC, 18 September 2003 for Hurricane Isabel. (Figure 1 of Ferraro, Section 3) Plate 8. (a) Fraction of precipitation events with snowfall at the surface over the northern Atlantic for October 2002; (b) total number of precipitation events found per 0.5 × 0.5 degree box; (c) same as (a) but for January 2003; (d) same as (b) but for January 2003. (Figure 4 of Bennartz, Section 3) Plate 9. Mean surface rain rates, convective rain proportions, and latent heating rates at 7 and 3 km altitude, derived from TMI observations from January 2000, using the GPROF algorithm. (Figure 5 of Olson et al., Section 3) Plate 10. Final combined HQ+VAR precipitation field computed with the Real-Time MPA for 1500 UTCon 15 September 2004. Zero values are colored white if computed by the HQ and gray if by VAR; all other precipitation values have the same color value for both sources. (Figure 3 of Huffman et al., Section 4)
CP 1
CP 2
Color Plates
Plate 11. Cumulated rainfall (mm h–1) from the MICRA algorithm on November 1–17, 1999, within the Meteosat sector. (Figure 2 of Marzano et al., Section 4) Plate 12. Input feature maps of the rainfall estimation technique based on cloud patch classification maps of Hsu et al. (Figure 8 of Hsu et al., Section 4) Plate 13. Precipitation (mm day–1) for January 1994 as observed in the satellite estimates of GPI, SSM/I scattering (SCT), SSM/I emission (EMS), OPI, MSU, the gauge-based analysis, the NCEP/NCAR reanalysis (REANAL), and the merged analysis. (Figure 1 of Xie et al., Section 4) Plate 14. 10 November 2001, 0300 UTC; upper left: BOLAM model; upper right: combined IR-MW NRL Turk algorithm estimate; bottom: IR NRE (neural rain estimator) estimate. (Figure 5 of Kästner, Section 5) Plate 15. 12-h accumulated precipitation (mm h–1), 1200 UTC, 10 Nov. 2001 (Algiers flood): estimated from satellite (upper left), reference forecast (R, upper right), and for the assimilation run (N, bottom) of the BOLAM model. (Figure 5 and 6 of Buzzi and Davolio, Section 6) Plate 16. Analysis increments of TCWV (in kg m-2) on April 7, 2003, 0000 UTC from all observations (a) and rain observations (b). (c) 48 h-24 h precipitation forecast (in 10-3 mm) initialized on April 1, 2003, 1200 UTC with rain observations. (Figure 5 of Bauer et al., Section 6) Plate 17. 6 August 2002. Rain intensity maps in mm h–1 for the Emilia-Romagna storm case study. All times are UTC. (a) Radar map at 0012; (b) NRLT for the slot starting at 0000; (c) Radar map at 0042; (d) NRLT for the slot starting at 0030; (e) NRLT for the slot starting at 0630; (f) PMW NESDIS algorithm for the SSM/I orbit 08D (F13) starting at 0627; (g) NRLT for the slot starting at 0700; (h) Radar map at 0642; (i) NRLT for the slot starting at 0800; (j) PMW NESDIS algorithm for the SSM/I orbit 10D (F14) starting at 0828; (k) NRLT for the slot starting at 0800; (l) Radar map at 0842; (m) NRLT for the slot starting at 0930; (n) PMW NESDIS algorithm for the SSM/I orbit 12D (F15) starting at 0941; (o) NRLT for the slot starting at 1000; (p) Radar map at 0942. (Figure 1 of Torricella et al., Section 7) Plate 18. January 1-10 (decade 1) 2003 rainfall accumulation estimate in mm (top), and maize yield projection (% yield potential) based on estimated rainfall and an empirical equation (bottom). White spaces denote zero yield potential. (Figure 7 and 8 of Liu et al., Section 7) Plate 19. MM5 and discrete ordinate Jacobian simulations at 424.763 ± 4 GHz of Hurricane Bonnie at landfall (courtesy of A. J. Gasiewski). (Figure 8 of Bizzarri et al., Section 8) Plate 20. (a) CWCs of six hydrometeor species [cloud droplets (top left), rain drops (top middle), graupel (top right), pristine crystals (bottom left), snowflakes (bottom middle), and aggregates (bottom right)] for inner grid of UW-NMS snowstorm simulation at time step 1800 s (i.e., 0600 UTC, 25 January 2000). Vertical lines indicate path of vertical cross-section in Plate 18b. (b) Synthetic snow IWC retrievals for selected cross-section of eastern U.S. snowstorm simulation. Left panels from top to bottom are: (1) EGPM radiometer retrieval using only four lower window frequencies, i.e. 18.7, 23.8, 36.5 and 89 GHz (similar to four SSM/I frequencies), (2) as previously but using all five window frequencies including 150 GHz, (3) as previously but using all five window frequencies plus four pairs of sounding channels in 50–54 and 118 GHz regions, (4) combined EGPM radar-radiometer retrievals, and for comparison (5) model “truth”. For each retrieval case, associated right panels show estimated average profile (blue solid line) together with model “truth” average profile (red line), plus retrieval error standard deviations at various atmospheric levels (blue error bars). (Figure 3 and 7 of Mugnai et al., Section 8) Plate 21. Distributions of monthly rainfall accumulation over global tropics for Feb 1998 produced by most recent version (V6) of standard TRMM L2 / L3 algorithms: (1) top panel shows TMI-only (L2 alg 2a12) (see Kummerow et al. 1996, 2001; Olson et al. 2001, 2007); (2) 2nd from top panel shows PR-only (L2 alg 2a25) (see Iguchi et al. 2000; Meneghini et al. 2000); (3) 3rd from top panel shows TMI-PR Combined (L2 alg 2b31) (see Haddad et al. 1997; Smith et al. 1997); and (4) bottom panel shows TMIonly (L3 alg 3a11) (see Wilheit et al. 1991; Hong et al. 1997; Tesmer and Wilheit 1998). Color bar denotes average rainrate in mm day–1. (Figure 3 of Smith et al., Section 8)
Color Plates
CP 3
Plate 1
Plate 2
Color Plates
CP 4
Plate 3
Plate 4
Color Plates
CP 5
Plate 5
Plate 6
Color Plates
CP 6
(a)
(b) Plate 7
Plate 8
Color Plates
CP 7
Plate 9
Color Plates
CP 8
Plate 10
Plate 11
Plate 12
Color Plates
CP 9
Plate 13
Color Plates
CP 10
Plate 14
Plate 15
Color Plates
CP 11
Plate 16
Color Plates
CP 12
a)
b)
c)
d)
e)
f)
g)
h)
i)
j)
k)
l)
m)
n)
o)
p)
Plate 17
Color Plates
CP 13
Plate 18
Color Plates
CP 14
Plate 19
Plate 20a
Color Plates
CP 15
Plate 20b
Color Plates
CP 16
Plate 21
Section 5 Validating Satellite Rainfall Measurements
27 METHODS FOR VERIFYING SATELLITE PRECIPITATION ESTIMATES Elizabeth E. Ebert Bureau of Meteorology, Melbourne, Australia
1 INTRODUCTION Satellite precipitation estimates are widely used to measure global rainfall on monthly timescales for climate studies (e.g., Huffman et al. 1997). Near realtime satellite precipitation estimates are becoming increasingly available to the wider community. These precipitation estimates are potentially very useful for applications such as numerical weather prediction (NWP) data assimilation, now-casting and flash flood warning, tropical rainfall potential, and water resources monitoring, to name a few. As with any observational data, it is important to understand their accuracy and limitations. This is done by verifying the satellite estimates against independent data from rain gauges and radars. The terms validate and verify are used here to mean the same thing, “to determine or test the truth or accuracy of”. The former is preferred by the satellite community, while the latter is preferred by the NWP modeling community. Standard measures such as bias, correlation, and root mean square error (RMSE) error are very useful in quantifying the errors in climate-scale satellite precipitation estimates. Users of near real-time precipitation estimates often require more specific information on expected errors in rain location, type, mean and maximum intensities. Diagnostic validation approaches can reveal additional information about the nature of the errors. This paper discusses validation methods that give useful information to (a) help algorithm developers to improve their products, and (b) help users of satellite precipitation estimates to understand the accuracy and limitations of those products.
345 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 345–356. © 2007 Springer.
346
E. E. Ebert
2 VALIDATION OF SHORT-TERM VERSUS CLIMATE-SCALE PRECIPITATION A great deal of work has been done to validate climate-scale precipitation estimates against gauge data over land and for island stations (Adler et al. 2001). The quantity of interest in this case is mean rain amount integrated over space and time (e.g., 2.5° monthly rainfall), and the statistics used to measure the accuracy of the estimates are usually the bias, correlation coefficient, and RMSE. For climate-scale rainfall the requirement is that the algorithm provides a good estimate on average, and so errors on shorter time and space scales are unimportant. It matters little, for example, that the single thresholdbased GOES Precipitation Index (GPI) erroneously associates rain with cirrus clouds and fails to detect rain from warm clouds because these errors cancel out over large space and timescales. Users of short-term precipitation estimates generally cannot afford to tolerate such errors. Hydrologists need accurate estimates of rain volume at the catchment scale. In NWP data assimilation of satellite rainfall, experience suggests that detecting the correct rain location and type may be more important than getting the correct amount (e.g., Xiao et al. 2000). For flash flood warning and tropical rainfall potential, it is not only important to correctly detect the occurrence of rainfall, it is also important to be able to estimate the maximum rain rates (e.g., Kidder et al. 2001). The methods used to validate near real-time precipitation estimates must give additional information not provided by the simpler scores. Validating shortterm satellite rainfall estimates is akin to verifying quantitative precipitation forecasts (QPFs) from NWP models. Indeed, some of the methods described here were developed for QPF verification.
3 VALIDATION DATA The main sources of rainfall data for satellite algorithm validation are observations from rain gauges, radar rainfall estimates, and objective analyses of one or both types of observations. Each has distinct advantages and disadvantages. Rain gauges are the only instruments that give direct measurements of rain accumulation. However, because they are point measurements, they are likely to be unrepresentative of the areal value estimated by the satellite. The timescale is generally also unrepresentative, i.e., a gauge accumulation over several minutes to several hours versus a satellite “snapshot”. Averaging over a long period of time is required to remove the representativeness errors. If satellite estimates are verified directly against individual gauge observations that are irregularly distributed in space (e.g., having higher sampling density in more populous regions), then the results will be biased toward the better-
Methods for Verifying Satellite Precipitation Estimates
347
sampled regions. Rain-gauge observations are limited to land regions, islands, and atolls, and therefore have limited usefulness for verifying oceanic satellite precipitation estimates. Radar observations are similar to satellite estimates in that they give “snapshots” in time and areal values in space. They have the advantage of high spatial and temporal resolution (~1 km and 5–10 min), which means that some of the random error will be averaged out when scaling up to the (usually) lower satellite resolution. The disadvantage with using radar data for verification is that they are themselves indirect estimates of rainfall, and are prone to errors of various kinds (Collier 1996). Careful quality control of the data and bias correction using nearby rain-gauge observations can correct most of the error in radar rainfall estimates. Like gauges, they are confined to land and near-coastal regions. Objective analyses of rain gauges and/or radar data combine the observations onto a spatial grid. The merging of data within grid boxes reduces some of the random noise and regularizes the spatial distribution. When used to validate satellite precipitation estimates, the satellite and analysis data must be mapped onto a common grid. This may result in reduced spatial detail, including reduced maximum rain rates. It is important to note that the quantitative verification results depend on the chosen grid scale, with coarser grids generally producing more favorable results. The most appropriate choice of validation data and grid depends on the availability of the data, the resolution of the satellite estimates, and the needs of the user. For instantaneous and high temporal and spatial resolution estimates, gauge-corrected radar estimates or analyses are generally preferable to gauge observations. At slightly larger space and timescales (6 h to daily), rain-gauge analyses or combined gauge/radar analyses are more accurate and should be used in preference to raw gauge or radar observations. For pentad (5-day), monthly, and longer timescales, rain-gauge analyses are usually accurate enough to provide good validation data. Measurement and sampling errors in the observations increase the apparent error of the satellite estimates. This effect is difficult to remove (Habib and Krajewski 2002) and is often neglected in rainfall verification. We can tolerate this as long as the observational error is random (unbiased) and is much smaller than the error in the satellite estimates. Such imperfect validation data can be reliably used to intercompare estimates from different algorithms.
4 STANDARD VERIFICATION METHODS Although subjective in nature, simple visual comparison of mapped estimates and observations (“eyeball” verification) is one of the most effective verification methods. The human mind has an acute ability to discern and interpret differences that scientists are still struggling to objectify in a meaningful way.
E. E. Ebert
348
Most standard objective verification statistics give quantitative measures of accuracy or skill. Some of the most commonly used statistics will be presented briefly here; excellent discussions can be found in the textbooks of Wilks (1995) and Jolliffe and Stephenson (2003), or the “Forecast Verification – Issues, Methods and FAQ” web page (2004). The advantages of the standard statistics are (a) they are familiar and easily understood by most people, (b) they are simple to compute, and (c) they are useful for monitoring and intercomparing the performance of different algorithms. Each statistic gives only one piece of information about the error, and so it is necessary to examine a number of statistics in combination in order to get a more complete picture.
4.1 Continuous verification statistics Continuous verification statistics measure the accuracy of a continuous variable such as rain amount or intensity. These are the most commonly used statistics in the validation of satellite estimates. In the equations to follow Yi indicates the estimated value at point or grid box i, Oi indicates the observed value, and N is the number of samples. The mean error (bias) measures the average difference between the estimated and observed values. The mean absolute error (MAE) measures the average magnitude of the error. The RMSE also measures the average error magnitude but gives greater weight to the larger errors. Mean error = MAE =
1 N
RMSE =
1 N
N
∑ (Y − O ) i
i =1
i
N
∑ Y −O i =1
1 N
i
(1)
i
N
∑ (Y − O ) i =1
i
2
i
Sometimes the errors are stratified by rain range, contingent either on the observations or the estimates, as shown in Fig. 1. This box and whiskers plot shows the inter-quartile range for each bin sample (middle 50% of the values) as boxes, the full range as a vertical line, and the median value as a horizontal line. This type of verification information is useful for providing error estimates to accompany the rainfall estimates. The correlation coefficient r measures the degree of linear association between the estimated and observed distributions. It is independent of absolute or conditional bias, however, and therefore must be used along with other measures when verifying satellite estimates.
Methods for Verifying Satellite Precipitation Estimates
349
N
r=
∑ (Y − Y )(O − O ) i =1
N
i
i
N
∑ (Y − Y ) ∑ (O − O ) i =1
2
i
i =1
(2) 2
i
Figure 1. Distribution of mean absolute errors as a function of estimated (top) and observed (bottom) rainfall magnitude for two competing algorithms.
Any of the error statistics can be used to construct a skill score that measures the degree of improvement over a reference estimate. The most frequently used scores are the MAE and the mean squared error. The reference estimate is usually climatology or persistence (the most recent set of observations), but it could also be an estimate from another algorithm. Skill score =
scoreestimate − score reference score perfect − score reference
(3)
4.2 Categorical verification statistics Categorical verification statistics measure the correspondence between the estimated and observed occurrence of events. Most are based on a 2 × 2 contingency table of yes/no events, such as rain/no rain, shown in Table 1. The elements in the table (hits, misses, etc.) give the joint distribution of
E. E. Ebert
350
events, while the elements below and to the right (observed yes, observed no, etc.) are called the marginal distributions. The bias score gives the ratio of the estimated rain area (frequency) to the observed rain area (frequency), regardless of how well the rain patterns correspond with each other. The probability of detection (POD) measures the fraction of observed events that were correctly diagnosed, and is sometimes called the “hit rate”. The false alarm ratio (FAR) gives the fraction of diagnosed events that were actually nonevents. hits + false alarms hits + misses hits POD = hits + misses false alarms FAR = hits + false alarms BIAS =
(4)
Perfect values for these scores are BIAS = 1, POD = 1, and FAR = 0. The POD and FAR should always be used together. The threat score (TS), also known as the critical success index, measures the fraction of all events estimated and/or observed that were correctly diagnosed. Since this score is naturally higher in wet regimes, a modified version known as the equitable threat score (ETS) was formulated to account for the hits that would occur purely due to random chance. The ETS, though not a true skill score, is often interpreted that way since it has a value of 1 for perfect correspondence and 0 for no skill. It penalizes misses and false alarms equally, and for this reason it is commonly used in NWP QPF verification. hits hits + misses + false alarms hits − hits random ETS = hits + misses + false alarms − hits random
TS =
hits random
1 = (observed yes × estimated yes) N
(5)
To measure algorithm performance for different rain rates it is useful to plot the categorical scores as a function of an increasing rain threshold, as shown in Fig. 2. The bias plot (Fig. 2a) shows that the algorithm underestimated the frequency of light rain and overestimated the frequency of heavy rain. The POD (Fig. 2b) in the tropics fell between 0.6 and 0.7 (60–70% of observed
Methods for Verifying Satellite Precipitation Estimates
351
Figure 2. Categorical verification of summertime (DJF) daily satellite precipitation estimates from the GPCP 1DD algorithm (Huffman et al. 2001) over Australian mid-latitudes (stars) and tropics (crosses) during 1997–2003.
rain was detected) for all rainfall except the heaviest, while values of about 0.5 were achieved in mid-latitudes. The algorithm had particular difficulty with false alarms in mid-latitudes (Fig. 2c), where 50% of all rain detections above 5 mm per day were incorrect, and 75% of rain detections above 20 mm per day were incorrect. The ETS (Fig. 2d) shows that the maximum detection skill was achieved for rain exceeding 2–5 mm per day in both midlatitudes and tropics, with values of 0.28 and 0.30, respectively. Table 1. 2 × 2 contingency table. The off-diagonal elements characterize the errors. Observed Yes Estimated
Yes No
Hits Misses Observed yes
No False alarms Correct negatives Observed no
Estimated yes Estimated no N = total
5 DIAGNOSTIC VERIFICATION METHODS Diagnostic, or “scientific”, verification methods provide greater detail about the nature of the errors than do the standard statistics. These methods consider distributions of values, rather than simply matched pairs of independent estimates and observations. Because diagnostic methods are
352
E. E. Ebert
often more complex than standard methods, they are more likely to be used in research than in operations. The familiar scatter plot of observed versus estimated values is frequently used to show the degree of correspondence between the two, and whether the errors show any systematic behavior. By binning the values into K categories, a numerical scatter plot can be constructed from the K × K contingency table (e.g., Brooks and Doswell 1996). This provides more information about the nature of the estimation errors than does the simple yes/no table, including the relative frequency of small versus large errors. The marginal distribution of the estimates can be plotted as a histogram and compared to that of the observations to check for bias and skewness in the distribution. For short-period estimates, we are frequently interested in how well a satellite estimate represents the spatial pattern of rainfall. Two classes of spatial verification techniques are described next, scale decomposition methods and entity-based methods.
5.1 Scale decomposition methods Scale decomposition methods measure the correspondence of the estimates and observations at different spatial scales. The goal is to discover which scales are best represented, and which scales are poorly represented by the estimates. Ideally, the algorithm developer would then put some effort into improving the algorithm at those poorly represented scales; if this is not possible or practical then it may be desirable to filter those scales out of the final product. Implicit in this strategy is the assumption that the estimates and observations are available at high spatial resolution. Early scale decomposition methods used 2D Fourier transforms. In recent years discrete wavelet decomposition has received greater attention, as it has some attractive properties. Wavelets are locally defined functions characterized by a location and a spatial scale, and do not have periodic properties as does Fourier decomposition. Briggs and Levine (1997) were among the first to use wavelets as a verification tool for forecasts of 500 hPa geopotential height fields. Casati et al. (2004) have developed a new verification method that first removes the bias component of the estimated field by histogram-matching recalibration against the observations, then generates binary rain images by thresholding. A wavelet decomposition is performed on the (binary) error field to obtain the structure of the errors as a function of scale. Categorical verification statistics are then computed as a function of scale to measure the contribution of each scale to the total error.
Methods for Verifying Satellite Precipitation Estimates
353
Figure 3. (a) Hourly radar rainfall and (b) power law rain-rate estimate (Vicente et al. 1998) frontal rainfall at 2230 UTC on 16 September 2002 in southeastern Australia. The verification shows (c) the threat score for rain ≥ 0.2 mm h–1, (d) conditional rain intensity, and (e) spatial variability as a function of scale.
In the upscaling approach proposed by Zepeda-Arce et al. (2000), verification scores and spatial variability measures are recomputed as the estimates are averaged over increasingly larger grid boxes. The rate of improvement with scale is a measure of the quality of the estimate. The motivation is to give credit to high-resolution estimates that look good in a broad sense, but do not match the observations very well in a point-to-point sense. An example of the upscaling approach is shown in Fig. 3. The satellite estimate gives rain rates that are similar to those observed by the radar, but the satellite rain extends too far in the southeast direction, possibly a result of misdiagnosed cirrus. The TS improves with increasing scale, but achieves a perfect value of 1 only at the coarsest scale. The conditional rain rates are much higher for the satellite estimate than for the radar. However, the spatial variability of the satellite and radar estimates are nearly identical for all but the largest scales.
5.2 Entity-based methods An alternative approach is the evaluation of rain entities, or contiguous rain areas (CRAs) (“blobs” on a rainfall map), in which the bulk properties of the satellite estimated rain entity are verified against the bulk properties of the observed rain entity (Ebert and McBride 2000). The verification quantities of interest include the position error, the difference between the estimated and
E. E. Ebert
354
Figure 4. Schematic diagram of a contiguous rain area (CRA), with the arrows showing the optimum shift of the satellite rain patterns.
observed rain area, volume, and mean and maximum rain rates, and the correlation between the position-corrected estimate and observations. The approach is shown schematically in Fig. 4. Entities are defined by a threshold isohyet, which may be assigned a very low value to capture all rainfall or a high value if heavy rainfall is the focus. The position error is determined (to the nearest grid point) by pattern matching, where the estimate is horizontally translated over the observations until a best fit criterion is satisfied. Possible best-fit criteria include minimization of the total-squared error, maximization of the spatial correlation coefficient, or maximum overlap; the correlation criterion appears to give the most satisfactory fit for satellite precipitation estimates. The disadvantage of entity-based verification is that if the estimate does not resemble the observations sufficiently, it may not be possible to reliably associate two entities objectively. a)
b)
c)
Displacement (km)
k
Observed
Observed
k
Figure 5. Verification of (a) rain volume, (b) maximum rain rate, and (c) frequency of location errors, for 289 CRAs from the NRL experimental geostationary algorithm (Turk et al. 2004) during April 2001–March 2002.
CRA verification lends itself well to characterizing systematic errors when a large number of rain systems are verified. In the example shown in Fig. 5, the algorithm produced good estimates of rain volume within systems,
Methods for Verifying Satellite Precipitation Estimates
355
although it had a tendency to overestimate their intensity. There appears to be a very slight location bias of estimates toward the south.
6 FINAL REMARKS Users of satellite precipitation estimates need to understand the expected accuracy and error associated with the estimates. Satellite algorithm developers need to validate their algorithms in order to make improvements. The standard continuous and categorical verification statistics give quantitative measures of the accuracy of the satellite-estimated rain amount and occurrence. There is no one score that summarizes the ability of an algorithm to make correct rainfall estimates. Rather, it is necessary to examine several scores together in order to characterize both the skill and the errors. Diagnostic verification methods give more detailed information on the nature of the errors, and are therefore appealing to algorithm developers. Examples of several standard and diagnostic verification methods applied to satellite rainfall estimates can be accessed from the International Precipitation Working Group’s web site (2004). One of the major problems in validating satellite precipitation estimates is the imperfect “truth” data that is available for doing the validation. Although we cannot expect the situation to improve enormously in the next few decades, polarimetric radar technology and improved gauge–radar analysis methods will lead to better quality validation data. Methods must be developed and put in place to quantify the uncertainty in the validation results that arise from uncertainty in the observations.
7 REFERENCES Adler, R. F., C. Kidd, G. Petty, M. Morissey, and H. M. Goodman, 2001: Intercomparison of global precipitation products: The Third Precipitation Intercomparison Project (PIP–3). Bull. Amer. Meteor. Soc., 82, 1377–1396. Briggs, W. M. and R. A. Levine, 1997: Wavelets and field forecast verification. Mon. Wea. Rev., 125, 1329–1341. Brooks, H. E. and C. A. Doswell III, 1996: A comparison of measures-oriented and distributions-oriented approaches to forecast verification. Wea. Forecasting, 11, 288–303. Casati, B., G. Ross, and D. B. Stephenson, 2004: A new intensity-scale approach for the verification of spatial precipitation forecasts. Meteorol. Appl., 11, 141–154. Collier, C. G., 1996: Applications of Weather Radar Systems: A Guide to Uses of Radar Data in Meteorology and Hydrology. Wiley, Chichester, 390 pp. Ebert, E. E. and J. L. McBride, 2000: Verification of precipitation in weather systems: determination of systematic errors. J. Hydrol., 239, 179–202. Forecast Verification – Issues, Methods and FAQ, 2004: http://www.bom.gov.au/bmrc/ wefor/staff/eee/verif/verif_web_page.html Huffman, G. J., R. F. Adler, P. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 5–20.
356
E. E. Ebert
Habib, E. and W. F. Krajewski, 2002: Uncertainty analysis of the TRMM ground-validation radar-rainfall products: application to the TEFLUN-B field campaign. J. Appl. Meteor., 41, 558–572. Huffman, G. J, R. F. Adler, M. M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2000: Global precipitation at one-degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 36–50. International Precipitation Working Group, 2004: http://www.isac.cnr.it/~ipwg/ Jolliffe, I. T. and D. B. Stephenson (eds.), 2003: Forecast Verification. A Practitioner’s Guide in Atmospheric Science. Wiley, Chichester, 240 pp. Kidder, S. Q., J. A. Knaff, and S. J. Kusselson, 2001: Using AMSU data to forecast precipitation from landfalling hurricanes. Symposium on Precipitation Extremes: Prediction, Impacts, and Reponses, Amer. Meteor. Soc., Albuquerque, NM, 14–19 January 2001, 344–347. Turk, F. J., E. Ebert, B.-J. Sohn, V. Levizzani, E. Smith, and R. Ferraro, 2004: Over-land validation of a global operational multi-satellite precipitation analysis at variable space and time scales. J. Appl. Meteor., submitted. Vicente, G. A., R. A. Scofield, and W. P. Menzel, 1998: The operational GOES infrared rainfall estimation technique. Bull. Amer. Meteor. Soc., 79, 1883–1898. Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. An Introduction. Academic Press, San Diego, 467 pp. Xiao, Q., X. Zou, and Y.-H. Kuo, 2000: Incorporating the SSM/I-derived precipitable water and rainfall rate into a numerical model: a case study for the ERICA IOP-4 cyclone. Mon. Wea. Rev., 128, 87–108. Zepeda-Arce, J., E. Foufoula-Georgiou, and K. K. Droegemeier, 2000: Space-time rainfall organization and its role in validating quantitative precipitation forecasts. J. Geophys. Res., 105 (D8), 10129–10146.
28 ASSESSMENT OF SATELLITE RAIN RETRIEVAL ERROR PROPAGATION IN THE PREDICTION OF LAND SURFACE HYDROLOGIC VARIABLES Emmanouil N. Anagnostou Civil and Environmental Engineering, University of Connecticut, Storrs, CT, USA
1 INTRODUCTION Precipitation is probably the most important component of a mixture of hydrologic cycle parameters (precipitation, soil moisture, evapotranspiration, runoff, etc.), mostly accountable for shaping the climatic state of water in the earth stores, its variability, and climatic trends. Understanding the interactions and feedback between precipitation and land surface processes is key to advancing the predictability of the water cycle at scales varying from mesoscale to regional and climate. Unfortunately, precipitation simulated by global climate models is not sufficiently accurate for directly forcing surface energy and water budgets in coupled land–atmosphere data assimilation systems. This uncertainty has an effect on the study of interactions between land surface process and atmospheric convection. A perceived solution is the use of radiation budget and precipitation estimates from ground and/or space-based sensors to force off-line land data assimilation systems (LDAS). In a recent report, Mitchell et al. (1999) emphasized on the viability of the land surface alternative to coupled systems. This approach uses uncoupled physically based LDAS systems (Koster and Suarez 1996; Liang et al. 1996; Mitchell et al. 2000) driven by surface forcing anchored by independent observation-based precipitation and radiation budget fields. A critical aspect associated with such an approach is the need to resolve the precipitation variability over large regions with high temporal (of the order of 1–3 h) and spatial (0.05° × 0.25°) resolution. Precipitation estimates over large regions (in particular over the tropical convective zones) are primarily 357 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 357–368. © 2007 Springer.
358
E. N. Anagnostou
available from microwave (MW) and infrared (IR) sensor observations onboard earth-orbiting and geostationary satellite platforms, respectively. Overland MW channels measure the emission and scattering by precipitating ice and hydrometers, which is physically related to the water content, and indirectly to surface rainfall rate. The IR channel measures the thermal radiation emitted by clouds (or the surface) that is related to the temperature of the emitting media (cloud tops). As a consequence, cold clouds are assumed to have large vertical extent and be associated with precipitation beneath. Although, MW channels are associated with more definitive precipitation estimates than IR, observations from earth-orbiting satellites carrying MW sensors offer intermittent coverage (currently this is about six observations per day counting all available satellite platforms). Consequently, there has been significant effort to bridge those gaps estimating precipitation from proxy parameters inferred from the half-hourly geostationary IR observations (e.g., Anagnostou et al. 1999; Adler et al. 2000). Improvement of geosynchronous satellite precipitation retrievals through regional calibration by the use of MW rain estimates, and by combining estimates from multiple sensors is currently approached in a number of studies (Hsu et al. 1999; Todd et al. 2001; Huffman et al. 2003). A common observation from these studies is that MW-calibrated IR-based rain retrievals are limited in terms of their ability to sufficiently characterize the surface rainfall variability. This is due to the weak physical relationship of IR measurement to convective rain processes. For example, Todd et al. (2001) showed 0.3 correlations and 350% standard error at high spatial (0.1 degree) and temporal (hourly) resolution in comparison with radar data. At coarser scale (1-degree daily) the same data comparison demonstrated a reduction of standard error to 90%. To improve upon this limitation recent studies have combined hourly IR rain retrievals with the infrequent passive MW estimates using data fusion approaches (see papers in Section 4 of this book). It is not well understood, though, how to optimally combine rain products of different resolutions, sampling frequencies, and retrieval error characteristics. For example, which is better representative for a 3-hourly/0.5-degree grid rain estimate: an infrequent MW retrieval that is closer related to precipitation or a 6-hourly rain average from frequent IR observations but indirectly related to rainfall? A merging of estimates that is followed by spatial and temporal aggregation would most likely correspond to the best estimate. Given, though, the large influence that sampling and retrieval error characteristics have on the merged product, a definitive answer about data fusion cannot be obtained without proper error investigation studies. The impact of satellite rain estimation error on the simulation of hydrological variables (else known as error propagation) has not been well investigated. Hossain and Anagnostou (2004) and Hossain et al. (2004)
Assessment of Satellite Rain Retrieval Error Propagation
359
recently studied the adequacy of current satellite rain retrievals in terms of flood prediction for medium-sized watersheds. An earlier study by Nijssen and Lettenmaier (2003) examined the effect of precipitation error due to satellite sampling alone on the simulation of runoff and other hydrological variables derived from a macroscale hydrological model. Although, these studies offered valuable information on the hydrological error characteristics of satellite retrievals they are based on stochastic error models with limited representation of the true estimation error characteristics. There is a clear need to further our understanding on this issue on the basis of experimental investigations that would use actual satellite rain retrieval data and hydrologic models. In this paper, we present such an experimental framework aimed at evaluating hydrologic relevant rain retrieval merging techniques. The framework includes a physically based land surface model (LSM) implemented in a data-rich region, and uses satellite rain retrievals from MW and IR observations. The selected region for this experiment is an area covered by the Oklahoma hydrometeorological station network (MESONET). The area is under the coverage of the National Weather Service Weather Surveillance Radar (WSR-88D) network, and of an array of MW satellites (TRMM, SSM/I). The community land surface model (CLSM) (Bonan et al. 2002) was used in this study. It was forced by radiation and meteorological (winds, temperature, and relative humidity) field data measured by the Oklahoma MESONET. High-resolution (hourly at 4 × 4 km2) rain gauge-calibrated radar (WSR-88D) rainfall fields were used as ground reference of precipitation (Fulton et al. 1998). This experimental framework was the basis for assessing two satellite rain retrieval schemes and evaluating their error structure in the prediction of hydrological and other land surface parameters. We will argue that limiting the evaluation of rain retrievals at the level of rain rate estimation error is not sufficient to identify retrieval techniques that are optimal in terms of other hydrological variables (e.g., runoff, soil moisture). The propagation of rain estimation error characteristics to the various hydrologic variables is complex and highly nonlinear, which calls for integrated error studies in rain estimation land surface modeling. In the subsequent sections we present a short description of CLM followed by the study area and data. We will then describe the satellite rain retrievals and present the simulation experiment where we evaluated the error statistics in the estimated rainfall and land surface hydrological variables. We close with a summary of our findings, and suggestions on the research direction for developing hydrologic relevant satellite rain retrievals.
360
E. N. Anagnostou
2 THE LAND SURFACE MODEL The CLM is devised in this study to simulate land surface parameters. It is a column (one-dimensional [1D]) model developed on the basis of the Bonan (1996) LSM, which was modified to incorporate the best features of the Dickson et al. (1986) Biosphere–Atmosphere Transfer scheme and the Dai et al. (1997) IAP94 LSM. CLM has one vegetation layer as in most LSMs, ten unevenly spaced vertical soil layers with variable hydraulic conductivity, and up to five snow layers depending on the total snow depth. The surface grid cell can be subdivided into a number of tiles that consist of a single type of land cover. The model performs the water and energy balance calculations over each tile at every time step. Most surface processes such as evaporation from the ground, transpiration from the plants root zone, soil and snow water propagation, leaf temperature and fluxes, soil and snow temperature, and phase change are parameterized through physical–empirical equations. Nevertheless, the parameterization of runoff-related processes in soil–vegetation– atmosphere interaction is not well formulated, which is perhaps one of the weak points of CLM. The runoff parameterization is based on TOPMODEL (Beven and Kirkby 1979) concept, while river routing model is not included. With externally forcing data (precipitation, radiation, wind speed, air temperature, and humidity) CLM computes a number of prognostic variables that include runoff, soil moisture, temperature in the soil layer, water intercepted on the canopy, leaf temperature, latent heat flux, and sensible heat flux.
3 STUDY AREA AND DATA The study region covers two 1-degree grid areas in Oklahoma associated with distinct vegetation cover. The center of the lowly vegetated 1-degree site (hereafter named Lveg) is at 34.6861N latitude and –99.8339W longitude; and it includes five MESONET stations. The dominant vegetation type is non-arctic grass (type 13 in CLM), while the soil sand and clay percentages are 47% and 19%, respectively. The average value of leaf area index (LAI) for this site in June is 1.3. The center of the highly vegetated 1-degree site (hereafter named Hveg) is 35.4360N latitude and –94.7740W longitude; it also includes five MESONET stations. The dominant vegetation type for Hveg is broadleaf deciduous temperate tree (type 7 in CLM), while the soil sand and clay percentages are 54% and 23%, respectively. The average value of LAI for this site in June is 5.3. Two types of data are needed to run CLM: land surface data, and hydrometeorological forcing data. The land surface data that include vegetation cover type, vegetation fraction, monthly leaf and stem area index, canopy top and bottom heights, and soil texture and color are available in CLM global
Assessment of Satellite Rain Retrieval Error Propagation
361
database and considered along with the model parameterizations as true representations of the land surface processes. Those data are available at 1-km spatial resolution and were mostly derived from space-borne remote-sensing observations such as the IGBP DISCover data set (Loveland et al. 2000), University of Maryland tree cover data set (DeFries et al. 2000a,b), advanced Very High Resolution Radiometer (AVHRR), and the IGBP soil data set (Global soil Data task, 2000). The hydrometeorological data include gridaveraged values of (1) meteorological measurements from the MESONET stations (Brock et al. 1995); and (2) rainfall rates from WSR-88D. The radar rainfall fields are extracted from the National Radar Rainfall Mosaic available at the Hydrologic Rainfall Analysis Product (HRAP) ~4-km resolution (Fulton et al. 1998). These estimates have been corrected for a number of radar error sources including mean field bias based on a scheme that uses hourly areaaveraged rain gauge-to-radar rainfall ratios (Seo et al. 1999, 2000). The WSR88D rain estimates are used here as the reference ground rainfall data set, against which we will assess the rainfall fields derived from the satellite retrievals. The experimental data span a period from 1 January to 30 September 2002.
4 SATELLITE RAINFALL ESTIMATES Two satellite rain retrieval schemes are evaluated in this study. The first is based on MW-calibrated IR rain estimates, while the second on a merging of those estimates with instantaneous precipitation fields derived from TRMM Microwave Imager and three Special Sensor Microwave/Imager (F13, F14, and F-15) passive MW observations on the basis of Grecu and Anagnostou (2001) overland algorithm. The IR rain retrieval is part of a variable rainfall product (VAR) array produced in real time at NASA Goddard Space Flight Center by Huffman et al. (2003), and is referred to as 3B41RT product (ftp://aeolus.nascom.nasa.gov). The VAR retrieval incorporates the NOAA Climate Prediction Center half-hourly Global IR composites (Janowiak et al. 2000) aggregated to 0.25-degree spatial grid resolution and hourly accumulations. The technique uses coincidental IR observations with MW rain retrievals from TMI and SSM/I observations to dynamically (on a month-tomonth basis) calibrate the IR rain algorithm at discrete 5-degree grid areas. The calibration is based on matching the probability density histograms of IR brightness temperatures and MW rain rates falling within a common data set. The second technique is a merging MW/IR scheme. The scheme identifies for every hourly IR rainfall value at 0.25-degree pixel resolution all available instantaneous MW estimates that are within a ±6 h time window. The following approach is then used to merge the MW and IR rain values for any given pixel:
E. N. Anagnostou
362
• • •
If MW > 0 and IR > 0: the merged value is based on an error variance-weighted averaging. If MW = 0 and IR = 0: the merged value is zero. If MW > 0 and IR = 0 or MW = 0 and IR > 0: the rain/no-rain decision is made according to the sensor with the highest rain detection Heidke skill score (HSS) (Heidke 1926).
The error variance and HSS statistics for MW and IR rain retrievals were evaluated using as reference matched hourly WSR-88D rainfall fields. Table 1 shows those statistics. Note a deviation in the rainfall error statistics between the SSM/I and TMI MW sensors. The MW error statistics were augmented by a representativeness error term used to account for the large time lags between a MW overpass and a certain hourly rain estimate. For this purpose we used hourly radar rainfall fields to determine the error variance and rain detection HSS score as function of time lag. The error statistics are shown in Fig. 1.
Figure 1. Representativeness error variance (left) and HSS score (right) presented as function of time lag. Table 1. Error variance and HSS of MW and IR rainfall estimates.
Variance HSS
MW (TMI) 0.65 0.41
MW (SSM/I) 0.8 0.37
IR 1.65 0.1
5 SIMULATION EXPERIMENT AND RESULTS As mentioned in the Introduction, this study aims at studying the effect of rainfall retrieval error on the simulation of land surface parameters assuming
Assessment of Satellite Rain Retrieval Error Propagation
363
that our knowledge of the physical processes and other input parameters is accurate. The CLM, which provides a physically based framework for land surface processes, is used to model the 1D grid-averaged processes at two distinct, in terms of vegetation cover, sites. Three simulation scenarios are considered where CLM is forced with the three different rainfall data sets (radar, IR, and merged MW/IR). During the spin-up period of the model (January to April) the grid-averaged radar rainfall was commonly used as input to the model. The other meteorological forcing parameters such as radiation budget, air temperature, wind speed, and relative humidity were based on grid-averaged MESONET measurements, which were considered to be associated with insignificant error. Consequently, the only source that would differentiate the predicted land surface variables among the three simulation scenarios is the differing precipitation input. This difference is defined in relative terms as following:
εx =
Vx − Vref Vref
(1)
V is used to symbolize the different hydrologic variables (including precipitation). Subscript “x” refers to the variables predicted from satellite IR or merged rain estimates and subscript “ref” refers to variables predicted on the basis of grid-averaged rain gauge-calibrated radar rainfall. The land surface variables evaluated in this study are: latent heat flux (LE), sensible heat flux (SH), surface runoff (Roff), soil moisture content at 21 cm depth ( θS ), and soil temperature at 21 cm depth ( θ T ). In Table 2 we summarize the relative error (ε) statistics (mean and standard deviation [STD]) of the two satellite retrievals in terms of precipitation and CLM-predicted variables. The statistics are presented for three spatial scales (0.25, 0.5, and 1 degree) and the two vegetation regimes (Hveg and Lveg). A first general observation is that the merged rain retrieval is associated with less bias and error variance in all predicted variables compared to the IR rain retrieval. In terms of bias (or mean error) we note an almost 250% increase between low (Lveg) and high (Hveg) vegetation sites for the IR retrieval, while the corresponding increase in the case of merged rain estimates is moderate (~80%). An interesting observation for the IR retrieval is that the rain estimation bias magnifies significantly in runoff prediction (~3 times) in the case of Lveg, while the corresponding runoff-to-rainfall bias ratio for the merged rain retrieval is almost one. In the case of Hveg site the runoff bias
E. N. Anagnostou
364
Table 2. Error statistics (bias and STD) of rainfall and CLM-predicted variables derived from the simulation experiment.
Lveg
P
Mean error (bias) 0.25° 0.5° 60 (26) 60 (26)
LE SH Roff
34 (19) –26 (–15) 115 (37)
31 (18) –25 (–14) 115 (35)
32 (18) –27 (–16) 116 (30)
θS
7 (0.5)
7 (0.2)
8 (0.2)
θT
–0.4(–0.2)
–0.4(–0.2)
θS
–0.5(– 0.3) –12 (–15) –11 (–6) 13 (7) –7 (–24) –5 (–4)
0.25° 124 (113) 76 (69) 35 (30) 187 (144) 113 (117) 40 (34)
–12 (–15) –12 (–7) 15 (9) –7 (–25) –5 (–4)
–12 (–15) –14 (–8) 17 (10) –6 (–27) –5 (–3)
93 (88) 49 (44) 29 (26) 93 (79) 71 (52)
88 (81) 49 (44) 29 (26) 91 (71) 63 (43)
76 (67) 47 (42) 27 (24) 92 (59) 60 (41)
θT
0.02 (0.0)
0.02 (0.0)
0.05 (0.0)
15 (13)
15 (13)
13 (10)
1.0° 60 (26)
(21 cm)
Hveg
(21 cm) P LE SH Roff
STD 0.5° 129 (113) 74 (67) 34 (30) 213 (148) 108 (110) 39 (32)
1.0° 130 (109) 68 (60) 32 (27) 232 (146) 118 (120) 36 (29)
(21 cm) (21 cm)
seems to reduce (smoothing effect) by about 30%. In terms of soil moisture and other variables, we note a significant reduction of bias for both satellite retrievals and vegetation sites. In terms of the relative STD error statistic our observations are the following. We demonstrate a moderate reduction of STD from Lveg to Hveg, and notable spatial scale dependence for STD in the Hveg regime. The STD of rainfall, runoff, and soil moisture error for the merged (IR) rain estimates reduces by an average of 23% (12%) going from 0.25 to 1-degree grid resolution. We show magnification in the error STD going from rainfall to runoff variable, and a smoothing effect for all other variables. The rate of STD magnification for the IR retrieval exhibits a spatial scale dependence (ranging from 60% to 80%) in the case of Lveg, while the corresponding rate of magnification for the merged rain estimate is consistent across the scales at ~30%. In Hveg site the IR retrieval error STD magnification in runoff is shown to vary from about 5% (at 0.25 degree) to 25% (at 1 degree), while for the merged rain estimates we note a consistent smoothing effect (~10% error reduction). In all other variables the STD error propagation is associated with smoothing, which varies across the different variables. In soil moisture, for example, the STD reduces by about 12% (8%) in the merged (IR) retrieval case. The error STD smoothing is significant for the energy-related variables and fluxes
Assessment of Satellite Rain Retrieval Error Propagation
365
(θT, LE, and LH), indicating that errors in precipitation alone would have minimal effect in those variables. We finally used our experimental data to investigate the spatial structure of error in the various hydrological and land surface variables. This is presented in Fig. 2 in terms of spatial correlation of error (ε) as function of distance. We present two curves in each panel, one representing the IR and the second the merged rain retrievals. Left and right panels represent Lveg and Hveg sites, respectively. We make the following observations. First, error in rainfall and runoff decorrelate faster than in the other variables. Second, the IR retrieval error is associated always with lower lag-correlation than the corresponding IR retrieval error. This explains why spatial averaging resulted in a more significant STD reduction in the merged case compared to the IR retrieval case.
Figure 2. Spatial lag-correlation of error. Left and right panels correspond to Lveg and Hveg sites, respectively.
E. N. Anagnostou
366
6 CLOSING REMARKS This paper presented a framework for investigating hydrologically relevant satellite rain retrievals. The strategy adopted is, firstly, to compile the hydrometeorological variables needed by an LSM including three different sources of precipitation data, i.e., gauge-calibrated radar rainfall and two satellite retrieval (an IR and a merged MW/IR rain retrieval), assuming that the model parameters are well representative of the land surface processes. The CLM was selected to simulate the land surface processes at three scales, 0.25, 0.5, and 1 degree, and two distinct vegetation conditions. Simulated results from the two satellite rain retrieval-forcing parameters were compared to corresponding simulations derived from the radar rainfall input considered as reference. The primary conclusions of the present study are as follows: •
• • •
Rainfall error propagates nonlinearly in hydrologic simulation uncertainty-precipitation error structure and land surface conditions affect this error propagation (we noted this specifically in the case of runoff, and to a lesser extent in soil moisture). Error variance reduces (nonlinear smoothing effect) in most of the hydrological variables and fluxes, but runoff. We noted lower error statistics when using the merged MW/IR rain input compared to the IR-only input. Overall, spatial averaging reduces the error in most of the land surface variables, but this reduction depends on the rain retrieval error characteristics and land surface conditions.
An important observation from this study is that the effect of vegetation and structural characteristics of rain retrieval error are factors affecting the error propagation in land surface variables. This study investigated only a limited number of land surface conditions and satellite retrieval schemes and, thus, should only be viewed in a qualitative sense. Apparently, using different rain retrievals and hydrometeorological regimes could lead to different error characteristics. We would like to view this experimental error propagation framework as the basis for achieving a more comprehensive hydrologic assessment of satellite retrievals, and/or for developing merging techniques that would optimize hydrological prediction error statistics.
7 REFERENCES Adler, R. F., G. J. Huffman, D. T. Bolvin, S. Curtis, and E. J. Nelkin, 2000: Tropical rainfall distributions determined using TRMM combined with other satellite and rain gauge information. J. Appl. Meteor., 39, 2007–2023.
Assessment of Satellite Rain Retrieval Error Propagation
367
Anagnostou, E. N., A. J. Negri, and R. F. Adler, 1999: A satellite infrared technique for diurnal rainfall variability studies. J. Geophys. Res., 104 (D24), 31477–31488. Beven, K. J. and M. J. Kirkby, 1979: A physically based variable contributing area model of basin hydrology. Hydrol. Sci. Bull., 24, 43–69. Bonan, G. B., 1996: A land surface model (LSM version 1.0) for ecological, hydrological, and atmospheric studies: Technical description and user’s guide. NCAR Technical Note, NCAR/TN-417+STR, Boulder, CO. Bonan, G. B., S. Levis, L. Kergoat, and K. W. Oleson, 2002: Landscapes as patches of plant functional types: an integrating concept for climate and ecosystem models. Global change Biochem Cycles, 16 (2), Art No. 1021. Bonan, G. B., K. W. Olsen, M. Vertenstein, S. Levis, X. B. Zeng, Y. J. Dai, R. E. Dickinson, and Z. L. Yang, 2002: The land surface climatology of the community land model coupled to the NCAR community climate model. J. Climate, 15, 3123–3149. Brock, F. V., K. C. Crawford, R. L. Elliot, G. W. Cuperus, S. J. Stadler, H. L. Johnson, and M. D. Eilts, 1995: The Oklahoma Mesonet: a technical overview. J. Atmos. Oceanic Technol., 12, No. 1, 5–19. Dai, Y. J. and Q. C. Zeng, 1997: A land surface model (IAP94) for climate studies 1: formulation and validation in off-line experiments. Adv. Atmos. Sci., 14, 433–460. DeFries, R. S., M. C. Hansen, and J. R. G. Townshend, 2000a: Continuous fields of vegetation characteristics: a linear mixture model applied to multi-year 8 km AVHRR data. Int. J. Remote Sens., 21, 1389–1414. DeFries, R. S., M. C. Hansen, J. R. G. Townshend, A. C. Janetos, and T. R. Loveland, 2000b: A new global 1-km dataset of percentage tree cover derived from remote sensing. Global Change Biol., 6, 247–254. Dickson, R. E., A. Henderson-Sellers, P. J. Kennedy, and M. F. Wilson, 1986: BiosphereAtmosphere Transfer Scheme (BATS) for the NCAR community climate model. Technical Note; NCAR/TN-275+STR, NCAR, Boulder, CO. Fulton, R. A., J. P. Breidenbach, D. J. Seo, D. A. Miller, and T. O’Bannon, 1998: The WSR88D rainfall algorithm. Wea. Forecasting, 13 (2), 377–395. Global Soil data Task, 2000: Global soil data products CD-ROM (IGBP-DIS). International Geosphere-biosphere Programme – data and Information Available Services. Available at: http://www.daac.ornl.gov Grecu, M. and E. N. Anagnostou, 2001: Overland precipitation estimation from TRMM passive microwave observations. J. Appl. Meteor., 40 (8), 1367–1380. Hossain, F., E. N. Anagnostou, and T. Dinku, 2004: Sensitivity analyses of satellite rainfall retrieval and sampling error on flood prediction uncertainty. IEEE Trans. Geosci. Remote Sens., 42, 130–139. Hsu, K., X. Gao, S. Sorooshian, and H. V. Gupta, 1997: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36 (9), 1176–1190. Huffman, G. J., R. F. Adler, E. F. Stocker, D. T. Bolvin, and E. J. Nelkin, 2003: Analysis of TRMM 3-hourly multi-satellite precipitation estimates computed in both real and postreal time. Proc. 12th Conf. Sat. Meteor. and Oceanog., 9–13 Feb. 2003, Long Beach, CA. Huffman, G. J, R. F. Adler, M. M. Morrisey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 36–60. Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2000: A real-time global half-hourly pixel resolution infrared dataset and its applications. Bull. Amer. Meteor. Soc., 82, 205–217. Koster, R. and M. Suarez, 1996: Energy and water balance calculations in the MOSAIC LSM. NASA Tech Memo 104606, Vol. 9. Liang, X., E. Wood, and D. Lettenmaier, 1996: Surface and soil moisture parameterization of the VIC-2L model: Evaluation and modifications. Global Planet. Change, 13, 195–206.
368
E. N. Anagnostou
Loveland, T. R., B. C. Reed, J. F. Brown, D. O. Ohlen, Z. Zhu, L. Yang, and J. W. Merchant, 2000: Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR data. Int. J. Remote Sens., 2, 1303–1330. Mitchell, K. and Co-authors, 1999: GCIP Land Data Assimilation System (LDAS) project now underway. GEWEX News, 9 (4), 3–6. Mitchell, K. and Co-authors, 2000: Recent GCIP-sponsored advancements in coupled landsurface modeling and data assimilation in the NCEP Eta mesoscale model. Prepr. 15th AMS Conf. on Hydrology, Long Beach, CA, Paper P1.22. Nijssen, B. and D. P. Lettenmaier, 2004: Effect of precipitation sampling error on simulated hydrological fluxes and states: Anticipating the Global Precipitation Measurement Satellites. J. Geophys. Res., 109, D02103, doi: 10.1029/2003JD003497. Seo, D. J., J. Breidenbach, and R. Fulton, 2000: Real-time adjustment of range-dependent biases in WSR-88D rainfall estimates due to nonuniform vertical profile of reflectivity. J Hydrometeor., 1 (3): 222–240. Seo, D. J., J. P. Breidenbach, and E. R. Johnson, 1999: Real-time estimation of mean field bias in radar rainfall data. J. Hydrology, 223 (3–4): 131–147. Todd, M. C., C. Kidd, D. Kniveton, and T. J. Bellerby, 2000: A combined satellite infrared and passive microwave technique for estimation of small scale rainfall. J. Atmos. Oceanic Technol., 18, 742–754.
29 EURAINSAT ALGORITHM VALIDATION AND INTERCOMPARISON EXERCISE Martina Kästner DLR, Oberpfaffenhofen, Germany
Abstract
This study compares four satellite rain estimations based on microwave (MW), infrared (IR) or combined MW–IR techniques and contrasts them with the mesoscale Bologna local area model (BOLAM) rain analysis or the network-based gauge data from the Global Precipitation Climatology Centre (GPCC) for a period from 08 to 13 November 2001 over the western Mediterranean Sea during a severe weather event, which resulted in a disastrous flood in Algeria. The PR-adjusted TMI Estimation of Rainfall (PATER) and frequency difference algorithm (FDA) are applied to MW TRMM data, the neural rain estimator (NRE) uses geostationary IR Meteosat data and the combined NRL Turk algorithm uses both MW data from low orbiting satellites and IR data from a geostationary orbit. The unique gridded data provide an effective basis to compare instantaneous space measurements with different algorithms. Validation results indicate that there is generally a better performance for heavy rain than for weak rain. Both MW algorithms, PATER, and FDA, perform rather similarly whereas PATER is applicable exclusively over the ocean and shows some rain detection problems due to thick aerosol loads originating from the desert. The BOLAM model performs rather well in this case study; although only a small location error of a heavy rain area was analyzed. The IR-based techniques have the advantage of a high temporal repetition rate but both algorithms, NRL and NRE, have problems with identifying the correct rainy areas compared to MW results. Overall, the results suggest combining both advantages, the wellknown rain physics of the MW channels with the high temporal resolution of IR algorithms, to retrieve precipitation from satellite data.
Keywords
Satellite rain retrieval, microwave, infrared, validation, Algerian flood 2002, severe weather event
369 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 369–380. © 2007 Springer.
370
M. Kästner
1 INTRODUCTION Various climate models predict a decrease of precipitation in the future over many parts of the subtropics, particularly in the winter (Bolle 2003). Therefore, it is essential to have not only climatological data from land, but also from over the seas. Remote areas not covered by conventional observation networks can now be continuously monitored by low orbiting and geostationary satellites. The first satellite rain retrievals in both the infrared (IR) and the microwave (MW) spectra date back to the 1970s, some recent works are Turk et al. (2000), Bauer (2000, 2001), Bauer et al. (2001), Levizzani et al. (2001), Grose et al. (2002), Oh et al. (2002), Tapiador (2002), and Kidd et al. (2003) among others. A cross-comparison of TRMM and GPCP rainfall data sets are described in Adler et al. (2002). The evaluation of passive MW precipitation algorithms which are directly linked to the 3D structure of the precipitating system use measurements from different sensors on different satellites, like SSM/I on DMSP, TMI/PR on TRMM, or AMSU on NOAA. Passive MW techniques perform much better over the oceans than over land. The MW techniques are directly related to the hydrometeors through scattering and emission, but the low earth orbits and less frequent coverage hinders tracking of developing severe storms. While the daily course of precipitation is not easily obtained from TRMM data, IR-based techniques from geostationary satellites have been widely used due to the high revisit period. However, they have an inherent weakness regarding the physical relation between cloud top temperatures and underlying rain rate (RR). Further, the rain characteristics vary with different climate regimes, hence, any developed method has to be validated against appropriate in situ measurements taken over the region of interest. An intercomparison of PMW and/or IR-based algorithms with BOLAM model data and GPCC gauge measurements at different spatial resolutions is performed for the Algerian flood in early November 2001. Although the validity of the results obtained is restricted to the case studies some general information could be extracted. Further details can be found in Kästner (2003) and Kästner et al. (2006).
2 DATA The validation of various rain algorithms is performed for a severe weather event between 08 and 12 November 2001 on the Algerian coast and the Balearic Islands. The synoptic situation was characterized by strong surface winds and heavy rainfall. An intense upper-level trough pushed far to the south of Europe where a cutoff low developed. The METEOSAT-7 IR image (Fig. 1) shows the clouds with heavy rainfall on the Algerian coast. The rainfall started on late 9 November and ended the next day at about
Eurainsat Algorithm Validation and Intercomparison Exercise
371
noon on 10 November, when 150 L m–² within 6 h were observed. Together with the cutoff low process heavy thunderstorms developed in a cyclogenesis over the Balearic Islands the next day. The precipitation was reported to be greater than 400 mm over 2 days, with a maximum of 68 L m–² in 6 h (Thomas et al. 2003). Two processes intensified the convective development: (1) the cold maritime arctic air that crossed over the still 18°C warm Mediterranean Sea where it picked up moisture, destabilized, and met initially maritime subtropic air and (2) the strong surface winds blowing against the high mountains along the African coast (>2,300 m) caused intense orographic rainfall which led to the flooding disaster in Algiers with more than 750 deaths.
Figure 1. 10 November 2001; left: TRMM TMI brightness temperatures for 19h (top), 19v (mid ) and 85v (bottom) GHz channels (v = vertical, h = horizontal polarization), orbit time 0025 UTC; right: METEOSAT-7 IR, 1200 UTC. A = Algiers.
Input data for PMW algorithms (PATER, FDA) are the TMI brightness temperatures (TB) of nine channels (10.7v,h, 19.4v,h, 21.3v, 37.0v,h, 85.5v,h GHz) with varying resolutions from 70 to 6 km. Figure 1 shows differences of emission over water and over land for both polarizations and for different TMI channels. Over water, the rainy areas appear to be warmer than their surroundings, while over land they appear to be colder due to the high MW emission of land. The applied rain retrievals and rain data are briefly described in the following: The BOLAM is a hydrostatic, primitive equation, grid point model in σ -coordinates, using horizontal wind components, potential temperature, specific humidity, and surface pressure as basic dependent variables. The initial and boundary conditions are obtained from the ECMWF 6-hourly analyses. The frequency difference algorithm (FDA) of Kidd et al. (2003) uses the 19v and 19h GHz channels and relates it to the rain rate (RR), as described
372
M. Kästner
in Ebert et al. (1996). It is a PMW satellite rain algorithm that is operationally applied to TRMM as well as SSM/I orbit data. Independent 1-degree daily (1DD) data from the Global Precipitation Climate Centre (GPCC) base on recordings of the very dense rain-gauge network in Europe, the SYNOP reports are automatically checked and corrected for systematic measuring errors (Rudolf et al. 1996). The neural rain estimator (NRE) is an operational rapid update IR-based algorithm to diagnose half-hourly near-surface rainfall. It is an empirical technique suited for acquisitions from geostationary satellites. It uses some relevant features of the cloud top evolution and structure and information from an NWP model. The basic strategy for the Naval Research Laboratory (NRL)-blended technique (Turk et al. 2000) draws upon the probability matching methods developed in the radar meteorology field for specific “tuned” Z–R relations. Time- and space-coincident IR and MW pixels are collected from different satellites and used to produce dynamically updated TB–RR lookup tables. The over-ocean satellite rainfall algorithm PATER is a physical algorithm that uses only two empirical orthogonal functions instead of the nine TBs from the TMI channels. The retrieval database is generated from several 3D cloud model simulations including the melting layer (Bauer 2001). The algorithm (Bauer et al. 2001) has a stand-alone PMW component based on TRMM TMI (1B11) data and an optionally carefully colocated calibration with PR (2A25) data (~5 km) downscaled to the lower TMI spatial resolution for the 10 GHz (~50 km). Currently, it is foreseen for operational implementation in the ECMWF assimilation scheme.
3 METHOD OF ANALYSIS In a pre-study the RR of the PATER over-ocean algorithm were merged from three TRMM overflights daily and then downscaled to a 1° × 1° grid for comparison with the independent RR gauge data from the 1DD GPCC data. For the main study a common area, period, grid, and format were essential for combining data sets with different temporal or spatial resolutions sometimes describing different physical observables. All subsequent tasks like calibration, sampling, or error analysis needed a common grid that allowed an equivalent evaluation. For the joint effort of validation and intercomparison of several rain algorithms applied in the scope of the EURAINSAT project, continuous and categorical statistics were used according to Ebert et al. (1996, 1998). The Algerian severe weather event was used as a common case study for an intercomparison of PMW, IR, combined MW/IR rainfall algorithms and the BOLAM model. The common area extended from 15 W to 20 E and from 30 to 60 N; with a common period from 09 to 11 November 2001; and with RR
Eurainsat Algorithm Validation and Intercomparison Exercise
373
data resampled into a 0.25° × 0.25° lat–long grid (~28 km). TRMM and IRE data did not both consistently cover the complete common area, therefore, the subsequent comparison was conducted only over the available common area. The temporal coincidence was optimal for the different PMW algorithms, because PATER and FDA have the same TRMM database, otherwise the temporal window was better than ±15 min for comparisons with IR (NRL, NRE) and in most cases better than ±90 min for comparisons with the independent model data, which have a 3-hourly temporal resolution. Comparisons were made for single orbits as well as for merged data within 3-h periods.
4 RESULTS 4.1 1° intercomparison – PATER retrieval with GPCC Although the GPCC data set was over land and the PWM PATER data set was over water; coastal boxes contained in both data sets, could be compared (see Fig. 2).
Figure 2. Rain rate (RR) analysis in mm day–1 on 10 November 2001; left: 1DD GPCC (Global Precipitation Climate Centre) RR distribution; right: PATER retrieval, RR from merged orbits (0025 and 0210 UTC) downscaled to 1DD.
In order to get an adequate number of observations (n = 146) all the coastal boxes of the southwestern Mediterranean Sea over 5 days (different symbols in Fig. 3) during a severe rain and flood event were taken into account. The result of this validation study is a rather high correlation coefficient of 0.71 between rain gauges and PATER estimates. A closer examination shows three outcomes. (1) Most of the data points are within the red ellipse indicating a good agreement. (2) A few data points are within the green ellipse showing higher rain-gauge values than PWM RRs . This is explained
M. Kästner
374
by the different temporal data structure of instantaneous orbit data versus accumulated gauge data. (3) Biased (too high) PATER RRs for low rain gauge measurements (blue ellipse). This third outcome was expected, because the sensitivity of the PATER algorithm is given by Bauer (2001) with lwc = 0.07 g m–3 equivalent to RR = 24 mm day–1. In fact the false alarm rate indicates an even higher sensitivity to about 17 mm day–1 in this case.
Figure 3. Rain rate comparison of the 1-degree daily (1DD) GPCC gauge data vs. PMW PATER rain rate estimations in mm per day for the period 09–13 November 2001 (Algerian case).
4.2 0.25° intercomparison – PATER and other EURAINSAT retrievals A validation study of the satellite rainfall estimations has been performed for the Algerian case using independent intercomparison of different rainfall retrievals, including pure PMW, pure IR, combined MW/IR techniques, and model results using BOLAM before nudging with satellite rainfall data. Both PMW algorithms, PATER, and FDA, rely on the same TMI orbit data and so it was expected that their comparison would result in a rather similar rainfall region and intensity. Both algorithms are assessed to be of equal quality in this heavy rainfall event, considering that the PATER algorithm is restricted to ocean surfaces and to rain events above 1 mm h–1. The most successful
Eurainsat Algorithm Validation and Intercomparison Exercise
375
comparison of pure PMW with other techniques was in this case the BOLAM model followed by the NRL blended MW/IR technique both of which performed better than the pure IR technique. The results of the intercomparison of different rain retrievals are given for visual inspection for a single date (10 November 2001, 0300 UTC) (Fig. 4 and 5). A complete analysis of the categorical statistics of all possible combinations of the above-described algorithms within the period 08 to 13 November 2001 is given in Table 1.
Figure 4. 10 November 2001, TMI orbit 22741 + 22742, 0025 + 02:10 UTC; left: PATER algorithm – MW; right: FDA algorithm – PMW.
Figure 5. 10 November 2001, 0300 UTC; upper left: BOLAM – model; upper right: NRL Turk algorithm – combined IR–MW; bottom: NRE (neural rain estimator) – IR. (see also color plate 14)
M. Kästner
376
Figure 4 shows three areas of heavy rainfall, one west of the Canary Islands, the biggest one is around Algiers, and one southeast of Sardinia. The three rain areas coincide very well with results of both MW algorithms. Even the rain intensities are similar except for the Sardinia area. FDA works over land and ocean, PATER exclusively over ocean and for rainfall rates above 1 mm h–1. The low RRs erroneously detected over the sea south of Sicily (Fig. 4 left) are attributed to strong desert aerosol also detected as aerosol fallout in Rome the next day. The rainy speckles over the Atlantic are due to cumulus convective showers within the cold air. Compared to the PMW techniques, the BOLAM model rainfall (before nudging) shows wide agreement of the strong rain bands (Fig. 5 upper left). The rainfall intensities were rather similar; only the position of the rainfall had an error for the Sardinia area. On the other hand, the shower pattern over the Atlantic is not well matched and there are too many areas with light rain. The two IR-based algorithms, NRL and NRE (Fig. 5), give heavy rain areas over the Mediterranean Sea, but not at the correct position. The Algerian coast, where the maximum precipitation fell, is hardly classified as a heavy rain area. The combined MW/IR NRL algorithm shows a much better performance than the IR algorithm NRE alone. Overall it seems to be worthwhile to combine the high temporal resolution of IR with the better rainfall identification performance of MW techniques for monitoring purposes. The NRL algorithm or the now available TRMM 3B42RT products belong to this category. The low MW pixel resolution makes a 0.25° lat–long grid appropriate, but better spatial resolution is desired by the users. Table 1. Categorical statistics for different satellite rain retrievals.
P vs. B P vs. NRE P vs. NRL B vs. NRE B vs. NRL NREvs.NRL F vs. P F vs. B F vs. NRL F vs. NRE Total
Compared pairs
Hit
Miss
False alarm
Correct negative % 62.2 56.1 67.2 39.2 65.9 52.6 79.7 66.3 75.9 56.3
Heidke skill score Best= 1 0.67 0.63 0.70 0.61 0.76 0.73 0.82 0.78 0.81 0.69
N = 100% 8.346 2.595 14.619 3.774 13.880 6.320 3.353 6.128 7.142 931
% 4.9 7.2 3.3 22.3 9.9 20.6 2.5 11.2 5.0 12.5
% 21.1 16.6 14.3 10.0 13.3 15.6 8.9 18.3 15.2 23.3
% 11.9 20.0 15.2 28.6 10.9 11.1 8.9 4.2 3.9 7.8
67.088
9.9
15.7
12.3
Bias Best = 1 0.65 1.14 1.05 1.58 0.89 0.88 1.01 0.52 0.44 0.57
62.1
0.72
0.87
Eurainsat Algorithm Validation and Intercomparison Exercise
377
The event occurrences of the contingency table for comparisons within the time period 09 to 11 November 2001 consider the above-mentioned temporal windows (Table 1). The number of compared pairs (N) differs for each algorithm pair; thus, the occurrences are given in percent of each N for better comparison between the cases, the maxima (hits) or minima (false alarms) are italicized. The following abbreviations are used: B = BOLAM, F = FDA, P = PATER. The accuracy of all intercomparisons is measured in terms of Heidke skill score (HSS), and ranges from 0.61 (BOLAM vs. NRE), to respectable 0.78 (BOLAM vs. FDA), to optimal 0.82 (PATER vs. FDA, Fig. 6). No bias between the data sets is indicated when the bias value is unity, thus, PATER and NRL have only a small bias, while again PATER performs best when compared with FDA. Note that even in this heavy rain event only 10% of the gridded pixels are hits (rain/rain), whereas the majority (62%) is correct negatives (no-rain/no-rain), thus, the correct negatives dominate the statistics.
Figure 6. Comparison of PATER with FDA. The accuracy via HSS*1,000 (thick) depends on the rain/no-rain threshold RR-min.
378
M. Kästner
5 DISCUSSION A heavy rainfall situation was selected for a validation analysis of different satellite rainfall algorithms representing MW, IR, or combined MW/IR retrievals and the BOLAM model. The rainfall data were gridded on 1° and 0.25° in a common area over Western Europe and categorical statistics are used. Special interest was laid on the PATER algorithm that uses TRMM active and passive MW data. The PATER retrieval is crucially dependent on the forward 3D cloud model calculations. There is a need for better and more comprehensive cloud models that cover the whole spectrum of natural clouds, particularly clouds with moderate and light rain and snow. The correlation coefficient r is often used in comparison tasks. It has already been shown that r is dependent on the choice of the grid size (Turk et al. 2002). In this study the maximum r is reached for both PMW techniques, namely r is 0.88 for PATER vs. FDA (orbit 22,741, 0.25° grid). The accuracy is measured in terms of the HSS, which is a combination of hits, false alarms, misses, and correct negatives. HSS is truly dependent on the threshold RR-min, discriminating rain from no-rain pixels. Figure 6 results from varying the minimum detectable RR (RR-min) for the rain algorithm pair PATER vs. FDA. The event occurrences for each category are given on the y-axis. For RR-min = 1 mm h–1, false alarms (red) and misses (blue) are of the same order, which means that in this case both the algorithms are unbiased. The impact of RR-min on the accuracy is evident; the strong increase of HSS from 0.3 to 0.8 for RR-min greater than 0.7 mm h–1 indicates that higher RRs are more accurately detected than low ones. This implies that not only the chosen grid size but also the problem on where the rain/no-rain threshold is set is inherently associated with the accuracy problem. As the statistics are dominated by the correct negatives and not by the hits, maybe the use of entity-based methods, like contiguous rain area (CRA) verification give further insight into algorithm performances (Ebert 2000). In this case study the PMW techniques performed better than IR techniques. Overall, the results suggest combining both advantages, the wellknown rain physics of the MW channels with the high temporal resolution of IR algorithms, to retrieve precipitation from satellite data. Acknowledgment: This research is funded in by the EURAINSAT project, a shared-cost project (contract EVG1-2000-00030), co-funded by the Research DG of the European Commission within the RTD activities of a generic nature of the Environment and Sustainable Development subprogramme 5th Framework Programme).
Eurainsat Algorithm Validation and Intercomparison Exercise
379
6 REFERENCES Adler, R. F., G. Huffman, and D. Bolvin, 2002: TRMM and GPCP initial cross-comparison. GEWEX News, 12, 5–6. Bauer, P., 2001: Over-ocean rainfall retrieval from multisensor data of the tropical rainfall measuring mission. Part I: design and evaluation of inversion databases. J. Atmos. Oceanic Technol., 18, 1315–1330. Bauer, P., D. Burose, and J. Schulz, 2000: Rain detection over land surfaces using passive microwave satellite data. ECMWF Tech. Memo., No. 330, Reading, England. Bauer, P., P. Amayenc, C. D. Kummerow, and E. A. Smith, 2001: Over-ocean rainfall retrieval from multisensor data of the Tropical Rainfall Measuring Mission. Part II: algorithm implementation. J. Atmos. Oceanic Technol., 18, 1838–1855. Bolle, H.-J. (ed.), 2003: Mediterranean Climate – Variability and Trends. Springer, Berlin, 372 pp. Buzzi, A., M. D’Isidoro, and S. Davolio, 2003: A case study of an orographic cyclone formation south of the Alps during the MAP-SOP. Quart. J. Roy. Meteor. Soc., 129, 1795–1818. Ebert, E. E., 1996: Results of the 3rd Algorithm Intercomparisons Project (AIP-3) of the Global Precipitation Climatology Project (GPCP). Revision I. Bureau of Meteorology Research Centre, Melbourne, Australia, 199 pp. Ebert, E. E. and M. J. Manton, 1998: Performance of satellite rainfall estimation algorithms during TOGA COARE. J. Atmos. Sci., 55, 1537–1557. Ebert, E. E. and J. L. McBride 2000: Verification of precipitation in weather systems: determination of systematic errors. J. Hydrol., 239, 179–202. Grose, A., E. A. Smith, H.-S. Chung, M. L. Ou, B. J. Sohn, and F. J. Turk, 2002: Possibilities and limitations for QPF using nowcasting methods with infrared geosynchronous satellite imagery. J. Appl. Meteor., 41, 763–785. Kästner, M., 2003: Inter-comparison of precipitation estimations using TRMM microwave data and independent data. In: Proc. 3rd GPM Workshop – Consolidating the Concept, Noordwijk, The Netherlands, 24–26 June 2003, AP-5. http://www.estec.esa.nl/conferences/ 03C06/ and in: Proc. CD-ROM 5th EGS Plinius Conference on Mediterranean Storms, Ajaccio, France, 1–3 Oct 2003. Kästner, M., F. Torricella, and S. Davolio, 2006: Intercomparison of satellite-based and model-based rainfall analyses. Meteor. Appl., 13, 213–223. Kidd, C., D. Kniveton, M. Todd, and T. Bellerby, 2003: Satellite rainfall estimation using a combined passive microwave and infrared algorithm. J. Hydrometeor., 4, 1088–1104. Levizzani V., J. Schmetz, H.J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2001: Precipitation estimations from geostationary orbit and prospects for METEOSAT Second Generation. Meteorol. Appl., 8, 23–41. Oh, H. J., B. J. Sohn, E. A. Smith, F. J. Turk, A. S. Seo, and H. S. Chung, 2002: Validating infrared-based rainfall retrieval algorithms with 1-minute spatially dense raingauge measurements over the Korean peninsula. Meteor. Atmos. Physics, 81, 273–287. Rudolf, B., H. Hauschild, M. Reiß, and U. Schneider, 1992: Beiträge zum Weltzentrum für Niederschlagsklimatologie – Contributions to the Global Precipitation Climatology Centre. Meteorologische Zeitschrift, 1, 7–84. Tapiador, F., 2002: A new algorithm to generate global rainfall rates from satellite infrared imagery. Revista de Teledeteccion, 18, 57–61. Thomas, W., F. Baier, T. Erbertseder, and M. Kästner, 2003: Analysis of the Algerian severe weather event in November 2001 and its impact on ozone and nitrogen dioxide distributions. Tellus B, 55B, 993–1006.
380
M. Kästner
Turk, F. J., J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000: Combining SSM/I, TRMM and infrared geostationary satellite data in a near real-time fashion for rapid precipitation updates: advantages and limitations. Proc. The 2000 EUMETSAT Meteorological Satellite Data Users’ Meeting, Bologna, Italy, 29 May – 2 June 2000; EUM P 29, 452–459. Turk, F. J., E. E. Ebert, H. J. Oh, B. J. Sohn, V. Levizzani, E. A. Smith, and R. R. Ferraro, 2002: Validation of an operational global precipitation analysis at short time scales. In: Proc. 1st Workshop Intern. Precip. Working Group (IPWG), Madrid, 225–248.
30 GROUND VALIDATION FOR THE GLOBAL PRECIPITATION CLIMATOLOGY PROJECT Mark M. Morrissey and Scott Greene University of Oklahoma, Norman, OK, USA
Abstract
This paper discusses the role of the Surface Reference Data Center (SRDC) and the activities associated with data collection, error characterization, and validation associated with the Global Precipitation Climatology Project (GPCP). Housed in the Environmental Verification and Analysis Center (EVAC) at the University of Oklahoma, the EVAC/SRDC has built upon work from past NOAA-supported projects to become a unique location for scientists to obtain scarce rain gauge data and to conduct research into verification activities. These data are continually analyzed to produce error-assessed rainfall products. Scientists need only to access the EVAC/SRDC web site (http://www.evac.ou.edu/ srdc) to obtain critical global rain gauge data sets. Many of these data sets are impossible to obtain elsewhere. In this paper we will discuss the data collection, analysis, and validation methodology activities of the SRDC.
1 INTRODUCTION During the initiation of the Global Precipitation Climatology Project (GPCP) in the early 1980s it was recognized that confidence could only be placed in a satellite-derived global precipitation analyses if an active verification program was set up concurrently with the GPCP. Thus, the idea of the Surface Reference Data Center (SRDC) was formed and initially located at the US National Climatic Data Center (NCDC) in Ashville, North Carolina. The primary mission of the SRDC at that time was to collect and analyze rain gauge data from special high density and quality networks located around the world. In addition, the data from these networks were to be interpolated in an optimal fashion to produce rainfall estimates of sufficient quality to provide useful comparisons with GPCP estimates at the same time
381 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 381–392. © 2007 Springer.
382
M. M. Morrissey and S. Greene
and space scales. The interpolated rain gauge estimates were placed on the SRDC web site to be downloaded for verification purposes by GPCP scientists. During the mid-1990s the SRDC was transferred to the Environmental Verification and Analysis Center (EVAC) at the University of Oklahoma. EVAC specializes in the development of creative verification methodology, which is primarily stochastic in nature. At this time, the SRDC took on an additional responsibility of performing verification analysis of GPCP products. EVAC specializes in locating and providing hard-to-find rain gauge data sets to the research community. For example, the Comprehensive Pacific Rainfall Database (PACRAIN, Morrissey et al. 1995a), which is part of the SRDC data base, is the most extensive Pacific island rain gauge data base in the world. Data have been collected from hundreds of Pacific island stations, with some records going back as far as the 1800s. It is currently available to scientists via the internet at www.evac.ou.edu/pacrain. Data analysis and quality control of these data are essential and considered an operational task of the SRDC. This is especially important for the Pacific region since data collection there is a variable and sometimes inconsistent process. The main web site for access to the SRDC/EVAC data, analyses, research results, and project descriptions is located at www.evac.ou.edu.
2 DATA COLLECTION AND AVAILABILITY The mission of the SRDC is to collect historical and near-real time rain gauge data only from regions of the world where satellite precipitation verification is especially important, such as in the tropical oceanic regions of the world and where special, high quality, dense rain gauge networks can be found. In addition, these data sets must be independent from the GPCP algorithms. In other words, if the GPCP products are calibrated they should not be calibrated using SRDC data sets. This would, of course, invalidate any verification efforts for such an algorithm. In regions such as the tropical Pacific, dense rain gauge networks are not in existence. However, the SRDC has developed a specialized stochastic method (i.e., the NCR method, see below; Morrissey 1991; Morrissey and Greene 1993) that allows very useful comparisons of the statistical properties of rain gauge-collected data with satellite rainfall estimates. This method was used in the Precipitation Intercomparison Project (PIP-3, Adler et al. 2003) to assess the uncertainties associated with different satellite rainfall estimates over the tropical Pacific region. In regions of the world where high time and space resolution rain gauge networks exist (e.g., the Oklahoma Mesonetwork, Brock et al. 1995), the SRDC maintains a database of interpolated areal estimates of gauge data at scales compatible with the GPCP products. For such networks, the SRDC
Ground Validation for The Global Precipitation Climatology Project
383
has developed stochastic methods to assess the sample error variance and the signal to noise ratio associated these networks (Morrissey et al. 1995b). The mathematical development of these statistics is available on the SRDC web site.
2.1 Pacific rain gauge data Due to important role of the tropical Pacific as the primary driver of the earth’s circulation through the tremendous release of latent heat and its variations, the amount and changes in rainfall in this region must be measured accurately for global climate models to simulate and predict changes in the earth’s weather and climate. These measurements cannot come from the existing rain gauge networks in the Pacific due to the lack of sufficient gauge density. They must come from satellite estimates. However, for verification purposes the existing rain gauge networks are invaluable. A review of the 3rd Algorithm Intercomparison Project (Ebert 1996; Ebert and Manton 1998) indicated that during a comparison of 57 different satellite rainfall algorithms using two shipboard couple ocean atmosphere response experiment (COARE; Webster and Lukas 1992) radars, the radars significantly underestimated rainfall by 30% compared with the average satellite estimate over the same region and time period. A comparison of islandbased rain gauge statistics (Morrissey et al. 1994) showed a strong consistency in the expected rain rate given that it is raining among the four widely separated tipping bucket gauges in the western Pacific. When the expected rain rate given rain is multiplied by the fraction of time raining (within a month) a reasonable estimate of monthly rainfall is obtained. A comparison of monthly rainfall with both the averaged satellite estimates and the COARE radar estimates strongly suggested that it was the radar estimates that contained the largest bias. This was mostly likely the result of calibrating the radar estimates to largely untested optical rain gauges on moored buoys in the western Pacific (Ciesielski 1998). Considering the size of the Pacific basin and importance of rainfall as a tracer of latent heat released, an error of this magnitude most likely results in a similarly sized error in model predictions. Thus, the importance of Pacific rain gauge data cannot be underestimated. The SRDC/EVAC Pacific rain gauge database, PACRAIN, consists of daily and monthly rain gauge data from 643 stations throughout the tropical Pacific, with some records going back into the 1800s. Much of these data have been collected through arrangements with local Pacific meteorological services, as well as New Zealand’s National Institute of Water and Atmospheric Research (NIWA), Meteo-France and US NCDC. The database currently contains over 1.3 million daily observations from 653 sites, extending from 1971 to the present. There are more than 40,000 monthly observations from 201 sites, extending from 1874 through 1970. Daily data
384
M. M. Morrissey and S. Greene
are organized into files by observation site, and monthly data are organized into files by month. There is also an interactive query form, which allows the user to select data based on criteria like date and location. A key component of the SRDC and the Pacrain database is the Schools of the Pacific Rainfall Climate Experiment (SPaRCE; www.evac.ou.edu/ sparce/). The SPaRCE program, funded by NOAA, has over 200 schools, technical centers, and other local organizations across the Pacific interested in taking part in the global climate research effort. Each organization is equipped with a direct read rain gauge, a GPS, a camera, and detailed instructions on setup and maintenance of a professional weather observation site. Several sites are equipped with instruments in addition to rain gauges, such as thermistors and hygrometers. Each local participant group takes daily rainfall measurements, which are quality controlled and then provided to the research community through inclusion in the PACRAIN database. The SPaRCE program also supplies participants with education materials (e.g., books, video tapes) and workshops in an effort to increase awareness of the necessity of enhancing the quality and quantity of Pacific environmental data. It is extremely important to the participants that their efforts make a direct and vital contribution to the global effort of understanding climate change and the potential effects of such a change on their specific locales. The SPaRCE program is an internationally recognized program, and just received an excellent review from the United Nations Economic and Social Commission for Asia and the Pacific (ESCAP) (www.unescap.org/drpad/ vc/conference/ex_pi_167_spr.htm).
2.2 Specialized rain gauge networks In addition to rain gauge data from the Pacific region, the SRDC also archives rainfall estimates from specialized networks located in a variety of countries and the USA. For example, the Oklahoma Mesonetwork (Brock et al. 1995; Morrissey and Greene 1998) has over 100 well-located sites within Oklahoma. The gauge data are collected at 5-min intervals which allows comparisons of analyzed satellite estimates within almost zero time differences. Some of the specialized gauge networks in the SRDC database are located in Darwin, Australia, Florida, Texas, Kenya, Brazil, and a very highdensity gauge network located in South Korea. In addition to these networks, EVAC has installed a very high-density gauge network (i.e., the Piconet) which consists of 15 pairs of gauges, uniformly spaced within 1.0 km area at the Will Rogers World airport in Oklahoma City, Oklahoma. The primary purpose of installing this network was to study the fine scale of rainfall events in an effort to better quantize areal rainfall for larger, less dense gauge networks. A complete description of this network can be found at www.evac.ou.edu/piconet/powerpoint_ams/ piconet_presentation_ files/frame.htm.
Ground Validation for The Global Precipitation Climatology Project
385
3 VALIDATION METHODOLOGIES A main thrust of the SRDC is to provide research in validation methodology as well as provide routine validation of GPCP products. The sections to follow provide an example of some of the statistical validation procedures developed as part of the ground validation efforts of the GPCP. Following that, a few examples of routine validation and simple comparative analysis of rain gauge and GPCP product estimates are shown.
3.1 The spatial sampling error variance A measure of the relative accuracy of a time-space average of areal rainfall taken from a network of rain gauges having any distribution in space can be obtained using a method developed by Morrissey et al. (1995b) based upon a method developed by Parker (1984) for time series. The method provides a baseline error statistic from which to decide whether to utilize a more sophisticated spatial averaging method such as block Kriging (Journel and Huijbregts 1989; Delfiner and Delhomme 1975; Bastin et al. 1984; Bras and Rodriguez-Iturbe 1985; Lebel et al. 1987; Davis 1986). Given a network of gauges, one can place a large square grid containing many sub-grid squares over the network where each sub-grid square contains one or zero gauges. Using this grid setup, the following relation for the error variance was obtained (more details are in Morrissey et al. 1995b):
σ e2 = σ p2 (
m m 2δ ( j ) ρ ( d i , j ) 1 1 m−1 m 2 ρ (d i , j ) + +∑ ∑ − ∑i =1 ∑ j =1 2 m n i =1 j =i +1 m mn
+ ∑i =1
m −1
∑
m j =i +1
2δ (i )δ ( j ) ρ (d i , j )
(1)
n2
where Φ 2p is the variance of the point values (about the long-term mean rainfall), ∆(di,j) is the distance correlation between values located at sub-grid square boxes i (i = 1, 2, … east) and j ( j = 1, 2, south), and ∗(i) is equal to 1 if small box i, j contains a value and zero if it does not. The total number of sub-grid boxes is n and the total number of these boxes that contain a rain gauge is m. The numbering scheme associated with the grid system has no bearing on the resulting standard error equation so long as an appropriate transformation between the index-lagged correction and the distance-lagged correlation is made. Using the method described above, representative locations having low standard error can be selected to perform validation exercises.
386
M. M. Morrissey and S. Greene
3.2 The signal to noise ratio A statistic with which to assess the usefulness of areal rainfall from a gauge network is a combination of two statistics, i.e., the spatial sampling error variance (see above), and the time-space rainfall variance (i.e., Φ 2T). The ratio of the time-space rainfall variance to the spatial sampling error variance (i.e., Φ 2T/Φ 2e) provides an excellent estimate of the signal to noise ratio. The development of the time-space rainfall variance is described by RodriguezIturbe and Mejia (1974) and Morrissey (1991). A detailed description of this method is provided on the SRDC web site.
3.3 The noncontiguous rain gauge method When all that is available within an important region where satellite verification must be conducted is a sparsely distributed network of rain gauges, such as that in the tropic Pacific region, another stochastic method is available for useful assessment of the uncertainty in satellite rainfall estimates. The non-contiguous rain gauge method (i.e., NCR, Morrissey 1991; Morrissey and Greene 1993) uses the second-order statistics of rainfall obtained from widely scattered rain gauges to assess the natural variance of time-space rainfall at the desired scales associated with given values of satellite estimates. This allows one to put error limits on the satellite estimates. The primary assumption required for this method to be valid is that the satellite algorithm uncertainty is constant within the test region. The rain gauge values are also assumed to be homogeneous and stationary. Thus, it is necessary to test for inhomogeneities before this method is applied.
4 EXAMPLE OF VALIDATION WORK: GPCP’S 1.0 DEGREE DAILY AND 2.5 DEGREE MONTHLY RAINFALL PRODUCTS TESTED OVER OKLAHOMA AND THE PACIFIC Much work has been completed comparing the GPCP’s satellite rainfall products over two very different climate regimes, the middle USA (i.e., Oklahoma) and over the tropical Pacific. A summary of a comparison GPCP’s 1.0 degree daily (1DD) products (Huffman et al. 2001) and the GPCP monthly product version 2 (i.e., V2; Huffman et al. 1995, 1997) with gauge data from the Okalahoma Mesonetwork is given below. Boxes at scales comparative with the satellite estimates were selected using the error characterization method described above. The GPCP produces two 1DD products, one which includes information from collocated rain gauge data (i.e., the PSG product) and one that does include this information (i.e., the PMS product). Both products were tested using the Mesonetwork data. It should be noted that the Mesonetwork data are not included in the
Ground Validation for The Global Precipitation Climatology Project
387
development of the PSG product and are thus produce rainfall estimates independent of the PSG product.
4.1 Oklahoma 1.0 degree daily A description of the GPCP 1DD rainfall product is given by Huffman et al. (2001). As an example of the approach the SRDC takes in analyzing the uncertainties associated with satellite estimates, we compared both 1DD product’s values from 1997 to August 2002 with simple areal-averaged rain gauge data from the Oklahoma Mesonetwork shown below. Three 1.0 degree boxes located in northwest, center, and southeast Oklahoma were selected for the comparison (Fig. 1). These boxes were selected due to the relatively low standard error and high signal to noise ratio (see description above).
Figure 1. Three 1 degree boxes contain mesonetwork rain gauges selected for their high signal to noise ratio. The results from each box are shown below using scatter diagrams (the box location is noted in the figure title).
A striking result is that all three boxes indicate very similar correlation and slope values (Fig. 2). A bias appears to be associated with the PSG product with that product overestimating high rainfall. Moreover, the bias which can be observed in each box also appears to be similar in magnitude. This is only an example of the work found at the SRDC web site (additional details and updated results can be found at www.evac.ou.edu/srdc). It should be noted that the non-normality and multi-colinearity inherent in rainfall data strongly suggests that standard linear regression analysis is inappropriate for this type of comparison. Thus, these analyses should only be considered as a first or “quick” look at the comparison between the satellite and rain gauge data.
M. M. Morrissey and S. Greene
388
Figure 2. A comparison between the boxaveraged daily rainfall data from the three Mesonetwork boxes and the GPCP 1 degree daily product.
4.2 Oklahoma 2.5 degree monthly Using the same Mesonet rain gauge data, two 2.5 degree boxes can be used to compare with the GPCP monthly product (Fig. 3). The GPCP V2 product data were compared with box-averaged gauge data from 1994 to August 2002. The results in scatter diagram form are shown in Fig. 4.
Figure 3. The two 2.5 degree monthly boxes selected for the comparison with the monthly GPCP product.
Ground Validation for The Global Precipitation Climatology Project
389
Figure 4. A comparison between the satellite and Mesonetwork box averages for the GPCP version 2 monthly product.
