Advances in Water Science Methodologies

Advances in Water Science Methodologies

Advances in Water Science Methodologies addresses ways and tools for using remote sensing data in various applications to underpin the study of interactions between the atmospheric, oceanic, and hydrological processes. Various water-related applications such as water resources management, environmental monitoring, climate prediction, agriculture, and preparation for and mitigation of extreme weather events, are characterized by widely varying requirements of spatial, temporal, and spectral resolutions. This volume was composed with greatest care to address the varying issues with the appropriate data, assimilation methodologies, and technology transfer practices, to cover various dimensions of the subject area and to illustrate potential growth aspects of remote sensing. By making the relation with cognate subjects such as data management and geomorphology and with the support of two case histories of water resource management dealing with water harvesting and water pollution, a complete picture of the area is provided. The book will be useful to university students and professionals in the area of remote sensing, water sciences and technologies, earth and environmental sciences, resource management, agriculture, civil engineering, ecology, and related areas. The editor, U. Aswathanarayana, has over 50 years of research and teaching experience both in southern (India, Tanzania, and Mozambique) and northern countries (United States of America, United Kingdom, and Canada). He has specialized in Nuclear Geology, Geochemistry, Economic Geology, and Natural Resources Management. The author is currently engaged in the development of the Mahadevan International Centre for Water Resources Management, Hyderabad, India, which is a part of the UNESCO–TWAS Network of Scientific Organizations. The editor was awarded the prestigious Excellence in Geophysical Education Award (2005) of the American Geophysical Union for his meritorious contributions to the paradigm shift in geoscience instruction. His previously published works are Principles of Nuclear Geology (Balkema, 1986), Water Resources Management and the Environment (Balkema, 2001), and Mineral Resources Management and the Environment (Balkema, 2003).

Advances in Water Science Methodologies

Edited by U. Aswathanarayana Mahadevan International Centre for Water Resources Management, Hyderabad, India

This edition published in the Taylor & Francis e-Library, 2005. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” Copyright © 2005 Taylor & Francis Group plc, London, UK All rights reserved. No part of this publication or the information contained herein may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, by photocopying, recording or otherwise, without written prior permission from the publisher. Although all care is taken to ensure the integrity and quality of this publication and the information herein, no responsibility is assumed by the publishers nor the author for any damage to property or persons as a result of operation or use of this publication and/or the information contained herein. Published by: A.A. Balkema Publishers, Leiden, The Netherlands, a member of Taylor & Francis Group plc www.balkema.nl and www.tandf.co.uk

ISBN0-203-08684-8 Master e-book ISBN

ISBN 0–415–37533–9 (Print Edition) Library of Congress Cataloging-in-Publication Data A catalog record for this book has been requested

Contents

List of figures List of tables Foreword Preface About the author Annotations of technical terms

vii xi xiii xv xvii xix

PART 1

Remote-sensing sensors, data retrieval, assimilation, and technology transfer 1 Remote sensing and hydrology

1 3

VENKAT LAKSHMI

2 Hydrologic data assimilation

25

JEFFREY P. WALKER AND PAUL R. HOUSER

3 Analysis of remotely sensed data

49

R. KRISHNAN AND B.L. DEEKSHATULU

4 Technology transfer in remote-sensing applications

61

S. KALLURI AND P. GILRUTH

PART 2

Remote-sensing data applications 5 Computing and mapping of evapotranspiration

71 73

RICHARD G. ALLEN, ANTHONY MORSE, MASAHIRO TASUMI, WILLIAM J. KRAMBER, AND WIM BASTIAANSSEN

6 Satellite remote sensing of soil moisture

91

THOMAS J. JACKSON

7 Ensemble streamflow forecasting: methods and applications BALAJI RAJAGOPALAN, KATRINA GRANTZ, SATISH REGONDA, MARTYN CLARK, AND EDITH ZAGONA

97

vi Contents 8 Regional climatic variability and its impacts on flood and drought hazards

117

B. GOZZINI, M. BALDI, G. MARACCHI, F. MENEGUZZO, M. PASQUI, AND F. PIANI

9 Climate drivers, streamflow forecasting, and flood risk management

135

GONZALO PIZARRO AND UPMANU LALL

10 Remote sensing in water resource management

157

D.P. RAO

11 Geospatial information technology in watershed management

171

I.V. MURALI KRISHNA

PART 3

191

Water resource management: case histories 12 Runoff agroforestry

193

P.R. BERLINER

13 Water pollution and its numerical modeling in coastal watersheds

201

A. GHOSH BOBBA AND VIJAY P. SINGH

Author index Subject index

221 227

Figures

1.1 1.2 1.3 1.4 1.5 1.6 2.1 2.2 2.3 2.4 2.5 2.6 2.7

3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 4.1 4.2

Land-atmosphere hydrological processes and variables attained by satellite remote sensing Comparison of surface and air temperatures and vapor pressure deficits retrieved using TOVS and AVHRR for (a) January 1989 and (b) August 1989 Comparison of surface, air temperatures and vapor pressure deficits retrieved using TOVS and AVHRR for (a) January 1990 and (b) August 1990 Schematic representation of the variable infiltration capacity model Monthly mean streamflow at Mississippi River at Grafton, IL, 1950–99 Illinois State averaged monthly soil moisture comparison 1981–99 Schematic of the hydrologic data assimilation challenge Satellite observations of near-surface soil moisture content made by SMMR Schematic of the (a) direct observer and (b) dynamic observer assimilation approaches Example of how data assimilation supplements data and complements observations True and prior surface soil saturation at three different times Comparison of snow simulations on January 5, 1987 over North America for snow water equivalent, snow depth, average snow temperature, and areal snow fraction Differences between simulated and reanalysis, assimilated and reanalysis mean skin temperature, and the resulting differences between simulated and reanalysis, and assimilated and reanalysis mean sensible heat fluxes for September–November 1992 Data volume evolution Asynchronous imaging mode Exposure principle of a TDI detector with three stages Example of decision tree classifier Classification accuracy Data dimension vs number of elements Accuracy vs dimensionality Scale space effect achieved through diffusion Hough transform SNAKES Road detection Remote sensing application development life cycle Cost of satellite data at different spatial resolutions

4 10 11 17 19 19 26 27 30 33 43 44

45 50 51 51 54 56 56 57 58 58 58 59 62 65

viii Figures 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 6.1 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 8.1 8.2 8.3 8.4

Comparison of ETr fractions derived from 7-day lysimeter measurements near Montpelier, Idaho during 1985 and values from METRIC for 4 Landsat dates Hourly measurements of ET, ETr, ETrF, and 24-h ETrF for clipped grass and sugar beets at Kimberly, Idaho on July 7, 1989 Comparison of daily ET predicted by METRIC using ETrF and SEBAL using EF on satellite image dates for sugar beets, potatoes, peas, and alfalfa Results by METRIC and ET by lysimeter as ETrF Comparison of cumulative METRIC ET with maximum water-right ET for 426 water-right polygons in Idaho Department of Water Resources Basin 35 (a) FCC image of T3NR1E of the Boise Valley, (b) land use–land cover polygons in T3NR1E of the Boise Valley, and (c) ET image of T3NR1E of the Boise Valley The scatter plot of pumpage vs METRIC ET for the period April–October, 2000 (a) April–October, 2000 METRIC ET compared with AgriMet ET extremes and (b) April–October, 2000 pumpage compared with AgriMet ET extremes Brightness temperature–soil moisture sensitivity as a function of microwave frequency Flow chart of the forecast framework Map of the Truckee–Carson Basin Gunnison River Basin and streamflow locations Climatology of streamflows and precipitation in the Truckee River, at the gauging station Farad Correlation of Carson River spring streamflows with winter climate variables (a) 500 mb geopotential height (Z500) and (b) SST Composites of vector winds, SST and Z500 during the winter of high and low streamflow years Correlation between PC1 of spring flows and November–March climate indices Composite of vector wind at 700 mb for (a) wet years and (b) dry years Residual resampling to obtain an ensemble forecast Skill scores of forecasts issued from the first of each month November–April for Truckee and Carson rivers PDF of the ensemble forecasts in a (a) dry year (1992) and (b) wet year (1999) for the Truckee River Median RPSS score for forecasts issued on January 1 and April 1 for the six streamflow sites Boxplots of ensemble streamflow forecasts at the station East River, Almont, for the dry, wet, and average years (a) Seasonal precipitation time series over the Sahel area and (b) boundaries of the Sahel region Composite difference of the storm track strength between strong and weak West African monsoon events during August (a) Location of the Arno River Basin and (b) orography of the Arno River Basin and location of the rain gauges Annual precipitation time series over the Arno River Basin: (a) annual frequency of rainy days and total precipitation and (b) annual frequency of rainy days and average daily precipitation intensity over the lower portion of the Arno River Basin

76 79 81 81 83 86 87 87 92 98 100 100 101 103 104 105 105 107 111 112 112 113 121 122 123

124

Figures ix 8.5 8.6 8.7 8.8 8.9

8.10 8.11 8.12 8.13 9.1 9.2 9.3

9.4 9.5 9.6 9.7 9.8 9.9 9.10 11.1 11.2 11.3 11.4 11.5 11.6 11.7 11.8

Winter season precipitation time series over the Arno River Basin: annual frequency of rainy days and total precipitation Frequency of rainy days over high thresholds: (a) lower Arno River Basin, (b) rain gauge in the upper Arno River Basin, and (c) rain gauge in the medium Arno River Basin (a) Global topography of the underlying surface in the CGCM2 climate scenario and (b) 1 km resolution topography and boundaries of the Arno River Basin in the grid cell of the CGCM2 climate scenario Annual time series of the frequency of rainy days over the threshold 20 mm around the Arno River Basin, simulated by means of the CGCM2-A2 scenario (a) Annual time series of the summer average rainfall intensity around the Arno River Basin and (b) annual time series of the summer total precipitation and frequency of rainy days around the Arno River Basin, simulated by means of the CGCM2-A2 scenario Annual time series of the seasonal average surface air temperature around the Arno River Basin, simulated by means of the CGCM2-A2 scenario Arno River discharge at the Subbiano section, during 1930–2003 Changes of the winter storm track strength, simulated by means of the CGCM2-A2 scenario Changes of the summer storm track strength, simulated by means of the CGCM2-A2 scenario Economic losses due to major weather-related natural disasters in the 1950–2000 period, 2001 prices, USD millions Normal state of (a) sea surface temperature and (b) sea level pressure Typical extent of (a) the warming, (b) cooling in the equatorial Pacific, (c) weakened Pacific high-pressure center, and (d) enhanced Pacific high-pressure center during developed El Niño (a and c) and La Niña (b and d) events Typical climatic anomalies attributable to fully developed El Niño or La Niña events Sea surface temperature anomalies during the (a) positive and (b) negative phase of the Pacific Decadal Oscillation Relationship of the highest annual daily flow at 137 locations for streams in the western United States to two slowly varying modes of global climate CO2 concentrations at different epochs Risk adequate premiums as a function of the size of the risk community Institutional structure for a catastrophe bond Concept diagram for climate-driven risk management approach for flood hazard A comparative study of spatial and attribute data in related information systems The relationship between GIS and base map/data Classification of GIS with reference to mapping and analysis capabilities Concept of GIS GIS project cost scenario Vector data model Vector data structure Topology for point, line, and polygon

125 126 127 128

129 130 130 132 132 135 138

139 140 141 142 144 148 149 152 173 174 174 175 176 177 178 178

x Figures 11.9 Grid/raster representation of point, line, and area features 11.10 Workflow diagram of the GIS layer creation approach 11.11 Typical examples of GIS topology build up for watershed features for area–area and area–line features 11.12 Typical examples of GIS topology build up for watershed features for point–line and point–area features 12.1 Total seasonal rainfall for the years 1962–2000 recorded at the Evenari Farm at Avdat 12.2 Schematic description of a runoff agroforestry system with P: precipitation, R: runoff water, B: cropping area, E, T: evaporation and transpiration, D: deep drainage, W: walls surrounding the cropping area, and S: spillway 12.3 Schematic description of an ideal sequence of events for a runoff agroforestry system 13.1 Different sources of pollutants in a coastal watershed 13.2 Hydrological pathways in a coastal watershed 13.3 Location of the Godavari Delta 13.4 Variation of potassium concentration in groundwater in the Godavari Delta 13.5 Variation of [Cl/HCO3 ⫹ CO3] ratio in groundwater in the Godavari Delta 13.6 (a) Simulated hydraulic heads and (b) freshwater depth of Godavari Delta in non-irrigation months 13.7 (a) Simulated hydraulic heads and (b) freshwater depth of Godavari Delta in irrigation season months 13.8 Simulated hydraulic heads of Godavari Delta in different seasons 13.9 Location of waste disposal site and Ra-226 concentration in Lake Ontario, Canada 13.10 Computed Ra-226 concentration in waste site, beach and observed concentration in coast

179 180 180 181 194

196 198 203 208 209 213 214 215 215 216 217 218

Tables

1.1 List of continental regions used for comparison of TOVS and AVHRR 1.2 Statistics of mean and standard deviation over the study regions averaged over a 2-year period 1.3 Comparison of monthly spatially distributed surface skin temperature, surface air temperature, and vapor pressure deficit derived using the TOVS and AVHRR using bias, standard deviation, and correlation coefficient 1.4 VIC vs TOVS surface temperature comparison 1981–99 1.5 Upper Mississippi River Basin averaged monthly soil moisture 2.1 Characteristics of hydrologic observations potentially available within the next decade 2.2 Commonly used data assimilation terminology 3.1 Dichotomies of classification 5.1 Summary of METRIC- and lysimeter-derived ET for weekly and monthly periods and the associated error for Bear River, 1985 5.2 Summary and computation of ET during periods represented by each satellite image and sums for April 1–September 30, 1989 for lysimeter 2 at Kimberly, Idaho 5.3 A comparison of METRIC ET with average ET for three cells of the Treasure Valley hydrologic model 5.4 Mean seasonal ET by land use–land cover class 6.1 Microwave satellite systems 7.1 Median skill scores for ensemble forecast issued on April 1, for all years, wet and dry, for the Truckee–Carson Basin 7.2 Median skill scores for ensemble forecast issued on April 1, for all years, wet and dry, for the Gunnison Basin 10.1 The morphological and spectral characteristics of different rock groups as expressed on the satellite imagery 10.2 Hydro-geological classification system 10.3 Hydro-geomorphological classification system 10.4 Classification of recharge conditions 11.1 A comparative analysis of the cost in INR per hectare for carrying out integrated resources mapping 11.2 Data quality components 13.1 Point sources of pollution in a coastal watershed 13.2 Non-point sources of pollution in a coastal watershed 13.3 Major water- and excreta-related pathogens

9 13 13 20 21 28 31 52 77 80 83 84 93 111 114 160 162 164 166 172 176 202 203 205

Foreword

Water science has to be the basis for good water management. Although modern society may be characterized by specialization and expert advice, we cannot all be specialists on everything. We depend on each other. Cooperation is a necessity, and for cooperation we need a common language. A common language and a basis of a common knowledge are necessary elements in our mutual sharing of new knowledge. Textbooks like Advances in Water Science Methodologies are therefore important building blocks in the dissemination of new knowledge. We are living in a period of time when one of the basic conditions of human life is threatened by increasing water consumption and loss of availability caused by uneven distribution and degrading quality of the world’s freshwater. There is, however, a broad consensus that global water resources are, and will continue to be, sufficient provided we manage these resources equitably and wisely. There is no ground for desperation, but new approaches need to be explored. Sustainability is the key factor in this effort. Sound water management is neither simple nor easy. There is strong need for political will to reform the water sector, to improve water-related legislation, to introduce economic tools when necessary, and to efficiently plan and control water supply and demand. Rational decision making in water management has to be based on water science and water technologies. Experience with good, successful practices should be disseminated. Again, the situation calls for cooperation. The emphasis of this book on remote sensing as a tool is very pertinent, and its chapters cover highly important aspects of water management such as soil moisture and agriculture, evaporation and drought, streamflow, climate change and flood hazard, and water pollution. The problems associated with water stress and scarcities are inherently most pronounced in the world’s arid and semi-arid zones, and the book will have a particularly important message for these parts of the world. However, application of remote sensing to solving water management problems is definitely of global relevance. It is a sad fact that our systems for monitoring the world’s water resources are far from sufficiently developed. We have little or no systematic infrastructure for providing early warnings when global, or even regional and national water resources are at risk. In fact, hydrological data collection networks are deteriorating in many countries, and statistics on water withdrawal are generally very poor. Remote sensing has definitely become a major tool for mitigating this lack of knowledge. It has for long been a promising avenue, and it is high time that we refine and make operational its use for water management purposes. This book provides both an updated background of the technological possibilities as well as examples of cost-efficient applications and useful case studies. By its very nature, water science is interdisciplinary. The chapters in this very current volume, imaginatively put together by Prof. U. Aswathanarayana, draw the attention of the university students and professionals to the high-technology observing systems, global and regional simulations, assimilation schemes, etc. available to understand and address the anthropogenic impact on the water resources and optimal ways of managing them.

xiv Foreword Scientific advances are always welcome and laudable. Their application for the benefit of humanity is even better. This book is an important step in that direction. Arne Tollan Norwegian Water Resources and Energy Directorate, Oslo, Norway December 2004

Preface

Two significant advances, which definitely need to be incorporated in water resources management methodologies, have been made since the publication of the well-received, broad-spectrum work of the author, Water Resources Management and the Environment (A.A. Balkema Publishers, The Netherlands, 2001). The first area of advance is conceptual. There is widespread recognition that only through synergy between the earth (including the atmospheric and oceanic realms), space and information sciences, is it possible: ● ●

●

to reduce the predictive uncertainty in hydrological sciences; to address the complex issues involved in the management of the four kinds of waters (rain water, surface water, groundwater, and soil water); and to generate employment in the process of utilizing them.