The monthly product over Oklahoma appears to be relatively unbiased and relatively accurate compared to the 1DD product. However, at large satellite value the Mesonetwork box averages tend to underestimate rainfall compared to the satellite value. Too much cannot be read into the accuracy of the product from this preliminary comparison due to de the reasons mentioned above. However, these “quick looks” do suggest areas of further, more in-depth research. Time series and error and bias comparison are also available on the SRDC web site.
4.3 PACRAIN 2.5 degree monthly Three boxes were selected within the Pacific region (Fig. 5) which contained a sufficiently low standard error and significantly high signal to noise ratio to allow a quick look at the accuracy of the monthly GPCP V2 product. It should be noted that gauge data are not assimilated into the monthly GPCP product. The scatter plots of the comparison for the three boxes are shown in Fig. 6. The record of comparison was from 1979 to August 2002. An initial analysis of the scatter diagrams suggests a statistically unbiased relationship with the possible exception of box 1,725. A quick look into the gauge data records for that box indicates that the records are quite sporadic and that the comparison may have been contaminated with poor gauge data and low spatial temporal and sampling. Assuming accurate gauge measurements, another look at the results seem to suggest perhaps a small underestimate by the GPCP monthly product of approximately 500 mm per month. While these comparisons are not statistically rigorous, they do provide the producers of the GPCP satellite algorithms with a quick look at the comparisons and allow them to pursue more stringent research methods if necessary.
390
M. M. Morrissey and S. Greene
Figure 5. Tropical western Pacific region. Outlined boxes are those having the highest signal to noise ratio for the monthly data. The crosses indicate rain gauge sites.
Figure 6. The scatter plots associated with each of the three Pacific 2.5 degree monthly boxes.
Ground Validation for The Global Precipitation Climatology Project
391
5 SUMMARY The SRDC’s role in the GPCP is to provide researchers with hard-to-find quality rain gauge data sets and to provide initial comparisons between the GPCP products and analyzed and interpolated rain gauge data sets. In addition, the SRDC researchers actively produce statistical methods which researchers can use to assess the usefulness of a set of rain gauge data and to perform statistical comparison of satellite estimates with sparsely distributed rain gauge data in region like the tropical Pacific.
6 REFERENCES Adler, R. F., C. Kidd, M. Goodman, A. Ritchie, R. Schudalla, G. Petty, M. Morrissey, and S. Greene, 1996: PIP-3 Intercomparison Results. Precip. Intercomparison Proj (PIP-3) Workshop, 18–20 November 1996, College Park, MD. Bastin, G., B. Lorent, C. Duqué, and M. Gevers, 1984 Optimal estimation of the average areal rainfall and optimal selection of rain gauge locations. Water Resour. Res., 20, 463–470. Bras, R. L. and I. Rodriguez-Iturbe, 1985: Random Functions and Hydrology, Dover, NY, 559 pp. Brock, F. V., K. C. Crawford, R. L. Elliott, G. W. Cuperus, S. J. Stadler, H. L. Johnson, and M. D. Eilts, 1995: The Oklahoma Mesonet: a technical overview. J. Atmos. Oceanic Technol., 12 (1), 5–19. Ciesielski, P. E. and R. H. Johnson, 1999: Precipitation estimates for the COARE intensive observing period. Proc. Conf. on the TOGA Coupled Ocean-Atmosphere Response Experiment (COARE98), Boulder CO, 7–14 July 1998, pp. 193–194, WCRP-107, Geneva, Switzerland. Davis, J. C., 1986: Statistics and Data Analysis in Geology. John Wiley, New York, 646 pp. Delfiner, P. and J. P. Delhomme, 1975: Optimum interpolation by kriging. In: Display and Analysis of Spatial Data, edited by J. C. Davis and M. J. McCullagh, John Wiley, New York, pp. 96–114. Ebert, E. E., 1996: Results of the 3rd algorithm intercomparison project (AIP-3) of the Global Precipitation Climatology Project (GPCP). BMRC Report No. 55, BMRC, GPO Box 1289K, Melbourne, Vic., Australia 3001, 199 pp. Ebert, E. E. and M. J. Manton, 1998: performance of satellite rainfall estimation algorithms during TOGA-COARE. J. Atmos. Sci., 55, 1537–1557. Huffman, G. J., R. F. Adler, M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multiobservations. J. Hydrometeor., 2, 36–50. Huffman, G. J., R. F. Adler, P. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The Global Precipitation Climatology satellite Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 5–20. Huffman, G. J., R. F. Adler, B Rudolf, U. Schneider, and P. R. Keehn, 1995: Global Precipitation estimates based on a technique for combining satellite-based estimates, raingauge analysis and NWP model precipitation information. J. Climate, 8, 1284–1295. Janowiak, J. E. and P. A. Arkin, 1991: Rainfall variations in the tropics during 1986–1989. J. Geophys. Res., 96, 3359–3373. Janowiak, J. E., P. A. Arkin, P. Xie, M. L. Morrissey, and D. R.Legates, 1995: An examination of the east Pacific ITCZ rainfall distribution. J. Climate, 8, 2810–2838. Journel, A. G. and C. J. Huijbregts, 1989: Mining Geostatistics. Academic Press, San Diego, CA, 600 pp.
392
M. M. Morrissey and S. Greene
Lebel, T., G. Bastin, C. Obled, and J. D. Creutin, 1987: On the accuracy of areal rainfall estimation: a case study. Water Resour. Res., 23, 2123–2134. Morrissey, M. L, 1991: Using sparse rain gages to test satellite-based rainfall algorithms. J. Geophys. Res., 96, 18561–18571. Morrissey, M. L. and J. S. Greene, 1993: Comparison of two satellite-based rainfall algorithms using Pacific atoll rain gage data. J. Appl. Meteor., 32, 411–425. Morrissey, M. L. and Y. Wang, 1995: Verifying satellite microwave rainfall estimates over the open ocean. J. Appl. Meteor., 34, 794–804. Morrissey, M. L., M. A. Shafer, S. E. Postawko, and B. Gibson, 1995a: Pacific rain gauge data. Water Resour. Res., 31, 2111–2113. Morrissey, M. L., J. A. Maliekal, J. S. Greene, and J. Wang, 1995b: The uncertainty of simple spatial averages using rain gauge networks. Water Resour. Res., 31, 2011–2017. Parker, D. E., 1984: The statistical effects of incomplete sampling of coherent data series. J. Climatol., 4, 445–449. Postawko, S. E., M. L. Morrissey, and B. Gibson, 1994: Schools of the Pacific Rainfall Climate Experiment: combining research and education. Bull. Amer. Meteor. Soc., 75, 2296–2311. Rodriguez-Iturbe, I. and J. M. Mejia, 1974: The design of rainfall networks in time and space. Water Resour. Res., 10, 713–728. Webster, P. J. and R. Lukas, 1992: TOGA COARE: The Coupled Ocean-Atmosphere Response Experiment. Bull. Amer. Meteor. Soc., 73, 1377–1416.
31 VALIDATION OF RAINFALL ALGORITHMS AT THE NOAA CLIMATE PREDICTION CENTER John Janowiak NOAA Climate Prediction Center, Camp Springs, MD, USA
1 INTRODUCTION The validation of precipitation estimates from satellite is an essential activity for numerous reasons, among them being to assess the skill of the estimation algorithms, to provide users with the accuracy of them, and to provide feedback to algorithm developers that may potentially help improve their methodology. The primary motivation for the validation activity that will be discussed in this paper is directed toward the latter, i.e., to provide useful feedback toward the improvement of satellite estimates of precipitation. Validation activities can be conducted at various time and space scales depending on the availability of reference data sets that are suitable to be used as “truth”. Ideally, these reference data sets have known error characteristics that can be incorporated into the validation process. In reality, however, this information is not widely available and is limited to relatively small-scale regions with high rain gauge density. In this paper, the focus is on a continental-scale validation effort over the USA. The system that has been implemented for the USA has been modeled after the excellent work of Dr. E. Ebert of the Australian Bureau of Meteorology Research Center (BMRC) who implemented a validation system over Australia for the International Precipitation Working Group (IPWG) and which is discussed in a separate paper of this section.
2 VALIDATION DATA Two sources of precipitation validation data are available over the USA. One is composed of rain gauge data while the other is radar. Rain gauges provide 393 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 393–401. © United States Government 2007.
J. Janowiak
394
the only totally direct measurement of precipitation at the surface and provide measurements that are continuous in time. However, rain gauges provide point measurements that are not spatially complete and formidable gaps in coverage exist in certain areas of the country (Fig. 1). Conversely, radar provides much better spatial coverage than gauges and the spatial coverage characteristics are quite similar to the satellite estimates, although radar does not provide a direct estimate of precipitation. By comparing the satellite estimates to both the rain gauge and radar data, a more complete validation exercise can be conducted than if only one source were used.
Figure 1. Typical distribution of rain gauge data.
2.1 Rain gauge analyses The main validation data that are used for this effort are objective analyses of over 7,000 rain gauge observations over the continental USA that provide rainfall totals over the 24-h period from 1,200 UTC to 1,200 UTC. These observations are objectively analyzed to a 0.25° × 0.25° latitude/longitude grid using a modified Cressman (1959) technique. While it is widely known that this analysis procedure tends to spread out light values of rainfall and also to deamplify heavy values, this rainfall validation data set provides the best available nationwide precipitation information on a daily basis in the USA at this point in time. A typical distribution of the rain gauge locations that are input to the analysis scheme is shown in Fig. 1. As part of the analysis process,
Validation of Rainfall Algorithms at the NOAA Climate Prediction Center
395
various types of quality control are performed, including climatological checks, “buddy” checks (i.e., comparisons with neighboring values), and a check against IR satellite data to ensure that clouds are present when rainfall has been reported. Further information is available from Higgins et al. (2000).
2.2 Radar All of the satellite-generated rainfall estimates are validated against the NEXRAD “Stage II” radar product that is produced by NOAA/NWS in addition to the rain gauge analyses. The Stage II radar estimates are not integrated with rain gauge observations, although a large-scale bulk correction based on rain gauge data is used to reduce bias in these estimates (R. Kuligowski, personal communication, Camp Springs 2005). Although the radar estimates are not considered as accurate as data from rain gauges, the spatial density that radar offers is more closely matched to the satellite estimates. The radar data are interpolated to a 0.25° × 0.25° latitude/longitude grid that matches the rain gauge analysis grid.
3 SATELLITE AND NUMERICAL MODEL ESTIMATES At the time of this writing, 14 different satellite estimates are validated along with short-term forecasts of precipitation from two numerical forecast models. The precipitation estimates are interpolated to match the validation grid, if necessary, using a Bessel interpolation scheme. There is a variety of information that is used by the various satellite precipitation estimation techniques. Some use only passive microwave information, some use only infrared data, and others use both. A list of the precipitation estimates that are presently validated is presented in Table 1. The validation system is designed with flexibility so that results from new algorithms can be incorporated rather easily.
4 VALIDATION STATISTICS AND RESULTS The validation over the USA is conducted on a continental scale, i.e., comparisons are made for the USA as a whole (excluding Alaska and Hawaii). Validation statistics that use the rain gauge analyses as the reference standard also include the northern half of Mexico. Only those locations where nonmissing estimates are available for all of the satellite techniques are used when computing statistics to ensure that the same domain is validated for each of the estimates. This is important over the USA particularly in winter because some techniques cannot provide precipitation estimates over snow-covered surfaces. Radar estimates are used both as a validating tool, and as a “contestant” in which the radar estimates are compared to the rain gauge analyses. The
J. Janowiak
396
statistics discussed in the following section are very elementary tools that are described in most statistics texts. The definitions below are condensed from Wilks (1995). Table 1. List of products that are presently being validated over the USA (TRMM Tropical Rainfall Measuring Mission (Simpson et al. 1988)).
Product name IR, passive microwave-blended methods: NRLGEO 3B42RT CMORPH PERSIANN Passive microwave only MWCOMB 3B40RT NRBLD AMSU IR only GPI IRRAIN HYDROE 3B41RT IR and visible data GMSRA GMSRAD NWP model forecasts GFS NOGAPS
Source, definition Turk et al. (2003) NASA/TRMM: merge of products “3B40RT” and “3B41RT” Joyce et al. (2004) Sorooshian et al. (2000) NOAA/CPC: merge of SSM/I, TMI, and AMSU-B precipitation estimates NASA/TRMM: merge of SSM/I and TMI-derived precipitation estimates Naval Research Lab: merge of SSM/I, TMI, and AMSU-B precipitation estimates Ferraro et al. (2000) Arkin and Meisner (1987) NOAA/CPC NOAA/NESDIS – also uses model data NASA/TRMM: IR estimates calibrated by passive microwave Ba and Gruber (2001) NOAA/NESDIS: also uses radar data NOAA/NWS: Global Forecast System model (formerly “MRF”) US Naval Research Laboratory global forecast model
4.1 Statistics for assessing errors in magnitude Mean rain rate. This is simply the average of the daily mean rainfall rate over the entire validation domain. The difference between the mean of the remotely sensed estimates and observations is the mean bias. Mean absolute error. This statistic is the country-wide average of the absolute difference (i.e., negative differences are changed to positive) between the estimates and observations. Absolute errors retain the differences in magnitude that would otherwise be reduced because positive and negative differences would cancel each other to some degree.
Validation of Rainfall Algorithms at the NOAA Climate Prediction Center
397
Root mean square error. Similar to mean absolute error, except that the differences are squared before summing, and the square root of the average is derived. Maximum rain rate. This is simply the largest value recorded each day regardless of location. This statistic is useful to define the range of values of the satellite estimates compared to observations when collected over a sufficient length of time.
4.2 Statistics for assessing errors in the spatial distribution of precipitation All of the statistics discussed in this section, with the exception of spatial correlation, are computed from a 2 × 2 contingency table that summarizes “hits” and “misses” for two categories: (1) daily rainfall amounts less than 1 mm and (2) daily rainfall of 1 mm or more. These statistics are used to assess the rainfall detection aspects of the estimates and not the magnitudes. Spatial correlation. This is the correlation between an estimated field and observed field that is conducted on daily data which yields information about the degree of agreement in the spatial variability of the two fields. Probability of detection (POD). The ratio between the number of occurrences where satellite estimates of rainfall > 1 mm day–1 were correctly observed and the total number of observations of rainfall exceeding that threshold. Values range from the worst possible score of “0” (precipitation never correctly detected by satellite) to the best possible score “1” (precipitation always correctly detected by satellite). False alarm ratio (FAR). The ratio between the number of satellite estimates of rainfall >1 mm day–1 that were detected incorrectly and the total number of satellite estimates that exceeded that threshold (whether correct or not). Values range from a perfect score of “0” (precipitation never observed when not detected by satellite) to the worst possible score of “1” (precipitation never observed when detected by satellite). POD and FAR should always be considered together, because good or even perfect values in either case individually are easily obtained. For example, if an algorithm has a wet bias such that it always detects rain everywhere all the time, the POD will be perfect (“1”). However, the FAR will also be close to the worst score (“1”). Bias ratio. The ratio between the number of satellite estimates > 1 mm day–1 and the number of observed amounts that exceeded that threshold. A value of “1” is perfect. Skill score. A score that attempts to assess the skill of the estimates with respect to random chance. The scores usually range from “0” (no skill over chance) to “1” (perfect skill).
398
J. Janowiak
Figure 2. Time series of statistics of a comparison with validating rain gauge analyses over the USA during June–August, 2003. Thick line is radar, dotted line is GFS model forecasts, thin solid line is the satellite estimate with the best statistic for each day.
4.3 Validation results A comparison of several estimates of precipitation with the validating rain gauge analyses over the USA are now discussed. Results are presented for both the warm season, when precipitation is primarily convective in nature, and for one cool season when stratiform precipitation dominates. The warm season results, which are displayed in Fig. 2, show time series of spatial correlation and Heidke skill score for radar (thick solid line), the NWS/NCEP global forecast model (GFS) 12–36-h precipitation forecast (dotted line), and the best score by any of the satellite estimates for a given day (thin solid line). For almost every day during June–August 2003, the radar performs best and the model forecasts worst compared to the rain gauge analyses. Note that the satellite estimates are very close to the radar values in both statistics over the entire 92-day period. In contrast, the GFS model predictions perform much better during the cool season (Fig. 3) and the performance measures are much closer among the radar, satellite, and model. In fact, the model forecasts often outperform the radar and satellite estimates both in terms of spatial correlation and skill during the cool season.
Validation of Rainfall Algorithms at the NOAA Climate Prediction Center
399
Figure 3. Same as Fig. 2, except for September–November.
Over the course of analyzing the validation results during the US summer season, a consistent bias in the satellite estimates was “rediscovered”, namely persistent overestimates in the western states. This observation is depicted in Fig. 4, which depicts areas of eastern Montana with rainfall amounts in excess of 40 mm day–1 from the satellite estimates but amounts of less than 5 mm day–1 from the rain gauge data. Note that radar also overestimates considerably although the amounts are somewhat lighter than the satellite estimates. To ensure that the gauge analysis was not in error, the gauge results were verified by contacting the Glasgow, Montana NWS forecast office, who in turn verified the precipitation measurements of cooperative observers in the region. This “rediscovery” is consistent with the earlier studies of Scofield (1987), Rosenfeld and Mintz (1988), and more recently McCollum et al. (2001) who found that significant evaporation occurs in semiarid regions between the cloud base and surface. In fact, Rosenfeld and Mintz (1988) estimate conservatively that 30% of the rainfall evaporates in the first 1.6 km below the cloud base in semiarid regions at rainfall intensities as high as 80 mm h–1. One way to account for this overestimation is to use relative humidity data to modulate the rainfall estimates. Scofield (1987) adopted this approach by using the mean humidity from the surface to 500 hPa from numerical forecast model analyses.
400
J. Janowiak
Figure 4. Rainfall over the 24-h period 1200 UTC 12 August 2003–1200 UTC 13 August 2003.
5 SUMMARY The example above is just one way that information from continental-scale validation efforts as described in this paper can provide helpful feedback to algorithm developers who can then modify and improve their estimation techniques. A similar continental-scale validation effort is underway over Australia and in the planning stages over Europe and Brazil. Certainly, several such efforts over different climatological regions have the potential to provide substantial useful feedback that will help the precipitation estimation algorithm community.
Validation of Rainfall Algorithms at the NOAA Climate Prediction Center
401
There are obviously other aspects of validation that can be conducted than have been presented here. In addition to such large-scale examinations, comparatively small-scale validation efforts over regions with dense groundbased validation data and known error characteristics would complement this validation effort in a positive fashion. Efforts such as these are underway in the USA under the direction of the IPWG and the Global Precipitation Climatology Project (GPCP), both of which are sponsored by the World Meteorological Organization.
6 REFERENCES Arkin, P. A. and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the western hemisphere during 1982–1984. Mon. Wea. Rev., 115, 51–74. Ba, M. B. and A. Gruber, 2001: GOES multispectral rainfall algorithm (GMSRA). Bull. Amer. Meteor. Soc., 40, 1500–1514. Cressman, G. F., 1959: An operational objective analysis system. Mon. Wea. Rev., 87, 367– 374. Ferraro, R. R., F. Weng, N. C. Grody, and L. Zhao, 2000: Precipitation characteristics over land from the NOAA-15 AMSU sensor. Geophys. Res. Lett., 27, 269–2672. Higgins, R. W., W. Shi, E. Yarosh, and R. Joyce, 2000: Improved United States precipitation quality control system and analysis. NCEP/Climate Prediction Center ATLAS No. 7, 40 pp., Camp Springs, MD 20746 USA. Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydromet., 5, 487–503. McCollum, J. R., W. F. Krajewski, R. R. Ferraro, and M. B. Ba, 2002: Evaluation of biases of satellite rainfall estimation algorithms over the continental United States. J. Appl. Meteor., 41, 1065–1080. Rosenfeld, D. and Y. Mintz, 1988: Evaporation of rain falling from convective clouds as derived from radar measurements. J. Appl. Meteor., 27, 209–215. Scofield, R. A., 1987: The NESDIS operational convective precipitation estimation technique. Mon. Wea. Rev., 115, 1773–1792. Simpson, J. R., R. F. Adler, and G. R. North, 1988: A proposed Tropical Rainfall Measuring Mission (TRMM) satellite. Bull. Amer. Meteor. Soc., 69, 278–295. Sorooshian, S, K. Hsu, X. Gao, H. V. Gupta, B. Imam, and D. Braithwaite, 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 2035–2046. Turk, F. J., E. E. Ebert, B.-J. Sohn, H.-J. Oh, V. Levizzani, E.A. Smith, and R. Ferraro, 2003: Validation of a global operational blended-satellite precipitation analysis at short time scales. 12th AMS Conf. on Sat. Meteor. and Ocean, CD-ROM, 13–17 February, Long Beach, CA. Wilks, D. S., 1995: Statistical methods in the atmospheric sciences. Academic Press, San Diego, CA, 467 pp.
32 GROUND NETWORKS: ARE WE DOING THE RIGHT THING? Witold F. Krajewski IIHR-Hydroscience & Engineering, University of Iowa, Iowa City, IA, USA
Abstract
The author discusses principal issues in designing ground-based networks of rainfall observing instruments for validation of space-based methods of rainfall estimation. He considers tipping bucket rain gauges, disdrometers, vertically pointing and X-band polarimetric scanning radars. The design issues emphasize the need for studies of small-scale rainfall variability, characterization of the measurement error of the instruments used, and observations of spatial and temporal variability of rainfall characteristics at scales smaller than those of the space-based methods.
1 INTRODUCTION As this entire volume testifies, there is a lot of excitement in the science community about using satellites for observing rainfall from space. There are many scientific applications where such observations offer new insights into the workings of the coupled atmosphere, land, and ocean system. For many of those applications, assessment of our ability to quantify rainfall using the signal measured by the satellite-based sensors is not critical. For other applications, such as agriculture, prediction of precipitation-induced natural hazards, and operation of water resources systems, considerable level of quantitative performance is necessary (e.g., Nijssen and Lettenmaier 2004.) The term validation is often used as a synonym of our attempts to evaluate algorithm or model performance. However, despite its popular use few authors bother to define the precise meaning of the term in their studies. To some, validation is the process of quantitative evaluation of the outcomes of algorithms and models and their subsequent improvements. Others view it as a statistical characterization of the outcome’s uncertainty (Krajewski and Smith 2002; Gebremichael et al. 2003). In this article, we adopt the latter view as 403 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 403–417. © 2007 Springer.
404
W. F. Krajewski
availability of such statistical description of outcome’s uncertainty provides a reference against which improvements can be measured. As the focus of this volume is on precipitation, we refer to the outcomes as precipitation products or maps. It is thus clear that mere comparison of products of two or more different algorithms cannot be considered as validation in the above sense. Still, such exercises are valuable and we will term them assessment or evaluation. These two terms are broad enough to include many different studies where the precise meaning of the nature of the investigation is not critical.
2 OBJECTIVES AND SCOPE Based on the above discussion, we define the objectives of this paper. We discuss a variety of issues related to the use of ground-based sensors as they promise to provide information relevant to validation of space-based rainfall products. We emphasize the estimation of rainfall rather than snowfall but much of the discussion is relevant to both. We set forth as the goal for validation to be the determination of the space-time joint distribution of the product errors. Determination of probability distribution function (pdf) of errors for a fixed space and time scale, the pdf conditional on rainfall intensity or accumulation for a fixed space and time scale, the spatial correlation function for fixed space and time scale and the temporal correlation function for fixed space and time scale are all example of special cases of the general problem and thus somewhat simpler. This by no means implies that these problems are simple to solve as we are still far away from solving any of them. In essence, this paper presents our view of a research agenda necessary for a practical solution of the validation problem for satellite-based rainfall estimation much as Krajewski and Smith (2002) did for radar-rainfall estimation. Before we proceed, let us define what we mean by error. Two fundamental quantities of interest are the difference and the ratio between the true and estimated quantity. The choice between them is not obvious and depends mainly on whether the mechanism causing error is additive or multiplicative. This is often unknown and therefore we recommend analysis of both, whenever possible. This approach has always provided interesting insights. Mathematical convenience is often a useful guiding aspect, which is fine, as long as it does not lead to unverifiable assumptions. As a general principle, based on the central limit theorem, uncertainties of products at large scale tend to be Gaussian and thus amenable to additive error models. How large is large is not well explored in the case of rainfall and thus, to some degree, is a subjective matter. The error of satellite rainfall products can be conceptualized as having three basic components: sampling, estimation, and instrumental effects. These components are not independent. Instrumental effects are typically small, which does not mean that they are negligible. The estimation error
Ground Networks: Are We Doing the Right Thing?
405
includes the conversion of the measured signal into the estimation quantity and thus contains algorithm error, the propagated effect of the instrumental error, the spatial sampling (known as field of view) error and the georeferencing error. If the product is rainfall accumulation over some time scale (e.g., daily), it also contains the temporal sampling error. Thus, the sampling error does not affect other errors if the product is rainfall intensity. Many studies have been conducted on the sampling error (e.g., Bell et al. 1990; North et al. 1993; Oki and Sumi 1994; Bell and Kundu 1996, 2000) of time-integrated products. Most assumed perfect (error-free) rainfall intensity fields. Relatively few studies address the total error (Krajewski et al. 2000; Gebremichael et al. 2003) or error decomposition (e.g., Short and North 1990; North and Polyak 1996; McCollum and Krajewski 1998). Since it is the total error that is of interest to users of products, we focus on a strategy for obtaining its statistical description using ground-based networks of sensors. In our discussion we emphasize the instrumental and estimation error as opposed to the sampling error on the basis that the latter has been studied much more extensively. Establishing ground-based observing systems for quantitative assessment of satellite-based estimates of rainfall is fundamentally important and challenging (e.g., North and Nakamoto 1989). According to our definition of validation, ground-based estimates of rainfall should serve as a reference for error studies of space-based products only if their own estimation error can be characterized, or if it can be shown that they are much more (order of magnitude) accurate than the space-based products. There are several types of sensors that need to be considered: single- and multiparameter scanning radar, vertically pointing radars (i.e., profilers) with their ability to provide detailed vertical view of precipitating atmosphere, disdrometers, and networks of rain gauges. Clearly, scanning radar offers probably the most promise due to its ability to provide spatially continuous and temporally frequent observations of rainfall. However, the situation with validation of its products is about the same as we discussed for the satellite case: we do not know the error structure of radar-rainfall products. Since Krajewski and Smith (2002) and Krajewski and Ciach (2003) provide recent relevant discussion, we limit our scope to other sensors and issues. The exception is networking of small, inexpensive radars which we discuss in Section 5.
3 RAIN GAUGE NETWORKS We begin with rain gauge networks and the characteristics they should have to serve as useful reference for satellite-based products. First, we address the issue of random error of tipping bucket rain gauge. To date we know of two rigorous studies on this subject. Habib et al. (2001) performed a data-based simulation and estimated the error distributions as function of rainfall
406
W. F. Krajewski
magnitude, time integration scale, and bucket size. Recently, Ciach (2003) conducted an experimental study and synthesized the results for two different methods of tip data processing. His results are in good agreement with the earlier study by Habib et al. (2001). A very interesting result is the scaling behavior of the errors. These results are also confirmed by analysis from the double gauge data at the Iowa City Municipal Airport. The main conclusion from these studies is that tipping bucket rain gauges, when well maintained and deployed as a pair, provide accurate observation of rainfall accumulations at scales from 10 min up. The standard errors decrease with increasing rain amount and time integration scale. Deploying the tipping bucket rain gauges in pairs, advocated by Ciach and Krajewski (1999) and Steiner et al. (1999), has many advantages. The most important one is data quality control. Since rainfall displays significant variability in space and time, analysis of “reasonableness” of single gauge record often passes cases of gauge malfunctioning. For example, a piece of dirt partially blocking inlet of the funnel will change estimated rain intensity without making it look suspicious. The main assumptions of the concept of using two gauges at a single location are: (1) it is highly unlikely that two gauges would fail in exactly the same way; and (2) that rainfall variability at the scale of gauge separation distance (~1 m) is negligible. Thus, when the gauges function well the data show good agreement with each other. A disagreement is a sign of at least one of them malfunctioning. The site should be checked by a qualified technician. Note, that if malfunction can be attributed to one of the gauges, the data is not lost at the given site as the second gauge worked properly. An added benefit of having a pair of gauges is a reduction of the random error (Ciach 2003). The second issue is that of designing networks of gauges for validation studies. Operational networks are of little help as their average separation distance is typically larger than the scale of interest to validation. The density of operational rain gauge networks varies from country to country, but it is safe to assume that it is rarely higher than 0.002 km–1. As satellite maps of rainfall can have a resolution on the order of 25 km2 , such density is clearly not adequate. Consider two objectives for design of validation networks: (1) characterizing statistical behavior of rainfall in space and time; and (2) estimating rainfall over an area as accurately as possible. It turns out that these two objectives lead to quite different network configurations. To illustrate this, let us pose a question relevant to direct validation (Krajewski and Smith 2002): “How many gauges are required in a 2 × 2 km2 area (equivalent to a typical radar-rainfall product) to obtain areal estimate with high accuracy (say, better than 5%)?” If the accuracy is specified in terms of mean square error, answering this question requires knowledge of the spatial covariance function of the relevant rainfall regime. To estimate the shape of covariance (or correlation) function of rainfall intensity or accumulation over short time
Ground Networks: Are We Doing the Right Thing?
407
scales the network has to sample several separation distances. In particular, accuracy of gauge-based estimated rainfall over an area comparable is size (linear) with the decorrelation distance of rainfall is sensitive to the shape of the covariance function near the zero separation distance (see Habib and Krajewski (2002) for an illustration). Thus, the network has to include gauges that are close to each other as well as far from each other. Formal statement of the design problem is not straightforward especially when little is known about the shape of the function the network is supposed to identify. Prior experience is limited but it could be helpful. Habib and Krajewski (2002) estimated correlation functions for Florida summer rainfall. Krajewski et al. (2003) estimated it for five different locations around the world and reviewed early experiments (in the 1960s and 1970s) motivated by the needs of the communication community (Crane 1990). However, these studies were based on short-lived experiments and the analyses are subject to significant sample size and other limitations. In particular, short-duration experiments and thus small samples prevent conditional analyses. Identifying the effects of rainfall intensity and amount, season, rainfall type, and linking these to synoptic situation and other observables is important for the validation problem but impossible with small samples. For example, when designing an experimental network at the Iowa City Municipal Airport in 1998, we included inter-gauge distance as short as 10, 100, 200, and 500 m. We have learned since that there is little variability at the scale of 200–500 m. Based on our experience thus far, distances as short as 250 m should be included in networks designed for studying the shape of covariance function in the tropics and 500 m in midlatitudes. Other statistics may require different separation distance considerations. Now let us turn our attention to the second objective, i.e., estimating rainfall over an area. Here, the network design problem has been well studied in the past as hydrologists need mean areal rainfall as input to their rainfall-runoff models (e.g., Bras and Rodriguez-Iturbe 1985). In general, the gauges should be organized on a uniform grid covering the area of interest. Such design maximizes rain cell detection. Estimation of network sampling errors associated with simple averaging of rain gauge values can be accomplished numerically using the methodology proposed by Morrissey et al. (1995). Thus, questions on network density required for achieving certain levels of accuracy, or questions of network expansion to improve accuracy can be easily studied if the rainfall spatial covariance function is known. Based on earlier studies of such function (e.g., Krajewski et al. 2003), we attempt to answer the question of the number of gauges needed to estimate rainfall over a 2 × 2 km2 pixel with accuracy better than 5%. We consider two cases: exponential decay with the correlation distance of 5 and 15 km. In the first case we need about 20 gauges uniformly covering the pixel to reach the 5% error level. In the second case only 5–8 gauges will achieve the same objective (Moore et al. 2000). The pixel size is more appropriate to the
408
W. F. Krajewski
validation of radar-rainfall problem. For the scale of a satellite rainfall product pixel (about 5 × 5 km2), 5% error level can be achieved with about 35 gauges for highly variable rainfall and with only about 10–12 for the less variable regime typical of midlatitudes. We present these cases to show that solving the validation problem is both feasible and not terribly expensive (i.e., compared to the cost of building and maintaining a satellite). Rain gauge networks of the size we mentioned above can be easily deployed and operated. As a matter of fact, the smaller the area the easier and cheaper it is to maintain a network. For example, clusters of the 2 × 2 km2 size can be placed at the airport where many other meteorological instruments are often placed. Cell phone-based data communication provides a simple way of near real-time operation and dual-gauge design provides means for demand-only maintenance visits (Kruger and Kanukurthy 2004). Since optimal network configurations may be difficult to archive in practice due to constraints imposed by the existing infrastructure or its lack, it is useful to study the sensitivity of a particular design using geographic information system (GIS) technology. Krajewski and Goska (2004) developed an ArcView GIS utility that calculates mean square error associated with a certain network configuration over an arbitrarily shaped area. Another rain gauge network design objective results from considerations of ground networks that include scanning radar. Although the subject of radar-rainfall estimation is outside the scope of the current paper, it is well known that radar-rainfall uncertainty depends on radar range. Thus, if we are to benefit from radar-estimated rainfall for the validation of satellite rainfall products, we need to organize observational networks that can address the radar-range effects (Krajewski and Ciach 2003). A simple solution is placing dense clusters along a radar beam. The question is “How many?” and “How should they be spaced?” The answer depends on the operating range of a given radar. There are two major effects that the spacing should be able to capture. The first is the detection of the so-called bright band (reference). In some rainfall regimes the effect may easily take place at a radar range shorter than 100 km. The second effect is overshooting precipitating clouds. The range of this effect is longer than the bright band effect and depending on the rain regime may take place anywhere between 100 and 200 km. Studies by Smith et al. (1996) demonstrate the shape of the range-dependent bias in early NEXRAD system estimates of rainfall in the USA while the theoretical considerations made by Krajewski and Vignal (2004) give a more general means of estimating both bias and error variance. Their model could be used as guidance in designing the cluster locations to capture the range effect shape. Another solution would be to place small cluster of about four sites every 10 km along the radar range.
Ground Networks: Are We Doing the Right Thing?
409
4 DISDROMETER NETWORKS Rain gauge networks provide surface-based information on the main variable of interest, i.e., rainfall on the ground as this is the variable that affects hydrologic response and thus is of concern to operators of water resources systems. However, much more general information about rainfall processes is included in the measurements of surface drop size distribution (DSD). From DSD data, not only point rainfall can be calculated, but also many other quantities relevant to remote sensing of rainfall and its hydrologic applications. For example, one can calculate variables such as radar reflectivity, optical extinction, and kinetic energy. Thus, why not simply replace the rain gauges in the experimental (i.e., validation) networks with disdrometers? The main answer is the cost of the instruments. The cost of a disdrometer may be an order or even two orders of magnitude higher than the cost of a tipping bucket rain gauge. This situation, however, is quickly changing. There are several new models of optical disdrometers on the market that are priced competitively with some high-end rain gauges. There are also efforts within the research community to further drive this cost down by exploring new design ideas. Another important consideration in wide-scale deployment of disdrometers is their power consumption and operation. Virtually all current design requires power supply and housing for a computer that control data acquisition system. Clearly, for the devices to be ready for remote operation both issues need to be addressed. The devices need to be low-power so that they can be operated from a solar-panel rechargeable battery and have to have a computer processor and data acquisition system embedded in their design. Technologies for addressing both issues exist and will not increase the overall cost of the devices significantly. Since disdrometers measure DSD indirectly and the cumulative experience with their operation is much less than in the case of rain gauges, they require thorough testing. Several intercomparison experiments point to sensitivity of the obtained results to the instrument type (Sheppard and Joe 1994; Campos and Zawadzki 2000; Williams et al. 2000; Tokay et al. 2001; Miriovsky et al. 2004.) These experiments used different types of instruments collocated or in a close proximity of each other. To distinguish the sampling error effects associated with a particular instrument from the crossinstrument differences it is necessary to compare several collocated disdrometers of the same kind. Ciach (2003) conducted and documented such an experiment using tipping bucket rain gauges but we do not know of a similar experiment using disdrometers. Another issue is the limited sampling area of the present-day disdrometers. The sampling area of some 50 cm2, typical for optical devices causes significant sampling errors (Smith 1993; Jameson and Kostinsky 2002) Mini radar-type devices such as the POSS (Sheppard 1990) suffer less from this
410
W. F. Krajewski
problem but require making assumptions on the velocity versus drop size relationship necessary for estimating the DSD. This is an important issue as some studies indicate often significant departures of the measured velocity from the often-used experimentally determined terminal velocity relationships (e.g., Gunn and Kinzer 1949; Beard 1976) It is also likely that the drop velocity near the ground surface is different from the terminal velocity aloft. This could be caused by turbulence, vertical air motions associated with updrafts and downdrafts, and other effects. Therefore, we recommend using instruments that have the ability to measure drop velocity independently of their size. Clearly, the POSS provides measurement of the drop velocity spectrum (through Doppler effect), but not independently of the size measurement. Other DSD measurement issues include the effect of the wind and the measurement of the drop shape. The two-dimensional (2D) video disdrometer (e.g., Kruger and Krajewski 2002) is capable of providing information on drop shape. This is important for the interpretation of polarimetric radar observations of rainfall (Bringi and Chandrasekar 2001). The wind effect is associated with air flow distortion by the instrument itself. For example, Nešpor et al. (2000) demonstrated that the large size and the shape of the 2D video disdrometer made by Johanneum Research in Austria, under certain conditions may cause significant distortion of the observed DSD. Other optical instruments with smaller structures are less susceptible to this effect, but may suffer from strong directional dependence on wind directions. With the current-day computational fluid dynamics technologies undertaking relevant studies is relatively straightforward although significant practical and theoretical issues remain (Habib and Krajewski 2001). Once the instrumental effects are well understood, we should undertake efforts to improve our knowledge of the spatial variability of variables relevant to remote-sensing rainfall. The most prominent variable is radar reflectivity. Its variability at the scale of radar-rainfall pixel directly affects quantitative interpretation of the estimated rainfall maps and products. The issues of observational network design, i.e., the number and distribution in space of the disdrometers are similar to those we discussed in the section above. As radar reflectivity is a higher-order moment of the DSD than rainfall rate, and there is no unique relationship between the two, it is likely that its characteristic scale (e.g., correlation distance) is significantly different from that of rainfall. Thus, we may need to organize experiments that will be able to capture a wide range of distances so that we can model the shape of the covariance function and other measures of association adequately. Only when we understand the error characteristics of the instruments we use will our interpretation of the results be meaningful. Thus, we need to continue supporting development of new, less-expensive disdrometers, conducting
Ground Networks: Are We Doing the Right Thing?
411
their intercomparisons, and designing and carrying out small-scale studies of spatial and temporal variability of various raindrop characteristics. These studies should be conducted in a continuous deployment mode as opposed to the more traditional campaign style typical for atmospheric experiments. This is necessary to enable conditional analysis of the collected data. Regarding the conditioning, it seems that the best strategy is to use the variables that can be observed by larger-scale remote sensors. For example, based on a vertically pointing radar we may be able to classify rainfall type into, say, convective and stratiform on a minute by minute basis, but if a nearby scanning radar is not able to resolve such scale conditional (on rain type) analysis is not very helpful in studies of radar performance. Another very appealing argument is that such conditioning would allow bringing data from many sites into the analysis provided they are conducted using the same or similar instruments. This is clearly the case using satellite-based observations. However, selection of appropriate variables is not an obvious task. It requires comprehensive analyses of the existing data via physically based modeling and/or data mining approaches.
5 OTHER GROUND-BASED OBSERVING SYSTEMS 5.1 Vertically pointing radars Vertically pointing radars provide crucial information for space-borne remote sensing. They are capable of observing vertical profiles of precipitating clouds and identify features affecting remote-sensing-based estimates. These features include thickness and height of the melting ice at cloud-based (i.e., bright band problem), precipitation phase, convective cores, updrafts and downdrafts, etc. They are also capable of providing estimates of the vertical profile of DSD. These estimates are more reliable if the profiler operates at multiple frequencies so that air and raindrop motion can be distinguished. However, for such radars the sampling volumes are not well matched. For high variability conditions such as convective cores this is a cause of concern as volume mismatch results in increased uncertainty of the measurements. Profiler-based studies of precipitation systems and the related instrumental and estimation issues have been well documented in a number of publications, for example, Wakasugi et al. (1986), Gage et al. (1999, 2000, 2002), Williams (2002), Williams et al. (2000), and Kollias et al. (2002). We propose to use these proven technologies to explore the spatial variability of the vertical profile of precipitating clouds. The scale of interest is the subscale of the resolution of space-based microwave and infrared sensors, i.e., about 15–25 km2. Although such scale domains can be explored by scanning radars, the main advantage of the profilers is their ability to
412
W. F. Krajewski
measure the Doppler velocity of the raindrops and thus provide estimates of the DSDs. They also identify the location and the thickness of the bright band much more accurately. Although the time resolutions of the VPRs is also very high, essentially providing continuous observations, our understanding of the space-time relation for rainfall processes is still limited (e.g., Fabry 1996.) Thus, having several, say five, such instruments evenly distributed over a single pixel, would provide additional insight into this relationship and the special variability of the processes affecting space-based sensors. This is particularly important for rainfall estimation over land where varied emissivity of the land surface introduces difficulties into interpretation of the satellite data.
5.2 Networks of small polarimetric radars Recently, another attractive technology is emerging that may offer many advantages to address the problem of validation of space-based remote sensing of precipitation: special purpose networks of inexpensive radars. Several groups have demonstrated advantages of using X-band polarimetric radars for rainfall estimation (Matrosov et al. 2002; Anagnostou et al. 2004). Because X-band radars are widely used for navigation, many manufacturers compete in this market. This drives down the cost of waveguides, microwave sources, and test equipment. Relatively small antennas can give high azimuthal resolution at X-band, compared to C-band and S-band radars. For example, to achieve 1.5° resolution requires an antenna with a diameter of 2 m, which translates to a tremendous cost advantage compared to C- and S-band. It is easy to install this size antenna on a small building or to mount it on a trailer. The polarimetric measurements at X-band also offer certain advantages, such as increased sensitivity to rainfall, as compared to longer wavelengths (e.g., Matrosov et al. 1999, 2002; Zrnic and Ryzhkov 1999). X-band (3 cm) waves are subject to more attenuation than the longer Cband or S-band waves in heavy rainfall. This is an issue if one needs a long operating range, but our focus is on short range and high resolution. If the network radars’ use is limited to 20 km, and there are multiple radars looking at the same area from different directions, the resultant multiradar estimates of rainfall will not suffer much from attenuation. The radars are polarimetric, and some polarimetric observations, such as KDP, are insensitive to partial attenuation. The physical concept behind polarization diversity is that, under aerodynamic forces, falling hydrometeors take oblate shapes, which depend on their size, and as a result, impact differently the propagation and backscattering of incoming horizontal (H) and vertical (V) electromagnetic waves. The most common polarimetric radar measurements are: (1) the reflectivity factors at H and V polarization (ZH, ZV); (2) the differential
Ground Networks: Are We Doing the Right Thing?
413
reflectivity factor (ZDR); and (3) the propagation differential phase (ΦDP). Over a certain radial distance ∆r, one can calculate the specific differential phase shift (KDP). All of these parameters, in various combinations, have been shown to contribute to improve rainfall estimation and enable retrieval of DSD (Bringi and Chandrasekar 2001; Anagnostou et al. 2004). Development and operation of a network of radars offers numerous advantages. For example, consider a network of four radars overlooking a regular dense network of rain gauges. Its operation leads to: Improved accuracy of rainfall algorithms. Sound algorithms can only be developed from large samples of data, since rainfall is highly variable in both space and time, and is highly intermittent (long periods of no rain between short duration events). This is contrary to past practice where a great deal of research and conclusions were based on case studies. As we emphasized in other sections of this paper, evidence is mounting that proper evaluation of remote sensing of rainfall can be established only from large sets of data. This need for large samples is increasing, as there is a trend to develop different algorithms for different atmospheric situations. To perform such conditioning for a sufficiently large sample, the data set from which the sample is drawn should be as large as possible. Increased reliability. It rains only about 5% of the time, so if the radar happens to be down we lose data. With four radars, it is unlikely that we will miss any rainfall events. We will be able to reduce the measurement error variance, and thus the uncertainty of the estimates of rainfall. Reduced development and operating costs. As network radars share spare parts and technical support, a network of four X-band radars may cost as little as $1 million. Repeatability. Credibility of the system, and therefore the confidence of users of the data will be greatly increased if the individual radars in the system demonstrate consistent performance when considered individually. On the other hand, viewing the same storm from different aspect angles will mitigate the adverse effects of signal attenuation and data noise. Still, much research remains to be done to fully realize the above benefits. These include technological advancements of radar hardware, software to operate the radar as a true network and not simply a collection of four individual radars, and, of course, rainfall estimation algorithms.
6 CONCLUSIONS AND RECOMMENDATIONS In this article we advocate an experimental framework for use of groundbased observation of rainfall for the quantitative evaluation of space-based methods of precipitation estimation and monitoring. We favor this approach over an alternative, i.e., the error propagation modeling on the basis of several arguments. The most important ones are: (1) the experimental approach
414
W. F. Krajewski
leads to total error estimates of the variable that is of ultimate interest, i.e., rainfall on the ground; (2) many of the parameters (characteristics) of the atmospheric and instrumental processes needed for the error propagation approach are not easy to observe or are not observable at all; (3) statistical distributions of the input variables are not known. We also caution against using approximate (engineering) methods of uncertainty propagation computations, i.e., the methods based on Taylor series expansion of the nonlinear functions governing the rainfall system and its observations. These methods give good results only when input variable errors are small (small coefficient of variation) and have symmetric distributions. We recommend a hierarchical approach to the validation problem. The hierarchy of instruments should cover the space-time scale gap that exists between the point observations by direct sensors such as rain gauges and observations from satellite platforms. The connecting device is weather radar. Used at different frequencies and configurations it brings down the resolution to the order of hundreds of meters in space and minutes in time. Such scales make the link with the point scale feasible via simple experimental setups. This in turn, permits error characterization of the radarbased estimates. With such knowledge we will be in a good position to attack the space-based technologies. Our hierarchical approach to the problem of validation of spacebased remote sensing of rainfall allows, of course, multisensor approaches (Krajewski 1987; Tustison et al. 2003). However, to optimally combine data from different sensors – motivated by the fact that they often have complementary characteristics in terms of sensing and relevance to rainfall – requires knowing error characteristics of the individual sensors. The combination could be done in a multisensor, multiscaling fashion or via a data assimilation schemes using physically based models. In either case error characterization is crucial. We hope that our discussion clearly illustrates the need for investing in basic research on rainfall via ground-based observational systems. The space-borne remote-sensing validation problem and the basic understanding of rainfall processes are strongly coupled. Without comprehensive understanding of the process, we will not be able to take advantage of the possibilities offered by remote sensing and without remote sensing we will not fully understand rainfall and its impact at global, regional, and local scale water cycle and other processes it affects. Therefore, what is the answer to the question we posed in the title? We do not know about many networks that were designed using the principles we discussed above. If we are serious about quantifying the uncertainty of space-based rainfall products we have to follow a systematic approach to designing, deploying, and using information from ground-based observational networks.
Ground Networks: Are We Doing the Right Thing?
415
Acknowledgments: Many of the ideas expressed in this article are result of collaborations and stimulating discussions over the years with Grzegorz Ciach, Anton Kruger, Mark Morrissey, Emad Habib, Paul Kucera, Brian Nelson, Mekonnen Gebremichael, Chris Kummerow, George Huffman, Bob Adler, Christopher Williams, Ken Gage, and James A. Smith, as well as with the participants of a recent workshop on small-scale variability of rainfall hosted by Jeff Austin in Auckland, New Zealand. These included Isztar Zawadzki, Frederic Fabry, Gyu Won Lee, Ian Cluckie, Chris Collier, Charles Lin, and Alan Seed, among others.
7 REFERENCES Anagnostou, E. N., M. N. Anagnostou, A. Kruger, W. F. Krajewski, and B. Miriovsky, 2004: High-resolution rainfall estimation from X-band polarimetric radar measurements. J. Hydrometeor., 43(1), 106–118. Beard, K. V., 1976: Terminal velocity and shape of cloud and precipitation drops aloft. J. Atmos. Sci., 33, 851–864. Bell, T. L., A. Abdullah, R. L. Martin, and G. R. North, 1990: Sampling errors for satellitederived tropical rainfall: Monte Carlo study using a space-time stochastic model. J. Geophys. Res., 95(D3), 2195–2205. Bell, T. L. and P. K. Kundu, 1996: A study of the sampling error in satellite rainfall estimates using optimal averaging of data and a stochastic model. J. Climate, 9, 1251–1268. Bell, T. L. and P. K. Kundu, 2000: Dependence of satellite sampling error on monthly averaged rain rates: Comparison of simple models and recent studies. J. Climate, 13, 449–462. Bras, R. L. and I. Rodriguez-Iturbe, 1993: Random functions and hydrology. Dover, 559 pp. Bringi, V. N. and V. Chandrasekar, 2001: Polarimetric Doppler weather radar: Principles and applications, Cambridge University Press, Cambridge, 636 pp. Campos, E. and I. Zawadzki, 2000: Instrumental uncertainties in Z–R relations. J. Appl. Meteor., 39, 1088–1102. Ciach, G. J., 2003: Local random errors in tipping-bucket rain gauge measurements. J. Atmos. Oceanic Technol., 20(5), 752–759. Ciach, J. G. and W. F. Krajewski, 1999: On the estimation of radar rainfall error variance. Adv. Water Resour., 22(6), 585–595. Crane, R. K., 1990: Space-time structure of rain rate fields, J. Geophys. Res., 95, 2001–2020. Fabry, F., 1996: On the determination of scale ranges for precipitation fields. J. Geophys. Res. 101, 12819–12826. Gage, K. S., C. R. Williams, W. L. Ecklund, and P. E. Johnston, 1999: Use of two profilers during MCTEX for unambiguous identification of Bragg scattering and Rayleigh scattering. J. Atmos. Sci., 56, 3679–3691. Gage, K. S., C. R. Williams, P. E. Johnston, W. L. Ecklund, R. Cifelli, A. Tokay, and D. A. Carter, 2000: Doppler radar profilers as a calibration tools for scanning radars. J. Appl. Meteor., 39, 2209–2222. Gage, K. S., C. R. Williams, W. L. Clark, P. E. Johnston, and D. A. Carter, 2002: Profiler contributions to Tropical Rainfall Measuring Mission (TRMM) ground validation field campaigns. J. Atmos. Oceanic Technol., 19, 843–863.
416
W. F. Krajewski
Gebremichael, M., W. F. Krajewski, M. Morrissey, D. Langerud, G. Huffman, and R. Adler, 2003: Error uncertainty analysis of GPCP monthly rainfall products: A data based simulation study. J. Appl. Meteor., 42(12), 1837–1848. Gunn, R. and G. Kinzer, 1949: The terminal velocity of fall for water droplets in stagnant air. J. Meteor., 6, 243–248. Habib, E., W. F. Krajewski, and A. Kruger, 2001: Sampling errors of fine resolution tippingbucket rain gauge measurements. J. Hydrol. Eng. 6(2), 159–166. Habib, E. and W. F. Krajewski, 2001: An example of computational approach used for aerodynamic design of a rain disdrometer. J. Hydraul. Res., 39(4), 425–428. Habib, E. and W. F. Krajewski, 2002: Uncertainty analysis of the TRMM Ground Validation radar-rainfall products: Application to the TEFLUN-B field campaign. J. Appl. Meteor., 41(5), 558–572. Jameson, A. R. and A. B. Kostinski, 2002: Spurious power-law relations among rainfall and radar parameters. Quart. J. Roy. Met. Soc., 128(7), Part B No. 584, 2045–2058. Kollias, P., B. A. Albrecht, and F. Marks Jr., 2002: Why Mie? Accurate observations of vertical air velocities and raindrops using a cloud radar. Bull. Amer. Meteor. Soc., 83, 1471–1483. Krajewski, W. F., 1987: Radar-rainfall data quality control by the influence function method. Water Resour. Res., 23(5), 837–844. Krajewski, W. F. and G. J. Ciach, 2003: Towards probabilistic quantitative WSR-88D algorithms: Preliminary studies and problem formulation. NOAA / N WS Report for Contract DG133W-02-CN-0089. Krajewski, W. F. and R. Goska, 2004: A GIS utility for observational network design for area-average estimation, to be submitted to Computers and Geosciences. Krajewski, W. F. and J. A. Smith, 2002: Radar hydrology: Rainfall estimation. Adv. Water Resour., 25, 1387–1394. Krajewski, W. F. and B. Vignal, 2004: Parameterization of the range dependent error in radar rainfall estimates based on a vertical profile of reflectivity analysis. to be submitted to J. Hydrometeorology. Krajewski, W. F., G. J. Ciach, and E. Habib, 2003: An analysis of small-scale rainfall variability in different climatological regimes. Hydrol. Sci. J., 48(2), 151–162. Krajewski, W. F., G. J. Ciach, J. R. McCollum, and C. Bacotiu, 2000: Initial validation of the Global Precipitation Climatology Project over the United States. J. Appl. Meteor., 39(7), 1071–1086. Kruger, A. and W. F. Krajewski, 2002: Two-dimensional video disdrometer: A description. J. Atmos. Oceanic Technol., 19, 602–617. Kruger, A. and K. Kanukurthy, 2004: A cell phone-based data logger and network for monitoring environmental variables. IEEE Trans. Instr. Meas. (near submission). Matrosov, S. Y., K. A. Clark, B. E. Martner, and A. Tokay, 2002: X-band polarimetric radar measurements of rainfall. J. Appl. Meteor., 41, 941–952. Matrosov, S. Y., R. A. Kropfli, R. F. Reinking, and B. E. Martner, 1999: Prospects for measuring rainfall using propagation differential phase in X- and Ka-radar bands. J. Appl. Meteor., 38, 766–776. McCollum, J. R. and W. F. Krajewski, 1998: Investigations of error sources of the Global Precipitation Climatology Project emission algorithm. J. Geophys. Res., 103(D22), 28711–28719. Miriovsky, B. J., A. A. Bradley, W. N. Eichinger, W. F. Krajewski, A. Kruger, B. R. Nelson, J.-D. Creutin, J.-M. Lapettite, G. W. Lee, I. Zawadzki, and F. L. Ogden, 2004: An experimental study of small-scale variability of reflectivity. J. Appl. Meteor. 5(1), 110–128. Moore, R. J., D. A. Jones, D. R. Cox, and V. S. Isham, 2000: design of the HYREX raingauge network. Hydrol. Earth System Sci., 4, 523–530.
Ground Networks: Are We Doing the Right Thing?
417
Morrissey, M. L., J. A. Maliekal, J. S. Greene, and J. Wang, 1995: The uncertainty in simple spatial averages using rain-gauge networks. Water Resour. Res., 31(8), 2011–2017. Nešpor, V., W. F. Krajewski, and A. Kruger, 2000: Wind-induced error of rain drop size distribution measurement using a two-dimensional video disdrometer. J. Atmos. Oceanic Technol., 17, 1483–1492. Nijssen, B. and D. P. Lettenmaier, 2004: Effect of precipitation sampling error on simulated hydrological fluxes and states: Anticipating the global precipitation measurement satellites, J. Geophys. Res., 109(D2), D02103 10.1029/2003JD003497. North, G. R., S. S. P. Shen, and R. Upson, 1993: Sampling errors in rainfall estimates by multiple satellites. J. Appl. Meteor., 32(2), 399–410. North, G. R. and S. Nakamoto, 1989: Formalism for comparing rain estimation designs. J. Atmos. Oceanic Technol., 6(6), 985–992. North, G. R. and I. Polyak, 1996: Spatial correlation of beam-filling error in microwave rainrate retrievals. J. Atmos. Oceanic Technol., 13(5), 1101–1106. Oki, R. and A. Sumi, 1994: Sampling simulation of TRMM rainfall estimation using radar AMeDAS composites. J. Appl. Meteor. 33(12), 1597–1608. Sheppard, B. E., 1990: Measurement of raindrop size distributions using a small Doppler radar. J. Atmos. Oceanic Technol., 7, 225–268. Sheppard, B. E. and P. I. Joe, 1994: Comparison of raindrop size distribution measurements by a Joss-Waldvogel disdrometer, a PMS 2DG spectrometer, and a POSS Doppler radar. J. Atmos. Oceanic Technol., 11, 874–887. Short, D. A. and G. R. North, 1990: The beam filling error in Nimbus-5 ESMR observations of GATE rainfall. J. Geophys. Res. 95, 2187–2193. Smith, J. A., D. J. Seo, M. L. Baeck, and M. D. Hudlow, 1996: An intercomparison study of NEXRAD precipitation estimates. Water Resour. Res., 32, 2035–2045. Smith, P., 1993: A Study of sampling-variability effects in raindrop size observations. J. Appl. Meteor., 32, 1259–1269. Steiner, M., J. A. Smith, S. J. Burges, C. V. Alonso, and R. W. Darden, 1999: Effect of bias adjustment and rain gage data quality control on radar rainfall estimates. Water Resour. Res., 35, 2487–2503. Tokay, A., A. Kruger, and W. F. Krajewski, 2001: Comparison of drop size distribution measurements by impact and optical disdrometers. J. Appl. Meteor., 40(11), 2083–2097. Tustison, B., D. Harris, and E. Foufoula-Georgiou, 2003: Scale-recursive estimation for multi-sensor QPF verification: A preliminary assessment. J. Geophys. Res., 108(D8), 8377–8390. Wakasugi, K., A. Mizutani, M. Matsuo S. Fukao, and S. Kato, 1986: A direct method for deriving drop-size distributions and vertical air velocities from VHF Doppler radar spectra. J. Atmos. Oceanic. Technol., 3, 623–629. Williams, C. R., 2002: Simultaneous ambient air motion and raindrop size distributions retrieved from UHF vertical incident profiler observations. Radio Sci., 37, 101029/ 2000RS002603. Williams, C. R., A. Kruger, K. S. Gage, A. Tokay, R. Cifelli, W. F. Krajewski, and C. Kummerow, 2000: Comparison of simultaneous rain drop size distributions estimated from two surface disdrometers and a UHF profiler, Geophys. Res. Lett., 27(12), 1763–1766. Zrnic, D. S. and A. V. Ryzhkov, 1999: Polarimetry for weather surveillance radars. Bull Amer. Meteor. Soc., 80(3), 389–406.
Section 6 Modeling Precipitation Processes and Data Assimilation for NWP
33 AEROSOL IMPACT ON PRECIPITATION FROM CONVECTIVE CLOUDS Alexander Khain, Daniel Rosenfeld, and Alexander Pokrovsky The Hebrew University of Jerusalem, Jerusalem, Israel
Abstract
Mechanisms through which atmospheric aerosols affect cloud microphysics, dynamics, and precipitation are investigated using a spectral microphysics cloud model. Significant effect of aerosols on cloud updrafts and cloud top height is found. Maritime aerosol leads to earlier formation of raindrops that fall down through cloud updrafts. This is one of the reasons of comparatively low vertical velocity in maritime convective clouds. An increase in the small cloud condensational nuclei (CCN) concentration leads to formation of a great number of small droplets with low collision rate. The direct consequence of this is a time delay in raindrop formation. This delay prevents the decrease in vertical velocity and increases the duration of the diffusion droplet growth stage, increasing latent heat release by condensation and freezing. As a result, vertical velocities in clouds developing in smoky (continental-type aerosol) air turn out to be larger and clouds attain higher levels. The decrease in precipitation efficiency of clouds arising in smoky air can be attributed to the higher loss of precipitating mass due to higher sublimation and evaporation. In case of very strong atmosphere instability and low air humidity (very continental thermodynamic conditions, forest fires, etc.), a great number of small droplets reach the upper troposphere and freeze with the formation of ice crystals that do not contribute to precipitation. Under more stable conditions, the delay in raindrop formation leads to the fact that the raindrops fall down from higher levels, as compared to those in case of clouds developing in clean air. In case of comparatively low air humidity and a certain wind shear, these raindrops fall trough a dry deep layer. As a result, precipitation from single cumulus clouds decreases significantly. Under certain conditions wide deep clouds developing in continental aerosol conditions produce stronger downdrafts and stronger convergence in the boundary layer. As a result, secondary clouds arise that can form a squall line. Under similar thermodynamic conditions, clouds developing under maritime aerosols do not produce strong downdrafts, and do not lead to the squall line formation. Formation of secondary clouds and squall lines increases precipitation over the area considered.Thus, the “aerosol effect” on precipitation can be understood only in combination with the “dynamical effect” of aerosols. This fact should be taken into account in schemes of parameterization of aerosol effects on precipitation.
421 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 421–434. © 2007 Springer.
422
A. Khain et al.
1 INTRODUCTION Analysis of the Tropical Rainfall Measuring Mission (TRMM) satellite data demonstrated that smoke from burning vegetation can practically shut off warm rain formation in tropical clouds (Rosenfeld 1999). Rosenfeld and Woodley (1999) observed that in polluted areas over Thailand and Indonesia smoky clouds do not precipitate at all because of the narrow spectra of small droplets. At the same time, similar clouds begin precipitating in clear air in 20–25 min after their formation. A decrease in precipitation in urban areas was reported (Rosenfeld 2000). Observations in the Amazon region (Kaufman and Nakajima 1993; Andreae et al. 2004) show significant effects of biomass burning on droplet spectra and precipitation formation. Some of these effects were numerically simulated by Khain et al. (1999, 2001) using the two-dimensional (2D) spectral microphysics Hebrew University Cloud Model (HUCM). Detailed investigation of raindrop formation in ascending cloud parcels (Segal et al. 2004) showed that the spectrum of cloud condensational nuclei (CCN) can be divided into three main ranges: CCN with radii rCCN < 0.01 µm are not usually activated and do not influence the cloud microphysical structure; CCN of the intermediate size with 0.01 µm < rCCN < ~1 µm are, as a rule, activated and give rise to droplet formation. An increase in the concentration of CCN of this size leads to an increase in the droplet concentration and slows the diffusional growth of droplets. Droplet spectra become narrower and the height of the collision triggering level increases. This leads to a delay in raindrop formation (Andreae et al. 2004; Khain et al. 1999). The last, CCNs with rCCN > ~1 µm give rise to formation of largest droplets, which foster raindrop formation at lower levels (e.g., Yin et al. 2000; Rosenfeld et al. 2002; Segal et al. 2004). A delay or acceleration in raindrop formation does not automatically lead to a decrease or increase in the accumulated rain. In this study we investigate physical mechanisms, through which aerosols influence cloud microphysics, dynamics, and accumulated rain.
2 NUMERICAL MODEL The model microphysics (see in more detail, Khain and Sednev 1996, Khain et al. 2000) is based on solving an equation system for eight size (number) distributions for water drops, ice crystals (columnar, plate-like, and dendrites), snowflakes (aggregates), graupel, and hail/frozen drops, and CCN. Each size distribution is represented by 33 mass doubling categories (bins), so mass mk in the category k is determined as mk = 2 mk–1, where k = 2,…,33. The minimum mass in the hydrometeor mass grids (except aerosols) corresponds to that of a 2 µm-radius droplet. The mass grids used for hydrometeors of all types are similar. This simplifies the calculation of interaction between
Aerosol Impact on Precipitation from Convective Clouds
423
hydrometeors of different bulk densities. The model microphysics is specifically designed to take into account the effect of atmospheric aerosols on cloud development and precipitation formation and effects of clouds on CCN concentration in the atmosphere. Nucleation (CCN activation) of droplets is based on the utilization of a separate size distribution function for CCN. In the current model version, the initial size distribution of CCN is calculated using a dependence of concentration N of activated CCN on supersaturation with respect to water Sw as described by Khain et al. (2000). In particular, the empirical dependence (Pruppacher and Klett 1997) can be written in the form of N = N0Skw, where Sw (in %), N0, and k are measured constants. Using supersaturation Sw calculated in the course of model integration, the value of critical size of dry CCN rcrit is determined at each model time step. Aerosol particles with radii rN > rN,crit are activated and transformed into droplets. Corresponding bins of the CCN size distributions become empty. In case there are no aerosol particles with rN > rN,crit in the CCN spectra in a particular grid point, no new droplet nucleation takes place at this point. The size of fresh nucleated droplets is calculated as follows. In case the radii of CCN rN < 0.03 µm, the equilibrium assumption (according to the Köhler equation), is used to calculate the radius of a nucleated droplet corresponding to rN (see Khain et al. 2000 for more detail). In case rN > 0.03 µm, the radius of water droplet formed on these CCN is simply equal to five times the radius of the dry aerosol particle (Kogan 1991; Khain et al. 1999; Yin et al. 2000). Since large CCN do not reach their equilibrium size at cloud base, this approach prevents nucleation of unrealistically large droplets and inhibit too fast raindrop formation. Nucleation of ice crystals is described proceeding from the formula presented by Meyers et al. (1992) relating the number concentration of deposition and condensation-freezing ice nuclei (IN), Nd, to supersaturation with respect to ice, Sice = Nd exp (ad + bdSice), where Nd = 10–3 m–3, ad = –0.639, bd = 12.96. Nucleation is prevented for temperatures warmer than –5°C. The number of newly activated ice crystals at each time step in a certain grid point, dNd, is calculated as follows:
⎧b N dS if dSice > 0 dN d = ⎨ d d ice 0 if dSice ≤ 0 ⎩
(1)
where dSice is calculated using a semi-Lagrangian approach (more details in Khain et al. 2000). The type of ice crystals nucleated depends on temperature. According to Takahashi et al. (1991) temperature-dependent nucleation proceeds as follows: plate-like crystals form at –8°C > Tc ≥ –14°C
424
A. Khain et al.
and –18°C > Tc ≥ –22.4°C, columnar crystals arise at –4°C > Tc ≥ –8°C, Tc < –22.4°C, and dendrites (branch-type crystals) form at –14°C > Tc ≥ –18°C. Secondary ice generation is described by a Hallett and Mossop (1974) mechanism, according to which at T = –5°C each 250 collisions of droplets having radii exceeding 24 µm with graupel particles leads to the formation of one ice splinter. According to measurements, this process is assumed to occur within –3 to –8°C temperature range. We suppose that the density of splinters is the same as that of pure ice (0.9 g cm–3) and, hence, the splinters are assigned to plate-type ice crystals. The rate of drop freezing is described following the observations of immersion nuclei by Vali (1975, 1994) and homogeneous freezing by Pruppacher (1995). The rate of freezing is calculated using a semi-Lagrangian approach allowing one to calculate changes in supersaturation and temperature in moving cloud parcels reaching model grid points (Khain et al. 2000). At each time step supersaturations with respect to water and ice were calculated by solving an equation system of corresponding differential equations (Khain and Sednev 1996). Besides droplet and ice nucleation, these values of supersaturation are used for calculation of diffusion growth/evaporation of water droplets and deposition/sublimation of ice particles. We take into account the shape of ice crystals to calculate diffusion growth of different ice crystals. An efficient and precise method of solving the stochastic kinetic equation for droplet collisions (Bott 1998) was extended to a system of stochastic kinetic equations that are used to calculate water–water, water–ice, and ice– ice collisions. The model uses height-dependent drop–drop and drop–graupel collision kernels calculated using a hydrodynamic method valid within a wide range of drop and graupel sizes (Khain et al. 2001; Pinsky et al. 2001). Ice–ice collision rates are assumed to be temperature dependent. An increase in the water–water and water–ice collision kernels by turbulent/inertia mechanism was taken into account following Pinsky et al. (2000). As a result of riming, ice crystals, and snowflakes can convert to graupel or to hail depending on temperature. Collisions between ice crystals lead to snow (aggregates) formation. Khain et al (2000) describe in detail the procedure for the conversion of hydrometeor types as a result of different kind of collisions. Recently, a description of collisional breakup has been implemented in the HUCM microphysics (Seifert et al. 2005). The changes of the drops size distribution due to breakup are represented by the well-known stochastic breakup equation (Pruppacher and Klett 1997). The coalescence efficiency and the fragment size distributions are parameterized following Low and List (1982) with some corrections for small raindrops using parameterizations given by Beard and Ochs (1995). The breakup is conducted for drops exceeding 100 µm in diameter.
Aerosol Impact on Precipitation from Convective Clouds
425
3 RESULTS: AEROSOL EFFECTS ON PRECIPITATION 3.1 Unstable continental conditions Simulations of cloud development under unstable conditions in Texas during summertime (Rosenfeld and Woodley 2000) were performed for a continental-type CCN (C-case), as well as for microphysically maritime CCN (M-case). In the simulations conducted, the initial size distribution of CCN was approximated as NCCN = ASwk. In C-case, A was set 1,260 cm–3 and k = 0.308. In addition, in C-case the maximum size of dry CCN was assumed equal to 0.6 µm. In M-case, coefficient A was set equal to 100 cm–3, while coefficient k = 0.462. It was also assumed, that there were no small CCN in the CCN spectrum that could be activated at supersaturation values exceeding 1.1% in the “maritime” case. This assumption is based on measurements by Hudson (1984, 1993) and Hudson and Frisbie (1991) indicating no increase in the CCN concentration in extreme maritime cases for S > 0.6%, which suggests a lack of small CCN. Thus, we assumed that under not very extreme conditions, maritime air does not contain small aerosols to be activated at S > 1.1%. This limitation allows us to keep droplet concentration to be typical of maritime clouds rain rates as the functions of time and x-coordinate are shown in Fig. 1. One can see a significant decrease in precipitation amount in clouds that developed in smoky (continental CCN) air. The difference in precipitation can be attributed to the following. When CCN concentration is low (M-case), raindrops and large graupel form at comparatively low levels, fall down and reach the surface without a significant evaporation. In C-case, high concentration of small droplets (up to 1,000 cm–3) arises by nucleation (Khain et al. 2001). These droplets have low collision efficiency, as well as low freezing rate. As a result, they reach the level of homogeneous freezing (~9.5 km) and give rise to formation of ice crystals with concentration of several hundred per cm–3. These crystals as well as small graupel and snowflakes formed at higher levels spread over a large area and sublimate, and do not contribute to the precipitation (they sublimate in a cloud anvil). Droplets falling from higher levels also experience significant evaporation. As a result, the increase in CCN concentration decreases the precipitation efficiency of clouds that develop in strong unstable low humidity air. Note that mass contents of water drops (with maximum of about 3 cm–3), as well as ice particles (total ice content maximum was also about 3 cm–3) were quite significant. Thus, precipitation efficiency (defined as the ratio of precipitation amount to the amount of hydrometeors formed in clouds by condensation and deposition) is quite low in these continental clouds (in C-case).
A. Khain et al.
426
3.2 Single maritime clouds Aerosol effects on clouds and precipitation under maritime (as well as under comparatively wet and not extremely unstable conditions) are more complicated and are highly affected by vertical wind shear and air humidity. The investigation of aerosol effects is performed using the same initial CCN distributions as were used under the continental thermodynamic conditions. However, in these simulations the GATE (day 261) temperature profile (Ferrier and Houze 1989) was used. Since the maximum CCN radius in the model is 2 µm, the maximum radius of nucleated droplets in M-case did not exceed ~10 µm. The role of giant CCN was not investigated in the study.
Maritime CCN
Continental CCN
Precipitation inhibition
Figure 1. Rain rates as the functions of time and x-coordinate under unstable continental conditions.
The development of single clouds was triggered by 5 min of heating within the zone of the 0.5 km width in the surface layer. By varying the intensity of the heating, single clouds of different top height were simulated. The cloud evolution was simulated up to the cloud dissipation. Figure 2 shows the dependence of accumulated rain on the cloud top height (determined by the 10 dBZ level) of convective clouds in three sets of simulations with the GATE (day 261) temperature profile. In the first set no wind shear (NO WS) was assumed. In the second set (with a weak wind shear, WS) the wind speed increases from 0.4 m s–1 at z = 0 km to 5 m s–1 at z = 9 km. Above z = 9 km the wind speed was assumed equal to 5 m s–1. In both cases air humidity was quite high, with ~90% in the lower 2 km layer and about 50– 60% in the middle troposphere. The third set of simulations was similar to the second one except the humidity was decreased by 10% in the lower 2 km layer (RH = 80%) and by about 20–25% in the middle troposphere. Each set
Aerosol Impact on Precipitation from Convective Clouds
427
of the experiments consisted of experiments with maritime CCN (M-case) and continental CCNs (C-case).
Figure 2. Dependences of accumulated rain on cloud depth under different conditions in case of maritime (GATE, 261 day) temperature profile.
Figure 3. Vertical profiles of convective heating, cooling, and net heating calculated under continental and maritime CCN. The clouds chosen produced nearly similar precipitation amounts.
One can see that (a) single clouds that developed in M-cases produce larger amounts of accumulated rain. To produce the same rain amount, clouds that developed in C-case, should be higher; (b) wind shear, and especially, the air humidity significantly change the accumulated rain amount, as well as the
428
A. Khain et al.
relationship between the rain amounts in C- and M-cases. At the same time, sensitivity of the accumulated rain to the CCN concentration under comparatively stable maritime temperature conditions is smaller than that under unstable continental conditions; (c) while in all sets of simulations the accumulated rain monotonically increases with the cloud top height in Mcases, in C-cases with a nonzero wind shear, precipitation stops growing, beginning with a certain cloud top height. The physical explanation of the lower precipitation efficiency in single clouds developed in microphysically continental CCN conditions can be derived from the analysis of vertical profiles of convective heating/cooling. Figure 3 shows such profiles plotted for two clouds that developed in simulations with no wind shear. Firstly, the cloud developing in the C-case reaches the top height of 9 km. Secondly, the cloud that develops in M-case, reaches the maximum top height of 6.5 km. According to Fig. 2, these clouds produce comparable accumulated rain amounts. These profiles were obtained by averaging of heating/cooling in the horizontal direction over the computational area (64 km) and over time of simulation (4 h). Dashed and dasheddotted lines corresponding to positive values of latent heat release reflect contribution of condensation, freezing and deposition. Along similar lines, but corresponding to negative values of latent heat release, reflect cooling caused by the droplet evaporation, ice melting and ice sublimation. Thick and thin solid lines show profiles of net convective heating for continental and maritime CCN, respectively. The squares (integrals) formed by these solid lines reflect the total atmospheric heating due to phase transitions and are a measure of the precipitation amount. The increase in the CCN concentration leads to a strong increase in heating both due to the increase in diffusion growth and due to more intense droplet freezing. At the same time, the cooling in the C-case is larger as compared to the M-case because of the larger droplet evaporation and ice sublimation. The higher droplets evaporation and ice sublimation in the C-case can be attributed to the following. In the C-case raindrops and ice particles are smaller than in the M-case and ascend to higher levels. Besides, they have smaller sedimentation velocity. As a result, the time duration of their sedimentation is longer. The detrainment of liquid and ice particles at upper levels leads to the fact that ice particles and drops falling from higher levels tend to sediment through a comparatively dry air. At the same time, in the M-case raindrops form at lower levels, and fall down through the vertically narrow layer of cloudy wet air (or in the close vicinity of the cloud). As a result, the evaporation (and cooling) in the maritime aerosol case is lower as compared to the continental CCN case. The net heating in the C-case is extended to higher levels. The minimum in the net heating at ~4 km in the C-case is related to the cooling caused by the melting of ice (mainly graupel). No such minimum is seen in the profile of net convective heating in the M-case simulation indicating a smaller contribution of melted rain in the last case.
Aerosol Impact on Precipitation from Convective Clouds
429
Figure 4. Vertical profiles of convective heating for clouds developed in continental and maritime aerosols in case of a comparatively weak wind shear.
The higher loss in the precipitating mass by the drop evaporation and ice sublimation is the main cause of the lower precipitation efficiency of clouds developed in the continental CCN case. Another example of latent heat release profiles for clouds developed in case of upper level wind shear in the C- and M-case is shown in Fig. 4. These clouds reached the same top heights (about 10 km), but have quite different accumulated rain (see Fig. 2). Stronger wind at upper levels moves water droplets and ice downwind fostering their fall through dry air and their evaporation and sublimation. Since water droplets and ice particles ascend to higher levels in C-cases, wind shear leads to higher loss in precipitation for C-cases. We believe that this is the reason, why in the presence of wind shear the precipitation amount in C- cases does not increase with cloud top height (beginning with a certain cloud depths): a significant fraction of hydrometeors ascending above certain level evaporates and does not contribute to precipitation. This can explain the results of some simulations with high wind shears at upper levels (not shown), when the growth of cloud top height in C-cases was accompanied by a decrease in precipitation amount. The decrease in precipitation with a decrease in humidity (Fig. 2) is explained by the increase in the loss in precipitating mass by evaporation and sublimation. This loss is higher in clouds that develop under continental aerosol conditions. A comparatively low sensitivity of precipitation in M-cases to both wind shear and humidity is related to the
430
A. Khain et al.
formation of raindrops at low levels and to their fall within wet air in, or in the close surrounding of the cloud.
3.3 Dynamical aerosol effects It is well known that vertical velocities in maritime convective clouds are significantly smaller than in continental clouds. While in continental clouds vertical updrafts easily exceed 20–30 m s–1, only 5% of deep maritime clouds have maximum updraft velocity exceeding 10–15 m s–1 (Zipser and LeMone 1983; Emanuel 1994). It is widely accepted that this difference is caused by higher instability of the continental atmosphere. This is usually attributed to a potentially higher surface temperature (which, say, in summertime Texas conditions can exceed 36°C (Rosenfeld and Woodley 2000), while the sea surface temperature (SST) hardly exceeds 31°C. Note, however, that the land surface temperature often does not exceed the SST, while the existence of a higher vertical humidity gradient (a higher gradient of virtual temperature) and lower cloud base level should foster formation of strong updrafts in maritime clouds too. Nevertheless, under the same surface temperatures, vertical velocities in maritime clouds still remain lower than those in continental cumulus clouds. As it was shown in Section 3.2, one of the factors, decreasing cloud updrafts in maritime clouds is the low CCN concentration that leads to early formation of raindrops and their fall through the cloud updrafts. The dynamical effect was observed even in experiments with the Texas unstable conditions (not shown): both Figure 5. Maximal updrafts and downmaximum values of convective updrafts in clouds developed using the GATE drafts as well as downdrafts are profiles under continental and maritime aerohigher in clouds developed in smoky sol conditions. The development of secondair (continental aerosols). ary clouds in experiments cloud with a Since the instability of the marisuccesssive increase in the intensity of the first. time atmosphere is comparatively
Aerosol Impact on Precipitation from Convective Clouds
431
low, deep maritime convective clouds are often forced by surface level convergence caused by dissipating clouds (Ferrier and Houze 1989; Emanuel 1994). As a result, formation of secondary clouds turns out to be dependent on the properties of primary clouds. In Section 3.2, primary clouds were forced by an initial heating within a comparatively narrow zone of the subcloud layer. Besides, the wind shear was weak in these simulations. As a result, these clouds did not produce secondary clouds either in the M- and Ccases. As the width of the initial heating triggering the convection was successively increased from 2 to 4 km, the primary clouds became wider and the maximum vertical velocity increased. Besides, the wind shear was also increased to that measured at 261 day of the GATE (~ 7 m s–1 per 5 km in the middle troposphere). Figure 5 shows that in the C-case experiments the strengthening of the primary cloud leads to formation of a secondary cloud (seen by the formation of new vertical velocity maximum). At a certain stage, this secondary cloud gives rise to squall-line formation. Note that secondary clouds formed in corresponding M-case simulations were much weaker and did not develop into a squall line. This effect can be attributed to the following. Figures 3 and 4 show that convective heating by condensation and freezing signifycantly larger in C-cases than in M-cases. Because of wind shear transporting cloud hydrometeors downwind, cooling due to evaporation and sublimation takes place at some distance from the cloud updraft. This spatial shift between heating and cooling is accompanied by a decrease in the loading in the updraft zone (in contrast to M-cases, where raindrops can fall through cloud updraft). These factors lead to formation of dynamically induced vorticity that increases both vertical updrafts and downdrafts in clouds arising within microphysically continental aerosol. As a result, the maxima of updraft and downdraft vertical velocity are larger in the C-case clouds than in M-case clouds (Fig. 6).
Figure 6. Precipitation rate in M-case and C-case in simulations when GATE (261 day) sounding in case of squall-line formation.
432
A. Khain et al.
Figure 6 shows precipitation rate in both the M-case and C-case in simulations when the GATE (261 day) sounding was used, and the first cloud was triggered by a surface layer heating within the area of the 4 km width. Profiles of convective heating suggest (not shown) a significant precipitation rate, as the heating significantly exceeds the cooling. We attribute the increase in the precipitation efficiency to an increase in the relative humidity within the area of the squall line and to a significant increase in cloud cover (squall line is accompanied by significant melted rain falling from stratiform clouds behind the line).
4 CONCLUSION A 2D spectral microphysics cloud model has been used for the investigation of aerosol effects on cloud dynamics, microphysics, and precipitation. It is shown that (a) increase in concentration of CCN with radii (0.01 µm < rCCN < 1 µm) drastically decreases precipitation from clouds developed under unstable dry continental conditions; (b) an increase in the CCN concentration decreases precipitation from individual maritime clouds (and, as follows from supplemental simulations, under intermediate continental conditions) as well. This decrease is, however, smaller than in case of unstable continental conditions and highly depends on wind shear and air humidity. The reduction of the precipitation efficiency with the increase in the CCN concentration is caused by the increase in the loss of precipitation during sedimentation through a deep layer of dry air. An increase in the CCN concentration increases the intensity of convection and fosters the formation of squall lines. In this case precipitation rate significantly increases, indicating nonlinear effects in the cloud–aerosol interaction. The cloud lifetime as well as the area covered by clouds also increases with the increase in the concentration of CCNs of intermediate size. These aerosol effects should affect the radiation balance of the atmosphere. It is widely accepted that an increase in the concentration of aerosols in the atmosphere leads to atmospheric (and climatic) cooling. It was shown here that the increase in the aerosol concentration leads to an increase in cloud cover at higher levels in the atmosphere that leads to atmospheric heating. Thus, dynamical effects of aerosols on cumulus convection make net aerosol effects on climate not so obvious. This topic requires further investigation. The increase in cloud intensity with the CCN concentration, as well as the cloud top height should foster the lightning formation, and, possibly, other dangerous meteorological phenomena. It is shown that aerosols influence precipitation efficiency. Clouds of similar cloud top heights will precipitate differently under different aerosol
Aerosol Impact on Precipitation from Convective Clouds
433
conditions. Thus, the aerosol effects on precipitation should be taken into account in different rain retrieval algorithms. Acknowledgments: The study was performed under support of the Israel Ministry of Science (German–Israel collaboration in Water Resources, grant WT 0403), by the Israel Water Company (Shaham) as well as by EU project EURAINSAT.
5 REFERENCES Andreae, M. O., D. Rosenfeld, P. Artaxo, A. A. Costa, G. P. Frank, K. M. Longo, and M. A. F. Silva-Dias, 2004: Smoking rain clouds over the Amazon. Science, 303, 1337–1342. Beard, K. V. and H. T. Ochs, 1995: Collisions between small precipitation drops. Part II: Formulas for coalescence, temporary coalescence, and satellites. J. Atmos. Sci., 52, 3977– 3996. Bott, A., 1998: A flux method for the numerical solution of the stochastic collection equation. J. Atmos. Sci., 55, 2284–2293. Emanuel, K. A., 1994: Atmospheric convection. Oxford University Press, Oxford, 580 pp. Ferrier, B. S. and R. A. Houze, 1989: One-dimensional time dependent modeling of GATE cumulonimbus convection. J. Atmos. Sci., 46, 330–352. Hallett, J. and S. C. Mossop, 1974: Production of secondary ice crystals during the riming process. Nature, 249, 26–28. Hudson, J. G., 1993: Cloud condensational nuclei near marine cumulus. J. Geophys. Res., 98, 2693–2702. Hudon, J. G. and P. R. Frisbie, 1991: Cloud condensation nuclei near marine stratus. J. Geophys. Res., 96, 20795–20808. Hudson, J. G. and Y. Xie, 1999: Vertical distribution of cloud condensation nuclei spectra over the summertime northeast Pacific and Atlantic Ocean. J. Geophys. Res., 104, 30219– 30229. Kaufman, Y. J. and T. Nakajima, 1993: Effect of Amazon smoke on cloud microphysics and albedo-analysis from satellite imagery. J. Appl. Meteor., 32, 729–744. Khain, A. P. and I. Sednev, 1996: Simulation of precipitation formation in the Eastern Mediterranean coastal zone using a spectral microphysics cloud ensemble model. Atmos. Res., 43, 77–110. Khain, A., A. Pokrovsky, and I. Sednev, 1999: Some effects of cloud-aerosol interaction on cloud microphysics structure and precipitation formation: Numerical experiments with a spectral microphysics cloud ensemble model. Atmos. Res., 52, 195–220. Khain, A. P., M. Ovtchinnikov, M. Pinsky, A. Pokrovsky, and H. Krugliak, 2000: Notes on the state-of-the-art numerical modeling of cloud microphysics. Atmos. Res., 55, 159–224. Khain, A. P., M. B. Pinsky, M. Shapiro, and A. Pokrovsky, 2001: Graupel-drop collision efficiencies. J. Atmos. Sci., 58, 2571–2595. Khain, A. P., D. Rosenfeld, and A. Pokrovsky, 2001: Simulation of deep convective clouds with sustained supercooled liquid water down to –37.5°C using a spectral microphysics model. Geophys. Res. Lett., 28 (20), 3887–3890. Kogan, Y. L., 1991: The simulation of a convective cloud in a 3-D model with explicit microphysics. Part I: Model description and sensitivity experiments. J. Atmos. Sci., 48, 1160–1189. Low, T. B. and R. List, 1982: Collision, coalescence and breakup of raindrops, Part II: Parameterization of fragment size distributions. J. Atmos. Sci. 39, 1607–1618.
434
A. Khain et al.
Meyers, M. P. and W. R. Cotton, 1992: Evaluation of the potential for wintertime quantitative precipitation forecasting over mountainous terrain with an explicit cloud model. Part I: Two-dimensional sensitivity experiments. J. Appl. Meteor., 31, 26–50. Pinsky, M., A. P. Khain, and M. Shapiro, 2000: Stochastic effects on cloud droplet hydrodynamic interaction in a turbulent flow. Atmos. Res., 53, 131–169. Pinsky, M., A. P. Khain, and M. Shapiro 2001: Collision efficiency of drops in a wide range of Reynolds numbers: Effects of pressure on spectrum evolution. J. Atmos. Sci., 58, 742– 764. Pruppacher, H. R., 1995: A new look at homogeneous ice nucleation in supercooled water drops. J. Atmos Sci., 52, 1924–1933. Pruppacher, H. R. and J. D. Klett, 1997: Microphysics of clouds and precipitation, 2nd edn. Kluwer Academic, Dordrecht, 914 pp. Rosenfeld, D., 1999: TRMM observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 20, 3105. Rosenfeld, D., 2000: Suppression of rain and snow by urban and industrial air pollution. Science, 287 (5459), 1793–1796. Rosenfeld, D. and W. Woodley, 2000: Deep convective clouds with sustained supercooled liquid water down to –37.5°C. Nature, 405, 440–442. Rosenfeld D., R. Lahav, A. Khain, and M. Pinsky, 2002: The role of sea spray in cleaning air pollution over ocean via cloud processes. Science, 297, 1667–1670. Segal, Y., A. Khain, and M. Pinsky, 2004: Effects of atmospheric aerosol on precipitation in cumulus clouds as seen from 2000-bin cloud parcel microphysical model: sensitivity study with cloud seeding applications. Quart. J. Roy. Meteor. Soc., 130B (587), 561–582. Seifert, A., A. Khain, and U. Blahak, 2005: Possible effects of collisional breakup on mixedphase deep convection simulated by a spectral (bin) cloud model. J. Atmos. Sci., 62, 1917–1931. Takahashi, T., T. Endoh, and G. Wakahama, 1991: Vapor diffusional growth of free-falling snow crystals between –3 and –23°C. J. Meteor. Soc. Japan, 69, 15–30. Yin, Y., Z. Levin, T. Reisin, and S. Tzivion, 2000: The effects of giant cloud condensational nuclei on the development of precipitation in convective clouds: A numerical study. Atmos. Res., 53, 91–116. Vali, G., 1975: Remarks on the mechanism of atmospheric ice nucleation. Proc. 8th Int. Conf. on Nucleation, Leningrad, 23–29 Sept., I.I. Gaivoronsky, ed., Gidrometeoizdat, 265–269. Vali, G., 1994: Freezing rate due to heterogeneous nucleation. J. Atmos. Sci., 51, 1843–1856. Zipser, E. J. and M. A. LeMone, 1980: Cumulonimbus vertical velocity events in GATE. Part 2: Synthesis and model core structure. J. Atmos. Sci., 37, 2458–2469.
34 THE WISCONSIN DYNAMIC/MICROPHYSICAL MODEL (WISCDYMM) AND THE USE OF IT TO INTERPRET SATELLITE-OBSERVED STORM DYNAMICS Pao K. Wang Department of Atmospheric and Oceanic Science, University of Wisconsin–Madison, Madison, WI, USA
1 BRIEF DESCRIPTION OF THE CLOUD MODEL WISCDYMM AS CURRENTLY CONFIGURED This paper will briefly report on the cloud model that has been used in our group for various researches but especially those related with the physics and dynamics atop thunderstorms. The successful applications of this model to investigate several satellite-observed dynamical processes atop thunderstorms will also be summarized. The model is the Wisconsin dynamic and microphysical model (WISCDYMM), developed in the author’s research group. The earliest form was described in Straka (1989) and subsequently modified by others (Johnson et al. 1993, 1994; Lin and Wang 1997; Wang 2003). Its properties are described in the following sections.
1.1 Grid configuration and time step WISCDYMM uses a uniform staggered grid in all directions, placing the wind components on the normal grid cell faces and the remaining variables at the grid cell centers (Arakawa-C grid). Typically, these cells are assigned equal dimensions in both horizontal directions while it is usually smaller in the vertical. The time step size, assumed uniform, is dictated by quasicompressible computational stability requirements. The computational domain and resolution can be changed according to the needs of specific purposes. A typical setup of the grid for studying severe storm dynamics is given in the following: the grid mesh is 1.0 km horizontally and 0.5 km 435 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 435–446. © 2007 Springer.
436
P. K. Wang
vertically, with corresponding dimensions of 55 and 20 km for the model domain, while the time step is 3 s. In some cases, the resolution remains the same but the horizontal domain is expanded to 120 × 100 km and the top boundary is set at 30 km. For the purpose of studying the water vapor transport across the tropopause (Wang 2003), the vertical resolution was set at 200 m. The model also has been run with horizontal resolution of 500 m and good results were obtained.
1.2 Initialization The initial fields in WISCDYMM simulations consist of two components: (a) a horizontally homogeneous base state closely adapted from a prestorm rawinsounding, with no condensate and specifying the surface pressure from the sounding data and (b) an impulse to initiate the modeled storm. The rawinsounding data, which have irregular vertical spacing, are linearly interpolated (without smoothing) to the appropriate model grid levels as potential temperature, either water vapor mixing ratio or relative humidity, and the horizontal wind components relative to the earth. Base-state pressure values above ground are derived from the surface pressure by assuming a hydrostatically balanced environment. So far, the initial impulse has been an ellipsoidal warm bubble in the lower central part of the model domain, with the same relative humidities as in the base state.
1.3 Model physics WISCDYMM predicts the three wind components, turbulent kinetic energy, potential temperature, pressure deviation and mixing ratios for water vapor, cloud water, cloud ice, rain, graupel/hail, and snow. The model adapts the quasicompressible, nonhydrostatic primitive equation system of Anderson et al. (1985), rearranging the mass continuity equation to predict the pressure deviation much as in the fully compressible 3D cloud model of Klemp and Wilhelmson (1978), but allows time steps approximately three times larger by assigning acoustic waves a reduced pseudo-sound speed roughly twice the maximum anticipated wind speed. As in Klemp and Wilhelmson, subgrid transports are parameterized via 1.5-order “K-theory” to predict turbulent kinetic energy, from which a time- and space-dependent eddy coefficient is diagnosed for momentum and set 35% larger for the heat and moisture predictands (Straka 1989). As elaborated by Straka (1989), a version of WISCDYMM called hail parameterization model (HPM) features a bulk microphysics parameterization that entails water vapor and five hydrometeor types: cloud water, cloud ice, rain, graupel/hail, and snow, with 37 individual transfer rates (source/sink terms). Adapted largely from Lin et al. (1983) and Cotton et al. (1982, 1986), this package treats all hydrometeors as spheres except for cloud ice, which is treated as small hexagonal plates. Cloud water and cloud ice are assumed
The Wisconsin Dynamic/Microphysical Model
437
monodisperse, with zero fall speed relative to the air. All three precipitation classes have inverse-exponential size distributions, with temperature-dependent intercepts for snow and graupel/hail, while the intraspectral variation of particle fall speed versus diameter for each is assumed to satisfy a power law. If necessary, WISCDYMM can also be run in the HCM mode in which the evolution of hailstones can be tracked by studying the growth of hail sizes in a number (e.g., 25) of size bins. The HCM has been tested successfully in Straka (1989). WISCDYMM is also programmed to activate one or more of the following three iterative microphysical adjustments (Straka 1989) where and if needed: •
• •
A saturation adjustment is performed to: (a) condense cloud water (or depose cloud ice) to eliminate supersaturation, releasing latent heat of evaporation (or sublimation), or (b) evaporate cloud water (or sublimate cloud ice) in subsaturated air until either saturation is reached or the cloud water (or cloud ice) is exhausted, absorbing latent heat instead. Cloud water is adjusted first and cloud ice second, incrementing the water vapor and temperature to suit. In-cloud saturation mixing ratios are weighted between their values with respect to water and ice in proportion to the relative amounts of cloud water and cloud ice respectively. Where no cloud is present, saturation is taken with respect to water or ice if the temperature is above or below 0°C, respectively. More than three iterations are rarely needed. Prior to the saturation adjustment, any cloud ice at temperatures above 0°C is melted, and any cloud water at temperatures below –40°C is frozen, respectively absorbing or releasing latent heat of fusion. After partial update of the moisture fields by advection and turbulent mixing, the decrement in each hydrometeor field due to the net sink (the sum of the individual microphysical sink terms) is compared to its available supply, defined as its partially updated mixing ratio plus the increment due to its net source (the sum of the individual microphysical source terms). If the net sink of a hydrometeor class exceeds its available supply, it is prorated downward along with each of its components so as to not exceed 25% of the available supply. The procedure is iterative because prorating down the sinks of one class also reduces the corresponding source terms for one or more other classes, but more than two iterations are rarely needed.
1.4 Boundary conditions The lateral boundary conditions are similar to those in Klemp and Wilhelmson (1978). Reflection of outward-propagating gravity waves is suppressed by “radiation” conditions which advect each horizontal normal wind component out with a velocity equal to itself plus a prescribed constant
438
P. K. Wang
outward gravity wave propagation of speed c*, except assigning zero advection in inflow of speed greater than c*. The other predictands at lateral boundary outflow locations are advected by upstream differencing. Both the upper and lower boundaries are rigid lids. Variables at the top are held undisturbed, while reflection of upward-propagating gravity waves off the lid is suppressed by imposing an upper-level Rayleigh damping layer that abuts it. The lower boundary has four options: free-slip with no surface heat flux; semi-slip with no surface energy budget; no-slip with heat flux and no surface energy budget; or no-slip with heat flux, insolation, and surface energy budget.
1.5 Interior numerics The current version of WISCDYMM uses forward-in-time differencing and sixth-order flux-conservative Crowley spatial differencing (Tremback et al. 1987). To suppress nonlinear instability, a fourth-order numerical diffusion operator with a constant coefficient, as in Klemp and Wilhelmson (1978), is added in the discretized predictive equations at each time step.
2 SIMULATION OF THE 2 AUGUST 1981 CCOPE SUPERCELL THUNDERSTORM Up to the present, WISCDYMM has been used primarily for simulating deep convective events. It has successfully simulated many cases of severe thunderstorms occurring in the USA and other parts of the world. To illustrate the capability of this model, the simulation results of the 2 August 1981 supercell storm that occurred in Montana in the Midwest of USA will be briefly presented. In later sections, the simulation results of this storm will be used as examples for understanding some thunderstorm dynamical processes as observed by meteorological satellites.
2.1 A brief description of the 2 August 1981 CCOPE supercell The storm chosen for the simulation for illustrating the plume-formation mechanism is a supercell that passed through the center of the Cooperative Convective Precipitation Experiment (CCOPE) (Knight 1982) observational network in southeastern Montana on 2 August 1981. The storm and its environment were intensively observed for more than 5 h by a combination of seven Doppler radars, seven research aircraft, six rawinsonde stations and 123 surface recording stations as it moved east–southeastward across the CCOPE network. Miller et al. (1988) and Wade (1982) provided many of the observations in this section, especially those on the history of the storm. This case was chosen because it is a typical deep convective storm in the US High Plains and it provides much detailed observational data for comparison with
The Wisconsin Dynamic/Microphysical Model
439
model results with regard to dynamics and cloud physics, and the author’s group has obtained successful simulations of it previously (Johnson et al. 1993, 1994).
2.1.1.
Environmental conditions
The initial conditions for the simulation are based on a 1,746 (mountain daylight time (MDT) sounding (Fig. 1) taken at Knowlton, Montana, approximately 90 km ahead of the storm. This sounding provided the most representative temperature and moisture profile available, with a massive convective available potential energy (CAPE) 3,312 J kg–1 distributed over a comparatively shallow layer from the level of free convection LFC = 685 hPa to the equilibrium level EL = 195 hPa. The subcloud layer (below 730 hPa) was nearly dry-adiabatic and well mixed, with a potential temperature close to 311.5 K, and also relatively moist because a surface low in north central Wyoming advected water vapor mixing ratios of 12–13 g kg–1 into the region on easterly winds. Above the subcloud region, a strong capping dry layer existed at approximately 710 hPa, caused by warmer and drier air that had unexpectedly moved into the region after 1,300 MDT. Wade (1982) gives some possible causes of this warming. The dry layer was significant in that it allowed the low-level air mass to continue warming for the remainder of the afternoon and become even more potentially unstable. From the dry layer to 450 mb, the environmental lapse rate was nearly dry adiabatic. The calculated indices from the Knowlton sounding (totals index = 60, lifted index = –9.4, and a K index = 38) indicated that the air mass over eastern Montana on 2 August was very unstable, and hence very favorable for the development of deep convection. Large vertical wind shear between lower and midlevels was also conductive to severe weather development. The 1,746 MDT Knowlton hodograph (not shown) indicated strong subcloud flow, veering nearly 70° from the surface layer to cloud base at 1.6 km AGL. The magnitude of the mean shear over the lowest 6 km was 0.008 s–1. There was little directional shear above the cloud base, but vertical speed shears between the cloud base and 9 km were 0.006 s–1 (Miller et al. 1988). Taking into account the vertical wind shear and buoyancy effects, the Bulk Richardson Number for the prestorm environment was 25, in the expected range for supercell storms. Some previous studies have pointed out that clockwise curvature of the wind shear vector over the lowest 2 km of the hodograph also favored development of the right-moving supercell.
2.1.2.
Examples of simulated microphysical and dynamical fields
To illustrate the model performance, Figs. 2 and 3 show the simulated hydrometeor mixing ratio and vertical velocity fields in the central east–west
P. K. Wang
440
Figure 1. The 1,746 MDT Knowlton, Montana sounding on 2 August 1981. The solid curve is for temperature and dashed curve for dew point. The portion of dew point curve above 300 hPa, which was missing in the original sounding, is constructed using an average August 1999 HALOE water vapor profile over 40–60N.
vertical cross section of the storm at t = 40 min. In general, the model results agree with the observed behavior of the supercell very well.
2.1.3.
Simulation of anvil top cirrus plumes
As indicated before, WISCDYMM has been used successfully to study cloud dynamical and physical processes atop deep convective storms. One recent example is the identification of the physical mechanism that produces the cirrus plumes above the anvils of some severe thunderstorms. Such plumes have been observed from satellite visible and infrared (IR) images and some details of them have been studied by a number of investigators (Setvak and Doswell 1991; Levizzani and Setvák 1996). Figure 4 shows an example of such plumes. Since the anvils of these severe storms were already at the tropopause level, the plumes must have been higher up in the stratosphere. In one case, Levizzani and Setvák (1996) estimated that the
The Wisconsin Dynamic/Microphysical Model
441
plume was located at ~15 km which was about 3 km above the local tropopause at the time. Since water vapor in the stratosphere has significant implications on the global climate process because of its strong absorption of IR radiation, it is important to understand where the water vapor source is and how much is injected into the stratosphere.
Figure 2. Simulated hydrometeor mixing ratios and storm-relative vector wind projection field for the CCOPE storm of 2 August 1981, in x–z (east–west) vertical cross sections through the maximum updraft as of 90 min. Solid – snow, dotted – graupel and hail, dashed – cloud droplets, dash dot – rain, dash dot dot – cloud ice.
CCOPE Supercell 2 Aug 1981 y = 27 km t = 40 min 0
0 0.5 0
18
1
0
-0.2 -1 5
01
-0.25
0 0. 5
7
-0.25
10
5 -0.2
-1
-0.5 3 -0.5 0.5 0-3 -1 7 10 -0.25 20 1 30
0
-1
1
.5 -0
-0.5
0
5
1 -1 5
3
1 0 0.5 -1
-0.5
-0 5 .2
25
0
-0.5
7
0.5 -3
20
0
15
-0 .2 5
0.5
10
-0.5
z (km)
0. 5
-3
30
7
0
10
-0 .5
-5
-0.5
0
5
-0.2 05.5 1
0
.2 5
-0.25
-0
0
20
.25
-3
-0
4
-1
-3
-1
-0.5
6
2
5
1
-0.5
0 0.5
8
0
0
-0.5 -1
-5 3 40
-0.2 5
10
5
12
-1
0
0. 5 1
-0.2
0
0.5
0
14
-0-0 .2.5 5 1
25 -0.
16
30
35
40
45
50
55
x (km)
Figure 3. The vertical velocity (w) field in the central east–west cross section of the simulated storm at t = 40 min. Solid (dashed) contours represent positive (negative) w (in m s–1).
442
P. K. Wang
Figure 4. GOES-8 composite image (cannel 1 + channel 3 + channel 4) on 6 May 2002 0015 UTC showing plumes on top of thunderstorms (white arrows) in Texas and Oklahoma border region (courtesy of NOAA).
Figure 5. Snapshots of modeled RHi (relative humidity with respect to ice) profiles at t = 24, 32, 40, 80, 96, and 112 min in the central east–west cross section (y = 27 km), showing the plume feature above the anvil. Only the portion near the cloud top is shown. The vertical axis range is 10–20 km and horizontal axis range 20–55 km (from Wang 2003).
The Wisconsin Dynamic/Microphysical Model
443
In order to identify the physical mechanism responsible for the plume formation and the source of its water vapor, we studied the simulation results of the CCOPE supercell to see whether or not the simulated storm produces the same plume phenomenon. The answer turned out to be positive. Figures 5 and 6 show the simulated plume phenomenon in the central cross-sectional view and 3D cloud top view, respectively (Wang 2003). The model-produced plumes exhibit nearly the same major characteristics of the observed ones, hence it is highly likely that the observed plumes must have been produced in a manner similar to that occurred in the simulated storm. Careful analysis of the model results indicate that the moisture forming the plumes come from the storm below, and the mechanism that eject moisture from the troposphere into the stratosphere is the breaking of cloud top gravity waves (Wang 2003). This demonstrates that the results generated by the cloud model are realistic and can be used as a substitute (when appropriate) for studying physical processes in thunderstorms whereas in situ or remote observations are either difficult or can provide only limited temporal and spatial coverage.
Figure 6. Snapshots of 3D renderings for the 30% RHi contour surface at t = 24, 32, 40, 80, 96, and 112 min, showing the plume features above the anvil. Data below 10 km are windowed out. (From Wang 2003.)
P. K. Wang
444
2.1.4.
Simulation of Fujita’s jumping cirrus
Fujita (1982) described the observation of the jumping cirrus phenomenon above a thundercloud from an aircraft as follows: One of the most striking features seen repeatedly above the anvil top is the formation of cirrus cloud which jumps upward from behind the overshooting dome as it collapses violently into the anvil cloud.
Figure 7. Snapshots of the RHi profiles in the central east–west cross section of the simulated storm from t = 1,320 –2,640 s. The RHi scale is similar to that in Fig. 5.
The Wisconsin Dynamic/Microphysical Model
445
In a later paper he made a more elaborated report on this phenomenon (Fujita 1989). Again due to the limited temporal and spatial coverage that can be provided by such observations, it is generally difficult to obtain detailed data for conclusively interpreting this phenomenon. At the time there were questions about whether it is possible to have cirrus clouds jumping upstream (i.e., against the wind). But by carefully studying the cloud top dynamics using the same WISCDYMM-simulated results of the CCOPE supercell, it can be seen that the gravity wave breaking related to the anvil top plume phenomenon is also responsible for the jumping cirrus phenomenon. The cirrus is not really jumping against the wind but it moves upstream only relative to the storm. If we keep in mind that the simulated storm is moving at ~30 m s–1 during its development, then the cirrus is still moving downwind relative to the surface. Figure 7 shows a series of the rendered cloud top humidity profiles that illustrate this process. Acknowledgments: This work is partially supported by NSF Grants ATM0234744 and ATM-0244505 to the University of Wisconsin-Madison.
3 REFERENCES Anderson, J. R., K. K. Droegemeier, and R. B. Wilhelmson, 1985: Simulation of the thunderstorm subcloud environment. Prepr. 14th Conf. Severe Local Storms. Indianapolis, IN., Amer. Meteor. Soc., 147–150. Cotton, W. R., G. J. Tripoli, R.M., and E. A. Mulvihill, 1986: Numerical simulation of the effects of varying ice crystal nucleation rates and aggregation processes on orographic snowfall. J. Climate Appl. Meteorol., 25, 1658–1680. Cotton, W. R., M. A. Stephens, T. Nehrkorn, and G. J. Tripoli, 1982: The Colorado state University three-dimensional cloud model – 1982. Part II: An ice phase parameterization. J. Rech. Atmos., 16, 295–320. Fujita, T. T., 1982: Principle of stereographic height computations and their application to stratospheric cirrus over severe thunderstorms, J. Meteor. Soc. Japan., 60, 355–368. Fujita, T. T., 1989: The Teton-Yellowstone tornado of 21 July 1987. Mon. Wea. Rev., 117, 1913–1940. Johnson, D. E., P. K. Wang, and J. M. Straka, 1994: A study of microphysical processes in the 2 August 1981 CCOPE supercell storm. Atmos. Res. 33, 93–123. Klemp, J. B. and R. B. Wilhelmson, 1978: The simulation of three-dimensional convective storm dynamics. J. Atmos. Sci., 35, 1070–1096. Knight, C. A. (ed.), 1982: The Cooperative convective precipitation experiment (CCOPE), 18 May–7 August 1981. Bull. Amer. Meteor. Soc., 63, 386–398. Levizzani, V. and M. Setvák, 1996: Multispectral, high resolution satellite observations of plumes on top of convective storms. J. Atmos. Sci., 53, 361–369. Lin, H.-M. and P. K. Wang, 1997: A numerical study of microphysical processes in the 21 June 1991 Northern Taiwan mesoscale precipitation system. Terres. Atmos. Oceanic Sci., 8, 385–404.
446
P. K. Wang
Lin, Y. L., R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22, 1065–1092. Miller, L. J., D. Tuttle, and C. A. Knight, 1988: Airflow and hail growth in a severe northern High Palins supercell. J. Atmos. Sci., 45, 736–762. Setvak, M., R. M. Rabin, C. A. Doswell III, and V. Levizzani, 2002: Satellite observations of convective storm tops in the 1.6, 3.7 and 3.9 µm spectral bands. Atmos. Res., 67–68, 607–627. Straka, J. M., 1989: Hail growth in a highly glaciated central High Plains multi-cellular hailstorm. Ph.D. Diss., Dept. Meteorology, University of Wisconsin, Madison, WI, 413 pp. Tremback, C. J. P., W. R. Cotton, and R. A. Pielke, 1987: The forward-in-time upstream advection scheme: Extension to higher order. Mon. Wea. Rev., 115, 540–555. Wade, C. G., 1982: A preliminary study of an intense thunderstorm which move across the CCOPE research network in southeastern Montana. Proc. 9th Conf. on Weather Forecasting and Analysis. Seattle, WA, Amer. Meteor. Soc., 388–395. Wang, P. K., 2003: Moisture pumes above thunderstorm anvils and their contributions to cross tropopause transport of water vapor in midlatitudes. J. Geophys. Res., 108 (D6), 4194, doi: 10.1029/2003JD002581. Wang, P. K. et al., 2001: A cloud model interpretation of the enhanced V and other signatures atop severe thunderstorms. Prepr. 11th Conference Satellite Meteorology and Oceanography, 15–18 October 2001, American Meteorological Society, Madison, WI, 402–403.
35 THE EUROPEAN CENTRE FOR MEDIUMRANGE WEATHER FORECASTS GLOBAL RAINFALL DATA ASSIMILATION EXPERIMENTATION Peter Bauer1, Philippe Lopez1, Emmanuel Moreau2, Frédéric Chevallier3, Angela Benedetti1, and Marine Bonazzola1 1
European Centre for Medium-Range Weather Forecasts, Reading, UK NOVIMET, Velizy, France 3 Laboratoire des Sciences du Climat et l'Environnement, Paris, France 2
1 INTRODUCTION The quality of today’s numerical weather prediction (NWP) systems is driven by the quality of data that is used to determine the present state of the atmosphere and the quality of the representation of physical processes in the model. The data usage is optimized if the analysis, that is the methodology for deriving the most realistic state of the atmosphere–land–ocean system at a given time, is capable of capturing the four-dimensional (4D) development of this highly variable system in accordance with observations that are distributed in space and time. Principally, this represents an optimization problem because a state is sought that agrees best with a priori information from a short-range forecast that is based on a previous analysis and observations. The observations may be in geophysical terms (e.g., temperature or humidity) or electromagnetic terms (e.g., radiance or reflectivity). As in all optimization problems, the errors associated with each component provide a weight for each component in the analysis. At ECMWF, 4D data assimilation was first implemented in 1997 (e.g., Rabier et al. 2000) and run very successfully for global analyses and forecasts since then. A fundamental principle of the assimilation at ECMWF is its “incremental formulation”. In essence, this means that the short-range forecast is accurate enough to assume a linear dependence of model physics on state variable increments in the vicinity of the forecasted (first-guess) 447 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 447–457. © 2007 Springer.
448
P. Bauer et al.
state. The increments are produced from the difference between model and observation related states (Courtier et al. 1994). Figure 1 summarizes the logical flow. A high-resolution (~45 km in the horizontal and on 60 model levels) calculation of the model trajectory over the (12-h) assimilation window is carried out that is initialized with a short-range forecast from the previous analysis. The model state is compared to observations and interpolated to a lower resolution. The first minimization loop (inner loop) is based on the lowresolution trajectory model fields and the departures between observations and the high-resolution trajectory model fields. The resulting increments serve for updating the high-resolution fields. At the first iteration, these originate from the first guess. After, these are replaced by the higher-iteration updates (outer loop).
Figure 1. Incremental 4D-Var data assimilation algorithm (for details see text). x = model control variable, S = interpolation operator, d = departures, H = observation operator, y = observations, J = cost function, δ = increment, ∇ = gradient, indices ‘0’, ‘b’, ‘i’, ‘a’ denote control variable before assimilation, background, during update, and from analysis, respectively. Block arrows indicate temporal integration (Tremolet 2004).
Data from several observation types is assimilated. The bulk of the observations is obtained from satellites and their impact has recently proven to exceed the impact of conventional observations even in the northern hemisphere (Simmons and Hollingsworth 2002). Secondly, the difference in
The European Centre for Medium-Range weather Forecasts
449
forecast skill for the northern and southern hemispheres has been reduced dramatically indicating the beneficial impact of satellite observations in otherwise data-sparse areas. While the forecast of the dry atmospheric state (wind, temperature, surface pressure) has reached an unprecedented accuracy, the quality of both humidity analysis and forecast are still unsatisfactory due to the insufficient accuracy of the parameterization of moist physical processes and due to insufficient observations. The parameterization schemes have to cover the response of the atmospheric moisture fields to large-scale dynamics, cloud and precipitation formation, and fluxes at the surface–atmosphere boundary. Beljaars (2003) assesses the ECMWF model performance with respect to the atmospheric moisture representation. His main conclusions are that too much precipitation is generated over the entire dynamic range of rain rates but that the model has a dry bias outside areas with precipitation. The precipitationbias and its reflection in total column water vapor (TCWV) was confirmed by Marécal et al. (2001, 2002) when comparing near-surface rain estimates from satellite data and model fields, as well as the atmospheric moisture required for producing the respective rainfall intensities. The clear-sky dry bias has been noted at least since the assimilation of satellite data that is almost entirely sensitive to the TCWV (Gérard and Saunders 1999) and its manifestation in the so-called tropical precipitation spin-down that is the overproduction of rain at the beginning of the forecast originating from the moistening of the atmosphere in the analysis. This, of course, feeds back into the large-scale dynamics through the release of latent heat. Apart from these problems, the global analyses are biased towards clearsky observations because almost without exception cloud and rain contaminated data are rejected. This is because: • Large discrepancies between model forecast and observations can be expected due to oversimplified parameterizations. • The observation operator, i.e., the model that translates between model state variables and observables may be nonlinear due to the strongly nonlinear response of cloud and precipitation schemes to moisture increments (Marécal and Mahfouf 2003), as well as due to the nonlinear relationship between radiances and water vapor/condensates. • Even if these observations were assimilated, the effect could dissipate over the subsequent forecasts. The first two aspects may be in conflict with the general linearity assumption in the incremental 4D-Var formulation and can lead to convergence failures in the minimization. The last issue may lead to a negligible impact on model forecasts and therefore may not solve the fundamental shortcomings in the models’ description of the water cycle. Therefore, the assimilation of observations related to clouds and precipitation represents a major challenge for NWP and requires the combination of
450
P. Bauer et al.
better parameterizations, a numerically efficient and stable observation operators, and quality-controlled observations. The experience with operational rainfall assimilation produced mixed results due to its little impact on model dynamics in existing data assimilation systems (e.g., since 2001 in NCEPGDAS, since 2002 in JMA-MSM). However, the ECMWF 4D-Var system provides an ideal environment for evaluating the optimum configuration for rainfall data assimilation given the current quality of moist physical parameterizations at the highest available horizontal and vertical resolution of the ECMWF model.
2 1D-VAR Since 1998, large efforts have been made to prepare the assimilation of precipitation information at ECMWF. A major part of the studies focused on the above issues for which Marécal and Mahfouf (2000, 2002) have laid the foundation for a methodology that became operational at ECMWF in 2005. In this methodology, a one-dimensional variational retrieval (1DVar) algorithm is applied with near-surface rain rates or passive microwave brightness temperatures (TBs) as observables and TCWV as the retrieval variables. The main reason for subsetting the 4D-Var analysis with a 1D-Var retrieval of TCWV is the nonlinear relationship between the control vector and the precipitation affected observations that may violate the incremental 4D-Var setup. Secondly, the lower model resolution and the simplified physics schemes in the inner loop (~120 km) may deteriorate the convergence if precipitation observations would be assimilated directly in 4D-Var. These limitations may be overcome in the near future with a better model resolution and more consistent physics included in the inner loop. The observation operator that is used for the 1D-Var algorithm uses new linearized cloud (Tompkins and Janisková 2004) and convection schemes (Lopez and Moreau 2004), as well as a radiative transfer model (Bauer 2002; Moreau et al. 2003a). The latter has recently been integrated into the operational RTTOV package (Saunders et al. 1999) that is available for a large NWP community. One of the issues in the 1D-Var retrieval is whether to use near-surface rainfall estimates or TBs as observables. Both options were analyzed by Moreau et al. (2003b). Figure 2 shows an example of TCWV increments that were produced by the 1D-Var retrieval using rainrates (1D-RR) or TBs (1DTB) from Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) observations over tropical cyclone “Mitag” on 3 March 2002. The different coverage originates from different screening methods. In this example, 1D-TB tends to produce larger increments due to less saturation at larger rain intensities if all lower TMI channels between 10 and 37 GHz are used (Moreau et al. 2003a, b).
The European Centre for Medium-Range weather Forecasts
10 5 3 2 1 0.5 −0.5 −1 −2 −3 −5 −10 −20 121E 123E 125E 127E 129E 131E 133E 135E 137E 139E
30
21
RbR0
40
50 60 −0.3 0.0 0.3 Mean T increments (K)
30 Model levels
21
RbR0
20
19N 18N 17N 16N 15N 14N 13N 12N 11N 10N 9N 8N 7N 6N
10 5 3 2 1 0.5 −0.5 −1 −2 −3 −5 −10 −20 121E 123E 125E 127E 129E 131E 133E 135E 137E 139E
21
21
RbR0
30
30
40
40
50
50
60 −1.5 0.0 1.5 -1 Mean Qv increments (g kg )
60 −0.3 0.0 0.3 Mean T increments (K)
Model levels
20 19N 18N 17N 16N 15N 14N 13N 12N 11N 10N 9N 8N 7N 6N
451
RbR0
40
50 60 −1.5 0.0 1.5 -1 Mean Qv increments (g kg )
Figure 2. TCWV increments (kg m–2) from 1D-Var retrievals using rain rates (top left panel) and brightness temperatures (top right panel). Bottom panels show averaged profiles of temperature and specific humidity increments, respectively. Cyclone “Mitag” on 3 March 2002, 1200 UTC.
Figure 2 also shows mean profiles of temperature and humidity increments produced by the 1D-Var. If temperature increments are transferred to humidity by assuming saturation, their values are 5–10 times smaller than those of humidity itself. This suggests the dominance of the sensitivity of cloud and convection schemes to moisture changes and that it is reasonable to only assimilate moisture increments in 4D-Var (through TCWV) following the 1D-Var retrieval. The profiles also show that increments are larger with 1D-TB. For 1D-RR, the profile shapes are rather homogeneous because they are only produced by the moist-physics parameterizations and the shape of the background error standard deviation profiles. For 1D-TB, also the sensitivity of the TBs to changes in hydrometeor profiles (TB-Jacobians) plays a role. These peaks at different altitudes per channel and hydrometeor type (cloud water, rain, snow). The maximum of increments near model level 50 (850–900 hPa) is a result of the domination of the background error standard deviation profile. When investigating the 1D-Var performance more thoroughly, Moreau et al. (2003b) found no clear indication that 1D-RR is outperformed by 1DTB. However, 1D-RR requires retrieval algorithms that invert TBs to rain rates and that provide a retrieval error estimate that is compatible with the
452
P. Bauer et al.
model rainfall error estimate (Bauer et al. 2002). There is a rather large range of products for the estimation of near-surface rainfall retrieval accuracy from comparisons to local reference measurements; however, efforts to estimate global retrieval accuracy have not been successful in identifying a single retrieval approach as being superior to others (e.g., Ebert and Manton 1998). The choice of TBs as observables circumvents this problem, but only partially, because a proper estimation of modeling errors from the observation operator (cloud/convection + radiative transfer) is required for the 1D-Var algorithm as well. Practically, the error estimation can be solved from the requirement that the departures between observed and modeled TBs are consistent with the errors associated with the model variables (expressed in TB-space) and those of the observations. Therefore, the error estimation aims at a balance between realistic observation errors, model forecast errors (again in TB-space), and the difference between observations and model simulations in the analysis. The biggest advantage of using TBs, however, is the larger flexibility with regard to which channel and which sensor is employed compared to algorithms that are usually sensor-specific. The results of Moreau et al. (2003b) also indicate that 1D-TB does not require the presence of rain or clouds in the background because TBs are sensitive to clear-sky TCWV as well. The usage of 1D-RR may increase/decrease rain when present in the background but cannot create rain where the background is rain-free because there, the derivative of model state with respect to rain is not defined. This is illustrated in Fig. 3 by comparing the model background, 1D-RR and 1D-TB, as well as TRMM precipitation radar (PR) data over tropical cyclone “Ami” on 14 January 2003, 18 UTC. All states are expressed in equivalent PR reflectivity. The PR observations show that the maximum rain intensities in cyclone “Ami’s” rain-band are displaced with respect to the model background. Both 1D-Var retrievals succeed to correct the background but 1D-RR fails to produce sufficient rainfall near the maximum of the rain intensities at 26S/172W due to missing rainfall in the background.
3 1D-VAR + 4D-VAR Figure 4 shows examples of precipitation forecasts based on a control experiment (operational ECMWF model) and 4D-Var analyses using 1D-RR and 1D-TB TCWV retrievals, respectively. The rain patterns and rain intensities are significantly modified by assimilating rain affected observations. For both 1D-RR and 1D-TB, the cyclone intensity is increased and this intensification is maintained over the selected forecast period. The fact that the rain assimilation can displace precipitating systems ensures that there remains a continuous impact in the forecast with respect to cloud system location. However, when regionally averaged precipitation intensities are analyzed, no significant change of the hydrological budget was observed (not shown here).
The European Centre for Medium-Range weather Forecasts
453
Figure 3. Radar reflectivities from TRMM PR (upper left), model background (upper right), 1D-RR and 1D-TB analyses. Tropical cyclone “Ami” on 14 January 2003, 1800 UTC.
Figure 4. Forecasts of accumulated precipitation (mm) from control, 1D-TB and 1D-RR assimilation experiments (from left to right) vs. rain-rate retrieval (lower right panel). Cyclone “Zoe” on 26 December 2002, 1200 UTC. Color scale: 0.1, 0.3, 1, 2, 3, 5, 10, 20, 50, 100 mm.
454
P. Bauer et al.
Figure 5 shows the global distribution of TCWV analysis increments from a 1D-Var + 4D-Var experiment using TMI rain-rate observations. Figure 5a contains the humidity increments of all observations while Fig. 5b isolates the increments that originate from the assimilation of precipitation information. The nature of the latter is featured in the finer structures and the change in increment sign on small scales. This is the result of the displacement of rain systems, i.e., the drying of the model background where there was no or little rain in the observation and vice versa. The magnitude of increments of all and rain observations is comparable. The 4D-Var analysis spreads humidity increments beyond the limits of data availability (40N– 40S) and therefore the assimilation of TMI data has a global impact. The direct impact of moisture increments on the precipitation forecast is illustrated in Fig. 5c that contains the accumulated precipitation between days 1 and 2 of the forecast period. Here, the displacement of precipitating systems is even more obvious than from the increments in Fig. 5b.
4 CONCLUDING REMARKS In the way the assimilation of precipitation information is carried out at ECMWF, an observation operator is required that translates between model control variables, e.g., temperature and humidity, and observables, e.g., rain rates, brightness temperatures, or reflectivities. This operator contains physical parameterizations for cloud and rain generation, as well as a radiative transfer model. The constraint on the model analysis that is produced by these observations depends on the balance between model background and observation error statistics. If profile information is assimilated, the vertical distribution of analysis increments is crucial as well because it will affect the response of the model physics to the assimilated data in the forecast. For example, identical moisture increments added near the bottom or the middle of the troposphere will have a different effect on cloud and rain generation during the forecast. All the above elements contain rather large uncertainties and the success of rainfall assimilation is strongly determined by the accuracy of the observation operator and the protection of the analysis from those observations that can lead to inconsistencies in the minimization. This safety requirement is even more difficult to fulfill because many other observations are assimilated in the vicinity of precipitation whose effect may interact with that of observations inside precipitation. Most of the recent work carried out at ECMWF dealt with the implementation of a computationally efficient yet accurate observation operator and with the investigation of the 4D-Var analysis performance once rain information is assimilated.
The European Centre for Medium-Range weather Forecasts
455
Figure 5. Analysis increments of TCWV (in kg m–2) on 7 April 2003, 0000 UTC from all observations (a) and rain observations (b). (c) 48–24 h precipitation forecast (in 10–3 mm) initialized on 1 April 2003, 1200 UTC with rain observations (see also color plate 16).
456
P. Bauer et al.
Due to the remaining uncertainties in this process, above all the comparably simple physical parameterization schemes applied in global models, some longer-term assimilation studies have been started to optimize the usage of rainfall information in the ECMWF assimilation system prior to its operational implementation. Since the model physics reacts sensitively to a change of moisture state, future options will include the combined use of microwave observations that are sensitive to liquid precipitation in lower layers and frozen precipitation and cloud water/ice in the upper layers. First studies on rainfall radar data assimilation at ECMWF have indicated that the vertical distribution of moisture increments can have a significant effect on the latent heating profile and therefore the relation between moisture and energy budget. In summary, the assimilation of rainfall and cloud information offers a large potential for future applications and its success can be considered as a benchmark towards improving moist physical parameterizations in global modes and data assimilation systems, as well as the prediction of severe weather systems.
5 REFERENCES Bauer, P., 2002: Microwave radiative transfer modeling in clouds and precipitation. Part I: Model description. NWP SAF Report No. 5. Available from The Met Office, Exeter, UK, pp. 27. Bauer, P., J.-F. Mahfouf, S. di Michele, F. S. Marzano, and W. S. Olson, 2002: Errors in TMI rainfall estimates over ocean for variational data assimilation. Quart. J. Roy. Meteor. Soc., 128, 2129–2144. Beljaars, A., 2003: Some aspects of modelling of the hydrological cycle in the ECMWF model. Proceedings of ECMWF/GEWEX Workshop on Humidity Analysis, 8–11 July 2002, ECMWF, Reading, UK, 191–202. Courtier, P., J.-N. Thépaut, and A. Hollingsworth: A strategy for operational implementation of 4D-Var using an incremental approach. Quart. J. Roy. Meteor. Soc., 120, 1367–1388. Ebert, E. and M. Manton, 1998: Performance of satellite rainfall estimation algorithms during TOGA-COARE. J. Atmos. Sci., 55, 1537–1557. Gérard, E. and R. Saunders, 1999: Four-dimensional variational assimilation of Special Sensor Microwave/Imager total column water vapour in the ECMWF model. Quart. J. Roy. Meteor. Soc., 125, 1453–1468. Lopez, P. and E. Moreau, 2004: A convection scheme for data assimilation: Description and initial tests. Technical report. ECMWF Technical Memorandum No. 411. Available from ECMWF Shinfield Park, Reading, UK, pp. 29. Mahfouf, J.-F., V. Marécal, and P. Bauer, 2003: The assimilation of SSM/I and TMI rainfall rates in the ECMWF 4D-Var system. Quart. J. Roy. Meteor. Soc., 128, 2737–2758. Marécal, V. and J.-F. Mahfouf, 2000: Variational retrieval of temperature and humidity profiles from TRMM precipitation data. Mon. Wea. Rev., 128, 3853–3866. Marécal, V. and J.-F. Mahfouf, 2002: Four-dimensional variational assimilation of total column water vapour in rainy areas. Mon. Wea. Rev., 130, 43–58.
The European Centre for Medium-Range weather Forecasts
457
Marécal, V. and J.-F. Mahfouf, 2003: Experiments on 4D-Var assimilation of rainfall data using an incremental formulation. Quart. J. Roy. Meteor. Soc., 129, 3137–3160. Marécal, V., J.-F. Mahfouf, and P. Bauer, 2002: Comparison of TMI rainfall estimates and their impact on 4D-Var assimilation. Quart. J. Roy. Meteor. Soc., 128, 2737–2758. Marécal, V., E. Gérard, J.-F. Mahfouf, and P. Bauer, 2001: The comparative impact of the assimilation of SSM/I and TMI brightness temperatures in the ECMWF 4D-Var system. Quart. J. Roy. Meteor. Soc., 127, 1123–1142. Moreau, E., P. Bauer, and F. Chevallier, 2003a: Variational retrieval of rain profiles from spaceborne passive microwave radiance observations. J. Geophys. Res., 108, ACL 11–1—11–18. Moreau, E., P. Lopez, P. Bauer, A. M. Tompkins, M. Janiskovà, and F. Chevallier, 2003b: Variational retrieval of temperature and humidity profiles using rain-rates versus microwave brightness temperatures. Quart. J. Roy. Meteor. Soc., 130, 827–852. Rabier, F., H. Jarvinen, E. Klinker, J.-F. Mahfouf, and A. Simmons, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. Part I: Experimental results with simplified physics. Quart. J. Roy. Meteor. Soc., 126, 1143– 1170. Saunders, R. W., M. Matricardi, and P. Brunel, 1999: An improved fast radiative transfer model for assimilation of satellite radiance observations. Quart. J. Roy. Meteor. Soc., 125, 1407–1426. Simmons, A. and A. Hollingsworth, 2000: Some aspects of the improvement in skill of numerical weather prediction. Quart. J. Roy. Meteor. Soc., 128, 647–677. Tompkins, A. and M. Janiskovà, 2004: A cloud scheme for data assimilation: Description and initial tests. Quart. J. Roy. Meteor. Soc., 130, 2495–2517. Tremolet, Y., 2004: Diagnostics of linear and incremental approximations in 4D-Var. ECMWF Technical Memorandum No. 399. Available from ECMWF Shinfield Park, Reading, UK, pp. 16.
36 RAINFALL ASSIMILATION INTO LIMITED AREA MODELS Andrea Buzzi and Silvio Davolio ISAC-CNR, Bologna, Italy
Abstract
A physical assimilation technique based on humidity nudging has been developed for application to satellite-derived rainfall fields, in the framework of the European project “EURAINSAT”. The aim of the forcing procedure is to improve the short-range precipitation forecasts with particular attention to specific meteorological phenomena, such as heavy orographic precipitation and small-scale “hurricane-like” cyclones in the Mediterranean area. The nudging scheme forces the model humidity profile in order to get model precipitation closer to the observed precipitation. The forcing is a function of the difference between the rain rates, observed and forecasted, and of precipitation type, convective, or stratiform. In addition, a modelling tool to reproduce the idealised development of midlatitude baroclinic unstable modes, including humidity in the atomsphere and a full water cycle, has been developed with the purpose of investigating the effects and capabilities of assimilation of precipitation in an idealised frame. More realistic experiments have been also performed by implementing a lagged forecast procedure, in order to evaluate, with an observing system simulation experiment (OSSE)-type strategy, the scheme’s performance in terms of improvements of short-range precipitation forecasts and impact on the dynamics of the meteorological evolution. Finally, satellite rain estimates, based on combined microwave (MW) and infrared (IR) techniques, have been assimilated into the limited area meteorological model trying to improve the short-range precipitation forecasts.
Keywords
Rainfall data assimilation, nudging, cyclogenesis, satellite estimates
1 INTRODUCTION Accurate quantitative forecasting of precipitation, especially during severe weather episodes, is one of the most challenging tasks of meteorological modelling. Data assimilation techniques are devoted to attain this aim. In 459 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 459–470. © 2007 Springer.
460
A. Buzzi and S. Davolio
particular, the frequent assimilation of variables directly related to the formation of precipitation and the water cycle may contribute to a better definition of model latent heating, vertical velocity, and moisture and consequently may lead to an improvement of the short-range precipitation forecasts (Manobianco et al. 1994). Recently, the problem of assimilating precipitation data from different sources (satellite, radar, rain gauge, etc.) into limited area meteorological models has received increasing attention, not only in the tropical area, but also in midlatitudes. Since precipitation is not a prognostic model variable, but an end result of complex dynamical and microphysical processes, it cannot be directly assimilated into numerical weather prediction (NWP) models. However, rain observations can be used to correct humidity and temperature profiles, and consequently latent heat release, in order to obtain simulated precipitation closer to the reality. Even if somehow empirical, the nudging is a quite simple, physically based method. In spite of its simplicity, it proved to be suitable for rainfall assimilation for synoptic-scale and mesoscale numerical forecasting, also in an operational framework (Falkovich et al. 2000; Macpherson 2001). The aim of this study is to investigate the effects and capabilities of the assimilation of precipitation both in an idealised and realistic frame. Therefore, the nudging technique has been implemented in a periodic channel version of the meteorological model BOLAM (Buzzi et al. 2003), which reproduces the idealised development of a midlatitude baroclinic unstable mode at finite amplitude. In particular, the application of rainfall assimilation to a concepttual model of cyclogenesis is useful to explain real assimilation results and persistence properties and provides a theoretical basis to the interpretation of the impact of assimilation on the dynamics of midlatitude-growing cyclones. Then the nudging scheme has been applied to the limited area version of the same model, used to analyse a couple of severe weather episodes.
2 THE NUDGING SCHEME The precipitation assimilation scheme has been developed on the basis of a procedure proposed by Falkovich et al. (2000) that modifies the specific humidity profile of the model according to the difference between observed and forecast rain rate. Moisture changes lead to changes of temperature and other dynamical variables through the model precipitation scheme (explicit and convective parameterisation). The procedure starts with a comparison between the forecast (Rm) and target (Rt) total precipitation accumulated over a suitable period of time. After rain-rate comparison, moisture profiles are nudged at grid points where the two values differ, according to the following equation:
Rainfall Assimilation Into Limited Area Models
461
∂q ( k ) = −ν S ,C ( k )τ −1 q ( k ) − ε S ,C q * ( k ) ∂t
{
}
(1)
where k is the model σ-level, q(k) is the specific humidity profile prior to the nudging, q*(k) is the saturation humidity profile (obtained from the model), τ is a relaxation time, εs,c is an over/undersaturation coefficient and νs,c(k) is a vertical modulation profile, whose value may vary in the interval [0–1]. A mean constant rain rate is assumed for the target precipitation within the accumulation interval. Once the model precipitation is available, that is at every convective adjustment time step (about 20 min), rain rates are computed (for both observed and forecast precipitation) using the rainfall accumulated up to the current time step, when the comparison takes place. Therefore, the scheme does not instantaneously adjust the rain rate at each time step, but rather compares and adjusts the rain accumulated until the current time step, seeking to recover the observed precipitation at the end of the accumulation interval. Convective precipitation and stratiform precipitation are handled differently. Model-generated precipitation is used in order to discriminate the precipitation type at a specific grid point. Different vertical modulation profiles νs,c(k) and different coefficients εs,c are used in the two cases in order to introduce/remove humidity only where it is needed. For stratiform precipitation νs(k) is defined in such a way that the humidity is changed only in the middle–lower troposphere, where it is assumed that most of the large-scale condensation takes place. Conversely, in case of pure convective precipitation νc(k) is such that humidity is changed mainly in the boundary layer, which is assumed to represent the source of humidity for the convective adjustment. In case of stratiform precipitation, if the model underestimates the rainfall with respect to the observed value (Rm < Rt), then q(k) is forced towards a slightly supersaturated profile ε S+ q*(k). If the model overestimates the precipitation, then q(k) is decreased gradually towards a subsaturated value ε S− q*(k). In case of convective precipitation, if Rm < Rt, then q(k) is forced
gradually towards a slightly undersaturated profile ε C+ q*(k). If Rm > Rt, then q(k) is decreased gradually towards a low relative humidity value ε C− q*(k). In case of coexistence of both types of precipitation, the sum of the modulating profiles does not exceed unity. At locations where rainfall is observed but not forecasted, both types of precipitation are provisionally considered, unless all the surrounding grid points are experiencing only one type of precipitation. If this is the case, only the appropriate modification, for large scale or convective rainfall, is applied.
462
A. Buzzi and S. Davolio
As for the computed convective (and all physical) tendencies, the nudging adjustment is distributed over all the time steps in the interval between two times at which rain rates are compared.
3 APPLICATION TO A CONCEPTUAL MODEL The basic conceptual models of mid- and high-latitude cyclogenesis were proposed by Petterssen and Smebye (1971), with a twofold classification (types A and B). Type A is considered to be closer to the classical theoretical Eady and Charney models of normal mode growth on a westerly baroclinic flow, while type B has a more local character, being related to the interaction between an upper level potential vorticity (PV) anomaly with a low-level thermal (or PV) anomaly. This basic classification has been supported also in recent observational studies (e.g., Deveson et al. 2002). In the last decade, however, the fundamental dynamical role played by humid processes in the development of cyclones has been emphasised in theoretical, numerical, and observational studies. The release of latent heat in general increases the growth rate, or even destabilises otherwise stable modes, at least for relatively uniform distributions and small amplitude perturbations of the mean westerly flow. In cases of finite amplitude and/or localised disturbances, however, the role of humidity is more complex and can be partly interpreted in terms of generation of low and midtropospheric-positive PV anomalies, interacting with the upper level PV anomalies (see, e.g., Fantini 2004). In this respect, a new type (C) of cyclogenesis has been identified in the literature, in which the latent heat exchange plays a major role and makes the cyclone characteristics very different from the classical conceptual models of Petterssen and Smebye (Plant et al. 2003). In view of the importance of the diabatic humid effects on cyclones and of the role played by precipitation assimilation techniques in altering the latent heat release in a growing cyclone, the effects of such assimilation on the growth properties (growth rate, phase speed) and structure of simple unstable modes have been investigated. In the present idealised framework, the impact of the application of the assimilation in terms of persistence of the modifications introduced can be evaluated more easily and more generally than in individual case studies. The simulations have been performed using a channel version of the BOLAM model. BOLAM is a primitive equation, hydrostatic, limited area meteorological model, whose description can be found in Buzzi et al. (2003). The precipitation assimilation system has been adapted to build up a conceptual model of baroclinic growth modified by the assimilation. The model geometry represents a midlatitude channel, comprised between 18 and 68 degrees N and extending in longitude with a period of 44 degrees. The adopted horizontal grid resolution is of about 50 km. An idealised initial state representing an unstable baroclinic zonal jet is defined, with a
Rainfall Assimilation Into Limited Area Models
463
meridional mean temperature difference of 28 K (applied also to the sea surface temperature), a maximum wind speed of about 30 m s–1 above the simulated tropopause, and a relative humidity decreasing from about 90% near the north wall to about 60% near the south wall (relative humidity is not constant to prevent the development of convective instability near the southern boundary). A barotropic wind perturbation in geostrophic equilibrium is prescribed in the initial condition, sinusoidal in longitude, and vanishing near the lateral walls. A 120-h simulation is considered as reference. Starting from the same initial condition, additional integrations have been performed, including a 24-h period of assimilation, starting after 36 h of unconstrained forecast. Therefore, the control and nudging simulations are identical during the first 36 h. Then, the rainfall assimilation modifies the nudging runs. Three data assimilation experiments have been performed. Two experiments are considered first in order to assess the effect of precipitation amount, one in which the assimilated precipitation (target) is null, the other in which the amount is doubled with respect to the reference experiment, but keeping the same spatial distribution. The target data are accumulated over 3-hourly intervals. Another experiment has been conducted to explore the response of the growing mode to the introduction of longitudinal phase changes by the precipitation assimilation procedure. The target precipitation fields are constituted by the 3-hourly accumulated fields obtained in the reference experiments, but shifted in longitude. The purpose is to investigate the possibility of altering (reducing) phase errors in model fields by means of the assimilation of precipitation. The phase shift is prescribed of the order of a few hundreds of kilometres, corresponding to a few degrees, in longitude. After a few days of integration, an unstable baroclinic wave develops into a large amplitude cyclone and anticyclone couplet, with fully developed PV anomalies, frontal structures, and precipitation areas. The mean sea level pressure field after 84 h is depicted in Fig. 1 (left panel). This is considered as a reference case with respect to cases in which rainfall assimilation is applied. The rainfall assimilation has a strong impact on precipitation fields for the experiments with modified target intensity. At the end of the nudging period, the rainfall has been completely suppressed when the target precipitation is null, while has been remarkably increased in the case of doubled target precipitation. Moreover, it appears immediately that the strength of the system has been altered substantially (Fig. 1). For example, doubling the precipitation target has the effect of deepening the cyclone by about 6 hPa, while the efficient suppression of precipitation, and hence of the associated latent heat release, during the assimilation period has increased the surface pressure in the cyclone minimum by about 10 hPa. The two experiments mentioned above indicate that the assimilation has a strong and long-lasting effect on this kind of development, by altering the growth
464
A. Buzzi and S. Davolio
rate of the system for a significant period of time, either increasing or decreasing the important contribution of latent heat release on the cyclone life cycle.
Figure 1. Mean sea level pressure evolution at day 3.5 (84 h) of an unstable model in the zonal channel. Left panel: reference experiment. Centre panel: assimilation to double rainfall rate than in the reference. Right panel: assimilation to null precipitation.
Figure 2. 3-h accumulated precipitation at the end of the assimilation period, after 60 h of integration. Left: reference experiment. Centre: target precipitation (as the reference run, but shifted in longitude). Right: assimilation run. Contour interval is 2 mm.
If, instead of altering the amount of target precipitation, the same pattern is maintained but shifted in longitude, the forcing during the assimilation phase seems again able to modify the forecast precipitation field (Fig. 2), even if the rainfall intensity is slightly weaker. A comparison between the tempe-
Rainfall Assimilation Into Limited Area Models
465
rature fields at 850 hPa, of the reference and nudged case, in the region of the warm intrusion associated with the precipitation area, at 72 h (i.e., 12 h after the end of the assimilation period) emphasises the persistence of the modified temperature structure and the intensification of the warm front. This reflects the local alteration of the dynamical structure of the mode introduced by the nudging. However, at variance with the case in which precipitation amount is altered substantially, in this case the changes due to the assimilation are reabsorbed in the subsequent further growth of the system, without introducing an appreciable durable change in the spatial phase of the global structure. This can be expected in cases in which the growing mode has a coherent structure, while the perturbation introduced by the transient nudging is likely to project on decaying components, or on components growing more slowly than the dominant mode. In the case in which the amount of precipitation, and hence of latent heat release, is changed substantially, the entire growing structure is altered.
Figure 3. Evolution of the equitable threat score for thresholds of 2 mm/6 h (left panel) and 10 mm/6h (right panel), for the forecast F (dashed) and the nudging run N (solid), computed for 6-h accumulated precipitation. Number in brackets indicates number of observations (control run) exceeding the threshold value. The vertical dashed line indicates the end of the nudging period.
4 APPLICATION TO A NUMERICAL METEOROLOGICAL MODEL The nudging procedure has been tuned and extensively tested in an idealised framework but using realistic meteorological fields (OSSE-type strategy), by implementing a lagged forecast scheme. Two severe weather events, both characterised by heavy rainfall, have been selected for this purpose. The first occurred in the region south of the Alps in September 1999 and was
466
A. Buzzi and S. Davolio
extensively observed during the Mesoscale Alpine Programme (MAP) fieldphase. It was associated with the passage of a frontal system and with heavy orographic rainfall. The second, which caused a major flood in the city of Algiers in November 2001, was characterised by the development of a very intense Mediterranean cyclone. Only the experiment results concerning the second event are presented here. This case study represents a suitable opportunity for evaluating the assimilation procedure, in terms not only of improvements in precipitation forecasts, but also of impact in short-range forecasting of a cyclogenesis event. Since the latent heat release and surface heat fluxes played a crucial role in the development of the Mediterranean cyclone, it is expected that a nudging procedure that modifies the humidity profiles had an appreciable impact on cyclogenesis. The nudging procedure has been applied to the BOLAM model and the lagged forecast scheme has been implemented, performing three simulations as follows. The first simulation (C) consisted of a 36-h run, initialised 10 November, 1200 UTC. The second (F) was initialised 12 h before, at 0000 UTC of the same day and lasted 48 h. C represents the reference state and provides the precipitation target data, while F is considered the “real” forecast to be improved (compared to C). Finally, a third 48-h simulation (N) was performed, starting from the same initial condition as F, but applying the nudging procedure for 12 h, from 10 November, 1200 UTC. Two-hourly model rainfall data extracted from C are used for the assimilation, with the aim of forcing the forecast towards C. The assimilation considerably improves the precipitation forecasts at the end of the nudging phase. The improvements of the rainfall assimilation extend well beyond the forcing period, as shown by the evolution of the equitable threat score (ETS) in Fig. 3. For both low and high threshold values of rainfall, the benefit of the nudging seems to last at least 18 h during the unconstrained forecast, following the forcing. Cross sections in correspondence to the rain bands show that mesoscale vertical motion has been generated by the nudging procedure and improvements in the potential temperature and relative humidity vertical distribution have been achieved. The evolution of the Mediterranean cyclone as described by the basic forecast (F) remarkably differs from the control (C): in place of a sharp low (985 hPa at 1200 UTC, 11 November) surrounded by an area of more levelled pressure, centred south-east of the Balearic Islands, F produces a weaker large-scale pressure system without an intense core, whose centre appears to be displaced northward. After the rainfall assimilation and for almost the entire following unconstrained forecast period, the evolution of the low is improved both in intensity, timing and trajectory of the pressure minimum (Fig. 4).
Rainfall Assimilation Into Limited Area Models
467
Figure 4. Surface low centre trajectory (left panel) and central MSLP (hPa) (right panel) for the control run (left: right line from the bottom; right: upper line from the left), forecast (left: left line from the bottom; right: lower line from the left)) and nudging run (left: middle line from the bottom; right: middle line from the left) computed at 2-h intervals, from 1200 UTC, 10 November to 0000 UTC, 12 November 2001.
Figure 5. 12-h accumulated precipitation at 1200 UTC, 10 November 2001 as estimated from satellite (left panel) and for the reference forecast (R, right panel) (see also color plate 15).
The nudging scheme has been also applied to the Algerian flood event in a more realistic framework, attempting to assimilate satellite precipitation estimates (combined IR-MW, Kidd et al. 2003). Rainfall estimates, available every 30 min, have been interpolated to the limited area model grid, accumulated over 2-hourly intervals and then used as target for the assimilation. In order to avoid using data possibly affected by large errors, the assimilation was performed only over the sea. Two 48-h simulations have been initialised at 0000 UTC, 10 November 2001. The first represents the reference forecast (R), the second (N) is forced by the nudging scheme during the first 12-h period, using 2-hourly satellite data.
468
A. Buzzi and S. Davolio
The 12-h accumulated precipitation field, as forecasted by the model, presents differences with respect to the satellite precipitation estimate, especially around Sardinia (Fig. 5). In particular, the model misses the rainfall area west of the island, while produces an intense rain band over the Tyrrhenian Sea. The latter rain band is present also in the observations, but weaker, displaced eastward and affecting a smaller area. The assimilation seems to reduce the forecast error in terms of both intensity and location of the rainfall patterns (Fig. 6). The precipitation nucleus west of Sardinia is correctly generated in the form of a rain band, whose intensity is however lower than the target. Over the Tyrrhenian Sea the excess of rainfall is reduced and the patterns is more similar to the observations. Similar experiments have been performed using the same rainfall data but accumulated over different intervals. Assimilating hourly data gives results that are almost the same as for 2-hourly data. Only a slight improvement is observed concerning the rainfall pattern over the Tyrrhenian Sea, but not elsewhere. A clear degradation of the assimilation impact emerges when rainfall estimates accumulated over longer interval (6 h) are used. In this case, both the reduction and the increase of the predicted rainfall becomes less evident, although still present, and the forecast is not efficiently modified by the assimilation scheme, confirming that target data are too smooth in time and space for this type of application.
Figure 6. 12-h accumulated precipitation at 1200 UTC, 10 November 2001, for the assimilation run (N) (see also color plate 15).
5 CONCLUSIONS An original technique of assimilation of precipitation into numerical meteorological models, separating stratiform and convective precipitation, has been successfully applied to the meteorological model BOLAM. The capability of the rainfall assimilation in altering the intensity of an idealised model of baroclinic cyclone development in midlatitudes has been shown
Rainfall Assimilation Into Limited Area Models
469
using a periodic channel version of the model. The experiments described above support the assumption that the water cycle in a baroclinic atmosphere is particularly effective in determining the growth and the structure of the modes. As a consequence, the assimilation of precipitation can induce a strong modification of the structure and properties of the unstable modes, affecting in a more or less permanent way the system trajectory in phase space, as a function of the projection of the perturbations introduced on the growing mode spectrum. In this sense, the results based on the conceptual numerical model can be used to interpret the results obtained in the realistic framework. The assimilation procedure has been applied to the real case studies of heavy precipitation, using both OSSE technique to generate model data and rainfall satellite estimates. In these cases, the impact of the assimilation on short-range precipitation forecasts and on cyclone growth appears to be quite important. However, as also found in the conceptual model application, the technique is less effective in introducing spatial phase correction into the meteorological evolution, so that in this case the benefit of the rainfall assimilation has a rather short duration in time. The modification of the humidity profiles, and consequently of the latent heat release through the model precipitation scheme, produces improvements in the short-range precipitation forecasts. The positive impact of the assimilation is confined, however, to a limited period of time of the unconstrained forecast, following the assimilation phase. The impact is retained for about 12–24 h, depending on the particular meteorological situation. However, it is important to point out that better rainfall forecasts are associated to a better reproduction of the vertical motion, latent heating, and vertical profile in the rainy areas. Acknowledgement: This research was funded by the EURAINSAT project, a shared-cost project (contract EVG1-2000-00030), co-funded by the Research DG of the European Commission within the RTD activities of a generic nature of the Environment and Sustainable Development Subprogramme, 5th Framework Programme).
6 REFERENCES Buzzi, A., M. D’Isidoro, and S. Davolio, 2003: A case study of an orographic cyclone formation south of the Alps during the MAP-SOP. Quart. J. Roy. Meteor. Soc., 129, 1795–1818. Deveson, A. C. L., K. A. Browning, and T. D. Hewson, 2002: A classification of FASTEX cyclones using a height-attributable quasi-geostrophic vertical-motion diagnostics. Quart. J. Roy. Meteor. Soc., 128, 93–118. Falkovich, A., E. Kalnay, S. Lord, and M. M. Mathur, 2000: A new method of observed rainfall assimilation in forecast model. J. Appl. Meteor., 39, 1282–1298.
470
A. Buzzi and S. Davolio
Fantini, M., 2004: Baroclinic instability of a zero-PVE jet: enhanced effects of moisture on the lifecycle of baroclinic cyclones. J. Atmos. Sci., 61, 1296–1307. Kidd, C., D. Kniveton, M. Todd, and T. Bellerby, 2003: Satellite rainfall estimation using a combined passive microwave and infrared algorithm. J. Hydrometeorology, 4, 1088– 1104. Macpherson, B., 2001: Operational experience with assimilation of rainfall data in the Met Office mesoscale model. Meteor. Atmos. Phys., 76, 3–8. Manobianco, J., S. Koch, V. M. Karyampudi, and A. J. Negri, 1994: The impact of assimilating satellite-derived precipitation rates on numerical simulations of the ERICA IOP 4 cyclone. Mon. Wea. Rev., 122, 341–365. Plant, R. S., G. C. Craig, and S. L. Gray, 2003: On a threefold classification of extratropical cyclogenesis. Quart. J. Roy. Meteor. Soc., 129, 2989–3012.
37 IMPLEMENTING AN OPERATIONAL CHAIN: THE FLORENCE LaMMA LABORATORY Alberto Ortolani1,3, Andrea Antonini2,3, Graziano Giuliani2,3, Samantha Melani2,3, Francesco Meneguzzo1,3, Gianni Messeri1,3, Andrea Orlandi1,2,3, and Massimiliano Pasqui1,3 1
Institute of BioMeteorology, National Council of Research (IBIMET-CNR), Florence, Italy Foundation for Applied Meteorology (FMA), Florence, Italy 3 Laboratory for Meteorology and Environmental Modelling (LaMMA), Florence, Italy 2
1 INTRODUCTION The Laboratory for Meteorology and Environmental Modelling (LaMMA, http://www.lamma.rete.toscana.it) was set up in 1997 due to an initiative of Regione Toscana (Tuscany Region Administration), which entrusted the Foundation for Applied Meteorology (FMA) to manage the laboratory in cooperation with the National Council for Research (CNR) and some Tuscany companies belonging to Finmeccanica Group, working in the field of space technology design and exploitation. From July 2002, LaMMA is managed by the Institute of BioMeteorology (IBIMET) of CNR. The main tasks of LaMMA are research, technological transfer, and service provision in support of the regional and national operational organizations, in the fields of meteorology and environmental monitoring. The research skills span from atmospheric and ocean modelling, air quality, and remote sensing of environmental parameters to geographic information systems for environmental management. The main research results are thus continuously integrated in the operational services of LaMMA, that acts as a regional meteorological service in Tuscany, with daily public forecasts for different TV and radio programs, as well as forecasts and nowcasts for supporting (even 24 h) Civil Protection decisions during severe weather events (storms, flash floods, snow, etc.). The Regional Atmospheric Modelling System (RAMS) (Pielke et al. 1992; MRC/*Aster 2000) is the atmospheric model used
471 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 471–482. © 2007 Springer.
472
A. Ortolani et al.
operationally at LaMMA from 1999 (Pasqui et al. 2000; Meneguzzo et al. 2001; Meneguzzo et al. 2004; Pasqui et al. 2002; Soderman et al. 2003). Hydrogeological disasters are among the sources of main concern (in terms of expense and health risk) in Tuscany, but generally in Italy and in several European and world countries. If their link to rain is widely apparent, precipitation itself (together with the hydrological mechanisms) remains largely a matter of study. The way rainfall happens, space-time distribution and intensity predictability, as well as methods for reliable homogeneous rainfall measurements are not completely understood and much efforts are needed for a satisfactory comprehension and prediction of these phenomena. In the framework of EURAINSAT a large part of these issues were addressed. The approach followed by LaMMA in the project activities was to focus on research with high and soon operational applicability, trying to maximize the potential benefits that the availability of quasi real-time rainfall fields (homogeneously produced for the whole Euro-Mediterranean area) could bring to the nowcasting and forecasting activities. For this purpose: • •
•
An automatic chain for real-time satellite rainfall estimation has been set up and a validation phase over the Tuscany area has been started and is still ongoing. In order to improve quantitative precipitation forecasts (QPFs) of RAMS, a method for diabatic assimilation of the convective part of rainfall fields in the first hours of the forecast simulation has been developed and tested, both with simulated and observed data. A method for soil moisture initialization in RAMS by means of antecedent observed precipitation has been designed and implemented with the aim of improving the description of the initial state of a simulation run.
All these methods have been designed in order to be integrable in a unique nowcasting/forecasting system, and in a way compatible with the operational constrains of the routine LaMMA activities. The following sections detail the work done and the main results.
2 THE OPERATIONAL CHAIN FOR RAINFALL FIELD ESTIMATION Precipitation estimates from satellites are a relevant component in environmental monitoring, from flash flood forecasting and landslips to assimilation in numerical weather prediction (NWP) models (Levizzani et al. 2002). In this frame, the LaMMA laboratory for either regional monitoring or hazardous prevention purposes has implemented an operational chain that produces real-time instantaneous half-hourly rainfall maps.
Implementing an Operational Chain
473
The procedure is based on a blended technique (Turk et al. 2000a, b) that dynamically correlates brightness temperatures as measured by geostationary sensors and instantaneous rain rates, as computed by microwave (MW) passive radiometer data (Ferraro and Marks 1995; Ferrraro 1997), by means of a statistical correlation (Crosson et al. 1996). The quantification of precipitation levels as implemented in the operational chain involves an automatic process that constantly colocates, in space and time, newly arriving MW and infrared (IR) geostationary data. In this frame, two automatic, independent processes for METEOSAT and SSM/I data have been developed. The procedures are scheduled to start running independently every 30 min, as shown in Fig. 1.
Figure 1. Operational chain flow chart.
The overall system has been implemented by means of a Workstation equipped with UNIX operating system and TeraScan Software (SeaSpace Corporation, Poway, CA) that allows the processing and visualization of satellite data. The Workstation performs the ingestion of both IR METEOSAT and MW SSM/I channels data. A PDUS acquires METEOSAT real-time high resolution (HR) data (every half hour). The data, supplied in digital format, also contain geolocation, rectification, and calibration parameters, as well as the temperatures for the various channels. The MW data, supplied in temperature data record (TDR) format, are downloaded by ftp connection from the SAA archive (http://www.saa.noaa.gov).
474
A. Ortolani et al.
The operational service provided by the laboratory consists of instanttaneous rain field estimation and temporal evolution (i.e., the animation of the last 6 h) of the precipitation events for the Euro-Mediterranean area, both updated every 30 min. The results can be viewed online through the LaMMA web site at http://www.lamma.rete.toscana.it/previ/ita/rainmeteosat.html. Imagery is geolocated trough a regular lat/long projection. Figure 2 shows an example for 15 November 2002, at 03:00 UTC, where an intense rainfall event occurred over the Ligurian Gulf.
Figure 2. Example of instantaneous rainfall map for 15 November 2002.
A validation phase with ground-based rain gauges measurements has been undertaken over the Tuscany area. The preliminary results show that the algorithm correctly (in terms of space-time phase) associates convective cloudiness to rainfall, but misinterprets non-raining stratiform cloudiness as light rain. Further developments will also include on the web site instantaneous rainfall maps and cumulated rainfall images over the last 1, 3, 6, 12, and 24 h for other geographical areas. Finally, the upgrade of the operational chain for the ingestion of the new MSG satellite data and other MW sources is under development.
Implementing an Operational Chain
475
3 DIABATIC PRECIPITATION ASSIMILATION IN RAMS The assimilation of high-resolution satellite rainfall estimates may be really effective in order to improve QPF. In particular, it can be exploited to minimize the spin-up problem of atmospheric models, which is related to the difficulty in providing appropriate divergence and moisture fields in the initial conditions. Such a difficulty causes inconsistencies in the representation of the latent heat release evolution, mainly in the first hours of model simulations. As a consequence the timing and location of precipitation events are often inaccurate, especially when the model spin-up phase clashes with the initial stages of precipitating phenomena. Several works have been devoted to the investigation of this problem (Carr and Baldwin 1991; Davidson and Puri 1992) and some other (Pereira Fo et al. 1999; Falkovitch et al. 2000) have pointed out how observed rainfall assimilation can potentially address this problem. In the last 15 years, various approaches for precipitation assimilation have been tried for global and large-scale models, as well as for mesoscale models. For a short review of publications on this topics see, for example, Orlandi et al. (2004). In mesoscale and limited area models the most promising results have been obtained by incrementing temperature and moisture throughout a pre-forecast period (latent heat nudging [LHN]), in order to approach precipitation observations (Jones and Macpherson 1997). It has been proven (Manobianco et al. 1994; Orlandi et al. 2004) that NWP models can retain useful information from precipitation data well beyond the assimilation period (sometimes up to 30 h). Sensitivity studies demonstrated that the specification of the space-time location of the rainfall event is more relevant than the information on precipitation intensity (Manobianco et al. 1994). Many implementations of the LHN are based on the inversion of the convective parameterization scheme. Several cumulus parameterization schemes have been developed, for large-scale and for mesoscale models, with various levels of complexity. Generally they are based on two main features: (1) resolvable-scale quantities are used to establish constraints on the amount of convection, and (2) a cloud model is used to estimate the vertical structure of the convective mass flux that satisfies the constraints. The outputs of the cloud model are re-ingested in the resolved dynamics of the model, as a feedback of the effects of parameterized cumulus convection. This feedback can be exploited for assimilating observed rainfall patterns. In particular the cloud model allows to redistribute along the column the latent heating and moisture derived from observed rainfall. The assimilation technique described here (Meneguzzo et al. 2002; Orlandi et al. 2004) is based on the inversion of the Kuo scheme (Kuo 1974;
A. Ortolani et al.
476
Molinari 1985; Molinari and Corsetti 1985) which is implemented in RAMS. The large-scale constraint of this scheme is the moisture convergence I at the base of the convective air column, which is computed from the modelresolved variables. It is split into two parts, by introducing the Kuo phenomenological parameter b, which embodies the microphysics of precipitation production. The fraction (1–b) of I is condensed and precipitated down. The convective precipitation rate is thus computed such as:
Pconv = (1 − b) I
(1)
The remaining fraction b of I is stored in the cloud, and acts to increase the moisture of the convective column. The feedback of cumulus convection is computed in terms of convective tendencies along the column for potential temperature and water vapour-mixing ratio:
L(1 − b) I ⎛ ∂θ ⎞ ⎜ ⎟ = Π ⎝ ∂t ⎠conv
Q1 ( z ) zct
∫ Q ( z ')dz ' 1
zg
,
⎛ ∂r ⎞ ⎜ ⎟ = bI ⎝ ∂t ⎠conv
Q2 ( z ) zct
(2)
∫ Q ( z ')dz ' 2
zg
where L is the latent heat of condensation for water, Π is the total Exner function, and b is computed, in RAMS, as a function of environmental wind shear (Fritsch and Chappell 1980). The vertical profiles of heating Q1(z) and moistening Q2(z) are computed through a cloud model (Molinari 1985; Tremback 1990), starting from the resolved variables. The convective tendencies are then inserted in the model equations, which are forward integrated in time at each time step. The scheme is activated with a fixed time cadence (default in RAMS is 20 min), to re-compute the convective rainfall and tendencies, only on those grid points whose air column results to be convectively unstable and with a sufficient moisture convergence. The inversion of the Kuo scheme is straightforward. The moisture convergence I is computed from the observed rainfall, by inverting Eq. (1). Convective tendencies are then computed by Eq. (2) and then inserted in the model equations. This realizes a feedback mechanism which allows to modify the atmospheric conditions consistently with the observed rainfall.
Implementing an Operational Chain
Cumulated rainfall [mm]
Cumulated rainfall [mm]
477
(
(
a)
Number of hours from 0:0 a.m. 1999/09/20
b)
Number of hours from 0:0 a.m. 1999/09/20
Figure 3. Comparison between rain gauges observations and results of assimilation and control runs. Averaged data over two hydrologic basins in Tuscany region: (a) Serchio and (b) middle Arno Valley.
The inverted Kuo scheme is activated, when there is a rainfall map available, on convectively unstable grid points with enough rainfall. The “observed convective rainfall”, Pconv, is evaluated (convective/stratiform partition) somehow roughly by subtracting the model-computed resolved rainfall from the observed rainfall. In the operational implementation the inverted scheme is activated during the first hours of model simulation (typically six), assimilating a satellite rainfall map each half an hour. Observed rainfall is weighted with respect to the model computed direct Kuo rainfall, by a gradually growing nudging function, so as to “gently” push the model towards the observed conditions. The assimilation is performed only in coarse grids, where convection is parameterized, but its effects are transmitted to high-resolution grids, where convection is explicit, through the nesting mechanism (Warner and Hsu 2000; Orlandi et al. 2001). Test and tuning experiments have been performed by assimilating synthetic convective rainfall produced by the model itself. These experiments allowed to tune some of the free parameters of the procedure, and demonstrated the permanence of positive effects of the assimilation up to 30 h. Comparisons with rain gauges measurements, when assimilating real satellite rainfall, confirm such relevant improvements. The pluviometric plot in Fig. 3 shows how the initial phase of the rainfall event is better described by the assimilating model run.
478
A. Ortolani et al.
4 INITIALIZATION OF RAMS WITH SOIL MOISTURE FROM ANTECEDENT PRECIPITATION FIELDS The soil-state initialization plays a major role in numerical weather modelling on a wide range of spatial temporal scales affecting forecast skills especially on surface atmospheric fields (Chen and Avissar 1994; Avissar and Schmidt 1998; Golaz et al. 2001; Pielke 2001; Meneguzzo et al. 2002). Furthermore, NWP models which include detailed schemes for the description of soil– vegetation–atmosphere interaction exhibit strong sensitivity to soil-state initialization. As a consequence a reasonable description of the initial state is crucial to improve forecasts reliability. Dealing with case studies, the general approach is to establish a reasonable choice of the initial soil state based on available data-sets, retrieved from satellite, weather stations or specific soil state bulletins. Another strategy is to derive the initial soil state from general circulation models, but such fields are generally not directly related to real observed precipitation fields and they have a coarse spatial resolution. The soil-state initialization is very important for the correct forecasting of a wide range of weather phenomena, but the availability of such information as first guess field is by no means trivial. The fifth version of RAMS provides a method to produce initial soil state computed from simulated atmospheric and observed precipitation fields. In other words it is possible to run the Land Ecosystem Atmosphere Feedback (LEAF) model (Walko et al. 2000) prescribing both the atmosphere state and the occurred rainfall, the latter being possibly the results of observations. The state of the atmosphere is provided by a previous atmospheric RAMS simulation. Clearly the real atmospheric state which produces the observed (and ingested) precipitation could exhibit important discrepancies with respect to the simulated atmospheric state (used as a forcing), leading to significant differences in the water exchange among soil, vegetation, and air. On the other hand, such soil first-guess field brings some benefits, firstly a more realistic evaluation of the water amount with respect to what is just simulated. Furthermore, it provides a better description of heterogeneity due to the hydrological model acting within LEAF, as well as a longer, so more accurate, reconstruction of the water cycle forcing with respect to a simple (even if somehow measured) initial estimation of the soil state. The soil initialization scheme is based on a modified version of RAMS named RAMS antecedent precipitation index (RAPI) (Pasqui et al. 2004). The RAPI model needs two different types of input: 1.
The precipitation field, as a distributed map of rainfall (from satellite estimates, radar, etc.) over the area of interest for the selected time period.
Implementing an Operational Chain
2.
479
The RAMS atmospheric fields, computed in a separate RAMS run on the same time period.
Figure 4. RAPI assimilation scheme.
Using such information, the RAPI model computes energy balance, as RAMS actually does, between atmosphere, prescribed from a previous RAMS run, and the provided rainfall fields. Using observed precipitation has the advantage of improving the computation of the water budget, at the soil level, both for heterogeneity and reliability. Several benefits of using RAPI are worth to be highlighted, both from a physical and an operational point of view. The observed precipitation, once projected on the target area, has the same topography and resolution as the model grids, so a basins budget could be more reliable than soil moisture interpolation coming from a coarser grid simulation (e.g., a GCM field). A pilot version of this assimilation techniques (Fig. 4) was set up at LaMMA using a three-nested grid RAMS configuration at 32, 8, 2 km horizontal resolution and 36 vertical levels, with a resolution ranging from 50 to 1100 m and 11 ground level down to –1.5 m with a stretched resolution. This 24-h daily simulation provides the atmospheric forcing with initial and boundary conditions every 6 h from NCEP/NCAR analysis fields. The hourly precipitation data-set, based on the satellite estimation algorithm described above, is collected over Europe and North Africa for the same 24-h period. Such data are stored on the RAMS standard lat/long format (same as SST, vegetation cover, soil textural classes) and represent the observed precipitation forcing data-set. Every day these two data-sets are ingested in RAPI to compute the initial soil state for the following day RAMS simulation. Note that simulations performed on DESMO, the LaMMA operational Linux Cluster, show the
480
A. Ortolani et al.
RAPI running about 30 times faster then RAMS. Thus, this approach does not affect critically the computational demand for the entire forecast system. Preliminary tests, performed on the surface temperature field, reveal promising benefits of this approach. In particular, a reduction of the spin-up errors during the first 6 h is found, in addition to a general improvement of the precision on the forecasted minimum temperatures (Pasqui 2004).
5 CONCLUSION The memory of well-known, sometimes very recent, floods and landslides in Tuscany and in other parts of Europe is still alive in the mind of the damaged people. Unforeseen critical events as well as great false alarms survive only in the memory of people in charge of technical responsibilities on the subject, but they are not matter of less concern due to their larger number. The work performed in the EURAINSAT context has demonstrated that also recent research results and technological means can be exploited in an operational context, and that they are able to improve real-time monitoring and prediction of critical precipitation events, in terms of precision and time ahead, opening new possibilities for bad-effect mitigation. We have addressed the problem of monitoring rainfall events by means of real-time precipitation field estimation based on merged operational satellite observations, and we have used such available rainfall fields to cope with the problem of the initial state reconstruction for prognostic models in order to improve forecast capabilities of precipitation as well as other atmospheric quantities. For this purpose the model RAMS was modified both to allow forcing convective profiles to be compliant to rainfall observations for an initial simulation period, and to compute a starting soil moisture field according to previous precipitation. In other words we have designed an integrated nowcasting/forecasting system for rainfall-driven events. The validation work is still in progress, and for the moment it is performed only on single pieces of the whole rainfall nowcasting/forecasting system. System tests and integration are going on, and all relevant results will be continuously published on LaMMA web pages. In any case the tests performed up to now have demonstrated that we are on the right way, but at the same time that a large room for improvements still exists, especially on the side of precipitation estimation, and consequently on the derived applications. New satellite missions that just started or are near to be launched could be definitely part of the solutions for the open problems.
6 REFERENCES Avissar, R. and T. Schmidt, 1998: An evaluation of the scale at which ground–surface heat flux patchiness affects the convective boundary layer using large-eddy simulations. J. Atmos Sci., 55, 2666–2689.
Implementing an Operational Chain
481
Carr, F. H. and M. Baldwin, 1991: Incorporation of observed precipitation estimates during initialisation of synoptic and mesoscale storms. 1st Int. Symp. on Winter Storms, New Orleans, LA, Amer. Meteor. Soc., 71–75. Chen, F. and R. Avissar, 1994: Impact of land–surface moisture variability on local shallow convective cumulus and precipitation in large-scale models. J. Appl. Meteor., 33, 1382– 1401. Crosson, W. L., C. E. Duchon, R. Raghavan, and S. J. Goodman, 1996: Assessment of rainfall estimates using a standard Z–R relationship and the probability matching method applied to composite radar data in central Florida. J. Appl. Meteor., 35, 1203–1219. Davidson, N. E. and K. Puri, 1992: Tropical prediction using dynamical nudging, satellite defined convective heat sources, and a cyclone bogus. Mon. Wea. Rev., 120, 2501–2522. Falkovich, A., E. Kalnay, S. Lord, and M. B. Mathur, 2000: A new method of observed rainfall assimilation in forecast models. J. Appl. Meteor., 39, 1282–1298. Ferraro, R. R., 1997: Special sensor microwave imager derived global rainfall estimates for climatological applications. J. Geophys. Res., 102 (D14), 16715–16735. Ferraro, R. R. and G. F. Marks, 1995: The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J. Atmos. Oceanic Technol., 12, 755– 770. Fritsch, J. M. and C. F. Chappell, 1980: Numerical prediction of convectively driven mesoscale pressure systems. Part I: Convective parameterization. J. Atmos. Sci., 37, 1722–1733. Golaz, J.-C., H. Jiang, and W. R. Cotton, 2001: A large-eddy simulation study of cumulus clouds over land and sensitivity to soil moisture. Atmos. Res., 59–60, 373–392. Kuo, H. L., 1974: Further studies of the influence of cumulus convection on large-scale flow. J. Atmos. Sci., 31, 1232–1240. Jones, C. D. and B. Macpherson, 1997: A latent heat nudging scheme for the assimilation of precipitation data into an operational mesoscale model. Meteorol. Appl., 4, 269–277. Levizzani, V., R. Amorati, and F. Meneguzzo, 2002: A review of satellite-based rainfall estimation methods. European Commission Project MUSIC Report (EVK1-CT-200000058), 66 pp. Manobianco, J., S. Koch, V. M. Karyampudi, and A. J. 1994: The impact of satellite-derived precipitation rates on numerical simulations of the ERICA IOP 4 cyclone. Mon. Wea. Rev., 124, 341–365. Meneguzzo, F., G. Menduni, G. Maracchi, G. Zipoli, B. Gozzini, D. Grifoni, G. Messeri, M. Pasqui, M. Rossi, and C. J. Tremback, 2001: Explicit forecasting of precipitation: sensitivity of model RAMS to surface features, microphysics, convection, resolution. In: Proc. 3rd Plinius Conf. on Mediterranean Storms. R. Deidda, A. Mugnai, F. Siccardi, eds., GNDCI Publ. N.2560, ISBN 88-8080-031-0, 79–84. Meneguzzo, F., V. Levizzani, A. Orlandi, A. Ortolani, M. Pasqui, F. Torricella, and B. Gozzini, 2002: Resolution and data assimilation issues in the operational numerical forecast of basin-scale rain storms. Proc. 4th EGS Plinius Conf. on Mediterranean Storms, October 2002 (a), Mallorca. Meneguzzo, F., M. Pasqui, G. Menduni, G. Messeri, B. Gozzini, D. Grifoni, M. Rossi, and G. Maracchi, 2004: Sensitivity of meteorological high-resolution numerical simulations of the biggest floods occurred over the arno river basin, Italy, in the 20th century. J. Hydrol., 288, 37–56. Molinari, J., 1985: A general form of Kuo’s cumulus parameterization. Mon. Wea. Rev., 113, 1411–1416. Molinari, J. and T. Corsetti 1985: Incorporation of cloud-scale and mesoscale down-drafts into a cumulus parameterisation: Results of one- and three-dimensional integrations. Mon. Wea. Rev., 113, 485–501.
482
A. Ortolani et al.
Orlandi, A., F. Meneguzzo, G. Messeri, A. Ortolani, M. Pasqui, M. Rossi, and A. Terzo 2001: Satellite rainfall assimilation to improve the quantitative precipitation forecasting, Proc. EUMETSAT Conference, Antalia, 453–460. Orlandi, A., A. Ortolani, F. Meneguzzo, V. Levizzani, F. Torricella, and F. J. Turk, 2004: Rainfall assimilation in RAMS by means of the Kuo convective parameterisation inversion: method and preliminary results. J. Hydrol., 288, 20–35. Pasqui, M. et al., 2000: Performances of the operational RAMS in a Mediterranean region as regards to quantitative precipitation forecasts. Sensitivity of precipitation and wind forecasts to the representation of the land cover. Proc. 4th RAMS Users Workshop, Cook College, Rutgers University, 22–24 May 2000, New Jersey. Pasqui, M. et al., 2002: Historical severe floods prediction with model RAMS over central Italy. Proc. 5th RAMS Users Workshop, Santorini, Greece. Pasqui, M., C. J. Tremback, F. Meneguzzo, G. Giuliani, and B. Gozzini, 2004: A soil moisture initialization method, based on antecedent precipitation approach, for regional atmospheric modeling system: a sensitivity study on precipitation and temperature. 18th Conf. on Hydrology, AMS, Seattle. Pereira Fo, A. J., K. C. Crawford, and D. J. Stensrud, 1999: Mesoscale precipitation fields. Part II: Hydrometeorologic modelling, J. Appl. Meteor., 38, 102–125. Pielke, R. A. and Coauthors, 1992: A comprehensive meteorological modelling system – RAMS. Meteor. Atmos. Phys., 49, 69–91. Pielke Sr., R. A., 2001: Influence of the spatial distribution of vegetation and soils on the prediction of cumulus convective rainfall. Rev. Geophys., 39, 151–177. Soderman, D., F. Meneguzzo, B. Gozzini, D. Grifoni, G. Messeri, M. Rossi, S. Montagnani, M. Pasqui, A. Orlandi, A. Ortolani, E. Todini, G. Menduni, and V. Levizzani, 2003: Very high resolution precipitation forecasting on low cost high performance computer systems in support of hydrological modeling. Proc. 17th Conf. on Hydrology, AMS, Long Beach, 9–13 February, CD-ROM, ISBN 1-878220-63-2. Tremback, C. J., 1990: Numerical simulations of mesoscale convective complex: model development and numerical results. Ph.D. dissertation, Dept. of Atmospheric Science, paper 465, Colorado state university, Forth Collins, CO. Turk, F. J., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000a: Meteorological applications of precipitation estimation from combined SSM/I, TRMM and geostationary satellite data. In: Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia, eds., VSP Int. Sci. Publisher, Utrecht, The Netherlands, 353–363. Turk, F. J., J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000b: Combining SSM/I, TRMM and infrared geostationary satellite data in a near-realtime fashion for rapid precipitation updates: advantages and limitations. Proc. 2000 EUMETSAT Meteorological Satellite Data Users’ Conf., Bologna, 29 May–2 June, 452– 459. Walko, R. L., L. E. Band, J. Baron, T. G. F. Kittel, R. Lammers, T. J. Lee, D. Ojima, R. A. Pielke, C. Taylor, C. Tague, C. J. Tremback, and P. L. Vidale, 2000: Coupled atmosphere-biophysics-hydrology models for environmental modeling. J. Appl. Meteor., 39, 931–944. Warner, T. T. and H.-M. Hsu, 2000: Nested-model simulation of moist convection: the impact of coarse-grid parameterised convection on fine-grid resolved convection. Mon. Wea. Rev., 128, 2211–2231.
Section 7 Applications to Monitoring Weather Events
38 SATELLITE PRECIPITATION ALGORITHMS FOR EXTREME PRECIPITATION EVENTS Roderick A. Scofield† and Robert J. Kuligowski National Environmental Satellite, Data, and Information Service Camp Springs, MD, USA
1 INTRODUCTION Floods and flash floods take a heavy toll each year in terms of both lives and property. According to the 2001 Disaster Report by the International Red Cross and Red Crescent Societies, floods accounted for over two-thirds of the 211 million people affected worldwide on average by natural disasters each year during the 1990s, and also for roughly 15% of the nearly 666,000 deaths from natural disasters during this period. Many of these events are triggered by extreme precipitation, often in conjunction with other factors. However, timely and reliable information on past, current, and future precipitation can be very difficult to obtain, especially in those portions of the world where resources are not available to build and support a comprehensive precipitation observing network. The three primary sources of precipitation information are rain gauges, radar, and satellite. Rain gauges have the clear advantage of directly measuring precipitation rather than to deriving it from a remotely sensed quantity. However, even relatively dense rain gauge networks are unable to depict the intensity and spatial extent of heavy precipitation (Smith et al. 1994, 1996). Furthermore, the rain gauge networks used in these studies are much more dense than those in most parts of the world. And although rain gauges are relatively low-cost on a unit basis, the expense required to install a rain gauge network suitable for extreme precipitation events is beyond the available resources for most nations. Radar offers widespread spatial coverage at high spatial and temporal resolution. However, there are difficulties both in obtaining accurate measures †
Louisville, 3 December 1942 – Pomfret, 25 February 2006
485 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 485–495. © United States Government 2007.
486
R. A. Scofield and R. J. Kuligowski
of reflectivity and in converting these reflectivity measurements into an accurate representation of the precipitation field at ground level. The former is the result of various phenomena including anomalous propagation, beam block, and beam overshoot, and is especially problematic in regions of high topographic relief where the range of useful radar data can be severely limited (e.g., Westrick et al. 1999; Young et al. 1999). Corrections have been developed and operationally implemented in an effort to mitigate these effects, such as hybrid scan strategies (Fulton et al. 1998) and vertical profile corrections (Seo et al. 2000) to improve the quality of the reflectivity data, and multiple reflectivity-rain rate (Z-R) relationships and correction of radar precipitation estimates using rain gauges (Fulton et al. 1998; Seo and Breidenbach 2002) to improve the resulting rain rate estimates. However, even beyond these issues lies the high monetary cost of building and maintaining a suitable network of radars. Precipitation estimates from sensors on geostationary satellite platforms offer an excellent complement to existing and planned rain gauge and radar networks. These estimates are available at high spatial resolution (3–5 km) and high temporal resolution (up to 15 min) for the whole globe between 60º N and 60º S latitude, and can compensate for the coverage shortcomings of rain gauge and radar data due to terrain or cost. However, the relationship between satellite-sensed radiances and rainfall rates at ground level is less robust than that between radar reflectivities and rainfall rates; consequently, these estimates should be viewed as a complement to other available sources of precipitation data. This paper describes real-time satellite precipitation algorithms for extreme precipitation events that are being developed at the National Oceanic and Atmospheric Administration (NOAA) National Environmental Satellite, Data, and Information Service (NESDIS). During the last 25 years, operational satellite precipitation estimates (SPE) based on data from Geostationary Operational Environmental Satellites (GOES) have evolved from a combination of manual effort and computer algorithms to a succession of fully automated algorithms. Efforts are now underway to further improve SPE by incorporating information from other sources such as microwave radiometers, and to enhance the impact of SPE by incorporating it into hydrologic and numerical weather prediction models and by producing 1–3 h SPE-based nowcasts. Section 2 describes this progression and the associated algorithms in more detail. A case study illustrating the performance of these algorithms is given in Section 3, followed by a discussion of future directions in Section 4.