The second advance is in the area of tools. Because of the advantages of repetitive coverage and capability for synoptic overview, satellite remote sensing has emerged as a powerful and cost-effective tool covering all aspects of water resources management. Advances in Water Science Methodologies was written to promote research, development, and education in this subject area, so that the scope of remote-sensing applications is enlarged. This way, it is aimed to make these methodologies commercially viable through launching of dedicated satellite systems, developing new retrieval algorithms for remote-sensing data, and formatting them for ingestion into GIS packages. In addition, the interaction with stakeholders, training of cadres, public policies, and other conditions varying per country should also be taken greatest care of. It is evident that this can only be done properly when customized to the biophysical and socioeconomic situations of a country. In this book, methods and means of using remote-sensing data to underpin the effect of interactions between the atmospheric, oceanic, and hydrological processes are treated together with their use in various applications. Technical terms used in the remote-sensing chapters have been annotated, to make it possible for non-specialists to understand them. Different water-related applications such as water resources management, environmental monitoring, climate prediction, agriculture, and preparation for and mitigation of extreme weather events are characterized by widely varying requirements of spatial, temporal, and spectral resolutions implying that different data assimilation methodologies and technology transfer practices are needed under different conditions. The themes of the chapters in the volume have hence been chosen carefully to cover various dimensions and potential growth areas of remote sensing such as the existing and projected satellite sensors, image analysis, data assimilation methods, technology transfer modalities, agriculture-related applications (involving evapotranspiration and soil moisture), prediction of runoff and flood risk, management of water resources in watersheds through linkages with geomorphology, etc. The last two chapters deal with two critically important themes of water resources management, namely, water harvesting and water pollution.

xvi Preface The contributors of the chapters are well-known experts in the subjects of their contribution. Some material from other anthologies edited by the author and for which he holds the copyright has been included in the volume to make the coverage comprehensive and self-contained. Although most chapters are essentially non-mathematical, a few chapters need knowledge of advanced mathematics for comprehension. The training and publications activities of the Mahadevan International Centre for Water Resources Management, Hyderabad, India, have been actively supported by Mr P.V.R.K. Prasad, Director General, MCR Human Resources Development Institute of the Andhra Pradesh Government, and Mr B.K. Rao, a senior civil servant of the Government of India, who has been closely associated with water resources management in the country. The volume carries a perceptive foreword by Arne Tollan, Senior Adviser, Norwegian Directorate of Water Resources and Energy, Oslo, Norway. The author started compiling and editing this volume when he was in Linkoping, Sweden, to visit his son’s infant daughter, appropriately named Neera (Nira), which means water in Sanskrit. His children Srinivas, Indira, and Vani, and his friend, Mr H.L. Hung, provided technical support. The book will be useful to university students and professionals in the area of remote sensing, water sciences and technologies, earth and environmental sciences, resource management, agriculture, civil engineering, and ecology. U. Aswathanarayana Boulder, Colorado, USA September 2004

About the author

U. Aswathanarayana who edited the volume has research and teaching experience of half-a-century in the countries of the South (India, Tanzania, and Mozambique) and of the North (United States of America, United Kingdom, and Canada). He received his BSc (Hons), MSc, and DSc degrees from Andhra University, Visakhapatnam, India. He specialized in Nuclear Geology, Geochemistry, Economic Geology, and Natural Resources Management. He is the author of over 100 original scientific papers. His first book titled Principles of Nuclear Geology (A.A. Balkema Publishers, The Netherlands, 1986) was followed by a quartet of books on the ecologically sustainable and employment-generating utilization of natural resources, namely Geoenvironment: An Introduction (A.A. Balkema Publishers, The Netherlands, 1995), Soil Resources and the Environment (Science Publishers, Enfield, USA, 1999), Water Resources Management and the Environment (A.A. Balkema Publishers, 2001), and Mineral Resources Management and the Environment (A.A. Balkema Publishers, 2003). While in Africa during 1980–2001, he served as a Consultant to UNIDO (Vienna), Commonwealth Secretariat (London), SIDA (Stockholm), World Bank (Washington, DC), Louis Berger International Inc. (New Jersey, USA), and the Ministry for the Coordination of Environmental Affairs (Mozambique). He is the Chairman of the Working Group on ‘Geochemical Training in the Developing Countries’ of the International Association of Geochemistry and Cosmochemistry (IAGC). He is presently engaged in developing the Mahadevan International Centre for Water Resources Management, Hyderabad, which is a part of the UNESCO–TWAS Network of Scientific Organizations. He is the recipient of the Excellence in Geophysical Education Award (2005) of the American Geophysical Union.

Annotations of technical terms

ADJOINT AIRS/AMSU ANALYSIS ANN ASTER AVHRR AVIRIS BACKGROUND CEOP

CERES COVARIANCE MATRIX DAAC DAO DBMS DEM DIAGNOSTIC DTC GAIN MATRIX GAPP GCMs GEOS GEWEX

Operator allowing the model to be run backwards in time An advanced version of HIRS2-MSU or the TOVS, which has a higher spatial and spectral resolution for the atmospheric soundings and land surface temperature Prediction after an update Artificial Neural Networks – a mathematical model used in classification which derives its inspiration from the working of the human brain Advanced Spaceborne Thermal Emission and Reflection sensor – senses the land surface temperature and emissivity Advanced Very High Resolution Radiometer – instrumental system mounted aboard NOAA’s Polar Orbiting Environmental Satellite, provides data about the temporal and spatial distribution of vegetation Airborne Visual and Infrared Imaging Spectrometer – a remote sensing Instrument Prediction prior to an update Coordinated Enhanced Observing Program – a system that uses the existing satellites and ground networks to understand the land-atmosphere-ocean states of the earth. This is a collaborative partnership between various countries/space agencies/meteorological organizations Clouds and Earth Radiant Energy System Describes the standard deviations and correlations Data Active Archival Center – a GSFC-based computer system that collects, archives, and helps distribute satellite data Data Assimilation Office – carries out some of the LDAS work at GSFC Database Management System Digital Elevation Model – this describes the altitude at every geographical point in the image; it is usually available in the form of a grid or as an irregular triangulated model of points A model state/flux diagnosed from the prognostic states – not required to propagate the model Decision Tree Classifier – this is a method of labeling Correction factor applied to the innovation GEWEX Americas Prediction Project General Circulation Models – a land surface – ocean model of the earth, which is used to predict future climate states Goddard Earth Observing System Global Water and Energy Experiment – project to better predict the water and energy flows in and out of the watersheds in North America

xx Annotations of technical terms GOES GSFC INNOVATION LAI LDAS

MLC MODIS MTF NDVI

NNC OBSERVATION PR PROGNOSTIC SSMI STATE SVM TANGENT LINEAR MODEL TDI TOVS TRMM UPDATE VCL VIC WCRP

Geostationary Earth Observing System, in the visible, near infrared and visible channels – provides information on incoming solar radiation, clouds, and surface temperature Goddard Space Flight Centre of NASA, Baltimore, USA Observation-prediction Leaf Area Index and NDVI are well correlated Land Surface Data Assimilation Scheme – use of estimation theory to “correct” the predictions made by the land surface model using the observations and the error characteristics of the model and the observations Maximum Likelihood Classifier – a supervised classification system Moderate Resolution Imaging Spectro-Radiometer – a satellite sensor in the visible infrared and thermal bands that characterizes the vegetation and the temperature of the land surface Modular Transform Function – a ratio of the output and input contrast of an imaging system Normalized Difference Vegetation Index is given by the ratio: Near Infra Red ⫺ Red/Near Infra Red ⫹ Red – Green leaf foliage is characterized by a strong absorption in the red region, and a strong reflectance in the Near Infra Red NIR region, due to scattering. A decrease in NDVI is indicative of reduced photosynthetic activity and green biomass Neural Network Classifier – based on ANN principle Measurement of a model diagnostic or prognostic Pattern Recognition – a method of labeling A model state required to propagate the model forward in time Special Sensor Microwave Imager – a four-frequency, seven-channel, microwave imager that provides information on surface temperature, wetness, atmospheric water vapor, and precipitation for land and oceans Condition of a physical system, that is, soil moisture Support Vector Machine – a method to create decision boundaries between classes Linearized using Taylor’s series expansion version of a non-linear model Time Delay Integration – a method of imaging which reduces the needed aperture TIROS Operational Vertical Sounder – contains HIRS High Resolution Infrared Sounder and MSU Microwave Sounding Unit, which sense the air temperature and water vapor at various levels in the atmosphere Tropical Rainfall Measuring Mission – a satellite package containing instruments that sense the rainfall in the tropics Correction to a model prediction using observations Vegetation Canopy Lidar – a Lidar system that determines the vertical distribution of the vegetation on the canopy Variable Infiltration Capacity – a hydrological model developed by the University of Washington World Climate Research Program – climate research agenda for the world

Part 1

Remote-sensing sensors, data retrieval, assimilation, and technology transfer

CHAPTER 1

Remote Sensing and Hydrology Venkat Lakshmi Department of Geological Sciences, University of South Carolina, Columbia SC 29223, USA

1.1

INTRODUCTION

The themes of the chapter are grouped into Satellite Remote Sensing (1.1–1.6), Satellite Validation Studies (1.7–1.10), and Hydrological modeling (1.11–1.15) (the remote-sensing technical terms used in the chapter are defined in Annotations). Satellite data sets offer many advantages to conventional in-situ ground-based observations. Traditional in-situ ground observations have limitations for input, validation, and assimilation in models. Point data is difficult to interpret over spatial domain of models that range from 1/8⬚ ⫻ 1/8⬚ for the high resolution Land Surface Data Assimilation Schemes (LDAS) to 2⬚ ⫻ 2.5⬚ in the case of Global Climate Models. Satellite data provides continuous spatial coverage and repeat temporal coverage. The spatial and temporal coverages are dependent on the orbit and swath of the satellite, and the resolution of the sensor. The use of satellite data sets is extremely important in the context of the EOS satellites that provide data sets on a wide number of atmospheric and land surface variables. The EOS Terra satellite has been launched in December 1999 and the EOS Aqua has been launched in May 2002. Furthermore, there are a variety of satellites such as those launched by Japan (ADEOS II), Europe (ENVISAT), and India (INSAT) that will also have global coverage using different sensors but sense similar/same variables at different overpass times. Together, these satellites carry new and enhanced sensors that will provide high-resolution data sets that will be made available to the scientific community through the Goddard Data Active Archival Center (DAAC). Figure 1.1 depicts the physical variables that can be sensed by multiple satellite remote sensors. Land surface modeling of hydrological and ecological processes on continental and global scales is an important research problem. Comprehensive observations of the land surface and near surface atmospheric variables needed as input for models or for validating model outputs are lacking. The lack of ground observations is a result of the prohibitive costs of establishing and maintaining the large number of sample stations required to characterize the spatial heterogeneity of the variables. Remote-sensing data are attractive to the modeling community as they are available at high spatial and temporal resolutions. 1.2

OBJECTIVES

The remotely sensed satellite data are utilized to fulfill the following objectives: 1 Input variables to offline land surface hydrological models. These input variables include vegetation content, air temperature, precipitation, total atmospheric precipitable water content, atmospheric temperature and water vapor profile, cloud fraction, and height to cloud base. 2 Validation of model output products such as surface temperature and soil moisture content. 3 Assimilation of satellite-derived products in land surface models. The products assimilated include surface temperature and soil moisture.

4 Venkat Lakshmi

Closing the terrestrial water budget using remote sensing ⌬W/⌬t = E + T – P – div Q
Soil moisture Microwave AMSR, SMOS, HYDROS∗

Atmospheric water balance

P Precipitation Microwave TRMM/TMI, SSM/I, GPM

Rn Radiation Shortwave GOES Longwave AIRS/AMSU H, G Surface temperature AIRS, AVHRR, MODIS Clouds GOES Water vapor (LE) AIRS/AMSU

⌬Z

T Transpiration/ND VI Visible/NIR MODIS, AVHRR, GLI, VCL

T

AIRS/AMSU T q

The land surface water and energy budgets are linked via evapotranspiration

E Evaporation/surface humidity Infrared/microwave AIRS/AMSU

P

E R Runoff/river level Laser HYDRASAT∗, TOPEX R




Groundwater flux

∗Planning phase

Water table Energy balance Rn + H + LE + G = 0

Water balance ⌬Z ⌬
/⌬t = P – E – T – R

Figure 1.1. Land-atmosphere hydrological processes and variables attained by satellite remote sensing (see Color Plate I).

4 Comparison of satellite-derived land surface products with the observations during field experiments and other data sets collected as a part of the coordinated enhanced observing program (CEOP). 1.3

REMOTE-SENSED DATA SETS

This section will outline the various variables that are retrieved using satellite data. These variables are classified according to their usage as stated in the previous section on objectives. Therefore, this section proposes utilization of single variables that may be derived from sensors with different spatial and temporal resolutions, coverage, and times of overpass. It may be noted that even though the same data sets have been mentioned in the validation and the assimilation modes these are designed to be complementary. The data used in the assimilation will not be used in validation and vice-versa. 1.3.1

Input variables and parameters in land surface models

Land surface models require various input data sets in order to characterize the properties of the land surface as well as to provide meteorological forcings. The input data sets include: 1 Leaf area index (LAI) derived from the Normalized Difference Vegetation Index (NDVI) from the Advanced Very High Resolution Radiometer (AVHRR) and/or Moderate Resolution Imaging Spectro-Radiometer (MODIS).

Remote sensing and hydrology 5 2 Surface roughness parameters – roughness length and zero plane displacement from the Vegetation Canopy Lidar (VCL). 3 Precipitation from Special Sensor Microwave Imager (SSM/I) and Tropical Rainfall Measuring Mission (TRMM). 4 Surface air temperature using TVX method from AVHRR or from the AIRS/AMSU and TIROS Operational Vertical Sounder (TOVS). 5 Surface specific humidity from AIRS/AMSU and TOVS. 6 Cloud cover fraction and height to cloud base derived from AIRS/AMSU, TOVS, and Clouds and Earth Radiant Energy System (CERES). 7 Atmospheric temperature and moisture profile from AIRS/AMSU and TOVS. 1.3.2

Validation data sets

Validation will be carried out using the following data sets: 1 Soil moisture derived using AMSR. 2 Surface temperature using AVHRR, ASTER, AIRS/AMSU, MODIS, TOVS, and Geostationary Earth Observing System (GOES). 1.3.3

Assimilation

Assimilation will be carried out using the following data sets: 1 Soil moisture derived using AMSR. 2 Surface temperature using AVHRR, ASTER, AIRS/AMSU, MODIS, TOVS, and GOES. 3 Air temperature and specific humidity profile of the atmosphere using AIRS/AMSU.

1.4

PREDICTION

GAPP will deal with numerous prediction issues on seasonal, annual, and interannual time scales. The use of satellite data in prediction models will have a major impact on model predictability. Assimilation of satellite data for soil moisture, surface temperature and precipitation in a real-time fashion will help in reducing forecast errors. These predictions and assimilation can be carried out on a regional scale and mesoscale as dictated by the particular application. In the case of detailed mesoscale applications, GOES-derived surface temperatures that have high spatial (1 km) and temporal (15 min) resolutions would be utilized for validation and assimilation purposes. In the case of seasonal predictions using the General Circulation Models (GCMs), coarser data sets can be used. The key objectives of an integrated seasonal prediction system can be realized by a better representation of the land surface system. This land surface system model will require inputs that have to be specified using satellite data. Data assimilation for the land surface will be carried out using remotely sensed data. In addition, prediction of land surface variables such as soil moisture and surface temperature can be validated using the satellite data over continental regions and extended time periods. Land surface states need to be initialized properly for accurate predictability. The initialization of land surface soil moisture and temperature can be carried out by the use of satellite data. Multi-scale downscaling of prediction components can be carried out using satellite data at appropriate resolution. For example, a large 1⬚ ⫻ 1⬚ forecast of land surface evapotranspiration can be disaggregated using the 1 km GOES surface temperatures and 250 m MODIS vegetation indices. The potential for use of satellite data to disaggregate the model forecasts onto

6 Venkat Lakshmi finer spatial resolutions is of prime importance in the future when satellite sensor resolutions will improve. 1.5

SCALING AND PROCESS INTERRELATIONSHIPS

Certain variables can be retrieved from different sensors with different spatial resolution and temporal overpass times. This leads to an interesting research issue of merging these data sets derived from different satellites. For example, surface temperature can be derived from GOES, AVHRR, MODIS, TOVS, AIRS/AMSU, and ASTER. Each of these sensors has different spatial resolutions. The spatial resolution for GOES is 1 km, AVHRR is either 1 km (raw data) or 4 km; MODIS has the thermal bands around 1 km, TOVS resolution is 60 km, AIRS/AMSU is at 12.5 km, and ASTER is around 90 m. The overpass times are different for each satellite. As a result, it will be possible to piece together the data from various satellite sensors to obtain a diurnal cycle. Therefore, two important tasks will be to merge these data sets for the same variable from different satellites in time and space so as to create a consistent data set. Such a merged data set will have to ensure spatial continuity between data from different sensors and temporal continuity between data from different platforms. This task is required before such data sets can be utilized. Process interrelationships can be studied using the data for different variables that are related to each other. An example of such a study would be the relationships between precipitation, soil moisture, surface temperature, and vegetation. Changes in precipitation patterns in time and space will affect vegetation, soil moisture, and surface temperature. However, there exists a feedback effect that would result in these affected variables (vegetation, soil moisture, and surface temperature) changing the precipitation patterns. Such a feedback could be positive, that is, changes in precipitation result in changes in the land surface variables which feedback to further change the precipitation. A negative feedback would result in a damping effect rather than amplification as described above. All of these variables can be derived using satellite data. Precipitation can be derived from TRMM, SSM/I, TOVS, and AIRS/AMSU, soil moisture from AMSR, surface temperature sources are mentioned above, and vegetation can be ascertained using AVHRR and MODIS. Regional scale process studies would focus on understanding the spatial distribution of these variables and the diurnal, seasonal, and interannual variations of these variables. Such studies can be carried out over seasonal, annual or multi-year time periods. These studies will provide useful comparisons with models for process interrelationships and will lead to better model parameterizations at a variety of space and time scales. 1.6

COORDINATED ENHANCED OBSERVATION PERIOD

Global Water and Energy Experiment (GEWEX) is a major sub-program of the World Climate Research Program, WCRP. GEWEX includes five continental scale experiments – BALTEX, GAME, GCIP, LBA and MAGS. In these continental scale areas, intensive measurement campaigns are being carried out using in-situ systems. The role of remote sensing (among others) is the validation of satellite algorithms used in the retrieval of land surface variables. The validation will be carried out using these in-situ measurements. In addition, the availability of spatially distributed satellite data will help in interpolation of these point-based measurements. The satellite data are available at specific times of the data but spatially continuous. The field measurements are at a point in space but temporally continuous. Therefore, schemes which use the spatial continuity of satellite data and the temporal continuity of the field measurements will help in creation of data sets that can be used for various process studies. The variables that will be the focal points of such a study include (but should not be limited to) – soil moisture, surface

Remote sensing and hydrology 7 temperature, precipitation, air temperature, and specific humidity near the surface. These variables will come from a variety of satellite sensors – AMSR, AIRS/AMSU, MODIS, GOES, TOVS, AVHRR, SSM/I, TMI, ASTER, etc. 1.7 1.7.1

SATELLITE VALIDATION STUDIES Introduction

This section examines, on global and continental spatial scales, the relationship between surface meteorological observations retrieved from AVHRR and TOVS. Comparisons of spatial and temporal patterns were carried out between January 1989 and December 1990 on a monthly time step for surface temperature, air temperature, and vapor pressure deficit. These variables play an important role in hydrological, ecological, and land surface–atmosphere interactions. A more detailed quantitative comparison was performed for seven continental land surface areas. The bias and standard deviation of the bias between the TOVS and AVHRR estimates were 5 K, 4 K, 1.5 kPa, for surface temperature, air temperature, and vapor pressure deficit, respectively for the continental areas. These comparisons show that there is a good agreement between the independent estimates of surface quantities from AVHRR and TOVS. In the absence of long-term and spatially extensive ground data sets, comparison studies such as the present one provide independent validation of various satellite estimates. Close correspondence of two independent estimates gives us greater confidence in the use of these satellite data. In order to use satellite data we must ensure that these data sets are unbiased when compared to surface measurements. In reality, the surface data are very difficult to ascertain. Satellite sensors obtain observations at a considerably coarser spatial resolution than ground sensors. When the variables in question contain significant spatial heterogeneity at a scale finer than that of the ground observations, errors in scaling up to the satellite sensor resolution are introduced. Modelers are often forced to interpret observations from a ground network as the average over the satellite sensor field-of-view, even when such errors are known to exist, due to the lack of more detailed information on spatial behavior. There have been numerous studies comparing land surface variables derived using satellite sensors with surface observations. These include (1) over forests (e.g. Lakshmi and Susskind, 2001), (2) over grasslands (e.g. Prince et al., 1998), (3) over mixed scrub vegetation and trees (e.g. Xiang and Smith, 1997), and (4) over airports (e.g. Lakshmi and Susskind, 1998). These studies carried out detailed analyses of the difference between satellite-derived land surface variables and the ground observations of the same variables over varying climatic regions, spatial extents, and time periods. The ground observations are by necessity acquired for small areas as a part of large interdisciplinary field experiments (e.g. Prince et al., 1998) or are routinely observed at airports (e.g. Lakshmi and Susskind, 1998). In this section we compare retrievals of land surface temperature, air temperature, and vapor pressure deficit from the AVHRR and the TOVS over a two-year period (1989–90) on a global scale. Regional comparisons over continental areas in Africa, North and South America, Europe, Australia, and Asia are also carried out. The spatial scale for the comparison is 1⬚ ⫻ 1⬚ and the temporal resolution is monthly. These statistics may prove useful for merging the two data sets complementing the high spatial resolution of the AVHRR with the high temporal frequency of the TOVS. 1.8

DATA AND METHODS

The data used in this study were derived from the AVHRR and the TOVS. These sensors were flown on the NOAA satellite series since 1978 and each captures global coverage twice a day. The sensors have different spatial and temporal resolutions as described below.

8

Venkat Lakshmi

1.8.1

AVHRR

Data from the AVHRR have been archived since July 1981 in the NOAA/NASA Pathfinder AVHRR Land Data Set (Agbu and James, 1994). This global data set contains afternoon overpass data from the NOAA-7, 9, and 11 satellites at a spatial resolution of 8 km. The AVHRR data is available as instantaneous values and biweekly averages. In this study we use the composited data to minimize cloud and atmospheric effects; composites are made up of the maximum per-pixel NDVI observations over the time period. The AVHRR sensor contains channels in the visible (channel 1: 0.55–0.68 ␮m), near infrared (channel 2: 0.73–1.10 ␮m), mid-infrared (channel 3: 3.55–3.93 ␮m), and thermal regions (channels 4 and 5: 10.3–11.3 ␮m and 11.5–12.5 ␮m). AVHRR data have been used extensively to determine land surface characteristics. Data from channels 1 and 2 have been used to estimate greenness via the NDVI (Tucker, 1979). Data from channels 4 and 5 have been used to derive land surface temperature via the split-window method (e.g. Becker and Li, 1990). The surface temperature algorithm uses satellite-specific coefficients and contains corrections for column water vapor and emissivity. Water vapor data was taken from Data Assimilation Office (DAO) reanalysis products (Schubert et al., 1993). A global emissivity map derived by the Surface and Atmosphere Radiation Budget working group for the CERES project (e.g. Gupta et al., 1999) was used for the emissivity correction. Canopy air temperature was estimated using the TVX method (Prihodko and Goward, 1997) in which a linear regression of surface temperature on NDVI is used to estimate canopy air temperature under the assumption that air temperature equals surface temperature for a full canopy. The vapor pressure deficit, or difference between saturated and actual atmospheric vapor pressure, was derived using TVX air temperature and amount of precipitable water in the atmospheric column extended to the surface (Prince and Goward, 1995). Column precipitable water was derived from the DAO reanalysis data set. For this comparison the variables were initially derived for the global land surface at a spatial resolution of 8 km and then aggregated to the TOVS resolution of 1⬚ using spatial averaging. 1.8.2

TOVS

The TOVS contains two instruments: the HIRS2 (High Resolution InfraRed Sounder) and the MSU (Microwave Sounding Unit). The TOVS has been operated on NOAA satellites from TIROS-N from 1978 to the present. Radiance data from the TOVS has been used to derive surface meteorological variables such as land surface temperature, air temperature, and specific humidity, atmospheric profiles of air temperature, water vapor and ozone burden, and cloud fraction and height (Susskind et al., 1997). These variables are calculated separately for each (instantaneous) overpass (2.30 am, 7.30 am, 2.30 pm, and 7.30 pm) and gridded to global 1⬚ ⫻ 1⬚ spatial resolution (land and ocean). The derived variables are aggregated into pentad (5 day) and monthly averages. The air temperature and water vapor profiles are calculated using an initial guess from the Goddard Earth Observing System (GEOS) GCM as input to a radiative transfer scheme (Susskind et al., 1984). The atmospheric profiles are adjusted so that the channel radiances calculated at the satellite equal observed channel radiances for the cloud free portions of the scene. Canopy air temperature and surface specific humidity are obtained by extrapolating the air temperature and water vapor profiles to the surface pressure. Land surface temperature is calculated directly using observations in the thermal and infrared regions (channels 8, 18, and 19: 11.14 ␮m, 3.98 ␮m, and 3.74 ␮m, respectively) and inversion of the Planck function. Surface emissivity values of 0.95 (channel 8) and 0.85 (channels 18, 19) were assumed for the surface temperature calculations. The atmospheric vapor pressure was calculated using surface specific humidity and surface pressure.

Remote sensing and hydrology 9 Table 1.1.

List of continental regions used for comparison of TOVS and AVHRR.

Region North America (East) North America (West) South America Africa Australia Central Asia Europe

1.8.3

Latitude

Longitude

Number of 1⬚ ⫻ 1⬚ boxes

30⬚N–50⬚N 30⬚N–50⬚N 10⬚S–10⬚N 10⬚S–10⬚N 20⬚S–35⬚S 50⬚N–65⬚N 45⬚N–50⬚N

80⬚W–100⬚W 100⬚W–120⬚W 40⬚W–80⬚W 10⬚E–30⬚E 115⬚E–140⬚E 60⬚E–80⬚E 0⬚E–5⬚E

400 400 800 400 375 300 25

Comparisons

AVHRR- and TOVS-derived surface temperature, air temperature, and vapor pressure deficit were compared over seven continental areas with different climatic, topographic, and vegetation characteristics. These continental areas were chosen to include different climatic regimes – temperate (North America East, Europe), desert (Africa and Australia), and tropical (South America). Also included are regions with uneven topography (North America West) and cold grasslands (Central Asia). The selection of these areas was designed to include a broad spectrum of conditions so as to achieve a balanced comparison of the two data sets. Table 1.1 contains the latitude and longitude boundaries and the number of 1⬚ ⫻ 1⬚ grid cells for each of the seven regions. Monthly data from AVHRR and TOVS for two years, January 1989–December 1990, were compared. For each region, the mean and standard deviation of the three variables were calculated. The variables were compared both spatially and temporally by simple linear regression of AVHRR on TOVS. The bias (AVHRR minus TOVS), standard deviation, and correlation coefficient were calculated for each of the seven regions. In addition, seasonal variability of the mean and standard deviation for each variable was compared using time series plots. A far less detailed yet still illustrative comparison for each variable was done at the global scale for the months of January and August using difference images. For each of the seven regions, a regression of AVHRR on TOVS was performed using monthly values of air and surface temperature and vapor pressure deficit. The bias, standard deviation of the bias, and coefficient of correlation were then calculated. In this analysis, the values in each 1⬚ grid cell rather than spatial averages were used. The bias was defined as the difference between the spatially averaged AVHRR and TOVS estimates, expressed as AVHRR ⫺ TOVS. The reported standard deviation (
) is the standard deviation of this bias; the coefficient of correlation () measures the linear dependence between the two estimates. 1.9

GLOBAL ANALYSIS

Global variation of air temperature, surface radiometric temperature, and vapor pressure deficit retrieved from the AVHRR and the TOVS data sets is shown in Figures 1.2(a) (January 1989), 1.2(b) (August 1989), 1.3(a) (January 1990), and 1.3(b) (August 1990). Black colors correspond to missing and/or flagged data from the AVHRR. The general spatial patterns of the three variables are similar across the two platforms. The air and surface temperatures show the expected seasonal cycles, increasing between January and August of each year for the northern hemisphere and decreasing for the southern hemisphere. Vapor pressure deficit

(a)

AVHRR Air temperature (K) 250

330

AVHRR Surface temperature (K)

250

330

AVHRR Vapor pressure deficit (kPa)

0

TOVS Air temperature (K) 250

330

TOVS Surface temperature (K)

250

0

–10

20

AVHRR–TOVS Surface temperature (K)

330

TOVS Vapor pressure deficit (kPa)

7.5

AVHRR–TOVS Air temperature (K)

–10

20

AVHRR–TOVS Vapor pressure deficit (kPa)

7.5

–1

5

(b)

AVHRR Air temperature (K)

250

330

AVHRR Surface temperature (K)

250

330

AVHRR Vapor pressure deficit (kPa) 0

7.5

TOVS Air temperature (K)

250

330

TOVS Surface temperature (K)

250

330

TOVS Vapor pressure deficit (kPa) 0

7.5

AVHRR–TOVS Air temperature (K)

–10

20

AVHRR–TOVS Surface temperature (K)

–10

20

AVHRR–TOVS Vapor pressure deficit (kPa) –1

5

Figure 1.2. Comparison of surface and air temperatures and vapor pressure deficits retrieved using TOVS and AVHRR for (a) January 1989 and (b) August 1989 (see Color Plate II).

(a)

AVHRR Air temperature (K) 250

330

AVHRR Surface temperature (K) 250

330

AVHRR Vapor pressure deficit (kPa) 0

7.5

TOVS Air temperature (K) 250

330

TOVS Surface temperature (K) 250

330

TOVS Vapor pressure deficit (kPa) 0

7.5

AVHRR–TOVS Air temperature (K) –10

20

AVHRR–TOVS Surface temperature (K) –10

20

AVHRR–TOVS Vapor pressure deficit (kPa) –1

5

(b)

AVHRR Air temperature (K) 250

330

AVHRR Surface temperature (K) 250

330

AVHRR Vapor pressure deficit (kPa) 0

7.5

TOVS Air temperature (K) 250

330

TOVS Surface temperature (K) 250

330

TOVS Vapor pressure deficit (kPa) 0

7.5

AVHRR–TOVS Air temperature (K) –10

20

AVHRR–TOVS Surface temperature (K) –10

20

AVHRR–TOVS Vapor pressure deficit (kPa) –1

5

Figure 1.3. Comparison of surface, air temperatures and vapor pressure deficits retrieved using TOVS and AVHRR for (a) January 1990 and (b) August 1990 (see Color Plate III).