Satellite Precipitation Algorithms for Extreme Precipitation Events
487
2 REAL-TIME GOES-BASED SPE AT NESDIS 2.1 Interactive Flash Flood Analyzer (IFFA) Research into using satellite-based IR measurements to estimate precipitation began in the late 1960s (Lethbridge 1967), but it was not until the advent of operational geostationary satellites in the mid-1970s and the development of techniques for operational use that satellite-based estimates of precipitation because suitable for real-time detection of extreme precipitation events. The first of these operational techniques was the Interactive Flash Flood Analyzer (IFFA; Scofield and Oliver 1977; Scofield 1987), which has been used operationally at the NOAA/NESDIS Satellite Services Division (SSD) Satellite Analysis Branch (SAB) since the 1970s (Borneman 1988). The IFFA was originally designed for intense Mesoscale Convective Systems (MCSs) and later extended to other types of extreme precipitation events. The basis of the IFFA is the presumed relationship between cloud-top brightness temperature and rainfall rate – colder cloud tops imply stronger convective updrafts and hence higher rainfall rates than warmer cloud tops. This relationship holds well for the raining cores in convective systems, but is less valid for stratiform areas of precipitation (such as those that trail mature MCSs) and is not valid at all for cirrus anvils (which have cold cloud tops but do not produce precipitation). SAB forecasters apply the IFFA technique by manually identifying the convectively active portions of mesoscale systems, which is done using not only individual images but also comparison of consecutive images to diagnose growth or decay of convective clouds. Once this is done, rainfall rates based on cloud-top temperature are determined, and adjustments are made to account for overshooting cloud tops, cloud mergers, available moisture, low-level inflow, and the speed of the convective system. Since the satellite cloud-top temperatures do not always reflect the extremely strong updrafts and heavy rainfall that can occur before the clouds reach maturity, a rain burst factor is used to make appropriate adjustments early in the life cycle of an MCS. Finally, since certain thermodynamic profiles will support strong updrafts but have a convective equilibrium level that precludes extremely low cloud-top temperatures, the convective equilibrium level temperature is used to enhance precipitation in such instances. For additional details, the reader is referred to Scofield and Oliver (1977) and Scofield (1987). In addition to estimating past and current precipitation, SPE information is also extrapolated into the future to produce 3-h precipitation nowcasts or outlooks (Spayd and Scofield 1984) that take into account the growth, decay, movement, and propagation of individual convective systems (Shi and
488
R. A. Scofield and R. J. Kuligowski
Scofield 1987; Juying and Scofield 1989; Corfidi 2003). SAB SPE and outlooks are sent out to NOAA National Weather Service (NWS) forecasters via the Advanced Weather Interactive Processing System (AWIPS); graphics of the estimates are also available on the SSD home page (http://www.ssd. noaa.gov/PS/PCPN/index.html).
2.2 Auto-Estimator (AE) and Hydro-Estimator (HE) A significant limitation of the IFFA approach is its highly interactive nature. The required labor limits the coverage of estimates to relatively small regions – especially problematic if multiple significant precipitation events are occurring simultaneously. To improve both spatial/temporal coverage and timeliness, NESDIS developed an automated SPE algorithm for highintensity rainfall called the Auto-Estimator (AE). The original AE, developed by Vicente et al. (1998), computes rain rates from 10.7-µm brightness temperatures (hereafter referred to as T10.7) based on a relationship derived from more than 6,000 collocated radar and satellite pixels. Areas of non-raining cold cloud are identified using the spatial gradients of T10.7 and on changes in T10.7 from the previous image. Amounts are then adjusted using a multiplicative moisture adjustment consisting of precipitable water (PW) in inches multiplied by relative humidity (RH) as a decimal value, both from numerical weather model data. During 1998 and 1999, a number of enhancements were made, including the use of 15-min WSR-88D reflectivity data to screen out non-raining cold cloud, an adaptation of the IFFA equilibrium-level temperature adjustment (using numerical model data), and adjustments for parallax and orographic enhancement of precipitation. Many of these enhancements are described in Scofield (2001), Vicente et al. (2002), and Scofield and Kuligowski (2003). However, the AE is highly dependent on radar data to correctly identify non-raining cold cloud pixels because the schemes for identifying them in the AE often incorrectly classify cirrus as raining cloud, resulting in significant overestimation of the spatial extent of heavy precipitation. Since one of the advertised strengths of satellite QPE is its usefulness in regions where radar and/or rain gauge coverage are unavailable, another version of the AE called the Hydro-Estimator (HE) was developed to address this and other issues via three significant new features: • Raining pixels are defined as those with T10.7 below the average value for cloudy pixels in the region surrounding the pixel of interest. This approach substantially reduced the exaggeration of rain area compared to the AE, which in turn eliminated the need for radar as a rain/no rain discriminator. • The rain rate curve is adjusted according to the difference between the pixel T10.7 and the average T10.7 of the nearby cloudy pixels, with the
Satellite Precipitation Algorithms for Extreme Precipitation Events
489
highest rain rates assigned to pixels that are coldest relative to their surroundings. • The components of the PW*RH adjustment have been separated, with the PW used to adjust the rain rate curve according to moisture availability and the RH used to derive an amount to be “evaporated” from the rain rate. These adjustments have improved the handling of stratiform events with embedded convection, and also of wintertime precipitation, which is typically associated with low PW values. The HE has been a very useful source of information for SAB forecasters, allowing them to monitor a greater number of heavy rainfall systems and to disseminate SPEs to NWS Forecast Offices in a more timely fashion than they could when relying solely on the IFFA. However, SAB forecasters have cautioned that the HE does have limitations that still make production of IFFA estimates necessary on many occasions. These include a tendency to overestimate the area and magnitude of heavy precipitation for cold-topped (below –58ºC) systems and to underestimate the heavy rain that can fall from warm-topped (above –58ºC) systems. Also, in regions of strong wind shear, there can be differences in the location of the cloud tops and the resulting rainfall. Finally, the aforementioned convective rain burst factor has not yet been implemented in the HE, resulting in underestimates of rain rates during the early stages of storm development. In spite of these limitations, the HE has been considered sufficiently robust to replace the AE as SAB’s operational automated algorithm and to be disseminated to NWS field offices via AWIPS. As of this writing 1-h estimates of precipitation for the continental USA and surrounding regions are updated hourly. In addition, the HE is run worldwide according to the availability of IR imagery (every 15 min over the continental USA) to produce real-time instantaneous rainfall rates and accumulations over 1, 3, 6, and 24 h.
2.3 GOES Multi-Spectral Rainfall Algorithm (GMSRA) Although much effort into SPE has focused on using a single channel (usually around 10.7 µm), the utility of other channels for precipitation applications has also been investigated by numerous authors. Some of this work has been implemented by Ba and Gruber (2001) into a real-time algorithm called the GOES Multi-Spectral Rainfall Algorithm (GMSRA) that is run every 15 min over the continental USA. The GMSRA uses the five GOES Imager channels as follows: • A threshold visible albedo value below 0.4 screens out thin cirrus (Rosenfeld and Gutman 1994). • Negative values of (T10.7–T6.9) distinguish overshooting tops from anvil cirrus (Tjemkes et al. 1993).
490
R. A. Scofield and R. J. Kuligowski
• During the daytime 3.9-µm reflectance derived from 3.9-, 10.7-, and 12.0-µm radiance data during the daylight hours is related to cloud particle size, and clouds with large particles (effective radius exceeding 15 µm) are considered to be raining even for relatively warm cloud-top brightness temperatures (Rosenfeld and Gutman 1994; Rosenfeld and Lensky 1998). • During the nighttime, this is replaced with a screen that identifies as nonraining those pixels where T6.9–T10.7 and T10.7–T12.0 both exceed threshold values. • Increases in T10.7 from the previous image identify inactive (non-raining) clouds. In addition, values of T10.7–T12.0 exceeding 1 K (Inoue 1987) indicate thin cirrus, but this function was turned off when the GOES-12 satellite (which replaced the 12.0-µm channel with a 13.3-µm CO2 absorption band) was activated. For those clouds classified as precipitating, both the probability of precipitation and the conditional rain rate are computed from T10.7 using different calibrations for different regions. An adjustment for sub-cloud evaporation similar to that used in the AE (PW*RH) is also made. Ba and Gruber (2001) contains additional details the algorithm. The original rain rate calibration for the GMSRA was based on data from a 17-day calibration period; however, a version of the GMSRA has recently been implemented that uses rain gauge-corrected radar to produce updated rain rate curves on a near-real-time basis (Ba et al. 2003). This version of the algorithm was implemented for real-time testing, in parallel with the original, beginning in early November 2003.
3 CASE STUDY AND INTERCOMPARISON Regular validation of SPE at NESDIS began in the spring of 2001 (Kuligowski et al. 2001) and is now automated and posted to the Web (http://orbit-net.nesdis.noaa.gov/arad/ht/ff/validation/validation.html). A case study is presented here in which heavy rains fell in the Tennessee River Valley during 5–7 May 2003, with accumulations exceeding 125 mm covering a significant area (Fig. 1). These rains triggered flash floods and river floods that lasted for over a week, producing US$17 million damage and 3 deaths, and displacing approximately 2,000 people. Flooding in many locations was the worst in 20–40 years. A comparison of 48-h precipitation totals from the AE, HE, and (fixedcalibration) GMSRA to the Stage IV radar/rain gauge product (Fulton et al. 1998) over this region is presented in Fig. 1. Note that, in this particular case, the AE and GMSRA significantly overestimated the spatial coverage of the heaviest precipitation – the result of the aforementioned tendency of
Satellite Precipitation Algorithms for Extreme Precipitation Events
491
satellite-based precipitation algorithms to incorrectly identify cirrus as precipitating cloud. The HE captures best the magnitude and spatial extent of the heaviest precipitation, though the HE maxima are located farther to the east than the Stage IV maxima. Table 1 contains a statistical summary of the performance of these algorithms, both against the Stage IV fields (for hourly amounts) and against rain gauges (for daily amounts) as a reliability check of the Stage IV data. As indicated by both Table 1 and Fig. 1, both the AE and the GMSRA significantly overestimated the amount of precipitation for this particular case, while the HE bias was closer to unity. (Note that the bias values are different because the hourly statistics are computed against the gridded Stage IV data while the daily statistics are computed against available rain gauges.) The AE and HE exhibit similar correlations with the validation data sets, while the GMSRA is less accurate for this case. However, the previously described improvements recently implemented in the GMSRA should lead to improvements in performance.
Figure 1. Comparison total accumulated precipitation for three satellite QPE algorithms and the Stage IV data set for the 48 h ending 1200 UTC 7 May 2003. A contour at 125 mm is provided for reference.
R. A. Scofield and R. J. Kuligowski
492
It is also clear from both the figure and the table that significant improvements are still needed in the quality of SPE. However, for regions without radar data or a dense rain gauge network, these data would prove to be quite valuable in alerting forecasters to the potential for significant flooding from heavy rainfall. Table 1. Performance statistics for satellite precipitation algorithms versus Stage IV (hourly) and versus rain gauges (daily) for the 48 h ending 1200 UTC 7 May 2003: bias ratio (“Bias”) and Pearson correlation coefficient (“Corr.”). Algorithm A-E H-E GMSRA
Hourly Bias 2.72 1.17 2.70
Corr. 0.46 0.42 0.32
Daily Bias 1.84 0.97 1.54
Corr. 0.51 0.52 0.45
4 SUMMARY AND OUTLOOK This paper has presented an overview of satellite QPE algorithms for extreme events that are produced routinely at NESDIS using IR and/or visible data from geostationary-based instrument platforms. As demonstrated by the case study in Section 3, SPE is a useful companion to radar and rain gauges; however, much work remains to be done to improve the accuracy of SPE and produce a product that is suitable for direct incorporation into multisensor precipitation analyses, hydrologic models, and numerical models without the need for manual corrections. Future research and applications in operational SPE will most likely focus on the following six areas: Improving the calibration of SPE by obtaining the best possible validation data and accounting for the scale differences between the validation data and the satellite estimates (e.g., comparing point rain gauge measurements with spatially averaged SPE). Incorporating an increasing volume of data from geostationary satellites into SPE. Not only will spatial resolution continue to improve, but the next generation of geostationary imagers will have additional channels, some of which have already been demonstrated to be useful for retrieving cloud characteristics pertinent to precipitation (e.g., Ackerman et al. 1990; Baum et al. 2000). The availability of hyperspectral data will further enhance this capability (e.g., Chung et al. 2000). Improved understanding of the physical relationship between precipitation and the signals observed in the visible and IR wavelengths will be needed to make optimal use of these new data. Blending of data from geostationary sensors with that from polar-orbiting microwave sensors. Microwave-based estimates of precipitation are considered
Satellite Precipitation Algorithms for Extreme Precipitation Events
493
to be more robust than IR-based estimates because microwave radiances are affected by the vertical profile of water and ice throughout the cloud rather than by only the cloud-top characteristics (see Section 3 for details). But since microwave data are not presently available from geostationary platforms (though work in this area continues to move forward) and thus are available relatively infrequently over a given loation, blending with IR data affords the best oppurtunity to take advantage of the strengths of both data sets. These issues were addressed in detail in Section 4, but an additional algorithm of interest for operational use is the Self-Calibrating Multivariate Precipitation Retrieval (SCaMPR) algorithm (Kuligowski 2003), which operates with latency comparable to IR/visible-only SPE. Blending of SPE with information from other sensors to produce an optimal precipitation data set. Numerous blending methods have already been developed for longer time scales (see Section 4), and methodologies for shorter time scales are also under development, including QPE-SUMS (Gourley et al. 2002) and the Multisensor Precipitation Estimate (MPE) which is planned for operational application by the NWS (Kondragunta and Seo 2004). The use of SPE in prediction applications. Improvements in SPE and useful expressions of its error characteristics will lead to increased use for initializing numerical weather models, especially over oceans and sparely populated regions where other sources of precipitation information are unavailable. SPE will also play an increasing role in hydrologic forecasting for regions where other sources of precipitation data are likewise difficult to obtain or inadequate. Furthermore, given the relatively poor performance of numerical weather prediction models at short lead times (e.g., Doswell 1986), direct use of extrapolated SPE as a nowcasting tool is being investigated, and experimental 1–3 h nowcasts for the continental USA are available in real time on the NOAA/NESDIS Flash Flood Home Page (http://orbit35i.nesdis.noaa.gov/arad/ht/ff). Finally, to realize the maximum benefit from these techniques, SPE will become more and more global in their focus. Many of the experimental algorithms described in this book are already being applied globally, including the HE described in this section. The operational implementation of SPE for real-time use by weather services outside the USA is also proceeding in Mexico (Fortune and Teran 2004) and in Central America (Alfaro 2003).
5 REFERENCES Ackerman, S. A., W. L. Smith, J. D. Spinhirne, and H. E. Revercomb, 1990: The 27–28 October 1986 FIRE IFO cirrus case study: Spectral properties of cirrus clouds in the 8–12 m window. Mon. Wea. Rev., 118, 2377–2388.
494
R. A. Scofield and R. J. Kuligowski
Alfaro, R., 2003: Validation of GOES precipitation estimates over Central America. Colorado State University, Cooperative Institute for Research in the Atmosphere Technical Report ISSN No. 0737-5352-58, 24 pp. Ba, M. B. and A. Gruber, 2001: GOES multi-spectral rainfall algorithm (GMSRA). J. Appl. Meteor., 40, 1500–1514. Ba, M. B., A. Gruber, and M.-J. Jeong, 2003: Frequency distribution of rain rate with cloud top brightness temperature and near-real time calibration of GOES Multispectral Rainfall Algorithm. Prepr. 12th Conf. Satellite Meteor. and Ocean., Long Beach, CA, Amer. Meteor. Soc., CD-ROM, P4.17. Baum, B. A., P. F. Soulen, K. I. Strabala, M. D. King, S. A. Ackerman, W. P. Menzel, and P. Yang, 2000: Remote sensing of cloud properties using MODIS airborne simulator imagery during SUCCESS 2. Cloud thermodynamic phase. J. Geophys. Res., 105(D9), 11781–11792. Borneman, R., 1988: Satellite rainfall estimating program of the NOAA/NESDIS Satellite Analysis Branch. Natl. Wea. Dig., 13, 7–15. Chung, S., S. Ackerman, P. F. Van Delst, and W. P. Menzel, 2000: Model calculations and interferometer measurements of ice-cloud characteristics. J. Appl. Meteor., 39, 634–644. Corfidi, S. F., 2003: Cold pools and MCS propagation: forecasting the motion of downwind developing MCSs. Wea. Forecasting, 18, 997–1017. Doswell, C. A., 1986: Short-range forecasting. Mesoscale Meteorology and Forecasting, P. S. Ray, Ed., Amer. Meteor. Soc., 689–719. Fortune, M. A. and A. R. Teran, 2004: Satellite-based rainfall estimates as a driver of river forecasts in Mexico. Prepr. 18th Conference on Hydrology, Long Beach, CA, Amer. Meteor. Soc., CD-ROM, P3.6. Fulton, R. A., J. P. Breidenbach, D. J. Seo, and D. A. Miller, 1998: The WSR-88D rainfall algorithm. Wea. Forecasting, 13, 377–395. Gourley, J. J., R. A. Maddox, K. W. Howard, and D. W. Burgess, 2002: An exploratory multisensor technique for quantitative estimation of rainfall. J. Hydrometeor., 3, 166–180. Inoue, T., 1987: A cloud type classification with NOAA 7 split-window measurements. J. Geophys. Res., 92, 3991–4000. Juying, X. and R. A. Scofield, 1989: Satellite-derived rainfall estimates and propagation characteristics associated with mesoscale convective systems (MCS). NOAA Tech Memo. NESDIS 25, 49 pp. Kondragunta, C. and D.-J. Seo, 2004: Toward integration of satellite precipitation estimates into the multi-sensor precipitation estimation algorithm. Prepr. 18th Conference on Hydrology, Long Beach, CA, Amer. Meteor. Soc., CD-ROM, J1.7. Kuligowski, R. J., 2002: A self-calibrating GOES rainfall algorithm for short-term rainfall estimates. J. Hydrometeor., 3, 112–130. Kuligowski, R. J., S. Qiu, R. A. Scofield, and A. Gruber, 2001: The NESDIS QPE verification program. Prepr. 11th Conf. on Satellite Meteor. and Ocean., Amer. Meteor. Soc., Madison, WI, 383–384. Lethbridge, M., 1967: Precipitation probability and satellite radiation data. Mon. Wea. Rev., 95, 487–490. Rosenfeld, D. and G. Gutman, 1994: Retrieving microphysical properties near the tops of potential rain clouds by multi spectral analysis of AVHRR data. Atmos.Res., 34, 259–283. Rosenfeld, D. and I. Lensky, 1998: Satellite-based insights into precipitation formation processes in continental and maritime convective clouds. Bull. Amer. Meteor. Soc, 79, 2457–2476. Scofield, R. A., 1987: The NESDIS operational convective precipitation estimation technique. Mon. Wea. Rev. 115, 1773–1792. Scofield, R. A., 2001: Comments on “A quantitative assessment of the NESDIS AutoEstimator”. Wea. Forecasting, 16, 277–278.
Satellite Precipitation Algorithms for Extreme Precipitation Events
495
Scofield, R. A. and R. J. Kuligowski, 2003: Status and outlook of operational satellite precipitation algorithms for extreme precipitation events. Wea. Forecasting, 18, 1037– 1951. Scofield, R. A. and V. J. Oliver, 1977: A scheme for estimating convective rainfall from satellite imagery. NOAA Tech. Memo. NESS 86, 47 pp. Scofield, R. A., M. DeMaria, and R. Alfaro, 2001: Space-based rainfall capabilities in hurricanes offshore and inland. Prepr. Symposium on Precipitation Extremes: Prediction, Impacts, and Response, Albuquerque, NM, Amer. Meteor. Soc., 297–301. Scofield, R. A., R. J. Kuligowski, and C. Davenport, 2004: The use of the Hydro-Nowcaster for mesoscale convective systems and the Tropical Rainfall Nowcaster (TRaN) for landfalling tropical systems. Prepr. Symposium on Planning, Nowcasting, and Forecasting in the Urban Zone, Seattle, WA, Amer. Meteor. Soc., CD-ROM, 1.4. Seo, D. J. and J. P. Breidenbach, 2002: Real-time correction of spatially nonuniform bias in radar rainfall data using rain gauge measurements. J. Hydrometeor., 3, 93–111. Seo, D. J., J. P. Breidenbach, R. Fulton, and D. Miller, 2000: Real-time adjustment of rangedependent biases in WSR-88D rainfall estimates due to nonuniform vertical profile of reflectivity. J. Hydrometeor., 1, 222–240. Shi, J. and R. A. Scofield, 1987: Satellite observed mesoscale convective system (MCS) propagation characteristics and a 3–12 hour heavy precipitation forecast index. NOAA Tech. Memo., NESDIS 20, U.S. Dept. of Commerce, Washington, D.C., 43 pp. Smith, J. A., A. A. Bradley, and M. L. Baeck, 1994: The space-time structure of extreme storm rainfall in the southern Plains. J. Appl. Meteor., 33, 1402–1417. Smith, J. A., D.-J. Seo, M. L. Baeck, and M. D. Hudlow, 1996: An intercomparison study of NEXRAD precipitation estimates. Water Resour. Res., 32, 2035–2045. Spayd, L. E., Jr. and R. A. Scofield, 1984: An experimental satellite-derived heavy convective rainfall short range forecasting technique. Prepr. 10th Conf. Weather Forecasting and Analysis, Clearwater Beach, FL, Amer. Meteor. Soc., 400–408. Tjemkes, S. A., L. van de Berg, and J. Schmetz, 1997: Warm water vapor pixels over high clouds as observed to Meteosat. Beitr. Phys. Atmos., 70, 15–21. Vicente, G. A., R. A. Scofield, and W. P. Menzel, 1998: The operational GOES infrared rainfall estimation technique. Bull. Amer. Meteor. Soc., 79, 1883–1898. Vicente, G. A., J. C. Davenport, and R. A. Scofield, 2002: The role of orographic and parallax corrections on real time high resolution satellite rainfall rate distribution. Int. J. Remote Sens., 23, 221–230. Westrick, K. J., C. F. Mass, and B. A. Colle, 1999: The limitations of the WSR-88D radar network for quantitative precipitation measurement over the western United States. Bull. Amer. Meteor. Soc., 80, 2289–2298. Young, C. B., B. R. Nelson, A. A. Bradley, J. A. Smith, C. D. Peters-Lidard, A. Kruger, and M. L. Baeck, 1999: An evaluation of NEXRAD precipitation estimates in complex terrain. J. Geophys. Res., 104, 19691–19703.
39 APPLICATION OF A BLENDED MW-IR RAINFALL ALGORITHM TO THE MEDITERRANEAN Francesca Torricella1,Vincenzo Levizzani1, and F. Joseph Turk2 1
Institute of Atmospheric Sciences and Climate, ISAC-CNR, Bologna, Italy Naval Research Laboratory, NRL, Monterey, CA, USA
2
1 INTRODUCTION The network of meteorological and environmental satellites is the only practicable means of monitoring and gauging rainfall on a global scale. For this reason an adequate estimation of the accuracy of global operational rain products is crucial. The first step for estimating error characteristics is perhaps a kind of local analysis, validation campaign and limited comparisons with reference data sets, in general taken from ground based instruments and networks, deemed to represent “truth” data sets. Historically, ground based radar and rain gauges supplied the reference data for such comparisons. With the advent of space based radars such as the precipitation radar of the Tropical Rainfall Measurement Mission (TRMM) reference data sets are also available from space. Moreover, the comparison with products derived from concurrent sensors (e.g., TRMM mixed radar-microwave products) is is another tool to check the overall performance of newly developed algorithms. A parallel approach consists in developing an error model of the retrieval process and assessing the uncertainties of its parameters. The key here is that the product being validated be derived on a physical basis with empirically verifiable assumptions. To date, the largest efforts of the scientific community have been aimed to the assessment of the accuracy of mean (monthly, weekly, daily) or cumulated rain products over suitable study periods and using standard evaluation statistics (Adler et al. 2001). Nevertheless, the application of satellite derived analysis to the characterization of severe rain events, meteorological applications, and flood management requires that the global achievements of the validation exercise 497 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 497–507. © 2007 Springer.
498
F. Torricella et al.
be verified on a local scale. It is necessary to understand whether useful and reliable estimates of precipitation fields can be obtained in particular climatic and geographic conditions. The goal is to analyze the performances of the method in producing instantaneous rainfall maps. It is conceivable that global algorithms need local optimization if snapshots of the precipitation field are required for the devised applications, because the physics of the rain processes strongly depends on the immediate environment. A global blended infrared-passive microwave (IR-PMW) technique producing rain rate fields at the time/space resolution and coverage of geostationary (GEO) observations is applied to rain events over Mediterranean countries. Two cases are examined: a series of intense rainstorms that affected the Emilia-Romagna area (Northern Italy) in early August 2002, and the November 2001 Algeria flood. In spite of their unusual intensity, they were selected among a set of analogous cases for they are representative of several common characteristics that a rain algorithm should be capable of coping with when functioning in such particular environment.
2 THE HYBRID PMW-IR RAINFALL ESTIMATION METHOD The idea of using data from GEO satellites to produce rain rate (RR) maps over large areas of the globe was largely exploited using visible (VIS) and IR stand-alone observations or combining them with information from different sensors, especially PMW instruments on polar platforms. At present, only GEO measurements have the spatial resolution (a few km2), repetition time (15–30 min), and spatial coverage suitable to properly follow the rapid variations of precipitation fields. Moreover, the long history and the robust technology of GEO instruments prompts for the reanalysis of historical events and guarantees a timely and reliable release of calibrated data. VIS and IR measurements give only indirect information on the precipitation field being limited to the uppermost cloud layer near the top. Their uncertainties are thus relevant per se since the precipitating hydrometeors do not interact directly with the photons collected aloft by space borne instruments at these wavelengths. Several methods have been developed that “calibrate” for example IR brightness temperature (TB) data by using the more physically based rain estimates derived from PMW instruments. This kind of blended techniques is intrinsically constantly evolving due to the ever-expanding suite of PMW sensors and multispectral GEO imagers. Moreover they should mitigate the sampling error deriving from noncontinuous precipitation sampling due to the orbital characteristics of the satellite and the spatio-temporal structure of precipitation associated with diurnal, synoptic, seasonal, and interannual variability cycles.
Application of a Blended MW-IR Rainfall Algorithm
499
The blended technique adopted hereafter (Turk et al. 2000b) has been recently validated using rain-gauge data and analyses by the Korean Meteorological Agency and the Australian Bureau of Meteorology. Looking at mean or cumulated rainfall amounts, the correlation fairly increases and bias and root mean square error decreases as either the integration/averaging period is increased (from a minimum of 1 h up to 30 days) or the grid size for spatial averaging is coarsened (from 0.1 to 3°). The original operational set up of the software (global, automatic, real time, using a suite of PMW and IR observations) was adapted to the task of analyzing test case studies. In the Turk’s method, hereafter referred to as Naval Research Laboratory technique (NRLT), rain rates derived from PMW measurements are used to create global, geolocated RR-TB relationships that are renewed as soon as new collocated data are available from both GEO and PMW instruments. The PMW RR data can be derived, in principle, from any source, provided they are geolocated rain intensities in mm h–1, and that the files containing them report the useful information on orbit (date, start time, sensor, satellite, etc.). For the present work the adopted PMW estimates are mainly derived from Special Sensor Microwave/Imager (SSM/I) data. From the brightness temperatures measured in seven polarized channels from 19.2 to 85.5 GHz, rain rates are derived by means of the NOAA-NESDIS operational algorithm (Ferraro and Marks 1995; Ferraro 1997). The NESDIS algorithm derives rainrates at the A-scan resolution of the SSM/I (~25 km) by means of nonlinear relationships involving the instrument channels (vertical and horizontal polarization) that have been calibrated using large sets of ground reference data collected by radar networks in different countries. The physical basis of such relationships are the scattering of MW radiation due to large ice particles above the freezing level occurring in precipitating clouds, and the emission from liquid water. This latter phenomenon can be sensed only above oceanic surfaces, due to high and largely unknown emissivity of land surfaces in the MW spectral range. Relying on PMW measurements only (no need of large input database of physical properties) and on simple but well founded relationships, this algorithm is very robust and lends itself to global applications. In the NRLT, to the end of calibrating IR measurements, the globe (or the study area) is subdivided in equally spaced LATLON boxes (2.5° × 2.5°). For each box, space and time coincident IR and PMW measurements are reduced to the worse spatial resolution and then collected. The colocation process allows for time and space offsets (15 min and 10 km, respectively). To form a meaningful statistical ensemble the method can look at older PMW orbit-IR slot intersections, until a certain box coverage is reached (say 75%) and a minimum number of coincident observations is gathered for a 3° × 3° boxes region. By means of this set of RR and corresponding TB, the RR-TB relationships are derived by applying a probability matching method (Calheiros and Zawadzki 1987).
F. Torricella et al.
500
a)
b)
c)
d)
e)
f)
g)
h)
i)
j)
k)
l)
m)
n)
o)
p)
Figure 1. 6 August, 2002. Rain intensity maps in mm h -1 for the Emilia-Romagna storm case study. All times are UTC. a) Radar map at 0012; b) NRLT for the slot starting at 0000; c) Radar map at 0042; d) NRLT for the slot starting at 0030; e) NRLT for the slot starting at 0630; f) PMW NESDIS algorithm for the SSM/I orbit 08D (F13) starting at 0627; g) NRLT for the slot starting at 0700; h) Radar map at 0642; i) NRLT for the slot starting at 0800; j) PMW NESDIS algorithm for the SSM/I orbit 10D (F14) starting at 0828; k) NRLT for the slot starting at 0800; l) Radar map at 0842; m) NRLT for the slot starting at 0930; n) PMW NESDIS algorithm for the SSM/I orbit 12D (F15) starting at 0941; o) NRLT for the slot starting at 1000; p) Radar map at 0942. (see also color plate 17)
501
Application of a Blended MW-IR Rainfall Algorithm
b)
d)
f)
h)
j)
Figure 2. 10 November, 2001. Rain intensity maps in mm h -1 for the Algeria case study. Time is UTC. a) SSM/I orbit 23A (F14) starting at 1919; b) SSM/I orbit 24A (F15) starting at 1958; c) NRLT for the slot starting at 0000; d) NRLT for the slot starting at 0200; e) PR for the TRMM orbit 22741 (area overpass around 0032); f) PR for TRMM orbit 22742 (area overpass around 0210); g) NRLT for the slot starting at 0000 (calibrated with PR data); h) NRLT for the slot starting at 0200 (calibrated with PR); i) 2A12 TMI rainrates for the TRMM orbit 22741; j) 2A12 TMI rainrates for the TRMM orbit 22742.
502
F. Torricella et al.
3 AUGUST 2002 INTENSE RAINSTORMS OVER EMILIA-ROMAGNA The Po Valley in Northern Italy is surrounded on three sides by high mountains and is therefore characterized by relatively high humidity and light winds at lower atmospheric levels, rather favorable conditions for the formation of line storms and Mesoscale Convective Systems. In August 2002 a number of intense storms hit the Italian peninsula, and attained a relevant intensity over a large area from Tuscany through Emilia-Romagna all the way up to northeastern Italy. Hailfall damages were widely registered in the belt between France, Switzerland, southern Germany, Austria, Hungary, and the Caucasus region. Early in the month instability conditions were fostered by a cyclonic area insisting over France, with cold air fluxes over north-central Italy, with associated strong winds, heavy rainfall, and hailstorms. The monthly rainfall accumulations reached unusually high values, with a maximum positive anomaly of 100 mm in Ferrara. Rainstorms started on the 4th and, after a few hours of pause, they intensified again in the afternoon of the 5th one after the other with similar characteristics and hitting a limited geographical area. The storm cells originated west of the Alps and moved rapidly eastward crossing the entire Po Valley. The electric activity remained quite impressive during the entire duration. The first useful SSM/I overflight is the orbit at 1601 UTC on day 5. Later on, the zone was imaged 6 times by the SSM/I sensors in about 24 h, thanks to the availability of data from three satellites (F13, F14, and F15). The starting time of the orbits on day 5 were 1601, 1802 and 1916 UTC, and there was no overpass during the night. The first overpass on day 6 corresponds to the orbit starting at 0627 UTC and the area was then covered by the orbits 0828, 0941 and 1547 UTC. After 1600 UTC the storm system left the region and moving eastward toward Slovenia. The rain maps for the overpass on day 6 are shown in Fig. 1. On the overall, the PMW algorithm detects the storm cells and gauges the high precipitation intensities up to 35 mm h–1, the maximum allowed rain intensity in the NESDIS algorithm (panels f, j, n). Note that all the SSM/I orbits cover the western part of the area (the gray shad delimits the area where the method/instrument gives results). The first row in the figure collects the results from NRLT for two slots during the night (panels b and d). For comparison nearly simultaneous radar maps are shown (panels a and c). Radar data are taken from C-band dual polarisation Doppler weather radar of the Servizio Meteorologico Regionale (SMR) in S. Pietro Capofiume, (44.654° N, 11.624° E, 11 m a.s.l.) in the southeastern sector of the Po valley. For these two slots, the last PMW overpass that calibrated the relationships for the blended rain intensities was the one on day 5 starting at 1916
Application of a Blended MW-IR Rainfall Algorithm
503
UTC, i.e., it was about 3 h old. This can be perceived by observing the precipitation cell over the Italy-Slovenia border. The NRLT correctly followed the movement of this cell, but the precipitation field became unrealistically extended, somewhat uniform, and the peak intensity was too low. Nonetheless, the NRLT located the new cell in the west, with peak intensity, location and shape in good agreement with radar measurements. This rather good performance is confirmed by looking at the NRLT map (panel e) just before the next PMW calibration (panel f) in early morning of day 6. The location of the rain cells is still preserved more than 11 h after the last calibration, and the rain field matches closely the radar data (panel h). The agreement obviously improves in picture g, due to the intervening PMW calibration (panel f). The capability of NRLT in extending the rain field information outside the PMW spatial coverage is testified in the next row of Fig. 1. The easternmost cell detected by radar in panel l (over Istria) is revealed also by NRLT even if no direct PMW measurements was available for this area since many hours. The last row in Fig. 1 (panels m, n, o, and p) confirms the NRLT performances with the exception of the feature in the upper-right corner of panel o that appears to be structureless.
4 THE NOVEMBER 2001 ALGERIAN FLOOD In early November 2001, a widespread frontal system and upper air trough from northeast Scandinavia to southwest Spain led to an extreme precipitation event in Algiers, causing severe flooding and huge mudslides. More than 120 mm of rain fell in 12 h during the night between 9 and 10 November and more than 130 mm during the next 6 h on the mountains behind Algiers. The unusually large rain rates were fed by the cold maritime arctic air that picked up moisture crossing the warm Mediterranean waters and met maritime subtropical air. An intense orographic enhancement was caused by strong surface winds oriented towards the high mountains of the African coast (>2300 m a.s.l.). The sudden onset of precipitation, the orographic complexity of the terrain, the vicinity to the coast are all elements that can introduce large errors and bias in PMW rainfall retrieval algorithms. Due to the short duration of the event (about 20 h), the area was imaged only a few times by PMW instruments. In such unfavorable condition, rapid update techniques are a powerful instrument to follow the evolution of the otherwise poorly observed rain field. The scarcity of validation data suggested a strategy involving not only SSM/I data but also data from the instruments onboard the Tropical Rainfall Measuring Mission (TRMM). The SSM/I derived RR fields were used for the calibration of the statistical relationships within the NRLT while the TRMM Microwave Imager (TMI) 2A12 operational rain product was retained as a source of comparison/validation data. Moreover,
504
F. Torricella et al.
due to the unsatisfactory results of the NESDIS SSM/I rain algorithm, the NRLT was re-applied using measurements of the TRMM Precipitation Radar (PR) (2A25 operational product) as a calibration source for the blended product. The rain maps for the central hours of the night between 9 and 10 November are collectively shown in Fig. 2 and document the peak intensity of the event. Note the following features for interpreting the rain maps: (a) the gray shaded area represents the swath coverage of the instrument or product; (b) the rain intensity in mm h–1 are represented according to the color scale included in the figure, from 0 to 35 mm h–1; (c) the first column (panels a, c, e, g, and i) refers to 0030 UTC and the second (panels b, d, f, h, and j) to 0100 UTC; (d) the town of Algiers is marked by a red circle. The inspection of PMW rain maps (panels a and b) reveals that coastal precipitation is missed altogether. Starting 50 km offshore and coming ashore the NESDIS algorithm uses the land module, i.e., precipitation in coastal environment is treated as it were over land. This is the most difficult zone to treat because of the discontinuity in atmospheric conditions and of the mixed sea-land signal collected within the field of view. In order to try to explain the complete lack of precipitation signal along the coast, the NESDIS algorithm was modified to eliminate the ad hoc treatment of coastal environment: overseas pixels were processed by means of the “sea” algorithm and overland with the “land” algorithm, no matter how far from the coast. The result somewhat unexpected of this exercise is that the “sea” part of the PMW algorithm works fairly well and detects convective cells along the coast, although at reduced rain intensity. Nonetheless, spurious rain signatures appear all along the coast even in cloud free conditions. On the contrary, no relevant rain signatures appeared inshore. It is thus evident that the “land” part of the algorithm is completely unfit for the particular rain type and/or surface characteristics. By analyzing the terrain classification applied prior to the rain computation it appears that some small (precipitation?) area is misclassified as snow. The largest part of precipitation over the Mediterranean sea is derived by means of the scattering algorithm. The heaviest precipitation is detected over the Mediterranean off the coast of North Africa, and the values never exceeded 13 mm h–1. Because of the lack of SSM/I overpasses after 1958 UTC (the successive orbit is at 0504 UTC the day after) the relationships derived from the two orbits are used to derive the rainfall maps during the night. Heavy rain started falling after this overpass, so that the algorithm faced very critical data input conditions. The analysis of the NRLT results for the same time period (b and d) do not bring about substantial improvements with respect to PMW’s. Almost no precipitation is detected over the African continent, excluding light rain over the Algeria-Tunisia border. One of the most prominent merit of the hybrid method is that it can eliminate discontinuities and mitigate the problem of
Application of a Blended MW-IR Rainfall Algorithm
505
directly deriving precipitation over the coast from PMW, but in this case even this type of technique is of no avail in detecting precipitation over Algiers. The geolocated statistical RR-TB relationships for the nine boxes covering the area show that the heaviest precipitation appear to be associated with the Algiers box and the one immediately to the North, with a maximum intensity of about 9 mm h–1. The adjacent boxes show similar relationships, but with a reduced rain intensity. The zero-rain threshold is always < –40°C (–47°C for the box containing Algiers), and the range of temperatures associated to precipitation is no more than 10°C if one considers intensities >1 mm h–1. These relationships are obviously derived from rainy pixels over the sea. Since the NRLT is based in the end on the IR TB, METEOSAT IR images for the whole night were analyzed, considering a 1°.×.1° box centered over Algiers. The TBs remained constantly higher than the zero rain threshold, and this explains why no precipitation is detected. In practice the behavior of the NRLT in the particular case reveals that the characteristics of the clouds and precipitation fields were definitely different over land (warm rain) with respect to the convective cells embedded in the storm system, which developed over the sea very close to the coastline. This is confirmed by the PR observations (panels e and f) that show very intense precipitation cells off the Morocco-Algeria coast that appear to be delimitated by the coastline. Very likely the orography played a major role in the event, especially the relief south of Algiers, giving rise to a precipitating system that, although embedded in a larger field, was neither detected by PMW (due to the low scattering) nor by the NRLT (due to the high TB). The NRLT was further calibrated using PR rain measurements. The results (panels g and h), however, look quite disappointing. The rain pattern is not much different with respect to panels c and d, but the rain intensity values rose to much higher values, according to the radar measurements. Indeed the PR data, due to very narrow swath of the instrument, drastically change only the relationships of the westernmost boxes, altering the rain intensity associated with each IR temperature, but not the overall shape of the relationship. Unfortunately, due to the low number of overpasses and the characteristics of the TRMM orbit, the PR did not take measurements over the area of the disaster. For comparison with the NRLT results TMI rainfall maps are shown in panels i and j. The maps reveal that most of the precipitation over Algiers is missed, even if the TMI rain product is derived by means of a completely different PMW algorithm, the Goddard Profiling Algorithm (GPROF; Kummerow et al. 2001; McCollum and Ferraro 2003). In general, quite low precipitation is detected over the land. Rain rates do not differ very much between the two methods, but are in general lower for NRLT. In the NRLT maps the rain field appears shifted to the North, preventing a meaningful
506
F. Torricella et al.
numerical comparison between the two, but the main features of the field seem fairly preserved, even 4 h after the calibration.
5 CONCLUSIONS Tests of a global blended PMW-IR satellite rainfall technique were conducted for two rainfall events in the Mediterranean area: (1) rainstorms over the Po Valley in Northern Italy and (2) a coastal heavy, flood-producing storm over Algiers. The technique constantly evolves in time due to the everexpanding suite of PMW sensors and multispectral GEO imagers. However crude the blending mechanism might be, as for instance in the use of a single thermal IR channel, the technique performed fairly well when the underlying MW algorithm supplies reliable input rain maps for its calibration. Results show that, on one hand, the technique suffers from the usual shortcomings of the adopted PMW algorithm, as for instance the unsatisfactory treatment of coastal environments and the possible misclassification of pixels in the screening procedure that precedes the rain retrieval. In certain environments these deficiencies prevent rainfall detection in the first place and therefore a complete representation of precipitation fields, as demonstrated by the analysis of the Algeria event. Moreover, the technique reveals intrinsic shortcomings connected mainly to the presence over the same area of quite different precipitation types, as is the case of very cold convective nuclei embedded in stratiform fields. The technique establishes geolocated rainrate/brightness temperature relationships that are the more correct the more homogeneous the characteristics of the precipitation field for the analysis area. This is obviously driven by the global design of the technique. On the other hand, if the previous unfavorable conditions do not arise, the technique is able to fairly reconstruct the precipitation field outside the space/ time domain covered by PMW observations, even in case of sparse and uneven PMW overpasses. The improvement of PMW rain estimations and the use of multispectral channel analysis will add more precision to the presently available blended products. It is conceivable that the advent of the ongoing international missions aimed to dramatically reduce the gaps between successive MW observations does not undermine the usefulness of blended techniques that remain the preferable method for producing reliable instantaneous rainfall maps, if they are required to be global, frequent and continuous in space.
6 REFERENCES Adler, R. F., C. Kidd, G. Petty, M. Morrissey, and H. M. Goodman, 2001: Intercomparison of global precipitation products: The third precipitation intercomparison project (PIP-3). Bull. Amer. Meteor. Soc., 82, 1377–1396.
Application of a Blended MW-IR Rainfall Algorithm
507
Calheiros, R. V. and I. Zawadzki, 1987: Reflectivity rain-rate relationship for radar hydrology and Brazil. J. Climate Appl. Meteor., 26, 118–132. Ferraro, R. R., 1997: Special sensor microwave imager derived global rainfall estimates for climatological applications. J. Geophys. Res., 102 (D14), 16715–16735. Ferraro, R. R. and G. F. Marks, 1995: The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J. Atmos. Oceanic Technol., 12, 755–770. Kummerow, C., Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. Turk, F. J., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000a: Meteorological applications of precipitation estimation from combined SSM/I, TRMM and geostationary satellite data. Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia, eds., VSP, Utrecht, The Netherlands, pp. 353–363. Turk, F. J., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000b: Analysis and assimilation of rainfall from blended SSMI, TRMM and geostationary satellite data. Proc. 10th AMS Conf. Sat. Meteor. Ocean., 9–14 January, Long Beach, CA., 66–69.
40 RETRIEVING PRECIPITATION WITH GOES, METEOSAT AND TERRA/MSG AT THE TROPICS AND MID-LATITUDES Christoph Reudenbach, Thomas Nauss, and Jörg Bendix Faculty of Geography, University of Marburg, Germany
1 INTRODUCTION Water affects all economic, cultural, social, and ecological aspects of daily life all over the world. Hence, investigations of sustainable water management strategies and risk assessment are main topics in today’s hydrological research. Therefore, reliable knowledge about the spatio-temporal dynamics of rainfall as a key input parameter in complex and high resolution (1 × 1 km², 1 h) decision support systems on the regional water cycle like DANUBIA for the upper Danube catchment (refer to Mauser 2003; Ludwig et al. 2003) is indispensable. At present, only adapted mesoscale weather models (Schipper 2004), locally restricted weather radar networks, and optical sensors onboard of geostationary satellites (e.g., MSG) can provide this information in the required spatial and temporal resolution however, with varying accuracy. While model results are necessary for scenario simulations under a changing climate, satellite retrievals are the only way to provide global coverage of rainfall data for climatological purposes, nowcasting and model validation. Because of the high temporal (10–30 min), spatial (3–10 km at nadir) and recently increased spectral (GOES-12, MSG) resolution of geostationary satellite sensors, this provides the opportunity to retrieve short term convective processes with lifecycles even less than 2 h. Moreover, the long-lasting GOES and Meteosat missions makes available long-term validation data sets for numerical models as well as possibilities for monitoring climatic extremes like El Niño (Bendix 1997, 2000). The current paper describes the Advective-Convective Technique (ACT) rainfall retrieval algorithm which is an improved version of the Enhanced 509 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 509–519. © 2007 Springer.
510
C. Reudenbach et al.
Convective-Stratiform-Technique (ECST) (refer to Reudenbach et al. 2001). It is designed for optical sensors of the GOES and Meteosat missions and recently updated by including cloud microphysics to benefit from the increased spectral resolution of the new generation of geostationary satellite sensors (e.g., MSG-SEVIRI). Three examples may present the different stages of the ACT development together with its possible range of application. The first example shows results from mostly convective heavy rain observations during the super El Niño event 1997/98. The second focuses on the combination of a convective and advective retrieval scheme which is necessary for a proper retrieval of mid-latitude precipitation processes over the upper Danube catchment area. The third study points on the severe European summer flooding in 2002 in order to demonstrate the benefits of including cloud microphysics in precipitation retrievals.
2 THE ADVECTIVE-CONVECTIVE TECHNIQUE ACT Operational satellite-based rainfall retrievals predominantly focus on tropical/subtropical regions but only case studies have been performed for the mid-latitudes (Levizzani et al. 2001; Levizzani 2003). These studies have proven that straightforward convective schemes which normally identify potential precipitating clouds by means of their infrared brightness temperature (TBIR) usually perform well in the tropics but cannot simply be applied to the complex situation of mid-latitude frontal precipitation. Hence, a new modular retrieval scheme, the ACT that is also applicable to advective precipitation in the mid-latitudes has been developed. It consist of three modules which deal with precipitation retrieval from convective core areas, from advective cloud regions and an enhanced classification scheme for precipitating clouds by using cloud microphysical properties. Because the first two modules only require brightness-temperatures from the infrared (TBIR) and water vapour (TBWV) channels, they can be used to investigate existing long time series of geostationary data (e.g., Meteosat). However, the increased spectral resolution of the latest generation of geostationary satellites (especially spectral bands at 0.6, 1.6 and 3.9 µm) is necessary for the third module. Figure 1 presents the principal outline of the ACT scheme. The ACT convective module is based on the Enhanced Convective Stratiform Technique (ECST; Reudenbach et al. 2001; Reudenbach 2003) that uses positive TBWV-TBIR differences (DWI) in order to discriminate between deep convective, optically thick clouds (DWI > 0) and non-raining cirrus (DWI < 0, refer to Tjemkes et al. (1997)). Pixels with positive DWI are then subdivided by analysing the frequency distribution of brightness temperatures (TBIR). Areas with TBIR < 1st quartile of the frequency distribution represent overshooting tops of convective cores, those who suit the 1st quartile reveal raining systems at tropopause level and pixels with TBIR < 3rd quartile identify potentially raining cloud systems of high vertical
Retrieving Precipitation with Goes, Meteosat and Terra/MSG
511
extension. As a result, isolated convective cores can be distinguished from directly adjacent stratiform raining areas.
Figure 1. Principal overview of the ACT convective, advective, and microphysics scheme.
The second module detects rainfall areas in warmer frontal systems (e.g., warm frontal clouds). The method is based on an iterative k-means clustering algorithm (Bradley and Fayyad 1998) that is applied to TBIR, TBWV and 3 ×3 infrared standard deviations (StdvIR). It integrates the classified cloud process patterns from the convective module as core raining areas. The resulting clusters represent potentially raining cloud types (advective-stratiform precipitation) if the maximum TBIR and StdvIR of the cluster are below respective higher than specified thresholds. Each cluster is reallocated into single cloud entities for which the compactness is calculated. The resulting cloud systems are classified as raining or non-raining (e.g., Cirrus, non-raining Nimbostratus) by means of a discriminant function based on the cluster centroid temperature, the compactness of the cloud entity and the number of embedded, isolated convective cores. After the classification of raining clouds, a specific rain rate is assigned to each pixel. The rain rates are derived from idealised 3D cloud model runs with the mesoscale Advanced Regional Prediction System (ARPS; Xue et al. 2003). For that, ten years of radiosonde data over central Europe were analysed with regard to specific convective indices (i.e., total-totals) using a 1D cloud model (Zock et al. 1995; Bendix and Bendix 1998). This analysis yielded a set of representative rain-bringing profiles which were used to initialise the ARPS model. The spatio-temporal assignment of rain rate in dependence on the brightness temperature of the raining pixel is performed by aggregating the simulated cloud top temperatures (ARPS) and rainfall
512
C. Reudenbach et al.
rates with respect to the viewing geometry and scan cycle of the used sensor. The model-based relations between cloud-top temperature and rain rate leads to specific transfer functions that are used for the final processing of the classified images. The ACT algorithm can be applied to almost every optical satellite system as long as it provides at least one water vapour and one infrared channel. However, a new cloud microphysics module could be implemented due to the increased spectral resolution of recent geostationary satellite sensors. This is important for the improvement of the ACT, because potential raining cloud systems require both a minimum optical thickness and a minimum effective cloud droplet radius (Lensky et al. 2003). Both parameters are simultaneously retrieved using an improved version of Nakajimas GTR code (Nakajima and Nakajima 1995; Kawamoto et al. 2001; Bendix 2002) that is adapted to Terra-MODIS and MSG-SEVIRI spectral bands at about 0.6 and 3.9 and 11 µm. It will be shown that knowledge about optical depth and effective radius allow an even more accurate separation between non-raining and raining advective clouds within the advective module of the ACT. To demonstrate the feasibility of the ACT, three examples present the chronological development of the different modules. The first example shows results from the convective scheme over Ecuador using GOES-8 data. The second study presents one year of rainfall retrieval over the upper Danube catchment area in Germany by means of both the convective and the advective module based on Meteosat-7 data. The third example reveals the potential enhancement of the ACT for the European summer flooding event of 2002 where cloud microphysics could be additionally considered using Terra-MODIS data in order to simulate the forthcoming potential of Meteosat-8 SEVIRI (MSG).
3 PRECIPITATION DYNAMICS DURING EL NIÑO 1997/98 IN ECUADOR – APPLYING THE ACT CONVECTIVE MODULE El Niño events cause heavy precipitation and significant economical losses in the normally dry costal areas of southern Ecuador and northern Peru while La Niña has similar impacts at the eastern-Andean slopes. In order to retrieve information about the spatio-temporal rainfall distribution and formation the convective module of the ACT was applied to half-hourly band 3 (6.47–7.02 µm) and band 4 (10.2–11.2 µm) GOES-8 data (refer to Bendix et al. 2003) during the centennial super event of 1997/98. The study area that is subdivided in the coastal plains, the Andean highlands including the eastern and western Cordillera and the inter-Andean basins passing into the Amazon region is shown in Fig. 2a. Normally, this region is characterised by two rainy seasons (March/April, October/November) except for
Retrieving Precipitation with Goes, Meteosat and Terra/MSG
513
the eastern Andean slopes between 1,000 m and 3,200 m with one rainy season in July and the more arid regions of southern Ecuador/northern Peru with only one weak precipitation peak in March (Bendix and Lauer 1992). Figure 2b shows total precipitation for an 11-day period during El Niño with maximum values exactly in the arid regions mentioned above and only slight anomalous effects within the inter-Andean basins. Precipitation is decreasing with increasing southern latitude.
Figure 2. Study region (a), total precipitation map (b) and diurnal course (c) for an 11-day period during El Niño 1997/98 retrieved with the ACT convective module over Ecuador/Peru.
The diurnal course (Fig. 2c) as a partial result of mesoscale thermal systems (land-sea breeze phenomenon) shows a maximum over land between 1300 and 0100 local time (LT) whereas during night and early morning precipitation dominates over coastal waters. From 0100 to 0700 LT, a clear maximum can be detected over coastal waters as a result of a welldeveloped land-breeze phenomenon which is slightly shifted westwards between 0700 and 1300 LT. During the afternoon, maximum rainfall is observed in the Amazon region and the coastal plains of Ecuador and northern Peru with a small coast-parallel line of reduced precipitation which indicates the divergence area from the back-flowing branch of the sea-breeze system. The great importance of the sea-breeze system on the spatial structure of El Niño can also be observed between 1900 and 0100 LT where
514
C. Reudenbach et al.
only the coastal plains show significantly enhanced precipitation (Bendix et al. 2003).
4 OPERATIONAL RAINFALL RETRIEVAL FOR THE UPPER DANUBE REGION – APPLYING THE ACT CONVECTIVE AND ADVECTIVE MODULES Within the framework of the German programme on global change in the hydrological cycle (GLOWA) the aim of GLOWA-Danube is to investigate new strategies for sustainable water management in the upper Danube catchment as a representative mountain-foreland region in the mid-latitudes (Ludwig et al. 2003). This is done by implementing a coupled webdistributed model system – DANUBIA – which integrates the different models of 16 socio-economic and natural science research groups by stateof-the-art Java network technology (Barth et al. 2003). The task of the authors is to provide reliable information about the spatio-temporal distribution of rainfall as a key input parameter for DANUBIA reference scenarios between 1995 and 2003. Therefore an AtmoSat object was integrated into the Java framework of DANUBIA using the convective and advective module of the ACT to retrieve precipitation from infrared and water vapour Meteosat-VISSR imagery. To ensure comparable time-lines to the old Meteosat system, the microphysics module is not applied to MSGSEVIRI data which is already used for recent retrievals. Two other objects which provide rainfall information as well are integrated in the DANUBIA framework. AtmoStations uses a distance dependent interpolation technique in order to extrapolate rain-gauge data of 250 stations from the German, Austrian and Swiss meteorological services to the 1 km grid of DANUBIA (Mauser 2003b). AtmoMM5 combines rainfall information which is derived from the mesoscale model MM5 (MM5 = Pennsylvania State University/National Center of Atmospheric Research Fifth Generation Mesoscale Model, Grell et al., 1995) and a subsequent downscaling technique that distributes the MM5 40 km grid result to the 1 km² DANUBIA resolution (Früh et al. 2004). Figure 3 shows a comparison of the amount of rainfall for February and July 1999 retrieved from the AtmoSat, AtmoMM5, and AtmoStations modules of DANUBIA. While for February all three models reveal similar patterns with maximum rainfall at the slopes of the northern Alps with only slight variations of ±5% about the common monthly mean rainfall of 127 mm, only AtmoSat and AtmoStations show intense precipitation in the south-eastern part in July which is mainly due to uncertainties in the MM5 convective parameterization scheme. This is also the cause for the almost complete absence of local induced thunderstorms in AtmoMM5 and AtmoStations precipitation pattern which are concentrated in the alpine foreland and
Retrieving Precipitation with Goes, Meteosat and Terra/MSG
515
eastern Bavaria. On the other hand, these convective systems can be clearly identified in the AtmoSat results. This example can only adumbrate the well performance of the satellite retrieval but an intercomparison of the 5-year period between the three rainfall products (not presented in the current paper) showed that the most realistic view on rainfall dynamics in DANUBIA is given by AtmoSat.
Figure 3. Monthly sum of rainfall derived from the AtmoMM5, AtmoSat and AtmoStations model of DANUBIA for the upper Danube catchment for February and July 1999.
Figure 4 presents the annual variation in monthly mean precipitation. Regarding winter, spring and autumn which are usually dominated by advective-stratiform rainfall events, all three techniques provide similar values. However, for mainly convective induced precipitation during summer, there is a clear deviation between AtmoStation/AtmoMM5 on the one and AtmoSat on the other hand. For AtmoMM5, this is due to the uncertainties in the MM5 cumulus parameterization scheme. For AtmoStations there exists the tendency that the spatio-temporal interpolation which is based on a sparse network of rain-gauge stations expands the area of convective precipitation patterns and therefore the mean amount of rainfall.
516
C. Reudenbach et al.
On the other hand, the half-hourly updated satellite data in AtmoSat enables the identification of borders of small convective systems more precisely which implies generally smaller rainfall areas compared to AtmoStations.
Figure 4. Annual course of monthly mean precipitation derived by the AtmoMM5, AtmoSat and AtmoStations model of DANUBIA for the upper Danube catchment in 1999.
5 PRECIPITATION RETRIEVAL FOR THE SEVERE 2002 ELBE FLOOD – INCLUDING THE ACT CLOUD MICROPHYSICS MODULE The severe Elbe flood event in 2002 caused an overall economic loss of about €18.5 billion (Munich Re 2003). The meteorological reason of this hazardous situation were three consecutive events of heavy rainfall. The first period from 1 to 4 August was dominated by heavy convective precipitation induced by regional destabilization due to the high-pressure system Elke over central Europe. From 5 August, Elke declined and the low-pressure systems Hanne and Ilse became decisive for the weather until on 10 August, a cyclone over the Gulf of Genoa started its way on a Vb track towards Poland. Figure 5a shows the complex situation on 5 August with Hanne centred over the north-western Netherlands causing extensive stratiform cloud areas along the occlusion in the north/north-eastern part and shallow convection over the alps indicating it’s a dissolving cold front. Between these frontal regions intensive convection due to a high pressure ridge from France to
Retrieving Precipitation with Goes, Meteosat and Terra/MSG
517
central Germany can be clearly identified in the Terra-MODIS infrared image.
Figure 5. Terra-MODIS data during the Elbe flood for 5 August 2002, 11:05 UTC, showing (a) infrared data overlayed by the synoptic situation, (b) the results of the ACT convective and advective module merged with data from the C-band radar network of the German weather service and (c) the same merge but with the additionally activated ACT microphysics module.
Figure 5b presents an overlay of radar network data (C-band) of the German Weather Service and the retrieval results using only the ACT convective and advective module. Both, the shallow convection along the altering cold front over the Alps/south-western Germany and the convective systems over France and the German mountain foreland are identified correctly as nonraining and rainfall is assigned to the cloud clusters along the border of the high pressure ridge. Nevertheless, only tropopause near areas in the northern stratiform band where identified as raining. This is due to the extreme heterogeneous structure of the occlusion with 3 × 3 infrared standard deviations greater than 1.5 K and TBIR higher than the actual derived threshold of 232.6 K. The potential improvement of the increased spectral resolution of e.g., Meteosat-8 can be seen in Fig. 5c. Here, the ACT microphysics module was
518
C. Reudenbach et al.
activated and applied to Terra-MODIS bands at 0.6, 3.7 and 11 µm together with the convective and advective module. The identified raining area now covers almost the entire northern cloud band and the isolated systems in the centre are also well detected. Hence, the ACT results based on all three modules fits the radar image significantly better than the two-module mode of the ACT as presented in Figure 5b.
6 CONCLUSION The presented method reveals the wide range of applications for rainfall retrievals based on geostationary satellites with high temporal and at least medium spatial resolution. Thereby the modular concept of the ACT ensures interoperability of many existing and forthcoming satellite sensors. While a combination of the convective and advective module retrieves reliable results for the Tropics and Mid-latitudes at least if aggregated over 3–4 h, the microphysics module applicable to, e.g., Meteosat-8 can significantly improve the ACT especially on a single slot basis and for extreme complex synoptic situations with dominant advective dynamics. The results of the ACT three-module mode shows that the derived rainfall structure agrees well with the corresponding radar information. Hence, the presented algorithm can provide global rainfall data in high temporal resolution especially on the basis of second generation geostationary satellites. The global coverage is a clear advantage in comparison to the locally restricted radar networks. Acknowledgements: Parts of the research described in this paper is funded by the German Federal Ministry of Education and Research as part of the German programme on global change in the hydrological cycle (GLOWADANUBE, Grant No. 07 GWK 04). The authors would like to thank EUMETSAT for providing 5 years of Meteosat data, T. Nakajima for supplying the GTR sources, W. Mauser for the AtmoStations results, A. Pfeiffer and H. Schipper for the AtmoMM5 data and the LCRS diploma student H. Scholz for his contribution to the development of the advective retrieval scheme.
7 REFERENCES Barth, M., R. Hennicker, A. Kraus, and M. Ludwig, 2003: An Integrated Simulation System for Global Change Research in the Upper Danube Basin. 1st World Congress on Information Technology in Environmental Engineering, ITTE. Bendix, J., 1997: Adjustment of the Convective-Stratiform Technique (CST) to estimate 1991/93 El Niño rainfall distribution in Ecuador and Peru by means of Meteosat-3 IR data. Int. J. Remote Sens., 18, 1387–1394. Bendix, J., 2000: Precipitation dynamics in Ecuador and Northern Peru during the 1991/92 El Niño – a remote sensing perspective. Int. J. Remote Sens., 21, 533–548.
Retrieving Precipitation with Goes, Meteosat and Terra/MSG
519
Bendix, J., 2002: A satellite-based climatology of fog and low-level stratus in Germany and adjacent areas. Atmos. Res., 64, 3–18. Bendix, J. and A. Bendix, 1998: Climatological Aspects of the 1991/92 El Niño in Ecuador. Bulletin de L’Institut Francaise d’Etudes Andines, 27, 655–666. Bendix, J. and W. Lauer, 1992: Die Niederschlagsjahreszeiten in Ecuador und ihre klimadynamische Interpretation. Erdkunde, 46, 118–134. Bendix, J., S. Gämmerler, C. Reudenbach, and A. Bendix, 2003: A case study on rainfall dynamics during El Niño/La Niña 1997/99 in Ecuador and surrounding areas as inferred from GOES-8 and TRMM-PR observations. Erdkunde, 57, 81–93. Bradley, P. S. and U. M. Fayyad, 1998: Refining Initial Points for K-Means Clustering. In: Shavlik, J. (Edt.): Proc. of the15th International Conf. on Machine Learning; 91–99. Früh, B., J. W. Schipper, A. Pfeiffer, V. Wirth, and J. Egger, 2005: Using mesoscale climate simulations as a predictor for highly resolved precipitation for the us in hydrological models. Quart. J. Roy. Meteor. Soc., submitted. Grell, G., J. Dudhia, and D. Stauffer, 1995: A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5). NCAR/TN 398+STR. Boulder, Colorado, USA: NCAR. Kawamoto, K., T. Nakajima, and T. Y. Nakajima, 2001: A Global Determination of Cloud Microphysics with AVHRR Remote Sensing. J. Climate, 14, 2054–2068. Levizzani, V., 2003: Satellite rainfall estimations: new perspectives for meteorology and climate from the EURAINSAT project. Annal. Geophysics, 46, 363–372. Levizzani, V., J. Schmetz, H. J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2001: Precipitation estimations from geostationary orbit and prospects for METEOSAT Second Generation. Meteorol. Appl., 8, 23–41. Ludwig, R., W. Mauser, S. Niemeyer, A. Colgan, R. Stolz, H. Escher-Vetter, M. Kuhn, M. Reichstein, J. Tenhunen, A. Kraus, M. Ludwig, M. Barth, and R. Hennicker, 2003: Webbased modelling of energy, water and matter fluxes to support decision making in mesoscale catchments – the integrative perspective of GLOWA-Danube. Phys. Chem. Earth, 28, 621–634. Mauser, W., 2003a: GLOWA-Danube: Integrative hydrologische Modellentwicklung zur Entscheidungsunterstützung beim Einzugsgebietsmanagement. Petermanns Geograp. Mitt., 147, 68–75. Mauser, W., 2003b: DANUBIA Software-Documentation. GLOWA-Danube Papers Technical Release No. 3. Munich Re, 2003: Topics 2002. Munich Re, Munich. Nakajima, T. Y. and T. Nakajima, 1995: Wide-area determination of cloud microphysical properties from NOAA AVHRR measurements for FIRE and ASTEX regions. J. Atmos. Sci., 52, 4043–4059. Reudenbach, C., 2003: Convective summer precipitation in Central Europe (in German). Bonner Geogr. Abh., 109, 152 pp. Sankt Augustin. Reudenbach, C., G. Heinemann, E. Heuel, J. Bendix, and M. Winiger, 2001: Investigation of summertime convective rainfall in Western Europe based on a synergy of remote sensing data and numerical models. Meteor. Atmos. Phys., 76, 23–41. Schipper, J. W., 2005: Sensitivity of MM5 precipitation to various configurations. Mon. Wea. Rev., submitted. Tjemkes, S. A., L. van de Berg, and J. Schmetz, 1997: Warm water vapour pixels over high clouds as observed by METEOSAT. Contr. Atmos. Phys., 70, 15–21. Xue, M., D.-H. Wang, J.-D. Gao, K. Brewster, and K. K. Droegemeier 2003: The Advanced Regional Prediction System (ARPS), storm-scale numerical weather prediction and data assimilation. Meteor. Atmos. Phys., 82, 139–170. Zock, A., G. Menz, and M. Winger 1995: Regionalisation of rainfall models in Eastern Africa using Meteosat Real-Time-Window data. Proceedings of the International Geoscience Remote Sensing Symposium (IGARSS’95), Florence, Italy (New York: I.E.E.E.); 250–252.
41 MODEL AND SATELLITE ANALYSIS OF THE NOVEMBER 9–10, 2001 ALGERIA FLOOD Carlo M. Medaglia1, Sabrina Pinori1,2, Claudia Adamo1, Stefano Dietrich1, Sabatino Di Michele1, Federico Fierli1, Alberto Mugnai1, Eric A. Smith3, and Gregory J. Tripoli2 1
Istituto di Scienze dell’Atmosfera e del Clima, Consiglio Nazionale delle Ricerche, Roma, Italy 2 Department of Atmospheric and Oceanic Sciences, University of Wisconsin, Madison, WI, USA 3 NASA/Goddard Space Flight Center, Greenbelt, MD, USA
Abstract
In this paper, we present the results of a numerical simulation study of the development of a devastating cyclonic storm that struck the Algerian coast on November 9–10, 2001, with over 200 mm of rainfall. The highresolution numerical simulation of the storm suggests that the cyclone was spawned by when a filament of potential vorticity shed from an intense tropopause fold balled up into an intense local maximum, just north of Algiers. The storm developed west of a frontal occlusion, similar to the classical formation of a polar low. Similarities include: (a) the preexistence of a major trough, (b) the occlusion of the frontal cyclone, (c) the isolation of the warm core low west of the frontal fracture, and d) the growth of the warm core vortex. Strong slantwise neutral uplift over the occluded front was instrumental in producing the heavy rains that affected the city. The numerical simulation is also used to perform more accurate rainfall rate estimates based on available TMI measurements.
1 INTRODUCTION In the Mediterranean Basin, cyclones grow mainly because of baroclinic instability (Holton 1992), which requires horizontal temperature gradients and vertical wind shear. Moreover, the interaction of a large-scale baroclinic wave with an orographic obstacle like the Alps, by virtue of the conservation of potential vorticity (PV), is the cause of a smaller-scale, orographically 521 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 521–534. © 2007 Springer.
522
C. M. Medaglia et al.
induced baroclinic lee cyclone (BLC), that is generated to the lee of the Alps in response to a larger-scale cyclone over the Atlantic. Mediterranean BLC’s have been studied in the last few years from observational, theoretical, and modeling perspectives (e.g., Romero et al. 1997, 2000; Doswell III et al. 1998; Homar et al. 1999; Campins et al. 2000). The interaction between orography and synoptic fluxes, even if applicable to a large fraction of Mediterranean cyclones, does not completely explain all the different types of events that occur over the Mediterranean Basin (Tudurì and Ramis 1997). In fact, at almost the same time as the Mediterranean BLC theory was being formalized, a different kind of subsynoptic cyclone, with similarities to tropical cyclones and polar lows, was observed. These systems are typically about 200 km in diameter and appear as small commashaped or near-circular cloud patterns, often with a clear eye at the center, and in general they are also connected to high PV anomalies that initiate the cyclonic disturbance. Some cases of such structures above the Mediterranean Sea were reported by Tsidulko and Alpert (2001). These strongly convective cyclones cannot be explained by dry baroclinic instability alone: latent heat release (Businger and Reed 1989) and sea interaction play a major role in their development. More recently, Rossa et al. (2000) have studied the evolution of a PV column formed by the concurrence of a PV anomaly in the upper troposphere with a cyclonic circulation on the ground. It appears that the PV positive anomaly maintains the cyclone, by feeding it from the higher layer – as already proposed by Hoskins et al. (1985), who suggested that a high PV anomaly induces a circulation cell of ascending motion eastward of the ridge and descending westward. In this paper, we use model simulations in conjunction with satellite observations to perform a detailed analysis of the processes leading to the devastating cyclonic storm that struck the Algerian coast on November 9–10, 2001 producing more than 200 mm of rainfall and winds of 33 m s–1 – see also Tripoli et al. (2004). In particular, this study focuses on the upper levels precursors and on the sea–air interaction to better understand the exact timing and location of upper-level PV coupling and associated low-level cyclogenesis.
2 SYNOPTIC DESCRIPTION An initial baroclinic instability associated with a large-scale tropopause fold over Western Europe was at the origin of the heavy rains that affected northwestern Algeria on November 9–10, 2001 and then the Balearic Islands. The meteorological situation was characterized by an infiltration of stratospheric air over central Europe, the Iberian peninsula and then the Gibraltar Gulf. As shown in Fig. 1, a high-pressure area centered west of Ireland and a lowpressure one centered on the Alpine region dominated the surface condition on November 9. A cold front was also present in southwestern Europe.
Model and Satellite Analysis of the Algeria Flood
523
At the same time, the 500 hPa situation showed a very different situation with an area of low pressure extending from the Scandinavian area to Gibraltar, where there were strong winds with velocity reaching 150 Kts (at 300 hPa).
Figure 1. Meteorological situation on November 9, 2001: UK Met Office surface analysis (left) and NCEP 500 hPa geopotential height (right).
On November 10, the upper level situation was the same, characterized by an occlusion of the low pressure area over Spain and North Africa. Nevertheless, the surface situation was different: a lowering of the pressure field occurred over the Algerian coasts within the lower levels of the atmosphere (850 and 700 hPa), creating a depression that evolved from southwest Algeria towards the north. Then this depression developed into a cyclone, which moved towards the Balearic Islands while growing slightly, and maintaining a perturbed flow on northwestern Algeria. The frontal situation occurring during the first hours of November 9 was suddenly modified by the injection of cold and dry air coming from higher latitudes, making the surface temperature in Southern Spain and Algeria drop 10 K at midday. This is demonstrated by Fig. 2 that shows the presence of a polar filament carrying a deep layer of stratospheric-originated cold air down into the troposphere – here, the polar filament is evidenced by the (dark) low water vapor (WV) values over Central Europe, Spain and Northwestern Africa in the METEOSAT image, as well as by the corresponding low values of columnar ozone as measured by the Total Ozone Mapping Spectrometer (TOMS) onboard the Advanced Earth Observing Satellite (ADEOS) – see Kramer (2002). Due to the presence of this cold air mass, the surface pressure rapidly dropped down, creating a vortex over the Algerian coast and allowing the formation of a cyclone during the day of November 10. On both days, extraordinary rainfalls of 120–140 mm in 12 h were reported, which led to the flooding disaster in Algeria. Then, during the night of November 10–11, this cyclone moved from the Algerian coast towards the Balearic Islands.
524
C. M. Medaglia et al.
Noteworthy, the low-level circulation developed without a correspondent low-level cold temperature field until late in the period, when a pool of cold air sat over the low-pressure surface – which is in marked contrast with developments in which surface baroclinicity is crucial (Hoskins et al. 1985).
Figure 2. Satellite observations for November 10, 2001: METEOSAT WV channel at 1200 UTC (left) and TOMS total ozone representative for the whole day (right).
3 SIMULATION OF THE EVENT We have performed a cloud resolving simulation of the event using the University of Wisconsin–Nonhydrostatic Modeling System (UW-NMS) developed by Tripoli (1992). The UW-NMS model is based on the nonBoussinesq quasi compressible dynamical equations, and employs a twoway multiply nested Arakawa “C” grid system – see Tripoli (1992), Tripoli and Cotton (1982, 1986), and Flatau et al. (1989) for a detailed description of the numerical integration scheme, of the transport equations, and of the microphysics and dynamic module of the model. To simulate this event, three nested grids were used: the first (120 × 90 grid points) with 37.5 km resolution over a large region spanning much of Europe and North Africa; the second (240 × 180 grid points) with 9.4 km resolution, covering the western Mediterranean Basin; and the finest one (160 × 96 grid points) with 2.4 km resolution, covering the region just around Algiers. The high-resolution grid was used only during the first 60 h to capture the storm around the city of Algiers. The model was initiated at 1200 UTC of November 8 from the NCEP AVN analysis and integrated for a period of 72 h. Nevertheless, different types of initialization were performed to evaluate the sensitivity of the analysis to different initial and boundary conditions – for instance, to evaluate the relative roles of the orography and sea surface latent heat flux versus the action of the upper-level PV centers.
Model and Satellite Analysis of the Algeria Flood
525
Figure 3. Top: Accumulated rainfall prediction over the 72 h simulation period; Bottom: accumulated rainfall measurements (mm) on the Algerian coast from November 9, 0600 UTC, till November 11, 1800 UTC (from the Algerian Meteorological Office).
The model simulation captured the flooding precipitation and reported winds of the event well, however, the heavy precipitation maximum over the city of Algiers was simulated to be slightly less than observations (see Fig. 3). This was not surprising, since convective storms not anchored to topography were the primary mechanisms for the heaviest precipitation. The simulated wind was more precise, as the simulated peak gust of 33 m s–1 near Algiers exactly matched reported peak gusts. The Algerian cyclogenesis process and its predictability had much in common with the “Storm of the Century” development on March 12, 1993 off the Texas coast reported by Bosart et al. (1996). Employing “PV thinking” (Hoskins et al. 1985) to examine the synoptic forcing, one finds that an exceptionally strong south–westward digging trough created an
526
C. M. Medaglia et al.
uncommonly deep tropopause fold in the days prior to the storm, forcing stratospheric PV values down to 650 hPa pressure (see Fig. 4).
Figure 4. Simulation of PV at 500 hPa (below 8 km) for November 10 1800 UTC.
Figure 5. View of simulated low level flow vectors on grid 2 (9.4 km resolution) at 10 m MSL for 1000 UTC, November 10, 2001. Topography is color shaded in θe (from 290 K in blue to 305 K in red), while mean sea level pressure isobars at 2 hPa intervals are also shown. Channeling around topography and transport of warm θe from Wind Induced Surface Heat Exchange (WISHE) are highlighted.
Model and Satellite Analysis of the Algeria Flood
527
The evolution and coupling of the surface air stream to the PV filament aloft resulted in the surface storm cyclogenesis. Figure 5 depicts the surface frontal evolution as the system moved south off of the coasts of Spain and France on the day prior to the storm genesis. Particularly interesting was the extended influence of the upstream mountain ranges in channeling the cold air movement. As the cold surges reached the southern Mediterranean shores, an easterly low-level barrier jet formed the next day north of Tunisia and Algeria. The low-level jet flowed westward into the developing upward ageostrophic circulation aloft forming in advance of the approaching upper level PV anomaly from the southwest. The strong shear across the contrasting air masses along the barrier jet resulted in a vortex sheet that would be the nucleus of future development, when convection produced strong local convergence maximum along the line. During the 24 h before the cyclogenesis, the wind induced surface heat exchange (WISHE) along the barrier jets increased the boundary layer equivalent potential temperature (θe) to critical convection levels. For coupling between the warm surface θe and PV aloft to occur, deep moist convection had to develop under the PV filament aloft. As shown in Fig. 6, simulated deep convection was initiated off shore within enhanced surface convergence formed beneath the left exit region of the upper level jet streak that was approaching from the southwest. The coupling of the surface to the forced warm θe filament aloft is strikingly evident.
Figure 6. View from the south of the simulated Algiers cyclone at 1000 UTC, November 9, 2001 on grid 2. Topography is color shaded according to the equivalent potential temperature (θe). Isobars of mean sea-level pressure are reported (white) at 1 hPa intervals.
A sensitivity experiment that was run without the Atlas Mountains showed that similar cyclogenesis did not occur without the presence of the mountains, suggesting the critical role of orography in inducing the mesoscale cyclogenesis process (see Fig. 7). Examination of the barrier jet
528
C. M. Medaglia et al.
depicted at this time on the fine grid suggests that cooler θe air sweeping of the African continent from the south, created the effect of a warm occlusion along the barrier jet, and even rotated the barrier confluence zone northward. Notably, the moist airstreams forced into the Atlas Mountains southwest of the upper level ageostrophic forcing shown in Fig. 5 did not result in deep convective storms, suggesting that there was a critical interaction between the ageostrophic forcing aloft and the orographically enhanced circulation below to create the deep convective plumes.
Figure 7. NMS 12 h accumulated precipitation for November 10, 1100 UTC: results from the runs with and without the Atlas Mountains (left and right panels), respectively. The isobars of MSPL at 4 hPa intervals are shown as while lines.
Immediately following the eruption of deep convection, coupling of the surface flow with the PV filament above took place. The surface vorticity began to strengthen dramatically, accelerating to levels in excess of 500 10–5 s–1. This appeared to be in response to the convergence of angular momentum into the localized regions of convective lifting. In effect, there was a “balling up” of the aforementioned vortex sheet (Jascourt 1997). A surface low formed by 0300 UTC and deepened to 989 hPa by 0830 UTC. Northerly near surface winds exceeding 37 m s–1 were occurring over the water on the western flank of the vortex. Convection concentrated the mesoα-scale ageostrophic convergence into a convective band 55 km across. The meso-β-scale vortex remained stationary along the shoreline for over 9 h, perhaps in response to conservation of potential vorticity in the downsloping flow on the storm’s eastern flank. This drove winds of 33 m s–1 and heavy rains at Algiers. The no-mountain sensitivity experiment seems to confirm the role of the orography in this scenario. During the intense stationary phase of the storm on the afternoon and evening of November 10, the warm θe anomaly at the storm’s surface inflow increased by 2 K due to strong WISHE forcing. The storm developed a 2 K weak warm core
Model and Satellite Analysis of the Algeria Flood
529
anomaly aloft, only about 60 km in diameter. This appeared as a shallow thermal perturbation within an otherwise much larger cold core structure. To the west, the strong north-easterly low-level air stream, that originated as the the barrier convergence zone and then spiraled cyclonically into the mesoscale vortex, ultimately decoupled from the vortex as the vortex became stationary, then drifted westward driving strong winds into the Atlas Mountains and depositing copious amounts of precipitation west of Algiers. The meso-β-scale vortex accelerated its movement southward by 2000 UTC on the 10th, apparently as the ageostrophic forcing aloft moved northeastward and decoupled. Immediately, surface pressures rose rapidly and the storm dissipated completely by 0000 UTC, November 11. In the meantime, a meso-α-scale cyclone began to strengthen to the north and east (see Fig. 8).
Figure 8. Surface equivalent potential temperature θe on November 11 0900 UTC.
Two additional sensitivity tests featuring (a) no latent heating and (b) no land surface over Africa, lead to the conclusion that the Atlas mountains and
530
C. M. Medaglia et al.
the effects of deep moist convection spawned by flow regimes created from the Atlas mountains were responsible for the meso-β-scale storm that affected Algiers (see Fig. 9). These studies demonstrate that the large meso-α-scale cyclone would have taken place anyway late on November 10 in direct response to the upper level PV forcing, even without the convective induced coupling, and would have affected the Mediterranean region to the north, but would have affected Algiers less severely. Indeed, this was already taking place in the simulation before the local orographically generated flow systems shortcircuited that scenario and plugged the upper level storm into the surface WISHE forcing via deep moist convection.
Figure 9. NMS 12 h accumulated precipitation for the “no Africa land “run for November 10, at 1100 UTC (left) and at 2400 UTC (right). The isobars of MSPL at 4 hPa intervals are shown as white lines.
4 PRECIPITATION MEASUREMENTS FROM TRMM The lack of ground radar coverage over Northern Africa and the scarcity of rain gauges make the precipitation measurements derived from satellite sensors especially important for this event. In another paper in this book, Torricella et al. (2007) analyze the results obtained using the combined microwave/infrared (MW/IR) method developed by Turk (see Turk et al. 2000). Here, we use measurements taken by the Precipitation Radar (PR) embarked on the Tropical Rainfall Measuring Mission (TRMM) satellite to test the corresponding rainfall rate estimates based on observations taken by the TRMM Microwave Imager (TMI) (see Kummerow et al. 1998 for a description of the TRMM sensor package).
Model and Satellite Analysis of the Algeria Flood
531
Figure 10. Rainfall rates measured by the TRMM PR on November 10, 2001 at 0205 UTC (left) and corresponding rainfall rates estimated by the GPROF algorithm from TMI observations for the same overpass (right). Values in mm h–1.
Figure 11. Rainfall rates (mm h–1) estimated by the BAMPR algorithm from TMI observations for the same overpass of Fig. 10. The right panel is a zoom for the northern Algeria-Morocco border region.
Results are shown in Fig. 10 for the Goddard Profiling Algorithm (GPROF) (Kummerow et al. 2001), which is the standard TMI rainfall algorithm, and in Fig. 11 for the Bayesian Algorithm for Microwave-based Precipitation Retrieval (BAMPR) (Mugnai et al. 2001; Di Michele et al. 2003, 2005). It is evident that the GPROF estimates are quite unsatisfactory – especially over land. While this may be partially due to an inadequate screening procedure, we notice that the GPROF cloud-radiation database is not specifically tailored for the Mediterranean area and it is based on events having different microphysical characteristics than the specific Algerian event. On the other hand, the BAMPR estimates match pretty well the PR measurements, both over land and sea. Apart from specific differences between the two algorithms, we believe that the main reason why the BAMPR algorithm works better is that it uses a cloud-radiation database
532
C. M. Medaglia et al.
which has been specifically built for the Mediterranean Basin within the EURAINSAT project (Tassa et al. 2003a, b). This database is composed by cloud and precipitation microphysical profiles, that have been generated by means of the UW-NMS model for several heavy precipitation events over the Mediterranean area, including the present Algerian storm. As a matter of fact, about 40% of the retrieved profiles belong to the Algerian simulation, which demonstrates the importance of having cloud-radiation databases that include the class of events under observation. At the same time, these results provide an indirect validation of the model microphysics.
5 CONCLUSIONS This study demonstrates that exact timing and location of upper level PV coupling and associated surface cyclogenesis, as described by Bosart et al. (1996), can be significantly altered by locally generated and unbalanced flow systems and driven by convection forcing. Fortunately, as is often the case for Mediterranean storms, these local flow systems seem to be well behaved and highly predictable given a model’s ability to finely resolve surface topographical and coastline features and skillfully simulate their interaction with the broader scale flow. Acknowledgments: This study has been funded by the Italian National Group for Prevention from Hydro-Geological Disasters (GNDCI), by the Italian Space Agency (ASI), and within the frame of EURAINSAT – a shared-cost project (contract EVG1-2000-00030) co-funded by the Research DG of the European Commission (5th Framework Program).
6 REFERENCES Bosart, L. F., G. J. Hakim, K. R. Tyle, M. A. Bedrick, W. E. Bracken, M. J. Dickinson, and D. M. Schultz, 1996: Large-scale antecedent conditions associated with the 12–14 March 1993 cyclone (Superstorm ’93) over Eastern North America. Mon. Wea. Rev., 124, 1865–1891. Businger, S. and R. J. Reed, 1989: Cyclogenesis in cold air masses. Wea. Forecasting, 4, 133–156. Campins, J., A. Genovès, A. Jansà, J. Guijarro, and C. Ramis, 2000: A catalogue and a classification of surface cyclones for the western Mediterranean. Int. J. Clim., 20, 969–984. Di Michele, S., F. S. Marzano, A. Mugnai, A. Tassa, and J. P. V. Poiares Baptista, 2003: Physically-based statistical integration of TRMM microwave measurements for precipitation profiling. Radio Sci., 38, 8072. Di Michele, S., A. Tassa, A. Mugnai, F. S. Marzano, P. Bauer, and J. P. V. Poiares Baptista, 2005: Bayesian algorithm for microwave-based precipitation retrieval: description and application to TMI measurements over ocean. IEEE Trans. Geosci. Remote Sens., 43, 778–791. Doswell III, C., C. Ramis, R. Romero, and S. Alonso, 1998: A diagnostic study of three heavy precipitation episodes in the western Mediterranean. Wea. Forecasting, 13, 102–124.
Model and Satellite Analysis of the Algeria Flood
533
Flatau, P., G. J. Tripoli, J. Berline, and W. Cotton, 1989: The CSU RAMS Cloud Microphysics Module: General Theory and Code Documentation. Technical Report 451, Colorado State University, 88 pp. Holton, J. E., 1992: An Introduction to Dynamic Meteorology. Academic Press, 319 pp. Homar, V., C. Ramis, R. Romero, S. Alonso, J. Garcìa-Moya, and M. Alarcòn, 1999: A case of convection development over the western Mediterranean sea: A study through numerical simulations. Meteor. Atmos. Phys., 71, 169–188. Hoskins, B. J., M. E. McIntyre, and A. W. Robertson, 1985: On the use and significance of isentropic potential vorticity maps. Quart. J. Roy. Meteor. Soc., 111, 877–946. Jascourt, S., 1997: Convective organizing and upscale development processes explored through idealized numerical experiments. PhD Thesis, University of Wisconsin–Madison, Madison, WI 53706, 267 pp. Kramer, H. J., 2002: Observation of the Earth and its Environment: Survey of Missions and Sensors. Springer-Verlag, 1510 pp. Kummerow, C. D., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15, 808–816. Kummerow, C. D., D. B. Shin, Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. Mugnai, A., S. Di Michele, F. S. Marzano, and A. Tassa, 2001: Cloud-model-based Bayesian techniques for precipitation profile retrieval from TRMM microwave sensors. Proc. ECMWF/EuroTRMM Workshop on Assimilation of Clouds and Precipitation, ECMWF, Reading, UK, 323–345. Romero, R., C. Ramis, and S. Alonso, 1997: Numerical simulation of an extreme rainfall event in Catalonia: Role of orography and evaporation from the sea. Quart. J. Roy. Meteor. Soc., 123, 537–559. Romero, R., C. A. Doswell III, and C. Ramis, 2000: Mesoscale numerical study of two cases of long-lived quasistationary convective systems over eastern Spain. Mon. Wea. Rev., 128, 3731–3751. Rossa, A. M., H. Wernli, and H. C. Davies, 2000: Growth and decay of an extra-tropical cyclone’s PV-tower. Meteor. Atmos. Phys., 73, 139–156. Tassa, A., S. Dietrich, S. Di Michele, S. Pinori, and A. Mugnai, 2003a: The EURAINSAT Cloud Radiative Dataset. EURAINSAT Technical Report Series, 1, 32 pp. Tassa, A., S. Di Michele, A. Mugnai, F. S. Marzano, and J. P. V. Poiares Baptista, 2003b: Cloud-model-based Bayesian technique for precipitation profile retrieval from TRMM Microwave Imager. Radio Sci., 38, 8074. Torricella, F., V. Levizzani, and F. J. Turk, 2007: Application of blended MW-IR rainfall algorithm to the Mediterranean. Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F.J. Turk, eds., Springer, 497–508. Tripoli, G. J. and W. R. Cotton, 1982: The Colorado State University three dimensional cloud/mesoscale model. Part I: General theoretical framework and sensitivity experiments. J. Rech. Atmos., 16, 185–200. Tripoli, G. J. and W. R. Cotton, 1986: An intense, quasi-steady thunderstorm over mountainous terrain. Part IV: Three-dimensional numerical simulation. J. Atmos. Sci., 43, 894–912. Tripoli, G. J., 1992: A nonhydrostatic mesoscale model designed to simulate scale interaction. Mon. Wea. Rev., 120, 1342–1359. Tripoli, G. J., S. Pinori, S. Dietrich, C. M. Medaglia, G. Panegrossi, A. Mugnai, and E. A. Smith, 2005: The 9–10 November 2001 Algerian flood: A numerical study. Bull. Amer. Meteor. Soc., 86, 1229–1235.
534
C. M. Medaglia et al.
Tsidulko, M. and P. Alpert, 2001: Synergism of upper-level potential vorticity and mountains in genoa lee cyclogenesis: A numerical study. Meteor. Atmos. Phys., 78, 261–285. Tudurì, E. and C. Ramis, 1997: The environments of significant convective events in the western Mediterranean. Wea. Forecasting, 12, 294–306. Turk, F. J., G. D. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000: Meteorological applications of precipitation estimation from combined SSM/I, TRMM and infrared geostationary satellite data. In: Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia eds., VSP, Utrecht, The Netherlands, pp. 353–363.
42 MODELING MICROPHYSICAL SIGNATURES OF EXTREME EVENTS IN THE WESTERN MEDITERRANEAN TO PROVIDE A BASIS FOR DIAGNOSING PRECIPITATION FROM SPACE Gregory J. Tripoli1, Carlo M. Medaglia2, Giulia Panegrossi1, Stefano Dietrich2, Alberto Mugnai2, and Eric A. Smith3 1
Department of Atmospheric and Oceanic Sciences, University of Wisconsin, Madison, WI, USA 2 Institute of Atmospheric Sciences and Climate, CNR, Rome, Italy 3 NASA/Goddard Space Flight Center, Greenbelt, MD, USA
1 INTRODUCTION Precipitation occurring over the Mediterranean basin is typically unusually difficult to measure due to variability resulting from the irregular terrain. Radar based precipitation measurements also are compromised by the terrain, which contaminates reflectivity with excessive ground cover and blocks the radar beam at low elevation angles. That same terrain also is instrumental in providing for the localization of rain producing storms that results in long term and flash flooding situations, especially in the fall season. These limitations are presumably overcome by precipitation measurements taken remotely from space based observing platforms. From space, instruments directly measure upwelling electromagnetic radiation. Radiation in the microwave bands are actively scattered, absorbed, and reflected by liquid and ice condensate present in the air. Observations of variability in the amount of microwave radiation reaching space are therefore tied to the structure of the liquid and ice present in the atmosphere from which the radiation originates. It is known that surface precipitation rate is also related to the vertical structure of the liquid and ice present. The question arises: “How unique are these two relationships?” and “Can we
535 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 535–547. © 2007 Springer.