12 Venkat Lakshmi was greater in summer than in winter for all regions, as the seasonal increase in saturation vapor pressure was greater than the seasonal increase in atmospheric vapor pressure. January air temperature (Figures 1.2(a) and 1.3(a)) ranged from 250–270 K for Canada to 270–285 K for most of the United States and Europe, 285–300 K for South America and Africa, 280–300 K for Asia, and 290–330 K for Australia. The surface temperature was a few degrees warmer than the air temperature over all regions. The vapor pressure deficit ranged between 0 and 1 kPa for Canada and the United States, 3 and 4 kPa for South America, 3 and 6 kPa for Africa, 0 and 4 kPa for Asia, and 2 and 7 kPa for Australia. In January, the highest temperatures and vapor pressure deficits occurred in the desert regions of Africa and Australia. The lowest temperatures and vapor pressure deficits occurred in the northern US, Canada, and northern Asia. Air and surface temperatures derived using AVHRR were warmer than the TOVS derived values over most of the global land surface. This was reflected in the AVHRR–TOVS panels for air and surface temperature where most of the global land surface values were positive. Differences ranged from ⫺10 K to ⫹20 K. Approximately 50–60% of the land surface had a difference between 0 and 10 K (green). There were small regions of negative value (TOVS greater than AVHRR) denoted by blue color; these comprise less than 5% of the total land surface area. In mountainous and desert regions the AVHRR temperatures were 10–20 K higher than the TOVS temperatures; this difference was most pronounced in the Himalayan region of Asia. Similarly, vapor pressure deficits calculated using AVHRR data were higher than the corresponding TOVS values over most of the global land surface. The difference between AVHRR and TOVS vapor pressure deficits ranged from ⫺1 kPa to 5 kPa, with the majority (60–70%) of the land surface area having a difference between 0 and 3 kPa (green–yellow). August air temperature (Figures 1.2(b) and 1.3(b)) ranged between 280 and 310 K for North America to 275 and 310 K for South America, 290 and 330 K for Africa, 280 and 310 K for Asia and Europe, and 270 and 300 K for Australia. Surface temperatures follow similar spatial patterns but are approximately 3–5 K higher. The vapor pressure deficit showed a variation between 0 and 6 kPa for North and South America, 2 and 7 kPa for Africa, 1 and 7 kPa for Asia, 0 and 5 kPa for Europe, and 0 and 5 kPa for Australia. The spatial patterns were forced by seasonality and vegetation variations, as were those in January. The AVHRR-derived air and surface temperatures and vapor pressure deficit were again larger than the TOVS values. However, the differences (AVHRR minus TOVS) appeared to follow a seasonal pattern as well. The differences were smaller in August than in January for the southern hemisphere and larger in August than in January for most of the northern hemisphere. This indicates a global seasonal cycle to the differences with the maximum differences in summer and the minimum differences in winter. 1.9.1

Regional regression analysis

Regional analyses were carried out by accounting for each 1⬚ grid value separately and determining the bias between the TOVS and AVHRR values, the standard deviation of the difference, and the correlation between these spatially distributed grid values. Results of the bias presented in Table 1.3 for surface temperature ranged from 4.35 K in Central Asia to 8.55 K in the North America West region. The standard deviation of the surface temperature bias ranged from 5.91 K in Europe to 10.8 K in the North America West region; correlation coefficients varied from 0.7 K for South America to 0.95 K for Central Asia. The high values of the correlation coefficient (⬎0.9 for 5 out of the 7 regions) indicate similarity in seasonal and spatial trends of surface temperature for the AVHRR and TOVS estimates. The bias and standard deviation for surface temperature were 2–3% and 4% of the mean (from Table 1.2). The bias in air temperature was lower than that for the surface temperature for all regions except Australia (Table 1.3). The air temperature bias ranged between 0.95 K for Central Asia

287.2 290.0 298.7 302.4 307.6 275.1 287.2

North America (East) North America (West) South America Africa Australia Central Asia Europe

8.2 9.2 4.1 6.4 8.9 4.8 1.9


298.6 292.8 305.1 306.9 315.7 279.4 292.2

Mean

AVHRR

9.4 7.9 4.4 5.3 6.2 6.5 2.1


286.5 286.1 299.9 300.6 299.6 275.1 287.1

Mean

TOVS

5.7 5.9 1.2 2.2 4.7 3.2 0.9


292.1 289.9 302.8 304.2 312.9 276.6 290.0

Mean

AVHRR

Air temperature (K)

8.4 7.3 3.7 3.9 6.8 5.4 1.6


0.6 1.1 0.64 1.1 2.5 0.4 0.6

Mean

TOVS

0.2 0.6 0.4 0.8 1.1 0.2 0.1




2.3 1.6 2.6 3.2 7.8 1.4 1.4

Mean

AVHRR

Vapor pressure deficit (kPa)

1.7 0.8 0.9 1.5 2.9 0.9 0.3




Bias

4.83 8.55 6.95 4.38 6.04 4.35 5.02

Region

North America (East) North America (West) South America Africa Australia Central Asia Europe

6.77 10.80 7.84 5.96 7.96 7.42 5.91

0.94 0.91 0.70 0.82 0.91 0.95 0.94

3.08 6.31 2.89 3.52 12.40 0.95 2.87

5.20 9.22 4.56 5.12 13.80 5.92 3.63




Bias






Air temperature (K)

Surface temperature (K)

0.95 0.88 0.36 0.58 0.92 0.99 0.97



0.99 1.34 2.02 2.02 5.11 0.91 0.79

Bias

1.29 2.21 2.28 2.48 6.44 1.34 0.94




0.82 0.69 0.34 0.53 0.87 0.66 0.89



Vapor pressure deficit (kPa)

Table 1.3. Comparison of monthly spatially distributed surface skin temperature, surface air temperature, and vapor pressure deficit derived using the TOVS and AVHRR using bias (mean difference), standard deviation (
), and correlation coefficient ().

Mean

TOVS

Surface temperature (K)

Statistics of mean and standard deviation (
) over the study regions averaged over a 2-year (1989–90) period.

Region

Table 1.2.

14 Venkat Lakshmi and 12.4 K for Australia. The standard deviation of the bias varies between 3.63 K for Europe and 13.8 K for Australia. The correlation coefficients were low in South America (0.36 K) and Africa (0.58 K) but high (⬎0.8) for all the other regions. The low value of  in South America emphasizes the lack of variation in the TOVS estimates for this region as opposed to the AVHRR. In Africa, the low value of  is probably due to the different trends in spatial variation seen in the TOVS and AVHRR estimates. The best agreement (low bias and standard deviation of bias, high correlation coefficient) was found in Central Asia. The high value of  in this region (0.99) depicts the strong agreement in the mean air temperature seasonal cycles. The bias and standard deviation expressed as a percentage of the mean air temperature were similar to those for surface temperature at around 2–3% and 4%, respectively. The bias in the vapor pressure deficit ranged from 0.79 kPa for Europe to 5.11 kPa for Australia. The standard deviation of bias lies between 0.94 kPa (Europe) and 6.44 kPa (Australia). The correlation coefficients were low for Africa (0.53), South America (0.34), North America West (0.69), and Central Asia (0.66). The correlation coefficients for the other three regions were much higher, ranging from 0.82 to 0.89. The lower values of
emphasize the relatively poor correlation between the AVHRR and TOVS derived vapor pressure deficits as compared to the agreement in the air and surface temperature estimates. The values of bias expressed as a percentage of mean vapor pressure deficits were much higher than the corresponding values for surface and air temperatures, ranging from 50% to 100%. 1.10

CONCLUSIONS AND DISCUSSIONS ABOUT SATELLITE VALIDATION STUDIES

Estimates of air temperature, surface temperature, and vapor pressure deficit were derived independently from TOVS and AVHRR for seven continental areas using monthly average values for the time period 1989–90. The results show that the bias and standard deviation of the bias between these two estimates vary regionally, with the largest differences in warm and arid areas. The ranges in bias were 4.3–8.5 K for surface temperature, 0.9–12.4 K for air temperature, and 0.7–5.11 kPa for vapor pressure deficit. The AVHRR estimates of all three variables were consistently higher than those derived from TOVS, although the spatial and temporal trends in temperatures were similar across platforms for most regions. The region with the highest overall bias was Australia; if we exclude the results for this region the range in the bias range drops to 0.9–6.3 K for air temperature and 0.7–2.0 kPa for vapor pressure deficit. When the bias was expressed as a percentage of the means (Table 1.2), the bias was very large (50–100%) in vapor pressure deficit but small (2–3%) in air and surface temperature when compared in degrees Kelvin and 10–30% when the temperatures are expressed in degrees centigrade. The results of this study were interpreted over similar spatial and temporal scales, namely continental and monthly. These results should not be considered representative of the local or instantaneous behavior of these variables. However, these results will provide valuable insights into environmental remotely sensed variables when used in conjunction with results of earlier studies (Lakshmi et al., 2001) of TOVS and AVHRR. Prince et al. (1998) determined that air temperature could be retrieved from AVHRR with a root mean squared error of 3.9 K, surface temperature with a root mean squared error of 3.5 K, and vapor pressure deficit with a root mean squared error of 1.09 kPa. These numbers were calculated by comparing AVHRR-derived estimates to ground observations during three field experiments – FIFE, HAPEX-Sahel, and BOREAS. Lakshmi and Susskind (2001) compared the surface temperature, air temperature, and vapor pressure for the same field experiments as Prince et al. (1998) and found that the long-term averages were unbiased but the standard deviation of the biases were 4 K, 3.5 K, and 3.5 mb, respectively. Lakshmi and Susskind (1998) compared TOVS retrievals of air temperature and vapor pressure to monthly observations at

Remote sensing and hydrology 15 four airports around the world (Washington DC, Abilene TX, and Cita and Minsk in the Former Soviet Union). They reported standard deviations of 1.1–2.4 K for air temperature and 0.1–0.2 mb for vapor pressure. Lakshmi et al. (2001) compared AVHRR and TOVS derived air temperatures to observations from a network of surface airway stations in the Arkansas Red River Basin; they reported standard deviations of 3–5 K. All these studies quoted above, show similar differences between satellite and ground data. These differences are expected as comparisons are made between a point (ground data) and a spatially averaged (satellite data) value. This indicates a good agreement between satellite and ground data. The global patterns of the TOVS and AVHRR differences for surface temperature, air temperature and vapor pressure deficit showed a seasonal pattern with maximum differences in the summer hemisphere (January for the southern hemisphere and August for the northern hemisphere). The estimates from the AVHRR had greater spatial standard deviation compared to those from TOVS, as a result of the former’s higher spatial resolution. The seasonal cycles of the mean and spatial standard deviation of these land surface variables exhibited reasonably good agreement with each other. The statistics of the agreement between the two satellite estimates for the different variables fall within the ranges determined by other investigators independent of this study (Prince et al., 1998; Lakshmi and Susskind, 2001). There are major differences between the TOVS and AVHRR fields especially in the vapor pressure deficit. It should be noted that the AVHRR has a better spatial resolution (4 km/GAC aggregated to 1⬚ for comparison purposes) than the TOVS data. The vapor pressure deficit is also the variable that is most difficult to calculate using satellite observations, as it is not a directly observed variable. In addition, the AVHRR uses the DAO analyses to calculate the vapor pressure deficit using the relative humidity. In the case of the other two variables (surface and air temperatures), major differences arise in Australia. This may be due to different specifications of emissivity/surface vegetation for desert/arid climates. Nonetheless, the variables have a good degree of agreement with each other. The surface temperature estimates from the two sensors showed agreement for both the average seasonal cycle and the spatial standard deviation over the continental regions. The air temperature estimates of the two sensors do not agree as well. There were examples of near perfect agreement between the estimates of spatial average seasonal cycle of air temperature (North America, East and West, South America, Africa, Asia, and Europe) and surface temperature (North America West, South America, Central Asia, and Europe). These agreements (and lack of the same) over a period of two years showed that there are some periods and regions of systematic bias in the retrieval algorithm for the sensors (the same systematic bias is unlikely due to different algorithm characteristics). This may be caused by the different treatment in each retrieval algorithm of the land surface effects and the atmosphere. In the case of vapor pressure, there was a close agreement between the two estimates for North America West and Europe for both mean and standard deviation. However, the other regions displayed significant differences. The standard deviations of the variables are different between the two sensors. This indicates that the characterization of spatial variability of the continental areas varies between the sensors. This is due to the differences in the input to the retrieval algorithms of land surface characteristics. This study differed from previous comparisons of satellite and surface measurements in that we have compared much larger spatial areas over a longer time period between satellite estimates of land surface variables. In addition, we do not have ground observations to serve as a reference point. Our results have a greater bias than those taken from point observations. We believe that the actual value lies somewhere in between these two estimates as prior studies of Lakshmi and Susskind (2001) and Prince et al. (1998) show lower biases for TOVS and AVHRR data, respectively. The similarity in spatial and temporal trends between the TOVS and AVHRR estimates in many of the selected regions was comparable and had high correlation coefficients and low bias expressed as a percentage of the mean. Our results indicate that the surface temperature and air temperature agree better than the vapor pressure deficit in these comparisons,

16 Venkat Lakshmi probably because of differences in the sources of near-surface humidity used by the AVHRR (DAO) versus TOVS (estimated). The cause of bias between the two estimates (of any variable) is due less to sensor technology than the retrieval algorithm. Different retrieval algorithms treat land surface effects, vegetation, bare soil, and the emissivity differently. In TOVS, we used preset values of emissivity, whereas in the case of AVHRR a global emissivity map derived from the CERES was used. In addition, the atmospheric correction is different in each of the retrievals. In the TOVS retrievals, the atmospheric correction was done using the simultaneously derived atmospheric water vapor and temperature profiles. In the AVHRR retrieval, this was done using DAO reanalyses. Another cause of differences between the two sensor estimates was related to the field of view (FOV) of the sensors. The HIRS2 has a FOV of 20 km and the AVHRR is on the order of 1 km. This results in a large difference in the type of land surface seen by the sensors, which in turn affects the retrieval. This study provides a good starting point for future work in merging the AVHRR and TOVS data sets, which would complement the high spatial resolution of the AVHRR with the high temporal resolution of TOVS. The 4 km spatial resolution of AVHRR can be used to disaggregate the four times of day 1⬚ data from the TOVS sensor. We can thereby generate global, four times of day, 4 km surface and air temperatures and vapor pressure deficit data. This synthesized data will provide a powerful addition to the global meteorological observation network.

1.11 1.11.1

HYDROLOGICAL MODELING Introduction

We have used a macroscale hydrological model (Variable Infiltration Capacity three-Layer (VIC-3L)) implemented to understand the water and energy balance for over a period of 50-years (1950–99) for the Upper Mississippi River Basin and also evaluated in detail the performance in the complete spectrum of droughts and floods. Simulations have been carried out between January 1950 and December 1999 at daily time step and 1/8⬚ spatial resolution for the water budget and at hourly time step and 1⬚ spatial resolution for the energy balance. The water balance simulations show seasonal variability of soil moisture, and, more importantly, the variability of soil moisture during the period of flood (1993) and drought (1988). Model-simulated soil moistures for the state of Illinois were validated with measured soil moisture data from Illinois State Water Survey. The model-simulated streamflows were compared with the USGS measured stream-gauge observations at the basin outlet at Mississippi River at Grafton, IL and the Illinois River at Valley City, IL, which showed a 15% relative bias. Model validations for the hourly surface temperature were performed for a period of 20 years (1980–99) by comparison with TOVS surface temperature. These comparisons show a good correlation coefficient of around 0.8 with a low bias of 1–2 K and a root mean squared difference (RMSD) of 6–7 K. The model simulated deep soil moistures have been analyzed to understand the spatial and temporal variability of droughts and flood. The study of land surface schemes has evolved over the years from simple linear parametric relationships to recent successes in better understanding the underlying physical processes of balancing the various components of the hydrological cycle. These developments have resulted in a distributed approach in modeling the response of the precipitation events on watersheds and basins to simulate the various components of the hydrological cycle. The land surface hydrological model used in the study is Variable Infiltration Capacity – three Layer (VIC-3L) model (Liang et al., 1998), a macroscale hydrological model that carries out complete water and energy balance on a gridcell basis. The VIC-3L model is based on a three layer Soil Vegetation Atmosphere Transfer Scheme to model different surface conditions. The model has been successfully validated and implemented on a variety of climatic conditions

Remote sensing and hydrology 17 and basins worldwide, including the Columbia and Delaware Rivers (Nijssen et al., 1997), the Arkansas and Red Rivers (Abdulla et al., 1996), the Weser River, Germany (Lohmann et al., 1998), Upper Mississippi River Basin, and the Mississippi River Basin (Maurer et al., 2002). The model has also been used to simulate runoff and soil moistures at continental scales (Schnur and Lettenmaier, 1997). Liang et al. (1994) describe in detail the formulation of the two-layered version of the model. 1.12

MODEL DESCRIPTION, DATA AND STUDY AREA

1.12.1 Model description The current implementation of the model consists of three soil layers – a top layer of 10 cm thickness and two bottom layers around 30 cm and 100 cm thickness each. We have utilized a predetermined thickness for each of the soil layers. The top layer characterizes dynamic behavior of soil columns responding to precipitation events and the bottom layer represents storm-soil moisture behavior responding to precipitation events only after the top two layers are saturated. The last layer (100 cm thick) responds to the long-seasonal time scales and displays greater inertia than the top layers. One of the main distinguishing features of the model is its sub grid variability of soil moisture (Wood et al., 1992). The model incorporates the various surface conditions to be described by n ⫽ 1, 2, 3, . . . , N types of vegetation as well the (N ⫹ 1)th type, the bare soil type. Each surface’s LAI, canopy resistance, root fraction depth, and also distinct soil moisture characteristics are defined during each time step. Also all the calculations of infiltration, base flow, and runoff are carried out for each of N ⫹ 1 land cover types. Figure 1.4 depicts the various components of the hydrological cycle in model. The amount of infiltration is controlled by a variable infiltration curve, which is based on the available moisture content of the top two layers. The water that cannot infiltrate is removed from a grid cell as runoff. The infiltrating moisture fills the top layer and then infiltrates into the lower layers. The bottom layer loses water by both transpiration and base flow generated using

n=2

n=N

n=1 n=4

P E1

n=N+1 Bare soil

Ec Et

Canopy Qd

Layer 1(0–10 cm) Layer 1(10–40 cm)

W 1c

Layer 1(40–140 cm)

W 2c Q12

Figure 1.4.

Qb

Land-cover (vegetation) classes Et = Evapotranspiration Ec = Canopy evaporation E1 = baresoil evaporation R = Runoff B = Baseflow Qd = Direct runoff Qb = Subsurface flow Q1,2 = Gravity flows layer 1 to 2 Q2,3 = Gravity flows layer 2 to 3 i = Infiltration P = Precipitation W1,W2,W3 = Water content in respective layers

Schematic representation of the variable infiltration capacity model.

18 Venkat Lakshmi an empirical relationship based on the soil moisture of the bottom layer (Liang et al., 1994). The model is implemented using a grid mesh for the entire basin. Evaporation, runoff, and base flow are predicted independently for each grid cell. Stream-flow is then generated at specified locations by routing runoff and base flow from each grid cell using the linearized Saint-Venant method as discussed by Lohmann et al. (1998). The surface temperature for each hourly time step is initialized from the ground heat flux and is iteratively determined by solving for the energy balance and by minimizing the residuals. Due to the combined solution of water and energy budgets by the model, the surface temperature estimates are dependant on the soil and vegetation types and also on the external forcing such as precipitation and air temperatures. 1.12.2

Model parameters, forcing, observations, and application

Soil data for the continental United States are obtained from Penn State’s Earth System Science Center’s State Soil Geographic Database (STATSGO) data (Miller and White, 1998) at 30 arc second resolution. For each of the available layers, most of the parameters including saturated hydraulic conductivity, porosity, soil moisture at field capacity, and wilting point are obtained based on the soil texture classes (Rawls et al., 1993). The vegetation classification data are based on University of Maryland (UMD) classification system. Vegetation parameters include the LAI and the fraction of vegetation cover within each 1/8th pixel as described by Maurer et al. (2001). The classification has 13 major classes and the data are time invariant. Terrain characteristics include the Digital Elevation model (DEM), stream network, and the basin boundaries. Meteorological data include daily precipitation, air temperature, wind speed, humidity, incoming shortwave and longwave radiation.

1.13

RESULTS OF MODELING STUDIES

1.13.1 Streamflow and soil moisture comparisons Previous applications of the model include a comparison of streamflows for the basin outlet of the Mississippi River at Grafton, IL and the Illinois River at Valley City, IL. VIC-3L was used to analyze the Upper Mississippi River Basin over a period of 50 years (1950–99). The threelayer VIC model simulations for the water and the energy balance were carried out using daily forcing – precipitation, maximum temperature, minimum temperature, and wind speed. The water balance simulations were performed at daily time step at 1/8⬚ spatial resolution and the simulations for the energy budget were carried out at hourly time step and 1⬚ spatial resolution. Figure 1.5 shows a reasonably good monthly streamflow comparison between the measured and simulated streamflow. The daily streamflow for Mississippi River at Grafton, IL between USGS measured discharge and model simulated streamflow for the period 1950–99 shows a reasonable R2 value of 0.74 and a bias of 32,438 cfs. The percentage difference of the mean flow for the bias (bias/mean flow) is approximately 15%. Figure 1.6 depicts the individual monthly average soil moisture comparisons for the different soil layers with the Illinois State Water Survey Board measurements for a period of 19 years (1981–99). The differences can be attributed to the fact that the model simulated soil moistures are average over a cell of approximately 12.5 km ⫻ 12.5 km area whereas those from the Illinois Water Survey data are point measurements. The R2 values improved with the depth of the layers from about 0.3 for layer 1 (0–10 cm) to 0.6 for the aggregated layer (0–140 cm). The comparisons from the integrated soil water for the simulated 0–140 cm layer (Figure 1.6) exhibit the same seasonal pattern as the observations. This highlights our ability to model water movement and conserve moisture.

Remote sensing and hydrology 19

6.00E + 05

Simulated Measured

Discharge (cfs)

5.00E + 05 4.00E + 05 3.00E + 05 2.00E + 05 1.00E + 05 0.00E + 00 Oct-54

Mar-60

Sep-65

Mar-71

Aug-76

Feb-82

Aug-87

Jan-93

Jul-98 Jan-04

Date (month-year)

Figure 1.5.

Monthly mean streamflow at Mississippi River at Grafton, IL, 1950–99.

600

Model Observation

Aggregated layer (0–140 cm)

Soil moisture (mm)

550 500 450 400 350 300 250 200 May-79

Feb-82

Nov-84

Aug-87

May-90

Jan-93

Oct-95

Jul-98

Apr-01

Date (month-year)

Figure 1.6.

1.13.2

Illinois State averaged monthly soil moisture comparison 1981–99.

Surface temperature comparisons with TOVS

The model simulated surface temperatures were compared with TOVS surface temperature data (Lakshmi and Susskind, 2001) for the basin for a period of 20 years (1980–99) for both the morning and afternoon overpasses of the satellite. The time of observation for each pixel is different depending on the latitude and distance from nadir. The exact time of each observation is included in the data. The VIC-3L model simulated surface temperature for the individual pixels were matched with the exact time of TOVS observations. If the TOVS data were not available, no comparisons were made for that particular record or pixel. Daily surface temperature comparison between the TOVS and VIC-3L simulated results for all the years from 1980 to 1999 for both morning and afternoon overpasses are given in Table 1.4. The R2 values range from about 0.72 to 0.89 and the average bias is around 1.3 K. The RMSD of about 6.77 K was

20 Venkat Lakshmi Table 1.4.

VIC vs TOVS surface temperature comparison 1981–99.

Year

Approximate overpass time (am/pm)

Best fit line

R2

Bias (K)

Root mean squared difference (K)

1981 1981 1982 1982 1983 1983 1984 1984 1985 1985 1986 1986 1987 1987 1988 1988 1989 1989 1990 1990 1991 1991 1992 1992 1993 1993 1994 1994 1995 1995 1996 1996 1997 1997 1998 1998 1999 1999

8 am 8 pm 8 am 8 pm 3 am 3 pm 3 am 3 pm 8 am 8 pm 3 am 3 pm 8 am 8 pm 8 am 8 pm 8 am 8 pm 8 am 7 pm 3 am 3 pm 3 am 2 pm 3 am 2 pm 8 am 8 pm 8 am 8 pm 8 am 8 pm 8 am 8 pm 8 am 8 pm 3 am 3 pm

y ⫽ 0.82x ⫹ 51.61 y ⫽ 0.90x ⫹ 31.39 y ⫽ 0.75x ⫹ 72.48 y ⫽ 0.86x ⫹ 43.89 y ⫽ 0.73x ⫹ 76.45 y ⫽ 0.84x ⫹ 48.56 y ⫽ 0.73x ⫹ 75.56 y ⫽ 0.87x ⫹ 38.65 y ⫽ 0.75x ⫹ 71.79 y ⫽ 0.84x ⫹ 50.77 y ⫽ 0.79x ⫹ 59.67 y ⫽ 0.86x ⫹ 44.04 y ⫽ 0.85x ⫹ 42.00 y ⫽ 0.92x ⫹ 27.284 y ⫽ 0.79x ⫹ 59.85 y ⫽ 0.89x ⫹ 31.81 y ⫽ 0.83x ⫹ 48.66 y ⫽ 0.92x ⫹ 24.29 y ⫽ 0.85x ⫹ 41.68 y ⫽ 0.89x ⫹ 35.76 y ⫽ 0.79x ⫹ 60.06 y ⫽ 0.83x ⫹ 53.02 y ⫽ 0.74x ⫹ 73.52 y ⫽ 0.75x ⫹ 73.29 y ⫽ 0.76x ⫹ 67.97 y ⫽ 0.84x ⫹ 47.57 y ⫽ 0.74x ⫹ 72.54 y ⫽ 0.88x ⫹ 37.09 y ⫽ 0.84x ⫹ 46.75 y ⫽ 0.89x ⫹ 34.44 y ⫽ 0.72x ⫹ 78.46 y ⫽ 0.80x ⫹ 59.50 y ⫽ 0.73x ⫹ 75.54 y ⫽ 0.83x ⫹ 50.58 y ⫽ 0.81x ⫹ 58.24 y ⫽ 0.83x ⫹ 60.23 y ⫽ 0.81x ⫹ 53.82 y ⫽ 0.80x ⫹ 60.80

0.72 0.85 0.73 0.83 0.82 0.82 0.79 0.78 0.83 0.78 0.81 0.74 0.77 0.87 0.82 0.88 0.78 0.86 0.75 0.81 0.82 0.74 0.77 0.68 0.80 0.69 0.75 0.83 0.78 0.84 0.75 0.80 0.76 0.80 0.82 0.79 0.82 0.73

1.01 4.38 2.29 4.17 2.12 4.07 1.22 2.98 1.86 4.62 0.54 4.39 0.08 3.83 0.14 3.45 0.29 3.21 0.09 3.33 1.45 3.36 1.85 2.93 1.83 2.42 1.16 3.13 1.00 2.62 1.33 2.42 0.83 2.11 0.79 3.16 0.49 2.17

6.85 6.45 7.57 6.91 5.60 8.14 5.47 7.69 5.54 8.92 4.94 8.96 6.04 5.92 6.46 6.11 6.56 6.14 6.22 6.12 5.05 8.86 5.12 8.76 5.40 8.30 6.96 6.26 6.29 6.09 7.17 6.92 6.28 6.55 5.69 7.89 4.40 7.98

obtained from the study. The statistics for the 20-years comparison (shown in Table 1.4) indicates a consistent model performance in simulation of land surface temperature. 1.14

MODEL CHARACTERIZATION OF EXTREME EVENTS – DROUGHTS AND FLOODS

The VIC-3L model simulated soil moisture for the Upper Mississippi River Basin were studied in detail to characterize the extreme events such as the Midwestern drought during summer 1988 and the Upper Mississippi flood during July 1993. The monthly average three-layer aggregated

Remote sensing and hydrology 21 Table 1.5.

January February March April May June July August September October November December Average

Upper Mississippi River Basin averaged monthly soil moisture (0–140 mm). 1950–99

1988

1988 ⫺ (1950–99)

1993

1993 ⫺ (1950–99)

404 408 428 452 450 427 391 357 353 373 396 406 404

398 406 426 433 405 332 265 252 263 289 325 355 346

⫺6 ⫺2 ⫺2 ⫺19 ⫺45 ⫺95 ⫺126 ⫺105 ⫺90 ⫺84 ⫺71 ⫺51 ⫺58

436 433 444 484 472 479 485 455 459 454 450 453 458.80

32 25 16 32 22 52 94 98 106 81 54 47 55

soil moisture for the years 1988 and 1993 along with the 50-year average is summarized in Table 1.5. The annual average soil moisture of the basin during the drought year 1988 was about 346 mm as compared to 458 mm during the flood year 1993 and the 50-year average was about 403 mm. During the month of July, the deficit in soil moisture in the drought year of 1988 when compared to the 50-year average was as much as 125 mm, whereas during the flood year (1993) for the same month the aggregated soil moisture for the basin is about 95 mm higher than the average for the 50-year period. 1.15

CONCLUSION AND DISCUSSION ABOUT HYDROLOGICAL MODELING

In this section, we use a hydrological model (VIC-3L), and apply it to the Upper Mississippi River Basin for the time period 1950–99. The model outputs are validated using observations of discharge, soil moisture, and surface temperature. We study in detail the range of land–atmosphere conditions from droughts to floods. In particular, the distribution of soil moisture at all depths (0–10 cm, 10–40 cm, 40–140 cm) for both these extreme conditions is examined in detail and evaluate the model characterization in depicting extreme conditions like droughts and floods. In this section we have studied the temporal dynamics exhibited by the deep layer soil moisture. Model simulated streamflows were validated with comparisons with USGS gauging station streamflow with reasonable accuracy. The streamflow simulations were seen to be within acceptable limits of around 15% for monthly comparisons and for daily comparisons the simulations were successful in predicting the discharge peaks and were within the range of about 27% over a period of 50 years. The work of Maurer et al. (2002) show a percent root mean square error for streamflow at the outlet (for Upper Mississippi Basin) as 25.6% and percentage bias of 13.8% with comparisons carried out for a 10-year period. Our corresponding calculations show reasonable comparisons with these values of 33% and 13.5%, respectively for RMSE and bias over a 50-year period. The actual differences could be attributed in part to the period of comparison (10 years versus 50 years) chosen. Sensitivity studies of spatial resolution on streamflow simulations were carried out by Haddeland et al. (2002), indicated a greater degree of dependence. Suggestions have been made for the use of diurnal cycles of streamflow in understanding the various hydrological components, namely, snowmelt, evapotranspiration, and infiltration. The study including seasonal changes in diurnal variations would

22 Venkat Lakshmi provide further insight in evaluating the various hydrological components described. Studies by Maurer et al. (2001, 2002) and Cherkauer and Lettenmaier (1999) over the Upper Mississippi River Basin provides similarly reasonable success in model performance in evaluating the various components of the hydrological cycle. The model simulated soil moisture was compared with Illinois Soil Water Survey Board soil moisture measurements. The seasonal patterns and the patterns in soil moisture variations are captured and the performance is reasonable for the aggregated layer (0–140 cm). With fewer number of point measurements over the state of Illinois, difficulties were encountered in performing a more detailed validation study. The simulated surface temperatures were validated by comparison with satellite observations – TOVS surface temperatures over a period of 20 years (1980–99), with reasonable agreement. The simulated daily values were within about 10–15% for the morning overpass and were about 15–20% for the afternoon overpass. The performance during the afternoon overpass is significant in terms of the peaks in energy flux exchange between the surface and atmosphere during the period. The surface temperature simulations at hourly time step were comparable with similar works in the Red Arkansas River Basin by Rhoads et al. (2001), where the comparisons were found to be within 25%. In their study (Rhoads et al., 2001) over Red Arkansas at 1⬚ spatial resolution and three hour time-step model time step yielded a regression coefficient of 0.76, bias 3.16 K, and RMSE of 8.33 K for the year 1985 and also showed similar trends in spatial distributions. REFERENCES Abdulla, F.A., Lettenmaier, D.P., Wood, E.F., and Smith, J.A. (1996) Application of a macroscale hydrologic model to estimate the water balance of the Arkansas-Red River Basin. Journal of Geophysical Research, 101(D3), 7449–7459, March 20. Agbu, P.A. and James, M.E. (1994) The NOAA/NASA Pathfinder AVHRR Land Data Set User’s Manual Version 3.0. Greenbelt, MD: Goddard Distributed Active Archive Center, NASA Goddard Space Flight Center. Becker, F. and Li, Z. (1990) Towards a local split window method over land surfaces. International Journal of Remote Sensing, 11, 369–393. Cherkauer, K.A. and Lettenmaier, D.P. (1999) Hydrological effects of frozen soils in the Upper Mississippi River basin. Journal of Geophysical Research, 104(D16), 19599–19610. Gupta, S.K., Wilber, A.C., and Kratz, D.P. (1999) Surface emissivity maps for satellite retrieval of longwave radiation budget. Tenth Conference on Atmospheric Radiation. Madison, WI: American Meteorological Society. Haddeland, I., Matheussen, B.V., and Lettenmaier, D.P. (2002) Influence of spatial resolution on simulated streamflow in a macroscale hydrological model. Water Resources Research, American Geophysical Union (in press). Lakshmi, V. and Susskind, J. (1998) Determination of land surface skin temperatures and surface air temperature and humidity from TOVS HIRS2/MSU data. Advances in Space Research, 22(5), 629–636. Lakshmi, V. and Susskind, J. (2001) Validation of TOVS land surface parameters using ground observations. Journal of Geophysical Research, 105(D2), 2179–2190, January 27. Lakshmi, V., Czajkowski, K.P., Dubayah, R.O., and Susskind, J. (2001) Potential of global air temperature mapping using TOVS and AVHRR. International Journal of Remote Sensing, 22(4), 643–662. Liang, X., Lettenmaier, D.P., Wood, E.F., and Burges, S.J. (1994) A simple hydrologically based model of land surface water and energy fluxes for GSMs. Journal of Geophysical Research, 99(D7), 14415–14428. Liang, X., Wood, E.F., and Lettenmaier, D.P. (1998) Modeling of ground heat flux in land surface parameterization scheme. Journal of Geophysical Research, 104(D8), 9581–9600. Lohmann, D., Raschke, E., Nijssen, B., and Lettenmaier, D.P. (1998) Regional scale hydrology: II. Application of the VIC-2L model to the Weser River Germany. Hydrological Sciences-Journal-des Sciences Hydrologiques, 43(1), February.