536
G. J. Tripoli et al.
infer surface precipitation rate from the measurement of the intensity of radiation from a limited number of microwave radiance measurements?” Typically space-borne microwave radiometers (like ones on board of TRMM and DSMP satellites) measure intensity of 4–6 carefully considered microwave frequencies (Spencer et al. 1989; Kummerow et al. 1998). This measurement would appear to have an insufficient number of degrees of freedom to distinguish the many complex vertical condensate structures that might evolve. However, we also understand that not all precipitation structures that relate to the observed radiance profile are actually likely to occur. The hypothesis of precipitation retrieval hinges on this observation and so that cloud structure and attendant precipitation can be inferred given knowledge of what cloud structures and attendant surface precipitation rate are most likely to lead to a given set of microwave radiance observations (Wilheit 1986).
2 USING CLOUD RESOLVING MODELS TO INTERPRET SATELLITE OBSERVATIONS The cloud structures that are most likely associated with a given radiance observation depend on: 1. Geographical location. 2. Season. 3. Basic dynamic mechanism of the storm being measured, i.e., vertical convective, slantwise convective or stratiform. 4. The location of the measurement within the storm, i.e. (anvil, convective core, warm front, cold front). We can determine those relationships through numerical cloud resolved models (CRMs) of “typical” precipitating storms (see Fig. 1). CRMs can simulate a cloud or systems of clouds through explicit simulation of the flow dynamics and the attendant microphysics and precipitation evolution. The result is a physically consistent prediction of the dynamics, microphysics and thermodynamics of the precipitating system. We can then apply a Passive Radiation Model (PRM) to the simulated atmospheric structure of the cloud to simulate the upwelling radiation that a satellite would observe. The model simulations of dynamics, microphysics and radiance are verified against conventional and special observations through Cloud Radiation Verification Studies (CRVSs). CRVSs can also be used to improve the model physics through these comparisons (Panegrossi 2004). The CRM/PRM combination is a Cloud Radiation Simulation (CRS) and results in a gridded data set containing the atmospheric dynamic, thermodynamic and microphysical structure, the surface precipitation rate, and the attendant upwelling microwave radiation. A database composed of a wide variety of CRS results is called a Cloud Radiation Data Base (CRDB).
Microphysical Signatures of Extreme Events in the Western Mediterranean
537
A CRDB then is used as a statistical basis from which one can retrieve precipitation from a space-based radiance observation. The analysis performed in the PRM is diagramed in Fig. 2. After choosing an “appropriate” model simulated vertical profile containing model simulated state parameters and microphysics, the upwelling microwave radiation must be calculated. This is accomplished through a PRM that calculates model inferred brightness temperature according to the particular microwave frequencies and path geometry viewed by the satellite.
Figure 1. Schematic describing the relationship between Cloud Radiation Simulation (CRS), space-based microwave observations, and precipitation measurement.
Figure 2. Cloud Radiation Database Generation Flow Diagram.
538
G. J. Tripoli et al.
Using the CRDB as a statistical reference from which to base retrieval has several important sources of error. First, there are errors in the model simulation itself including errors from the model’s inability to simulate a weather system dynamics or the microphysics of precipitation formation. Precipitation simulation is especially important since there are large sensitivities of the simulated upwelling radiation to the predicted hydrometeor size, habit and phase(s). Studies comparing simulated microphysics to actual cloud observations have shown mixed results. Overall the performance of bulk microphysics models in properly simulating microphysics observed in situ can be characterized as weak. However, there may be more realism in the simulation of the overall relationship between microphysics and precipitation rate, even if the microphysics simulated for a particular storm is poor. Fortunately, based on experiences comparing radiance implied by the simulated microphysics to satellite observations, the simulated microphysics occurring in Mediterranean storms appears to be less likely to produce unrealistic satellite signatures than other storms around the world that we have attempted. This is likely because of microphysical similarity between Mediterranean storms that allows a “tuned” bulk parameterization to perform well over a wide range of cloud applications and the predominance of liquid rain in most heaving precipitation situations.
Figure 3. Illustration depicting the challenges of “representativeness” for a mesoscale convective system containing a mixture of slantwise updrafts and downdrafts, vertical convective plumes and suspended anvil ice. The upwelling ration reaching the space borne sensor is highly dependent on the angle of view and the exit point of radiation ray.
Microphysical Signatures of Extreme Events in the Western Mediterranean
539
A second major source of error is in the PRM. Perhaps because of the precipitation is dominated by rain in most cases, the radiative transfer is straightforward and has minimal error. One would expect the light snow situations to be the most challenging, but they are relatively rare except in the very high terrain. A major source of error encountered when attempting to base the relationship between precipitation rate and radiance on entries from a CRDB is the representativeness of the entries employed (see Fig. 3). Representativeness is affected by the angle of the observation and whether the radiative path goes through a horizontally uniform region of the cloud or perhaps passes through differing dynamical regions of the clouds. How can one find a representative vertical atmospheric/radiation profile among the seemingly infinite possibilities of cloud samplings possible even from just a few model runs? This challenge becomes even greater as the resolution of the radiance observation increases and smaller parts of a storm are sampled by the satellite at a given time. The Mediterranean again provides a more forgiving environment for finding representative entries because the precipitation mechanisms tend to conform to topographical forcing and be less variable spatially and over different cases. This will be discussed more below. Finding an appropriate CRDB entry to reference for a precipitation retrieval calculation has been a continuing challenge since no two storms are exactly alike, a CRS is far from perfect, and only a limited number of entries are feasible either to create or store in the CRDB. Hence there is not only the question of which simulated radiance profiles of the CRDB are most appropriate, but also the question of whether any of the profiles are appropriate.
Figure 4. Mediterranean basin depicting mountainous terrain surrounding the sea.
540
G. J. Tripoli et al.
3 APPLICATION TO THE MEDITERRANEAN BASIN In view of these challenges, we propose the hypothesis that CRSs performed for the Mediterranean basin tend to be exceptionally well suited as a statistical basis for retrieval compared to many locations on the globe. This is because of the unique character of precipitating systems in that region that result in an increased robustness in structure. Over the last decade we have performed a number of CRSs of Fall season heavily precipitating storms in the Mediterranean and in each case, the results support our hypothesis.
3.1 Characteristic of Mediterranean precipitating storm Below we summarize what we have found about the uniqueness of the Mediterranean basin for the simulation and prediction of precipitation: 1. Local origins of atmospheric vapor: The Mediterranean basin (see Fig. 4) is characterized by a warm body of water ringed by severe topography on all sides. With the hot dry desert to the south and east, a cool Atlantic Ocean to the west, and a large land mass to the north, much of the moisture falling as precipitation in the basin evaporates from the Mediterranean Sea itself. 2. Since the sea surface temperature is well observed in the Mediterranean, atmospheric models can reasonably estimate surface water vapor and thermal fluxes from the sea surface. 3. Orographic and land use influences on convection initiation: The prediction of convection initiation is perhaps the greatest problem facing quantitative precipitation forecasting (QPF). Where the first convective towers occur is difficult to predict and in most places in the world, depend on initial placement of local density currents and boundaries of various types perhaps left over from previous storms, fronts land use variations, and topography. In the Mediterranean region, topographical forcing together with the land/sea effect dominates initiation (see Fig. 5) and because their influence is represented well in a high resolution model, convective initiation in the Mediterranean tends to be unusually predictable. 4. Orographic influences on stable precipitation: Stable rain is created by forced lifting by non-buoyant rising currents. These rising currents are formed primarily by baroclinic structures featuring flow-up slanted isentropic surfaces (surfaces of constant entropy or potential temperature) or flow up a topographical surface. Stable baroclinic rains are reasonably well predicted, but may have significant error in precise placement on the mesoscale simply due to a small relative error in the placement of the synoptic system. Precipitation tied to topographic lifting, however, is much less sensitive to the error in the large scale system prediction
Microphysical Signatures of Extreme Events in the Western Mediterranean
541
because the topography is well represented in a model at all times and the local slope geometries further localize the most intense lifting.
Figure 5. Schematic depicting the roles of topographic initiation, the EML, and WISHE in focusing and triggering Mediterranean convection.
5. Elevated mixed layer (EML): Elevated mixed layers are well known to provide for the capping of conditionally unstable boundary layers from the formation of unforced deep moist convection, allowing instabilities to build to high levels. Central North America is a good example of where EMLs formed over the Rockies move eastward over the High Plains and allow for extreme instabilities to build as Gulf air flows northward under the EML. The Mediterranean region also can feature an EML (see Fig. 5), especially in the fall season. At that time of the year, the Sahara is still very hot and features a deep dry Desert Boundary Layer (DBL), extending upward to 2–4 km above the surface. As a middle latitude upper level trough approaches the Mediterranean region from the northwest, the ageostrophic Sawyer-Elliasen circulation (Carlson 1991) will result in a surge of warm surface air from the south to the north. The strong thermal gradient between the cool air to the north and the hot desert air to the south will provide additional energy for and so enhance this circulation. As the warm surge moves offshore over the relatively cool Mediterranean Sea, a cool, moist internal Marine Boundary Layer (MBL) is developed adjacent to the surface. Continued Wind Induced Surface Heat Exchange (WISHE) (Yano and Emanuel 1991) occurs as the air flow northward, building high Convective Available Potential
542
G. J. Tripoli et al.
(CAPE) that is capped by the strong EML. Generally, under these circumstances, no significant convection can be triggered until the flow is pushed up a significant topographical barrier in the northern Mediterranean. This all works to further focus the areas of precipitation into predictable locations associated with the topography and land sea boundary. 6. Low Froude number flow and topographic channeling: The Froude number is defined to be the ratio of the inertial accelerations producing vertical lifting over a barrier to static stability induced decelerations to that lifting. It is expressed as: Fr = u/hN, where u is the velocity, N is the Brunt Vaisala frequency and h is the barrier height. The strong capping of the EML increases N within the southerly surge and so decreases Fr. This causes southerly low level flow to resist being drawn over topographical barriers and causes the flow to be diverted around the barriers instead. As a result, flow which is approximately southerly, tends to follow channels between topographical barriers such as the Atlas mountains, Sicily, Sardinia, Corsica, the Italian Peninsula, Crete, Cyprus, and so on (see Fig. 6). The flow also tends to be diverted around significant mountain ranges on the north shore, such as the Mediterranean Alps near Nice, France, because of the strong barrier effect. As a result, the eventual inevitable lifting tends to be focused into highly predictable locations such as Genoa, Italy, that may not contain the highest topography, but where lifting is strongest. There convection is triggered or else stable orographic rains are forced.
Figure 6. Channeling of flow focusing on Genoa Region. Flow is focused into Genoa by a combination of topographical channeling and the barrier effect of the Alps.
Microphysical Signatures of Extreme Events in the Western Mediterranean
543
7. Downslope wind storms: Mountain waves and downslope wind storms resulting from mountain waves can occur on the lee side of mountain ranges under certain atmospheric conditions depending primarily on the vertical shear and stability profile (Tudurì and Ramis 1997). These wind events can be simulated with a competent mesoscale model with a high degree of skill. Moreover, these events produce modulating effects on the initiation and maintenance of precipitating storms and even on the magnitude of surface fluxes. Our simulations of a strong precipitating storm in Friuli, Italy (Tripoli et al. 2000) was one case where the western extent of the heavy rains and the width of the Adriatic channel jet were modulated by the formation of downslope winds off the Apennine mountains (see Fig. 7). Because of their predictability, downslope wind storms also lend a degree of predictability to precipitating systems.
Figure 7. Depiction of downslope winds due to Apennines mountains affecting the channeled flow in the Adriatic Sea.
8. Land–sea breezes: Thermal circulations related to land/sea differences produce concentrated regions of convergence called sea or land breeze fronts that are preferred locations of convection initiation. These features are relatively well posed and accurately predicted in a mesoscale simulation (see Fig. 8). They play a large role in initiating convection that helps lend another degree of predictability to precipitation events.
544
G. J. Tripoli et al.
9. Obstacle flow effects: The mountain islands and mountain ranges surrounding the Mediterranean also create obstacle flow effects that behave differently as the Froude number varies. These obstacle flows produce predictable curved flows on the downstream side of islands such as Corsica that can further concentrate and enhance the surface CAPE being moved around and may act to locally initiate convection in certain regions. The interaction of these flows with each other or with other land breeze flows can produce strong predictable vortex sheets that can lead to the growth of waterspout producing deep moist convection (Golden 1974). Simulations by the authors of waterspouts occurring on 29 October were performed leading to the CAPE and vorticity distribution given in Fig. 8. Regions of intersection between the obstacle flow and the land breeze effect led to a strong vortex sheet simultaneous with strong convergence leading to the predictable initiation of waterspout producing storms.
Figure 8. CRM simulation of complex surface flows over the Tyrrhenian Sea occurring at night resulting from a combination of obstacle flow around Corsica, a land breeze of the Italian peninsula and downslope flow off of the mountains in Corsica. The CAPE field is shown to be formed by WISHE upstream of Corsica and then advected around to the south and lifted in the zone where the land breeze off of Italy meets the obstacle flow. This particular situation produced a vortex sheet along the converging flows in the Tyrrhenian Sea that led to the formation of cumulus congestus and eventually waterspouts.
10. Lee cyclogenesis: The movement of synoptic scale baroclinic waves into the Mediterranean basin is characterized by vorticity laden flow moving across major topography barriers. The principles of vorticity conservation (Maddox et al. 1979) lead to lee cyclogenesis, often over
Microphysical Signatures of Extreme Events in the Western Mediterranean
545
the Mediterranean basin. The effect of the mountains is to increase the predictability of cyclogenesis location by helping focus the location of vorticity buildup by attaching it to known topographical features. The precipitation associated with the cyclogenesis has enhanced predictability, because of the reasons listed above. 11. Potential vorticity–surface coupling: Cyclogenesis over the Mediterranean is typically associated with the injection of stratospheric potential vorticity into the troposphere as a short-wave disturbance moves around the northwest side of a major trough (Tsidulko and Alpert 2001). Coupling of the upper level feature with a surface potential vorticity maximum can lead to surface cyclogenesis. Surface potential vorticity maximum can be inferred through warm near surface anomalies found in either the potential temperature or in the equivalent potential temperature in the case where convection is occurring. Originally this coupling mechanism was explained for Gulf-of-Mexico cyclogenesis by Bosart (1996) and later shown for Mediterranean cyclogenesis by Tripoli et al. (2004) (see Fig. 9). The interaction with the well represented sea surface temperature field, the large-scale dynamics of an approaching trough and the interaction with the local orography help focus cyclogenesis and again aid in enabling atmospheric models to represent the effect deterministically.
Figure 9. Numerical simulation of cumulus arising from Mediterranean causing a coupling with an upper level tropopause fold and rapid surface cyclogenesis near Algeria. The lowered 306 θe surface is a result of a tropopause fold and is associated with a similar shaped potential vorticity anomaly. The cumulus near the surface are in the form of θe plumes also containing 306 θe. Surface isobars drawn every 2 hPa are also drawn. View is from the west-southwest showing the Atlas mountains of north Africa near the center of the figure.
546
G. J. Tripoli et al.
4 CONCLUSION AND FUTURE WORKS Overall, precipitation occurring in the Mediterranean basin has been shown to feature a high degree of predictability because of these characteristics. As a result, CRSs tend to capture the precipitation mechanisms applicable to particular situation better than in many other places of the world where there are far more degrees of freedom to storm formation and so more uncertainty to the relationship between observed radiance and precipitation rate. A CRDB consisting of a wide cross section of Mediterranean storm types, can be expected to perform exceptionally well in this environment. Nevertheless, the performance of precipitation retrieval could be improved if the storm type occurring could be more precisely defined. The overall good predictability of Mediterranean events also means that a realtime CRM model prediction of precipitation in the basin will likely predict most precipitation events albeit with some degree of error. If a real-time relatively short-time prediction is employed to form the primary CRDB entry for a particular real-time retrieval, one might expect the probability of the database being applicable to the case to be better than an attempt to find a similar case in a general “historic” database. Hence the deterministic character of many Mediterranean precipitation events will likely facilitate the use of real-time Cloud Radiation Prediction (CRP) models as the optimal methodology for creating a statistical database for retrieval. To carry this step further, an interactive use of satellite observations and the model assimilation process would likely further improve the representation of precipitation implied from space-borne observations. Increasing computing power is beginning to allow for the possibility of real-time ensemble simulation and Ensemble-Kalman Filter (Evensen 1994) (EnKf) data assimilation schemes. This technique will use the combined use of satellite and all other observations including other satellite, radar, radiosonde, ACARS, and others to determine the error covariance matrix of each ensemble member acting as a background field for a multivariate analysis. The effect will be to use the data to selectively nudge the straying ensemble members toward solutions consistent with all of the observations and preserve the solutions that are consistent with observations. The analysis then takes the form of an ensemble of analyses, the spread of which is related to the uncertainty in the analysis. The unique characteristics of the Mediterranean basin will likely produce a relatively narrow spread in a Mediterranean precipitation analysis compared to other regions of the world where the solution is not so highly forced by the local geography. In conclusion, the Mediterranean basin is an excellent laboratory for developing technologies to retrieve precipitation from space-borne observing platforms. It unique characteristics make storms more predictable and better able to be represented in CRMs. As a result, there is a greater ability to relate
Microphysical Signatures of Extreme Events in the Western Mediterranean
547
the space-borne observations to precipitation with the help of the models, than in other regions of the globe. Acknowledgments: The authors would like to thank Sabatino DiMichele of CNR (now at ECMWF) and Will Lewis of the UW for their contributions to this study. This work was supported under NASA grant PMM-0069-0153 and within the framework of EURAINSAT a shared-cost project (contract EVG1-2000-00030) co-funded by the Research DG of the European Commission within the RTD activities of a generic nature of the Environment and Sustainable Development sub-program (5th Framework Program). This study has been partially funded by the Italian Space Agency through “LAMPOS” project and within the framework of Community Initiative INTERREG IIIB CADSES - RISK AWARE project.
5
REFERENCES
Bosart, L. F., G. J. Hakim, K. R. Tyle, M. A. Bedrick, W. E. Bracken, M. J. Dickinson, and D. M. Schultz, 1996: Large-scale antecedent conditions associated with the 12–14 March 1993 cyclone (“Superstorm’93”) over Eastern North America. Mon. Wea. Rev., 124, 1865–1891. Carlson, T. N., 1991: Mid-Latitude Weather Systems. HarperCollins Academic, New York, 1991. Evensen, G., 1994: Sequential data assimilation with a nonlinear quasigeostrophic model using Monte Carlo methods to forecast error statistics. J. Geophys. Res., 99 (C5), 10 143– 10 162. Golden, J. H., 1974: Scale-interaction implication for the waterspout life cycle II. J. Appl. Meteor., 13, 693–709. Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15, 808–816. Maddox, R. A., C. F. Chappell, and L. R. Hoxit, 1974: Synoptic and meso-α scale aspects of flash flood events. Bull. Amer. Meteor. Soc., 60, 115–123. Panegrossi, G., 2004: Learning from passive microwave measurements to improve microphysics parametrization in explicit cloud resolving models. PhD Thesis, University of Wisconsin, Madison, 267pp. Spencer, R. W., M. H. Goodman, and R. E. Hood, 1989: Precipitation retrieval over land and ocean with SSM/I. Part I: Identification and characteristics of the scattering signal. J. Atmos. Oceanic Technol., 6, 254–273. Tripoli, G. J., G. Panegrossi, A. Mugnai, S. Dietrich, and E. A. Smith, 2002: A numerical study of the Friuli, 1998 and Genoa, 1992 floods. Proceedings of the 2nd Plinius Conference on Mediterranean Storms, Siena, 127–140. Tsidulko, M. and P. Alpert, 2001: Synergism of upper-level potential vorticity and mountains in Genoa lee cyclogenesis: A numerical study. Meteor. Atmos. Phys., 78, 261–285. Tudurì, E. and C. Ramis, 1997: The environments of significant convective events in the western Mediterranean. Wea. Forecasting, 12, 294–306. Wilheit, T. T., 1986: Some comments on passive microwave measurements of rain. Bull. Amer. Meteor. Soc., 67, 1226–1232. Yano, J. and K. Emanuel, 1991: An improved model of the equatorial troposphere and its coupling with the stratosphere. J. Atmos. Sci., 48, 377–389.
43 ONLINE VISUALIZATION AND ANALYSIS: A NEW AVENUE TO USE SATELLITE DATA FOR WEATHER, CLIMATE, AND INTERDISCIPLINARY RESEARCH AND APPLICATIONS Zhong Liu1, Hualan Rui2, William L. Teng2, Long S. Chiu1, Gregory Leptoukh1 , and Gilberto A. Vicente1 GSFC Earth Sciences Data and Information Services Center, Distributed Active Archive Center, NASA/Goddard Space Flight Center, Greenbelt, MD, USA 1 CEOSR, George Mason University, Fairfax, VA, USA 2 SSAI, Lanham, MD, USA
Abstract
This article describes a new avenue to use satellite data for weather, climate and interdisciplinary research and applications: the TRMM Online Visualization and Analysis System (TOVAS). The system description, application examples, as well as future plans are given.
Keywords
TRMM, precipitation, rainfall, remote sensing, visualization, rainfall analysis, flood, drought, crop monitoring, crop forecast, satellite measurement, rainfall climatology, rainfall image, ASCII rainfall data, rainfall, rain rate
1
INTRODUCTION
Precipitation data have been widely used in weather, climate, and interdisciplinary research and applications. For example, a real-time or nearreal-time precipitation product can be used to monitor heavy rain events (Rui et al. 2003). Historical precipitation data and time series can be used to study historical events, seasonal-to-interannual variations, El Niño/Southern Oscillation (ENSO) events, etc. Precipitation products can also be used in interdisciplinary research and applications, such as, tropical infectious diseases (e.g., Anyamba et al. 2000; Zhou et al. 2002; Masuoka et al. 1998; 549 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 549–558. © United States Government 2007.
550
Z. Liu et al.
Liu et al. 2002a,b), drought and flood monitoring (Liu et al. 2002c), and crop yield estimates (e.g., the United Nations World Food Program), etc. However, in many tropical regions and parts of the mid-latitudes, precipitation estimates still remain a major challenge due to sparse rain gauges. To better develop research and applications for these regions, it is necessary to have rainfall data with adequate spatial and temporal resolutions. The Tropical Rainfall Measuring Mission (TRMM), as an important part of the NASA Earth Sciences Enterprise (ESE), is a joint US-Japanese satellite mission to monitor tropical and subtropical (40º S–40º N) precipitation and to estimate its associated latent heating. The TRMM satellite provides the first detailed and comprehensive data set on the four-dimensional distribution of rainfall and latent heating over vastly undersampled tropical and subtropical oceans and continents. The TRMM satellite was launched on November 27, 1997. Data from the TRMM satellite are archived and distributed by the NASA Goddard Space Flight Center Earth Sciences Distributed Active Archive Center (GES DAAC). TRMM data products and services can be found at the GES DAAC website (http://daac.gsfc.nasa.gov). Detailed information about TRMM can be found at the TRMM official website (http://trmm.gsfc.nasa.gov). Despite the relatively short history, TRMM rainfall products have been widely used in many areas. One of the goals of NASA ESE is to maximize the use of ESE data products. With over 6 years’ collection of TRMM data products, it is a challenging task to make them available to users at all levels. At present, there are only few Internet websites that offer data sets that have global coverage. Most of these sites provide data for downloading only and few provide browse images that are often either hard to read or often cannot be customized. To obtain precipitation information, such as, a time series, one often needs to go through these steps: (1) order the data product; (2) obtain the processing software; (3) install the software if they have the right equipment and operating system; (4) learn to use the software. Users can easily have problems in any one of these steps. In many cases, users will be very likely to find out that the product they order does not fit in their research and application requirements, therefore their time and resources have been wasted. In short, it is important, for all levels of users, to have a simple and easy-to-use system that allows everyone to access precipitation data products. To meet this requirement, the Hydrology Data Support Team (HDST) at the NASA GES DAAC initiated a project, the TRMM Online Visualization and Analysis System (TOVAS). TOVAS is an Internet-based system. It is a simple but powerful tool that enables users to concentrate on doing science. No other software or libraries are required. To display results, users simply select a product, an area, a parameter, a color option, a plot type, a time range, and an output type. With a web browser and a few mouse clicks, any user can easily obtain precipitation information from around the world. This paper will describe the system and its main functions, the data
Online Visualization and Analysis
551
products included in TOVAS, application examples, conclusions and future plans.
2
SYSTEM DESCRIPTION AND DATA PRODUCTS
TOVAS consists of three main components: input, processing and output. Figure 1 is a schematic of the system. The input component collects user’s selections from the web interface. Scripts are composed in the processing component to generate graphic or data output files. The last component sends the output files to the user’s browser. There are two web interfaces, Java and non-Java. The Java interface allows users to easily select an area of interest with a mouse. The non-Java interface allows users to use the system when Java and Javascript are disabled. The Grid Analysis and Display System (GrADS), developed by the Center for Ocean-Land-Atmosphere Studies (COLA), is used for the analysis. Standard analysis scripts are used and users can easily regenerate analysis results offline. Because of its simple design, TOVAS can be easily configured for new applications.
Figure 1. A schematic of TOVAS.
Main functions and features of TOVAS are: • Area plot – averaged or accumulated over any available data period for any rectangular area. • Time plot – time series averaged over any rectangular area. • Hovmöller plots – image view of any longitude-time and latitude-time cross sections. • ASCII output – for all plot types. • Image animation – for area plot. • Color options – for more customized images.
552
Z. Liu et al.
The TRMM satellite flies at an altitude of 402.5 km. The TRMM satellite carries three rain-measuring instruments. NASA Goddard Space Flight Center provided the TRMM Microwave Imager (TMI), the Visible Infrared Scanner (VIRS), and the observatory, and operates the TRMM satellite via the Tracking and Data Relay Satellite System (TDRSS). The Japan Aerospace Exploration Agency (JAXA) provided the Precipitation Radar (PR), the first space-borne precipitation radar, and launched the TRMM observatory. TRMM standard products at three levels are available at the GES DAAC. Level 1 products are the VIRS calibrated radiances, the TMI brightness temperatures, and the PR return power and reflectivity measurements. Level 2 products are derived geophysical parameters at the same resolution and location as those of the Level 1 source data. Level 3 products are the time-averaged parameters mapped onto a uniform spacetime grid. An evaluation of the sensor, algorithm performance and first major TRMM results appear in the Special Issue on the Tropical Rainfall Measuring Mission (TRMM), the combined publication of the Journal of Climate and Journal of Applied Meteorology (2000). The monthly distribution statistics collected at GES DAAC shows that most users prefer Level-3 products. At present, we only include Level-3 products in TOVAS (see Fig. 2). These Level-3 and other rainfall products are, 3B42RT (0.25 degree and 3-hourly), 3B42 (1 degree and daily), 3B43 (1 degree and monthly), the historical monthly rainfall (0.5 degree and monthly) provided by Cort J. Willmott and Kenji Matsuura from Center for Climatic Research Department of Geography University of Delaware (Willmott and Matsuura 1995), and the monthly precipitation product (1 degree and monthly) provided by the Global Precipitation Climatology Centre. All of the selected products provide global precipitation at different temporal and spatial scales. Since the launch of TOVAS, it has been used in a wide variety of earth science applications, such as weather events, climate, and interdisciplinary studies, agricultural crop monitoring, rainfall algorithm study, and data product comparison. Recent examples of the applications, collected by the GES HDST, are: • Study on coastal urban heat island effect on rainfall. • Additional rainfall information to supplement ground stations in Sri Lanka. • Phenology study in Africa and North America. • Crop yield estimates and flood watch in Africa and Asia. • Rainfall information for a development project in Afghanistan. • Fire monitoring activities in Africa. • Data for hydrological modeling in Africa. • Range prediction of American butterflies.
Online Visualization and Analysis
• • • •
553
Intercomparison with other products in North America. Monitoring rain events in Balkan. Investigation of the 1997–98 El Niño/La Niña event. Investigation of insect activities in the USA.
Figure 2. An example of the TOVAS interfaces for the TRMM Level-3 monthly product, 3B43.
Figure 3. Time series of the average rainfall in the affected region.
Z. Liu et al.
554
3
EXAMPLES OF WEATHER, CLIMATE AND INTERDISCINPLINARY APPLICATIONS
3.1 Morning heavy rain events The remnants of Tropical Cyclone Japhet, which moved into Mozambique on March 2, 2003, brought heavy rains to the region during the first week of March. Parts of Mozambique, Malawi and Zimbabwe experienced strong gusty winds and locally torrential rainfall, which produced areas of flooding. In Mozambique, over 50,000 people were affected, leaving thousands cut off from desperately needed food supplies according to the United Nations World Food Program (WFP). Figure 3 shows the time series of the 3-hourly precipitation over the affected region. The heaviest rainfall event can be easily identified. Figure 4 shows the spatial distribution of the heaviest rainfall event.
Figure 4. Rainfall map at 1800 Z March 6 2003. The 3-hourly near-real-time 3B42RT can be used to monitor heavy rain events, especially those over oceans where radar and gauge data are scarce.
3.2 Climate research and applications 3.2.1. Monitoring rainfall in the mid-Atlantic region With Hovmöller plot options, it is easy to obtain seasonal-to-interannual precipitation information around the world. TOVAS allows users to plot Hovmöller maps with a fixed latitude/longitude, or an averaging area with fixed latitudes/longitudes. Users could also adjust the color bar options to
Online Visualization and Analysis
555
make customized plots. Users could also obtain data in ASCII format for additional analyses and applications. With TOVAS, one could monitor and study both present and historical rainfall conditions around the world. Here is an example of the application in the Mid-Atlantic region of the USA. Figure 5 is a Hovmöller diagram for the 3B43 monthly rainfall product during the period between January 1998 and September 2003. The plot allows an easy comparison of seasonal-tointerannual variations. A period of drier months (marked by a white arrow) in 2002 can be easily identified from the plot. It was a severe drought reported in the region and the state emergencies were declared to save water. From the plot, a period of excessive rainfall following the drought (marked by a red arrow) can be identified as well.
Figure 5. Monthly 3B43 rainfall Hovmöller diagram for the Mid-Atlantic region of the USA.
Figure 6. Hovmöller diagram of rainfall along the Equator in Pacific Ocean.
3.2.2. Monitoring rainfall along the Equator Precipitation is a very important physical parameter in studies of El Niño/Southern Oscillation (ENSO) (Curtis and Adler 2000). With TOVAS,
556
Z. Liu et al.
it is easy to explore the whole data archive. Figure 6 is a Hovmöller diagram for the rainfall between January 1998 and November 2003. A large variation of rainfall is found during this time period.
3.3 Interdisciplinary applications TOVAS allows the HDST to provide many customized services to a wide variety of users. For example, recently the HDST has been providing precipitation information to the WFP for crop yield estimates and flood/drought assessment in Africa and Asia. Figures 7 and 8 are examples of many customized products that the HDST has been providing to WFP. Figure 7 is the rainfall accumulation estimate for January 1–10, 2003. Figure 8 is the maize yield projection based on the rainfall and an empirical equation.
Figure 7. January 1–10 (decade 1) 2003 rainfall accumulation estimate in mm (see also color plate 18).
Figure 8. Maize yield projection (% yield potential) based on estimated rainfall, and an empirical equation. White spaces denote zero yield potential (see also color plate 18).
Online Visualization and Analysis
4
557
CONCLUSIONS AND FUTURE PLANS
TOVAS provides a convenient way to access TRMM and other precipitation data products. TOVAS is a simple but powerful tool that enables users to concentrate on doing science, thus providing a new avenue for weather, climate and interdisciplinary research and applications. For example, before submitting a science proposal, one usually needs to do a lot of investigation work. The capability of generating countless analytical graphics and data allows users to fully explore the data products. Users from the climate community will find it particularly useful in obtaining information on seasonal-tointerannual variation. K-12 users can easily use TOVAS for their classroom projects because it does not require additional installations of software and libraries and the interface can be learnt in a few minutes. In short, TOVAS targets users at all levels. Once one learns how to use it, he or she will have nearly all the controls of the data products. TOVAS has demonstrated one of many great potentials of information technology. The merge of computation and communication technologies allows timely providing data and information and enabling users to concentrate on doing science, which will have a very profound impact on how we do science in the future. TOVAS is just a beginning of many similar ongoing prototypes in Earth science applications at the GES DAAC. A recently released MODIS Online Visualization and Analysis System (http://lake.nascom.nasa.gov/movas), developed jointly by the Aerocenter and GES DAAC at NASA Goddard Space Flight Center, is a good example to demonstrate how the TOVAS model can be applied to other disciplinary data products. Future plans for TOVAS will be concentrating on the following areas. The first to be added is intercomparison of precipitation products. The uncertainty issue in rainfall measurements is a well-known issue in research and application communities. Timely and easy access of this information will have a great impact on both research and applications. For example, errors in rainfall measurements can easily propagate to other precipitation derived products, such as, a crop yield estimates. At present, most users have to rely on referred publications where the works of investigators often either focus on global or a specific area. The conclusions often cannot be applied to the areas of their interests. With an online intercomparison system, users will be able to identify the uncertainty by intercomparing different products. In the future, related environmental variables, such as, terrain, measurements from satellites, will be added to the system to help identifying sources of errors. Several climatological precipitation products (baselines) will also be included in the system, which is very crucial for many monitoring activities. For the 3-hourly near-real-time 3B42RT product, the focus will be on enhancing and improving forecasting capabilities, such as, rainfall tendency, movement information of rain clusters. Users will be able to use the system
Z. Liu et al.
558
to monitor rain events. For the monthly products, new features, such as, anomaly and normalized (by the climatological mean) anomaly, will be added. The anomaly information will be very useful in monitoring drought events. New data products and systems will be added for interdisciplinary users. For example, a 10-day rainfall product is particularly useful for many agriculture users. Other remote-sensing products, such as, NASA QuikSCAT sea surface wind, TRMM Microwave Imager (TMI) sea surface temperature, etc. will be integrated into the system for air–sea interaction studies and applications.
5
REFERENCES
Anyamba, A., K. J. Linthicum, and R. L. Mahoney, 2000: Application of NDVI time series data to monitor Rift Valley fever outbreak patterns. Proceedings of the 25th Annual Climate Diagnostics and Prediction Workshop, Palisades, New York. Curtis, S. and R. F. Adler, 2000: ENSO indexes based on patterns of satellite-derived precipitation. J. Climate, 13, 2786–2793. Liu, Z., L. Chiu, W. Teng, H. Rui, and G. Serafino, 2002a: TRMM rainfall products and tools for tropical infectious disease studies. 15th Conference on Biometerology/Aerobiology and 16th Congress of Biometeorology, Kansas City, MO. Liu, Z., L. Chiu, W. Teng, H. Rui, and G. Serafino, 2002b: TRMM rainfall for human health and environment applications. International Tropical Rainfall Measuring Mission (TRMM) Science Conference, Honolulu, Hawaii. Liu, Z., L. Chiu, W. Teng, H. Rui, and G. Serafino, 2002c: TRMM rainfall data for ecosystem studies and applications in arid and semiarid regions. AGU Spring Meeting, Washington, DC. Liu, Z., L. Chiu, W. Teng, and G. Serafino, 2002: A simple online analysis tool for visualization of TRMM and other precipitation data sets. Science Data Processing Workshop 2002, Greenbelt, MD. Masuoka, P., J. Gonzalez, S. Gordon, N. Achee, P. Pachas, R. Andre, and L. Laughlin, 2002: Remote sensing and GIS studies of Bartonellosis in Peru. Poster Presentation at High Spatial Resolution Commercial Imagery Workshop, Reston, VA. Rui, H., W. Teng, Z. Liu, and L. Chiu, 2003: TRMM scientific data at the GES DAAC and their applications to monitoring tropical cyclones. IUGG2003 MC02/07P/D-002, Abstracts Week B page B.406 Special Issue on the Tropical Rainfall Measuring Mission (TRMM), combined publication of the December 2000 Journal of Climate and Part 1 of the December 2000 J. Appl. Meteor., AMS, Boston, MA. Willmott, C. J., and K. Matsuura, 1995: Smart interpolation of annually averaged air temperature in the United States. J. Appl. Meteor., 34, 2577–2586. Zhou, J., Lau, W. K., P. Masuoka, R. G. Andre, J. Chamberlin, P. Lawyer, and L. W. Laughlin, 2002. El Niño and the spread of Bartonellosis epidemics in Peru. EOS Trans., American Geophysical Union, 83(14), 157–161.
Section 8 The Present and Future of Satellite Platforms
44 THE SPACE-BASED COMPONENT OF THE WORLD WEATHER WATCH’S GLOBAL OBSERVING SYSTEM (GOS) Donald E. Hinsman1 and James F. W. Purdom2 1
World Meteorological Organization, Geneva, Switzerland2 Cooperative Institute for Research in the Atmosphere, Colorado State University, Fort Collins, CO, USA Abstract
Satellite data are an exceptionally important component of the World Weather Watch’s (WWW) Global Observing System (GOS). Indeed, it would be difficult to find any WMO programme that does not use or depend on satellite data. During these first decades of the 21st century WMO members will continue to exploit operational meteorological satellite systems, while expanding utilization to include experimental satellites. During the next decades existing capabilities will be refined and improved, while new applications and advanced technology will migrate from the experimental realm into full operational use. This paper addresses the current space-based component of the GOS and takes a brief look to the future.
Keywords
WMO, climate, weather, satellite
1
INTRODUCTION
The space-based subsystem of the World Weather Watch’s (WWW) Global Observing System (GOS) is now (2004) comprised of three types of satellites: operational meteorological polar-orbiting and geostationary, and environmental research and development (R&D) satellites. With regard to meteorological satellites, both polar-orbiting and geostationary, they continue to prove invaluable to WMO National Hydrological and Meteorological Services (NMHS) through the provision of a multitude of services including imagery, soundings, data collection, and data distribution. In particular, the present operational meteorological satellites include the following geostationary and polar-orbiting satellites: GOES-10, GOES-12, NOAA-15, NOAA-16, and NOAA-17 operated by the USA; GMS-5 operated by Japan; GOMS N-1, METEOR 2-20, METEOR 2-21, METEOR 3-5, and METEOR 561 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 561–570. © 2007 Springer.
562
D. E. Hinsman and J .F. W. Purdom
3M N1 operated by the Russian Federation; Meteosat-5, Meteosat-6, Meteosat-7, and Meteosat-8 (formerly MSG-1) operated by EUMETSAT; and FY-2B, FY-1C, FY-1D operated by China. Additional satellites in orbit include GOES-8, GOES-9, and GOES-11 operated by the USA. It should be noted that most space agencies contributing operational polar-orbiting and geostationary satellites have in place contingency plans for satellite systems that guarantee the continued daily flow of satellite data, products and services WMO members have come to depend on. In this regard, Japan and the USA initiated a back-up operation of GMS-5 with GOES-9 on 22 May 2003. With regard to R&D satellites, NASA’s Aqua, Terra, NPP, TRMM, QuikSCAT, and GPM missions, ESA’s ENVISAT, ERS-1, and ERS-2 missions, NASDA’s ADEOS II and GCOM series, Rosaviakosmos’s research instruments on board ROSHYDROMET’s operational METEOR 3M N1 satellite, as well as on its future Ocean series and CNES’s JASON-1 and SPOT-5, either are, or will be after launch, part of the R&D constellation. The ability of geostationary satellites to provide a continuous view of weather systems make them invaluable in following the motion, development, and decay of such phenomena. Even such short-term events as severe thunderstorms, with a lifetime of only a few hours, can be successfully recognized in their early stages and appropriate warnings of the time and area of their maximum impact can be expeditiously provided to the general public. For this reason, its warning capability has been the primary justification for the geostationary spacecraft. Since 71% of the Earth’s surface is water and even the land areas have many regions which are sparsely inhabited, the polar-orbiting satellite system provides the data needed to compensate the deficiencies in conventional observing networks. Flying in a near-polar orbit, the spacecraft is able to acquire data from all parts of the globe in the course of a series of successive revolutions. For these reasons the polar-orbiting satellites are principally used to obtain: (a) daily global cloud cover; and (b) accurate quantitative measurements of surface temperature and of the vertical variation of temperature and water vapour in the atmosphere. There is a distinct advantage in receiving global data acquired by a single set of observing sensors. Satellite data are totally different in character from in situ data and have to be used in ways that reflect their characteristics. For example, numerical weather prediction (NWP) models now use satellite-measured radiances directly, instead of inverting satellite radiances into atmospheric temperatures. This direct insertion of radiances into models had a profound positive impact on NWP forecast accuracy. Some observations, such as vegetation indices, have no direct surface-based counterpart but have been found to be of great value. Often, many years of research are needed in order to use new forms of satellite data to best advantage. Indeed, new uses for cloud imagery are still being developed after four decades of routine use, taking advantage of the
Space-Based Component of the World Weather Watch’s GOS
563
improving spectral resolution of new generations of satellites and the vastly improved capabilities of computer systems. The thrust of the current generation of environmental satellites is aimed primarily at characterizing the kinematics and dynamics of the atmospheric circulation. The ability to achieve such objectives was demonstrated during the global weather experiment in 1979. This capability is now part of the global operations of the WWW. The existing network of environmental satellites, forming part of the GOS of the WWW, produces real-time weather information on a regular basis. This is acquired several times a day through direct broadcast from the meteorological satellites by more than 1,450 stations located in over 170 National Meteorological and Hydrological Services (NMHSs). Figure 1 shows the nominal configuration for the spacebased subsystem of the WWW’s GOS.
Figure 1. Nominal space-based component of the global observing system.
Note: Information on the characteristics, capabilities and uses of the current system of meteorological satellites is contained in the CGMS Directory of Meteorological Satellite Applications. Additional up-to-date information can be found via the WMO Space Programme Homepage: http://www.wmo.ch/hinsman/satsun.html.
564
D. E. Hinsman and J .F. W. Purdom
1.1 Current and future polar platforms Operational meteorological polar-orbiting satellites provide global coverage twice a day. Their orbital altitude at approximately 850 km makes it technically feasible to make high spatial resolution measurements of the atmosphere/surface. To provide a reasonable temporal sampling for many applications, the WMO requirement is for at least two satellites in the AM and two satellites in the PM orbits, thereby providing 3-hourly coverage. A backup capability for the polar orbit exists by reactivating “retired” platforms and this has been demonstrated recently. Since 1979, coverage with two polar-orbiting satellites has been achieved most of the time. They carry a multispectral imager (usually with 1 km resolution visible, nearinfrared (NIR), and infrared (IR) window bands for observing cloud cover and weather systems, deriving sea surface temperature, detecting urban heat islands and fires, and estimating vegetation indices), a multispectral IR sounder (usually with roughly 20 broad spectral bands of 20 km resolution for deriving global temperature and moisture soundings), and a multispectral microwave (MW) sounder (most recently with 20 MW channels of 50 km resolution for deriving temperature soundings even in nonprecipitating cloud-covered regions). Current operational meteorological polar orbiters include the NOAA series from the USA and the METEOR, RESURS, and OKEAN series from the Russian Federation and the FY-1 series from China. They provide image data that can be received locally. The NOAA satellites also enable generation of atmospheric sounding products that are disseminated to NWP centres on WMO’s Global Telecommunications System (GTS). In the future, the NOAA AM satellite will be replaced by the METOP satellites provided by EUMETSAT and the NOAA PM satellite will transition to the NPOESS series. The Russian Federation METEOR series will evolve into the METEOR 3M series and the Chinese FY-1 series will be replaced by the FY-3 series. R&D missions continue to make many contributions in the area of polarorbiting remote-sensing measurements. To maximize the impact of those data and the associated expenditures in resources (manpower and financial) by operational users, space agencies are committing to (a) open and timely access to the data in standard formats, (b) preparation of the community for new data usage, and (c) data continuity. NASA’s Earth Observing System includes multiple platforms. Terra has been in an AM orbit since December 1999 and is providing global data on the state of the atmosphere, land, and oceans, as well as their interactions with solar and earth radiation. Aqua followed in a PM orbit in May 2002 and will provide climate-related data with respect to clouds, precipitation, atmospheric temperature/moisture content, terrestrial snow, sea ice, and SST. Both provide X-band direct broadcast of their high resolution MODIS data which are being received by a
Space-Based Component of the World Weather Watch’s GOS
565
number of WMO Members. In addition, AIRS data from Aqua is being tested in several GDPSs across the globe. The NASA/CNES Topex/ Poseidon and Jason-1 satellites provide a wealth of observations on the status of the ocean topography, and this is used in both short-term storm analysis and climate studies. In 2004, Aura will provide a suite of chemistry measurements focusing on atmospheric trace gases in the upper troposphere and lower stratosphere. In addition, the Earth Observer series has started providing hyperspectral VIS/NIR data. ESA launched the ENVISAT platform in March 2002. It is designed to provide measurements of the atmosphere, ocean, land, and ice over a 5-year period. ENVISAT has an innovative payload that will ensure the continuity of the data measurements of the ESA Earth Remote-Sensing (ERS) satellites, as well as facilitating the development of operational applications. Thereafter, several Earth Explorer missions are planned to study the gravity field (2005), atmospheric dynamics (2007), polar ice (2004), and soil moisture and salinity (2006). NASA and JAXA launched in 1997 a joint mission, the Tropical Rainfall Measuring Mission (TRMM). TRMM is designed to monitor and study tropical rainfall and the associated release of energy that helps to power the global atmospheric circulation shaping both weather and climate around the globe. It also provides information on soil moisture. By combining TRMM precipitation data with ocean vector winds data from QuikSCAT (launched by the USA in 1999), researchers have demonstrated the ability to significantly improve hurricane track and landfall prediction. JAXA intends to launch the Advanced Land Observing Satellite (ALOS) in 2004 that will utilize advanced land-observing technology. Later this decade, the Global Change Observation Mission (GCOM) will be aimed at observing parameters over the long term (as long as 15 years), and to understand the mechanism of the global environmental change. GCOM-A1 will observe ozone and greenhouse gases and GCOM-B1 will monitor energy and the general circulation from a sunsynchronous orbit. The People’s Republic of China is providing the newest series of polar-orbiting satellites, the FY-1 series. The FY-1 series has greatly enhanced imaging capabilities from polar orbit with its 10 channel radiometer that includes the same five channels as found on NOAA’s AVHRR and five new channels.
1.2 Current and future geostationary platforms The geosynchronous orbit is over 40 times higher than a polar orbit, which makes measurements technically more difficult from geostationary platforms. The advantage of the geostationary orbit is that it allows frequent measurements over the same region necessary for now-casting applications and synoptic meteorology. Weather, and weather-related phenomena extend across a broad range of scales. In meteorology the link between the synoptic
566
D. E. Hinsman and J .F. W. Purdom
scale and the mesoscale is many times a key factor in controlling the intensity of local weather. The only observing tool capable of monitoring weather across those scales (and those scales interactions) is the geostationary satellite. Nowcasting is in many aspects mesoscale in nature, and it is here that satellite data can provide great benefit. While observations from polarorbiting satellites often detect mesoscale phenomena at the needed spatial resolutions, they lack the temporal resolution required for many now-casting applications. For example, sounder data from polar-orbiting satellites, particularly HIRS data, provide high spatial resolution data that can be used to derive information on atmospheric instability over land in cloud free areas generally once in every 6 h. Those data are valuable for assessing the broadscale features such as axes of deeper moisture and instability that may support convection, however, their temporal resolution is not optimal for convective nowcasting. Complementing the polar-orbiting sounder data, the USA’ GOES geostationary sounder provides high spatial resolution hourby-hour information of the atmosphere’s ability to support (and inhibit) deep convection. It is important to recognize that prior to geostationary satellites, the mesoscale was a “data sparse” region; meteorologists were forced to make inferences about mesoscale phenomena from macroscale observations. Today’s geostationary satellites provide multispectral high-resolution imagery at frequent intervals. Those data reveal meso-meteorological features that are infrequently detected by fixed observing sites. The clouds and cloud patterns in a satellite image provide a visualization of mesoscale meteorological processes. When cloud imagery (and products derived from sounding data such as lifted index) is viewed in animation, the movement, orientation, and development of important mesoscale features can be observed. Furthermore, such animation provides observations of convective behaviour at temporal and spatial resolutions approaching the scale of the mechanisms responsible for triggering deep and intense convective storms. From geostationary altitude, a fixed full disk view of the Earth is viewed from one satellite, thus, at least six equally spaced satellites around the equator are needed to provide global coverage; polar regions are either very poorly observed, or not observed at all because of the large zenith viewing angles. Currently, there is global coverage from geostationary orbit (more than six operational satellites for image data and products (e.g., cloud motion winds) and two satellites are also providing a sounding capability. Reactivating “retired” platforms provides backup and there have been several examples of this type of activity. Additionally, operational satellite agencies have developed the “help your neighbour” concept whereby adjacent agencies seek to provide continuity of data and services through regional contingency planning. The geoimagers typically have 1 km resolution visible and 5 km IR window bands for observing cloud cover and weather systems in motion and estimating atmospheric motion vectors. The
Space-Based Component of the World Weather Watch’s GOS
567
geosounders to date have 18 broad spectral bands of 10 km resolution for deriving atmospheric temperature and moisture trends in time. Some of the satellites provide a real-time transmission capability to allow immediate access to the imagery for real-time applications. Products are disseminated on the WWW’s GTS by the satellite operators for near realtime applications. At present there are satellites at 0° longitude and 63°E (Meteosat 8 and 5 operated EUMETSAT), 76°E (GOMS N1 operated by the Russian Federation), 105°E (FY2B operated by the People’s Republic of China), 140°E (GMS-5 operated by Japan and backed up by GOES-9), and 135°W and 75°W (GOES 10 and 12 operated by the USA. All geostationary satellite instruments are evolving to more spectral coverage and faster imaging. In 2004, EUMETSAT introduced METEOSAT Second Generation (MSG) to the operational suite of geostationary satellites, making high-resolution 12-channel imagery of the earth’s disc available every 15 min with the advanced imager SEVIRI. Japan will launch JAMI in 2004. China will launch another in the FY-2 series of imagers in 2004. The USA will evolve to an Advanced Baseline Imager and Sounder in 2012. MSG also carries the geostationary earth radiation budget (GERB) instrument, the first full spectrum earth radiation budget measuring device to fly in geostationary orbit.
2
ENHANCED MONITORING OF SELECTED EARTH SYSTEM COMPONENTS
New observing capabilities demonstrated in a research mode during the next decade will become part of an operational observing system of the future. New capabilities relevant to WMO member needs to include the areas addressed below.
2.1 Atmospheric sounding Continuous observation of tropospheric temperature/moisture profiles, wind pattern, and moisture inflow in the far field around weather systems, where the cloud cover is broken, are currently being demonstrated in polar orbit with NASA’s AIRS instrument. Operational polar-orbiting systems that will follow AIRS with high spectral resolution interferometers are IASA on METOP and CrIS on NPOESS. Currently, the only geostationary programme planning for a high spectral resolution sounding instrument is the USA with the introduction of its GOES-R series. It is anticipated that the very high spectral resolution data from GOES-R’s sounder will demonstrate a unique ability to peer continuously through many layers of the atmosphere from geosynchronous orbit with the precision and accuracy of atmospheric
568
D. E. Hinsman and J .F. W. Purdom
sounding capabilities being developed for low earth orbit systems (i.e., 1 K accuracy and 1 km vertical resolution). The instrument is intended to map the three-dimensional (3D) distribution of water vapour at different altitudes in the atmosphere and determine atmospheric temperature profiles for use in nowcasting.
2.2 Atmospheric winds Global wind field measurements will be directly applicable to NWP, and extremely valuable for scientific diagnosis of large-scale atmospheric transport, weather systems, and boundary layer dynamics. A space-based Doppler lidar system is being developed by ESA to deliver observations of tropospheric and stratospheric wind data; progress in space-borne laser technology will continue in order to make this active sounding technique available to operational uses.
2.3 Global precipitation Quantitative measurement of the time and space distribution of global precipitation is the next highest climate research priority beyond atmospheric temperature and moisture, and an essential requirement to understand the coupling among atmospheric climate, terrestrial ecosystems, and water resources. Satellite remote sensing is the only means to acquire global rainfall data, considering the paucity of surface observations over the ocean and sparsely populated land areas. Measurement of global precipitation would likely be based on frequent observations from a constellation of passive MW sensors with detailed vertical atmospheric distribution of rain data provided by a common rain radar satellite for refinement and validation of retrieval algorithms for all satellites in the constellation. An early demonstration of this concept is being conducted using the TRMM, Aqua, and ADEOS II research satellites in tandem with operational meteorological satellites. This extends TRMM-like precipitation measurements to extratropical parts of the world for the first time, and demonstrates the concept of 3-h global precipitation products with utility to a broad range of WMO Members’ objectives.
2.4 Soil water content At present, near-surface soil water content is the only primary hydrologic variable that is not measured at large spatial scale. Scientific evidence shows that near-surface soil water content is the most significant indicator of the state of the terrestrial hydrologic system, and is the governing parameter for partitioning rainwater among evaporation, infiltration, and run-off. Large antennae will be needed to meet these requirements at low MW frequencies; these remain a significant technological challenge.
Space-Based Component of the World Weather Watch’s GOS
569
2.5 Cold climate products Passive and active (radar) MW remote-sensing methods are being considered by Europe, the USA, and Japan to determine the most effective means to acquire information globally on snow extent and water equivalent, soil freezing, and thawing that strongly affect the hydrologic regime of river basins at high latitudes.
2.6 Vertical profiles of clouds and aerosols Atmospheric aerosol content is subject to substantial variation in amount and type, as concentrations are driven by natural and human activities, including agricultural and industrial practices. In the first half of this decade, the first attempt at global observation of the 3D structure of clouds and aerosol distributions will be undertaken. These involve active remote-sensing systems (i.e., lasers/lidars and MW radars) rather than the passive remotesensing systems such as radiometers that are common today. Due to the long-term nature of climate change research, such systems are likely candidates to become part of the operational climate-observing system in the future.
3
CONCLUSIONS
During the next several years the polar-orbiting component of the operational environmental satellite system will evolve to include at least six satellites in polar sun-synchronous orbits. Those satellites will carry highresolution multispectral imagers, IR interferometer sounders, advanced MW imagers and sounders, improved ozone monitoring instruments, and radio occultation (GPS) sensors. As in the early days of satellite meteorology, polar-orbiting satellites are again providing valuable information on which we will build the geostationary satellite systems of the future. In the geostationary arena, plans are for similar coverage as today with satellites operated by China, India, EUMETSAT, the Russian Federation, and the USA. The major step forward in the geostationary arena is that many operators are planning for with satellites with greatly improved spectral and temporal coverage for imaging, and the USA is planning to move to hyperspectral IR sounding. Data from research satellite instruments such as NASA’s Moderate Resolution Imaging Spectro-radiometer and AIRS, NASDA’s IMG, NASA’s Earth Observer-1 with the hyperspectral Hyperion instrument, and ESA’s Medium Resolution Imaging Spectrometer are clearly showing that innovative hyperspectral observing is the future for satellite imaging and sounding. Planning is underway for an important follow-on mission based on the success of TRMM. That mission, known as the Global Precipitation Measurement (GPM) mission, will allow for the derivation of
570
D. E. Hinsman and J .F. W. Purdom
global precipitation at least eight times a day (more frequent in polar regions due to multiple area coverage from the sun-synchronous satellite observations). Satellite data provides a user with powerful information that can be used to aid in a variety of WMO programmes. For some applications, optimal resolutions may not be attainable from any one satellite but may be approached using data from a series of satellites. The future of space-based remote sensing and its applicability to WMO member needs looks bright. The move towards an improved space-based component of the GOS, building on current capabilities should provide a firm foundation of observations upon which skill in many areas will advance. Acknowledgements: The space-based component of the GOS is a worldwide cooperative effort involving both research and operational satellite operators. It is through their dedication and hard work that this cooperative effort is being brought to fruition.
45 THE METEOSAT AND EPS/METOP SATELLITE SERIES Johannes Schmetz, Dieter Klaes, Alain Ratier, and Rolf Stuhlmann EUMETSAT, Darmstadt, Germany Abstract
The paper provides short overviews of the European meteorological satellite programmes: (1) the first generation of European geostationary Meteosat satellites, (2) Meteosat Second Generation (MSG), and (3) the future EUMETSAT Polar System (EPS). The features of the MSG satellites are presented along with examples of novel observations of cloud and atmospheric instability. Four MSG-type satellites will serve the community for the next decade and a half. Finally, the future EPS/Metop satellites jointly developed with the European Space Agency (ESA) are introduced. The first Metop satellite is scheduled for launch in 2005. EPS/Metop provides advanced observations of temperature and humidity profiles, wind, ozone, and trace gases.
Keywords
Meteosat, MSG, SEVIRI, GERB, EUMETSAT
1
INTRODUCTION
Until 1960, our knowledge of the present weather situation around the world had been almost entirely provided by ground-based observational systems. But this changed dramatically with the launch of the first US meteorological satellite from Cape Canaveral, Florida, on 1 April 1960. This experimental satellite, TIROS (Television and Infrared Observation Satellite) provided for the first time regularly pictures of the Earth’s weather systems over large areas. Europe started to contribute to the global space-borne observing system with the launch of Meteosat-1 on 23 November 1977; this satellite was the first in geostationary orbit that carried a water vapour (WV) channel in the 6.3 µm band. Today we look back to more than 40 years of meteorological satellites that have proven to be the best way to observe the weather on a large scale. Typically, operational meteorology utilizes two types of satellites to provide the required information: 571 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 571–586. © 2007 Springer.
J. Schmetz et al.
572
Low Earth-Orbiting (LEO) satellites fly at relatively low altitudes of around 800 km above the Earth, mostly in polar (sun-synchronous) orbits, and can provide information with high spatial resolution. The whole surface of the Earth can be observed twice a day. More than one polar satellite with different equatorial crossing times is required in order to attain more frequent observations. Geostationary satellites (GEO) flying in the equatorial plane at an altitude of about 36,000 km above the Earth have the same revolution time as the Earth itself and therefore always view the same area. They can perform frequent imaging, which, in animated mode, depicts the everchanging atmospheric processes. The disadvantage of the high altitude is that it limits spatial resolution and precludes the use of active instruments like radars. With the experience of seven geostationary Meteosat satellites of the first generation, the successful start of operations of Meteosat-8 (a new satellite series) and the EUMETSAT Polar System (EPS), with its Metop satellites, close to the first launch, Europe contributes significantly to the space-based global observing system, through the combined efforts of EUMETSAT, the European Space Agency (ESA), Centre National d’Etudes Spatiales, and other partners.
2
THE FIRST-GENERATION METEOSAT SATELLITES
In 1972 the Meteosat programme, originally a proposal from the French space and meteorological authorities, became “Europeanised” through legal arrangements between the European Space Research Organisation (ESRO), later to become the ESA, and the Centre National d’Etudes Spatiales (CNES). Only 5 years later the efforts culminated in the launch of the first satellite on 23 November 1977. The first image from Meteosat-1 was successfully acquired on 9 December 1977 (see Fig. 1) and the satellite continued to operate nominally until 24 November 1979, when an on-board electronic component failure resulted in the loss of all missions except that supporting the data collection systems (DCS). The launch of the second preoperational satellite, Meteosat-2, which had already been approved in 1977, took place on 19 June 1981. The undoubted success of the preoperational programme promoted the development of a follow-on operational programme. The Meteosat Operational Programme (MOP) began in 1983 and was executed by ESA, on behalf of EUMETSAT from 1986 onwards. As the construction of three new satellites took some years and Meteosat2 had a design lifetime of 3 years only, a decision was taken to refurbish the prototype satellite, Meteosat-P2, to a flight standard for launch on the first test flight of the Ariane-4 launcher. This satellite would then be capable of
The Meteosat and EPS/Metop Satellite Series
573
filling any potential gap between Meteosat-2 and the first of the new MOP satellites. Meteosat-3 (renamed from Meteosat-P2 after launch) was eventually launched in June 1988 and took over the operational missions during the following August. It was fortuitous that Meteosat-2 had been able to fulfil its prime functions of image taking and dissemination over 7 years after launch. After serving as an in-orbit spare for a further 3 years Meteosat2 was finally re-orbited, to an altitude over 300 km above geostationary orbit, in early December 1991. With the successful launch of Meteosat-4 on 6 March 1989 the MOP was underway. Then Meteosat-3 supported for nearly 4 years two important missions in support of the National Oceanic and Atmospheric Administration (NOAA). From 1 August 1991 through 27 January 1993 Meteosat-3 provided operational images of the Atlantic basin from a position at 50°W, under the called the “Atlantic Data Coverage (ADC)” mission (de Waard et al. 1992). The satellite then was moved to 75°W and provided an operational imaging service from 21 February 1993 through 1 May 1995, called the “Extended Atlantic Data Coverage (XADC)”. The MOP programme comprised two more satellites, both of which are, at this point in time, still in operational use. Meteosat-5 operates over the Indian Ocean at 63°E since July 1998 when operation began as a support to the Indian Ocean Experiment (INDOEX). A year later this support to INDOEX continued as a routine Indian Ocean Data Coverage (IODC) mission. Meteosat-6, the backup to Meteosat-7, provides so-called “rapid scan” imagery with a repeat cycle of 10 min observing Europe. Meteosat-7, launched in September 1997, is the last first-generation Meteosat. It operates in the nominal European orbit at 0° longitude.
2.1 Earth imaging with the first generation of Meteosat The main instrument on-board the satellite is a multispectral radiometer, which provides image data in three spectral bands (e.g., Mason and Schmetz 1992): • • •
0.5–0.9 µm visible band, 5.7–7.1 µm infrared (IR) WV absorption band, 10.5–12.5 µm thermal IR window band.
Images of the full Earth disk are taken in the three spectral bands every half hour. The radiometer scans the Earth from east to west by virtue of the spinning motion (100 rpm) of the satellite whilst the south to north scanning is achieved by stepping the radiometer through a small angle (1.25 × 10–4 rad) at each rotation of the satellite. There are two visible detectors (VIS1 and VIS2), that are placed in the focal plane of the primary telescope such that they observe adjacent lines of the Earth. Thus, by combining each
J. Schmetz et al.
574
individual visible image, consisting of 2,500 lines each of 5,000 pixels, it is possible to obtain a visible image of 5,000 lines of 5,000 pixels with a sampling distance of 2.5 km at the subsatellite point (SSP). The SSP sampling distance of the IR and WV images is 5 km, each image consisting of 2,500 lines of 2,500 pixels. On the preoperational satellites, i.e., Meteosat-1 to Meteosat-3, the idea to include a WV channel occurred at a late stage in the development of the radiometer and was made possible by using the electronic chain of one of the visible channels. This meant that with the first three satellites two modes of operation were possible: Either mode 1 including VIS2, WV, and IR or mode 2 with VIS1, VIS2, and IR. In practice during daylight hours the two modes were used alternately for successive slots whilst at night mode 1 was used exclusively. The VIS and WV images had a digitization of only 6 bits. With the start of Meteosat-4 (first of MOP series) full resolution VIS and WV images could be operated simultaneously and all image data were digitized with 8 bits. Noteworthy is also an improvement to the noise of the WV channel for Meteosat-4 to Meteosat-7, which was a significant step towards the successful operational derivation of atmospheric motion vectors (AMVs) in clear-sky areas from WV images (Laurent 1993). The clear-sky WV winds complemented the AMVs derived from the tracking of cloud features (Schmetz et al. 1993). Both became important in the assimilation of data for numerical weather prediction models. More recent advances in the AMV products addressed the quality control (Holmlund 1998).
3
METEOSAT SECOND GENERATION
As the first Meteosat series (i.e., Meteosat-1 through Meteosat-7), MSG (now Meteosat-8) and its successors are spin-stabilized, however, capable of greatly enhanced Earth observations. The satellite’s 12-channel imager, known as the spinning-enhanced visible and infrared imager (SEVIRI), observes the full disk of the Earth with an unprecedented repeat cycle of 15 min in 12 spectral wavelength regions or channels. The MSG programme covers a series of four identical satellites, expected to provide observations and services over at least 15 years. Each satellite has an expected lifetime of 7 years. As with the first Meteosat system, the new generation, starting with Meteosat-8, is planned as a dual-satellite service, where one additional satellite is available in orbit. As MSG is a new series of satellites (see Fig. 1), the period for commissioning of Meteosat-8 was longer than the typical 6-month period
The Meteosat and EPS/Metop Satellite Series
575
Figure 1. The first MSG satellite was launched in August 2002; it was renamed to Meteosat-8 with the start of operational services in January 2004.
and the operational service started in January 2004. A second MSG satellite, named Meteosat-9, was launched on December 2005.
3.1 Earth imaging with the second generation of Meteosat The primary mission of the second-generation Meteosat satellites is the continuous observation of the Earth’s full disk with a multispectral imager. The repeat cycle of 15 min for full-disk imaging provides multispectral observations of rapidly changing phenomena such as deep convection. It also provides for better retrieval of wind fields, which are obtained from the tracking of clouds, WV, and ozone features. The imaging is performed by utilizing the combination of satellite spin and scan mirror rotation, a process known as stepping. The images are taken from south to north and east to west. The eight thermal IR and three solar channels have a sampling distance of 3 km at nadir and scan the full disk of the Earth. The high-resolution visible channel provides images with one kilometre sampling at nadir. Data rate limitations confine the high-resolution visible images to half the Earth in an east–west direction; however, the exact coverage of the Earth is programmable.
J. Schmetz et al.
576
Table 1. Spectral channel characteristics of SEVIRI in terms of central, minimum, and maximum wavelength of the channels and the main application areas of each channel. Channel No.
Spectral band (µm)
Characteristics of spectral band (µm)
λcen
λmin
λmax
1
VIS0.6
0.635
0.56
0.71
2
VIS0.8
0.81
0.74
0.88
3
NIR1.6
1.64
1.50
1.78
4
IR3.9
3.90
3.48
4.36
5
WV6.2
6.25
5.35
7.15
6
WV7.3
7.35
6.85
7.85
7
IR8.7
8.70
8.30
9.1
8
IR9.7
9.66
9.38
9.94
9
IR10.8
10.80
9.80
11.80
10
IR12.0
12.00
11.00
13.00
11
IR13.4
13.40
12.40
14.40
12
HRV
Broadband (about 0.4–1.1 µm)
Main observational application
Surface, clouds, wind fields Surface, clouds, wind fields Surface, cloud phase Surface, clouds, wind fields Water vapour, high level clouds, atmospheric instability Water vapour, atmospheric instability Surface, clouds, atmospheric instability Ozone Surface, clouds, wind fields, atmospheric instability Surface, clouds, atmospheric instability Cirrus cloud height, atmospheric instability Surface, height of thin clouds, atmospheric instability
SEVIRI has eight spectral channels in the thermal IR, three channels in the solar spectrum, and a broadband high-resolution visible channel. The accompanying table provides more details of the characteristics of these channels, and indicates how each channel is used: for observations of clouds and surface temperatures, WV or ozone. Figure 2a and b show the location of the SEVIRI bands on top of a solar and typical thermal energy spectrum, respectively. Figure 3a and b give examples of the weighting functions of the thermal channels for a tropical standard atmosphere and nadir view (3a) and a subarctic winter atmosphere (3b) and a satellite viewing angle of 60°. The MSG level 1.5 data have a 10-bit digitization and provide the basis for all further processing and for the derivation of meteorological products. Concerning radiometric performance Meteosat-8 exceeds the specification by far; Fig. 4 shows a comparison of the specified radiometric noise in terms
The Meteosat and EPS/Metop Satellite Series
577
of NEDT at a reference temperature for the thermal IR channels with in-orbit performance measurements after Aminou et al. (2003). It should be noted that the specified NEDT refers to end of life, while the in-orbit measurements refer to beginning of the satellite’s lifetime. SEVIRI SOLAR CHANNELS
Standard Mid-Latitude Summer Nadir
2000 300
1500 0.60 LEAF REFLECTANCE
1000 0.40
SOIL REFLECTANCE
500
0.20
EBBT [K]
2
0.80
IRRADIANCE (W/ m )
REFLECTANCE / TRANSMITTANCE
1.00
250
TOA IRRADIANCE
0.00
0 0.5
1.0 WAVELENGTH (mm)
(a)
1.5
200
5
10
15
WAVELENGTH (mm)
(b)
Figure 2. Left: (a) MSG SEVIRI spectral response functions for the solar channels plotted with the spectral reflectance of vegetation, bare soil and the spectral irradiance at the top of the atmosphere. Right: (b) thermal terrestrial spectrum in terms of equivalent blackbody brightness temperature (EBBT) for a mid-latitude summer atmosphere and nadir view and MSG SEVIRI spectral response functions for the thermal infrared (IR) channels.
Figure 3. Left: (a) weighting functions for the MSG SEVIRI thermal channels, i.e., channels 4–11, for a satellite nadir view. A tropical summer standard atmosphere has been assumed for the simulation with a radiative transfer model. Right: (b) same as Fig. 3a except for a subarctic winter atmosphere and a viewing angle of 60°.
As an additional scientific payload, MSG carries a geostationary Earth radiation budget (GERB) instrument that observes the broadband thermal IR and solar radiances exiting the Earth’s atmosphere. The GERB instrument makes accurate measurements of the shortwave and longwave components
J. Schmetz et al.
578
of the radiation budget at the top of the atmosphere. GERB data are of interest to climatological studies and also for comparison with and validation of weather forecast models. As the first radiation budget experiment from geostationary orbit, the GERB instrument has great potential to shed new light on climatic processes related to clouds and WV. In particular, simultaneous observations with SEVIRI and GERB will reveal unknown physical elements of the process of deep convection (e.g., in the tropics) and its influence on the radiation budget. MSG Noise Specifications and In-Flight Measurements 2.00 1.80 1.60
NEDT (K)
1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00 3.9
6.2
7.3
8.7
9.6
10.8
12
13.4
Channe l Ce ntral Wav e le ngth (µm)
Specification
In-Flight Measurement
Figure 4. Noise equivalent temperatures of cold spectral channels of MSG (i.e., Meteosat-8). The reference temperatures for the NEDT are: 300 K at 3.9 µm; 250 K at 6.2 µm; 250 K at 7.3 µm; 300 K at 8.7 µm; 255 K at 9.7µm; 300 K at 10.5 µm; 300 K at 12 µm; 270 K at 13.4 µm. (From Aminou et al. 2003.) Note that specified NEDT refers to end of life, measured values to beginning of life.
3.2 Products from Meteosat Second Generation Continuity for meteorological products from the first-generation Meteosats is provided through core products centrally derived at EUMETSAT. Those meteorological products include atmospheric motion vectors, cloud analysis, and atmospheric humidity. In addition, there are novel products such as atmospheric instability and total ozone over the entire MSG field of view (FOV) (for details see Schmetz et al. 2002). Over and above the central processing at EUMETSAT, there are products from a geographically distribution network of services, Satellite Application Facilities (SAF), hosted by National Meteorological Services and other institutions. The idea behind the network of SAFs is that more products from MSG (and also from the future EPS) can be derived effectively capitalizing on the scientific expertise across the EUMETSAT member states. In January
The Meteosat and EPS/Metop Satellite Series
579
2004, seven SAFs exist addressing seven themes: (1) ocean and sea ice hosted by Météo-France, (2) numerical weather prediction hosted by the MetOffice, UK, (3) nowcasting hosted by the Institute for Meteorology, Spain; (4) GRAS Meteorology hosted by the Danish Meteorological Institute, (5) land surface analysis hosted by the Portuguese Meteorological Institute, (6) climate monitoring hosted by the German Met Service (DWD), (7) ozone monitoring hosted by the Finnish Meteorological Institute, and (8) support to operational hydrology and water management hosted by the Italian Meteorological Service (UGM). Each SAF provides operational services to end-users, including real-time and/or off-line product services, distribution of user software packages, and data management. The application of the second generation of Meteosat satellites ranges from short-term forecasting to numerical weather prediction and climatological studies. The most important products for numerical weather prediction are the spectral radiances themselves and the wind fields derived from tracking the displacement of clouds and WV features in successive satellite images. Both winds and radiances are assimilated into the numerical models that compute the change of the atmosphere in the future and provide the basis for weather forecasts.
Figure 5. B/W version the RGB full disk-image from Meteosat-8 observed on 5 June 2003. Channel 3 (at 1.6 µm); channel 2 (at 0.81 µm), and channel 1 (at 0.635 µm) were used. Ice clouds appear bright.
Figure 5 shows an RGB image from Meteosat-8 derived centrally at EUMETSAT, where channels 3, 2, and 1 have been assigned to the colours red, green, and blue, respectively. As ice clouds strongly absorb at 1.6 µm (channel 3) those ice clouds appear blue in Fig. 5. Another useful product is the global instability index, specifically the lifted index (König 2002).
J. Schmetz et al.
580
4
THE EUMETSAT POLAR SYSTEM/METOP
4.1 Overview The EPS is the European contribution to the joint European/US operational polar satellite system. The European part covers the mid-morning (AM) orbit, whereas the USA continues to cover the afternoon (PM) orbit. In the framework of the EPS programme (Bühler et al. 2001; Klaes et al. 2001, 2007) a space segment with associated launch services, a full ground segment are developed, expected to provide 14 years of operations. The space segment comprises three Metop satellites, which are developed in cooperation by the EUMETSAT and the ESA, the French Centre National d’Etudes Spatiales (CNES). The Metop satellite will also carry instruments provided by the National Oceanic and Atmospheric Administration (NOAA). The first Metop satellite has been launched on 19 October 2006. The first Metop spacecraft is being developed under the ESA Metop-1 Programme. This includes the development of some payload components as the Global Ozone Monitoring Experiment (GOME-2), Advanced Scatterometer (ASCAT) and the GPS Radio Occultation Sounder (GRAS). Furthermore an Advanced Very High Resolution Radiometer (AVHRR) and the Advanced TIROS Operational Sounder (ATOVS) package, comprised of High Resolution Infrared Radiation Sounder (HIRS-4), Advanced Microwave Sounding Unit (AMSU-A), and Microwave Humidity Sounder (MHS) are components of the Metop payload. MHS, which is a EUMETSAT development, replaces the AMSU-B instrument in the ATOVS suite, while NOAA provides the ATOVS and AVHRR instruments. The IASI instrument, developed by CNES, provides advanced sounding capabilities in the IR. Figure 6 provides a view of the Metop satellite and its payload.
Figure 6. Metop spacecraft of the EUMETSAT Polar System with instruments.
The Meteosat and EPS/Metop Satellite Series
581
4.2 ATOVS: Continuity Continuity is an important aspect to operational services. Today the information on temperature and humidity soundings and surface information (including clouds) for numerical weather prediction and other applications, is provided by the ATOVS suite supported by the AVHRR imager for both morning and afternoon missions (currently on the NOAA-16 (PM) and NOAA-17 (AM) satellites). The instrument suite is common to the two components of the Initial Joint Polar System (IJPS), i.e., Metop-1 and Metop-2 (AM satellites) and NOAA-N and NOAA-N’ (PM satellites) satellites.
4.3 IASI: New technology The Infrared Atmospheric Sounding Interferometer (IASI) introduces new technology into EPS. The purpose of IASI to measure temperature, WV and trace gases at a global scale. The measurement principle is Michelson interferometry providing 8,461 spectral channels, aligned in three bands between 3.62 µm (2,760 cm–1) and 15.5 µm (645 cm–1). The unapodised resolution is between 0.3 and 0.4 cm–1, with a spectral sampling of 0.25 cm–1. Included into the instrument is an integrated imaging system (IIS), consisting of a radiometer measuring between 10 and 12 µm with high spatial resolution. The FOV covers 64 × 64 pixels and provides information in the focal plane of the sounder, allowing to co-register with AVHRR, enabling an accurate navigation and also a detailed analysis of cloud properties inside the IASI sounder pixels. Figure 7 shows a typical simulated IASI spectrum for a clear and cloudy situation after Rizzi (1998).
Figure 7. IASI spectrum simulated for a clear and cloudy atmosphere. (From Rizzi 1999.) Gray = clear sky, black = cloudy sky.
J. Schmetz et al.
582
l Figure 8. Retrieval errors from simulated IASI retrievals for temperature (left) and specific humidity (right). (From Schlüssel and Goldberg 2000.) Figures are plots of RMS errors in temperature and humidity profiles, all angles, noise-averaged over IASI IFOV-quadruples, sea surface, no clouds.
The IASI data will be used in synergy with the microwave sounding instruments, to which the scan is synchronized. Products will include, besides level 1 spectra, vertical profiles of temperature, humidity, and ozone at global scale. Figure 8 depicts the simulated retrieval error for temperature and humidity after Schlüssel and Goldberg (2002) indicating that an accuracy for temperature of 1 K per 1 km layer can be achieved throughout most of the troposphere and lower stratosphere.
4.4 ASCAT and GOME-2: Operational use of research missions There are two instruments within EPS, which are in heritage of missions on the ESA Earth Remote Sensing (ERS) Satellites: The ASCAT and the GOME-2. 4.4.1 Advanced Scatterometer ASCAT is a real-aperture, polarised C-band radar in heritage from the AMI/Wind mode mission instrument on ERS-1 and ERS-2. The improved design aims at providing ocean surface winds at 50 km over a 25 × 25 km2 grid, along and across two 550 km wide swaths on both sides of the nadir track. A high-resolution wind product will be generated at 12.5 × 12.5 km2 grid, providing wind vectors at the sea surface at 25 km resolution.
The Meteosat and EPS/Metop Satellite Series
583
Further capabilities of ASCAT are the measurement of sea ice boundaries and sea ice type. Emerging applications are expected from land surface observations. 4.4.2 Global Ozone Monitoring Experiment The GOME-2 provides the possibility to monitor the ozone total column and profiles and the components related to ozone chemistry in the Earth’s atmosphere. The instrument on Metop profits from experience gained over many years of observation and data analysis with the GOME instrument on ERS-2. Improvements include: • • • • •
Increased spatial sampling of 40 × 40 km2 for total column products, Increased Earth coverage due to increased swath width (1,920 km). Improved polarization measurements (12 bands). Enhanced on-board calibration through added white light source. Increased spectral sampling.
GOME-2 measures the backscattered UV–Vis radiation in four band between 240 and 790 nm, at a spectral resolution between 0.25 and 0.5 nm. Expected accuracy of ozone total column and profiles is better than 5% and 15% above 30 hPa and better than 50% below 30, hPa respectively. The objective is 3% for columns and 10% for profiles at all levels. Additional intended products are vertical columns of BrO, OClO, NO2, and SO2, expected to be retrieved with accuracy better than 20%.
4.5 GPS Radio Occultation Sounder The GRAS makes use of the signals of the Global Positioning Satellite System (GPS) and introduces this technology into operational use for the first time. It follows experience obtained with the GPS/MET and CHAMP experiments. The basic product is the bending angle profile, which provides the basic parameter for the derivation of temperature profiles from atmospheric refractivity. Making the GPS occultation sounding operational in real time requires to introduce a complex subsystem into EPS. In order to correct the clock errors from the different involved satellites and to provide the required precise orbits of the satellites involved in the measurement, in particular the Metop satellite, the installation of a global GRAS Support Network (GSN) was required. This GSN needs to be operated in real time includes 24 global stations.
J. Schmetz et al.
584
4.6 EPS ground segment data and products The EPS ground segment comprises a central component and distributed elements, the former at EUMETSAT in Darmstadt, the latter at the SAF, hosted by EUMETSAT member states (see Section 3.2). The central facility provides global level 1 products with a latency of 2 h 15 min, and selected level 2 products (IASI and ATOVS retrieved profiles of temperature and humidity) with a latency of 3 h. The SAF provide level 2 and higher products, based on EPS products or as multi-mission products using other satellite data, including MSG data. The access to the products is assured in real-time through the Advanced High Resolution Picture Transmission (AHRPT) and the Low Resolution Picture Transmission (LRPT) service, where users have the possibility to receive (local) EPS data in real time, when the satellite is in view of their reception stations. HRPT comprises all mission data, whereas LRPT contains the full set of ATOVS sounding HIRS/4, AMSU-A, and MHS) data and JPEG compressed AVHRR data (three selected channels). LRPT replaces the analogue APT service. The near real-time access to global level 1 and level 2 products extracted at EUMETSAT is also provided to EUMETSAT member states and ECMWF. A subset of the centrally processed products will be made available to WMO users for the dissemination over the Global Telecommunication Service (GTS/RMDCN). SAF products are planned to be distributed via GTS/RMDCN as well. The EUMETSAT Unified Meteorological Archive and Retrieval Facility (U-MARF) provides off-line data access within 7 h from measurement. All centrally generated products will be archived in U-MARF, this includes the raw data. Most of the products generated in SAFs are archived locally SAFs, however catalogue information is available in U-MARF too.
5
SUMMARY AND OUTLOOK
The European geostationary Meteosat satellites have been a great success and are, nowadays, considered as indispensable for meteorological services. Seven Meteosat’s of the first generation provided operational services since 1977, imaging the full disk of the Earth every 30 min in three spectral bands. The last three of the generation, i.e., Meteosat-5, Meteosat-6, and Meteosat7, are still in operational use. With Meteosat-8 a new series of geostationary satellites has started. After the launch end of August 2002 an extended commissioning period ended with the start of operations in January 2004. Meteosat-8 and its successors, continue the successful 25-year long mission of the first-generation Meteosat satellites. The established services from the first-generation satellites will
The Meteosat and EPS/Metop Satellite Series
585
continue with a seamless operational transfer. In addition, Meteosat-8 and its successors offer a wealth of new observational capabilities: SEVIRI, the operational imager, has twelve spectral channels and observes the Earth (full disk) every 15 min. The multispectral imaging and the high temporal repeat cycle will benefit weather forecasting and will improve severe weather warning. Significant indirect benefits will come through the assimilation of products in numerical weather models by improved forecasts. Four Meteosat satellites of the second generation are foreseen to cover operational services until about 2015, when Meteosat third generation is expected to take over. The new EPS, developed jointly with the ESA and CNES, and launched in 2006 establishes the European contribution to the global polar meteorological space observing capabilities. The Metop satellites of EPS fly together with the last TIROS satellites and then with the future National Polar Orbiting Environmental Satellite System (NPOESS) and NPOESS Preparatory Programme (NPP) satellites, where Metop will be in the morning orbit around 0930 UTC. NPOESS satellites will have instruments equivalent to the ones on Metop, i.e., Cross Track Interferometer Sounder (CrIS) equivalent to IASI and ATMS equivalent to AMSU-A plus MHS. EPS provides on the one hand continuity to current systems, through continuation of the proven ATOVS instrument suite and the AVHRR imager, on the other hand it includes novel capabilities, i.e.: • IASI provides high spectral resolution sounding and radiances which will improve NWP and ensures the availability of such a service in the mid-morning orbit. • Instruments with heritage from ESA Earth observation missions (ASCAT and GOME) are utilized operationally and will provide continuous observations over a period of at least 14 years. • With GRAS the radio occultation principle is introduced for the first time into an operational system and will demonstrate the capability of such system to provide high quality soundings in near real time. • The mission duration of 14 years will assure long-term and consistent observations that provide a sustained basis for improved utilization in NWP. Furthermore, it provides the basis for climate-monitoring which could be enhanced through a regular reprocessing of data and products considering scientific advance.
6
REFERENCES
Aminou, D. M. A., H. J. Luhmann, C. Hanson, P. Pili, B. Jacquet, S. Bianchi, P. Coste, F. Pasternak, and F. Faure, 2003: Meteosat Second Generation: A comparison of on-ground and on-flight imaging and radiometric performances of SEVIRI on MSG-1. Proc. 2003 EUMETSAT Meteorological Satellite Conference, Weimar, Germany, 29 September–3 October 2003, 236–243.
586
J. Schmetz et al.
Bühler, Y., G. Mason, J. Perez, D. Klaes, and T. Brefort, 2001: The EUMETSAT Polar System: Mission, system and programmatics. 52nd International Astronautical Congress, October 1–5, 2001, Toulouse, France. de Waard, J., W. P. Menzel, and J. Schmetz, 1992: Atlantic data coverage by Meteosat-3. Bull. Amer. Meteor. Soc., 73, 977–983. Holmlund, K., 1998: The utilization of statistical properties of satellite-derived atmospheric motion vectors to derive quality indicators. Wea. Forecasting, 13, 1093–1104. König, M., 2002: Atmospheric instability parameters derived from MSG SEVIRI observations. Technical Memorandum, 9, EUMETSAT Programme Development Department, pp. 27. Klaes, D., J. Schmetz, M. Cohen, J. Figa, J.-P. Luntama, R. Munro, P. Schlüssel, and A. Ratier, 2001: The EUMETSAT Polar System within the Initial Joint Polar System: Mission objectives, expected capabilities and products. 1st Post MSG User Consultation Meeting, Darmstadt, 13–15 November 2001, 38 pp. Klaes, D., M. Cohen, Y. Bühler, P. Schlüssel, R. Munro, J.-P. Luntama, A. von Engeln, E.O. Clérigh, H. Bonekamp, J. Ackermann, and J. Schmetz, 2007: An introduction to the EUMETSAT Polar System (EPS). Bull. Amer. Meteor. Soc., in Press. Laurent, H., 1993: Wind extraction from Meteosat water vapour channel image data. J. Appl. Meteor., 32, 1124–1133. Mason, B. and J. Schmetz, 1992: Meteorological satellites. Int. J. Remote Sens., 13, 1153– 1172. Rizzi, R., 1998: Simulation of IASI radiances in presence of clouds. Final Report to EUMETSAT, Contract Number EUM/CO/96/390/DD, 34 pp. Schlüssel, P. and M. Goldberg, 2002: Retrieval of atmospheric temperature and water vapour from IASI measurements in partly cloudy situations. Adv. Space Res., 29, 1703–2706. Schmetz, J., K. Holmlund, J. Hoffman, B. Strauss, B. Mason, V. Gaertner, A. Koch, and L. van de Berg, 1993: Operational cloud motion winds from METEOSAT infrared images. J. Appl. Meteor., 32, 1206–1225. Schmetz, J., P. Pili, S. Tjemkes, D. Just, J. Kerkmann, S. Rota, and A. Ratier, 2002: An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc., 83, 977– 992.
46 THE EVOLUTION OF THE NOAA SATELLITE PLATFORMS W. Paul Menzel NOAA-NESDIS, Madison, WI, USA
Abstract
The history and upcoming changes to the constellation of satellites operated by the National Environmental Satellite Data and Information Service (NESDIS) of the National Oceanic and Atmospheric Administration (NOAA) are described.
Keywords
NOAA, NESDIS, NPOESS, satellite, polar, geostationary
1
THE GLOBAL OBSERVING SYSTEM
In the last 40 years, a remote-sensing capability on polar and geostationary platforms has been established that has proven useful in monitoring and predicting severe weather such as tornadic outbreaks, tropical cyclones, and flash floods in the short term as well as climate trends indicated by sea surface temperatures, biomass burning, and cloud cover in the longer term. This has become possible first with the visible and infrared window imagery of the 1970s and has been augmented with the temperature and moisture sounding capability of the 1980s. Satellite imagery, especially the time continuous observations from geostationary instruments, dramatically enhanced our ability to understand atmospheric cloud motions and to predict severe thunderstorms. These data were almost immediately incorporated into operational procedures. Sounder data are filling important data voids at synoptic scales. Applications include temperature and moisture analyses for weather prediction, analysis of atmospheric stability, estimation of tropical cyclone intensity and position, and global analyses of clouds. Polar orbiting
587 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 587–600. © United States Government 2007.