Remote sensing and hydrology 23 Maurer, E.P., O’Donnell, G.M., Lettenmaier, D.P., and Rhoads, J.O. (2001) Evaluation of the land surface water budget in NCEP/NCAR and NCEP/DOE reanalyses using an off-line hydrologic model. Journal of Geophysical Research, 106(D16), 17841–17862. Maurer, E.P., Wood, A.W., Adam, J.C., Lettenmaier, D.P., and Nijssen, B. (2002) A long-term hydrologically-based data set of land surface fluxes and states for the continental United States. Journal of Climate (submitted). Miller, D.A. and White, R.A. (1998) A conterminous United States multi-layer soil characteristics dataset for regional climate and hydrology modeling. Earth Interactions, 2, 1–15. Nijssen, B. et al. (1997) Streamflow simulation for contenental-scale river basins. Water Resources Research, 33(4), 711–724, April. Prihodko, L. and Goward, S. (1997) Estimation of air temperature from remotely sensed observations. Remote Sensing of Environment, 60, 335–346. Prince, S.D. and Goward, S.N. (1995) Global primary production: a remote sensing approach. Journal of Biogeography, 22, 2829–2849. Prince, S.D. et al. (1998) Inference of surface and air temperature, atmospheric precipitable water and vapor pressure deficit using AVHRR satellite observations: validation of algorithms. Journal of Hydrology, 212–213, 230–250. Rawls, W.J., Brakensiek, D.L., and Logsdon, S.D. (1993) Predicting saturated hydraulic conductivity utilizing fractal principles. Soil Science Society of America Journal, 57, 1193–1197. Rhoads, J., Dubayah, R., Lettenmaier, D.P., O’Donnell, G., and Lakshmi, V. (2001) Validation of land surface models using satellite-derived surface temperature. Journal of Geophysical Research, 106(D17), 20085–20099, September 16. Schnur, R. and Lettenmaier, D.P. (1997) A global gridded data set of soil moisture for use in general circulation models. Poster Presented at the 13th Conference on Hydrology, 77th AMS Annual Meeting, Long Beach, CA, February 7. Schubert, S., Rood, R., and Pfaendtner, J. (1993) An assimilated data set for earth science applications. Bulletin of the American Meteorological Society, 74(12). Susskind, J., Rosenfield, J., and Reuter, D. (1984) Remote sensing of weather and climate parameters from HIRS2/MSU on Tiros-N. Journal of Geophysical Research, 89(D3), 4677–4697. Susskind, J., Piraino, P., Rokke, L., Iredell, L., and Mehta, A. (1997) Characteristics of the TOVS pathfinder path a data set. Bulletin of the American Meteorological Society, 78(7), 1449–1472. Tucker, C.J. (1979) Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment, 8, 127–150. Wood, E.F., Lettenmaier, D.P., and Zartarian, V.G. (1992) A land-surface hydrology parameterization with sub-grid variability for general circulation models. Journal of Geophysical Research, 97, 2717–2728. Xiang, X. and Smith, E.A. (1997) Feasibility of simultaneous surface temperature-emissivity retrieval using SSM/I measurements from HAPEX-Sahel. Journal of Hydrology, 188–189, 330–360.

CHAPTER 2

Hydrologic Data Assimilation Jeffrey P. Walker Department of Civil and Environmental Engineering, The University of Melbourne, Australia Paul R. Houser Hydrological Sciences Branch, NASA Goddard Space Flight Center, Greenbelt, MD, USA

2.1

INTRODUCTION

Earth observing satellites have revolutionized our understanding and prediction of the Earth system over the last 30 years, particularly in the meteorologic and oceanographic sciences. However, historically remote sensing data has not been widely used in hydrology. This can be attributed to (1) a lack of dedicated hydrologic remote sensing instruments, (2) inadequate retrieval algorithms for deriving global hydrologic information from remote sensing observations, (3) a lack of suitable distributed hydrologic models for digesting remote sensing information, and (4) an absence of techniques to objectively improve and constrain hydrologic model predictions using remote sensing data. Three ways that remote sensing observations have been used in distributed hydrologic models are (1) as parametric input data including soil and land cover properties, (2) as initial condition data, such as initial snow water storage, and (3) as timevarying hydrologic state data such as soil moisture content to constrain model predictions. This chapter focuses on the latter. The historic lack of hydrologic missions and observations has been the result of an emphasis on meteorologic and oceanographic missions and applications due to the large scientific and operational communities that drive those fields. However, significant progress has been made over the past decade on defining hydrologically relevant remote sensing observations through focused ground and airborne field studies. Gradually, satellite-based hydrologic data are becoming increasingly available, though little progress has been made in understanding their observation error. Land surface skin temperature and snow cover data have been available for many years, and satellite precipitation data are becoming available at increasing space and time resolutions. In addition, land cover and land use maps, vegetation parameters (albedo, leaf area index, and greenness), and snow water equivalent data of increasing sophistication are becoming available from a number of sensors. Novel observations such as saturated fraction and changes in soil moisture, evapotranspiration, water level and velocity (i.e. runoff ), and changes in total terrestrial water storage are also under development. Further, near-surface soil moisture, a parameter shown to play a critical role in weather, climate, agriculture, flood, and drought processes, is currently available from non-ideal sensor configuration observations. Moreover, two missions targeted at measuring near-surface soil moisture with ideal sensor configuration are expected before the end of the decade. Though remote sensing can make spatially comprehensive measurements of various components of the hydrologic system, it cannot provide information on the entire system, and the

26 Jeffrey P. Walker and Paul R. Houser

Remote sensing satellite

Logger

Near-surface moisture skin temperture etc. Land surface model u(z) = f [qs, Dc(u), c(u)]

Stream “gauge”

Figure 2.1.

Point measurement

Schematic of the hydrologic data assimilation challenge (see Color Plate IV).

measurements represent only a snapshot in time. Land surface hydrology process models may be used to predict the temporal and spatial hydrologic system variations, but these predictions are often poor owing to model initialization, parameter and forcing errors, and inadequate model physics and/or resolution. Figure 2.1 illustrates the hydrologic data assimilation challenge to optimally merge the spatially comprehensive but limited remote sensing observations with the complete but typically poor predictions of a hydrologic model to yield the best possible hydrologic system state estimation, and utilize limited point measurements to calibrate the model(s) and validate the assimilation results. While hydrologic data assimilation is still very much in its infancy, a few hydrologic models have been developed that can use remotely sensed observations. The key is that the remote sensing observations of interest can be directly related to a prognostic state(s) of the hydrologic model. Figure 2.2 demonstrates how satellite observations of the near-surface soil moisture content may be used to constrain the hydrologic model soil moisture prediction using state-ofthe-art hydrologic data assimilation techniques (Walker et al., 2003). This example uses actual space-borne near-surface soil moisture observations from a historic satellite record in a data assimilation framework, and highlights the significant benefit of using these techniques. However, quantifying the hydrologic model prediction improvement by assimilating remote sensing data requires targeted field campaigns, and such data are lacking for these historic satellite records. Because of its importance, and our increasing ability to observe relevant hydrologic information remotely, it is expected that the amount of hydrologic remote sensing data will grow exponentially over the next decade. However, its usefulness will be limited by our ability to analyze and integrate diverse hydrologic information with hydrologic models. Quantifying hydrologic process variability will require innovative interpretation of potentially large hydrologic observation volumes owing to observation type, scale, and error disparities (Table 2.1). The effect of variations in instrument type, placement, calibration, and accuracy of both remote sensing and in-situ hydrologic observations must also be quantified. The complexities of future hydrologic observation scenarios will require systematic methods to organize and comprehend this information. Therefore, a comprehensive hydrologic data assimilation framework will be a critical component of future hydrologic observation and modeling systems.

Model root zone soil moisture (v/v) from 11 to 20 June 1983 125 155 185 215 245 275 305 335 365 395 425 455 110E 115E 120E 125E 130E 135E 140E 145E 150E 155E 160E 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Assimilated root zone soil moisture (v/v) from 11 to 20 June 1983 125 155 185 215 245 275 305 335 365 395 425 455 110E 115E 120E 125E 130E 135E 140E 145E 150E 155E 160E 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Figure 2.2. Satellite observations of near-surface soil moisture content made by the scanning multifrequency microwave radiometer (SMMR) are used to constrain hydrologic model predictions of soil moisture throughout the root zone using data assimilation (see Color Plate V).

28 Jeffrey P. Walker and Paul R. Houser Table 2.1.

Characteristics of hydrologic observations potentially available within the next decade.

Hydrologic quantity Precipitation

Surface soil moisture

Remotesensing technique Thermal infrared Passive microwave Active microwave Passive microwave

Time scale Hourly 1 day 15 days 3 hr Daily 1–3 days

Active microwave

3 days 30 days

Surface skin temperature

Thermal infrared

Snow cover

Visible/ thermal infrared

1 hr 1 day 15 days 1 hr 1 day 15 days

Snow water equivalent

Passive microwave Active microwave

Water level/ velocity

Total water storage changes Evaporation

2.2

1–3 days 30 days

Laser

10 days

Radar

30 days

Gravity changes

30 days

Thermal infrared

1 hr 1 day 15 days

Space scale

Accuracy considerations

4 km 1 km 60 m 10 km

Tropical convective clouds only

4 km 1 km 60 m

Significant assumptions

Example sensors GOES MODIS, AVHRR Landsat, ASTER TRMM, SSMI, AMSR-E, GPM TRMM, GPM

Land calibration problems 10 m Land calibration problems 25–50 km Limited to sparse AMSR-E, SMOS, vegetation, low Hydros topographic relief 3 km Significant noise from ERS, JERS, 10 m vegetation and RadarSat roughness 4 km Soil/vegetation average, GOES 1 km cloud contamination MODIS, AVHRR 60 m Landsat, ASTER 4 km Cloud contamination, GOES 500 m– vegetation masking, MODIS, AVHRR 1 km bright soil problems Landsat, ASTER 30–60 m 10 km Limited depth AMSR-E penetration 100 m Limited spatial SnoSat or CLPP coverage (proposed missions) 100 m Cloud penetration ICESAT problems 1 km Limited to large TOPEX/POSEIDON rivers 1000 km Bulk water storage GRACE change GOES MODIS, AVHRR Landsat, ASTER

HISTORY OF HYDROLOGIC DATA ASSIMILATION

Charney et al. (1969) first suggested combining current and past data in an explicit dynamic model using the model’s prognostic equations to provide time continuity and dynamic coupling among the fields. This concept has evolved into a family of techniques known as data assimilation. In essence, hydrologic data assimilation aims to utilize both our hydrologic process knowledge, as embodied in a hydrologic model, and information that can be gained from observations. Both, model predictions and observations, are imperfect and we wish to use both synergistically to obtain a more accurate result. Moreover, both contain different kinds of

Hydrologic data assimilation 29 information that when used together provide an accuracy level that cannot be obtained when used individually. For example, a hydrologic model provides both spatial and temporal near-surface and root zone soil moisture information at the model resolution including errors resulting from inadequate model physics, parameters, and forcing data. On the other hand, remote sensing observations contain near-surface soil moisture information at an instant in time but do not give the temporal variation or the root zone moisture content. While the remote sensing observations can be used as initialization input for models or as independent evaluation, providing we use a hydrologic model that has been adapted to use remote sensing data as input, we can use the hydrologic model predictions and remote sensing observations together to keep the simulation on track through data assimilation (Kostov and Jackson, 1993). Moreover, large errors in nearsurface soil moisture content prediction are unavoidable because of its highly dynamic nature. Thus, when measured soil moisture data are available their use to constrain the simulated data should improve the overall estimation of the soil moisture profile. However, this expectation is based on the assumption that measurement errors are smaller than simulation errors (Arya et al., 1983). Data assimilation techniques were pioneered by meteorologists (Daley, 1991) and have been used very successfully to improve operational weather forecasts for decades. Data assimilation has also been widely used in oceanography (Bennett, 1992) for improving ocean dynamics prediction. However, hydrologic data assimilation has just a few case studies demonstrating its utility. Fortunately, we have been able to jumpstart hydrologic data assimilation by building on knowledge derived from the meteorologic and oceanographic data assimilation experience, with significant advancements being made over the past decade. One of the primary areas of hydrologic data assimilation application has been soil moisture content. Other observation types such as surface temperature, snow, terrestrial water storage, and streamflow have been used only in more recent applications. The study by Jackson et al. (1981) was among the first to update soil moisture predictions using near-surface soil moisture observations. In this application, the soil moisture values in both layers of the United States Department of Agriculture Hydrograph Laboratory model were substituted with observed nearsurface soil moisture observations as they became available. The model’s performance improvement was evaluated by annual runoff values. Ottlé and Vidal-Madjar (1994) used a similar approach but with the assimilation of thermal infrared derived near-surface soil moisture content. Another early study based on the direct insertion assimilation method was that of Bernard et al. (1981). Here synthetic observations of near-surface soil moisture content were used to specify the surface boundary condition of a classical one-dimensional soil water diffusion model, in order to estimate the surface flux. They found that large soil moisture content variations resulting from rainy periods required special handling of the upper boundary condition. Prevot et al. (1984) repeated this study with real observations and a similar approach was used by Bruckler and Witono (1989). A more popular approach for the improved estimation of land surface fluxes has been the assimilation of screen-level measurements of relative humidity and temperature (Bouttier et al., 1993; Viterbo and Beljaars, 1995). To date only one study has explored the assimilation of remotely sensed land surface flux observations (Schuurmans et al., 2003). The first known study to use an “optimal” assimilation approach is that of Milly (1986). In this study, a Kalman filter (KF – a statistical assimilation approach) was used to update a simple linear reservoir model with near-surface soil moisture observations. It was not until Entekhabi et al. (1994) that this approach was extended, when synthetically derived vertical and horizontal polarized passive microwave and thermal infrared observations were assimilated into a one-dimensional soil moisture and temperature diffusion model using KF. This synthetic study was further extended by Walker et al. (2001a) to more realistic observation times and Walker et al. (2001b) to a field application. Since then there has been a plethora of one-dimensional KF and variational assimilation studies.

30

Jeffrey P. Walker and Paul R. Houser

Georgakakos and Baumer (1996) were one of the first to use the KF to update a hydrologic basin model with near-surface soil moisture measurements. Results showed that even when the observations carried substantial measurement errors, estimation of soil moisture profiles and total soil moisture storage was possible with an error that was smaller than that achieved without the use of remotely sensed data. Walker et al. (2002) were also among the first to use a three-dimensional KF based assimilation in a small catchment distributed hydrologic model. Houser et al. (1998) was the first detailed study of several alternative assimilation approaches including direct insertion, statistical correction, Newtonian nudging, and optimal interpolation. Both the Newtonian nudging and optimal interpolation approaches, pathologic cases of KF, showed the greatest improvement. 2.3

SUMMARY OF DATA ASSIMILATION

The data assimilation challenge is that given a (noisy) model of the system dynamics the best estimates of system states X from (noisy) observations Z are to be found. Most current approaches to this problem are derived from either the direct observer (i.e. KF) or dynamic observer (i.e. variational through time) techniques. Figure 2.3 illustrates schematically the key (a)

State value

x

x x x x

Observation Model Background x Analysis Time (b)

Window 1

Window 2

x

State value

x x

x x Observation Model Background x Analysis Time

Figure 2.3.

Schematic of the (a) direct observer and (b) dynamic observer assimilation approaches.

Hydrologic data assimilation 31 Table 2.2.

Commonly used data assimilation terminology.

State Prognostic Diagnostic Observation Covariance matrix Prediction Update Background Analysis Innovation Gain matrix Tangent linear model Adjoint

Condition of a physical system, that is, soil moisture A model state required to propagate the model forward in time A model state/flux diagnosed from the prognostic states – not required to propagate the model Measurement of a model diagnostic or prognostic Describes the standard deviations and correlations Model estimate of states or covariances Correction to a model prediction using observations Prediction prior to an update Prediction after an update Observation-prediction Correction factor applied to the innovation Linearized (using Taylor’s series expansion) version of a nonlinear model Operator allowing the model to be run backwards in time

differences between these two approaches to data assimilation. To help the reader through the large amount of jargon typically associated with data assimilation, a list of terminology has been provided (Table 2.2). 2.3.1

Direct observer assimilation

The direct observer techniques sequentially update the model forecast, using the difference ˆ , known as the “innovation,” whenever between observation Z and model predicted observation Z observations are available. The predicted observation is calculated from the model predicted or “background” states, indicated by the superscript b. The correction added to the background state vector is the innovation multiplied by a weighting factor K known as the “gain” (sometimes called the Kalman gain). The gain represents the relative uncertainty in the observation and model variances, and is a number between 0 and 1 in the scalar situation. The resulting estimate of the state vector is known as the “analysis,” as indicated by the superscript a.

ˆ ak ⫽ Xbk ⫹ K(Zk ⫺ Zˆ k) X

(2.1)

The subscript k refers to the time of the update. If the uncertainty of the predicted observation (as calculated from the background states and their uncertainty) is large relative to the uncertainty of the actual observation then the analysis state vector takes on values that will yield the actual observation. Conversely, if the uncertainty of the predicted observation is small relative to the uncertainty of the actual observation then the analysis state vector is unchanged from the original background value. The commonly used direct observer methods are: 1 2 3 4 5 6 7 8

direct insertion; statistical correction; successive correction; optimal interpolation/statistical interpolation; analysis correction; nudging; 3D variational; KF and variants.

While approaches like direct insertion, nudging, and optimal interpolation are computationally efficient and easy to implement, the updates do not account for observation uncertainty or

32 Jeffrey P. Walker and Paul R. Houser utilize system dynamics in estimating model background state uncertainty, and information on estimation uncertainty is limited. Although computationally demanding in its pure form, KF can be adapted for near-real-time application and provides information on estimation uncertainty. However, it has only limited capability to deal with model errors, and necessary linearization approximations can lead to unstable solutions. The ensemble Kalman filter (EnKF), while it can be computationally demanding (depending on the size of the ensemble), is well suited for near-real-time applications, is robust, very flexible and easy to use, and is able to accommodate a wide range of model error descriptions. 2.3.2

Dynamic observer assimilation

The dynamic observer techniques find the best fit between the forecast model state and the observations, subject to the initial state vector uncertainty ⌺ and observation uncertainty R, by minimizing over space and time an objective function J such as

J⫽

⫺1 1 1N ⫺ 1 ˆ ( X0 ⫺ Xb0)T ⌺b0 (X0 ⫺ Xb0) ⫹ (Zk ⫺ Zˆ k)T R⫺1 k (Zk ⫺ Zk) 2 2 0

兺

(2.2)

where the superscript b refers to the initial or “background” estimate of the state vector, the subscript refers to time, and N is the number of time steps. To minimize the objective function over time, an assimilation time “window” is defined and an “adjoint” model is typically used to find the derivatives of the objective function with respect to the initial model state vector X0. The adjoint is simply a mathematical operator that allows one to determine the sensitivity of the objective function to changes in the solution of the state equations by a single forward and backward pass over the assimilation window. Although an adjoint is not strictly required (i.e. a number of forward passes can be used to numerically approximate the objective function derivatives with respect to each state), it makes the problem computationally tractable. The dynamic observer techniques can be considered simply as an optimization or calibration problem, where the state vector – not the model parameters – at the beginning of each assimilation window is “calibrated” to the observations over that time period. The dynamic observer techniques can be formulated with 1 strong constraint (variational); 2 weak constraint (dual variational or representer methods). Strong constraint is where the model is assumed perfect, as in equation 2.2, while weak constraint is where errors in the model formulation are taken into account as process noise. This is achieved by including an additional term in equation 2.2 so that J⫽

⫺1 1N ⫺ 1 1N ⫺ 1 T ⫺1 1 ˆ k)T R⫺1 ˆ ( X0 ⫺ Xb0)T ⌺b0 (X0 ⫺ Xb0) ⫹ ( Zk ⫺ Z w Q w k (Zk ⫺ Zk) ⫹ 2 2 0 2 0 k k k (2.3)

兺

兺

where w is the model error vector and Q is the model error variance–covariance matrix. Dynamic observer methods are well suited for smoothing problems, but provide information on estimation accuracy only at considerable computational cost. Moreover, adjoints are not available for many existing hydrologic models, and the development of robust adjoint models is difficult due to the nonlinear nature of hydrologic processes.

Hydrologic data assimilation 33 2.3.3

Features of data assimilation

The potential benefit of data assimilation for hydrologic science is tremendous and can be summarized as follows (adapted from Rood et al., 1994): ●

●

●

●

●

Organizes the data by objectively interpolating from the observation space to the model space. The raw observations are organized and given dynamic consistency with the model equations, thereby enhancing their usefulness. Supplements the data by constraining the model’s physical equations with parsimonious observations, which can be used to estimate unobserved quantities. This allows the progress of research that would be impossible without assimilation, because it allows for a more complete understanding of the true state of a hydrologic system (see Figure 2.4(a)). Complements the data by propagating information into regions of sparse observations using either observed spatial and temporal correlations, or the physical relationships included in the model (see Figure 2.4(b)). Quality controls the data through comparison of observations with previous forecasts to identify and eliminate spurious data. By performing this comparison repeatedly, it is possible to calibrate observing systems and identify biases or changes in observation system performance. Validates and improves the hydrologic models by continuous model confrontation with real data. This helps to identify model weaknesses such as systematic errors and correct them.

(a) Model prediction

Statistically optimal model update

Depth

Depth

Model prediction

Direct replacement with observations

Model update True profile

True profile

Matric head

Matric head

(b) Model with 4DDA

Observation

Model

Tombstone,AZ AZ Tombstone, 0%

20%

Scale (km) 0

5

Figure 2.4. Example of how data assimilation supplements data and complements observations. (a) Numerical experiment results demonstrating how near-surface soil moisture measurements are used to retrieve the unobserved root zone soil moisture state using (left panel) direct insertion and (right panel) a statistical assimilation approach (Walker et al., 2001a) and (b) six Push Broom Microwave Radiometer (PBMR) images gathered over the USDA-ARS Walnut Gulch Experimental Watershed in southeast Arizona were assimilated into the TOPLATS hydrologic model using several alternative assimilation procedures (Houser et al., 1998). The observations were found to contain horizontal correlations with length scales of several tens of kilometers, thus allowing soil moisture information to be advected beyond the area of the observations (see Color Plate VI).

34 2.3.4

Jeffrey P. Walker and Paul R. Houser Quality control for data assimilation

One of the major components of any data assimilation system is quality control of the input data stream. Quality control is a pre-assimilation rejection or correction of questionable or bad observations, which begins where the remote sensing product quality control activities leave off. The observation data from remote sensing systems contain errors that can be classified into two types: 1 natural error (including instrument and representativeness error); 2 gross error (including improperly calibrated instruments, incorrect registration or coding of observations, and telecommunication error). These errors can be either random or spatially and/or temporally correlated with each other; inversion techniques and instrument biases can be correlated in time and space, and calibrations of remote sensing instruments can drift. To address these problems, a number of quality control operations are performed. The quality control process consists of a set of algorithms which examine each data item, individually or jointly, in the context of additional information. Their primary purpose is to determine which of the data are likely to contain unknown (incorrigible) gross errors, and which are not. Quality control proceeds in a three step process – (1) test for potential problem observations, (2) attempt to correct the problem observation, and (3) decide the fate of the observation (data rejection). The quality control algorithms can be categorized as follows: ●

●

●

●

Quality control flags are used to check the data for inconsistencies noted during the measurement, transmission, pre/post processing, and archiving stages. Consistency or sanity checks to see if the observation absolute value or time rate of change is physically realistic. This check filters such things as observations outside the expected range, unit conversion problems, etc. Buddy checks compare the observation with comparable nearby (space and time) observations of the same type and reject the questioned observation if it exceeds a predefined level of difference. Background checks examine if the observation is changing similarly to the model prediction. If it is not, and the user has some reasonable confidence in the model, the observation may be questioned.

2.3.5

Validation using data assimilation

The continuous confrontation of model predictions with observations in a data assimilation system presents a rich opportunity to better understand physical processes and observation quality in a structured, iterative, and open-ended learning process. Inconsistencies between observations and predictions are easily identified in a data assimilation system, providing a basis for observational quality control and validation. Systematic differences between observations and model predictions can identify systematic error. This methodology clearly illustrates the importance of a good quality forecast and an analysis that is reasonably faithful to the observations. If the hydrologic model makes reasonably good predictions, then the analysis must only make small changes to an accurate background field. The validation of observations in a data assimilation system is centered on (1) comparisons of new observations with the model forecast and the data assimilation analysis, and (2) interpretation of the forecast error covariances. The data assimilation validation algorithms can be categorized as follows: ●

Innovation evaluation compares the observation with the model prediction as either a single point in time or change over time; large or obvious deviations from the model prediction are

Hydrologic data assimilation 35

●

●

●

●

probably wrong. Means, standard deviations, and time evolution of observed minus predicted fields are examined with the goal of detecting abrupt changes. Analysis residual evaluation compares the observation with the data assimilation analysis. Examination of the means, standard deviations, and time evolution of observed minus predicted fields will help to diagnose systematic or abrupt observation system changes. This technique is useful to diagnose the performance of the analysis, and whether the observations are being used effectively (Hollingsworth and Lonnberg, 1989). Observation withholding is a stringent method for validation in an assimilation system where some of the observational data are withheld from the analysis procedure in data-dense regions. This allows the analysis to be validated against the withheld observations. Error propagation is undertaken and changes in the regional distribution or absolute value of these errors could indicate observational problems. Model and observation bias is generally assumed to be zero and uncorrelated in space. These assumptions work reasonably well for in-situ observations, but satellite observations are usually biased by inaccurate algorithms, and their errors are usually horizontally correlated because the same sensor is making all the observations. With recent work by Dee and Todling (2000) the bias of the model and observations can be continuously estimated and corrected for. Evaluation of these bias estimates in space and time may lead to additional insights into the observational characteristics.

2.4

DIRECT OBSERVER ASSIMILATION METHODS

Land surface hydrology process models are typically nonlinear, and can be considered to forecast the system state vector X at time k ⫹ 1 as a function of the system state vector estimate at the previous time step k and a forcing vector U. The model state forecast is subject to a model error vector w, which represents errors in the model forcing data, initial conditions, parameters, and physics. The state equation is given by Xk ⫹ 1 ⫽ ak(Xk, Uk)⫹wk

(2.4)

where a is a nonlinear operator. This equation can be linearized to obtain the “tangent linear model” as Xk ⫹ 1 ⫽ AkXk ⫹ BkUk ⫹ wk

(2.5)

The state space equation is subject to the initial state vector X0 ⫽ X(t0) ⫹ e0

(2.6)

with error vector e. The observation equation is given by

ˆ k ⫽ hk(Xk) ⫹ vk Z

(2.7)

where h is a nonlinear operator. This equation can also be linearized as

ˆ k ⫽ HXk ⫹ Yk ⫹ vk Z with error vector v.

(2.8)

36 Jeffrey P. Walker and Paul R. Houser The key assumptions of this assimilation approach are that the error terms w, v, and e are uncorrelated (white) through time and have Gaussian distributions as represented by their covariance matrices Q, ⌺, and R, respectively. That is E(wk) ⫽ 0

E(wkwTk ) ⫽ Qk

E(e0) ⫽ 0

E(e0eT0) ⫽ ⌺b0

E(vk) ⫽ 0

E(vkvTk ) ⫽ Rk

(2.9)

where E is the expectation operator. The assumption that observation and model errors are unbiased relative to each other and the “truth” is the most restrictive assumption, most commonly violated assumption, and most detrimental assumption in terms of predictive performance. One key question in the direct observer data assimilation technique, and the fundamental difference between the various methods, is the choice of the gain matrix K. Ultimately, Kk should be chosen such that Xak approaches the expectation of Xk as k approaches infinity. This can be achieved by choosing K as the optimal least squares estimator or Best Linear Unbiased Estimator (BLUE) analysis obtained as a solution to the variational optimization problem posed in equation 2.2 – that is, choosing K such that objective function J is a minimum. This can be shown analytically to produce (Bouttier and Courtier, 1999) K ⫽ ⌺b HT(H ⌺b HT ⫹ R)⫺1

(2.10)

ˆ is the covariance matrix of the predicted observation Zˆ . Thus, on assimilawhere H⌺b HT ⫽ R tion interval k ∈ [0, N], the analysis XaN given by KF should be equal to the converged solution obtained by the adjoint method at time k equal to N. From application of standard error propagation theory on the correction equation it can also be shown that the updated uncertainty of the states is given by ⌺a ⫽ (I ⫺ KH) ⌺b(I ⫺ KH)T ⫹ KRKT

(2.11)

where I is the identity matrix. Equations 2.1, 2.5, 2.8, 2.10, and 2.11 form the basis of the KF approach (Kalman, 1960) to data assimilation. Apart from the assumption that errors are unbiased and normally distributed, the difficulty associated with applying these equations is an estimate of the background variance–covariance ˆ b), matrix ⌺b , and that to find the analysis Xa one must compute ⌺b HT( H ⌺b HT ⫹ R)⫺1(Z ⫺ Z which is computationally expensive. As a result, approximations to these equations and/or alternative methodologies of solving the key equations are sought. Ultimately, it is approximations to K that are typically made. 2.4.1

Direct insertion

One of the oldest and most simplistic approaches to data assimilation is direct insertion. As the name suggests, the forecast model states are directly replaced with the observations by essentially assuming that K ⫽ 1. This approach makes the explicit assumption that the model is wrong (has no useful information) and that the observations are right, which both disregards important information provided by the model and preserves observation errors. A further key disadvantage of this approach is that model physics are solely relied upon to propagate the information to unobserved parts of the system (Houser et al., 1998; Walker et al., 2001a).

Hydrologic data assimilation 37 2.4.2

Statistical correction

A derivative of the direct insertion approach is the statistical correction approach, which adjusts the mean and variance of the model states to match those of the observations. This approach assumes the model pattern is correct but contains a non-uniform bias. First, the predicted observations are scaled by the ratio of observation field standard deviation to predicted field standard deviation. Second, the scaled predicted observation field is given a block shift by the difference between the means of the predicted observation field and observation field (Houser et al., 1998). This approach also relies upon the model physics to propagate the information to unobserved parts of the system. 2.4.3

Successive correction

This is an iterative type approach that uses weights W to smooth observations into the model states by modifying the states at all grid points within a specified radius of influence r of each observation s (Bratseth, 1986). Any weighting system can be used, but the Cressman weights given by

W sij ⫽

冦

(r 2 ⫺ d2ij) Ⲑ (r 2 ⫹ d2ij ),

dij ⬍ r

0,

dij ⱖ r

(2.12)

are commonly used, where dij is the distance between grid point i, j and the observation. In practice the approach is usually applied consecutively to each observation s from 1 to sf as

ˆ k) Xks ⫹ 1 ⫽ Xsk ⫹ Wsk(Zk ⫺ Z

(2.13)

˜ sf in equation 2.1 where W ˜ sf is and then setting Xak ⫽ Xskf . This is equivalent to using Kk ⫽ W calculated from ˜ 1 ⫽ W1 W ⯗ ˜ s⫹1⫽Ws⫹1(I⫹HW ˜ s) ⫹ W ˜s W

(2.14)

⯗ This approach assumes that the observations are more accurate than model forecasts, with the observations fitted as closely as is consistent. Moreover, it is ineffective in data sparse regions (Nichols, 2001). 2.4.4

Optimal interpolation

The optimal interpolation (OI) approach, sometimes referred to as statistical interpolation, approximates the “optimal” solution from equation 2.10 by choosing ˜ b HT( H⌺ ˜ b HT ⫹ R)⫺1 K⫽⌺

(2.15)

˜ b is an approximated background covariance matrix with a “fixed” structure for all where ⌺ time steps, and is often given by prescribed variances and a correlation function given only by distance (Lorenc, 1981).

38 Jeffrey P. Walker and Paul R. Houser 2.4.5

Analysis correction

This is a modification to the successive correction approach that is applied consecutively to each observation s from 1 to sf as (Lorenc et al., 1991)

ˆ k) Xks ⫹ 1 ⫽ Xsk ⫹ WkVk(Zsk ⫺ Z

(2.16)

where the observation vector Zs is also successively updated by

ˆ k) Zsk ⫹ 1 ⫽ Zsk ⫺ Vk(Zsk ⫺ Z

(2.17)

and the weight matrices W and V given by ˜ b HT R⫺1 Wk ⫽ ⌺ k k k

(2.18a)

Vk ⫽ (I ⫹ HkWk)⫺1

(2.18b)

In practice Zs is not updated and Vk is approximated to avoid inversion. The result of these assumptions is an update equation equivalent to that for optimal interpolation (Nichols, 2001). 2.4.6

Nudging

The nudging approach approximates the gain matrix by the empirical function K 艐 G(WT⌰W)(WI)⫺1

(2.19)

where G is a nudging factor that gives the magnitude of the nudging term and has a value from 0 to 1, ⌰ is an observational quality factor with a value from 0 to 1, I is the identity matrix, and W is a temporal and spatial weighting function also with a value from 0 to 1. The function W is given by wxywzwt, where wxy is a horizontal weighting function (i.e. Cressman), wz is a similar vertical weighting function, and wt is a temporal weighting function. Each of these temporal/ spatial weighting functions has a value from 0 to 1 (Stauffer and Seaman, 1990). 2.4.7

3D Variational

This approach directly solves the iterative minimization problem given by equations 2.2 or 2.3 for N ⫽ 1 (Parrish and Derber, 1992). The same approximation for the background covariance matrix as in the optimal interpolation approach is typically used. The solution gives an analysis which is similar in nature to the direct insertion approach. 2.4.8

Kalman filter

The family of KF data assimilation approaches calculate the gain matrix in equation 2.10 by directly forecasting the background covariance matrix. In the traditional KF approach this is achieved by application of standard error propagation theory on the (tangent) linear model in equation 2.5. (The only difference between KF and the extended Kalman filter (EKF) is that the forecast model is linearized using a Taylor’s series expansion; the same forecast and update equations are used for each.) The state covariance forecast equation is ⌺bk ⫹ 1 ⫽ Ak ⌺bk ATk ⫹ Qk

(2.20)

Hydrologic data assimilation 39 where A is the linear operator from equation 2.5 and Q is the model error covariance matrix given in equation 2.9. Thus, the (extended) KF requires propagation of the state covariances along with the states. Whereas the approach gives an optimal analysis for the assumed statistics, the initial state error covariance matrix ⌺0 and more seriously the model error covariance matrix Q are difficult to define and are often assumed ad hoc. The standard EKF approach assumes an explicit model, which is limiting in terms of computational runtime as a result of the small step size necessary to satisfy stability criteria. However, it is also possible to apply the same update and state covariance forecast equations to an implicit formulation, such as the Crank–Nicholson scheme ⌽1Xk ⫹ 1 ⫹ ⍀1 ⫽ ⌽2Xk ⫹ ⍀2

(2.21)

by making the substitutions that Ak ⫽ ⌽⫺1 1 ⌽2

(2.22a)

BkUk ⫽ ⌽⫺1 1 [⍀2⫺⍀1]

(2.22b)

but the inverse and multiplication required to calculate A is costly for large systems (Walker et al., 2001b). The standard EKF update and state covariance forecast equations can also be applied directly with a nonlinear state forecast model. This is achieved by numerically approximating the Jacobians A and H as required by ⭸Xbk⫹1

Ak ⫽

⭸Xbk

Hk ⫽

⭸Zk

(2.23a)