588
W. P. Menzel
microwave measurements help to alleviate the influence of clouds for all weather soundings. Geostationary depiction of temporal and spatial changes in atmospheric moisture and stability are improving severe storm warnings. Atmospheric flow fields (wind field composites from cloud and water vapour drift) are helping to improve hurricane trajectory forecasting. Applications of these data also extend to the climate programmes; archives from the last twenty years offer important information about the effects of aerosols and greenhouse gases and possible trends in global temperature. In the next decade more improvements will be realized. The anticipated accuracies, resolutions, and cycle times for some of the meteorological parameters derived from satellite systems planned for this decade are summarized in Table 1; these include temperature and humidity profiles (from infrared and microwave radiometers) and upper winds (from tracking the movement of cloud and water vapour features). Infrared radiometers provide the highest quality profiles but only in clear sky conditions. Microwave radiometers provide data under cloudy conditions but have lower vertical resolution. Observations from both are currently widely available over the oceans, but over land varying surface emissivity is currently limiting soundings to the upper atmosphere. Upper wind observations are provided on a global basis but only where suitable tracers are available and usually only at one level in the vertical. Improvements in resolution and capability are expected over the next 5–10 years; better utilization of sounding data over land is also near at hand. As the space based remote-sensing system of the future develops and evolves, four critical areas (all dealing with resolution) will need to be addressed in order to achieve the desired growth in knowledge and advanced applications. They are: (1) spatial resolution – what picture element size is required to identify the feature of interest and to capture its spatial variability; (2) spectral coverage and resolution – what part of the continuous electromagnetic spectrum at each spatial element should be measured, and with what spectral resolution, to analyse an atmospheric or surface parameter; (3) temporal resolution – how often does the feature of interest need to be observed; and (4) radiometric accuracy – what signal to noise is required and how accurate does an observation need to be? Each of these resolution areas should be addressed in the context of the evolving space based observing system wherein the satellite(s) exist, or will exist. Higher temporal resolution is becoming possible with detector array technology; higher spatial resolution may come with active cooling of infrared detectors so that smaller signals can be measured with adequate signal to noise. Higher spectral resolution is being demonstrated through the use of interferometers and grating spectrometers. Advanced microwave radiometers measuring moisture as well as temperature profiles are being introduced in polar orbit; a geostationary complement is being investigated. Ocean colour observations with multispectral narrow band visible measurements are being
The Evolution of the NOAA Satellite Platforms
589
studied. Active sensors are being planned to supplement passive sensors with measurements of ocean height and atmospheric motions. The challenge of the future is to further the progress realized in the past decades so that environmental remote sensing of the oceans, atmosphere, and earth increases our understanding the processes affecting our lives and future generations. Table 1. Anticipated accuracies, resolutions, and cycle times for some of the meteorological parameters derived from satellite systems planned for this decade. Note that (1) the horizontal and vertical interval refers to the sampling distance between consecutive measurements, (2) cycle is the time interval between measurements and assumes two satellites in polar orbit, (3) delay is the time delay between the observation and receipt of data by the end user, (4) satellite winds are single level data with vertical sampling typically ~1 km, and (5) for ATOVS, 1–4 profiles are provided per second per satellite (per 100 km2). Element and Instrument
Humidity Sfc – 300hPa NOAA & METOP (ATOVS)6 METOP (IASI) & NPOESS (CrIS) Humidity below 50hPa METOP (GRAS) & NPOESS (GPSOS) Temperature Sfc – 10hPa NOAA & METOP (ATOVS) 6 METOP (IASI) & NPOESS (CrIS) Temperature 500hPa – 10hPa METOP(GRAS) & NPOESS (GPSOS) Wind Sfc – 200hPa MSG (SEVIRI) & GOES (Imager)
2
Horiz Res1 (km)
Vert Res2 (km)
Cycle 3
Delay4
Accuracy
(h)
(h)
(rms)
50 15
3 1
6 12
2 2
15% 10%
500
~1
12
2
10%
50 15
3 1
6 12
2 2
1.5 K 1K
500
~1
12
2
1K
50/100
One level5
1
2
2–5 m s–1
MEETING REMOTE-SENSING REQUIREMENTS IN THE NEXT TWO DECADES
Monitoring of the Earth’s environment has become an international endeavor. No one country has the observational systems necessary to provide the data it needs for its environmental monitoring and prediction operations. In the last decade and more so in the next decade, satellite remote-sensing contributions to the Global Observing System are being made by an increasing number of international partners. The collaboration and coordination among the international satellite community continues to increase. The demands for environmental data are enormous, ranging across all components of the Earth system – atmosphere, oceans, and land. The data
590
W. P. Menzel
requirements cover a broad range of spatial and time scales – 100s of metres and minutes to global, seasonal and inter-annual to decadal and centennial time frames. A number of research satellites are starting to provide advanced observations of the Earth and its atmosphere; these instruments will provide data on atmospheric, ocean, and land surface conditions with accuracy and spatial resolutions never before achieved. It is not uncommon for today’s research satellites to achieve lifetimes of several years, and many nonoperational space agencies are becoming increasingly aware of the importance of utilization of their satellite’s data streams. It is not unrealistic to foresee a time when there will be more and more of these special data made available for operational uses (blurring the distinction between research and operational platforms). Data from satellites are making contributions not only in the weather forecasting arena, but also in the fields of climate and ocean research. Satellite remote sensing is beginning to establish the level of continuity and calibration in their worldwide observations so that it will be possible to understand the physical, chemical, and biological processes responsible for changes in the Earth system on all relevant spatial and time scales. Coastal and ocean satellite remote-sensing services are expanding. A new generation of improved resolution, coastal and ocean remote-sensing capable satellites is rapidly emerging. For the first time ever, one can envision an operational coastal and ocean remote-sensing system that will assist marine fisheries management, coastal planning, and environmental quality stewardship. In the past, the satellite systems evolved mainly from a series of technology demonstrations. In the current planning, improvements in satellite remote sensing are being driven by user requirements for improved measurements and products. While these will rely on new technology demonstrations, the push is coming from user requirements more than opportunities to realize new technological capabilities. Finally, it is important to note that satellites are but one component of an Integrated Global Observing System (IGOS). In situ observations from a variety of instruments on balloons, planes, ships, buoys, and land surfaces are the other part of an IGOS. The challenge is to identify the best mix of satellite and in situ observations that will meet environmental monitoring and prediction requirements. This is a daunting task that must build upon the current systems and anticipate the future systems.
3
CURRENT AND FUTURE POLAR PLATFORMS
Polar orbiters allow a global coverage to be obtained from each satellite twice a day. To provide a reasonable temporal sampling for many applications at least two satellites are required, thereby providing 6-hourly coverage. A backup capability exists by reactivating “retired” platforms and this has been demonstrated recently. Since 1979, coverage with two polar orbiting
The Evolution of the NOAA Satellite Platforms
591
satellites has been achieved most of the time. The orbital altitude of 850 km makes it technically feasible to make high spatial resolution measurements of the atmosphere/surface. Current operational polar orbiters include the NOAA series from the USA and the METEOR, RESURS, and OKEAN series from Russia and the FY-1 series from China. They provide image data that can be received locally. The NOAA satellites also enable generation of atmospheric sounding products that are disseminated to NWP centres on the GTS. In the future, the NOAA AM satellite will be replaced by the METOP satellites provided by EUMETSAT and the NOAA PM satellite will transition to the NPOESS series. Instruments that have been or will soon be a part of the polar orbiting series of satellites include: visible and infrared radiometers, atmospheric temperature and humidity sounding, microwave all-weather radiometers, ozone monitoring, scatterometers, radiation budget, and positioning sensors. Table 2. VIIRS channel number, wavelength (µm), and primary application. Bands I are sensed with a spatial resolution of 400 m and bands M at 800 m. The signal to noise ratio in the reflective bands ranges from 25 to 1,000; the noise equivalent temperature difference in the emissive bands ranges from 0.03 to 0.4 K (larger values for the I bands). Ch number M1 M2 M3 M4 I1 M5 M6 M7 I2 M8 M9 M10 I3 M11 M12 I4 M13 M14 M15 I5 M16
Wavelength (µm) Reflective Bands 0.412 0.450 0.488 0.555 0.630 0.672 0.751 0.865 0.865 1.24 1.378 1.61 1.61 2.26 Emissive Bands 3.7 3.74 4.05 8.55 10.8 11.55 12.0
primary application ocean colour/ aerosol ocean colour/ aerosol ocean colour / aerosol ocean colour / aerosol imagery ocean colour / aerosol atm corr atm corr NDVI cld particle size cirrus snow fraction snow map clouds SST imagery / clouds SST / fires cld top properties SST cloud imagery SST
592
W. P. Menzel
3.1 Visible and infrared radiometers The Advanced Very High Resolution Radiometer (AVHRR), flown in October 1978 on TIROS N, measures radiation in five visible and IR windows at 1 km resolution. This will transition to a more capable visible and infrared imager called the Visible Infrared Imaging Radiometer Suite (VIIRS, see Table 2), when the NOAA satellites become the NPOESS series, starting with a demonstration program in 2006, called the NPOESS Preparatory Project (NPP). VIIRS will be better calibrated than the AVHRR, have higher spatial resolution (400 m vs. 1 km at nadir), and have additional spectral capability including channels that can be utilized to determine ocean colour. Parameters that may be derived from the VIIRS for use in operational as well as climate monitoring include sea surface temperature, aerosols, snow cover, cloud cover, surface albedo, vegetation index, sea ice, and ocean colour.
3.2 Atmospheric temperature and humidity sounding An important development was the remote sounding of vertical temperature and humidity profiles in the atmosphere on a worldwide basis with the TIROS Operational Vertical Sounder (TOVS). TOVS has evolved to an advanced version in 1998 and consists of the High resolution Infrared Radiation Sounder (HIRS) and the Advanced Microwave Sounding Unit (AMSU). These IR and microwave sounders can produce soundings in clear and cloudy (non-precipitating) skies every 50 km. NOAA will be transitioning to more capable sounders in the NPOESS series, starting with a demonstration program in 2006 on NPP. HIRS will be replaced by the Cross Track Infrared Sounder (CrIS), a Michelson interferometer that is designed to enable retrievals of atmospheric temperature profiles at 1 degree accuracy for 1 km layers in the troposphere, and moisture profiles accurate to 15% for 2 km layers. This is accomplished by the CRIS working together with the Advanced Technology Microwave Sounder (ATMS), being designed to be the next generation cross track microwave sounder. Comparable sounding capability will be realized on the METOP series by the Infrared Atmospheric Sounding Interferometer (IASI) in conjunction with the advanced microwave temperature sounding units (AMSU-A) and microwave humidity sounders (MHS/HSB). CrIS/ATMS will fly on afternoon (1330 ascending) and IASI/AMSU/MHS will fly in the morning (0930 descending) orbit.
3.3 Microwave all-weather radiometers A complementary series of DMSP satellites in polar orbit fly a scanning microwave radiometer called the Special Sensor Microwave Imager (SSM/I), flown since June 1987. These provide night–day, all-weather imaging of the
The Evolution of the NOAA Satellite Platforms
593
land and ocean surface because of the ability of microwave radiation to penetrate clouds. NOAA has used the DMSP SSM/I data extensively. A conical scanning version of the microwave sounder will be flown on NPOESS. The Conical Microwave Imager Sounder (CMIS) will combine the microwave imaging capabilities of Japan’s Advanced Microwave Scanning Radiometer (AMSR) on EOS PM-1, and the atmospheric sounding capabilities of the Special Sensor Microwave Imager/Sounder (SSMI/S) on the current DMSP satellites. Polarization for selected imaging channels (vertical, horizontal, and +/– 45 degrees) will be utilized to derive ocean surface wind vectors similar to what has previously been achieved with active scatterometers. Although demonstrated on airborne platforms, space based validation of the passive microwave technique for wind vector derivation is being tested in the Windsat Coriolis mission in 2003. CMIS data can be utilized to derive a variety of parameters for operations and research including all weather sea surface temperature, surface wetness, precipitation, cloud liquid water, cloud base height, snow water equivalent, surface winds, atmospheric vertical moisture profile, and atmospheric vertical temperature profile.
3.4 Monitoring ozone Another important sounding approach used the ultraviolet portion of the electromagnetic spectrum to sound atmospheric ozone. The Solar Backscatter Ultraviolet (SBUV), which provides information on ozone amounts for atmospheric 7–10 km layers, was incorporated into the operational series of NOAA polar satellites (POES) beginning with NOAA-9 in 1984. The Total Ozone Mapping Spectrometer (TOMS) flew on Nimbus-7 and provided critical image data that first identified the Antarctic ozone hole, but it has not been made operational. The Nimbus-7 TOMS lasted into the 1990s and was replaced subsequently by TOMS sensors flying on a Russian Meteor spacecraft, the Japanese ADEOS, and a NASA Earth Probe. The TOMS equivalent capability will be continued with the flight of the Dutch provided Ozone Mapping Instrument (OMI) on NASA’s Chemistry mission in 2004 and subsequently the Ozone Mapping and Profiler Suite (OMPS) on NPOESS, being developed for flight on afternoon (1330 ascending) NPOESS platforms. It consists of a nadir scanning ozone mapper similar in functionality to TOMS and a limb scanning radiometer that will be able to provide ozone profiles with vertical resolution of 3 km. Depending upon its ultimate design, the OMPS may be able to provide some of the same capability as limb scanning sensors on NASA’s UARS and EOS Chem. However in the near term, there is concern about a possible gap in TOMS type data coverage.
594
W. P. Menzel
3.5 Scatterometers The first active radar scatterometer to determine wind speed and direction over the ocean surface was flown on Seasat in 1978. Great progress in this area was possible in the 1990s with the SCAT data on ERS-1 and ERS-2; the SCAT on ERS-2 is still providing quasi-operational data. A NASA scatterometer termed NSCATT flew on Japan’s ADEOS-1 from August 1996 to June 1997; scientists were able to show a significant positive impact in predicting marine forecasting, operational global numerical weather prediction, and climate forecasting. A follow-on mission, Quickscat, launched in 1999, carries the NSCATT successor instrument, Seawinds. Another Seawinds sensor flew on ADEOS-2 in 2002. No additional US scatterometer missions are planned before NPOESS, which plans to use a passive microwave approach to determining the ocean vector wind field. This passive microwave technique is being tested as part of the Windsat Coriolis mission in 2003. Europe’s METOP series of satellites, scheduled to begin flying in 2005 include an Advanced Scatterometer (ASCAT) sensor, but ASCAT alone may not be able to provide the required geographic coverage and frequency of observation needed for operations and research. The Japanese are planning to carry a Seawinds follow-on provided by NASA on GCOM early next decade.
3.6 Radiation budget The Earth’s radiation budget and atmospheric radiation from the top of the atmosphere to the surface will be measured by the Clouds and Earth Radiant Energy System (CERES) on its afternoon (1330 local time ascending orbit) NPOESS platforms. The predecessor Earth Radiation Budget (ERB) sensors flew on Nimbus in 1978, as well as on a free flyer and on NOAA-9 and NOAA-10 in the mid-1980s. The first CERES is currently flying on the Tropical Rainfall Measuring Mission (TRMM) which was launched in November 1997. Two CERES scanners (one each working in the biaxial and cross track mode) are in orbit with EOS Terra since December 1999 and EOS Aqua since May 2002.
3.7 Altimetry Altimeters flew on the European ERS-1 and ERS-2 satellites in the 1990s and provided a major quasi-operational contribution. NOAA is planning to manifest a dual frequency microwave radar altimeter for its morning (0530 descending) NPOESS platforms. The type of altimeter, realized with JASON-1 in 2001, measures the ocean topography which provides information on the ocean current velocity, the sea level response to global warming/cooling and hydrological balance, the marine geophysical processes (such as crustal deformation), and the global sea state.
The Evolution of the NOAA Satellite Platforms
595
3.8 Positioning sensors Geometric determinations of location depend on inferences about the atmospheric temperature and moisture concentrations; they provide valuable complementary information to tropospheric infrared and microwave sounders about the tropopause and stratosphere. Ray bending and changes in the phase and amplitude of the transmitted signals allowing inference of the upper atmosphere temperature profile to the order of 1 K or better between altitudes of 8–30 km in layers (with footprints ranging between 1 × 30 km2 and 1 × 200 km2 extent) with near global coverage. The coverage would be expected to be evenly spread over the globe, excepting polar regions. The system measures upper atmospheric virtual temperature profiles so data from the lower atmosphere would require alternate data to separate vapour pressure and temperature traces. The Global Positioning System Occultation Sensor (GPSOS) will measure the refraction of radiowave signals from the GPS constellation and Russia’s Global Navigation Satellite System (GLONASS). This uses occultation between the constellation of GPS satellite transmitters and receivers on LEO satellites. The GPSOS will be used operationally for spacecraft navigation, characterizing the ionosphere, and experimentally to determine tropospheric temperature and humidity. A similar system, GPSMET, flew in 1995. NOAA is planning to manifest a GPSOS on all NPOESS platforms. A promising research GPS system is COSMIC (Constellation Observing System for Meteorology, Ionosphere and Climate). The National Space Program Office (NSPO) in China, the University Corporation for Atmospheric Research (UCAR), the Jet Propulsion Laboratory (JPL), the Naval Research Laboratory (NRL), the University of Texas at Austin, the University of Arizona, Florida State University and other partners in the university community are developing COSMIC, a project for weather and climate research, climate monitoring, space weather, and geodetic science. COSMIC plans to launch eight LEO satellites later this decade, each COSMIC satellite will retrieve about 500 daily profiles of key ionospheric and atmospheric properties from the tracked GPS radio-signals as they are occulted behind the Earth limb. The constellation will provide frequent global snapshots of the atmosphere and ionosphere with about 4000 daily soundings (see: http://www.cosmic.ucar.edu/).
4
CURRENT AND FUTURE GEOSTATIONARY PLATFORMS
The geosynchronous orbit is over 40 times higher than a polar orbit, which makes measurements technically more difficult from geostationary platforms. The advantage of the geostationary orbit is that it allows frequent measurements over the same region necessary for now-casting applications
W. P. Menzel
596
and synoptic meteorology. A disadvantage is that a fixed full disk view of the Earth is viewed from one satellite. Thus, five equally spaced satellites around the equator are needed to provide global coverage; polar regions are reviewed poorly at large zenith angles. Currently, there is global coverage from geostationary orbit (>5 operational satellites for image data and products (e.g., cloud motion winds) and two satellites are providing a sounding capability as well. Reactivating “retired” platforms provides backup and there have been several examples of this. Some of the satellites provide a real-time reception capability to allow immediate access to the imagery for real-time applications. Products are disseminated on the GTS by the satellite operators for near real-time applications. Instruments that have been or will soon be a part of the geostationary series of operational satellites include: Table 3. Advanced Baseline Imager Spectral Bands and Objectives. Spatial resolution is 2 km infrared and 0.5 km visible, full disk coverage will require 5 min. Future GOES Imager (ABI) Band 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Wavelength Range (µm)
Central Wavelength (µm)
Sample Objective(s)
0.45-0.49 0.59-0.69 0.84-0.88 1.365-1.395 1.58-1.64 2.235 - 2.285 3.80-4.00 5.77-6.6 6.75-7.15 7.24-7.44 8.3-8.7 9.42-9.8 10.1-10.6 10.8-11.6 11.8-12.8 13.0-13.6
0.47 0.64 0.86 1.38 1.61 2.26 3.90 6.19 6.95 7.34 8.5 9.61 10.35 11.2 12.3 13.3
Daytime aerosol-over-land, Color imagery Daytime clouds fog, insolation, winds Daytime vegetation & aerosol-over-water, winds Daytime cirrus cloud Daytime cloud water, snow Day land/cloud properties, particle size, vegetation Sfc. & cloud/fog at night, fire High-level atmospheric water vapor, winds, rainfall Mid-level atmospheric water vapor, winds, rainfall Lower-level water vapor, winds & SO 2 Total water for stability, cloud phase, dust, SO 2 Total ozone, turbulence, winds Sfc. & cloud Total water for SST, clouds, rainfall Total water & ash, SST Air temp & cloud heights and amounts
4.1 Visible and infrared radiometers The Visible and Infrared Spin Scan Radiometer (VISSR), flown since 1974, has been the mainstay of geostationary imaging on GOES, Meteosat, and GMS. VISSR enabled 5–7 km images of the full earth disk every 30 min in two or three visible and infrared windows and one water vapour sensitive band. The USA changed to a staring Imager with 5 channels of visible and infrared measurements at 5 km resolution with full disk coverage in 30 min in 1993. More changes are underway. The USA will evolve to an Advanced Baseline Imager (ABI, see Table 3) that makes full disk images in 16 spectral bands in 5 min at 2 km infrared and 0.5 km visible resolution. ABI offers improved performance over current GOES in all dimensions (routine
The Evolution of the NOAA Satellite Platforms
597
full Earth disk imaging while enabling mesoscale sub 1 min interval imaging, better navigation, more noise free signals, and additional spectral bands for improved moisture feature detection).
4.2 Infrared sounding With the three axis stable platform on GOES-8, NOAA was able to introduce geostationary infrared sounders. Measuring the infrared radiation in 18 spectral bands, these sounders provide temperature and moisture sounding over North America and nearby oceans every hour every 30 km (in clear skies). A variety of products and applications are described in the literature. NOAA plans to evolve to the Hyperspectral Environmental Sounder (HES) in 2012, using an interferometer, focal plane detector arrays, and on board data processing to cover 3.7–15.4 µm with 2,000 plus channels measuring radiation from 10 km resolution; contiguous coverage of 6,000 × 5,000 km2 will be accomplished in less than 60 min. NASA is planning to demonstrate the technology necessary for HES, with the Geostationary Imaging Fourier Transform Spectrometer (GIFTS) in 2008. GIFTS will improve observation of all three basic atmospheric state variables (temperature, moisture, and wind velocity) with much higher spatial, vertical, and temporal resolutions. Water vapour, cloud, and trace gas features will be used as tracers of atmospheric transport. GIFTS observations will improve measurement of the atmospheric water cycle processes and the transport of greenhouse and pollutant gases. GIFTS and HES represent a significant advance in geostationary sounding capabilities and brings temporal and horizontal and vertical sounding resolutions into balance for the first time ever.
5
THOUGHTS ON THE FUTURE GLOBAL OBSERVING SATELLITE SYSTEM
At present, soundings of temperature and humidity are primarily the domain of the polar-orbiting meteorological satellite constellation. The reason for this is historical. Sounding instruments were first developed for the polarorbiting satellites since they flew closer to the Earth and provided a more complete global coverage. However, the present user requirement is for hourly soundings that cannot be satisfied with the present constellation of polar-orbiting satellites. Additionally, there is already proven technology for soundings from geostationary orbit. Finally, there are firm plans to continue at least some of the geostationary satellites with a sounding capability. Thus, soundings of temperature and humidity should be provided from both constellations of satellites. There are several user requirements for wind vector over the ocean surface. The technology has been proven for well over a decade. There are
598
W. P. Menzel
plans to fly operational scatterometers on operational polar-orbiting satellites. Thus, the operational polar-orbiting satellites should have the capability to provide surface wind vectors over the ocean. To assure continuity, as well as to provide sufficient overlap between geostationary satellites, there is a need for at least six geostationary satellites. The concept of requiring a satellite to have the capability to make several different types of concurrent observations, e.g., soundings, imagery and scatterometry from the same polar-orbiting satellite, should be reviewed. It is possible that a series of smaller single purpose satellites would be more cost effective. Although not yet implemented on operational satellites, the concept has been successfully demonstrated on experimental and single satellite missions. Experimental observation satellites pose unique challenges for open and timely access to experimental data in standard formats, preparation of the community for new data usage, and data continuity. It is expected that a set of guidelines will be developed and agreed upon by the satellite operators. Assuming that such assurances can be obtained, a new constellation of satellites could be added to the space-based GOS. Certainly, user requirements exist in abundance for parameters not provided by the meteorological satellites including, but not limited to aerosol, cloud ice, cloud water, ozone and other trace gas profiles, land cover, land surface topography, ocean wave period and direction, ocean topography, ocean colour, significant wave height, snow water equivalent, soil moisture, and vegetation type. There is now a convergence of needs since the R&D satellite operators have also shown a keen interest for operational evaluations of their new data. The existence of experimental satellite missions capable of measuring these as yet unsatisfied requirements provides ample proof of the availability of technology and plans – although not necessarily for a continuous series of satellites. Thus, the space-based GOS could add a constellation of experimental satellites covering several different mission areas such as oceanographic, atmospheric chemistry, high-resolution land use and hydrological. Such a constellation would probably require a variety of mission oriented polarorbiting experimental satellites. Coordination of equator crossing times and geostationary positions, frequency allocations for communications, standard data formats, and open data policy remain challenges for the global community when planning the future GOS. Thus it is envisioned that the space based component of the GOS in 2015 will involve • quantitative measurements for input to 4-D continuous global data assimilation systems, • synergistic multi-satellite / multi-sensor systems,
The Evolution of the NOAA Satellite Platforms
• • • • •
599
atmospheric state variables (T,q,V) with the required accuracy and frequency, multi-level cloud and H2O imagery, small-satellite GPS occultation density profiles, higher spatial resolution multispectral (hyperspectral) land and ocean surface observations, air quality (CO and O3) monitoring.
User requirements in several applications areas (including numerical weather prediction) indicate the need for a four-polar and six-geostationary satellite system. Microwave, altimetry, scatterometry, radio occultation, and lidar systems remain unique to the low earth orbiting satellites. In the evolved GOS, (a) sounding will be accomplished with combined radiometric (infrared and microwave) and geometric (radio-occultation) systems, (b) passive and active remote sensing are combined to offer the best measurement of water vapour at resolutions commensurate with its variability in nature (c) altimetry will be pursued with a two-orbit system fully operational with real time capability wide swath (non-scanning) altimeters to enhance mesoscale capabilities, (d) atmospheric wind profiles will be accomplished with Doppler lidar systems, (e) ocean surface wind vectors now achieved with active techniques will be derived from passive measurements, (f) SST will evolve from a combined LEO and GEO systems of measurements, (g) expanded ocean colour capabilities will include increased horizontal resolution for coastal areas, and (h) SAR will belong to a multi-satellite system with a “wave mode” and sea ice/wave monitoring service. Expansion of the space based component of the GOS will be an international collaboration. There will be efforts to facilitate contributions of single instruments to larger platforms or flying in formation; the latter will mitigate the need for launching the full platform upon the loss of one critical instrument. Replacement strategies of the current or near future GOS satellites by the next generation satellites will proceed with a phased implementation approach. The role of small satellites in the GOS will be expanded. Coordination of international contributions to the polar orbiting observing system to achieve optimal spacing for a balance of spectral, spatial, temporal, and radiometric coverage will be a goal. Operational continuation of research capabilities with proven utility to the GOS will occur as much as possible without interruption of the data flow. There will be a commitment for adequate resources to sustain research developments necessary for improved utilization of these measurements. As much as possible, preparation for utilization of a given new measurement will begin prior to launch with distribution of simulated data sets that test processing systems; this will improve the percentage of the instrument post
600
W. P. Menzel
launch lifetime that is used operationally (the current 6 months to 2 years of post launch familiarization will be reduced). International algorithm development will assure best talent participation and enhance uniformity in derived products.
47 JAPAN’S ROLE IN THE PRESENT AND FUTURE SATELLITE OBSERVATION FOR GLOBAL WATER CYCLE RESEARCH Riko Oki and Yoji Furuhama Japan Aerospace Exploration Agency (JAXA), Japan
1
INTRODUCTION
1.1 Background It is said that the year-by-year increase of the effect of global warming has recently been felt as reality. Global observations from satellites have increased their importance, and have long been recognized indispensable in Japan as well. As a worldwide activity, the implementation plan of the World Summit on Sustainable Development (WSDD) that was held in 2002 in Johannesburg refers to the promotion of global water cycle research and observation including satellite remote sensing. Under such circumstances, the Council for Science and Technology Policy (CSTP), which is the organization to make decisions on governmental policy on science and technology in the Cabinet Office in Japan, has summarized in 2002 a report on the long-term vision of the future development and application of space activities. The report says that Japan should give priority to Earth environment monitoring along with national security, information and telecommunications, and positioning in the future space development and applications. In the area of Earth environment monitoring and observation, the report emphasizes that continuous observations of carbon dioxide and global water cycle should be realized with Earth observing satellites to deepen our understanding of earth science. The Space Activities Commission (SAC) also summarized in September 2003 a long-term plan of space development in which observation of global warming and water cycle was selected as a weighty program.
601 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 601–609. © 2007 Springer.
602
R. Oki and Y. Furuhama
JAXA has developed satellites and space-borne sensors for environment monitoring and measurement. For example, the Advanced Earth Observing Satellite (ADEOS) was launched in 1996, and ADEOS-II in 2002. The Precipitation Radar (PR) on NASA’s Tropical Rainfall Measuring Mission (TRMM) satellite and AMSR-E (the Advanced Microwave Scanning Radiometer for the Earth Observing System) on NASA’s Aqua satellite have been providing invaluable data for atmospheric science and other research areas. Related to the above mentioned trend of Japan’s policy, study for satellite development (so-called phase B study) has started for GOSAT (Greenhouse Gas Observation Satellite) and GPM (Global Precipitation Measurement) missions recently in 2003 and 2004, respectively.
1.2 Scientific and social needs for observation of global water cycle by satellites One of the major concerns of the present day is the global change of our environment like global warming. To assess the effect of changes in each environmental factor on the global environment, we need to know its present status, relationship to other factors, and the mechanism of how it affects the environment. Among the effects of global warming, changes in global water cycle and in the distribution of water resources as a consequence of the former should be emphasized because of their close connection with our life. In a sense, monitoring global water cycle is more important than monitoring temperature environment. In fact, changes in global water cycle are not caused only by global warming. Among many environmental factors related to the global water cycle, precipitation is one of the most important components, because it affects everyone’s life and work. Too much precipitation causes floods, and too less of it causes droughts. Agricultural production depends on precipitation. It is also a true global variable that determines the general circulation through latent heating and reflects climate changes. It is a key component of air–sea interaction and eco-hydrometeorological modeling. Japan is located in the east edge of Asia where the Asian monsoon affects directly her weather and climate. Monitoring the present status and predicting the variation of global water cycle and the Asian monsoon have been an important challenge to us. Even though precipitation is such an important component of our environment, it is one of the least known physics components of cloud–weather–climate prediction models. Because of its large variability in space and time, its distribution over the globe is not very accurately known. We can never obtain accurate global rain distribution data by rain gauges and ground-based rain radars alone. Knowledge of the spatial and temporal distribution of global precipitation is a key to improve our understanding of weather and climate systems.
Japan’s Role in the Present and Future Satellite Observation
603
Japan in collaboration with the USA launched the TRMM satellite in 1997 to investigate tropical rainfall. TRMM was a pilot project to measure rain from space and the world’s first space-borne precipitation radar technology was demonstrated. TRMM’s concept, that is, precipitation observation by radar in combination with a microwave radiometer, will be succeeded by the Global Precipitation Measurement (GPM). It is also important to observe 3-dimensional (3D) cloud structure globally by active sensors in the next step to improve cloud-weather-climate models.
2
TRMM
The TRMM satellite was launched in November 1997 and has collected more than a 6-year record of tropical and subtropical rain. Great strides in rain observation technology by satellite sensors were made in the TRMM mission. The results can be summarized in two points. Firstly, it carries the world’s first space-borne precipitation radar (PR developed by Japan) that enabled measurement of the detailed vertical structure of rain systems uniformly over the globe. Secondly, by combining the PR data with data simultaneously obtained by the visible and infrared scanner (VIRS) and the TRMM microwave imager (TMI), the uncertainties in rain estimates that originated in measurement principles and sensor performances themselves have been greatly reduced. Many scientific results were obtained by analyzing PR data in Japan, and major results can be summarized as follows: • • • • • • • •
Substantial increase in quantitative measurements of rain distribution in tropical and subtropical areas. Accurate rain measurements over ocean and land in nearly equal quality (PR) for a long time period. Impact of PR observation to TMI algorithms, resulting in more physically consistent and quantitative estimates of rain rates. Revealing diurnal, annual, and long-term variations of precipitation. Observation of 3D rain structure. Improvement of short-term weather forecasting by 4D data assimilation with TMI data. Estimation of sea surface temperature from TMI data. Estimation of soil moisture from the characteristics of surface cross sections measured by PR.
The success of the PR measurements that is proven by the achievements exemplified above and the effectiveness of simultaneous measurements by radar and microwave radiometer will be inherited by GPM, the next generation mission, and help to realize a further developed observation system.
R. Oki and Y. Furuhama
604
3
AQUA/AMSR-E AND ADEOS-II/AMSR
The Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) was developed and provided to NASA’s EOS Aqua satellite by the National Space Development Agency of Japan (NASDA) as one of the indispensable instruments for Aqua’s mission (Kawanishi et al. 2003; Shibata et al. 2003). AMSR-E is a modified version of AMSR that was launched in December 2002 aboard the Advanced Earth Observing Satellite-II (ADEOS-II). Both of them are dual-polarized total-power passive microwave radiometers that observe water-related geophysical parameters to clarify the mechanism of global water and energy circulation. The frequency bands include 6.925, 10.65, 18.7, 23.8, 36.5, and 89.0 GHz. AMSR also has 50.3 and 52.8 GHz bands. The hardware improvements over the existing space-borne microwave radiometers for Earth imaging include the largest main reflector of its kind and the addition of 6.925-GHz channels that has been desired for a long time. These improvements provide finer spatial resolution and the capability to retrieve sea surface temperature and soil moisture information on a global basis. In addition to the variables such as water vapor, precipitation, and sea surface wind speed, that have been proved to be measured with a radiometer, AMSR and AMSR-E can retrieve novel geophysical parameters, including sea surface temperature (SST) and soil moisture, by using new frequency channels. Near-real-time products will be used to investigate satellite data assimilation into weather forecasting models and to contribute to improve forecasting accuracy. Very high spatial resolution of AMSR and AMSR-E improves precipitation retrievals, because it lessen the effect of nonlinear response (i.e., beam filling problem) of the brightness temperatures to the total rain amount within the footprint. Many applications over land as well as sea ice investigation also benefit from this improved spatial resolution. Although TMI realized the higher spatial resolution thanks to the lower orbit altitude of the TRMM, AMSR-E improves the resolution by using a large antenna and extends its capability to global measurements. Unfortunately, the observation with AMSR abruptly terminated in less than one year after launch because of the unexpected accident of ADEOS-II. The possibility of deploying an AMSR-like instrument in space is being discussed in order to restart a ADEOS-II follow-on mission as early as possible, because it is believed necessary to collect climate data at least for 15 years continuously to study the variation of Earth environment.
4
GPM
Based on the success of TRMM, a follow-on mission was proposed both in Japan and the USA. It is the Global Precipitation Measurement (GPM) that is planned to start in approximately 2008. Feasibility studies, related science,
Japan’s Role in the Present and Future Satellite Observation
605
and mission planning have been carried out in both countries. Details of GPM are described in the article by E. Smith et al. in this book and Furuhama et al. (2002). Only the significance of the mission from Japanese point of view is reaffirmed in this article. GPM will expand the TRMM mission in two ways. Firstly, GPM will make accurate measurement of precipitation that was established with the TRMM PR. Secondly, by international collaboration it will achieve frequent measurement of precipitation that was not realized by the TRMM satellite alone. In these two points, the GPM mission is not a single satellite program, but a program that utilizes the information from multiple of satellites in international collaboration. The entire GPM system consists of the core satellite and a fleet of constellation satellites each of which carries a microwave radiometer for precipitation measurement. The core satellite will carry the dual-frequency precipitation radar (DPR) which is an improved version of the TRMM PR and the GPM microwave imager (GMI). The constellation satellites will join GPM by providing microwave radiometer data in international partnership. The frequent and accurate global precipitation measurement will be realized with this whole system. GPM is expected to give answers to many questions. For example, how are the rainfall and rainfall structure responding to changes in the Earth’s temperature and other climate variables? How directly is the surface hydrology coupled to the rainfall and evaporation? We need to observe, understand, and model the Earth’s system to learn how it is changing, and consequences for life on Earth. To do so, we need to establish the existence of trends in the rate of global water cycle. Acceleration would lead to faster evaporation, increased global average precipitation, and a general increase in extremes, particularly droughts and floods. In addition, GPM may impact many other surrounding research areas. They included studies in cloud system and radiation, ocean–land atmosphere interactions, freshwater forcing on ocean processes, ocean salinity modeling, development of hydrometeorology and carbon assimilation models, soil moisture and its impact on flood/drought prediction, and water vapor transport, to mention a few.
4.1 DPR measurement Precipitation measurement with higher accuracy than TRMM is expected with the DPR because the DPR should be capable of differentiating solid particles from liquid particles and providing some information of the drop size distribution (DSD). The DPR consists of Ku-band (13.6 GHz) and Kaband (35.5 GHz) precipitation radars. The minimum detectable rainfall rate will be improved to 0.2 mm h–1 by the Ka-band radar. Table 1 shows the main characteristics of the DPR. Accurate rainfall estimates will be provided by a dual-frequency algorithm using the matched beam data observed simultaneously by the two radars. Figure 1 shows the concept of dual-frequency
R. Oki and Y. Furuhama
606
measurement of precipitation. Because the two radars have different dynamic ranges, the Ka-band radar will be used to measure snow and light rain whereas the Ku-band radar will measure light to heavy rain. In the overlapped dynamic range, DSD information and accurate rainfall estimates will be provided by a dual-frequency algorithm. The reader should refer to Nakamura and Iguchi (2007), and Adhikari and Nakamura (2003) for the details of the DPR and the rain profiling algorithms. Table 1. Main characteristics of DPR. Frequency Swath width Horizontal resolution Tx pulse width Range resolution Observation range Tx peak power Minimum detectable rainfall rate * Measurement accuracy Beam-mating accuracy Data rate Mass Power consumption Size
KuPR 13.597/13.603GHz 245 km 5 km (at nadir) 1.6 µs (×2) 250 m 18 km to –5km ASL >1000 W 0.5 mm h–1 (18 dBZ) within ±1 dB <1000 m <112 kbps <370 kg <380 W 2.4 × 2.4 × 0.6 m
KaPR 35.547/35.553 120 km 5 km (at nadir) 1.6/3.2 µs (×2) 250 m / 500 m 18 km to –3 km ASL >140 W 0.2 mm h–1 (12 dBZ) within ±1 dB <1000m <78 kbps <300 kg <300 W 1.44 × 1.07 × 0.7 m
* The minimum detectable rainfall rate is defined by Ze = 200 R1.6
The central idea of GPM is to realize frequent (three-hourly is the goal) and high precision measurement of precipitation by combining the microwave radiometer data from multiple satellites with information provided by the DPR and GMI on the GPM core satellite. The DPR will provide 3D information of hydrometeor distribution with high spatial resolution. Such data are very valuable for the study of storm structure. The major importance of the DPR, however, lies in the fact that it can provide the regional and seasonal statistics of storm structure together with DSD parameters. Since rain retrieval algorithms for passive microwave radiometers have to assume a vertical structure of storm either deterministically or statistically, reliable storm structure information is crucial for the accuracy of rain estimation. The statistics from the DPR can be used as a database in radiometer algorithms to reduce the uncertainties of the storm models.
Japan’s Role in the Present and Future Satellite Observation
607
Figure 1. Concept of dual-frequency measurement of precipitation.
4.2 Operational use of GPM data Frequent measurement of precipitation to be realized by a fleet of constellation satellites with microwave radiometers is expected to reduce substantially the sampling error, or the estimation error that originates in insufficient frequency of measurement. Moreover, if frequently observed data with high precision are delivered to the users in near-real time, weather forecasts and flood warning systems will greatly benefit from them. While TRMM’s objectives are purely scientific, GPM’s objectives also include applications of data in operational use and the mission is considered as a bridge mission that transfers the developed technology to the operational world.
5
FUTURE PLANS AND PROSPECT
5.1 Realization of an ADEOS-II follow-on mission The ADEOS-II satellite unfortunately stopped its operation in slightly less than one year after the start of data collection due to malfunction. To contribute the continuous observation for climate change and global water cycle research, planning of a follow-on mission is under way in Japan. The follow-on mission may have a basic sensor for Earth observation like the
R. Oki and Y. Furuhama
608
AMSR and the GLI. An AMSR follow-on mission will be expected as a conponent of GPM, and a GLI follow-on mission will be useful for reducing the uncertainty of global change prediction. The follow-on mission might be realized around 2010.
5.2 Observation of 3D distribution of clouds by radar One of the major issues in climate research such as estimating the future global warming is the uncertainty of the effect of clouds on the radiation balance. To reduce its uncertainty, it is necessary to grasp the current global distribution of cloud in the 3D space. Research and development of spaceborne cloud radars has started in Japan to realize such cloud measurement from space. The EarthCARE mission, which is a joint research mission between ESA and JAXA and realizes simultaneous measurements of clouds and aerosols with a combination of radar and lidar, is a possible core mission in ESA’s Earth Explorer Program. Although the CloudSat mission in the USA will precede the EarthCARE mission, radar measurement of cloud simultaneously with lidar and the Doppler function in the radar together with other data from Earth observing satellites should bring new insights into Earth’s radiation, cloud-precipitation processes, and cloud-aerosol interactions.
6
CONCLUSION
This article shows that Japan has been contributing to Earth observation with satellites. New knowledge has been accumulated in meteorology and climatology from the data obtained by advanced technology developed in Japan. Accumulation of many years of data and their application are required in research of Earth environment. From this viewpoint, researchers in Japan make requests for Earth observation in addition to new technology development. Requirements for conventional observations have also changed to demand more frequent observation and higher resolution. To realize such requirements, a well organized Earth observation structure with more international collaboration must be established. In the future, Japan will continue to participate in research of the Earth’s environment change in Japan’s leading areas in such an organization in collaboration with other countries.
7
REFERENCES
Adhikari, N. B. and K. Nakamura, 2003: Simulation-based analysis of rain rate estimation errors in dual-wavelength precipitation radar from space. Radio Science, 38, 1066. Furuhama, Y., R. Oki, T. Iguchi, and E. A. Smith, 2002: Precipitation observation from space in the next generation: the Global Precipitation Measurement (GPM). URSI-F Open Symposium, Garmisch-Partenkirchen, Feb. 12–15.
Japan’s Role in the Present and Future Satellite Observation
609
Kawanishi, T., T. Sezai, Y. Ito, K. Imaoka, T. Takeshima, Y. Ishido, A. Shibata, M. Miura, H. Inahara, and R. W. Spencer, 2003: The Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E), NASDA’s contribution to the EOS for global energy and water cycle studies. IEEE Trans. Geosci. Remote Sensing, 41, 184–194. Nakamura, K. and T. Iguchi, 2007: Dual-wavelength radar algorithm. In: Measuring Precipitation from Space – EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 225–234. Shibata, A., K. Imaoka, and T. Koike, 2003: AMSR/AMSR-E level 2 and 3 algorithm developments and data validation plans of NASDA. IEEE Trans. Geosci. Remote Sensing, 41, 195–203.
48 INTERNATIONAL GLOBAL PRECIPITATION MEASUREMENT (GPM) PROGRAM AND MISSION: AN OVERVIEW Eric A. Smith1, Ghassem Asrar2, Yoji Furuhama3, Amnon Ginati4, Alberto Mugnai5, Kenji Nakamura6, Robert F. Adler1, Ming-Dah Chou7, Michel Desbois8, John F. Durning1, Jared K. Entin2, Franco Einaudi1, Ralph R. Ferraro9, Rodolfo Guzzi10, Paul R. Houser11, Paul H. Hwang1, Toshio Iguchi12, Paul Joe13, Ramesh Kakar2, Jack A. Kaye2, Masahiro Kojima14, Christian Kummerow15, Kwo-Sen Kuo16, Dennis P. Lettenmaier17, Vincenzo Levizzani18, Naimeng Lu19, Amita V. Mehta20, Carlos Morales21, Pierre Morel22, Tetsuo Nakazawa23, Steven P. Neeck2, Ken’ichi Okamoto24, Riko Oki25, Garudachar Raju26, J. Marshall Shepherd1, Joanne Simpson1, ByungJu Sohn27, Eric F. Stocker1, Wei-Kuo Tao1, Jacques Testud28, Gregory J. Tripoli29, Eric F. Wood30, Song Yang31, and Wenjian Zhang32 1
NASA/Goddard Space Flight Center, Greenbelt, MD, USA Science Division, NASA Headquarters, Washington, DC, USA 3 Japan Aerospace Exploration Agency, Tokyo, Japan 4 ESA/European Space Research & Technology Center, Noordwijk, The Netherlands 5 Ist. di Scienze dell’Atmosfera e del Clima, Consiglio Nazionale delle Ricerche, Roma, Italy 6 Inst. of Hydrology & Atmospheric Science, Nagoya Univ., Furocho, Chikuaku, Nagoya, Japan 7 Dept. of Atmospheric Science, National Taiwan University, Taipei, Taiwan 8 Centre National de la Recherche Scientifiique, Ecole Polytechnique/LMD, Palaiseau, France 9 National Environmental Satellite Data Information Service, NOAA, Camp Springs, MD, USA 10 Agenzia Spaziale Italiana, Osservazione della Terra, Roma, Italy 11 School of Computational Sciences, George Mason University, Fairfax, VA, USA 12 National Inst. of Information & Communications Technology, Koganei, Tokyo, Japan 13 Environment Canada, Toronto, Ontario, Canada 14 Japan Aerospace Exploration Agency, Tsukuba City, Japan 15 Dept. of Atmospheric Sciences, Colorado State University, Fort Collins, CO, USA 16 Caelum Inc., NASA/Goddard Space Flight Center, Greenbelt, MD, USA 17 Dept. of Hydrology, Univ. of Washington, Seattle, WA, USA 18 Ist. di Scienze dell’Atmosfera e del Clima, Consiglio Nazionale delle Ricerche, Bologna, Italy 19 National Satellite Meteorological Center, China Meteorological Administration, Beijing, China 20 JCET/UMBC, NASA/Goddard Space Flight Center, Greenbelt, MD, USA 21 Dept. of Atmospheric Sciences, University of Sao Paulo, Sao Paulo, Brazil 22 Univ. of Paris, Paris, France 2
611 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 611–653. © United States Government 2007.
E. A. Smith et al.
612 23
Meteorological Research Inst., Japan Meteorological Agency, Tokyo, Japan Dept. of Aerospace Engineering, Osaka Prefecture Univ., Saki, Osaka, Japan 25 JAXA/Earth Observation Research & Application Center, Chuo-ku, Tokyo, Japan 26 Indian Space Research Organization, Bangalore, India 27 Dept. of Earth and Atmospheric Science, Seoul National University, Seoul, Korea 28 Novel Initiative for Meteorological and Environment Technologies, Vélizy, France 29 Dept. of Atmospheric & Oceanic Sciences, University of Wisconsin, Madison, WI, USA 30 Dept. of Civil Engineering, Princeton Univ., Princeton, NJ, USA 31 GMU, NASA/Goddard Space Flight Center, Greenbelt, MD, USA 32 Dept. of Observation & Telecom., China Meteorological Administration, Beijing, China 24
Abstract
The purpose of the International Global Precipitation Measurement (GPM) Program is to develop a next-generation space-based measuring system which can fulfill the requirements for frequent, global, and accurate precipitation measurements. The associated GPM Mission is being developed as an international collaboration of space agencies, weather and hydrometeorological forecast services, research institutions, and individual scientists. The design and development of the GPM Mission is an outgrowth of valuable knowledge and published findings enabled by the Tropical Rainfall Measurement Mission (TRMM). From the TRMM experience, it was recognized that the GPM Mission must consist of a mixed nonsunsynchronous and sunsynchronous orbiting satellite constellation in order to have the capability to provide physically based retrievals on a global basis, with ~3-h sampling assured at any given Earth coordinate ~90% of the time. The heart of the GPM constellation is the Core satellite, under joint development by NASA and the Japan Aerospace Exploration Agency (JAXA), which will carry a dual frequency Ku/Kaband precipitation radar (PR) and a high-resolution, multichannel passive microwave (PMW) rain radiometer. The core is required to serve as the calibration reference system and the fundamental microphysics probe to enable an integrated measuring system made up of additional constellationsupport satellites, each carrying at a minimum some type of PMW radiometer. In this article the background, planning, design, and implementation of the GPM is described.
Keywords
NASA, JAXA, GPM, DPR, TRMM, precipitation, water, climate, forecasting, hydrology
1
INTRODUCTION
Globally distributed, continuous, and high-quality measurements of accumulation, intensity, and temporal evolution of precipitation are valued for a wide range of basic- and applications-oriented research of international interest and consequence. These include: (1) diagnostic analyses of regional and global-scale water budgets and the modeling of water cycling in the atmosphere and within the surface, (2) climate reanalyses and climate simulations with atmospheric, oceanic, and coupled global climate models (AGCM, OGCM, CGCM), (3) numerical weather prediction (NWP) and rainfall data assimilation for global and mesoscale NWP models, (4) hydrometeorological/agrometeorological modeling and prediction, (5) modeling of
International GPM Program and Mission
613
fresh water flux impacts on oceanic and large lake circulations, (6) monitoring and quantitative precipitation forecasting (QPF) of landfalling tropical cyclones and severe continental storms, (7) specification of forcing for distributed water routing and flood-hazard models, (8) fresh water resources assessment and prediction, (9) development of regional and global rainfall climatologies, and (10) the very important physical-radiative methodologies for retrieving and validating rainfall retrievals, to name some of the more important topics intrinsically requiring high-quality rainfall information. However, unlike acquisition of more homogenous meteorological fields such as pressure and temperature, obtaining high-quality precipitation measurements is particularly challenging due to precipitation’s stochastic and rapidly evolving nature. In general, precipitation systems exhibit spatially heterogeneous rain rates over local to regional domains, and highly fluctuating rain rate intensities over time, up to large spatial scales. Thus, it is commonplace to observe a broad spectrum of rain rates within a few hours time frame for a given locale or even a region. For this reason and the fact that most of the world is not equipped with precision rain-measuring sensors (i.e., reliable rain gauges and/or radars), the current rain measurement of choice for regional to global analysis is obtained from satellite remote sensing. The purpose of the International Global Precipitation Measurement (GPM) Program is to develop a next-generation space-based measuring system which can fulfill the requirements for frequent, global, and accurate precipitation measurements, continuously acquired along with well-defined and quantitative metrics of the measurements’ systematic and random errors. The ultimate goal of the associated GPM Mission, which is being developed as an international collaboration of space agencies, weather and hydrometeorological forecast services, research institutions, and individual scientists, is to serve as the flagship satellite mission for a variety of water-related research and applications programs. These include international research programs involved with the global water and energy cycle (GWEC) such as the World Climate Research Program (WCRP), Global Energy and Water Cycle Experiment (GEWEX), and to support basic research, applicationsoriented research, and operational environmental forecasting throughout individual nations and consortiums of nations. Because water cycling and the availability of fresh water resources, including their predicted states, are of such immense concern to most nations, and because precipitation is the fundamental driver of virtually all environmental water issues, developing a space-based, globally inclusive precipitation measuring system has become a pressing issue for a large body of nations. In fact, over 30 nations came together in Washington, DC during July 2003 for the Earth Observation Summit (EOS), to promote the development of an Earth-observing capability among governments and the international community, designed to understand and address global environmental and
614
E. A. Smith et al.
economic challenges. As a result of the EOS, an ad hoc Group on Earth Observations (GEO) was established to prepare a 10-year implementation plan for a coordinated, comprehensive, and sustained Earth observation system. The National Aeronautics and Space Administration (NASA), along with its US interagency and international partners, is playing a key role in GEO activities. Notably, the international nature of the GPM Mission positions it as a viable prototype for the GEO effort. The design and development of the GPM Mission is an outgrowth of valuable knowledge and published findings enabled by the Tropical Rainfall Measurement Mission (TRMM) and produced by various US, Japanese, and European Union (EU) research teams, and dedicated individual scientists. From the TRMM experience, from consideration of basic physical principles associated with direct sensing of precipitation from space, and from a realistic view of contemporary economic constraints, it is now recognized that the GPM Mission must consist of a constellation of satellites, some dedicated, and some conveniently available through other experimental and operational missions supported by various of the world’s space agencies, i.e., in the vernacular of GPM, “satellites of opportunity”. The heart of the GPM constellation is the Core satellite, under joint development by NASA and the Japan Aerospace Exploration Agency (JAXA). As with TRMM, the basic workshare arrangement between NASA and JAXA is that JAXA will provide the radar and the launch, while NASA will provide the radiometer, the satellite bus, and the ground segment. The Core satellite is the central rain-measuring observatory which will fly both a dual frequency (Ku/Ka-band) precipitation radar (PR) called DPR, and a high-resolution, multichannel passive microwave (PMW) rain radiometer called GMI. The core is required to serve as the calibration reference system and the fundamental microphysics probe to enable an integrated measuring system made up of, typically, eight additional constellation-support satellites. Each support satellite is required to carry one or more precipitation-sensing instruments, but at a minimum, some type of PMW radiometer measuring at several rain frequencies. Fortunately, the GPM constellation has had the welcome attention of the European Space Agency (ESA) and a consortium of European and Canadian scientists, who are planning for the contribution of a European GPM (EGPM) satellite whose instrument capabilities would strengthen the core measurement scheme. This observatory will be specially outfitted with an advanced rain radiometer using a mix of window and molecular O2 sounding frequencies, and a Ka-band, high-sensitivity (5 dBZ) radar – a combination of instruments suitable for measurements of light and warm rainfall, moderate to heavy drizzle, and light to moderate snowfall. All these types of precipitation, which are largely outside the dynamic range of the Core satellite’s instruments, are very important contributors to the Earth’s water cycle at mid- to high-latitudes, while warm rain and drizzle are significant
International GPM Program and Mission
615
contributors in the tropics, particularly in the extended marine stratocumulus regions. It is anticipated, based on mission-life specifications and expected launch dates, that the time-frame for the core and EGPM satellites to be in place to support the constellation architecture is approximately 2009–2014, recognizing that this period could be extended were mission lifetimes to exceed specifications. The GPM Mission consists of four main components. The first is the space hardware making up the constellation measuring system as described. The second is the data information system, referred to as the GPM Precipitation Processing System (PPS), a system whose functions will be distributed amongst NASA, JAXA, and ESA, with the main node at the NASA/Goddard Space Flight Center (GSFC). The main responsibilities of the PPS are: (1) to acquire level 0 and 1 sensor data, (2) to produce and maintain consistent level 1 calibrated/earth-located radiometer brightness temperatures (TB) and radar reflectivities (Zs), (3) to process level 1 data into consistent level 2 and 3 standard precipitation products, (4) to disseminate precipitation products through both “push” and “pull” data transfer mechanisms, and (5) to assure archival of all data products acquired or produced by the PPS, either within the PPS or through suitable arrangements with other data archive services. The third mission component is the internationally organized GPM ground validation (GV) program, which will consist of a worldwide network of GV-measuring sites and their associated scientific and technical support organizations. A subset of these sites are referred to as “GV Supersites”, designating that they will operate in a semicontinuous, near-real-time mode under a well-defined GV data reporting protocol supported by the GSFC PPS. The main function of the GPM GV program will be: (1) to acquire ground-based sensor data relevant to the validation of and/or comparison with satellite sensor measurements and standard precipitation product retrievals, (2) to produce, archive, and publicly make available on the Internet, standard GV products, (3) if a Supersite, to provide near-real-time error characteristics concerning instantaneous rain-rate retrievals from the core-level satellites (i.e., Core and EGPM satellites), consisting of bias, bias uncertainty, and spatial error covariance information, and (4) if a Supersite, to support ongoing standard algorithm improvement by reporting significant errors in instantaneous retrievals from the core-level satellites to scientific groups authoring and maintaining the standard rain-rate algorithms, including with the reports, essential core (and core-type) satellite and GV data needed to interpret effectively algorithm breakdowns. Error characteristics of ~7.5% accuracy and ~20% precision for instantaneous retrievals – which will produce 3-hourly GPM precipitation products – represent projected estimates of the underlying retrieval uncertainties. These anticipated error
616
E. A. Smith et al.
factors are based on results from recent TRMM validation analyses being used to help design the GPM Mission’s International GV Program. The fourth mission component is the most valuable of the entire mission; that being the people involved, i.e., the collection of individual scientists, engineers, and program officials making up the various science teams from participating nations, as well as the oversight committee/working group infrastructure that will manage and coordinate the international aspects of the mission. Because the GPM Mission is expected to be flexible and fluid in design, enabling space hardware assets to come and go as the situation evolves (referred to as the “rolling wave” constellation approach), allowing for new and changing PPS and GV site facilities and capabilities, and accepting that the underlying scientific effort is a shared responsibility, the people involved will practice international diplomacy as well as adhere to their fundamental responsibilities and commitments within their own organizations and sovereign nations. With these four components, the GPM Mission will have the capability to provide physically based retrievals on a global basis, with ~3-h sampling assured at any given Earth coordinate ~90% of the time; such frequent diurnal sampling made possible by a mixed nonsunsynchronous/sunsynchronous satellite orbit architecture. Ultimately, however, it will be the people involved that will demonstrate how an internationally sanctioned collective effort can be used to acquire a long-sought measuring capability of one of the Earth’s most fundamental variables and one of life’s most precious commodities.
2
CAPTURING THE HYDROLOGICAL CYCLE FROM SPACE
It is recognized that both natural and human-induced climate variations manifest themselves in the global water cycle; see Chahine (1992). If in this context the Earth’s climate is changing, i.e., if global temperatures are increasing as now generally accepted based on many independent observations (e.g., Graham 1995; Karl and Knight 1997; Houghton et al. 2001; Levitus et al. 2001), higher evaporation and precipitation rates might occur. This, in turn, could lead to an overall acceleration of the global water cycle and concomitant increases in weather extremes and durations of major flood and drought episodes, e.g., Gleick (1989), Easterling et al. (2000), Houghton et al. (2001), McCarthy et al. (2001), and Milly et al. (2002). Notably, Ziegler et al. (2003) report that this process might take place so slowly that it would not become recognizable from observations until some 50 years of high-quality global water budget data are assembled. Figure 1 helps illustrates one of the fundamental riddles concerning Earth’s climate modulation and precipitation extremes, i.e., whether established climate cycles such as the
International GPM Program and Mission
617
El Niño – Southern Oscillation (ENSO), the Pacific Decadal Oscillation (PDO), and the North Atlantic Oscillation (NAO) are in some way related to the extremes of flood and drought.
Figure 1. Monthly ENSO, PDO, and NAO indices from 1990 to present plotted along with major flood/drought indices for continental USA, noting correlations between each climate index and both flood/drought indices are intermittently nonnegligible.
A verifiable rate of increase (decrease) in a regional or global water cycle requires the total flux accumulation of freshwater at distinct Earth interfaces, specifically the most relevant atmosphere–ocean, atmosphere–land, and ocean– land interfaces over a sufficient time period, say 1 year, to increase (decrease) in a gradual fashion over a number of years, at least for a decade and possibly for multiple decades. Such an increase (decrease) must be consistent with other measures of water cycle variability, such as: (a) the mean residence time of water vapor in the atmosphere; (b) the change in water storage in the land reservoirs including soil moisture, permanent snow-ice features, deep ground water, lakes, and inland seas; (c) the change in accumulated surface
618
E. A. Smith et al.
runoff, interflow, and baseflow out of major continental water basins; and (d) the total E – P flux (evaporation minus precipitation) across the ocean– atmosphere interface over the ocean basins. It is important to recognize that the current generation of climate models is widely divergent concerning these terms (Lau et al. 1996). Figure 2 illustrates the current uncertainties and ambiguities in various compilations of the global water budget.
Figure 2. Recent compilation of global water budget developed by Dyck and Peshke (1995), along with tabulation of uncertainties based on published sources, as well as estimation of reduction of P uncertainties by this paper’s lead author using TRMM measurements.
Understanding the Earth’s climate and how global/regional water cycle processes create and react to climate perturbations relies on what is known about atmospheric moisture transport, clouds, precipitation, evaporation, transpiration, latent heating, runoff, and large-scale circulation variations in concert with changing climate conditions; e.g., Chen et al. (1994), Soden (2000), Chang and Smith (2001), and Mariotti et al. (2002). The process linking these quantities is precipitation, the variable indicating the rate at which water cycles through the atmosphere. Precipitation also has a profound influence on the quality of human lives in terms of availability of fresh water for consumption and agriculture. Thus, it is no understatement to note that high-quality precipitation measurements with global, long-term coverage
International GPM Program and Mission
619
and frequent sampling are crucial to understanding and predicting the Earth’s climate, weather, and GWEC – and the consequences for human civilization. Relatively high-quality, multidecadal compilations of the historical, global-scale rainfall record for selective continental areas based on raingauge measurements, have been available since the 1950s. However, oceanic rainfall remained largely unobserved prior to the beginning of the satellite era. In fact, it was only after the launch of the first Special Sensor Microwave Imager (SSM/I) on the Defense Meteorological Satellite Program (DMSP) satellite series in July 1987 that representative oceanic precipitation measurements became available on a regular basis (Wilheit et al. 1994; Smith et al. 1998). In response, and recognizing that satellites had become the foremost tools for measuring global precipitation, the NASA renewed its emphasis on quantitatively measuring precipitation from space with the Mission to Planet Earth (MTPE) program in the 1990s; Simpson et al. (1988). This resulted in the Tropical Rainfall Measuring Mission (TRMM), a collaborative mission between the US’s NASA and its counterpart agency in Japan, i.e., the National Space Development Agency (NASDA); see Simpson et al. (1996) and Kummerow et al. (2000). [Note, NASDA has recently been renamed as JAXA]. The TRMM satellite was launched on 27 November 1997 (USA’s Thanksgiving Day) from Japan’s Tanegashima launch facility. Fundamentally, the TRMM research program is dedicated to measuring tropical–subtropical rainfall over a lengthy time period (now expected to be ~7.5 years), and by so doing, acquiring the first accurate, representative, and consistent ocean climatology of precipitation. TRMM was initially launched into a lowaltitude (350 km), non-sunsynchronous orbit inclined 35 degrees to the Earth’s equatorial plane, with a nominal mission lifetime of 3 years – but with expectations for a longer lifetime. (At this juncture the TRMM satellite is expected to remain operational until early 2005, after which time it will likely have to undergo a “safe” end-of-mission re-entry procedure). TRMM carried to space, for the first time, a PR developed by NASDA’s partner agency in Japan, the Communications Research Laboratory (CRL); see Okamoto et al. (1988), Meneghini and Kozu (1990), Nakamura et al. (1990), and Okamoto and Kozu (1993). [Recently, CRL was renamed the National Institute of Information and Communications Technology (NiCT)]. The PR consists of a noncoherent, phased array Ku-band radar (13.8 GHz) based on slotted waveguide technology – with a 2 × 2 m2 aperture. The PR has produced remarkable new measurements of precipitation’s vertical structure and unprecedented views of the Earth’s rich assortment of clouds, convection, frontal zones, precipitating storms, and tropical cyclones. The TRMM satellite also carries a conical-scanning, V/H-polarized, 9-channel
620
E. A. Smith et al.
(10.7 V/H, 19 V/H, 21.3 V, 37 V/H, 89 V/H) PMW radiometer (with 0.6 m diameter antenna) called the TRMM Microwave Imager (TMI). The more than three times greater swath width of the TMI relative to the PR (780 vs. 220 km) enables TMI to measure precipitation over a wide-swath track at high duty cycle, while measuring the detailed physics of precipitation along the narrow-swath radar track in coincidence with the inner-swath radiometer track. It is also notable that the TRMM observatory is the first case of a rain radiometer observing precipitation from space at low altitude (350–400 km), delivering enlightening high spatial resolution views from an attenuationlike perspective. However, the most important achievement of the TRMM research program is the high quality of the retrieval data sets produced by the different instruments and level 2 (L2)/level 3 (L3) standard algorithms used to produce the TRMM climatological estimates – particularly over ocean. [L2 algorithms produce instantaneous rain rate retrievals at full spatial resolution while L3 algorithms produce monthly averaged rain rate retrievals at 5 × 5° spatial resolution]. Figure 3 compares oceanic monthly averaged rain maps from the most recent version (May 2004) of the four standard TRMM L2 and L3 algorithms to illustrate the close agreement between the different physical retrieval approaches: (1) L2 TMI-only, (2) L2 PR-only, (3) L2 TMI-PR combined, and (4) L3 TMI-only. The PR and TMI rain measuring instruments are accompanied by three additional instruments which can be used to study precipitation indirectly, these being: (1) the 5-channel Visible-Infrared Scanning Radiometer system (VIRS) used to image the Earth’s cloud field and surface at optical-infrared wavelengths, (2) the Lightning Imaging System (LIS) used to optically detect and image the Earth’s lightning flash rate, and (3) the prototype version of the Cloud and the Earth Radiant Energy System (CERES) used to investigate the Earth’s radiation budget and the impact of clouds on net radiant energy. The study of Kummerow et al. (1998) provides detailed descriptions of all five instruments on the TRMM observatory. Motivated by the successes of TRMM, well articulated in three special scientific journal issues (JRSSJ 1998; JAM 2000; JCLIM 2000), an American Meteorological Society (AMS) scientific monograph (AMM 2002) dedicated to Dr. Joanne Simpson, and a mission highlights compendium (Sumi 2002), and recognizing the need for a more comprehensive global precipitationmeasuring program, NASA and JAXA conceived a new mission called the GPM Mission. The idea of a GPM Mission was first proposed at NASA’s Post-2002 Mission Planning Workshop at Easton, MD in August 1998 (see Kennel et al. 2002). The objective of that forum was to define possible space-based missions in support of NASA’s Earth Science Enterprise (ESE) research program, which is a part of the US’s evolving Climate Change Science Program (CCSP), the WCRP, and the International Geosphere-Biosphere
International GPM Program and Mission
621
Figure 3. Distributions of monthly rainfall accumulation over global tropics for Feb1998 produced by most recent version (V6) of standard TRMM L2/L3 algorithms: (1) top panel shows TMI-only (L2 alg 2a12) (see Kummerow et al. 1996, 2001; Olson et al. 2001, 2007); (2) 2nd from top panel shows PR-only (L2 alg 2a25) (see Iguchi et al. 2000; Meneghini et al. 2000); (3) 3rd from top panel shows TMI-PR combined (L2 alg 2b31) (see Haddad et al. 1997; Smith et al. 1997); and (4) bottom panel shows TMI-only (L3 alg 3a11) (see Wilheit et al. 1991; Hong et al. 1997; Tesmer and Wilheit 1998). Color bar denotes average rain rate in mm day–1 (see also color plate 21).
622
E. A. Smith et al.
Program (IGBP). These programs were established to develop a scientific understanding of the Earth as a system and its climatic response to natural and human-induced changes; see Asrar et al. (2001) and ESE (2003). The primary scientific goal of the ESE is to answer the question: “How is the Earth changing and what are the consequences for life on Earth?”, a goal central to man’s concern with the water cycle though the connections between (among others) water and climate, water and food, etc. Because the global water cycle is so central to Earth’s system variability, one of the major goals in ESE’s research strategy is to focus on the GWEC and answer the following question: How are global precipitation, evaporation, and cycling of water changing? Recognizing the need for improved and frequent precipitation measurements in answering this question, the GPM Mission was selected by NASA’s Division of Earth Science (Code Y) as the highest priority mission in support of ESE’s GWEC initiative. A fundamental scientific goal of the GPM Mission is to make substantial improvements in global precipitation observations, especially in terms of measurement accuracy, sampling frequency, Earth coverage, and spatial resolution – and in doing so, extend TRMM’s rainfall time series (Shepherd et al. 2004). To achieve this goal, the mission will consist of a constellation of low earth-orbiting satellites carrying various passive and active microwave measuring instruments (Smith et al. 2004a). One of these satellites will be similar to the TRMM satellite – referred to as the GPM Core satellite (Smith et al. 2002). The Core will have for its main payload, two rain measuring instruments. The first is a dual-frequency (Ka/Ku-band) PR under development at JAXA and NiCT – referred to as the DPR. The second is an advanced, large aperture, multichannel PMW rain radiometer – referred as the GPM Microwave Imager (GMI). The GMI has been designed by NASA/GSFC and is current being developed under a competitive industry solicitation. Additional satellites in the constellation (notionally eight constellation support members) will be internationally contributed and will carry a variety of multichannel PMW rain radiometers, which will use Core satellite measurements for their calibration reference. This approach assures an unbiased global data set made up of multi-instrument precipitation retrievals. The GPM Mission is, primarily, a science-driven research program which will be used to address important GWEC issues central to improving the predictions of climate, weather, and hydrometeorological processes, but also to stimulate operational forecasting and to underwrite an effective public outreach and education program including near-real-time dissemination of televised regional and global rainfall maps. In regards to the latter, in 2002, when the United Nations (UN) requested NASA and NASDA to recommend satellite missions for peaceful uses of space, they showcased the GPM mission as an example of such a mission within the framework of a program entitled “Remote Sensing for Substantive Water Management in Arid and Semi-Arid Areas”.
International GPM Program and Mission
623
Past observations and model simulations have shown that regionally, precipitation varies with timescales ranging from minutes to years; see Nykanen et al. (2001) and Dai and Wigley (2000), respectively. While local weather-related processes affect precipitation variability on hourly to daily time scales, global-scale phenomena, such as the ENSO, are found to be responsible for interannual precipitation variability (typically, 2–6 years), entailing major perturbations to the atmospheric water budget; see Trenberth et al. (2003) and Sohn et al. (2004). Moreover, there is evidence of precipitation variability on timescales of decades to centuries, the so-called DecCen timescales; see Huang and Mehta (2004) and Huang et al. (2005). Extreme events such as tropical cyclones, severe electrified storms, blizzards, floods, landslides, and droughts associated with multiscale precipitation variability remain as some of the greatest natural hazards to human lives. Thus, understanding the links between climate variability, weather changes, hydrometeorological anomalies, and small-scale cloud macrophysics and microphysics, in the context of multiscale precipitation variability, are central scientific objectives of the GPM research program. Given its constellation design and the use of improved space hardware, the GPM Mission is expected to provide improved measurements of precipitation on space scales ranging from millimeters (the microphysical scale) to 105 km (the global scale) and on timescales from hours to decades.
3
GPM CONSTELLATION MEASURING PROGRAM
Frequent, 3-hourly and globally distributed measurements of rainfall will be enabled by the GPM Mission’s constellation design. Moreover, because of the additional radar and radiometer channel capacity on the GPM Core and an additional core-like satellite being provided by the ESA called EGPM (to be described), detailed measurements of vertically distributed rain rate, water content of component hydrometeor species, latent heating, and bulk microphysical DSD parameters will be available from these two constellation members.
3.1 NASA–JAXA core satellite Under a NASA–JAXA partnership similar to that developed for the TRMM project, the GPM Core satellite is being developed to serve as the “point-ofreference” for the GPM constellation. The Core observatory will carry the new JAXA–NiCT dual-frequency, cross-track scanning, Ku/Ka-band (13.6/ 35.5 GHz) PR, i.e., the DPR (Iguchi et al. 2002), along with NASA’s advanced, large aperture (high resolution), conical-scanning, multichannel GMI PMW radiometer (Smith et al. 2002). The radiometer will employ nine rain-measuring channels at H and/or V polarizations across the 10.7–89 GHz spectrum (similar in wavelength to those on the TMI), and possibly 3–5
624
E. A. Smith et al.
additional high frequency (HF) channels positioned at the 166 GHz window and near the 183 GHz H2O absorption line (these latter HF sounding channels pending NASA’s final mission budget constraints). The Core satellite’s precipitation measurements will be unprecedented in two ways. The GMI’s spatial resolution will exceed that of any current or future planned microwave radiometer, including the new Conical-scanning Microwave Imager/Sounder (CMIS) radiometer with its 2.5 m antenna being developed for the NOAA Polar Orbiter Environmental Satellite System (NPOESS) platforms. This results from the combined effect of the GMI’s 1.2 m diameter antenna aperture flown at 400 km, i.e., the orbit altitude at which the GPM Core satellite will be flown (less than half the altitude of NPOESS). [Note, 400 km is the same altitude that the TRMM satellite has been flown since a boost maneuver was invoked during August 2001, elevating it from its initial launch altitude of 350 km and taking it out of the relatively high-drag atmosphere at 350 km – thus extending mission life by preservation of fuel]. Improving spatial resolution up to a critical limit (order 5 km at 19 GHz) is paramount in improving precipitation retrieval accuracy from a PMW radiometer. This is because the current dominant source of error in PMW retrieval is from heterogeneous beam filling which necessarily produces significant retrieval error, i.e., because algorithm transforms of brightness temperatures to rain rates are invariably nonlinear, a problem only correctable by increasing spatial resolution of the radiometer measurements. Second, because of the DPR’s two well-separated attenuating frequencies within the Mie regime vis-à-vis raindrops, it will be able to measure differential reflectivity and thus provide measurements sensitive to fluctuations in rain DSD properties (Doviak and Zrnič 1984; Meneghini et al. 1989, 1990, 1992; Kuo et al. 2004). This feature allows for recovery of at least two of three parameters needed to describe the bulk distribution of the generalized raindrop size distribution (DSD). Such measurements are important because increased knowledge of DSD factors improves the retrievals of rain rate for both radar and radiometer retrieval schemes, and helps in interpreting climate change, since cloud microphysics and ambient DSDs are directly influenced by climatic perturbations associated with temperature, moisture, and aerosol loading. Thus, with the DPR, the GPM Core satellite will provide enhanced information on precipitation microphysics relative to what has been possible from the single frequency TRMM PR, thereby enabling improvements in latent heating algorithms and mass spectra properties associated with the highly varying DSD. In addition, the GMI with its 1.2 m antenna being flown in a 400 km orbit, will provide unprecedented spatial resolution microwave measurements at the critical emission and scattering frequencies of 10.7, 18, 22, 37, and 89 GHz relative to any PMW radiometer of its generation. The current expectation for launch of the GPM Core satellite is
International GPM Program and Mission
625
within the 2009–2010 time frame. The launch date of the GPM Core satellite is currently a subject of negotiations between the Earth Science Divisions of JAXA and NASA, prior to their signing of a Memorandum of Understanding (MOU) concerning their respective responsibilities for the Core satellite component of the mission. NASA’s Mission Confirmation Review (MCR) for the Core satellite is now anticipated for the third quarter of 2005. Furthermore, the anticipated measurement improvements from the GPM Core satellite have stimulated a new type of rain-retrieval data processing design under development within GPM’s PPS. The PPS will be run in distributed fashion at NASA-GSFC, JAXA-Earth Observation Research and Application Center (EORC), and ESA-European Space Research Institute (ESRIN), with the NASA-GSFC node serving as the main facility. With one exception, the Core satellite’s precipitation retrievals will be used as the de facto calibration reference for the remainder of the retrievals from other constellation members. The exception is the EGPM satellite which will be able to produce additional calibration-standard results for the light rain, warm rain, drizzle, and snowfall portions of the precipitation spectrum (see Mugnai et al. 2007).
3.2 ESA EGPM satellite The EGPM satellite will be a critically important member of the GPM constellation because it can augment the Core satellite’s reference measurements over the low optical depth portion of the rainfall spectrum by virtue of its own unique radar and radiometer instrument features. As elaborated in the study of Smith et al. (1998), involving an intercomparison of over 20 PMW-based precipitation retrieval algorithms, and as an ongoing intercomparison investigation of the various TRMM standard algorithms (Dr. Song Yang, 2004, personal communication), detection and quantification of light precipitation has been the most difficult problem area from which to obtain agreement amongst the variety of contemporary radiometer and radar algorithms. This is because the standard rain radiometer frequencies being flown today are not optimal for low rain rate and snowfall retrieval, and because the sensitivities of the current PR technology (i.e., TRMM PR’s 17 dBZ sensitivity at its 13.8 GHz frequency and GPM Core DPR’s 17 and 12 dBZ sensitivities at its 13.6 and 35.5 GHz frequencies) are not high enough to detect much of the low rain rate and light-to-medium snowfall spectrum (Mugnai et al. 2007). As noted in the latter study, light rainfall and light to moderate snowfall are the dominant contributions to rain water equivalent accumulation within and poleward of the midlatitudes (see Mugnai et al. 2007). To address this problem, the EGPM satellite will carry: (1) a new, innovative 17-channel microwave radiometer (EMMR) employing six standard H/V window channels between 18.3–89 GHz (at 18, 37 and 89 GHz),
626
E. A. Smith et al.
one 23.8 GHz (H-Pol) water vapor absorption channel, a 150 GHz (2 mm) H/V channel pair, and two 4-channel sets of single polarization sounding channels within the 50–54 and 118 GHz molecular oxygen absorption wing regions; and (2) a Ka-band, 3-beam Nadir-pointing Precipitation Radar (NPR) with a sensitivity of ~5 dBZ. The EMMR’s two sets of oxygen absorption channels (producing differential frequency scattering because the associated cross-band frequencies are selected so their clear-sky weighting functions are matched), in combination with the high-frequency 89 and 150 GHz channels, permit exceedingly high-quality retrieval performance for light rain, warm rain, drizzle, and snowfall. In addition, the NPR’s high sensitivity permits vertical profiling of most liquid rainfall and snowfall situations, and when the reflectivity profile information is combined with EMMR’s concomitant brightness temperature information, the retrieval of low optical depth precipitation can be performed at accuracies commensurate with accuracies for moderate to heavy rainfall and heavy snowfall from the GPM Core satellite’s DPR and GMI measurements. In essence, the EGPM instrument requirements have been established to enable its measurements to serve as the calibration standard for: (1) the light/warm/drizzle rainfall portions of the liquid precipitation spectrum (from ~1.5–0.05 mm h–1) over which the DPR has lost its differential reflectivity acuity, (2) the domains poleward of the 65 degree parallels beyond which the Core satellite does not measure (noting the Core satellite’s orbit inclination will be 65 degrees), and (3) light to moderate snowfall which will not be detectable by the DPR because its Ka-band sensitivity cutoff is ~12 dBZ when operated in its low vertical-resolution mode (i.e., 0.5 km). [Note, the DPR’s Ka-band radar sensitivity for its high verticalresolution mode (0.25 km) is more like 17 dB, equivalent to the DPR Kuband radar sensitivity]. The anticipated orbit profile for the EGPM satellite is sunsynchronous with a 1,430 DN equator crossing, at an approximately 500 km altitude – with an anticipated launch in the 2009–2010 time frame (as of this writing the mission status of EGPM is as a potential mission under ESA’s Earth Watch operational satellite program). The combined radar and radiometer instrument suites on the GPM Core and EGPM core-like satellites will be able to reduce errors in precipitation retrievals introduced by nonprecipitating clouds, diverse macro- and microscale cloud physics, and high-resolution heterogeneous precipitation features (which can produce beam-filling errors in radiometer estimates without coincident knowledge of the vertical radar reflectivity structures). This means that measurements from these two satellites, used together, will serve to eliminate systematic differences from precipitation estimates generated by the other seven members of the constellation. Thus, the GPM Core and EGPM satellite instruments used together with the additional seven PMW radiometers on the constellation fleet will provide a sensor network in space
International GPM Program and Mission
627
providing frequently sampled, bias-free precipitation estimates with fullEarth coverage.
3.3 GPM constellation design The most distinctive aspect of the GPM mission is that it represents an across-the-board effort by an international collection of space agencies and partner agencies/organizations to develop a “virtual” constellation of Earth science satellites, each member of which is underwritten with its own unique experimental or operational purpose and agenda, but from which the collective set of platforms produces an integrated, first-order environmental measurement, i.e., global rainfall. Notably, the foremost meteorological variables currently achievable from space are temperature, cloud-drift winds, cloud cover, and rainfall. The first three variables can be measured from geostationary earth-orbiting (GEO) satellites and thus for a given region, do not necessarily require a constellation of satellites. On the other hand, rainfall cannot yet be measured directly from GEO orbit. This is because the required antenna sizes for making an order 10 km resolution microwave measurement from a notional GEO altitude of ~35,000 km are order 25 m. Dimensions of this magnitude are currently out of reach in terms of technological readiness – and could remain so for at least a decade or more. Therefore, low earth-orbiting (LEO) satellites are required for direct detection of rainfall, and consequently, in order to obtain frequent sampling with LEOs (order 3-h), an 8–10-member constellation is required. [If cloud-drift winds now produced by GEO satellites are eventually superseded by more accurate coherent lidar-based winds, it is conceivable that an LEO satellite constellation would be used for the needed sampling – an advance that could also benefit research on the global water cycle]. A meaningful routine measurement of rainfall must be able to resolve precipitation’s daily cycle. The maximum allowable measuring interval to resolve the principal daily harmonics (i.e., diurnal, semidiurnal, and asymmetric semidiurnal) is 3 h, which requires 8–10 LEOs in a “constellation of opportunity” when considering total global coverage. In the GPM constellation, the Core satellite’s orbit is purposefully non-sunsynchronous in order to provide high-quality, diurnal-sampled calibration reference estimates, and assures coincident orbit intersections with all other sunsynchronous and nonsunsynchronous constellation members. Orbit intersections between the Core satellite and each of the constellation’s remaining members are essential for producing closely coincident data in time needed to determine the systematic biases for the entire set of constellation members. Moreover, for the expected constellation members, their orbit planes and orbit plane positions produce sufficiently distributed and robust diurnal sampling to guarantee 3hourly return times some 90% of the time. Thus, the Core satellite’s intrinsic diurnal sampling property and capability of eliminating biases from other
628
E. A. Smith et al.
constellation data sets, enables global, bias-free precipitation estimates across the daily cycle. No individual space agency can afford the immense expenditure required to develop an “orthodox constellation system”, which would consist of one non-sunsynchronous satellite flown at a relatively high inclination angle, equipped with the essential radar-radiometer payload enabling it to be used as a calibration reference system for a complement of eight additional constellation satellites distributed two each, 180 degrees apart in four evenly spaced orbit planes. Therefore, international cooperation is required to develop the “constellation of opportunity” design. Moreover, for international cooperation to be effective all main space agencies must be involved to develop the major space hardware and attract the participation of smaller agencies – meaning effectively that ESA, JAXA, and NASA must be involved. It is the authors’ opinion, with no prejudice intended, that by the time the GPM Mission is operational, some ten agencies from seven countries plus the EU could be participating in the space constellation. The countries and agencies potentially involved are Canada (CSA), China (NSMC), the EU (ESA), France (CNES), India (ISRO), Italy (ASI), Japan (JAXA/NiCT), and the USA (NASA, NOAA, IPO). Moreover, the space agencies of Brazil, Germany, Israel, Korea, Taiwan, and the UK have all expressed varying degrees of interest in participating. This is an explicit case of cooperation and shared responsibility amongst nations that is central to the activity of the GEO that was an outgrowth of the 1st EOS, held on 31 July 2003 in Washington, DC (EOS 2003). One might ask what would be the effect if one of the three major space agencies did not participate in the GPM constellation development. The answer is that the mission’s scientific robustness and sampling capability would be degraded, which may then lead to other agencies withdrawing their support or never engaging in the first place. This would not derail the mission, but it would very likely diminish it. One might also ask how does a given agency participate in a unique fashion? The answer is by providing a constellation member that produces a unique measuring capability – and if that is not possible, by producing a uniquely focused scientific thrust and uniquely talented science team. The GPM Core and EGPM satellites provide unique measuring capabilities. The combination of enhanced radiometer resolution and differential radar reflectivity on the Core satellite, enable unprecedented retrieval accuracies and new knowledge concerning microphysical variability. The EGPM satellite, with its coordinated 50/118GHz sounding channel radiometer measurements coupled with its highsensitivity 35-GHz radar measurements, provide an entirely new means to measure light/warm rainfall, drizzle, and snowfall. Notably, as summarized in Table 1, each additional GPM constellation satellite also provides a unique resource for the full mission. This bodes well for other Earth sciencemeasuring problems for which the constellation approach is applicable.
International GPM Program and Mission
629
The ESA science community’s motivation to contribute to GPM research stems from their focus on scientific precipitation problems somewhat unique to Europe and Canada, just as the factors motivating American and Japanese science communities to engage in GPM research stem from each of their country’s specific research and applications interests under the auspices of R&D programs sanctioned by NASA and JAXA, and various sister agencies sponsoring precipitation-related research. Besides the common interests by this body of scientists in monitoring and predicting the Earth’s water cycle, what also binds them together is their common scientific experience resulting from participating in focused TRMM research. Precipitationcentric TRMM research has expanded knowledge and understanding in a number of traditional Earth science topic areas. In the case of American and Japanese scientists, a large body of research related to: (1) rainfall climatologies, (2) GV performance, (3) NWP-based rainfall data assimilation, (4) climate reanalyses, (5) hydrometeorological modeling, (6) effects of freshwater fluxes on ocean processes, (7) cloud-resolving model (CRM) verification, and (8) precipitation microphysics has been published within an international spectrum of journals as highlighted in the four special TRMM volumes referred to earlier. In the case of European and Canadian scientists, for which the EuroTRMM research project was enabled by ESA-ESTEC some 2 years after launch of the satellite (Dr. J.P.V. Poiares Baptista, 2004, personal communication), and for which an associated stream of publications is just now beginning to appear, their interests tend to converge on topics related to: (1) NWP-based radiance data assimilation, (2) retrieval error characterization, (3) new strategies for GV, (4) cloud-radiation modeling, (5) flash-flood predicttion applications, (6) over-land precipitation retrieval, (7) Mediterranean-basin water budget analyses, (8) theoretical analysis of K-band radar scattering and attenuation processes, and (9) theoretical basis for snowfall retrieval. What about the other expected members of the GPM constellation? What motivates their science interest groups and what unique capabilities do they have to offer in the way of precipitation measuring and scientific focus. As noted, these issues are summarized in Table 1, which identifies the “dedicated” and “satellite of opportunity” type constellation members, plus a group of potential backup satellites. This table provides information on: (1) sponsoring agency(ies), (2) relevant instruments (and features), (3) mission potential (and lifetime), (4) probable launch window (and likely orbit), (5) unique measuring capabilities, and (6) and main scientific capabilities. [It is emphasized that some of the data in this table represents educated guesswork on the part of the authors, and thus the contents should not be taken as official information].
E. A. Smith et al.
630
Table 1. Summary of GPM’s anticipated and potential constellation members. SS=sunsynchronous orbit, NSS = non sun-synchronous orbit, DN = local time decending node, AN= local time ascending node, DFR = dual-frequency radar. Relevant instruments
Probable launch window (likely orbit)
Satellite
Sponsoring agency
GPM Core
JAXA NASA
GMI DPR
2009-10 (400 km, 65-deg, NSS)
EGPM
ESA
EMMR NPR
2009-10 (600 km, 1430 DN, 2320 AN, SS)
Megha Tropiques
CNES ISRO NASA Partner (TBD)
Madras ScaRab GMI TBD
2009-10 (800-km, 20-deg, NSS) 2011-12 (TBD)
Unique measuring capabilities
Dedicated constellation members
NASA–Partner
Ku/Ka-band DFR O2 dual-band channels, 5-dBZ sensitivity Ka-band radar Radiation budget TBD
Satellites of opportunity DMSP (F19,F20)
DOD
SSMIS
2006-09 (833 km, 0530-0930 DN,18202220 AN, SS)
Operational satellite asset Operational satellite asset with eight complimentary meteorological instruments
NPOESS (C1,C2,C3)
IPO (NASA– NOAA–DOD)
CMIS
2009-32 (833 km, 0530-0930-0040 DN, 1820-2220-1330 AN, SS)
FY-3 (V1,V2,V3,V4)
NSMC
PMWR
2007-12 (TB km, TBD DN, TBD AN, SS)
Operational satellite asset
GCOM-B1
JAXA
AMSR
2012-TBD (800 km, 1030 DN, 2320 AN, SS)
Experimental satellite asset with three complimentary hydromet instruments
Potential backup satellites TRMM
JAXA NASA
TMI, PR
Orbit since 1997 (400 km, 35-deg, NSS)
First PMW radiom in low inc/alt orbit, first space-borne precip radar
CORIOLIS
IPO (NASA– NOAA–DOD), Navy
WindSat
Orbit since 2003 (830 km, 0510 DN, 1800 AN, SS)
First PMW with full stokes vector
AMSR-E
Orbit since 2002 (705 km, 0040 DN, 1330 AN,SS)
1.8-m aperture radiom at 705 km, both window and sounding freqs
AQUA
NASA
NPP
IPO (NASA– NOAA–DOD)
ATMS
2006-07 (830 km, 1030 DN, 2320 AN, SS)
31 GHz plus sounding freqs
METOP
ESA EUMETSAT
AMSU-A
2005-15 (800–850 km, 0930 DN, 2240 AN, SS)
Operational satellite asset with eight complimentary met instruments
3.3.1 Dedicated constellation members Besides the GPM Core and EGPM satellites, which are the first two dedicated GPM constellation members, there are two other possible dedicated constellation members. Number 3 is Megha-Tropiques, which is
International GPM Program and Mission
631
likely to be launched in the 2009–2010 time frame as a partnership mission between the Centre Spatial Nazionale (CNES) headquartered in Paris (i.e., the French National Space Agency) and the Indian Space Research Organisation (ISRO) headquartered in Ahmedabad. Of relevance to GPM, insofar as Megha Tropique’s instrument suite, is its PMW rain radiometer called Madras and a radiation budget instrument called ScaRab, the pair of which makes Megha Tropiques an ideal satellite for conducting GWEC-type studies. In fact, a central scientific objective of the Megha Tropiques mission is to conduct GWEC studies over the Asian monsoon domain. Perhaps of greatest importance to the GPM Mission is the orbit profile, i.e., an 800-km high satellite within a 20 degree inclined orbit plane – yielding a nonsunsynchronous orbit which produces frequent, diurnally sampled, wideswath measurements over the tropics out to 30-degree latitudes. The sampling capability alone makes Megha Tropiques unique within the GPM constellation and, in fact, enables the 3-h sampling metric to be preserved within tropical latitudes (see Lin et al. 2004). (The CNES–ISRO partnership calls for CNES to develop the Madras instrument front-end (i.e., primarily the antenna, feed-horns, receivers, and other electronics), the Scarab instrument, as well as a third instrument for water vapor profiling named SAPHIR, while ISRO is to provide the Madras mechanical assembly, a satellite bus, and launch). The fourth dedicated constellation member, referred to as the NASA– Partner satellite, is tentative. This intended partnership mission between NASA and a yet-to-be-identified international partner would have NASA provide a copy of the GMI, the Partner providing a satellite bus and their choice of an additional instrument(s), with launch terms to be negotiated. 3.3.2 Satellites of opportunity Four types of satellites carrying PMW rain radiometers have been identified as strong possibilities for “satellites of opportunity” during the GPM Mission era. The most important of these are the last two satellites of the operational DMSP, payloads F19 and F20, each of which will carry copies of the Special Sensor Microwave Imager-Sounder (SSMIS), a PMW radiometer suitable for precipitation retrieval and temperature and moisture sounding), and the first three satellites of the operational NPOESS, the next-generation LEO weather satellite program. These latter satellites will carry the CMIS, a large aperture PMW radiometer with many channels, an instrument designed for virtually all standard environmental retrieval problems requiring PMW measurements, including precipitation. A third type of operational satellite anticipated to be carrying a multichannel PMW radiometer, including rain frequencies, will be China’s nextgeneration operational LEO satellite, FY-3. (The National Satellite Meteorological Center (NSMC) of the Chinese Meteorological Administration (CMA)
632
E. A. Smith et al.
is currently engaged in an industry search for four copies of the required radiometer, and is expected to participate in the GPM constellation though some type of partnership arrangement). Finally, JAXA is considering the launch of a follow-up to the Advanced Earth Observing Satellite-II mission (ADEOS-II or MIDORI II), the Earth Observing System (EOS) satellite which was launched in mid-December 2002 carrying the second copy of JAXA’s AMSR PMW radiometer but which stopped communicating to ground in late-October 2003. The follow-up mission, now referred to as GCOM-B1, would include a third radiometer copy called AMSR-Follow On (AMSR-F/O), however, this mission is not confirmed. 3.3.3 Potential backup satellites There are five potential backup satellites for use during the GPM mission, although the three which carry the preferred type of PMW rain radiometer, have varying degrees of probability for service during the notional 2009– 2014 GPM era. The most doubtful of these, as a backup, is TRMM satellite, which will likely be decommissioned in the near future because of NASA’s required safe reentry policies. Since this satellite is currently 7.5 years old, the likelihood of it lasting into the 2010 time frame is extremely slim under any circumstances. However, the performance of the TMI and PR instruments, as well the satellite bus itself, have been superb – and thus it is not out of the realm of engineering possibility that TRMM could last until the GPM era. Nevertheless, without accentuating circumstances, NASA’s policy would prevent the TRMM satellite lifetime to be extended beyond a critical fuel reserve level, needed to produce a controlled reentry. This point is expected to be reached sometime in mid-2004. Another questionable backup satellite is CORIOLIS, a satellite being flown by the US Navy and the US’s Integrated Program Office (IPO) (i.e., the NASA–NOAA–DOD consortium developing a converged, civilianauthority LEO weather satellite program to be maintained operationally by NOAA). This satellite carries an elegant PMW radiometer called WindSat designed for surface wind retrieval. Because the essential frequencies needed for over-ocean surface wind retrieval are equivalent to those needed for over-ocean rain retrieval, and because of the advanced polarization diversity of various WindSat channels, this satellite would be a genuine plus for the GPM constellation. However, CORIOLIS was launched in 2003, and as with TRMM, given its design life, the probability of it lasting into the GPM era is low. A third satellite currently in orbit, but one which has meaningful probability of surviving into the GPM era is AQUA, which carries the first of the JAXA AMSR instruments (i.e., Advanced Microwave Scanning Radiometer). This satellite’s instrument suite (AMSR-E plus five other instruments)
International GPM Program and Mission
633
would augment the GPM constellation with a valuable capability – although, this is strictly a wait-and-see issue concerning AQUA’s survivability as a data-producing space asset. There are two additional satellites carrying PMW radiometers (albeit at nonideal sounding frequencies) that are very likely to be available during the GPM era. The first is the NPOESS Preparatory Program (NPP) satellite, which will carry the Advanced Technology Microwave Sounder (ATMS), an instrument designed primarily for stratospheric temperature sounding, but including a 23.8 water vapor absorption channel, a 31.5 GHz off-window emission channel, and a standard 90 GHz scattering channel that used together, have limited applications for rain retrieval. Figure 4 shows an assessment of the ATMS radiometer precipitation retrieval capability relative to the TRMM TMI (along with a equivalent comparison assessments of the SSM/I and WindSat radiometers). The second is the METOP satellite. In late 2005, the ESA–EUMETSAT consortium will launch the first of three successive
Figure 4. Error assessment of NPP-ATMS and CORIOLIS-WindSat radiometers for precipitation retrieval, relative to TRMM-TMI and DMSP-SSM/I radiometers – based on PMW radiometer simulator described in Shin and Kummerow (2003). SSM/I is 33% more uncertain in terms of random error than TMI, ATMS degrades almost 100% (twice as uncertain as TMI), while WindSat indicates slight improvement (2%). All three bias factors are negligible. [Calculations provided courtesy of Prof. Dong-Bin Shin, George Mason University].
E. A. Smith et al.
634
operational LEO weather satellites by this name, each of which will carry seven instruments, one being an AMSU-A sounder which could be used in a backup role for rain retrieval, if needed. Finally, the GPM program will support the use of rapid-sequence infrared (IR) imagery from GEO satellites for precipitation estimation, specifically to increase the sampling frequency above 3-hourly. While recognizing there are significant uncertainties in retrievals of IR-based rainfall, there are certain applications, particularly hydrometeorological applications that can tolerate the degree of error intrinsic to short term IR precipitation estimates. Furthermore, as discussed by Levizzani et al. (2000, 2001), given recent advances in radiometer technology for GEO satellites and in rain microphysics retrieval science in general, improvements in GEO-IR retrieval techniques can be anticipated. Moreover, combination-merger methods such as described by Huffman et al. (2001) and Turk et al. (2004), which calibrate and normalize the statistical properties of GEO-IR estimates with microwave-based estimates, thus producing remotely sensed blended rainfall products at high temporal resolution, have already shown promise and are expected to continue to improve. Figure 5 compares the percentage of time that 3-h sampling is achievable from a possible and robust current constellation made up of the TRMM satellite (used as the core) plus a set of five support satellites – and a hypothetical constellation made up of the GPM Core satellite plus a set of eight support satellites. It is evident that the 9-member GPM constellation provides 3-hourly coverage approximately 90% of the time everywhere, whereas
Percentage of 3-Hour Intervals Sampled in 7-Day Period Precipitation Sampling Worldwide : Constant Area Pixels
TRMM Era
(6-member constellation)
TRMM, DMSP-F13, -F14, & -F15, AQUA, & CORIOLIS
GPM Era
(9-member constellation)
GPM Core, EGPM, Megha-Tropiques, NASA-Partner, DMSP-F19 & -F20, NPOESS#1, & Two Additional 600km Constellation Members @ 34° and 84° Inclination
Figure 5. Comparison of percentage 3-h sampling during TRMM and GPM eras.
International GPM Program and Mission
635
the 6-member TRMM constellation only provides 3-h coverage approximately 75% of the time in the mid- to high-latitudes, deteriorating to approximately 50% of the time in the tropics. Figure 6 then provides a schematic illustration of a possible scenario for a GPM constellation design, including dedicated constellation members, satellites of opportunity, and roles for various potential back-up satellites. It should be noted, as evident from Table 1, that since there are various unknowns concerning the possible constellation assets during the GPM era (~2009–2014), this diagram must be viewed in the context of the realm of possible, not the realm of prescribed.
Figure 6. Schematic diagram of potential GPM Mission constellation architecture.
4
GPM MISSION’S PRECIPITATION PROCESSING SYSTEM
The GPM PPS is an evolution of the existing mission-specific TRMM Science Data and Information System (TSDIS) to a generic system capable of supporting many microwave precipitation missions. The PPS will provide the science data processing facility for precipitation missions within NASA and partnering institutions, having the responsibility of receiving satellite
E. A. Smith et al.
636
data streams provided by NASA and other partner satellite sites and generating the global precipitation products established by mission Science Team members. In addition, PPS will have the capability to receive small amounts of calibration reference data and software from GV Supersites that are an integral part of establishing the calibration and error characterization for mission products; see Fig. 7 illustrating the PPS system components.
Figure 7. Schematic diagram of PPS system components.
The PPS will ingest level 0 and 1 instrument data from ground systems that support automated protocols. The data are then processed into higher-level products using science algorithms developed by mission Science Teams. Finally, PPS will generate spatially gridded science products using software developed by PPS or the mission Science Teams as negotiated by mission Project Scientists on behalf of the mission Science Teams. The main PPS node will reside at the NASA/GSFC and will be operated 8 h a day, 5 days a week, staffed by a PPS Operation Team. [The PPS will also be staffed outside of the normal schedule in response to spacecraft or ground system anomalies and emergencies.] The PPS core system is based on the concept of automated processing, data management, and error reporting. The front end of PPS contains automated ingest software with a built-in capability for incorporation of custom modules to interface with a variety of external systems. Once data have been ingested, they are maneuvered through the system as needed by automated data handling and archive functions. The heart of the PPS is its automated scheduler backed up by a production database, which executes the above functions, in addition to
International GPM Program and Mission
637
managing processing schedules and product delivery to external sites. Should any problems arise, PPS provides a comprehensive error reporting system that will automatically notify the operations and support staff of the problem regardless of on-site presence. The PPS will also provide an external interface for mission scientists to access various system services, such as data ordering and instrument planning aids and reports. Finally, the PPS provides mission scientists with a test bed for evaluating new science algorithms. These capabilities are illustrated in Fig. 8 in terms of a system architecture diagram. These features translate into a generic framework that is portable to any precipitation mission allowing addition, deletion, and modification to processing threads and equipment suitable to individual mission requirements. It is this feature that will enable the PPS to operate in a distributive fashion, with additional nodes envisioned at the JAXA-EORC and ESAESRIN centers in Tokyo, Japan, and Frascati, Italy.
Figure 8. Schematic diagram of NASA’s central PPS architecture.
638
E. A. Smith et al.
Figure 9. Current status of potential GPM ground validation site network.
5
GPM MISSION’S INTERNATIONAL GV RESEARCH PROGRAM
As noted previously, an important component of the GPM Mission will be the GV research program. Because precipitation is a global variable, and globally diverse in nature, it stands to reason that, to the extent possible, the GV research program should be globally distributed. This means at the outset, that just as the space hardware component of the mission must be international in design because the mission requires a constellation of loworbiting satellites much too expensive for any single space agency, the GV component must be international in design since no single nation can practically deploy (or afford) a global network of GV measuring sites. Fortunately, there is great interest from a number of worldwide organizations and individuals in establishing an internationally based GPM GV network. At an initial GPM GV kick-off workshop held during October 2004 in Abingdon, UK (hosted by John Goddard of Rutherford-Appleton Laboratories), 20 nations plus a consortium of West African nations were represented in a forum designed to establish the basic requirements and principal scientific objectives of GPM’s International GV Research Program. A document describing the outcome of this opening workshop has been prepared (Smith et al. 2005). Figure 9 provides a map of participating and/or interested nations, and in the case of Japan and the USA, their two (2) and three (3) anticipated GV sites, respectively.
International GPM Program and Mission
639
5.1 Importance of GV research program to GPM mission The importance of a GV research program tailored for the GPM Mission stems from three basic concerns relevant to precipitation measurement. The first is measurement uncertainty and the need to quantify uncertainty in detail, so that users of GPM precipitation retrievals can interpret their research and applications results with all due caution and constraint. The second involves the natural inclination to continually improve the retrieval algorithms used to make the space-based measurements. This process cannot be done in isolation from independent precipitation data sets which are needed to assess algorithm performance, both physically and statistically, and at scales below those associated with the level 1 satellite measurements. The third concern is associated with the need for continually improving the independent ground-based GV measurements themselves. Historically, ground-based measurements of precipitation have been beset with difficulties, whether they have been acquired from collection rain gauges, optical rain gauges, disdrometers, submerged acoustic phonics, nonattenuating radars, attenuating radars, long-wave Doppler profilers, or other less known devices. Nonetheless, there have been impressive engineering innovations applied to ground precipitation sensors in the last two decades. Wide aperture rain gauges with improved protection from wind effects, dropcounting rain gauges, extended spectrum disdrometers, common aperture/ multifrequency Doppler profilers, and dual-polarization/dual-frequency radars are just some of the engineering developments that have produced more accurate ground-based precipitation measurements (e.g., Bringi and Chandrasekar 2001). Given this background, it is incumbent that groundbased sensors continue to improve. Moreover, for the GPM Mission to be engaged in improving global precipitation datasets, it should sponsor improvements in independent, ground-based measuring systems, an issue that needs attention within GPM’s International GV Research Program.
5.2 Design of GV research program GPM’s International GV Research Program requires the deployment of a heterogeneous mix of measuring sites operating under a straightforward scientific strategy. Ultimately, participating organizations must represent their own self-interests and concerns. By the same token, the GPM Mission requires some degree of consistency across the eventual site network. Thus, GV sites are classified into two groups: (1) those that conduct their operations according to their own institutional requirements, and serve the greater GPM science community by engaging themselves in forums and publication processes essential in communicating scientific findings (this first type of site to be referred to as a GPM GV Site), and (2) those, which in addition to their independent operations, also periodically report information
E. A. Smith et al.
640
to the NASA-GSFC arm of GPM’s PPS, at near real time and according to a simple protocol, information needed to calculate ongoing retrieval error characteristics (bias, bias uncertainty, spatial error covariance), and to support in ongoing fashion, a process to improve the standard rain retrieval algorithms, accomplished by detecting and reporting algorithm breakdowns associated with the instantaneous core-level satellite retrievals. This second type of site is referred to as a GPM GV “Supersite” – designating that it conducts routine GV operations according to the protocol and for the two basic purposes described. As evident from Fig. 9, these two types of sites are expected to provide ample sampling of the variety of precipitation systems distributed over the Earth. Thus, the design of GPM’s International GV Program provides for the participating organizations to deploy a heterogeneous network of GV sites on their own terms and with their own preferred instrumentation – and in the event they agree to participate in “Supersite” mode, they would be required to transmit routine data to the GSFC PPS needed for retrieval error characterization (i.e., error factors desired at forecast centers utilizing NWP rainfall data assimilation for optimizing the assimilation process), and needed for algorithm improvement. Because of the necessity for routine operations, Supersites would also be required to be affiliated with some type of on-site or remote-site Science Information Center (SIC). The resultant virtual SIC network would exchange data with the GSFC node of the PPS, receiving Core and EGPM overpass data from the PPS, while transmitting low bandwidth error characterization information and algorithm error reports back to the PPS – all according to the protocol. Participating scientists and engineers at any of the sites would be invited, as a matter of course, to participate in GPM’s International GV scientific forums (plus other forums), and to communicate their relevant scientific findings through the open literature. This design is explained in the Abingdon Workshop proceedings.
5.3 Main scientific role of GV research program In summary, four principal objectives define GPM’s International GV Research Program: 1. Quantify uncertainties and error characteristics of GPM satellite retrievals, in metrics and at data latencies desired by GV client communities, to enable optimal use of GPM precipitation retrieval products for scientific research and applications. 2. Foster process of continuous rainfall algorithm improvement by use of fluid and low latency data communication between GPM GV Supersites and GSFC-PPS.
International GPM Program and Mission
641
3. Develop, archive, and make publicly available through the Internet, standard GV data products applicable for GV research and for developing improved methodologies in conducting GV process. 4. Create new GV technologies, emphasizing technologies that lead to higher accuracies, greater precisions, and ultimate greater understanding of precipitation’s physical processes.
6
GPM MISSION SCIENCE RESEARCH AND APPLICATIONS PROGRAM
For clarity, it is generally useful to condense the various goals and objectives of any scientific project into one overarching goal. In essence, the one central goal to which the GPM Mission is dedicated is to significantly improve the physical and quantitative understandings of the Earth’s water cycle at a hierarchy of scales – to enable improved climate, weather, and hydrometeorological forecasting. This is a realistic and achievable goal from both theoretical and practical perspectives, because of the various improvements being made to the GPM Mission’s precipitation measurements insofar as: (1) sampling, (2) global coverage, (3) measurement consistency (4) dynamic range, (5) accuracy and precision of instantaneous retrievals, (6) microphysical acuity, and (7) realism in diagnosed DSD properties latent heating. Given the new capability for high-frequency diurnal sampling, GPM measurements lend themselves to: •
• • •
An enhanced capability to achieve water budget closure and the associated understanding of how to better formulate the water budget in prognostic models, including climate, weather, and hydrometeorological forecasting models. A better stabilized and more accurate/precise climatic record of precipitation and its variations, DSD variability, and four-dimensional structure of latent heating. An advanced means of conducting diabatic initialization and continuous rainfall data assimilation in NWP models. A greatly improved means of forcing hydrometeorological models coupled with process models for such applications as flood routing, regional flood–drought forecasting, determination of agricultural water supplies, and evaluation–prediction of fresh water resources.
The global coverage and measurement consistency provided by the GPM constellation architecture, the use of calibration referencing to generate biasfree global data sets, and extending the dynamic range of the measurements
E. A. Smith et al.
642
into the light rain, warm rain, drizzle, and snowfall spectrums enabled by the Core satellite’s DPR radar and the EGPM satellite’s EMMR/NPR radiometerradar suite, ensure the acquisition of a far superior global climatic record of precipitation, including the important mid- and high-latitude zones. Also, because of the improved instruments on the Core and EGPM satellites, the accuracy and precision of the reference retrievals will improve significantly, an effect that will extend across all measurements from the constellation members, and thus all scientific applications insofar as water cycle research and prognostic modeling. Insofar as microphysical acuity, the DPR’s capability of measuring differential reflectivity, besides improving the quality of the reference retrievals, will add a new dimension in climate monitoring in conjunction with a fundamental property of precipitating clouds, i.e., their three-term microphysical DSD vector. There are various ways to formulate this vector, e.g., given in terms of average median drop size (), normalized concentration offset (), and shape factor (<μ>) (see Testud et al. 2001) – or given in terms of effective radius (reff), effective variance (νeff), and liquid water content (LWC) (see Kuo et al. 2004). Finally, because the vertical structure of the retrieved precipitation profiles will improve insofar as accuracy, precision, and microphysical acuity, the diagnostic calculations of the latent heating profiles will improve accordingly. This will have beneficial effects, including: (1) stimulating new techniques in diabatic initialization and continuous data assimilation in NWP models, (2) improving understanding of how latent heating released by precipitation influences the atmosphere’s general circulation and the resultant influence of general circulation change on climate change, and (3) ascertaining the degree to which climate change feeds back on precipitation and latent heating – the latter two mechanisms representing fundamental processes within the global water cycle.
6.1 GPM’s principal scientific objectives Given the broad scientific capabilities enabled by the GPM Mission, a set of four principal mission scientific objectives have emerged: 1. Improve water budget closure and thus quantitative understanding and physical formulation of Earth’s water cycle through: (a) More accurate and precise precipitation measurements enabled by improved radar-radiometer instrument suites on Core and EGPM satellites, as well as additional improvements on other microwave radiometers and ancillary instruments on additional constellation members. (b) Improved capability of achieving water budget closure from more robust measuring system (i.e., sampling and coverage) and
International GPM Program and Mission
643
understanding of how to better formulate water budget process in physical process models. (c) Better understanding of uncertainties in precipitation measurements (which drive uncertainties in water budget) and thus better consequent understanding of uncertainties in other water budget terms. 2. Improve climate simulations and climate reanalyses through: (a) Progress in accurate quantification of trends and space-time variations of rainfall (and associated error bars). (b) Improvements in achieving water budget closure from low to high latitudes. (c) Use of rainfall data assimilation in GCM reanalysis. (d) Better understanding of relationship between rain microphysics (as given by three-term DSD vector) plus associated latent heat profile and general circulation/climate variations as mediated by accompanying accelerations of both atmospheric and surface branches of global water cycle. 3. Improve diabatic data initialization, continuous rainfall data assimilation, and resultant weather forecasting through: (a) More accurate, more precise, diurnally sampled, and globally distributed measurements of instantaneous precipitation rate and latent heat release. (b) Development of advanced NWP techniques in diabatic initialization and continuous rainfall assimilation – with retrieval error characterization. 4. Improve hydrometeorological models and their application to hazardous flood forecasting, seasonal–regional flood–drought outlooks, agrometeorological conditions, and assessment–prediction of fresh water resources through: (a) Capability of forcing hydrometeorological models with accurate, diurnally sampled, and full-continental-coverage high-resolution precipitation measurements (including snowfall). (b) Better closure of continental water budget and consequent understanding and formulation of continental water cycle. (c) More innovative designs of hydrometeorological models used for applications, particularly hazardous flood forecasting, seasonal/regional flood–drought outlooks, determination of agrometeorological conditions, and fresh water resources assessment–prediction.
6.2 GPM mission’s principal scientific themes In seeking to achieve the scientific objectives of the GPM Mission, various scientific strategies have evolved within the different NASA, JAXA, and
644
E. A. Smith et al.
ESA Science Teams that will accomplish the lion’s share of the initial research. In the case of NASA, the strategy is one of dividing the research process into different themes, emphasizing the fundamental barrier problems. Thus, for NASA’s GPM Scientific Implementation Plan (GPM SIP – see Smith et al. 2004b), there are nine principal research themes which help organize the research and guide how research working groups will evolve. These nine themes and main topic areas for research are as follows: 1. Climate diagnostics (a) Refinement and extension of precipitation climatologies including snowfall climatologies. (b) Detection of statistically significant global and regional precipitation trends and space-time variations. 2. GWEC and hydrometeorological predictability (a) GWEC analyses and modeling (b) Water transports (c) Water budget closure (d) Hydrometeorological modeling and applications 3. Climate change and climate predictability (a) Climate-change analyses and prediction (b) Climatic water-radiation states and interactions (c) GWEC response to climate change, particularly water cycle accelerations and feedbacks 4. Data assimilation and precipitating storm predictability (a) Diabatic initialization and continuous rainfall data assimilation (b) NWP technique development 5. Marine boundary layer fluxes and surface processes (a) Air-sea interface processes and flux modeling (b) Generation of E – P data sets (c) Ocean mixed layer salinity changes (d) Freshwater subduction 6. Land surface processes (a) Land–atmosphere interface processes and heat-moisture fluxes/flux modeling (b) Integrated radiation-energy-water-carbon budget process modeling 7. Coupled cloud-radiation models (a) Precipitating storm simulations and diagnosis of cloud dynamics, macrophysical/microphysical processes, and evolving response of three-dimensional radiation field including latter’s relationship to hydrometeor and latent heat profiles
International GPM Program and Mission
645
(b) Physical parameterizations of microphysics through bulk and bin schemes (c) Advanced radiative transfer in nonhydrostatic mesoscale cloud resolving models. 8. Precipitation retrieval, validation, and synthesis (a) Algorithm development for physical retrieval of precipitation, DSD vector, latent heating, and scale-normalized measurement blending (b) Algorithm calibration, referencing, and normalization techniques (c) Algorithm validation and quantification of retrieval error characteristics (d) Synthesis of validation measurements for algorithm improvement 9. Public outreach, education, and service applications (a) Creation and broadcast of real-time global precipitation maps and other media products (b) K-12 and college-level Internet-based precipitation data access and educational display-analysis tools (c) Weather forecasting (d) Hydrometeorological applications involving flood forecasting, seasonal-regional flood-drought outlooks, agrometeorology, fresh water resources management, and other relevant applications (e) Precipitation alerts for human health system, transportation industry, construction industry, and other clients (f) Involvement of community organizations (e.g., schools, churches, nonprofit clubs) in rain-gauge-based data collection for micrometeorological verification studies.
6.3 GPM mission’s principal scientific challenges The achievement of success in any major scientific project is inevitably faced with challenges – the GPM mission being no exception to the rule. However, in the view of the authors, all of the major GPM challenges are intellectual in nature – not challenges of the engineering, technological, financial, or political kind. This is fortunate because intellectual challenges can be overcome with the standard tools of open debate, scientific persuasion, and cooperation. Most key challenges arise because in moving from the TRMM to the GPM era, it has become incumbent to use precipitation measurements in a proactive manner for research, rather than the more contemplative manner of traditional research. This is because access to frequent, global and accurate rainfall measurements has the benefit
646
E. A. Smith et al.
of improving environmental forecasts, emphasizing that improving such forecasts has become an issue of expedience. To summarize, the foremost of these challenges are: 1. Shifting of intellectual inquiry paradigm from “curiosity-driven” to “key problem-driven” through Science Team and Science Working Group coordination at national and international levels, thus ensuring that participating nations are deriving full benefit from overall mission and setting stage for follow-up generation’s Operational Global Precipitation Mission (OGPM). 2. Shifting of research focus paradigm from “measuring takes precedent” to “both measuring and prediction take precedent” through recommendations to various GPM Science Teams, thus enabling GPM products to achieve maximum scientific impact. 3. Shifting of precipitation phase paradigm from “liquid rain-centric retrieval” to “liquid/frozen precipitation-centric retrieval” through exploitation of HF radiometer channels on any and all constellation members plus well thought out strategy for integration of Core and EGPM satellite measurements into calibration reference data pool – thus expanding measurement of precipitation spectrum and enhancing richness of consequent datasets for studies of Earth’s water budget and water cycle. 4. Shifting of derived products release paradigm from “cautiousdelayed release policy” to “cautious-aggressive release policy” through direct modeler involvement in assessment of GPM precipitation products (liquid and frozen precipitation rates, DSD vectors, latent heat profiles, retrieval error characteristics, algorithm breakdown reports), thus expediting testing and impact of GPM precipitation products on environmental forecasts. 5. Shifting of fast data delivery paradigm from “only operational users need them” to “research users need them too” through transfer of specialized GPM data products from GPM’s PPS nodes to research partners conducting experimental real-time and near-real-time prediction experiments, thus guaranteeing that experimental forecast centers conduct research under equivalent “available knowledge” conditions of operational centers, a critical element in producing meaningful and relevant improvements in prediction models. 6. Shifting of GV paradigm from “post-mission assessment, generally involving comparison of scatter diagrams” to “near-realtime delivery to clients, including use of physical error modeling” through deployment of high-quality GV instrumentation over modest global network of GV Supersites which include GV-SICs, thus allowing low bandwidth, site-specific retrieval data to flow out from PPS nodes at realtime to GV-SICs, who in turn return low-bandwidth, site-
International GPM Program and Mission
647
specific GV information to GSFC-PPS node at near real time to support client-based retrieval error characterization and algorithm improvement. 7. Shifting of cloud and precipitation paradigm from “these are separate and distinct problems” to “this is microphysical-dynamic continuum” through integration of scientific deliberations and field program activities involving cloud-focused missions (particularly CloudSat and EarthCare if approved) and precipitation-focused missions (i.e., GPM), thus producing better understanding, and modeling of cloud-precipitation life cycles, and ultimately better textbooks and more advanced teaching of cloud and precipitation physics.
6.4 GPM mission’s anticipated systemic advances One noteworthy benefit of the GPM Mission, which involves a diverse array of scientists, engineers, and program managers, is that by the end of the mission, there will be systemic advances in various scientific, technological, instrumental, and programmatic areas involving rain measuring. In the author’s opinion, six advances are envisioned to take place during the GPM era: 1. Demonstration of efficacy of global precipitation measuring program for improvements in various types of numerically based environmental prediction problems. 2. Evolution of Operational Global Precipitation Measurement (OGPM) system. 3. Establishment of permanent Global Precipitation Processing System (GPPS) capability for acquiring, processing, distributing, and archiving global precipitation data products defined by GPM partners and stakeholders. 4. Development of new airborne and space flight-qualified PR s, involving dual- and perhaps triple-frequency systems with common apertures, and eventually including multiparameter sensing capabilities, particularly broadband Doppler velocity and send–receive polarization-diversity. 5. Creation of long-lived precipitation validation capability comprised of global network of GV Sites and Supersites – allowing flexibility for change, involving standard rain gauge, disdrometer, and weather radar systems and radar networks, and also including high-tech raingauge disdrometer type instruments, multiparameter radars and radar systems, and specialized upward-looking radars, radiometers, and LF Doppler profilers.
E. A. Smith et al.
648
6. Improvement in design and reduction of costs for both real and synthetic aperture PMW radiometers.
7
CONCLUSIONS
This paper provides a current overview of the GPM Program and eventual GPM Mission. Whereas much remains to be accomplished to ensure a successful mission, including the positive outcome of various important decisions by the engaged agencies, particularly NASA, JAXA, and ESA (but also including the IPO and CNES/ISRO consortiums, CMA’s NSMC in China, a number of nations and organizations who intend to provide GV facilities and/or science and engineering expertise, and other current or yetto-be-committed participants), it is fair to say that the basic contours of the mission have been defined and the fundamental scientific objectives identified and outlined. Hopefully, the GPM measurement concept, the internationally based GPM research program, and the eventual GPM Mission presented here address these concerns and will adjust to the changing needs and requirements of the world insofar as the need for water experts to measure and understand precipitation, and most importantly, to use this information wisely in confronting issues pertaining to the global water cycle, fresh water resources, and the water needs of the Earth’s population. Acknowledgements: Research support for the lead author has been provided by the GPM Mission Formulation Project within NASA’s Division of Earth Science. Support for additional coauthors has been provided by their respective institutions under targeted funding for the development of the GPM and EGPM missions.
8
REFERENCES
AMM, 2002: Cloud Systems, Hurricanes, and the Tropical Rainfall Measuring Mission (TRMM): A tribute to Dr. Joanne Simpson (W.-K. Tao and R. Adler, eds.). AMS Meteorological Monographs, 59, 234 pp. Asrar, G., J. A. Kaye, and P. Morel, 2001: NASA research strategy for earth system science: Climate component. Bull. Amer. Meteor. Soc., 82, 1309–1329. Baumgartner, A. and E. Reichel, 1975: The World Water Balance. Oldenburg Verlag, München, 179 pp. Bringi, V. N. and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar. Cambridge University Press, Cambridge, 636 pp. Chahine, M. T., 1992: The hyrdrological cycle and its influence on climate. Nature, 359, 373– 380. Chang, F.-C. and E. A. Smith, 2001: Hydrological and dynamical characteristics of summertime droughts over US Great Plains. J. Climate, 14, 2296–2316. Chen, T.-C., J. Pfaendtner, and S.-P. Weng, 1994: Aspects of the hydrological cycle of the ocean-atmosphere system. J. Phys. Oceanogr., 24, 1827–1833.
International GPM Program and Mission
649
Dai, A. G. and T. M. L. Wigley, 2000: Global patterns of ENSO-induced precipitation. Geophys. Res. Lett., 27, 1283–1286. da Silva, A. M., C. C. Young, and S. Levitus, 1994: Atlas of Surface Marine Data 1994, Vol. 1: Algorithms and Procedures. NOAA Atlas NESDIS 6–8, US Dept. of Commerce, NOAA, NESDIS, 83 pp. Doviak, J. D. and D. S. Zrnič, 1984: Doppler Radar and Weather Observations. Academic Press, New York, 458 pp. Dyck, S. and G. Peschke, 1995: Grundlagen der Hydrologie. Verlag für Bauwesen, Berlin, 536 pp. Easterling, D. R., G. A. Meehl, C. Parmesan, S. A. Changnon, T. R. Karl, and L. O. Mearns, 2000: Climate extremes: Observations, modeling, and impacts. Science, 289, 2068–2074. EOS, 2003: Proceedings. , multiple documents. ESE, 2003: NASA Earth Science Enterprise Strategy. , 69 pp. Gleick, P. H., 1989: Climate change, hydrology, and water resources. Rev. Geophys., 27, 329– 344. Graham, N. E., 1995: Simulation of recent global temperature trends. Science, 267, 666–671. Haddad, Z. S., E. A. Smith, C. D. Kummerow, T. Iguchi, M. R. Farrar, S. L. Durden, M. Alves, and W. S. Olson, 1997: The TRMM ‘Day-1’ radar/radiometer combined rainprofiling algorithm. J. Meteor. Soc. Japan, 75, 799–809. Hagemann, S., K. Arpe, and L. Bengtsson, 2003: Validation of the hydrological cycle in the new ERA40 dataset. EGS-AGU-EGU Joint Assembly/Session on Climate: Past, Present, Future, Held 6–11 April 2003, Nice, France. Hong, Y., T. R. Wilheit, and W. R. Russell, 1997: Estimation of monthly rainfall over oceans from truncated rain-rate samples: Applications to SSM/I data. J. Atmos. Oceanic Technol., 14, 1012–1022. Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, and D. Xiaosu, 2001: Climate change 2001: The scientific basis. Cambridge University Press, Cambridge, 944 pp. Huang, B. and V. M. Mehta, 2004: The response of the Indo-Pacific warm pool to interannual variations in net atmospheric freshwater. J. Geophys. Res., 109, C06022, doi:10.1029/ 2003JC002114. Huang, B., V. M. Mehta, and N. Schneider, 2005: Oceanic response to idealized freshwater flux in the Pacific at decadal timescale. J. Phys. Oceanogr., 35, 2467–2486. Huffman, G. J., R. F. Adler, M. M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 36–50. Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain-profiling algorithm for the TRMM precipitation radar. J. Appl. Meteor., 39, 2038–2052. Iguchi, T., R. Oki, E. A. Smith, and Y. Furuhama, 2002: Global Precipitation Measurement program and the development of dual-frequency precipitation radar. J. Comm. Res Lab., 49, 37–45. JAM, 2000: Joanne Simpson Special Issue on the Tropical Rainfall Measuring Mission (TRMM). AMS Journal of Applied Meteorology (M. Garstang, C.D. Kummerow, eds.), Vol. 39 (No. 12/Part 1), 1961–2301. JCLIM, 2000: TRMM Special Issue. AMS Journal of Climate (D. Randall, C.D. Kummerow, eds.), Vol. 13 (No. 23), 4069–4228. JRSSJ, 1998: A Special Issue on TRMM. J. Remote Sensing Society of Japan (T. Kozu, ed.), Vol. 18 (No.5). Karl, T. R. and R. W. Knight, 1997: The 1995 Chicago heat wave: How likely is a recurrence? Bull. Amer. Meteor. Soc., 78, 1107–1119. Kennel, C., E. Frieman, B. Moore, and L. Shaffer, 2002: Report on the Post-2002 Mission Planning Workshop [24–26 August 1998; Easton, Maryland]. NASA Earth Science Enterprise, , 28 pp.
650
E. A. Smith et al.
Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Rem. Sens., 34, 1213–1232. Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The tropical rainfall measuring mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15, 809–817. Kummerow, C., J. Simpson, O. Thiele, W. Barnes, A. T. C. Chang, E. Stocker, R. F . Adler, A. Hou, R. Kakar, F. Wentz, P. Ashcroft, T. Kozu, Y. Hong, K. Okamoto, T. Iguchi, H. Kuroiwa, E. Im, Z. Haddad, G. Huffman, B. Ferrier, W. S. Olson, E. Zipser, E. A. Smith, T. T. Wilheit, G. North, T. Krishnamurti, and K. Nakamura, 2000: The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. J. Appl. Meteor., 39, 1965–1982. Kummerow, C., Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, D.-B. Shin, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. Kuo, K.-S., E. A. Smith, Z. Haddad, E. Im, T. Iguchi, and A. Mugnai, 2004: Mathematicalphysical framework for retrieval of rain DSD properties from dual-frequency Ku/Ka-band satellite radar. J. Atmos. Sci., 61, 2349–2369. Lau, K.-M., Lau, J. H. Kim, and Y. Sud, 1996: Intercomparison of hydrologic processes in AMIP GCMs. Bull. Amer. Meteor. Soc., 77, 2209–2227. Levitus, S., J. I. Antonov, J. L. Wang, T. L. Delworth, K. W. Dixon, and A. J. Broccoli, 2001: Anthropogenic warming of Earth’s climate system. Science, 292, 267–270. Levizzani, V., P. P. Alberoni, P. Bauer, L. Bottai, A. Buzzi, E. Cattani, M. Cervino, P. Ciotti, M. J. Costa, S. Dietrich, B. Gozzini, A. Khain, C. Kidd, F. S. Marzano, F. Meneguzzo, S. Migliorini, A. Mugnai, F. Porcù, F. Prodi, R. Rizzi, D. Rosenfeld, L. Schanz, E. A. Smith, F. Tampieri, F. Torricella, J. F. Turk, G. A. Vicente, and G. Zipoli, 2000: Use of the MSG SEVIRI channels in a combined SSM/I, TRMM and geostationary IR method for rapid updates of rainfall. Proc 1st MSG-RAO Workshop (May 17–19; Bologna, Italy), European Space Agency Publ. Div., ESA Sp-452, 63–66. Levizzani, V., P. Bauer, A. Buzzi, S. Davolio, D.E. Hinsman, C. Kidd, F. S. Marzano, F. Meneguzzo, A. Mugnai, J. P. V. Poiares Baptista, F. Porcú, F. Prodi, J. F. W. Purdom, D. Rosenfeld, J. Schmetz, E. A. Smith, F. Tampieri, F. J. Turk, and G. A. Vincente, 2001: EURAINSAT – Looking into the future of satellite rainfall estimations. Proc. 2001 EUMETSAT Meteorological Satellite Data User’s Conf., EUMETSAT, Darmstadt, Germany, 375–384. Lin, X., E. A. Smith, G. Asrar, W. J. Adams, J. P. V. Poiares Baptista, M. Byers, M. Cleave, M. Debois, J. Durning, D. F. Everett, Y. Furuhama, A. Ginati, T. Iguchi, P. Ingman, E. Im, C. Ishida, P. Joe, R. Kakar, D. Kendall, M. Kojima, C. Kummerow, V. Levizzani, A. Mugnai, K. Nakamura, S. P. Neek, R. Oki, G. Raju, D. A. Randall, S. Schneider, J. M. Shepherd, B. J. Sohn, J. Testud, G. Withee, and W.-J. Zhang, 2004: OSSE analysis of Global Precipitation Measurement (GPM) Mission’s satellite constellation using GCM. Bull. Amer. Meteor. Soc., in preparation. Mariotti, A., M. V. Struglia, N. Zeng, and K.-M. Lau, 2002: The hydrological cycle in the Mediterranean region and implications for the water budget of the Mediterranean Sea. J. Climate, 15, 1674–1690. McCarthy, J. J., O. F. Canziani, N. A. Leary, D. J. Dokken, and K. S. White (eds.), 2001: Climate Change 2001: Impacts, Adaptation and Vulnerability. Cambridge University Press, Cambridge, 1000 pp. Meneghini, R., K. Nakamura, C. W. Ulbrich, and D. Atlas, 1989: Experimental tests of methods for the measurement of rainfall rate using an airborne dual-wavelength radar. J. Atmos. Oceanic Technol., 6, 637–651. Meneghini, R. and T. Kozu, 1990: Spaceborne Weather Radar. Artech House, Boston, 199 pp.
International GPM Program and Mission
651
Meneghini, R., T. Kozu, H. Kumagai, and W. C. Boncyk, 1992: A study of rain estimation methods from space using dual-wavelength radar measurements at near-nadir incidence over ocean. J. Atmos. Oceanic Technol., 9, 364–382. Meneghini, R., T. Iguchi, T. Kozu, L. Liao, K. Okamoto, J. Jones, and J. Kwiatkowski, 2000: Use of the surface reference technique for path attenuation estimates from the TRMM Precipitation Radar. J. Appl. Meteor., 39, 2053–2070. Milly, P. C. D., R. T. Wetherald, K. A. Dunne, and T. L. Delworth, 2002: Increasing risk of great floods in a changing climate. Nature, 415, 514–517. Mugnai, A., S. Di Michele, E. A. Smith, F. Baordo, P. Bauer, B. Bizzarri, P. Joe, C. Kidd, F.S. Marzano, A. Tassa, J. Testud, and G. J. Tripoli, 2007: Snowfall measurements by proposed European GPM mission. Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 655–674. Nakamura, K., K. Okamoto, T. Ihara, J. Awaka, and T. Kozu, 1990: Conceptual design of rain radar for the Tropical Rainfall Measuring Mission. Int. J. Sat. Comm., 8, 257–268. Nykanen, D. K., E. Foufoula-Georgiou, and W. M. Lapenta, 2001: Impact of small-scale rainfall variability on larger-scale spatial organization of land-atmosphere fluxes. J. Hydrometeor., 2, 105–121. Okamoto, K., J. Awaka, and T. Kozu, 1988: A feasibility study of rain radar for the Tropical Rainfall Measuring Mission. 6: A case study of rain radar system. J. Comm. Research Lab., 35, 183–208. Okamoto, K. and T. Kozu, 1993: TRMM Precipitation Radar algorithms. Proc. IGARSS’93, IEEE Geoscience and Remote Sensing Society, 426–428. Oki, T., 1999: The global water cycle. Global Energy and Water Cycles (K.A. Browning and R.J. Gurney, eds.), Cambridge University Press, New York, 10–27. Olson, W. S., Y. Hong, C. D. Kummerow, and J. Turk, 2001: A texture-polarization method for estimating convective – stratiform precipitation area coverage from passive microwave radiometre data. J. Appl. Meteor., 40, 1577–1591. Olson, W. S., S. Yang, J. Stout, and M. Grecu, 2007: The Goddard Profiling Algorithm (GPROF): Description and current applications. Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 179– 188. Peixoto, J. P. and A. H. Oort, 1992: Physics of Climate. American Institute of Physics, New York, 520 pp. Roads, J. and A. K. Betts, 2000: NCEP-NCAR and ECMWF reanalysis surface water and energy budgets for the Mississippi river basin. J. Hydromet., 1, 88–94. Roedel, W., 1992: Physik unserer Umwelt: Die Atmosphäre. Springer Verlag, Berlin, 457 pp. Schulz, J., 2002: Freshwater flux. Encyclopedia of Atmospheric Science: Vol. I, J. R. Holton, J.A. Curry, and J.A. Pyle, eds., Academic Press, Amsterdam, 75–83. Shepherd, J. M., E. A. Smith, G. Asrar, and R. Kakar, 2004: Scientific case for international Global Precipitation Measurement (GPM) Mission. Bull. Amer. Meteor. Soc., in preparation. Shin, D.-B. and C. Kummerow, 2003: Parametric rainfall retrieval algorithms for passive microwave radiometers. J. Appl. Meteor., 42, 1480–1496. Simpson, J., R. F. Adler, and G. North, 1988: A Proposed Tropical Rainfall Measuring Mission (TRMM) satellite. Bull. Amer. Meteor. Soc., 69, 278–295. Simpson, J., C. Kummerow, W.-K. Tao, and R. F. Adler, 1996: On the Tropical Rainfall Measuring Mission (TRMM) satellite. Meteor. Atmos. Phys., 60, 19–36. Smith, E. A., J. Turk, M. Farrar, A. Mugnai, and X. Xiang, 1997: Estimating 13.8 GHz path integrated attenuation from 10.7 GHz brightness temperatures for TRMM combined PRTMI precipitation algorithm. J. Appl. Meteor., 36, 365–388. Smith, E. A., J. Lamm, R. Adler, J. Alishouse, K. Aonashi, E. Barrett, P. Bauer, W. Berg, A. Chang, R. Ferraro, J. Ferriday, S. Goodman, N. Grody, C. Kidd, D. Kniveton, C. Kummerow, G. Liu, F. Marzano, A. Mugnai, W. Olson, G. Petty, A. Shibata, R. Spencer,
652
E. A. Smith et al.
F. Wentz, T. Wilheit, and E. Zipser, 1998: Results of WetNet PIP-2 project. J. Atmos. Sci., 55, 1483–1536. Smith, E. A., W. J. Adams, G. M. Flaming, S. P. Neek, and J. M. Shepherd, 2002: The Global Precipitation Measurement (GPM) Mission. Mediterranean Storms 2001 (edited by R. Deidda, A. Mugnai, and F. Siccardi), Proceedings of 3rd EGS Plinius Conference on Mediterranean Storm (Baja Sardinia, Italy; October 2001), 175–183. Smith, E. A., G. Asrar, A. Mehta, J. M. Shepherd, P. Bauer, R. Bras, V. Chandrasekar, D. Carson, H. J. Cooper, M. Debois, J. Entin, R. Ferraro, E. Foufoula-Georgiou, A. Ghazi, B. Goodison, Z. Haddad, A. Hou, P. Houser, T. Iguchi, E. Im, P. Joe, J. Kaye, K.-S. Kuo, C. Kummerow, W. Lau, D. Lettenmeir, V. Levizzani, G. Liu, N. Lu, F. Marks, V. Mehta, R. Meneghini, P. Morel, A. Mugnai, K. Nakamura, T. Nakazawa, K. Okamoto, R. Oki, G. Raju, F. R. Robertson, S. Rutledge, D.-B. Shin, G. Skofronik-Jackson, B. J. Sohn, S. Sorooshian, G. Stephens, E. Stocker, W.-K. Tao, J. Testud, G. Tripoli, F.J. Turk, J. Weinman, T. Wiheit, E. Wood, S. Yang, and S. Yuter, 2004a: Global Precipitation Measurement (GPM) Mission: Past progress and future prospects. Bull. Amer. Meteor. Soc., in preparation. Smith, E. A., K.-S. Kuo, A. Mehta, J. M. Shepherd, E. Stocker, and S. Yang (eds.), 20004b: NASA’s Science Implementation Plan for GPM Mission. GPM Report Series #11, NASA/Goddard Space Flight Center, Greenbelt, MD, in preparation. Smith, E. A., C. V. Chandrasekar, E. Ebert, J. Goddard, P. Hwang, A. Mugnai , K. Nakamura, J.-C. Nam, C.-H. Sui, C. Wilson, and S.E. Yuter (eds.), 2005: International Ground Validation Research Programme of Global Precipitation Measurement (GPM) Mission: Report of 1st International GPM GV Requirements Workshop [4–7 November 2003; Abingdon, UK]. GPM Report Series #13, NASA/Goddard Space Flight Center, Greenbelt, MD, 102 pp. Soden, B. J., 2000: The sensitivity of the tropical hydrological cycle to ENSO. J. Climate, 13, 538–549. Sohn, B. J., E. A. Smith, F. R. Robertson, and S.-C. Park, 2004: Derived over-ocean water vapor transports from satellite-retrieved E–P datasets. J. Climate, 17, 1352–1365. Spencer, R. W., R. E. Hood, F. J. LaFontaine, E. A. Smith, J. Galliano, and E. Lobl, 1994: High resolution imaging of rain systems with the Advanced Microwave Precipitation Radiometer. J. Atmos. Oceanic Technol., 11, 849–857. Sumi, A., 2002: Rain as Seen from Space. Japan Advance Plan Co., Inc., Tokyo, 75 pp. Tesmer, J. R. and T. T. Wilheit, 1998: An improved microwave radiative transfer model for tropical oceanic precipitation. J. Atmos. Sci., 55, 1674–1689. Testud, J., S. Oury, R. A. Black, P. Amayenc, and X. Dou, 2001: The concept of “normalized” distribution to describe raindrop spectra: A tool for cloud physics and cloud remote sensing. J. Appl. Meteor., 40, 1118–1140. Todini, E. and L. Dümenil, 1999: Estimating large scale runoff. Global Energy and Water Cycles, K.A. Browning and R.J. Gurney, eds., Cambridge University Press, New York, 265–277. Trenberth, K. E., A. Dai, R. M. Rasmussen, and D. B. Parsons, 2003: The changing character of precipitation. Bull. Amer. Meter. Soc., 84, 1205–1217. Turk, F. J., E. A. Smith, E. Amitai, E. Ebert, J. Hawkins, F. S. Marzano, A. Mehta, A. Mugnai, and B. J. Sohn, 2004: Blending coincident TRMM, SSM/I, AMSU, AMSR, and GEO-IR satellite data for rapid updates of global rainfall. Bull. Amer. Meteor. Soc., in preparation. Wang, S. A., 1996: Modeling the beamfilling correction for microwave retrieval of oceanic rainfall. Ph.D. Dissertation, Department of Meteorology, Texas A&M University, College Station, TX, 100 pp. Wilheit, T. T., A. T. C. Chang, and L. S. Chiu, 1991: Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution functions. J. Atmos. Oceanic Technol., 8, 118–136.
International GPM Program and Mission
653
Wilheit, T., R. Adler, S. Avery, E. Barrett, P. Bauer, W. Berg, A. Chang, J. Ferriday, N. Grody, S. Goodman, C. Kidd, D. Kniveton, C. Kummerow, A. Mugnai, W. Olson, G. Petty, A. Shibata, E. Smith, and R. Spencer, 1994: Algorithms for the retrieval of rainfall from passive microwave measurements. Remote Sens. Rev., 11, 163–194. Ziegler, A. D., J. Sheffield, E. P. Maurer, B. Nijssen, E. F. Wood, and D. P. Lettenmaier, 2003: Detection of intensification in global- and continental-scale hydrological cycles: Temporal scale of evaluation. J. Climate, 16, 535–547.
49 SNOWFALL MEASUREMENTS BY PROPOSED EUROPEAN GPM MISSION Alberto Mugnai1, Sabatino Di Michele1, Eric A. Smith2, Fabrizio Baordo1, Peter Bauer3, Bizzarro Bizzarri1, Paul Joe4, Christopher Kidd5, Frank S. Marzano6, Alessandra Tassa1, Jacques Testud7, and Gregory J. Tripoli8 1
Istituto di Scienze dell’Atmosfera e del Clima, Consiglio Nazionale delle Ricerche, Roma, Italy Goddard Space Flight Center, National Aeronautics and Space Administration, Greenbelt, MD, USA 3 European Centre for Medium-range Weather Forecasts, Shinfield Park, Reading, UK 4 Meteorological Service of Canada, Downsview, Ontario, Canada 5 School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston Birmingham, UK 6 Dipartimento di Ingegneria Elettrica, Università dell’Aquila, L’Aquila, Italy 7 Novel Initiative for Meteorological and Environment Technologies (NOVIMET), Vélizy, France 8 Department of Atmospheric and Oceanic Sciences, University of Wisconsin, Madison, WI, USA 2
Abstract
The European Space Agency (ESA) is currently engaged in the study of the European Global Precipitation Measurement (EGPM) Mission. This represents both a dedicated ESA mission primarily supporting research and applications in the European Community and Canada, and a contributed spacecraft element and scientific research component for the International GPM Constellation Mission. The scientific payload of the EGPM satellite consists of a multispectral, conicallyscanning microwave radiometer and a 3-beam, nadir-pointing Ka-band rain radar. The radiometer operates at four window frequencies from cm to mm wavelengths (18.7, 36.5, 89.0, and 150.0 GHz), at one water vapor absorption frequency on the far wing of the 22.235 GHz H2O line (23.8 GHz), at four temperature sounding frequencies within 50–54 GHz on the low frequency wing of the 60 GHz molecular Oxygen complex, and at four complimentary-paired sounding frequencies in the vicinity of the strong 118.75 O2 line. The EGPM measurements will make contributions to a number of important scientific topics vis-à-vis precipitation science and the Earth’s climatically varying water cycle. In this investigation, we examine the specific contribution of the EGPM satellite to snowfall measurement. This is achieved by synthetically calculating both the radiative observations generated by the EGPM radar – radiometer instrument suite and the consequent precipitation estimates for a numerically simulated snowstorm. The storm simulation is performed by means of a time-dependent, 3-dimensional, cloud/ mesoscale model using explicit multi-species microphysics. The results demonstrate that because of its unprecedented sensor/channel capacity, the EGPM mission is expected to make significant improvements in the measurement of snowfall, to extend the observed precipitation spectrum as retrieved from space, and ultimately to improve the understanding and closure of the global water cycle.
655 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINST and the Future, 655–674. © 2007 Springer.
A. Mugnai et al.
656 Keywords EGPM, ESA, radar, precipitation, snow
1
INTRODUCTION
The Earth has a yearly precipitation average of some 690 mm of which about 5% is produced in the form of snowfall. There is a strong latitudinal influence in which the snowfall proportion increases dramatically as a function of latitude, particularly light snowfall, while the total precipitation amounts undergo steady decreases from the relative maxima positioned at the average latitudes of the northern and southern hemisphere midlatitude storm tracks (~40º N/45º S). As a consequence, in the central and northern regions of Europe, North America, and Asia, since snowfall represents a significant portion of the total precipitation amount, it becomes the main driver of the regional water cycle process, whose influence extends to the entire globe. For instance, while the typical annual average precipitation total in Canada is 535 mm, with 36% falling as snow, in northern Canada the proportion of snowfall to total precipitation ramps up to approximately 90%, almost all of which is produced in the form light snowfall. Therefore, light snowfall, and a smaller component of light rainfall, are key drivers to water cycling in mid to northern latitudes. This issue is manifest in seeking to understand how snow accumulation is changing in the context of the current global warming epoch (e.g., see Zwally 1989). These elements of precipitation are brought out in the latitude-precipitation rate category diagram shown in Fig. 1a, a diagram produced from the Comprehensive Ocean-Atmosphere Data Set (COADS). [COADS is a climatology of oceanic meteorological and sea surface observations recorded on ships; see da Silva et al. 1994.] The figure illustrates the latitudinal distribution of cumulative percentage of light snowfall occurrence (indicated in gray as frozen precipitation) and light rain occurrence (indicated in blue as liquid precipitation), as given by the COADS climatology. Note the white region above the colored region of the diagram indicates the cumulative remainder of moderate to heavy rain occurrences, where 1 mm h–1 is taken as the approximate threshold between light and moderate rain rate. Also, note there are minor contributions by mixed phase precipitation at the ocean surface. It is immediately evident from this figure how important light rainfall and snowfall, particularly light snowfall, are to the water budget poleward of the 30 degree parallels.
Snowfall Measurements by Proposed European GPM Mission
657
Figure 1a. Latitude-precipitation rate category diagram as derived from COADS dataset. [See text for explanation.]
Figure 1b provides a further illustration of the relevance of light rainfall and snowfall to the water budget poleward of ~45º N in the northern hemisphere. In the upper-left panel of this figure, there is a comparison of rain rates between Europe and the Tropics which emphasizes the great disparity between rain rate intensities between these two regions of the globe. In the upper-right panel, there is a set of cumulative distribution functions (CDFs), given in terms of snow water equivalent rate, at four measurement locations from mid to northern latitudes – showing that at least 90% of the precipitation occurs at rates less than 3 mm h–1. Finally, in the lower-left panel, there are a set of three graphs indicating snow to total precipitation ratios at three locations situated between 45ºN to 82ºN in Canada, indicating that in moving from the southern-most point to the northern-most point, snowfall fractions rise from ~30% to ~90%. In examining these two figures, it should be recognized that the core satellite of the GPM mission, which will also fly an advanced radarradiometer payload (i.e., a dual-frequency Ku/Ka-band radar and a hiresolution conically scanning radiometer), will only reach the ±65 degree parallels given its anticipated orbit inclination of 65 degrees. Therefore, all measuring of the high-latitude water budget will become the realm of the EGPM mission. Moreover, all measurement of light rainfall and snowfall producing radar reflectivity factor signatures below ~12 dBZ (i.e., sensitivity of the GPM core satellite’s Ka-band radar), must be done by the EGPM radar whose sensitivity will be considerably higher (~5 dBZ).
658
A. Mugnai et al.
Figure 1b. Upper left panel indicates differences in properties of rainfall intensity (abscissa) between Europe and Tropics in terms of CDFs of accumulation and occurrence. Upper right panel indicates CDFs as function of snow water equivalent rate (mm h–1) for four locations in mid to northern latitudes: (1) Fort Simpson, NW Territories, (2) Woodlands, Manitoba, (3) Franktown, Ontario, and nation of Finland (latitude-longitude coordinates given in parentheses). Lower left panel indicates snow to total precipitation ratio (in percent) for three different locations: (1) Ottawa, Ontario at ~45º N, (2) Yellow Knife, NW Territories at ~62º N, and (3) Alert, Ellesmere Island (NW Territories) at ~82º N.
Detailed knowledge of the spatial distribution of snowfall is vital for a number of scientific applications. Additionally, since climate models suggest that climate change will be a first most evident in polar regions, monitoring snowfall may provide the optimal means to test the hydrological component of the models. However, snowfall measurements are difficult to make due to strong wind effects on snow gauges – especially for light snowfall, which drifts, blows away, evaporates, and melts, often before it can be measured. In addition, ground-based measurement sites are sparse in remote or mountainous regions, thus resulting in poor knowledge of snowfall at such locations. Therefore, satellites are the only viable means for providing areally extended snowfall measurements – although current satellite sensor systems have not been specifically designed to measure snowfall accumulation or snowfall rate. In principle, such measurements can be made by space-borne microwave radiometers employing frequencies above 30 GHz that respond to scattering by ice particles. However, highly variable radiation emission by the underlying land surface generally obfuscates retrieval of snowfall based on measurements at microwave window frequencies (i.e., frequencies for which the atmosphere is relatively transparent with respect to the neighboring spectrum). That is why Special Sensor Microwave/Imager (SSM/I) radiometers and multichannel precursor radiometers that began operating in 1978
Snowfall Measurements by Proposed European GPM Mission
659
at window frequencies up to 85 GHz (Kramer 2002) have not been used successfully to measure snowfall from space. This problem has been partially overcome with the launch in 1998 of the first of the 5th generation of operational polar orbiting satellites of the US National Oceanic and Atmospheric Administration (i.e., NOAA-15). This platform carries the Advanced Microwave Sounding Unit (AMSU) – see Kramer (2002) for details. AMSU is a 20-channel microwave radiometer, comprised of three separate cross-scanning units: AMSU-A1, AMSU-A2 and AMSU-B. AMSU-A1 has 12 channels in the 50–60 GHz molecular oxygen band for atmospheric temperature profiling, while AMSU-B has two highfrequency window channels at 89 and 150 GHz and three split-band channels (183.31±1, 183.31±3, 183.31±7 GHz) surrounding the strong water vapor line at 183.31 GHz for atmospheric humidity profiling. These channels are sensitive to scattering by precipitating snow, and since they permit partial to full obscuration of emission from the underlying land surface, they can be used to retrieve snowfall within the atmosphere. Empirical relationships to detect snowfall from AMSU measurements have been recently published by Kongoli et al. (2003), while Skofronick-Jackson et al. (2004) are using a physical approach to retrieve snowfall over land from AMSU-B measurements. Snowfall detection and measurement is a main driving requirement of the EGPM mission, now in progress of a Phase A study by the European Space Agency (ESA). The EGPM mission has been proposed by a precipitation/ water budget-focused international scientific community, primarily from Europe and Canada, but also including scientists from the USA and Japan. The main scientific objectives of the EGPM mission are centric to current research problems in Europe and Canada, but the mission also has been proposed as a participant in the International Global Precipitation Measurement (GPM) Mission as a dedicated member of the GPM constellation. The article by Smith et al. (2007) in this same volume describes the International GPM Mission. The EGPM mission is specifically designed to detect and measure snowfall, light rain, and warm rain – both over land and ocean – particularly in mid and northern latitude climates. That is why the mission requires an advanced and unprecedented scientific payload. The payload consists of: (1) an innovative conically scanning microwave radiometer combining the conventional rain-measuring window channels of 18, 23, 37, and 85 GHz, a high-frequency window channel at 150 GHz, and four crosspaired temperature-sounding channels near 52 and 118 GHz (four channels in each O2 absorption region); and (2) a Ka-band (35.6 GHz) 3-beam nadirpointing precipitation radar having a sensitivity of ~5 dBZ. The sounding channels respond to scattering by snow and are less sensitive to surface emission than the window channels. Additionally, coupled radiometric observations within the strong oxygen absorption band between 50–54 GHz and near the strong O2 absorption line at 118.75 GHz, permit the
660
A. Mugnai et al.
accurate retrieval of precipitation information by exploiting differential atmospheric absorption within each band and differential hydrometeor scattering between the two bands. Choosing pairs of frequencies in the two O2 absorption regions that have similar weighting functions for clear skies over precipitating clouds, enables vertical partitioning (slicing) of the frozen cloud constituents. This partitioning results from a combination of: (a) differences in cross-frequency band upwelling brightness temperatures (TBs) originating from large hydrometeor scattering (with higher frequency TBs appearing colder than lower frequency TBs, i.e., differential scattering), and (b) differences in within-frequency band upwelling TBs originating from molecular oxygen emission (with far-wing TBs providing greater cloud depth penetration than near-wing TBs, i.e., differential absorption). In essence, a sequence of corresponding channel pairs provides profile information on snow amount from cloud top to monotonically increasing depths inside the cloud (depth being defined from the top-down perspective). It is important to recognize that the AMSU instrument does not have similar snow measuring potential because it lacks any type of coupling between its oxygen band and water vapor line sounding channels. The potential of this approach was first investigated by Gasiewski and Staelin (1990). Lately, Blackwell et al. (2001) performed an analysis of the observations that were taken in these two oxygen bands by the airborne NPOESS Aircraft Sounder Testbed-Microwave (NAST-M) instrument during the Convection and Moisture Experiment (CAMEX-3) campaign in Summer 1998. This instrument consists of two radiometers, one with 8 channels covering the 50–57 GHz portion of the oxygen complex (the so-called 54GHz radiometer), and the other with 9 split-band (double-sideband) channels around the 118.75 oxygen line (the so-called 118-GHz radiometer). More recently, Bauer and Mugnai (2003) used NAST-M observations of hurricane Bonnie to demonstrate the potential of these temperature-sounding channels for precipitation profile retrieval from space. Based on the above noted analyses, four pairs of sounding channels are proposed for the EGPM radiometer. The lower four channels match the NAST-M 54-GHz radiometer (53.75, 52.8, 51.76, and 50.30 GHz), these channels enabling ever-increasing penetration depths into the precipitating cloud. The corresponding channels near 118.75 GHz are 118.75±8.5, 118.75± 4.2, 118.75±2.3, and 118.75±1.2 GHz, respectively. These are determined by matching their clear-sky weighting functions with the four corresponding 50–54 GHz channels based on calculations with some 55,000 atmospheric profiles. In the following analysis, we examine the performance of the EGPM scientific payload (radar + radiometer) for snowfall measuring, by estimating with a radiative transfer model, the relevant radiometer and radar observations and respective precipitation retrievals associated with a major snowstorm over the eastern USA. The storm is initially simulated by means of a
Snowfall Measurements by Proposed European GPM Mission
661
time-dependent, 3-dimensional (3D) mesoscale model run in cloud-resolving model mode (CRM-mode) with explicit 2-water/4-ice microphysics.
2
SIMULATION OF US EAST COAST SNOWSTORM
We use a numerical simulation of a meteorologically prominent snowstorm which occurred over the US east coast during January 24–26, 2000. The simulation was produced by the University of Wisconsin Nonhydrostatic Modeling System (UW-NMS), a 3D, time-dependent mesoscale model, run in CRM mode with explicit/six-species microphysics. The six species consist of suspended liquid and frozen cloud hydrometeors (cloud drops and pristine ice crystals), precipitating rain drops, and three forms of precipitating ice particles (snowflakes, aggregates, and graupel-hail). A detailed description of the original model is found in Tripoli (1992). The simulation was run for a 53.5-h period from 00:00 UTC January 24 through 05:30 UTC January 26, initialized over the outer grid with ECMWF global analysis information configured within a 2.5-degree mesh. A 3-nest grid system was used with an outer grid resolution of 37.5 km, a transition mesh of 9.38 km grid resolution, and an inner, fine mesh grid resolution of 2.34 km. The inner domain covers a horizontal area of 468 × 468 km2, extending 17 km in the vertical – resolved to 39 levels with a height-dependent spacing using highest resolution at lowest levels. Over the first 12 h of the simulation, only the outer nest was invoked; the second nest was added at 12 h while the third was inserted at 24 h. The UW-NMS microphysics scheme was used for each of the three grids, predicting the mixing ratios of the six different hydrometeor species. The medium and inner grids were positioned in such a way to best capture the snow fields, being guided to move northward so as to follow the location of minimum pressure. Figure 2 provides a schematic illustration of the grid nest design. For the analysis, we use model outputs from the inner grid at 06:00 UTC of January 25, 2000. Figure 3 illustrates columnar equivalent water content (CWC) for the six hydrometeor species at this time step. It is evident that heavy rainfall is localized in the top-right (northeastern) portion of the domain (over warm ocean), with heavy snow present across an extended inland region (over cold terrain).
662
A. Mugnai et al.
Figure 2. Nested grid system used to perform UW-NMS simulation of January 24–26, 2000 snowstorm over US east coast.
Figure 3. CWCs of six hydrometeor species [cloud droplets (top-left), rain drops (top-middle), graupel (top-right), pristine crystals (bottom-left), snowflakes (bottom-middle), and aggregates (bottom-right)] for inner grid of UW-NMS snowstorm simulation at time step 1800 s (i.e., 06:00 UTC, January 25, 2000). Vertical lines indicate path of vertical crosssection given special attention in latter sections of the paper (see also color plate 20).
Snowfall Measurements by Proposed European GPM Mission
3
663
MICROWAVE RADIATIVE TRANSFER MODELING
In order to simulate the upwelling TBs that would be observed by the EGPM radiometer, we apply a 3D-adjusted plane-parallel radiative transfer (RTE) model to the microphysical outputs of the simulated snowstorm. The RTE model as been described in the studies of Roberti et al. (1994), Liu et al. (1996), Bauer et al. (1998), and Tassa et al. (2003). For most applications, RTE model-generated upwelling TBs over simulated satellite footprints would be calculated at the model resolution (generally higher than the radiometer resolution) and at the radiometer’s incident viewing angle (for EGPM this is 53°). The TBs would then be spatially filtered in order to reconcile with the radiometer’s effective resolutions for different channels. However, in this sensitivity study, we use upwelling TBs at nadir (0° incidence) and at model resolution (2.34 km) since they are more suitable to understand and evaluate the behavior of the different frequencies. The required inputs to the RTE model are suitable temperature/moisture profiles and temperature/emissivity of the surface, as well as vertical profiles of liquid/ice water contents (LWCs/IWCs) of the various hydrometeors – along with their single-scattering properties. Surface temperature and vertical profiles are provided by the UW-NMS simulation. Absorption by atmospheric gases at microwave frequencies are calculated according to the Liebe and Gimmestad (1978) and Liebe (1985) clear-moist air refractivity model that provides a combined water vapor – oxygen volume absorption coefficient. Surface emissivity is assigned dependent upon frequency and surface characteristics (land/ocean, surface roughness, type of soil and soil cover, soil humidity, etc.). While it can be highly variable depending on these parameters, for this study we assume constant values of 0.9 and 0.65 for land and ocean backgrounds, respectively (see Hewison and English 2000 for the emissivity of many land surfaces). The uncertainty in surface emissivity is taken into account within the Bayesian retrieval scheme developed by Mugnai et al. (2001) and Di Michele et al. (2003). This is accomplished by means of an error covariance matrix (see Di Michele et al. 2004), which accounts for TB sensitivity to parameter uncertainties and approximations used in the forward RTE model (e.g., Tassa et al. 2003, 2004). Finally, there is a need to determine the single-scattering properties of the various hydrometeor species. This is a straightforward calculation in Mie scattering if only pure water and ice spheres are considered, but can be a major challenge due to the wide variety of sizes, densities, and shapes of natural ice hydrometeors – especially for snowflakes and ice aggregates which are hydrometeors large enough to interact with the radiometer frequencies of interest, but whose shapes are typically in radical departure
664
A. Mugnai et al.
from spheres. However, whereas the UW-NMS model provides hydrometeor size distributions and densities, information on shape is not available from its microphysical parameterization scheme – which is designed strictly to exchange H2O mass between vapor and the different species of hydrometeors (under the allowed gas-hydrometeor or hydrometeor-hydrometeor mass transfer schemes) without considering the habits under which the mass transfers take place. For this reason, we take the customary assumption that all particles are spherical and homogeneous so that Mie theory can be applied (Bohren and Huffman 1983; Wiscombe 1980). Graupel particles are assumed to be spherical with densities nearly that of pure ice (i.e., ~0.92 g cm–1). They are assumed to be “equivalent homogeneous spheres” having an effective dielectric function obtained by combining the dielectric functions of ice and air (or water, in case of melting) according to the effective medium Maxwell–Garnett mixing theory for a two-component mixture of inclusions of air (water) in an ice matrix; see Bohren and Huffman (1983). On the other hand, snowflakes and aggregates are low-density, fluffy ice particles (as long as they are completely frozen). As a result, they should not be modeled according to Maxwell-Garnett mixing theory. This is especially the case for equivalent soft-ice spheres having very large asymmetry factors (>0.9). In essence, the application Maxwell–Garnett theory to assumed “equivalent homogeneous spheres” at the higher microwave frequencies does not adequately “cool” the upwelling radiation. This problem has been addressed by Skofronick-Jackson et al. (2004) in their attempt to retrieve snowfall over land from AMSU-B measurements, using a physically based retrieval algorithm relying on a CRM simulation of a winter storm. These authors were able to simulate TBs that match the AMSUB observations, and to obtain retrieved snow masses that are consistent with available radar measurements. They did so by extending the procedure of Grenfell and Warren (1999) to AMSU-B frequencies. They show that the single-scattering properties of nonspherical ice particles across the ultraviolet, optical, and infrared portions of the electromagnetic spectrum are well approximated by a collection of solid ice spheres having the same volume-tosurface-area ratios as nonspherical particles (see also Neshyba et al. 2003). Note that due to mass conservation, each nonspherical particle is represented not by just one sphere, but rather by a number (n) of equal size spheres. Based on these results, we represent each pristine ice crystal, snowflake, or aggregate particle with an ensemble of equivalent solid ice spheres, all having a diameter determined by the volume to cross-sectional area ratio (V/A) of the original nonspherical ice particle. The volume (V) is provided by the UW-NMS simulation. For calculating cross-sectional area (A), we use the observational relationship A/(πD2/4) = C0 * DC that has been published by Heymsfield and Miloshevich (2003) for several different individual particle habits. [Here, D (in cm) is the maximum diameter of the particle and the coefficients C0 and C (in appropriate cgs units) depend on ice particle habit.]
Snowfall Measurements by Proposed European GPM Mission
665
For pristine ice crystals and aggregates, we use values C0 = 0.18 and C = –0.2707 (indicated by the same authors) as appropriate averages for midlatitude, continental mixed-habit cirrus clouds. For snowflakes, we use values C0 = 0.6 and C = 0., which are consistent with those given in the same paper for several snow habits. As a result, the diameter (Ds) and the number (ns) of the equivalent solid-ice spheres are given by:
Ds =
ρ D ρ ice C 0 D C
ns =
C 0 D 3+ C 1 − C Ds 3
where a given density ρ of each of three ice categories (i.e., pristine ice crystal, snowflake, aggregate) is given by the UW-NMS model – and ρice = 0.916 g cm–3. To estimate reflectivities that would be measured by the EGPM 35.6 GHz radar, all ice particles can been modeled by use of Maxwell–Garnett mixing theory which is found to be adequate at this frequency according to Meneghini and Liao (2000).
4
SIMULATED EGPM OBSERVATIONS
To demonstrate the main features that would be observed by the EGPM payload instruments, we now focus on a vertical cross section along the most intense portion of the snowstorm (as is given in Fig. 3). This region is contains frozen precipitation over land. To better understand the relationships between the upwelling TBs that would be measured by the EGPM radiometer and the microphysical structure of the simulated storm, we illustrate in Fig. 4 the simulated TBs together with the columnar water contents for snowflakes and aggregates – and the vertical profiles of ice water content of total snow along the selected cross section. These results highlight the potential of the various radiometer frequencies for the retrieval of precipitating snow over land. Specifically: While the two lower window frequencies (18.7 and 23.8 GHz) show little sensitivity to precipitating snowfall over land, significant TB-depressions (more than 10ºC) are evident at 36.5 GHz in correspondence to the wide portion of the storm (between ~260–390 km along the cross-section), a region characterized by intense snowfall.
666
A. Mugnai et al.
Figure 4. Simulated EGPM radiometer observations for the eastern US snowstorm (over land). From top to bottom, individual panels show brightness temperatures (TBs) at: (1) lower frequency window channels, (2) higher frequency window channels, ( 3) lower frequency sounding channels, and (4) higher frequency sounding channels, plus (5) CWC) of key hydrometeors (snowflakes and ice aggregates), and (6) vertical cross-sections of snow IWC.
The higher window frequencies (89 and 150 GHz) respond very strongly to scattering by snow over land. TB-depressions up to some 80ºC are observed in correspondence to the deepest snow layer between 340–390 km along the cross-section. The fact that the 89 GHz TBs are significantly lower (up to about 40ºC) than those at 150 GHz between 250–320 km, can only be explained by the presence of super-cooled water and small ice particles which reduce the scattering efficiency at 150 GHz more than at 89 GHz. Note that between about 280–300 km, the 89 GHz TBs are warmed considerably (up to
Snowfall Measurements by Proposed European GPM Mission
667
about 30ºC) by the concurrent presence of a shallow layer of rain mixed with precipitating snow. Overall, the lower frequency (50–54 GHz) and higher frequency (118 GHz) sounding channels exhibit similar behavior. They show TB-depressions due to scattering by snow that increase with snow layer penetration depth – a feature less evident at 53.75 GHz and at 118.75±1.2 GHz because their weighting functions peak above the snow layers. In addition, the various channel pairs exhibit a similar response to rain/snow mixing between ~280– 300 km. Nevertheless, the magnitudes of the TB-depressions are much larger for the higher frequency channels due to the increase in scattering efficiency with frequency. They are as large as 20ºC and 80ºC for the 50–54 GHz band and the 118 GHz band, respectively, in correspondence to the deepest snow layer. To evaluate the performance of the EGPM radar, we illustrate in Fig. 5 the simulated attenuated radar reflectivity at 35.6 GHz for the same snowstorm cross-section of Fig. 4. The three dotted lines represent the limits of detection for 0, 5, and 16 dBZ.
Figure 5. Simulated radar reflectivity at 35.6 GHz for selected cross-section of eastern US snowstorm simulation. Dotted lines correspond to three different limits of radar detection sensitivity, specifically 0, 5, and 16 dBZ.
These results demonstrate that setting a requirement for the EGPM radar sensitivity of ~5 dBZ is very important for the observation of solid precipitation. There is a significant benefit in using this sensitivity limit, rather than something along the lines of the 17 dBZ sensitivity limit used for the Precipitation Radar (PR) onboard the Tropical Rainfall Measuring Mission (TRMM) satellite. As noted in Kramer (2002), the higher sensitivity threshold allows for the detection of light precipitation (which is found at <100 km along the cross-section) and precipitating aggregates (found at >380 km). In addition, with a sensitivity limit of ~5 dBZ, about 1 km of the topmost cloud layer will be detected, a cloud region that will nearly always interact with passive microwave radiometer observations at the higher frequencies.
A. Mugnai et al.
668
Moreover, there is little to gain in extending the radar sensitivity beyond 5 dBZ – as clearly illustrated in Fig. 6a–b.
Figure 6a. Same as upper right panel of Fig. 1b, except abscissa is given in term of radar reflectivity factor (dBZ).
5
SIMULATED EGPM RETRIEVALS
To highlight the improvement in snowfall measurement from space that would be obtained by the EGPM mission, a series of synthetic retrievals have been performed using a snow profiling algorithm (Di Michele et al. 2003) for different combinations of the EGPM radiometer channels, as well as for the EGPM radar-radiometer payload. Results for the selected cross-section of the snowstorm simulation are shown in Fig. 7, while Table 1 summarizes the average retrieval accuracy of the near-ground snow IWC for the four different retrievals illustrated in the Fig. 7 panels. These results can be summarized as follows: Retrievals based on the four lower window frequencies (i.e., SSM/I-like retrievals) do not provide an acceptable estimate of the snow profiles, but only gross differentiations among them. In particular, the heaviest profiles as well as snowfall just above the ground are greatly underestimated, while the structure of the lightest profiles is entirely missed (this being a consequence of a larger impact of surface emissivity on the observed TBs). Addition of the highest window frequency at 150 GHz largely improves the retrievals of the heaviest profiles and of snowfall just above the ground, by improving retrieval of the columnar snow contents. Nevertheless, there are cases (near 350 km) in which the snow profiles are incorrectly retrieved
Snowfall Measurements by Proposed European GPM Mission
669
because the snow content is mainly associated with upper levels rather than just above the ground – as with the model “truth”. In addition, retrieval of the lightest profiles remains unsatisfactory.
Figure 6b. Radar data analyses performed on Chill radar measurements obtained during heavy (left panel) and light (right panel) snowstorm cases from Fort Collins, CO – indicating 90% CDF accumulation coordinates with respect to reflectivity factor axes (abscisses). Associated 90% snow water equivalent coordinates are shown below CDF diagrams according to three different Z-R relationships for snow. [Data provided courtesy of Prof. S. Rutledge, Colorado State Univ., Ft. Collins.]
A. Mugnai et al.
670
Further addition of the sounding channels significantly improves the retrievals, because of their small sensitivity to land surface emission compared to the window channels and because of their vertical profiling capability. In particular, retrieval of the shallow snow layer above the ground at 250–330 km is improved; the double-layer snow profiles near 350 km are better retrieved; and even the main characteristics of the lightest profiles are captured by the retrieval – except the lack of precipitating snow above the ground between 400–450 km. Note that in this case, the average accuracy (bias and rms error) of near-ground snow profile retrievals is increased by a factor of 2. Finally, if the radiometer retrieval database is calibrated by vertical profile information obtained from radar observations, the combined radar-radiometer retrieval produces excellent results in every situation. Only in this case is a very low bias for the near-ground snow retrieval obtained. Also noteworthy is that a calibration of the radiometer database can be extended across the full swath of the radiometer. Table 1. Average retrieval accuracy of near-ground (up to 500 m) snow IWC for four different retrievals shown in Fig. 5. Percentage values are calculated with respect to “truth” average value provided by UW-NMS model (0.784 g m–3). Configuration
6
Bias [g m-3]
r.m.s. error [g m-3]
Radiometer with channels at 18.7, 23.8, 36.5, 89 GHz
–0.418 (–53%)
0.677 (86 %)
Radiometer with additional 150 GHz channel
–0.289 (–37%)
0.476 (61 %)
Radiometer with further additional (sounding) channels in the 50–54 and 118 GHz regions
–0.145 (–18%)
0.301 (38 %)
Radar-calibrated EGPM Radiometer
–0.026 (–3%)
0.104 (13 %)
CONCLUSIONS
A simulation study has been performed assessing the EGPM satellite’s radar radiometer sensitivity and retrieval performance for a meteorological relevant situation, i.e., a winter snowstorm. This type of precipitation event represents the primary observation target for the EGPM satellite. The main results of the investigation are summarized as follows: Precipitating snow over land or snow cover has distinct signatures at 89 and 150 GHz as well as at the paired sounding frequencies. The brightness temperature depressions due to scattering and the differential signatures at the 89–150 GHz frequencies and the 50/118 GHz frequency pairs contain
Snowfall Measurements by Proposed European GPM Mission
671
detailed information on the presence of snow, on the vertical depth of the snow layer and concomitant cloud structure, on the snowfall rate, and on coexisting liquid precipitation.
Figure 7. Synthetic snow IWC retrievals for selected cross-section of eastern US snowstorm simulation. Left panels from top to bottom are: (1) EGPM radiometer retrieval using only four lower window frequencies, i.e., 18.7, 23.8, 36.5 and 89 GHz (similar to four SSM/I frequencies), (2) as previously but using all five window frequencies including 150 GHz, (3) as previously but using all five window frequencies plus four pairs of sounding channels in 50–54 and 118 GHz regions, (4) combined EGPM radar-radiometer retrievals, and for comparison (5) model “truth”. For each retrieval case, associated right panels show estimated average profile (blue solid line) together with model “truth” average profile (red line), plus retrieval error standard deviations at various atmospheric levels (blue error bars) (see also color plate 20).
672
A. Mugnai et al.
The set of window frequencies provides near-ground snow retrieval accuracies just meeting EGPM requirements in terms of rms error (50%), but exhibit too large a bias. The sounding channels are significantly less sensitive to surface emissivity variations than the window channels and provide significantly more information on snowfall and snow cover over land surfaces. As a consequence, these retrieval experiments show an increase in average accuracy (reduction in both bias and rms error) of near-ground snow profile retrievals by a factor of 2 when the sounding channels are used – an improvement that is sufficient to meet EGPM requirements. However, both the high frequency window channels and the sounding channels are necessary to optimize the snowfall retrievals. The high sensitivity (5 dBZ) EGPM precipitation radar will be able to observe the full vertical profile of precipitating snow, even in the case of light snowfall. The radiometric retrievals can be significantly improved if the radar measurements are used as a constraint for the radiometer inversions. Only in this case do the near-ground snow retrievals exhibit the preferred low bias. These results demonstrate the potential of the full EGPM payload for snowfall measuring from space, and the enhancement to the International GPM mission made possible by EGPM participation in Earth Science’s first international constellation partnership. Acknowledgements: We wish to thank the EGPM Mission Advisory Group (MAG) of the European Space Agency (ESA) for all relevant discussions and inputs concerning the EGPM mission. [Besides five of the authors (A. Mugnai, P. Bauer, P. Joe, C. Kidd, and J. Testud), additional members of ESA’s EGPM MAG are Dr. Paul Ingmann of ESA-ESTEC (Noordwijk, The Netherlands), Professor Maria del Carmen Llasat of the University of Barcelona (Barcelona, Spain), and Professor Giorgio Roth of the University of Genova (Genova, Italy).]. We also wish to thank Dr. Pedro Poiares Baptista of ESA-ESTEC, Dr. Ralf Bennartz of the University of Wisconsin (Madison, Wisconsin, USA), Dr. Vincenzo Levizzani of CNR-ISAC (Bologna, Italy), and Dr. Jim Weinman of the University of Washington (Seattle, Washington, USA) for useful discussions and insights regarding the snowfall retrieval problem. This study has been funded by ESA within the context of the EGPM supporting scientific studies carried out in the frame of Phase A, by the Italian National Group for Prevention from HydroGeological Disasters (GNDCI), by the Italian Space Agency (ASI), and within the frame of EURAINSAT – a shared-cost project (contract EVG12000-00030) co-funded by the Research DG of the European Commission (5th Framework Program).
Snowfall Measurements by Proposed European GPM Mission
7
673
REFERENCES
Bauer, P., L. Schanz, and L. Roberti, 1998: Correction of the three-dimensional effects for passive microwave remote sensing of convective clouds. J. Appl. Meteor., 37, 1619– 1632. Bauer, P. and A. Mugnai, 2003: Precipitation profile retrieval using temperature-sounding microwave observations. J. Geophys. Res., 108(D23), 4730–4743. Blackwell, W. J., J. W. Barrett, F. W. Chen, R. V. Leslie, P. W. Rosenkranz, M. J. Schwartz, and D. H. Staelin, 2001: NPOESS Aircraft Sounder Testbed-microwave (NAST-M): Instrument description and initial flight results. IEEE Trans. Geosci. Remote Sens., 39, 2444–2453. Bohren, C. F. and D. R. Huffman, 1983: Absorption and Scattering of Light by Small Particles. Wiley, New York, 530 pp. da Silva, A. M., C. C. Young, and S. Levitus, 1994: Atlas of Surface Marine Data 1994, Volume 1: Algorithms and Procedures. NOAA Atlas NESDIS 6–8, U.S. Dept. of Commerce, NOAA, NESDIS, 83 pp. Di Michele, S., F. S. Marzano, A. Mugnai, A. Tassa, and J. P. V. Poiares Baptista, 2003: Physically-based statistical integration of TRMM microwave measurements for precipitation profiling. Radio Sci., 38, 8072–8088. Di Michele, S., A. Tassa, A. Mugnai, F. S. Marzano, P. Bauer, and J. P. V. Poiares Baptista, 2005: Bayesian algorithm for microwave-based precipitation retrieval: description and application to TMI measurements over ocean. IEEE Trans. Geosci Remote Sens., 43, 778– 791. Gasiewski, A. J. and D. H. Staelin, 1990: Numerical modeling of passive microwave O2 observations over precipitation. Radio Sci., 25, 217–235. Grenfell, T. C. and S. G. Warren, 1999: Representation of a nonspherical ice particle by a collection of independent spheres for scattering and absorption of radiation. J. Geophys. Res., 104, 31697–31709. Hewison, T. and S. English, 2000: Fast models for land surface emissivity. Radiative Transfer Models for Microwave Radiometry, C. Matzler, ed., COST Action 712, Directorate-General for Research, European Commission, Brussels, Belgium, 117–127. Heymsfield, A. J. and L. M. Miloshevich, 2003: Parameterizations for the cross-sectional area and extinction of cirrus and stratiform ice cloud particles. J. Atmos. Sci., 60, 936– 956. Kongoli, C., P. Pellegrino, R. R. Ferraro, N. C. Grody, and H. Meng, 2003: A new snowfall detection algorithm over land using measurements from the Advanced Microwave Sounding Unit (AMSU). Geophys. Res. Lett., 30, 1756–1759. Kramer, H. J., 2002: Observation of the Earth and its Environment: Survey of Missions and Sensors. Springer, 1510 pp. Liebe, H. J. and G. G. Gimmestad, 1978: Calculation of clear air EHF refractivity. Radio Sci., 13, 245–251. Liebe, H. J., 1985: An updated model for millimeter wave propagation in moist air. Radio Sci., 20, 1069–1089. Liu, Q., C. Simmer, and E. Ruprecht, 1996: Three-dimensional radiative transfer effects of clouds in the microwave spectral range. J. Geophys. Res., 101, 4289–4298. Meneghini, R. and L. Liao, 2000: Effective dielectric constants of mixed-phase hydrometeors. J. Atmos. Oceanic Technol., 17, 628–640. Mugnai, A., S. Di Michele, F. S. Marzano, and A. Tassa, 2001: Cloud-model based Bayesian techniques for precipitation profile retrieval from TRMM microwave sensors. Proc. ECMWF/EuroTRMM Workshop on Assimilation of Clouds and Precipitation, ECMWF, Reading, UK, 323–345.
674
A. Mugnai et al.
Neshyba, S. P., T. C. Grenfell, and S. G. Warren, 2003: Representation of a nonspherical ice particle by a collection of independent spheres for scattering and absorption of radiation: 2. Hexagonal columns and plates. J. Geophys. Res., 108, 4448–4465. Roberti, L., J. Haferman, and C. Kummerow, 1994: Microwave radiative transfer through horizontally inhomogeneous precipitating clouds. J. Geophys. Res., 99, 16707–16718. Skofronick-Jackson, G., M.-J. Kim, J. A. Weinman, and D.-E. Chang, 2004: A physical model to determine snowfall over land by microwave radiometry. IEEE Trans. Geosci Remote Sens., 42, 1047– 1058. Smith, E. A., G. Asrar, Y. Furuhama, A. Ginati, A. Mugnai, K. Nakamura, R. Adler, M.-D. Chou, M. Desbois, J. Durning, J. Entin, F. Einaudi, R. Ferraro, R. Guzzi, P. Houser, P. Hwang, T. Iguchi, P. Joe, R. Kakar, J. Kaye, M. Kojima, C. Kummerow, K.-S. Kuo, D. Lettenmaier, V. Levizzani, N. Lu, A. Mehta, C. Morales, P. Morel, T. Nakazawa, S. Neeck, K. Okamoto, R. Oki, G. Raju, J. M. Shepherd, J. Simpson, B.-J. Sohn, E. F. Stocker, W.-K. Tao, J. Testud, G. Tripoli, E. Wood, S. Yang, and W. Zhang, 2007: International Global Precipitation Measurement (GPM) program and mission: an overview. In: Measuring precipitation from space. EURAINSAT and the future. V. Levizzani, P. Bauer, and J. F. Turk, eds., Springer, 611–654. Tassa, A., S. Di Michele, A. Mugnai, F. S. Marzano, and J. P. V. Poiares Baptista, 2003: Cloud-model based Bayesian technique for precipitation profile retrieval from TRMM Microwave Imager. Radio Sci., 38, 8074–8086. Tassa, A., S. Di Michele, A. Mugnai, F. S. Marzano, P. Bauer, and J. P. V. Poiares Baptista, 2004: Evaluation of cloud-radiation modeling errors for space-borne passive microwave retrieval of precipitation. IEEE Trans. Geosci. Remote Sens., submitted. Tripoli, G. J., 1992: A nonhydrostatic model designed to simulate scale interaction. Mon. Wea. Rev., 120, 1342–1359. Wiscombe, W. J., 1980: Improved Mie scattering algorithms. Appl. Opt., 19, 1505–1509. Zwally, H. J., 1989: Growth of Greenland ice sheet: Interpretation. Science, 246, 1589–1591.
50 OBSERVING RAIN BY MILLIMETRE– SUBMILLIMETRE WAVE SOUNDING FROM GEOSTATIONARY ORBIT Bizzarro Bizzarri 1, Albin J. Gasiewski 2, and David H. Staelin 3 1
CNR Istituto di Scienze dell’Atmosfera e del Clima (ISAC), Roma, Italy NOAA Environmental Technology Laboratory (ETL), Boulder, CO, USA 3 MIT Research Laboratory of Electronics (RLE), Cambridge, MA, USA 2
Abstract
A satellite sensor for estimating precipitation from sensors onboard geostationary orbits is described. The system relies upon the use of higher frequencies, in the submillimetre wavelength range, where antenna diameters can be more practical The principle is illustrated using temperature soundings at two distinct oxygen bands separated by approximately one octave. A proposed instrument is described, the Geostationary Observatory for Microwave Atmospheric Sounding (GOMAS), based on a 3-m diameter mechanically scanned antenna.
Keywords
Precipitation, geostationary, millimetre, sounding, GOMAS, GEM.
1
INTRODUCTION
Precipitation is a rapidly evolving parameter with a space-time structure typical of a fractal field; therefore, the sampling problem is particularly stringent. Specifically critical applications requiring rapid sampling include nowcasting, since fractal fields are difficult to extrapolate (Fig. 1), and hydrology, which requires accumulated precipitation data over several time intervals, that can be accurately computed only if actual measurements are available, since for a fractal field even interpolation is difficult. It is evident that geostationary satellites are the only suitable platform for sub-hourly global observations. At the local scale, rain gauges provide suitable temporal sampling, but are not representative of area-averaged rain rates and suffer reduced accuracy in high winds, snow and very light rain. Rain radars provide frequent and representative data, but are not suitable for mountainous regions or observation more than ~100 km offshore. 675 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 675–692. © 2007 Springer.
676
B. Bizzarri et al.
Precipitation rate estimation based on visible/infrared (VIS/IR) imagery from geostationary satellites (GEO) is a long-standing application. However, this technique is at best indirect, since visible and infrared radiance fields provide information only on cloud top reflectances, temperatures, cloud morphology, and dynamics. Much external information is required when processing the data, and the results are applicable only under observing conditions consistent with the retrieval model and the external information provided. The EC-supported EURAINSAT project (Levizzani et al. 2001), centred on the use of the newly available VIS/IR images from the Meteosat Second Generation SEVIRI, has explored and assessed the performance capabilities of these techniques.
Figure 1. AMSU-B imagery from NOAA-15 and NOAA-16 showing a rapidly evolving east-coast snowstorm hitting New England on March 5, 2001, around 23 UTC. These three images (near 183 ± 7 GHz) reveal no significant snowfall only ~4.5 h before nor ~8 h after the event. The event extended over nearly 1,000 km and was over 50 km across. Despite its enormous extent, the major features of this storm evolved on a timescale of 1–2 h. (From Staelin 2001.)
More direct precipitation information is provided by microwave (MW) sensors that penetrate the cloud interior where the signal is controlled by emission and scattering (from ice), and by polarisation and depolarisation effects (over the sea). In addition, if the MW radiation is actively generated by the instrument (radar), then measurement of the delay and intensity of the backscattered radiation provides the precipitation vertical profile and drop size distribution. However, these techniques have so far been applied successfully only from low-Earth orbiting satellites (LEO) since diffraction
Observing Rain by Millimetre-Submillimetre Wave Sounding
677
in the applicable range of frequencies (~ 6–90 GHz) strongly limits the achievable resolution. For example, for 10-km resolution at nadir, the antenna diameter would be 15 m at 90 GHz, or 35 m at 37 GHz and 70 m at 19 GHz. A second problem is that the viewing geometry from GEO precludes exploitation of differential polarisation. In the absence of MW measurements from GEO, techniques have been developed (Turk et al. 2000) for exporting “calibration” from the infrequent microwave LEO images to the frequent VIS/IR GEO images or, rather, using GEO images to interpolate between LEO images. The accuracy of this methodology in “extrapolating” precipitating cloud observation for the purpose of nowcasting, or “interpolating” between MW observations for accumulated precipitation estimates needs to be assessed, and is expected to be strongly dependent on the precipitating cloud type. A promising attempt to develop MW sensors in GEO orbit relies on the use of higher frequencies, in the submillimetre wavelength range, where antenna diameters can be more practical (Gasiewski 1992). The use of submillimetre frequencies, particularly in atmospheric absorption bands, reduces the sensitivity to polarising surfaces, thus, the new principle is equally applicable over both water and land. Indeed, rapidly evolving precipitation is more frequent and more important from an application viewpoint over land. Figure 2 illustrates the electromagnetic absorption spectrum of the atmosphere between ~2 and 1,000 GHz, i.e., in the centimetre, millimetre, and submillimetre ranges. As can be seen, there are sufficient O2 and H2O absorption bands at millimetre and submillimetre wavelengths, including the AMSU 50–57 and 183 GHz bands, that enable use of reasonably sized antennas for GEO imaging of precipitation. The higher-frequency O2 bands are centred around 118 and 425 GHz, while those for H2O are around 325 and (preferred) 380 GHz. Table 1 illustrates the resolution achievable from GEO at these frequencies for several antenna diameters. It is observed that an antenna diameter of 3 m provides ~10 km resolution at the 425 GHz O2 band, and that at lower frequencies the resolution is comparable to AMSU, considering that the AMSU values of 48 km (at 54 GHz) and 16 km (at 183 GHz) refer only to the sub-satellite-point (s.s.p.) and degrade to over 100 km and 30 km at the edges of the swath (in the GEO image projection, the resolution degrades more slowly away from the s.s.p.).
2
THE MEASUREMENT PRINCIPLE
Using absorption bands to retrieve precipitation is an experimental albeit well-justified application of AMSU-A/B data. The principal advantages of absorption bands relative to atmospheric windows are the monotonically decreasing ability to detect rain intensity by altitude probing. Figure 3 shows
B. Bizzarri et al.
678
Figure 2. Atmspheric spectrum in the centimetre, millimetre, and submillimetre wavelength ranges (from Klein and Gasiewski 2000).
Table 1. Resolution (at s.s.p.) versus frequency and antenna size for GEO imaging (from Bizzarri et al. 2002).
Antenna Ø 54 GHz
118 GHz
183 GHz
380 GHz
425 GHz
1m 2m 3m 4m
112 km 56 km 37 km 28 km
73 km 36 km 24 km 18 km
35 km 18 km 12 km 8.8 km
31 km 16 km 10 km 7.8 km
242 km 121 km 81 km 60 km
an example of frontal precipitation observed over land using AMSU-A/B 54and 183-GHz bands compared with the NEXRAD radar network. The principle is further illustrated using temperature soundings at two distinct oxygen bands separated by approximately one octave. Figure 4 shows experimental data collected by the airborne NPOESS Aircraft Sounder Testbed – Microwave (NAST-M) that includes channels in the O 2 bands around 54 and 118 GHz. In the top figure the ratio between temperature profiles obtained independently from the 118 and the 54 GHz bands is reported as the aircraft travels. If there is no precipitation the ratio of the two temperature profiles is unity throughout the entire vertical range. When precipitation is present the ratio becomes less than unity below the altitude of the precipitation cell due to higher attenuation at 118 GHz than at 54 GHz. The signature of precipitation is the result of the similar “clear-air
Observing Rain by Millimetre-Submillimetre Wave Sounding
679
Figure 3. Precipitation images from a cold front on 7 October, 1998. Left: NEXRAD map smoothed to 50 km resolution; right: NOAA/AMSU map obtained using a neural net retrieval technique implying the use of both AMSU-A and AMSU-B (from Staelin and Chen 2000).
weighting functions” of channels in each of the two bands and the difference in ice scattering characteristics for the two wavelength regions (Gasiewski and Staelin 1990; Gasiewski 1992). The bottom part of Fig. 4 reports the precipitation profile simultaneously recorded by a Doppler radar (EDOP) onboard the aircraft. The horizontal map of Fig. 3 and the vertical cross section of Fig. 4 suggest that, if geostationary MW soundings are obtained in several frequency bands at, say, 15-min intervals, meteorologists would have a proxy rain radar operating over continental fields of view, and particularly over oceans and mountainous terrain.
Figure 4. Comparison between the 118/54 GHz profile ratio from the NAST-M microwave radiometer on the NASA ER-2 aircraft and simultaneous EDOP Doppler radar reflectivity observation. Hurricane Bonnie at 1700 GMT on August 26, 1998 (from Tsou et al. 2001).
680
B. Bizzarri et al.
Figure 5. Stripmap images of convective precipitation cells over ocean obtained using the MIR multi-channel airborne radiometer. The scenes are ~40 km (width) × ~200 km (length) (from Gasiewski et al. 1994).
While conventional AMSU-A/B bands can provide precipitation sensitivity, the use of higher-frequency microwave bands (380 and 425 GHz) is a practical necessity from geostationary orbit because only at submillimetre wavelengths can the resolution from GEO approach the natural spatial variability of rain (~10 km). Figure 5 shows experimental data from the MIR airborne radiometer equipped with channels in both atmospheric windows and in the 183-GHz and 325-GHz water vapour bands (the 325 GHz band is similar to that at 380 GHz). See, for instance, the increasing cloud area moving from the peak 183 ± 1 GHz to 183 ± 3 GHz and then 183 ± 7 GHz (nearly window); or from 325 ± 1 GHz to 325 ± 3 GHz and then 325 ± 9 GHz (nearly window); and moving across windows from 90 GHz to 150 GHz, 183 ± 7 GHz, 220 GHz and 325 ± 9 GHz. It is noted that many precipitation cells are missed at the 90 GHz frequency due to the bimodal response of this channel to maritime rain: at low rain rates the signature over water is warmed, but at higher rates the signature is cooled by ice scattering in the cell top. As for the 425 GHz band, experimental data are available from an upgraded version of NAST-M, that now includes the 183 and 425 GHz bands in addition to the 54 and 118 GHz bands of Fig. 4. Figure 6 shows pictures taken in these four bands. In each case, over the four octaves
Observing Rain by Millimetre-Submillimetre Wave Sounding
681
covered by these images the raincells are observed with increasing sensitivity, but remain clearly delineated while underneath cirrus anvil clouds. The 52.6-GHz images clearly respond strongly only to the narrow convective cores of these cells wherein hydrometeors are sufficiently large and/or frozen to have strong absorption and/or scattering signatures. The apparent width of these cores is ~9 km. The apparent diameters of cells observed using 118 GHz are markedly larger because Rayleigh scattering from both ice and liquid is inversely proportional to the fourth power of wavelength, and thus stronger by a factor of ~26. This trend continues as the frequency progresses to 183 and 425 GHz, for which these cell diameters approach ~16–17 km. A related trend continues for the minimum brightness temperatures at each of the cell tops. In order of increasing frequency, these minima are approximately 6, 18, 20–30 and 24–40 K below their baselines in all cases (note that these are “temperature perturbations” in respect of the clear-air brightness temperature).
Figure 6. Images of four convective cells (A, B, C and D) as observed at four different frequencies with weighting functions peaking, in clear-air conditions, at approximately the same altitude (∼3 km). NAST-M flight on North Pacific on 14 March 2003 (from Leslie and Staelin 2004).
682
B. Bizzarri et al.
Figure 7. Nonlinear KL decomposition of three-channel brightness imagery over the maritime precipitation cells of Fig. 5 (on the left). The plots in the centre illustrate the eigenmodes and KL coefficient dependencies for cell top brightness temperatures. The nonlinear decomposition shows that at least 2 independent observable degrees of freedom are obtained freedom are obtained using 150, 220 and 325 ± 9 GHz data (courtesy of A.J. Gasiewski).
The relevance of the higher frequency bands is not only in their improved spatial resolution, but also in the additional information that they provide regarding precipitation. Figure 7 shows Karhunen–Loeve (KL) (empirical orthogonal) coefficient maps of maritime raincells obtained using three window channels from Fig. 5 (150, 220, and 325±9 GHz). The KL coefficients are nonlinearly mapped to make the final coefficients statistically independent, thus imposing a stronger condition of independence than that of only decorrelation. The resulting number of independent channels of information is seen to be at least two, as evident from the unique morphology of the cells observed in the k1 and k2 maps. This conservative estimate of the number of observable degrees of freedom suggests that unique information on raincell dynamics will be provided by a broad set of submillimetre spectral observations. For example, the use of k2 would facilitate separation of convective cores and cirrus anvil regions, thus improving rain rate estimates. Although the higher frequency bands provide new information and are necessary because of the achievable resolution, their relationship with precipitation is more indirect. Since the H2O and N2 continuum (see Fig. 2) are so strong that only the middle-high troposphere is sensed, thus cloud ice and its scattering properties provide the dominant effects. The response of the highest frequencies to absorption and scattering has been studied by
Observing Rain by Millimetre-Submillimetre Wave Sounding
683
means of numerical simulation. Figure 8 shows the results of a study using MM5 simulations of Hurricane Bonnie near landfall with 6-km grid spacing for the most transparent channel of the 425 GHz band.
Figure 88.. MM5 and discrete ordinate Jacobian simulations at 424.763 ± 4 GHz of Hurricane Bonnie at landfall (courtesy of A.J. Gasiewski) (see also color plate 19).
The discrete ordinate Jacobians for the scattering coefficient (∂TB/∂as), absorption coefficient (∂TB/∂aA) and thermodynamic temperature (∂TB/∂T), illustrate probing depths of submillimetre channels of up to several kilometres into heavy precipitation, along with a seamless transition into a clear-air weighting function at the periphery of the system. Of particular interest is the bimodal response to absorption by precipitation. In regions of little scattering the derivative of brightness with respect to the absorption coefficient is negative – as would be expected in clear-air for increases in water vapour or liquid cloud content. Over heavy scattering, however, the derivative becomes positive as a result of a reduction in cell-top reflectivity caused by an increased probability of photon absorption. It is important to note that the quantitative nature and computational speed of these Jacobians make the data suitable for radiance assimilation in NWP models (Voronovich et al. 2004). The rapid updates provided by a geostationary imager further provide the potential for “precipitation locking” of a NWP model, much as a phase-locked loop is phase-synchronised with an incoming signal.
684
B. Bizzarri et al.
It is noted that whereas the 54, 118, and 183 GHz bands have channels that penetrate the lower troposphere down to surface, the 380 and 425 GHz bands are basically insensitive below 2–5 km in typical mid-latitude and tropical atmospheres. However, the connection between cell-top hydrometeor dynamics and surface rain rate has been statistically proven, and use of a precipitation locked NWP model will only enhance this connection, and thus improve the low-level rain rate retrieval accuracy. It is concluded that, notwithstanding the need for further intensive modelling of the relationships between cloud ice and precipitation at submillimetre wavelengths, the information content of the identified set of absorption bands (54, 118, 183, 380, and 425 GHz) in terms of potential precipitation retrieval is substantial. As mentioned, the retrieval of precipitation by absorption bands passes through the retrieval of temperature and humidity profile. Temperature and humidity profiling is not the primary objective of MW in geostationary orbit; nonetheless, the expected results are rather interesting. Figure 9 shows simulated retrieval errors for temperature and humidity. It can be observed that, for temperature, 118 + 425 GHz channels as stand-alone are rather good for the Upper Troposphere and Lower Stratosphere (UT/LS). When added to an IR sounder of the class of IASI or AIRS (“GAIRS”) the performance is as good as for the reference AIRS + AMSU/A + HSB of EOS/Aqua, except in the lower troposphere. The 54 GHz band therefore continues to be necessary. For humidity, the 380 GHz band (aided by 425) performs better than the 183 GHz band (aided by 118) in the UT/LS, but worse in the medium and lower troposphere. The full set of 118, 183, 380, and 425 bands is seen to be optimal. The accuracy of the water vapour profile retrieved in the UT/LS is better than 10%, very valuable for climatology. This level of accuracy in the
Figure 9. Retrieval errors for bands 118, 183, 380, and 425 GHz, stand-alone or associated with an IR sounder of the IASI or AIRS class (“GAIRS”). Temperature is shown on the left hand, and humidity on the right. On the left hand, GAIRS + 118 + 425 is compared to GAIRS alone and to 118 + 425 alone. For reference, the EOS/Aqua AIRS + AMSU/A + AMSU/B complex also is shown (from Blackwell and Staelin 1996).
Observing Rain by Millimetre-Submillimetre Wave Sounding
685
UT/LS region is difficult to achieve using any other technique including Thermal IR spectroscopy (e.g., by IASI- and AIRS-like instruments): only Far IR spectroscopy and lidar could be competitive, but these techniques are not suitable for image scanning and anyway impossible from GEO.
3
PROPOSED MISSIONS
The concept of MW sounding from geostationary orbit dates from 1978 (Staelin and Rosenkranz ed. 1978). In Europe, MW sounding was placed in 1984 as a requirement for the Meteosat Second Generation, and led to an industrial study presented in 1988 after a definition study dated 1985 (Chedin et al. 1985). Those early initiatives were premature because of the state-of-the-art technology of both antennas and radiometers at the time. Consideration of the use of submillimetre wave frequencies enabled a new approach to the antenna problem (Gasiewski 1992). Accordingly, the concept of millimetre–submillimetre sounding from geostationary orbit has been reconsidered in both the USA and Europe. In the USA, a “Geosynchronous Microwave Sounder Working Group” carried out a study for NOAA/NESDIS delivered in 1997 (Staelin et al. 1997) and gave rise to the proposal for GEM (Geostationary Microwave Observatory) in 1998 (Staelin et al. 1998). In Europe, an analysis of the scientific and technological scenario was prepared for EUMETSAT in 2000 (Bizzarri 2000) and finally, in January 2002, a proposal was submitted to ESA in the framework of the Earth Explorer Opportunity Missions for Geostationary Observatory for Microwave Atmospheric Sounding (GOMAS). The GOMAS proposal includes GEM heritage and was submitted by a group of 40 European and US proponents (Bizzarri et al. 2002). The key difference between GEM and GOMAS is that due to the average high latitude of Europe a better resolution at the equatorial s.s.p. is required. The reference resolution at 425 GHz is thus 10 km for GOMAS, which implies (according to Table 1) the need for a 3-m diameter antenna compared to the 2-m diameter antenna and 15 km resolution for GEM. A second reason to limit the antenna size for GEM is to maintain platform compatibility with GOES-R, whereas GOMAS is considered to be a dedicated mission. Whereas the resolution requirement drives the antenna diameter of GEM/GOMAS, the radiometric characteristics are driven by the fact that, although temperature and humidity profiling is not the primary objective, the principle exploited to measure precipitation benefits from full profiling over a broad range of bands. Profiles from different bands are impacted differently by liquid and ice hydrometeors of different drop size and shape so that a number of degrees of observational freedom is required for simultaneous retrieval of all these quantities. One consequence is that the number of channels in each band (from 6 to 11), their bandwidth (<1%) and the radiometric accuracy (SNR > 100) must be sufficient for accurate
B. Bizzarri et al.
686
profiling. Table 2 lists the radiometric channels currently foreseen for GEM and GOMAS. Figure 10 shows the clear-air Incremental Weighting Functions (IWFs) for the listed channels. Table 2. Channels selected for GEM and GOMAS (from Bizzarri et al. 2002). 54
GHz
118 GHz
ν (GHz) 56.325 56.215 56.025 55.520 54.950 54.400 53.845 53.290 52.825 51.760 50.300
∆ν (MHz) 50 50 250 180 300 220 190 360 300 400 180
ν (GHz) 118.750±0.018 118.750±0.035 118.750±0.080 118.750±0.200 118.750±0.400 118.750±0.700 118.750±1.100 118.750±1.500 118.750±2.100 118.750±3.000 118.750±5.000
183 GHz ∆ν (MHz) 6 12 20 100 200 400 400 400 800 1000 2000
380 GHz
ν (GHz) 183.310±0.300 183.310±0.900 183.310±1.650 183.310±3.000 183.310±5.000 183.310±7.000 183.310±17.000
∆ν (MHz) 300 500 700 1000 2000 2000 4000
425 GHz
ν (GHz) 380.197±0.045 380.197±0.400 380.197±1.500 380.197±4.000 380.197±9.000 380.197±18.000 340.0 optional/auxiliary
∆ν (MHz) 30 200 500 900 2000 2000 8000
118.7503 GHz
ν (GHz) 424.763±0.030 424.763±0.070 424.763±0.150 424.763±0.300 424.763±0.600 424.763±1.000 424.763±1.500 424.763±4.000
424.7631 GHz
40
40
15−21
25−35 30
29−41 70−90
20
150−250 300−500 500−900 900−1300 ← 1300−1700 ← 1700−2500 ← 2500−3500
10 4000− 6000↓
0
0.05
0.1
Altitude (km)
Altitude (km)
30
0 −0.05
60−80 120−180
20
250−350 500−700 800−1200 1200−1800
10
3500−4500
0
0.15
0
−1
183.3101 GHz
0.15
0.2
0.25
380.1974 GHz & 340 GHz 20
150−450 650−1150 1300−2000 2500−3500 4000−6000
30−60 300−500
15 Altitude (km)
Altitude (km)
0.1
Temperature IWF (km )
20
15
0.05
−1
Temperature IWF (km )
Figure 10. Clear-air incremental weighting functions (IWFs) for temperature and humidity in the bands 118, 183, 380 and 425 GHz. The numbers attached to the curves indicate the offset (in MHz) from the associated line centre (from Klein and Gasiewski 2000). The IWF's for the more familiar 54 GHz band also are plotted (courtesy of A.J. Gasiewski).
∆ν (MHz) 10 20 60 100 200 400 600 1000
10 ← 6000−8000
5
1250−1750
10 3550−4450 8000−10000
5 17000−19000↑
15000−19000
0 −2
0
2
4 −1
Water Vapor IWF (K km )
6
0 −2
336−344 GHz
−1.5
−1
−0.5
0
0.5
−1
Water Vapor IWF (K km )
The radiometric accuracy required for sounding (SNR > 100) and the narrow channel bandwidths have to be confronted with the required observing cycle, assumed to be 15 min. Preliminary studies have shown that this is possible
Observing Rain by Millimetre-Submillimetre Wave Sounding
687
using current technology if the scanned area is limited to about 1/12 of the Earth’s visible disk, e.g., the sector shown in Figure 11 or an equivalent region. Although sampling is performed at 10 km s.s.p. for all channels and bands, in order to recover the required SNR for all altitudes some pixels will need to be averaged. Anticipated GOMAS products are: – temperature profiles, all-weather and inside clouds, resolution ~30 km; – humidity profiles, all-weather and inside clouds, resolution ∼20 km; – cloud liquid/ice water total column or gross profile, resolution ∼20 km; – precipitation rate, resolution ∼10 km; with updates each 15 min, over ∼1/12 of the disk covering sea and land.
Figure 11. Earth’s disk observed by Meteosat and reference coverage in 15 min from GOMAS (from Bizzarri et al. 2002).
4
TECHNICAL CONCEPT
The GOMAS instrument concept is based on a 3-m diameter mechanically scanned antenna. The reflector surface has a quiescent accuracy of ∼10 µm rms. Thermal and inertial deformations are monitored by a series of sensors on the antenna border and actively compensated using a tilting and deforming subreflector, which also provides for limited fast raster scanning. Gross movements of the antenna are performed using elevation and azimuth motors with momentum compensation, although the possibility of using the
B. Bizzarri et al.
688
satellite attitude control system in combination is being studied. A single virtual focal point with quasi-optical feed multiplexer is baselined so as to provide hardware co-alignment of all feedhorns for the five bands. One option employing a feed cluster to simplify the multiplexer design is still being studied. The baseline receiver set includes five individual spectrometers with uncooled (ambient temperature) components, for a total of ~43 channels (see Table 2 for detailed frequencies and passband characteristics). A concept drawing of the instrument is shown in Fig. 12. 3” Thick Composite Reflector Nodding / Morphing Subreflector Space Calibration Tube
Receiver Package
50-430 GHz Feeds
Thin Struts
Elevation Motor & Compensator
Azimuth Motor & Compensator Backup Structure
Figure 12. Concept view of the GOMAS instrument with 3-m antenna. Scaled from the GEM (2-m antenna) concept (from Bizzarri et al. 2002).
Mass, size, and power specifications are: • Antenna: ∅ = 3 m, 40 kg, 40 W; • Radiometer: 30 × 50 × 50 cm3, 67 kg, 95 W; • Total: 107 kg, 135 W, 115 kbps. Whereas GEM is designed to be compatible for deployment on a multipurpose satellite of the GOES-R series, the GOMAS plan is being studied for implementation by a dedicated satellite. This configuration would provide the necessary programmatic flexibility for a demonstration mission. A dedicated mission would favour decoupling programmatic aspects from
Observing Rain by Millimetre-Submillimetre Wave Sounding
689
the Meteosat Third Generation development and, in case the satellite is implemented in cooperation with the USA, would allow drifting the satellite in orbit to alternatively serve Europe and America, as well as possible other interested areas. Figure 13 shows a concept view of the GOMAS satellite as currently envisioned. Its specifications are: • Mass: 860 kg (“dry”: 430 kg). • Electrical power: 500 W. • Volume: 3.0 × 3.0 ×3.0 m3 (stowed). • Data rate: 128 kbps (S-band, compatible with the existing low-cost stations deployed to receive low-rate Meteosat Second Generation data).
Figure 13. Concept view of the GOMAS satellite (from Bizzarri et al. 2002).
B. Bizzarri et al.
690
5
CONCLUSION
The case for frequent precipitation observation from geostationary orbit is well known, and economically solvable only by using a geostationary platform. The driving applications include nowcasting (continuous monitoring of actual weather and extrapolation to a few hours) and hydrology (computing cumulated precipitation over short time intervals). Visible and IR imagery provide frequent enough observation, but only qualitative inference. MW imagery from low-orbiting satellites provides accurate measurement, but only infrequently (at most 3 h even if a full GPM constellation is deployed). Data fusion between VIS/IR from GEO and MW from LEO is becoming a practice, but the accuracy is not as high as required. MW imagery in GEO cannot be implemented by simply exporting the LEO principle based on observation in atmospheric windows, since this would imply antennas of unaffordable size. The proposed principle uses absorption bands instead of windows. By measuring temperature and humidity profiles over a range of frequencies spanning over nearly one order-of-magnitude (50–430 GHz), the signatures will differ as a consequence of monotonically increasing sensitivity to absorption and scattering from liquid water drops or ice crystals, as well as precipitation. Although intensive modelling effort is still necessary to establish the relationships between ice and precipitation, there is enough experimental and theoretical evidence to recognise that unique and valuable information exists in millimetre–submillimetre imagery. The use of absorption bands at millimetre and submillimetre wave frequencies enables a major reduction in the antenna size to the extent that a payload to be flown in GEO is economically viable. Preliminary system studies – first in the USA (GEM) and later in Europe (GOMAS) – lead to the conclusion that a demonstration mission is possible using existing antenna and radiometer technology, provided that the observation is performed over a limited area (some 1/12 of the Earth’s disk in 15 min). The system would enable simultaneous retrieval of nearly all-weather temperature and humidity profiles, columnar content or gross profile of cloud ice or liquid water, and ultimately precipitation over both sea and land, of convective, frontal and stratiform types. Though nowcasting and hydrology are the driving applications, direct assimilation of brightness temperatures in NWP could be beneficial as well. Both technological and scientific studies are being performed in both the USA (NOAA and NASA) and Europe (EUMETSAT and ESA). Because of the increasing interest of the operational hydrological community for using satellites, the request for frequent observation of precipitation from GEO will strongly increase in the near future. The implementation of millimetre– submillimetre sounding from GEO would be an appropriate response.
Observing Rain by Millimetre-Submillimetre Wave Sounding
6
691
REFERENCES
Bizzarri, B., 2000: MW/Sub-mm sounding from geostationary orbit. Report to EUMETSAT Science W.G., EUM/STG/SWG/9/00/DOC/11, 11 pp. Bizzarri, B. et al. (the 40 GOMAS proponents), 2002: Requirements and perspectives for MW/sub-mm sounding from geostationary satellite. EUMETSAT Meteorological Satellite Conference, Dublin, Ireland, 2–6 September 2002, 97–105. Blackwell, W. J. and D. H. Staelin, 1996: Comparative performance analyses of passive microwave systems for tropospheric sounding of temperature and water vapor profiles. GOES-8 and Beyond, Denver Co., August 7–9 1996 - SPIE Proceedings Series, 2812, 472–478. Chedin, A., D. Pick, and R. Rizzi, 1985: Definition study and impact analysis of a microwave radiometer on a geostationary spacecraft. ESA Report dated March 1985, pp.58. Gasiewski, A. J. and D. H. Staelin, 1990: Numerical analysis of passive microwave O2 observations over precipitation. Radio Science, 25, 217–235. Gasiewski, A. J., 1992: Numerical sensitivity analysis of passive EHF and SMMW channels to tropospheric water vapor, clouds and precipitation. IEEE Trans. Geosci. Remote Sensing, 30, 859–870. Gasiewski, A. J., D. M. Jackson, J. R. Wang, P. E. Racette, and D. S. Zacharias, 1994: Airborne imaging of tropospheric emission at millimeter and submillimeter wavelengths. Proc. Int. Geoscience and Remote Sensing Symposium, Pasadena, Ca., August 8–12, 1994, 663–665. Klein, M. and A. J. Gasiewski, 2000: The sensitivity of millimeter and sub-millimeter frequencies to atmospheric temperature and water vapor variations. J. Geophys. Res., 13, 17481–17511. Leslie, R. V. and D. Staelin, 2004: NPOESS aircraft sounder testbed – M: Observations of clouds and precipitation at 54-, 118-, 183 and 425-GHz. IEEE Trans. Geosci. Remote Sensing, 42, 2240–2247. Levizzani, V., P. Bauer, A. Buzzi, S. Davolio, D.E. Hinsman, C. Kidd, F.S. Marzano, F. Meneguzzo, A. Mugnai, J.P.V. Poiares Baptista, F. Porcù, F. Prodi, J.F.W. Purdom, D. Rosenfeld, J. Schmetz, E.A. Smith, F. Tampieri, F.J. Turk, and G.A. Vicente, 2001: EURAINSAT – Looking into the future of satellite rainfall estimations. Proc. 2001 EUMETSAT Meteorological Satellite Data Users’ Conf., EUMETSAT, Antalya, 1–5 October 2001, 375–384. Staelin, D. H., 2001: Microwave Sensors for LEO and GEO. Presentation at the CGMS/WMO Workshop on the Long-Term Future of the Basic Sounding and Imaging Missions, Geneva, April 23–24. Staelin, D. H. and F. W. Chen, 2000: Precipitation observation near 54 and 183 GHz using the NOAA 15 satellite. IEEE Trans. Geoscience Remote Sensing, 38, 2322–2332. Staelin, D. H. and P. W. Rosenkranz, editors, 1978: “Applications Review Panel: High resolution passive microwave satellites”. Report for NASA Contract NAS5-23677, MIT Research Laboratory of Electronics. Staelin, D. H., J. P. Kerekes, and F. J. Solman III, 1997: Final report of the geosynchronous microwave sounder working group. Prepared for the NOAA/NESDIS GOES Program Office, MIT, Lexington Mass., pp.51. Staelin, D. H., A. J. Gasiewski, J. P. Kerekes, M. W. Shields, and F. J. Solman III, 1998: Concept proposal for a Geostationary Microwave (GEM) Observatory. Prepared for the NASA/NOAA Advanced Geostationary Sensor (AGS) Program, MIT, Lexington, MA, 23 pp. Tsou, J. J., W. L. Smith, P. W. Rosenkranz, G. M. Heymsfiels, W. J. Blackwell, and M. J. Schwartz, 2001: Precipitation study using millimetre wave temperature sounding channels. Specialist Meeting on Microwave Remote Sensing, Boulder, CO, November 2001.
692
B. Bizzarri et al.
Turk, J. F., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000: Meteorological applications of precipitation estimation from combined SSM/I, TRMM and geostationary satellite data. In: Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia, eds., VSP, Utrecht, The Netherlands, 353–363. Voronovich, A., A. J. Gasiewski, and B. L. Weber, 2004: A fast multistream scattering-based Jacobian for microwave radiance assimilation. IEEE Trans. Geosci. Remote Sensing, 42, 1749–1761.
51 THE CGMS/WMO VIRTUAL LABORATORY FOR EDUCATION AND TRAINING IN SATELLITE MATTERS James F.W. Purdom1 and Donald E. Hinsman2 1
Cooperative Institute for Research in the Atmosphere, Colorado State University, Ft Collins, CO, USA 2 World Meteorological Organization, Geneva, Switzerland
Abstract
The Virtual Laboratory for Education and Training in Satellite Matters (VL) has been established to maximize the exploitation of meteorological satellite data across the globe. It is a collaborative effort joining four major operational meteorological satellite operators with six WMO “centers of excellence” in satellite meteorology. Those “centers of excellence” serve as the satellite-focused training resource for WMO members. The VL continues to evolve and has been used successfully in a number of WMO-sponsored training events since 2000.
Keywords
WMO, CGMS, satellite, training
1
INTRODUCTION
1.1 Pre-VL activity – CIRA contribution The Virtual Laboratory for Education and Training in Satellite Matters (VL) officially came into existence in 2000; however, its origins trace back to work done by the Cooperative Institute for Research in the Atmosphere (CIRA), at Colorado State University in 1994 and a project known as RAMSDIS. RAMSDIS stands for RAMM Advanced Meteorological Satellite Demonstration and Interpretation System. At that time, the RAMSDIS’ project goal was to disseminate real-time, high quality, digital GOES data to selected National Weather Service Forecast Offices via a powerful, low-cost,
693 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and theFuture, 693–704. © 2007 Springer.
694
J. W. F. Purdom and D. E. Hinsman
PC-based workstation for use in advanced satellite data display and analysis1 (Molenar et al. 2000). The project was designed to: •
• • •
provide online case study data and selected real time data (using the Internet for data distribution) from GOES for training United States National Weather Service (NWS) office staff to fully utilize digital data from NOAA’s new generation of GOES satellites; provide forecaster familiarization with the next generation GOES data sets, in preparation for NOAA’s Advanced Weather Interactive Processing System (AWIPS) deployment; determine future forecaster training requirements to ensure full utilization of the advanced GOES data sets; and provide a platform for forecaster evaluation of GOES imagery and products before the GOES data became operationally available via NOAAPORT.
By 1999, RAMSDIS workstations were being utilized in over half of the National Weather Service Forecast Offices and the project was hugely successful in meeting its goals. The concept soon expanded to providing both online case study data and near real time data to the WMO Regional Meteorological Training Centers (RMTCs) in Barbados and Costa Rica. As with the NWS effort, CIRA provided RAMSDIS specialized software and data analysis systems to the RMTCs. Using common software and hardware allowed work done on algorithm research at CIRA and other institutes to be used at RMTCs in Barbados and Costa Rica on real time digital satellite imagery. This cooperative arrangement benefited all involved and through the development of new products that are regionally relevant has resulted in marked improvements in training capability. Today RAMSDIS capability exists at each of the centers of excellence and satellite sponsoring agencies referred to in the abstract. In addition, research RAMSDIS workstations at CIRA are utilized to demonstrate new and experimental products, with realtime products from those workstations available on the Internet at a site known as RAMSDIS OnLine.
1
The workstation is based on the University of Wisconsin Space Science and Engineering Center McIDAS software, with automatic product ingest, display and analysis applications developed by the RAMM Team.
The CGMS/WMO Virtual Laboratory for Education
695
1.2 Genesis of the VL within WMO Recognizing the importance of a coordinated, worldwide approach to improving satellite data utilization, the WMO’s CBS OPAG IOS2 Expert Team on Satellite Systems Utilization and Products, ET-SSUP, initiated discussions at a meeting in Locarno,3 Switzerland, on how to attain that goal through the WMO strategy “train the trainer.” The next meeting of the ETSSUP, held in Melbourne,3 Australia, noted that satellite training institutions and their sponsoring satellite agencies must utilize modern technology to provide a range of training opportunities and materials to WMO Members. That meeting noted that a key ingredient of the VL would be to build strong links with science groups. Annex IV of the Melbourne meeting summarizes the ET’s discussion on the background, objectives, status, and guidelines for the VL. Later in Lannion,3 France, the ET-SSUP identified the need for two streams of learning skills (basic and specialist) and a virtual resource library within the VL. Subsequent to the Lannion meeting, the concept of the VL was brought before Coordination Group for Meteorological Satellites, CGMS, in October 2000. CGMS presented an action to the WMO to work with CGMS to initiate the establishment of a Focus Group on satellite data utilization and training4 within the VL Framework. A major function of the VL Focus Group was to help foster the VL to realize the challenges set forth by the WMO Executive Council Panel on Education and Training and in support of the WMO Strategy for Education and Training in Satellite Matters. The meeting of the Focus Group occurred during mid-May 2001, and may be thought of as the “official birth” of the VL. The Focus Group defined the various roles and responsibilities of participants, as well as the relationships between various components of the VL.
2
STRUCTURE AND FUNCTIONS OF THE VL
2.1 Guidelines Several tenets help guide the VL its activities and evolution: (1) satellite training institutions and their sponsoring satellite agencies must utilize modern technology to provide a range of training opportunities and materials to WMO Members; (2) a key ingredient is strong linkage with relevant
2
See http://www.wmo.ch/web/www/BAS/CBS-teams.html. http://www.wmo.ch/hinsman/Pubs_OPAG_IOS_ET.html. 4 Membership consist of the VL sponsors and collaborators. 3
696
J. W. F. Purdom and D. E. Hinsman
science groups; (3) there are two streams of learning skills (basic and specialist); (4) training requirements are developed as a part of the WMO’s interaction with its members; (5) the satellite operators and RMTCs are linked, with an agreed upon statement of objectives; and (6) a virtual resource library exists within the VL.
2.2 Relationships and structure Each of the major satellite operators sponsors one or more of the centers of excellence. NOAA/NESDIS sponsors the RMTCs in Bridgetown, Barbados and San Jose, Costa Rica. EUMETSAT sponsors the RMTCs in Niamey, Niger and Nairobi, Kenya. CMA/NSMC sponsor the RMTC located in Nanjing, China, and JMA sponsors the center of excellence located at the Australian Bureau of Meteorology training center in Melbourne, Australia. The six “centers of excellence” and four satellite operators are linked together through the Internet. All of the satellite operators have connectivity speeds in excess of T1, as do some of the centers of excellence, while a few of the centers of excellence only have 64 K connectivity. It should be noted that the goal is for all centers of excellence to have at least T1 capacity by 2005. The centers of excellence and satellite operators share a number of resources to help meet their two major strategic goals: To provide high quality and up-to-date training resources on current and future meteorological and other environmental satellite systems, data, products and applications. To enable the “centers of excellence” to facilitate and foster research as well as the development of socioeconomic applications at a local level by the NMHS through the provision of effective training and links to relevant science groups. At the core of the VL, shown schematically in Fig. 1, lies the Virtual Resource Library, VRL. VRLs are supported by each of the major satellite operators as well as a few of the centers of excellence. A goal is for each “center of excellence” to have a VRL by 2005. Although each of the VRLs has a different slightly focus, generally training tools, real time and retrospective satellite data and products, algorithms, tutorials and satellite imagery are freely available to all users through one or more of the VRL sites. For example, JMA has made SATAID available and NOAA/NESDIS has made RAMSDIS, RAMSDIS OnLine and VISITview available. SATAID, RAMSDIS OnLine and VISITview are important components of the VRL. VISITview is a source of online training materials based on case studies. It serves as a powerful platform for teletraining and collaboration during training events. VISITview software may be copied for local use for training others remotely and is freely available over the Internet. SATAID is a small application that runs on MS Windows computers allowing users to view and
The CGMS/WMO Virtual Laboratory for Education
697
interact with satellite, observational, and Numerical Weather Prediction data. It is very useful for analysis of imagery in conjunction with conventional meteorological fields and for interacting with those data fields. RAMSDIS OnLine provides access to multi-channel real-time gif format loops of GOES imagery and products. It is a valuable tool for reinforcement of training concepts when used with map discussions and analysis of real-time weather events. RAMSDIS OnLine also contains convenient help buttons for those unfamiliar with the characteristics of various image channels and their utilization. It is important to note that among the resources available to the VL is support from organized scientific groups: the International TOVS Working Group (ITWG), the International Winds Working Group (IWWG), and the International Precipitation Working Group (IPWG). Formal links between those science groups and the VL, while fruitful today, will become increasingly important as the VL evolves.
Virtual Laboratory Basic Skills
for
Virtual Laboratory for Specialist Skills
Virtual Resource Library
Documented competencies for the use of satellite data (WMO Pub. No. 258), biennial questionnaire
Satellite Operators
Science Community IWWG, ITWG, IPWG
Visualization and data manipulation tools (RAMSDIS, SATAID, VISITView, RAMSDIS
Available learning resources and guides (COMET, SATAID, EuroMET, printed material, CIRA, BMTC and other CAL) and Figure 1. Schematic illustration of the VL and its resource base.
J. W. F. Purdom and D. E. Hinsman
698
Welcome to the Virtual Laboratory for Satellite Training and Data Utilization The Virtual Laboratory for Satellite Training and Data Utilization (VL) has been established to maximize the exploitation of satellite data across the globe. It is a collaborative effort joining the major operational satellite operators across the globe with WMO “centers of excellence” in satellite meteorology. Those “centers of excellence” serve as the satellitefocused training resource for WMO Members. Click here to access a more detailed description of the VL. To access VL resources or different components of the VL click on the appropriate item below. To access VL resources or different components of the VL click on the appropriate item below. Virtual Lab Resources available from this site (CIRA) Satellite Imagery
Satellite Products
Tutorials
Training Tools
Software
Digital Satellite Imagery
Live Training Events
Online Courses and Quiz’s
Search
Centers of excellence resource sites and sponsors’ resource libraries Centers of excellence at five WMO Regional Meteorological Training Centers at San Jose, Costa Rica, Bridgetown, Barbados, Niamey, Niger, Nairobi, Kenya, and Nanjing, China, and Australian Bureau of Meteorology Training Center (ABOMTC) Resource libraries at CIRA, EUMETSAT, JMA, NSMC, and WMO Supporting Science Groups International TOVS Working Group (ITWG) International Winds Working Group (IWWG) International Precipitation Working Group (IPWG) Virtual Lab Sponsors USA (NESDIS) Europe (EUMETSAT) China (NSMC) Japan (JMA)
Figure 2. CIRA VL home page.
The CGMS/WMO Virtual Laboratory for Education
699
Entry into a VL through Internet opens a homepage similar to that of the NOAA/NESDIS supported CIRA site, Fig. 2. Once into a VL site users can access a variety of VRL reviewed information and resources (as mentioned above), or different components of the VL through moving a cursor to the appropriate item and then clicking on that resource link.
3
VL TRAINING EVENTS
Three major WMO training events have been conducted since the concept of the VL was introduced: Nanjing 2000, APSATS 2002, and Barbados 2003. Highlights are summarized in the subsections below.
3.1 Nanjing 2000 Although occurring while the VL was in its conceptual stage, the Nanjing training event marks the first time the VL model was utilized in a WMO sponsored training event. The Nanjing training event followed the guidelines for the WMO Strategy for Education and Training in Satellite Matters. The training event was a “first” in several respects – it was the first training event held in Nanjing as a “specialized centre of excellence” under the Strategy for Education and Training in Satellite Meteorology and it was the first event held that embraced the recently approved concept of the Virtual Laboratory for Education and Training in Satellite Meteorology. It focused on “training the trainers.” The Nanjing RMTC, located at the Nanjing Institute of Meteorology, NIM, provided a network of over twenty computers with greater than T1 online Internet capabilities to allow each participant to view the presentations. Most lectures were made utilizing Microsoft Powerpoint presentations or stand alone with RAMSDIS, and some lectures were made from CIRA over the Internet. All materials presented electronically were mastered onto two CD-ROMs and given to each participant at the end of the training seminar thus enabling them to return to their country and further train other members of their NMHS. In addition, JMA provided a CD for SATAID and EUMETSAT provided CDs with ASMET modules and for the CGMS Directory of Applications. At the end of each day, there was a map discussion covering the weather situation within the field of view of data received by NIM. The Nanjing event was exceptional and set a standard for future training events.
3.2 APSATS 2002 Asia Pacific Satellite Applications Training Seminar (APSATS) 2002 was conducted at the Australian Bureau of Meteorology Training Center, one of the six WMO “specialized centre of excellence” under the Strategy for Education and Training in Satellite Meteorology. The objective for this workshop was to provide structured learning opportunities for participants to increase their skills, knowledge, and attitudes in the use of a wide range of
700
J. W. F. Purdom and D. E. Hinsman
satellite data and products and thus contribute to an increase in the effective use of satellite data and products by National Meteorological and Hydrological Services in the Asia Pacific region. In conjunction with the WMO strategy, workshop participants were expected to train other staff on return to their workplace. To assist the participants and presenters before, during and after the workshop an APSATS 2002 Learning and Action Guide was created. The Learning and Action Guide has several purposes: it contains workshop information such as the program, contact details, and biographies of the participants and presenters, and information about the workshop facilities; it also contains suggestions on different ways for participants to maximize their learning during and after the workshop; outlines and information about each session, what the session goals are, major references and includes room for participants to add their own notes and things they wish to follow up on for that session; and, the final sections contain templates for them to enter in their overall action plans and the course evaluation sheets. Presenters provided session outlines and goals for their sessions. The workshop was a major training success, in large part due to the resources and applications of the VL. During APSATS, instructors were able to demonstrate the effective use of several of the Virtual Lab tools, primarily VISITview for distance education and SATAID for case studies, and specialized analysis tools for manipulating hyperspectral data. Lecturers also demonstrated the usefulness of the RAMSDIS OnLine website during the workshop with morning briefing sessions illustrating the use of remotely sensed data. During the workshop we collected some 20Gb of case studies and resources were made ready for CD duplication and subsequently were distributed to all of the participants and the centers of excellence. As part of the Research and Development Satellite Operators recent commitment to providing data to the World Weather Watch, NASA presented a series of lectures and workshops on the use of Moderate Resolution Imaging Spectroradiometer (MODIS) instrument data. Each major operational satellite operator supported the workshop by providing experts to speak on a variety of topics. Two lectures on the use of satellite data and products in the tropics were presented: the difference with those two sessions from others was that the presenter was in the USA (at CIRA) and used two Internet based applications to present the material. The VISITview application was used for the interactive graphics and text component and the voice segment was done using Yahoo messenger service. This was the first time that a distant lecturer has been used on a WMO or Bureau course. The same technologies were also used for the live global image discussion (also a first) on the Friday of the first week with remote presentations from the USA and Europe. The JMA SATAID application was one of the backbones of this workshop. For APSATS 2002 participants were asked to nominate a case study (times and dates and geographic area) to the organizers prior to the workshop. This data was then extracted from either the Bureau’s satellite and NWP archives or
The CGMS/WMO Virtual Laboratory for Education
701
JMA’s archives and converted to the SATAID format for the hands on sessions during the workshop. Most participants found the case studies to be very useful. Participants took home copies of their case studies and the SATAID application when they left. This data and the available CDs form the basis of their resource material to begin their training tasks. As noted in the final evaluations many participants thought this was one of the best, if not the best training course they had undertaken.
3.3 Barbados 2003 Barbados 2003 was sponsored by the Caribbean Institute for Meteorology and Hydrology (CIMH), one of WMO’s six “centers of excellence” for education and training for satellite meteorology. The VL achieved another major milestone in continuing its preiemince in meeting the WMO Space Program’s training objectives. The first day of the event marked the beginning of a long line of firsts, foremost being the complete removal of the language barrier, often common to regional training events, by a seminar simultaneously conducted for participants in Barbados as well as at the RMTC in Costa Rica (another first) through the use of VISITview. The Barbados participant received several hour long lectures from the RMTC in Costa Rica that utilized live Internet audio and VISITview. Since several of the Barbados participants had Spanish as their mother tongue, lectures alternated between English and Spanish. The VL, its training material, resources, and personnel were central to the success of the workshop. The participants benefited from presentations by world-renowned lecturers, both live and using the Internet and VISITview, that covered a variety of topics on satellite meteorology. In addition, instruction included how to use various VL tools: (1) the capabilities of the JMA software package SATAID and its applicability to the data streams available in RA III and IV; (2) an in-depth description of VISITview and how to establish the necessary links to take advantage of its true power; (3) use of RAMSDIS OnLine as a primary tool for daily map discussions; and (4) the use of RAMSDIS for fire detection and cloud type identification. As in APSATS 2002, a research space agency participated in the training event. ESA provided a lecture and materials relevant to its satellite missions and in particular ENVISAT. Another in the list of firsts – tropical storms Odette and Peter broke a 100-year record for two storms in the month of December and RAMSDIS Online was there to provide near real-time imagery to the participants and some extremely interesting map discussions. All the Barbados participants were trained in how to obtain the free VISITview software as well as how to establish similar dual location lectures. The participants were so buoyed by this new capability that they established a new Caribbean Focus Group to perpetuate and build a new and stronger dialogue amongst trainers and forecasters in the region, another first
702
J. W. F. Purdom and D. E. Hinsman
for the VL. As with the previous two VL training events, CIMH’s will be providing each participant with all the lectures on CD-ROM to enable each “trainer” to pass on the expertise gained in Barbados to colleagues in their own NMHS. Barbados 2003 was an excellent satellite data and products training course. It relied heavily on resources and training material from the VL, the crown jewel of the WMO Space Programs’ training activities, and demonstrated the potential of remote teaching using software such as VISITview.
3.4 Future training event guidelines Based on experience from the training events cited above, the VL Focus Group agreed to the following principles for the planning and running of VL training workshops: Planning should begin a year prior to the event. All the course information must be included in a Learning and Action guide that conforms with the VL template. This acts to focus the participants and lecturers on the purpose of the training activity and its relationship to the course objectives and other learning activities. Courses should include a wide range of learning activities. Those activities are designed to engage participants taking into account their culture, language and skill level. Typical activities will include traditional face to face content lectures and their accompanying consolidation workshops, online lectures, talks in nonformal sessions, group discussions, poster presentations, participant presentations on how they utilize satellite data and products, real-time imagery discussion, and one-on-one sessions with lecturers. The course must cover the three facets of learning: skills, knowledge, and attitude. Must provide resource material for participants to take back to their home institution to assist them in providing training at their NMHS. Should form a task group composed of the training participants, the appropriate satellite operator, and the “centers of excellence” to become a self-help team. Should conduct a six-month assessment after the training workshop and if necessary provide follow-up online training covering relevant workshop material. Should report back to the other “centers of excellence” on lessons learnt from running this training event to aid in the running of future training events.
The CGMS/WMO Virtual Laboratory for Education
4
703
SETTING PRIORITIES AND ASSESSING UTILIZATION FOR WMO TRAINING ACTIVITIES SUPPORTED BY THE VL
The establishment of a VL user tracking and feedback mechanism is recognized as an exceptionally important activity, and is addressed in part by the Future Training Event Guidelines (above) agreed upon by the focus group. Other feedback mechanisms are under review. Use of WMO guidelines for forecaster skill in use of satellite data and products, as well as using information developed through the Expert Team on Satellite Systems Utilization and Products, ET-SSUP, evaluation of the biennial questionnaire, will help guide activities undertaken within the VL. To keep abreast of user requirements for the VL (baseline being WMO Pub No. 258), the analysis of user responses focused on education and training to ETSSUP questionnaire will be shared with the “centers of excellence” within their region to aid them in planning future training events. In addition, questionnaire results and other user feedback activity that is carried out by “centers of excellence” will be reported back to VL focus group for use in reporting to CGMS and WMO.
5
CONCLUSIONS
The Virtual Laboratory for Education and Training in Satellite Matters (VL) has been established to maximize the exploitation of satellite data across the globe. Meeting the demands of this challenge is only possible because of the combined efforts of the World Meteorological Organization (WMO) and the world’s producers of operational meteorological satellite data through their Coordination Group for Meteorological Satellites (CGMS). To date three exceptionally successful WMO-sponsored training events have occurred within the VL construct. The VL’s current makeup (as of 2004) is WMO six “centers of excellence” and four major satellite operators. It is expected to grow in both membership and scope as it matures. Exciting times lie ahead for those interested in training and full utilization of satellite data. Reflection on the state of the art in personal computer technology and communications since the initial CIRA activity in 1994 and their extraordinary growth to today’s capabilities bode well for the VL as it moves into the future where communications and computer technology are centric to success. The increase in satellite observing capabilities during the coming decade, with the expectation of high quality data for operational use
J. W. F. Purdom and D. E. Hinsman
704
being provided by both operational and research5 satellite operators, promising improved spatial resolution, the addition of spectral bands and increased temporal observing capabilities (rapid scanning by geostationary satellites as well as more operational polar satellites) place an enormous responsibility on those undertaking the challenge of full utilization of those resources: the VL for satellite training and data utilization is well poised to meet that challenge. Acknowledgment: The VL is a joint venture between the WMO and CGMS, whose strong support has been central to its formation. The various partners in the VL are the fiber bringing it to reality. The VL focus group played a pivotal role in the VL’s formation and will continue to guide its evolution.
6
REFERENCES
Molenar, D. A., K. J. Schrab, and J. F. W. Purdom, 2000: RAMSDIS contributions to NOAA satellite data utilization. Bull. Amer. Meteor. Soc., 81, 1019–1030.
5
It should be recalled that the fifty-third WMO Executive Council, held in June 2001 decided to expand the space-based component of the Global Observing System to include appropriate Research and Development satellite missions. ESA, NASA, NASDA (now JAXA), Rosaviakosmos and CNES have all made firm commitments for the participation of their satellite missions, i.e., Aqua, Terra, NPP, GPM, ENVISAT, etc. in the new R&D constellation.
52 THE INTERNATIONAL PRECIPITATION WORKING GROUP: A BRIDGE TOWARDS OPERATIONAL APPLICATIONS Vincenzo Levizzani1 and Arnold Gruber2 1
Consiglio Nazionale delle Ricerche, Istituto di Scienze dell’Atmosfera e del Clima, Bologna, Italy 2 National Oceanic and Atmospheric Administration, National Environmental Satellite, Data and Information Service, College Park, MD, USA Abstract
The International Precipitation Working Group (IPWG) was established in 2001 as a scientific-technical group co-sponsored by the Coordination Group for Meteorological Satellites (CGMS) and the World Meteorological Organization (WMO). It is established to foster the development of better measurements, and improvement of their utilization, the improvement of scientific understanding, and the development of international partnerships. The IPWG has a worldwide partnership among operational bodies and research institutions, and aims towards a wider and improved use of satellite-derived rainfall products by operational organizations.
Keywords
Rainfall, weather forecasts
1 INTRODUCTION The plan to establish the IPWG as a permanent Working Group of the Coordination Group for Meteorological Satellites (CGMS) was developed at a Workshop held on 20–22 June 2001 at the Colorado State University. The IPWG main scope is to focus the scientific community on operational and research satellite–based quantitative precipitation measurement issues and challenges. In the area of quantitative precipitation estimation, the IPWG builds upon the expertise of scientists who are currently involved in precipitation measurements from satellites with emphasis on derivation of products. The IPWG was established to foster:
705 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 705–712. © 2007 Springer.
V. Levizzani and A. Gruber
706
• • •
Development of better measurements, and improvement of their utilization. Improvement of scientific understanding. Development of international partnerships.
The objectives of the IPWG are: (a) to promote standard operational procedures and common software for deriving precipitation measurements from satellites; (b) to establish standards for validation and independent verification of precipitation measurements derived from satellite data, including: • reference standards for the validation of precipitation for weather, hydrometeorological, and climate applications; • standard analysis techniques that quantify the uncertainty of groundbased measurements over relevant time and space scales needed by satellite products; (c) to devise and implement regular procedures for the exchange of data on intercomparisons of operational precipitation measurements from satellites; (d) to stimulate increased international scientific research and development in this field and to establish routine means of exchanging scientific results and verification results; (e) to make recommendations to national and international agencies regarding the utilization of current and future satellite instruments on both polar and geostationary platforms; (f) to encourage regular education and training activities with the goal of improving global utilization of remote-sensing data for precipitation measurements.
2 THE FIRST IPWG SCIENCE WORKSHOP The first IPWG science workshop was held at the EUMETSAT Nowcasting Satellite Applications Facility (SAF) in Madrid, Spain from 23–27 September 2002 (Levizzani and Gruber 2003). The workshop promoted the exchange of scientific and operational information between the producers of precipitation measurements, the research community, and the user community, and developed pathways forward for a variety of activities within the IPWG. The workshop contributed to set the framework of the IPWG activities by identifying the following topical areas: • • • •
Operational Estimation of Rainfall. Missions and Instruments. Research Activities. Validation.
The International Precipitation Working Group
707
3 OPERATIONAL APPLICATIONS Operational applications of satellite rainfall estimation are far from being satisfactory in qualitative and quantitative terms and the IPWG is called to contribute products and strategies that transfer scientific achievements into the operational meteorological practice. The IPWG web site (http://www.isac.cnr.it/~ipwg/) is established as an interface towards science and operational users. First, an IPWG subgroup is formed for the development of instantaneous precipitation estimation export algorithms for users of geostationary and low Earth-orbiting satellite data. The idea is to collect current, operationallyoriented, instantaneous satellite-based precipitation algorithms to be made available through the IPWG website. The first step is the compilation of an inventory of routinely produced precipitation estimates; either operational or experimental/research. The two most important criteria for the inventory are: (1) the retrieval algorithm be published so that others can study what was done and (2) the algorithm be in current use and producing precipitation estimates on a routine or regular schedule. The current status of the inventory is reported in Table 1. A second key activity of the IPWG aims to provide information on current and research-oriented climatological precipitation techniques. An ad hoc group is established to survey the requirements for space/time, accuracy, data latency time, and temporal coverage. A coordinated research on satellite based precipitation estimation research with the needs and requirements for climatological applications is needed. While climate models typically analyze precipitation on time scales of one month and space scales of 2.5 degrees, there is an increasing need for daily-scale precipitation estimates and space scales approaching one degree. These data can be produced with a latency time, but need to have data records extending back over a sufficient number of years in order to be useful for identifying long time scale variability. Tables 2 and 3 describe the available algorithms (Arkin 2002). IPWG has recognized the necessity of coordinating efforts and activities contemplating the use of satellite precipitation data for “non-traditional” applications. This implies broadening the application of satellite precipitation products by cooperation with communities other than meteorology and hydrology, e.g., snow models, irrigation models, pest and disease models, mud slide and avalanche models, dispersion models, surface pollution models, and others.
V. Levizzani and A. Gruber
708 Table 1. IPWG algorithm inventory. IR based algorithms Algorithm name CMA Convective-Stratiform Technique (CST) EURAINSAT/A 1.0 High resolution Precipitation Index (HPI) JMAMSC Multiple precipitation estimations blend Algorithm name GOES Multispectral Rainfall Algorithm (GMSRA) GPCP 1 Degree Daily GPCP Satellite – Gauge Combination Hydro-Estimator TRMM var (3B41RT) UOB Advection 1.0 UOB NET 1.0 MW based algorithms Algorithm name AMSU operational global rain rates AMSU global monthly and pentad rainfall SSM/I operational global rain rates SSM/I global pentad and monthly rainfall TRMM HQ (3B40RT) Blended MW-IR algorithms Algorithm name CPC Morhping technique (CMORPH) EURAINSAT/B 1.0 NRL Blended Satellite Technique PERSIANN TRMM HQ/VAR (3B42RT)
Institution China Meteorological Agency, People Rep. of China NASA /GSFC, USA EURAINSAT project, EU NASA/GSFC, USA Japan Meteorological Agency, Japan Institution NOAA/NESDIS, USA NASA /GSFC, USA NASA/GSFC, USA NOAA/NESDIS, USA NASA/GSFC, USA University of Birmingham, UK University of Birmingham, UK Institution NOAA/NESDIS, USA NOAA/NESDIS, USA NOAA/NESDIS, USA NOAA/NESDIS, USA NASA/GSFC, USA Institution NOAA, USA EURAINSAT project, EU Naval Research Laboratory, USA University of California, Irvine, USA NASA/GSFC, USA
Table 2. IPWG algorithm inventory for climate applications. The table contains algorithms that are merged products based on various inputs, including satellite-derived estimates from geostationary and low earth orbit IR, SSM/I and AMSU B. In general, they all incorporate gauge observations in some fashion (courtesy of P. A. Arkin 2002). Algorithm
Application
GPCP monthly
Monitoring, verification, research Monitoring, verification, research Monitoring, research
GPCP pentad GPCP daily
Update frequency Monthly
Latency 3 mon
Space/time scales 2.5°/monthly
Seasonal
3 mon
2.5°/5-day
Monthly
3 mon
1°/daily
African
Monitoring
Daily
6h
10 km/daily
South Asian
Monitoring
Daily
6h
10 km/daily
CAMS/OPI
Monitoring
Monthly
6h
2.5°/daily
Areal/temporal coverage Global/ Jan 1979 – present Global/ Jan 1979 – present Global – 50°N-50°S/ November 1997 – present Africa/ April 2000(?) – present South Asia/ April 2001 - present Global/ January 1979 – present
The International Precipitation Working Group
709
Table 3. IPWG algorithm inventory for climate applications using satellite data only (courtesy of P. A. Arkin 2002). Algorithm
Input data
Application
Update frequency
Latency
Space/ time scales
Goddard 3-hourly
IR and MW (SSM/I, TRMM)
Monitoring
3h
6h
0.25°/3hourly
GPI
IR
Baseline, input to merged algorithms
30 min
15 min
Various
NESDIS/ FNMOC Scattering index (Ferraro) NESDIS High Frequency (Weng and Ferraro) Goddard Profiling Algorithm (GPROF)
SSM/I
Baseline, input to merged algorithms, monitoring, research Baseline, input to merged algorithms, monitoring, research Monitoring research
Daily
6h
Daily
4h
Daily
4h
Wilheit/ Chang
SSM/I, TRMM(?)
Monthly
3 months
TOVS
TOVS/ HIRS
Daily
1 months (?)
1°/daily
OPI
AVHRR
Baseline, input to merged algorithms, research Input to merged algorithms, research Input to merged algorithms
0.25°/daily 1.0°/pentad, monthly 2.5°/pentad, monthly 0.25°/daily 1.0°/pentad, monthly 2.5°/pentad, monthly 0.25°/daily 1.0°/pentad, monthly 2.5°/pentad, monthly 2.5°/monthly
Daily
1 day
2.5°/daily
AMSU
TMI, AMSR
Areal/ temporal coverage Global – 50°N-50°S/ January 2002 – present Global – 40° N–40° S/ January 1986 – present Global/ July 1987– present January 1999 – present November 1997– present Global ocean/ July 1979 – present Global/ January 1979 – present Global/ January 1979 – present
Finally, the need for operational algorithms brings the IPWG to encouraging a continued development, refinement, and validation of the various researchstatus satellite sensor precipitation estimation techniques, i.e., • •
MW satellite precipitation techniques; Combined or “blended” satellite precipitation techniques (IR+MW); Satellite + NWP precipitation techniques; Improved validation techniques to properly analyze the error characteristics of satellite-derived precipitation estimates at different space and time scales; Assimilation into NWP models at short time scales (instantaneous to 6-hourly) and relatively fine spatial scales (0.25 degree or less).
V. Levizzani and A. Gruber
710
4 RESEARCH ACTIVITIES The first goal of the IPWG for supporting research on rainfall algorithms is to provide a generally accessible platform on data and algorithms for the research community. In particular, there is the need for developing elements of algorithm transition required for general applications: quantitative precipitation estimation (QPE), bias/error estimate, limits of applicability, documentation, comprehensive bibliography, compliance with user requirements (as defined by WMO, EUMETSAT, NASA-DAO, ESA, and others), compilation of user requirements with respect to precipitation. Practical actions that are being pursued are: •
•
Set up an inventory of field campaign data with colocated satellite and ground data relevant for validation purposes, i.e., provide the necessary data sets for validation exercises; Enable cooperation and training through the exchange of software libraries; Develop new strategies on flexible and global structures for physical algorithm development, validation, and data fusion. Establishing a framework for o o o
Physical algorithm development (global and regional). Developing a testbed for algorithm validation (comparison data sets etc.). Product merging and blending.
The principal problem of under-constrained precipitation retrievals requires different types of analysis/retrieval approaches such as: • • • •
more complex analysis methodologies; multiple sensors, multiple satellites, new instrument developments (e.g., geo-MW, lightning sensors, dual-frequency radars); climatological information; simplified dynamical physical models.
The usefulness and complexity of the needed methodology and information depends on the application. Only by stronger constraints on algorithms special retrieval problems can be solved such as orographic precipitation, light precipitation, frozen precipitation, resolution enhancement [spatial (vertical/horizontal) and temporal]. The IPWG is active on the recommendations and indications for future sensors that should emerge based on the identified scientific outstanding areas. In view of future sensor development a long-term strategy for frequency protection has to be developed and integrated in the current ITWG activities
The International Precipitation Working Group
711
(http://cimss.ssec.wisc.edu/itwg/groups/frequency/). A first report was compiled for the protection of MW-Sub-mm frequency range (Bizzarri 2002).
5 VALIDATION ACTIVITIES Validation activities will have to represent a large part of the overall efforts on algorithm development and will involve an increasing international coordination. Baseline validation standards for satellite precipitation algorithm(s) are to be provided in terms of the needs of users in NWP data assimilation, nowcasting, hydrology, climate, and algorithm development communities. Validation metrics are to be defined for NWP, nowcasting, hydrology, and climate. This involves the active participation of members from these communities to the IPWG validation activities. The IPWG has started a validation effort in 2004 in cooperation with the Global Precipitation Climatology Project (GPCP, Gruber et al. 2007) to monitor the performance of operational precipitation algorithms on a large scale and on a daily basis, preferably in connection with NWP forecast validation. The ongoing effort is unprecedented and its results will provide indications on how to improve operational algorithms in the near future. The reader is directed to two web pages in particular: (1) Access pages for IPWG validation activities http://www.isac.cnr.it/~ipwg/validation.html http://www.bom.gov.au/bmrc/SatRainVal/validation-intercomparison.html (2) Australia: Australian Bureau of Meteorology Research Center http://www.bom.gov.au/bmrc/SatRainVal/sat_val_aus.html (3) USA: National Oceanic and Atmospheric Administration, Climate Prediction Center http://www.cpc.ncep.noaa.gov/products/janowiak/us_web.shtml (4) Europe: University of Birmingham, UK http://kermit.bham.ac.uk/%7Ekidd/ipwg_eu/ipwg_eu.html The definition of suitable approaches to the rainfall validation problem, including alternative approaches to error estimation such as physical error modeling and cloud/system classification to obtain global error estimates (Ebert 2007), is an outstanding activity supported by the IPWG. Other activities will concern the creation of an inventory of existing high quality reference data, the sharing of data from Intensive Observation Periods in large-scale experiments, and the development of new methods for the error characterization of reference datasets. Finally, surface reference data need a special attention by the community and the IPWG (e.g., Krajewski 2007; Morrissey 2007) is active in:
V. Levizzani and A. Gruber
712
• • •
Encouraging the use of dual-gauge systems and optimal network design in operational rain gauge networks, to improve the reliability and quality of rainfall observations. Investigating the quality and availability of surface reference networks for the validation of hard-to-measure (orographic, light, solid) precipitation. Developing an assessment software package, incorporating both basic and advanced techniques, to facilitate validation of satellite rainfall estimates by algorithm developers and users.
6 REFERENCES Arkin, P. A., 2002: Climate monitoring principles and requirements. IPWG Rep., http://www.isac.cnr.it/~ipwg/reports/IPWG-Climate-Report.pdf, 4 pp. Bizzarri, B., 2002: Requirements of IPWG for MW/Sub-mm frequencies protection. IPWG Rep., http://www.isac.cnr.it/~ipwg/reports/IPWG-frequency-protection-requirements.pdf, 3 pp. Ebert, E. E., 2007: Methods for verifying satellite precipitation estimates. In: Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 345–356. Gruber, A., B. Rudolf, M. M. Morrissey, T. Kurino, J. Janowiak, R. Ferraro, R. Francis, A. Chang, and R. F. Alder, 2007: The Global Precipitation Climatology Project. In: Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 25–36. Krajewski, W. F, 2007: Ground networks: are we doing the right thing? In: Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 403–418. Levizzani, V. and A. Gruber, 2003: The 1st International Precipitation Working Group (IPWG) Workshop. In: Proc. 1st IPWG Workshop, EUMETSAT EUM P34, ISBN 92-9110-045-5, pp. VII-VIII. Morrissey, M. M. and S. Greene, 2007: Ground validation for the Global Precipitation Climatology Project. In: Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F. J. Turk, eds., Springer, 381–392.
List of acronyms ABI ACARS ACT ADC ADEOS AE AGCM AGPI AHRPT AIP AIRS ALOS AMS AMSR-E AMSU AMV ARPS ASCAT ASI ATMS ATOVS ATSR AVHRR AWIPS AWS BALTEX BAMPR BB BLC BMRC BMTC BOLAM BoM CAL CAMEX CAMS CAPCOM CBS CCN CCOPE CCSP CERES CG CGCM CGMS CHAMP CIMH CIRA CLM
GOES-R Advanced Baseline Imager Aircraft Communications Addressing and Reporting System Advective-Convective-Technique (rainfall) Meteosat Atlantic Data Coverage Advanced Earth Observing Satellite Auto-Estimator (NOAA rainfall estimation technique) Atmospheric Global Climate Model Adjusted Geostationary Operational Environmental Satellite (GOES) Precipitation Index EUMETSAT Advanced High Resolution Picture Transmission Algorithm Intercomparison Programme (satellite rain estimates) Advanced InfraRed Sounder Advanced Land Observing Satellite American Meteorological Society Advanced Scanning Microwave Radiometer – Earth Observation Satellite Advanced Microwave Sounding Unit Atmospheric Motion Vectors Advanced Regional Prediction System Advanced Scatterometer Agenzia Spaziale Italiana Advanced Technology Microwave Sounder Advanced TIROS Operational Vertical Sounder Along Track Scanning Radiometer Advanced Very High Resolution Radiometer Advanced Weather Interactive Processing System Automated Weather Station BALTic sea Experiment Bayesian Algorithm for Microwave-based Precipitation Retrieval Bright Band Baroclinic Lee Cyclone BoM Research Centre BoM Training Center (Australia) BOlogna Limited Area Model Bureau of Meteorology (Australia) Computer-Aided Learning Convection and Moisture Experiment Climate Analysis and Monitoring System Comprehensive Analysis Program for Cloud Optical Measurement Commission for Basic Systems (WMO) Cloud Condensation Nuclei Cooperative Convective Precipitation Experiment Climate Change Science Program (USA) Clouds and Earth Radiant Energy System Cloud to Ground (lightning) Coupled Global Climate Model Coordination Group for Meteorological Satellites CHAllenging Minisatellite Payload Caribbean Institute for Meteorology and Hydrology (Barbados) Cooperative Institute for Research in the Atmosphere (Colorado State Univ.) Community Land surface Model 713
714 CloudSat CMA CMAP CMIS CMORPH CNES CNR COADS COARE COLA COMET COSMIC CPC CPL CRAs CrIS CRM CRP CRS CRVSs CSA CSAVs CST CSTP DAAC DBL DCS DMSP DoD DPR DWD Earth CARE ECMWF ECST EDOP EGPM EML ENSO EORC EOS EOS EPS ERBE ERS ESA ESCAP ESDIS ESE ESMR ESRIN ESRO ESTEC ET-SSUP
List of acronyms Cloud satellite (NASA) China Meteorological Agency CPC Merged Analysis of Precipitation Conical Microwave Imager/Sounder CPC MORPHing technique Centre National d’Études Spatiales (France) Consiglio Nazionale delle Ricerche (Italy) Comprehensive Ocean – Atmosphere Data Set TOGA Couple Ocean Atmosphere Response Experiment Center for Ocean-Land-Atmosphere Studies (Univ. Maryland) UCAR’s Cooperative Program for Operational Meteorology, Education and Training Constellation Observing System for Meteorology, Ionosphere and Climate NOAA Climate Prediction Center Cloud Physics Lidar Contiguous Rain Areas Cross-track scanning Infrared Sounder Cloud Resolving Model Cloud Radiation Prediction Cloud Radiation Simulation Cloud Radiation Verification Studies Canadian Space Agency Cloud System Advection Vectors Convective-Stratiform-Technique Council for Science and Technology Policy (Japan) Distributed Active Archive Center Desert Boundary Layer Data Collection Systems Defense Meteorological Satellite Program (USA) Department of Defense (USA) Dual-frequency Precipitation Radar Deutscher Wetterdienst (Germany) Earth Clouds, Aerosol and Radiation Explorer European Centre for Medium-range Weather Forecasts Enhanced Convective-Stratiform-Technique NASA ER-2 Doppler radar European contribution to GPM (ESA) Elevated Mixed Layer El Niño – Southern Oscillation Earth Observation Research and application Center (Japan) Earth Observation Summit NASA Earth Observing System EUMETSAT Polar System Earth Radiation Budget Experiment Earth Remote Sensing satellites European Space Agency UN Economic and Social Commission for Asia and the Pacific NASA Earth Sciences Distribution Information Systems NASA’s Earth Sciences Enterprise Electronically Scanning Microwave Radiometer European Space Research Institute (ESA) European Space Research Organisation European Space Research and Technology Centre (ESA) Expert Team - Satellite Systems Utilization and Products (WMO)
List of acronyms EU EUMETSAT EURAINSAT EUROMET EVAC FNMOC FDA FP6 FY GARP GATE GCE GCM GCOM GCOS GDAS GEM GEO GERB GEWEX GFS GHCC GHCN GHG GIFTS GIS GISS GLI GLONASS GLOWA GMES GMI GMS GMSRA GMSRAD GNDCI GNSS GOES GOMAS GOME GOMS GOS GOSAT GPC GPCC GPCP GPI GPM GPPS GPROF GPSOS GrADS
715 European Union European Organization for the Exploitation of Meteorological Satellites EUropean SATellite rainfall analysis and monitoring at the geostationary Scale European Collaboration in Measurement Standards Environmental Verification and Analysis Center U.S. Navy’s Fleet Numerical Meteorology and Oceanography Center Frequency Difference Algorithm (satellite rain estimation) Framework Programme 6 of the European Commission Fengyun satellite series (China) Global Atmospheric Research Program GARP Atlantic Experiment Goddard Cumulus Ensemble model General Circulation Model or Global Climate Model Global Change Observation Mission Global Climate Observing System Global Data Assimilation System Geostationary Microwave Observatory Geostationary Earth Orbit (also, a satellite in GEO) Geostationary Earth Radiation Budget instrument Global Energy and Water cycle Experiment (WCRP WMO) Global Forecast System (NOAA NWS) Global Hydrology and Climate Center (NASA) Global Historical Climate Network Greenhouse Gases Geostationary Imaging Fourier Transform Spectrometer Geographic Information System NASA Goddard Institute for Space Studies Global Imager sensor Global Navigation Satellite System GLObalen Wasserkreislauf (Global Change in the Hydrological Cycle, Germany) Global Monitoring for the Environment and Security GPM Microwave Imager Geostationary Meteorological Satellite GOES Multispectral Rainfall Algorithm GMSRA with radar addition Gruppo Nazionale per la Difesa dalle Catastrofi Idrogeologiche (Italy) Global Navigation Satellite System Geostationary Operational Environmental Satellite Geostationary Observatory for Microwave Atmospheric Sounding Global Ozone Monitoring Experiment Geostationary Operational Meteorological Satellite Global Observing System Greenhouse Gas Observation Satellite Global Positioning System Global Precipitation Climatology Centre Global Precipitation Climatology Project GOES Precipitation Index Global Precipitation Measurement mission Global Precipitation Processing System Goddard Profiling algorithm Global Positioning System Occultation Sensor Grid Analysis and Display System (COLA)
716 GRAS GSFC GSN GTOS GTS GV GWEC HALOE HCM HDST HES HIRS HPM HRAP HSB HUCM IASI IC IFFA IGBP IIS IMS INDOEX IODC IOS IPCC IPO IPWG ISSCP ISRO IR IR Tb ITCZ ITWG IWC IWWG JAM JAXA JCLIM JMA JRSSJ KMA KWAJEX LaMMA LDAS LEAF LEO LEO-IR LHN LIS LRPT LSM MADRAS
List of acronyms GNSS (GPS) Receiver for Atmospheric Sounding NASA Goddard Space Flight Center GRAS Support Network Global Terrestrial Observing System Global Telecommunication System Ground Validation Global Water and Energy Cycle HALogen Occultation Experiment Hail Category Model NASA’s Hydrology Data Support Team Hyperspectral Environmental Sounder High Resolution Infrared Radiation Sounder Hail Parameterization Model Hydrologic Rainfall Analysis Product Humidity Sounder for Brazil Hebrew University Cloud Model Infrared Atmospheric Sounding Interferometer Intra Cloud (lightning) NOAA Interactive Flash Flood Analyzer International Geosphere-Biosphere Programme Integrated Imaging System NESDIS Interactive Multi-sensor Snow and Ice Mapping System INDian Ocean EXperiment Meteosat Indian Ocean Data Coverage WMO CBS OPAG team on Integrated Observing Systems Intergovernmental Panel on Climate Change Integrated Program Office (USA) International Precipitation Working Group International Satellite Cloud Climatology Project Indian Space Research Organisation Infrared Infrared brightness temperature Inter-Tropical Convergence Zone International TOVS Working Group Ice Water Content International Winds Working Group Journal of Applied Meteorology Japan Aerospace eXploration Agency (formerly NASDA) Journal of Climate Japan Meteorological Agency Journal of Remote Sensing Society of Japan Korean Meteorological Agency Kwajalein Islands EXperiment Laboratorio di Meteorologia e Modellistica Ambientale (Italy) Land Data Assimilation Systems Land Ecosystem Atmosphere Feedback model Low Earth Orbit (also, a satellite in LEO) IR observed by a LEO satellite Latent Heat Nudging Lightning Imaging Sensor EUMETSAT Low Resolution Picture Transmission Land Surface Model Multi-frequency Microwave Scanning Radiometer
List of acronyms MAP MCC MCS McIDAS MCR MCS METEOR METOP MHS MICRA MIRRA MJO MLP MM5 MODAPS MODIS MOP MPA MSFC MSG MSM MSPPS MSU MTG MTP MTPE MW MWCOMB NAO NASA NASDA NAST-M NCAR NCDC NCEP NCR NESDIS NEXRAD NICT NIR NIWA NLDN NMHS NN NOAA NOGAPS NPOESS NPP NPR NRE NRL NSCATT
717 Mesoscale Alpine Programme Mesoscale Convective Complex Mesoscale Convective System Man computer Interactive Data Access System Mission Confirmation Review Mesoscale Convective System Russian polar orbiting weather satellite series METeorological Operational weather satellite Microwave Humidity Sounder Microwave Infrared Combined Algorithm MW IR Rainfall Algorithm Madden-Julian Oscillation MultiLayer Perceptron Fifth-Generation NCAR/Penn State Mesoscale Model MODIS Adaptive Processing System MODerate resolution Imaging Spectroradiomenter Meteosat Operational Programme Multi-satellite Precipitation Analysis NASA Marshall Space Flight Center Meteosat Second Generation Mesoscale Spectral Model Microwave Surface and Precipitation Products System Microwave Sounding Unit Meteosat Third Generation Meteosat Transition Programme NASA Mission To Planet Earth Microwave MW COMBination (daily combination of passive MW satellite data) North Atlantic Oscillation National Aeronautics and Space Administration National Space Development Agency of Japan (now JAXA) NPOESS Aircraft Sounder Testbed - Microwave National Center for Atmospheric Research (USA) National Climatic Data Center (USA) National Center for Environmental Predictions (USA) Non-Contiguous Rain gauge method National Environmental Satellite Data and Information Service (USA) NEXt generation weather RADar National Institute of Information and Communications Technology (Japan) Near IR New Zealand’s National Institute of Water and Atmospheric research National Lightning Detection Network (USA) National Meteorological and Hydrological Service (WMO members) Neural Network National Oceanic and Atmospheric Administration (satellite) US Navy Operational Global Atmospheric Prediction System National Polar-orbiting Operational Environmental Satellite System NPOESS Preparatory Project Nadir-pointing Precipitation Radar Neural Rain Estimator Naval Research Laboratory (USA) Adeos-1 scatterometer
718 NSMC NWP NWS OGCM OGPM OKEAN OMI OMPS OPAG OPI ORNL OSSE PACRAIN PARASOL PATER PBL PDO PDUS PERSIANN PIP PMM PMS PMW PMWR POES POLDER PPS PR PRM PSG QBO QPE QPF QuikSCAT RAMM RAMS RAMSDIS RAPI RESURS RMTC Rosaviakosmos Roshydromet RTM RTTOV SAC SAF SAFARI SAPHIR
List of acronyms National Satellite Meteorological Center (China) Numerical Weather Prediction National Weather Service (USA) Ocean Global Climate Model Operational Global Precipitation Measurement Russian oceanographic satellite series Ozone Mapping Instrument Ozone Mapping and Profiler Suite Open Program Area Group (WMO) OLR Precipitation Index Oak Ridge National Laboratory (USA) Observing System Simulation Experiment Comprehensive Pacific Rainfall Database Polarization & Anisotropy of Reflectances for Atmospheric Sciences coupled with Observations from a Lidar PR Adjusted TMI Estimation of Rainfall Planetary Boundary Layer Pacific Decadal Oscillation Primary Data User Station (Meteosat) Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks Precipitation Intercomparison Project Probability Matching Method (radar-satellite) Precipitation Multi-Satellite product Passive MW Passive MW Radiometer Polar Operational Environmental Satellite program POLarization and Directionality of the Earth's Reflectances GPM Precipitation Processing System TRMM Precipitation Radar Passive Radiation Model Precipitation Satellite-Gauge product Quasi Biennial Oscillation Quantitative Precipitation Estimation Quantitative Precipitation Forecast NASA Quick Scatterometer Mission Regional and Mesoscale Meteorology (Colorado State Univ.) Regional Atmospheric Modelling System RAMM Advanced Meteorological Satellite Demonstration and Interpretation System RAMS Antecedent Precipitation Index Russian mapping satellite series Regional Meteorological Training Center (WMO) Russian Aviation and Space Agency Russian Federal Service for Hydrometeorology and Environmental Monitoring Radiative Transfer Model Radiative Transfer for TOVS model Space Activities Commission (Japan) EUMETSAT MSG Satellite Application Facilities Southern African Fire-Atmosphere Research Initiative Sondeur Atmosphérique du Profil d'Humidité Intertropicale par Radiométrie
List of acronyms SAR SBUV SeaWiFS SCaMPR ScaRaB SCSMEX SeaWinds SEVIRI SMMR SOFM SORCE SPaRCE SPCZ SRDC SRT SSCR SSD SSM/I SSMIS SSP TCI TDR TDRSS TIROS TMI TOGA TOMS TOVAS TOVS TRMM TSDIS UAGPI UARS UCAR U-MARF UN UTC UW-NMS VAR VDOM VIIRS VIRS VIS VISSR VL VLF VPR VRL WCRP WFP WHOM WINDSAT WISCDYMM WISHE
719 Synthetic Aperture Radar Solar Backscatter Ultraviolet Sea-viewing Wide Field-of-view Sensor Self-Calibrating Multivariate Precipitation Retrieval SCAnner for RAdiation Budget South China Sea Monsoon Experiment QuikSCAT scatterometer for sea surface winds Spinning Enhanced Visible and InfraRed Imager Scanning Multichannel Microwave Radiometer Self-Organizing Feature Map Solar Radiation and Climate Experiment Schools of the Pacific Rainfall Climate Experiment South Pacific Convergence Zone Surface Reference Data Center (University of Oklahoma, Norman) Surface Reference Technique (radar) Signal Simulator for Cloud Retrieval NESDIS Satellite Services Division Special Sensor Microwave/Imager Special Sensor Microwave Imager/Sounder Sub Satellite Point TRMM Combined Instrument Temperature Data Record format Tracking and Data Relay Satellite System (NASA) Television-Infrared Observation Satellite TRMM Microwave Imager Tropical Ocean Global Atmospheres experiment Total Ozone Mapping Spectrometer TRMM Online Visualization and Analysis System TIROS Operational Vertical Sounder Tropical Rainfall Measuring Mission TRMM Science Data and Information System Universally Adjusted GPI Upper Atmosphere Research Satellite University Corporation for Atmospheric Research (USA) EUMETSAT Unified Meteorological Archive and Retrieval Facility United Nations Coordinated Universal Time University of Wisconsin Non-hydrostatic Modeling System Variable Rainrate precipitation algorithm Vector Discrete Ordinate Method (RTM) Visible Infrared Imaging Radiometer Suite Visible/Infrared Scanner Visible Visible Infrared Spin-Scan Radiometer WMO’s Virtual Laboratory for Satellite Training and Data Utilization Very Low Frequency Vertically Pointing Radar Virtual Resource Library World Climate Research Programme (WMO) UN World Food Programme Web Hierarchical Ordering Mechanism Windsat/Coriolis mission for wind speed and direction (IPO/DoD/NASA) Wisconsin Dynamical/Microphysical Model Wind Induced Surface Heat Exchange
List of acronyms
720 WMO WSDD WSR-88D WV WVCA WWW XADC 1DD
World Meteorological Organization World Summit on Sustainable Development Weather Surveillance Radar-88 Doppler Water Vapor Water Vapor Column Amount World Weather Watch (WMO) Meteosat Extended Atlantic Data Coverage 1 Degree Daily product
List of symbols and functions BRN BTD CAPE CDF CRAs CRDB CSI CTH CTT CWC CWV D D0 D0e dB DSD EBBT EOF EEOF EnKf ETS ewvc FAR FMR FOV FSE FVR GHz HBB Hc Hfreeze HF HKI hPa Hstorm HSS ISE IWC IWFs IWP K K kCLW KDP keff KL kO2 kP kWV LAI LI
Bulk Richardson Number Brightness Temperature Difference Convective Available Potential Energy Cumulative Distribution Function Contiguous Rain Areas Cloud Radiation Data Base Critical Success Index Cloud Top Height Cloud Top Temperature Columnar equivalent Water Content Column Water Vapor Drop diameter Median volume drop diameter Equilibrium median volume drop diameter Decibel Drop Size Distribution Equivalent Blackbody Brightness Temperature Empirical Orthogonal Function Extended EOF Ensemble-Kalman Filter Equitable Threat Score Equivalent water vapour of the cloud layer False Alarm Ratio Fractional Mean Reduction Field of View Fractional Standard Error Fractional Variance Reduction Gigahertz (109 Hertz) Height of bright band Cloud height Height of freezing level High Frequency Hansen and Kuiper Index HectoPascal (10.19716 kg m-2 or 0.001 bar) Height of storm top Heidke Skill Score Index of Simmetry of Error Ice Water Content Incremental Weighting Functions Ice Water Path Degrees Kelvin K index (also Whiting index) Specific attenuation due to cloud liquid water Specific differential phase Effective specific attenuation Karhunen-Loeve (empirical orthogonal) coefficient Specific attenuation due to molecular oxygen Specific attenuation due to precipitation Specific attenuation due to water vapor Leaf Area Index Lifted Index 721
List of symbols and functions
722 LUT LWC MAE MPM N(D,r) NEB NEDT N0 OLR OT PDF PIA POD PODR PODNR PV PW R rCCN RDSD Re RMSE SOFM SOM SNR SST STD T TB TCWV TS UPM Z ZDR Ze ZH Zm Zmmax ZV ∆s⊥
φ ΦDP
Look-Up Table Liquid Water Content Mean Absolute Error Multivariate Probability Matching Number density of drop as a function of diameter and radar range Normalized Error Bias Noise Equivalent Delta Temperature Normalized drop concentration offset Outgoing Longwave Radiation Optical Thickness Probability Density Function Path Integrated Attenuation Probability of Detection Probability of Detection of Rain Probability of Detection of No Rain Potential Vorticity Precipitable Water Rainrate (also RR) Radius of CCN Raindrop Size Distribution Effective radius (also re) Root Mean Square Error Self-Organizing Feature Map (neural networks) Self Organizing Map (neural networks) Signal-to-Noise Ratio Sea Surface Temperature Standard deviation Temperature Brightness temperature Total Column Water Vapor Threat Score Univariate Probability Matching Radar reflectivity factor Differential reflectivity Effective reflectivity factor Reflectivity factor at horizontal polarization Measured reflectivity factor Maximum Zm Reflectivity factor at vertical polarization Absolute horizontal displacement of radiation in the direction Perpendicular to the viewing direction between the emission and the Sensor position Azimuth angle (also Φ) Propagation differential phase
θ
Satellite zenith angle
θ0 Θv
Solar zenith angle Viewing angle (conical scanning) Reflectance (radiation) and hydrometeor density Standard deviation Normalized surface backscatter cross section Optical thickness Surface albedo
ρ σ
σ m0
τ
ωs
Advances in Global Change Research 1. 2.
3. 4. 5.
6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.
P. Martens and J. Rotmans (eds.): Climate Change: An Integrated Perspective. 1999 ISBN 0-7923-5996-8 A. Gillespie and W.C.G. Burns (eds.): Climate Change in the South Pacific: Impacts and Responses in Australia, New Zealand, and Small Island States. 2000 ISBN 0-7923-6077-X J.L. Innes, M. Beniston and M.M. Verstraete (eds.): Biomass Burning and Its InterISBN 0-7923-6107-5 Relationships with the Climate Systems. 2000 M.M. Verstraete, M. Menenti and J. Peltoniemi (eds.): Observing Land from Space: ISBN 0-7923-6503-8 Science, Customers and Technology. 2000 T. Skodvin: Structure and Agent in the Scientific Diplomacy of Climate Change. An Empirical Case Study of Science-Policy Interaction in the Intergovernmental Panel on Climate Change. 2000 ISBN 0-7923-6637-9 S. McLaren and D. Kniveton: Linking Climate Change to Land Surface Change. 2000 ISBN 0-7923-6638-7 M. Beniston and M.M. Verstraete (eds.): Remote Sensing and Climate Modeling: ISBN 0-7923-6801-0 Synergies and Limitations. 2001 E. Jochem, J. Sathaye and D. Bouille (eds.): Society, Behaviour, and Climate Change ISBN 0-7923-6802-9 Mitigation. 2000 G. Visconti, M. Beniston, E.D. Iannorelli and D. Barba (eds.): Global Change and ISBN 0-7923-6818-1 Protected Areas. 2001 M. Beniston (ed.): Climatic Change: Implications for the Hydrological Cycle and for ISBN 1-4020-0444-3 Water Management. 2002 N.H. Ravindranath and J.A. Sathaye: Climatic Change and Developing Countries. 2002 ISBN 1-4020-0104-5; Pb 1-4020-0771-X E.O. Odada and D.O. Olaga: The East African Great Lakes: Limnology, PalaeolimnolISBN 1-4020-0772-8 ogy and Biodiversity. 2002 F.S. Marzano and G. Visconti: Remote Sensing of Atmosphere and Ocean from Space: ISBN 1-4020-0943-7 Models, Instruments and Techniques. 2002 F.-K. Holtmeier: Mountain Timberlines. Ecology, Patchiness, and Dynamics. 2003 ISBN 1-4020-1356-6 H.F. Diaz (ed.): Climate Variability and Change in High Elevation Regions: Past, ISBN 1-4020-1386-8 Present & Future. 2003 H.F. Diaz and B.J. Morehouse (eds.): Climate and Water: Transboundary Challenges ISBN 1-4020-1529-1 in the Americas. 2003 A.V. Parisi, J. Sabburg and M.G. Kimlin: Scattered and Filtered Solar UV MeasureISBN 1-4020-1819-3 ments. 2004 C. Granier, P. Artaxo and C.E. Reeves (eds.): Emissions of Atmospheric Trace ComISBN 1-4020-2166-6 pounds. 2004 M. Beniston: Climatic Change and its Impacts. An Overview Focusing on Switzerland. 2004 ISBN 1-4020-2345-6 J.D. Unruh, M.S. Krol and N. Kliot (eds.): Environmental Change and its Implications ISBN 1-4020-2868-7 for Population Migration. 2004 H.F. Diaz and R.S. Bradley (eds.): The Hadley Circulation: Present, Past and Future. 2004 ISBN 1-4020-2943-8 A. Haurie and L. Viguier (eds.): The Coupling of Climate and Economic Dynamics. Essays on Integrated Assessment. 2005 ISBN 1-4020-3424-5
Advances in Global Change Research 23.
24. 25. 26. 27. 28.
U.M. Huber, H.K.M. Bugmann and M.A. Reasoner (eds.): Global Change and Mountain Regions. An Overview of Current Knowledge. 2005 ISBN 1-4020-3506-3; Pb 1-4020-3507-1 ISBN 1A.A. Chukhlantsev: Microwave Radiometry of Vegetation Canopies. 2006 4020-4681-2 J. McBeath, J. Rosenbery : Comparative Environmental Politics. 2006 ISBN 1-4020-4762-2 M.E. Ibarraran ´ and R. Boyd: Hacia el Futuro. Energy, Economics, and the Environment in 21st Century Mexico. 2006 ISBN 1-4020-4770-3 N.J. Roserberg: A Biomass Future for the North American Great Plains: Toward Sustainable Land Use and Mitigation of Greenhouse Warming. 2006. ISBN 1-4020-5600-1 V. Levizzani, P. Bauer and F.J. Turk (eds.): Measuring Precipitation from Space. EURAINSAT and the Future. 2007 ISBN 978-1-4020-5834-9
springer.com