(2.23b)

⭸Xbk

However, the cost of doing this is n ⫹ 1 times the standard model runtime, where n is the number of state variables to be updated by the assimilation. Note that only states with significant correlation to the observation need be included in the state covariance forecast and update (Walker and Houser, 2001). A further approach to estimating the state covariance matrix is the EnKF. As the name suggests, the covariances are calculated from an ensemble of state forecasts using the Monte Carlo approach rather than a single discrete forecast of covariances. In this case m ensembles of n model predicted states X are stored as x using different initial conditions and forcing (Turner et al., 2004), different parameters and/or models, different model error (additive/ multiplicative/etc.), etc., in order to get a representative spread of state forecasts amongst the ensemble members. While this is quite straightforward, the question of what model error w to apply, and how, is still a major unknown. Moreover, special care is required when m is less than the number of observations n. Using this approach, the background state covariance matrix is basically calculated as

⌺bk ⫽

(xbk ⫺ xbk)(xbk ⫺ xbk)T m⫺1

(2.24)

40 Jeffrey P. Walker and Paul R. Houser This could then be used in equation 2.10 directly, except some smart math is typically used so only matrices of size (n ⫻ m) are required (Evensen, 1994; Houtekamer and Mitchell, 1998). Thus, ⌺b is never calculated explicitly. Here the analysis equation is presented as Xak ⫽ Xbk ⫹ BTk bk

(2.25)

where BTk ⫽ ⌺bk HTk

(2.26a)

ˆ k) bk ⫽ (Hk ⌺bk HTk ⫹ Rk)⫺1(Zk ⫺ Z

(2.26b)

By rearranging equation 2.26a and letting y ⫽ Z ⫹ , where  is a zero mean random observation error term with covariance matrix R, b is solved for each ensemble from

(Hk ⌺bk HTk ⫹ Rk)bk ⫽ (yk ⫺ Zˆ k)

(2.27)

where qkqTk m⫺1

(2.28)

qk ⫽ Hk(xbk ⫺ xbk) ⫽ (ˆzk ⫺ ˆz k)

(2.29)

Hk ⌺bk HTk ⫽ and

The vector zˆ is the predicted observation vector for each of the respective ensemble members. In this case it is not necessary to solve for H either, and the updates are made individually to each of the ensemble members. Finally, B can be estimated from

BTk ⫽

(xbk ⫺ xbk) m⫺1

qTk

(2.30)

Reichle and McLaughlin (2001) applied the EnKF to the soil moisture estimation problem and found it to perform as well as the numerical Jacobian approximation approach to the EKF, with the distinct advantage that the error covariance propagation is better behaved in the presence of large model nonlinearities. This was the case even when using only the same number of ensembles as required by the numeric approach to the EKF, that is, n ⫹ 1.

2.5

DYNAMIC OBSERVER ASSIMILATION METHODS

In its pure form, the “variational” (otherwise known as Gauss-Markov) dynamic observer assimilation methods use an adjoint to efficiently compute the derivatives of the objective function J with respect to each of the initial state vector values X0. This adjoint approach is derived

Hydrologic data assimilation 41 by defining the Lagrangian ᏸ as the adjoining of the model to the model response using Lagrange multipliers  ᏸ⫽ J ⫹

N⫺1

兺0 Tk ⫹ 1[Xk ⫹ 1 ⫺ ak(Xk,Uk)]

(2.31)

where ideally the second term is zero. Thus the Lagrange multiplier is chosen such that ⵱ᏸ ⫽ 0 and N ⫽ 0, yielding (i.e. backward pass)

ˆ k ⫽ ATk k ⫹ 1 ⫺ HTk R⫺1 k (Zk ⫺ Zk)

(2.32)

The derivative of the objective function is given from the Lagrange multiplier at time zero by ⫺T0 (Castelli et al., 1999; Reichle and McLaughlin, 2001). Note that AT, the adjoint operator, is from the tangent linear model in equation 2.5, and effectively needs to be saved during the forward pass (Bouttier and Courtier, 1999). Solution to the variational problem is then achieved by minimization and iteration. In practical application, the number of iterations is usually constrained to a small amount. While “adjoint compilers” are available (see http://www.autodiff.com/tamc/) for automatic conversion of the nonlinear forecast model into a tangent linear model, application of these is not straightforward. It is best to derive the adjoint at the same time as the model is developed.

2.6

CASE STUDIES

Significant advances in hydrologic data assimilation have been made over the past decade from which we have selected a few case studies to demonstrate the utility of hydrologic data assimilation. 2.6.1

Case study 1: soil moisture assimilation

A one-dimensional KF soil moisture assimilation strategy was developed by Walker and Houser (2001) that provides a framework to constrain model predicted soil moisture with observations, using covariances that represent their respective uncertainty. A one-dimensional EKF was used because of its computational efficiency and the fact that horizontal correlations between soil moisture prognostic variables of adjacent catchments at the scales of interest to climate modeling are likely only through the large-scale correlation of atmospheric forcing. A set of numerical experiments was undertaken for North America to illustrate the effectiveness of the KF assimilation scheme in providing an accurate soil moisture estimate. When assimilating surface soil moisture once every three days, the scheme was generally able to retrieve the “true” profile soil moisture after only one month, and the predicted evapotranspiration and runoff fluxes were significantly improved. Walker and Houser (2004) used this same EKF framework to address soil moisture satellite mission accuracy, repeat time, and spatial resolution requirements through a numerical twin study. Simulated soil moisture profile retrievals were made by assimilating near-surface soil moisture observations with various accuracy, repeat time, and spatial resolution. It was found that (1) near-surface soil moisture observation error must be less than the model forecast error required for a specific application and must be better than 5% v/v accurate to positively impact soil moisture forecasts, (2) daily near-surface soil moisture observations achieved the best soil moisture and evapotranspiration forecasts, (3) near-surface soil moisture observations should have a spatial resolution of around half the model resolution, and (4) satisfying the

42 Jeffrey P. Walker and Paul R. Houser spatial resolution and accuracy requirements was much more important than repeat time. This kind of study is important for planning future observation systems, but it must be recognized that observation requirements are also highly application specific; for example, flood forecasting and precision agriculture requirements will likely have different requirements than climate modeling and policy planning, as they operate at different scales. Walker and Houser (2002) was the first known study to use space-borne measurements of near-surface soil moisture content to estimate the spatial and temporal variation of soil moisture content at the continent scale by the process of data assimilation. Near-surface soil moisture measurements from the 6.6 GHz (C-band) channels of the Scanning Multi-channel Microwave Radiometer (SMMR) were assimilated into a land surface model over North America using a KF to correct for soil moisture estimation errors. Comparison with the limited ground-based point measurements of soil moisture content found a net improvement when near-surface soil moisture observations were assimilated. Walker et al. (2003) used a similar approach to estimate the spatial and temporal variation of soil moisture content across Australia (Figure 2.2). Unfortunately, the lack of appropriate soil moisture evaluation data and mismatch in scale between model output and available data made it difficult to draw any conclusive statements regarding improvements in soil moisture predictions. There was, however, an obvious increase in correlation between soil moisture predictions and NDVI data when SMMR surface soil moisture data were assimilated. This provides some encouragement for pursuing assimilation experiments using the new AMSR-E data and the collection of more appropriate ground-based soil moisture data for validation purposes. 2.6.2

Case study 2: downscaling with data assimilation

In a data assimilation framework, it may be possible to effectively increase the resolution of observations by making use of forecasts that include higher resolution meteorologic, land cover, and soil texture information. Figure 2.5 demonstrates how low-resolution brightness temperature observations have been used to estimate soil moisture at high resolution through variational assimilation into a land surface model (Reichle et al., 2001). Spatial structures at scales well below the scale of the observations were resolved satisfactorily. This means that brightness images with a resolution of a few tens of kilometers are useful, even if the estimation scale of interest is of the order of a few kilometers, provided that fine-scale information is available on the meteorologic forcing, land cover, and texture. The downscaling properties of data assimilation may also be able to help overcome the limitations of passive radiometric remote sensing (higher accuracy but lower spatial resolution) and active radar remote sensing (higher spatial resolution but lower accuracy) for sensing soil moisture. Zhan et al. (2004) tested this hypothesis by conducting an Observation System Simulation Experiment (OSSE) where the feasibility of retrieving surface soil moisture at a medium spatial scale (10 km) from both coarse scale (40 km) radiometer brightness temperature and fine-scale (3 km) radar backscatter cross-section observations was evaluated using the EKF. In this case, the background field was an inversion of the passive data. Compared with the results from traditional soil moisture inversion algorithms, the combined active-passive EKF retrieval algorithm significantly reduced soil moisture error at the medium scale. There is the potential to further enhance this downscaling approach by additionally using the information contained in overlapping observations. 2.6.3

Case study 3: snow assimilation

Because of snow’s high albedo, thermal properties, feedback to the atmosphere, and its nature of being a medium-term water store, improved snow state estimation has the potential to greatly increase climatologic and hydrologic prediction accuracy. An analysis scheme to assimilate

Hydrologic data assimilation 43

True

Prior

Est (1: 4)

Est (1:16)

Day of year 169.7 y (km)

160 0.6

120

0.4 40 0.2

0

Day of year 174.66 y (km)

160 0.6

120

0.4 40 0.2

0

Day of year 178.66 y (km)

160 0.6

120

0.4 40 0

0.2 0

40 x (km)

80

0

40 x (km)

80

0

40 x (km)

80

0

40 80 x (km)

Figure 2.5. True and prior surface soil saturation (in the first two columns) at three different times. Corresponding estimates for downscaling ratios of (1 : 4) and (1 : 16) are shown in the third and fourth columns. The resolution of the observations used to compute the downscaled estimates are indicated with solid black grid lines (Reichle et al., 2001) (see Color Plate VII).

observed snow water equivalent into a land surface model has been developed (Sun et al., 2004). Using a set of numerical “twin” experiments, the scheme is shown to be successful in retrieving the snow states (snow water equivalent, snow depth, and snow temperature) from observations of snow water equivalent alone. The study illustrates that, by assimilating remotely sensed snow water equivalent observations, the errors in forecast snow states from poor initial conditions can be removed (Figure 2.6), and the prediction of runoff and atmospheric fluxes can be improved. A comparison between monthly averaged runoff and atmospheric fluxes showed negligible differences between the assimilation and truth simulations. Moreover, the assimilation significantly improved both upward shortwave and longwave radiation, and runoff predictions, as compared to no assimilation. Snow has several properties that make it uniquely challenging to assimilate, as follows. First, snow cover and depth observations provide an incomplete description of the multi-layer snow water equivalent, temperature, and density states used in most physical snow models. For example,

(a)

(b)

(c) 120

90N

108

80N

96 70N 84 SWE

60N

72

50N

60

40N

48 36

30N

24 20N 12 10N 90N

0 600

80N

540 480

70N Snow depth

420 60N

360

50N

300

40N

240

30N

180 120

20N

60 10N

0 0

90N

–5

80N

–10

Snow temperature

70N

–15 60N –20 50N

–25

40N

–30

30N

–35 –40

20N –45 10N 90N

–50 1.00

80N

0.90 0.80

70N Snow fraction

0.70 60N 0.60 50N

0.50

40N

0.40

30N

0.30 0.20

20N

0.10 10N 0.00 105W 104W 103W 102W 101W 100W

60W 5W

104W 103W 102W 101W 100W 60W 5W

104W 103W 102W 101W 100W 60W

Figure 2.6. Comparison of snow simulations on January 5, 1987 over North America for snow water equivalent (in mm, top row); snow depth (in mm, second row); average snow temperature (in ⬚C, third row); and areal snow fraction (bottom row) from (a) truth run (using spin up initial condition), (b) assimilation run (with degraded initial condition and assimilation of daily total snow water equivalent observations), and (c) control run (with degraded initial condition) (Sun et al., 2004) (see Color Plate VIII).

Hydrologic data assimilation 45 snow cover observations provide a binary snow presence description without snow quantity information. This is generally incompatible with data assimilation schemes that act on snow states. Rodell et al. (2002) overcame this problem by adding an arbitrarily thin layer of snow to model elements that have no snow when snow cover was observed. Second, snow is a highly transient model state that disappears and is not predicted for long periods of time during the year. This is generally incompatible with modern data assimilation techniques that seek to propagate error covariances; when there is no snow state prediction (i.e. when the temperature is above freezing), then there can also be no error propagation. This problem was overcome by Sun et al. (2004) by simply reinitializing the error covariances when snow reappeared, but may not be so easily overcome in an EnKF context where ensemble members may vary significantly. Finally, in the presence of temperature bias, snow assimilation may have an undesirable water budget impact. Cosgrove et al. (2004) show that large water balance errors occur when imperfect snow melting processes interact with the direct insertion of perfect snow observations. Constraining these snow melt biases is important for achieving optimal assimilation results and is an important topic for future research. 2.6.4

Case study 4: skin temperature assimilation

The land surface skin temperature state is a principle control on land–atmosphere fluxes of water and energy. It is closely related to soil water states, and is easily observable from space and aircraft infrared sensors in cloud-free conditions. The usefulness of skin temperature in land data assimilation studies is limited by its very short memory (on the order of minutes) due to the very small heat storage it represents. Radakovich et al. (2001) have demonstrated skin temperature data assimilation in a land surface model (Figure 2.7) using three-hourly observations from the International Satellite Cloud Climatology Project (ISCCP). Incremental and SON 1992 sensible heat flux (Wm–2) model – NCEP Bias = 54.409; SD = 32.11

SON 1992 skin temperature (K) model – NCEP Bias = 1.1067; SD = 4.465 80N 70N 60N 50N 40N 30N 20N 10N EQ 10S 20S 30S 40S 50S 60S 180

10 5 3 2 1 –1 –2 –3 –6 –10 120 W

60 W

0

60E

10 5 3 2 1 –1 –2 –3 –5 –10

120W

60W

0

60E

120E

150 100 50 25 –25 –50 –100 –150 –200 180

120E

Assim.V – NCEP Bias = 0.1841; SD = 3.446 80N 70N 60N 50N 40N 30N 20N 10N EQ 10S 20S 30S 40S 50S 60S 180

200

80N 70N 60N 50N 40N 30N 20N 10N EQ 10S 20S 30S 40S 50S 60S 120W

60W

0

60E

120E

Assim.V – NCEP Bias = 88.633; SD = 99.43

80N 70N 60N 50N 40N 30N 20N 10N EQ 10S 20S 30S 40S 50S 60S

200 150 100 50 25 –25 –50 –100 –150 –200

180

120W

60W

0

60E

120E

Figure 2.7. Differences between simulated and reanalysis (top left), assimilated and reanalysis (bottom left) mean skin temperature (K), and the resulting differences between simulated and reanalysis (top right), and assimilated and reanalysis (bottom right) mean sensible heat fluxes (Wm⫺2) for September through November 1992. Global terrestrial mean bias and standard deviation (SD) for September through November are also noted (Radakovich et al., 2001) (see Color Plate IX).

46

Jeffrey P. Walker and Paul R. Houser

semi-diurnal bias correction techniques based on Dee and da Silva (1998) were developed to account for biased skin temperature forecasts. The assimilation of ISCCP-derived surface skin temperature significantly reduced the bias and standard deviation between model predictions and the NCEP reanalysis (Kalnay et al., 1996). However, the assimilation of ISCCP-derived surface skin temperature has a substantial impact on the sensible heat flux, due to an enhanced gradient between the surface and 2 m air temperatures. If the near-surface air temperature were interactive, as in a coupled land-atmosphere model, then it would respond to this enhanced flux rather than maintain the artificial temperature gradient.

2.7

SUMMARY

Hydrologic data assimilation is an objective method to estimate the hydrologic system states from irregularly distributed observations. These methods integrate observations into numeric prediction models to develop physically consistent estimates that better describe the hydrologic system state than the raw observations alone. This process is extremely valuable for providing initial conditions for hydrologic system prediction and/or correcting hydrologic system prediction and for increasing our understanding and improving parameterization of hydrologic system behavior through various diagnostic research studies. Hydrologic data assimilation is still in its infancy, with many open areas of research. Development of hydrologic data assimilation theory and methods is needed to (1) better quantify and use model and observation errors, (2) create model independent data assimilation algorithms that can account for the typical nonlinear nature of hydrologic models, (3) optimize data assimilation computational efficiency for use in large operational hydrologic applications, (4) use forward models to enable the assimilation of remote sensing radiances directly, (5) link model calibration and data assimilation to optimally use available observation information, (6) create multivariate hydrologic assimilation methods to use multiple observations with complementary information, (7) quantify the potential of data assimilation downscaling, and (8) create methods to extract the primary information content from observations with redundant or overlaying information. Furthermore, the regular provision of snow, soil moisture, and surface temperature observations with improved knowledge of observation errors in time and space are essential to advance hydrologic data assimilation. Hydrologic models must also be improved to (1) provide more “observable” land model states, parameters, and fluxes, (2) include advanced processes such as river runoff and routing, vegetation and carbon dynamics, and groundwater interaction to enable the assimilation of emerging remote sensing products, (3) have valid and easily updated adjoints, and (4) have knowledge of their prediction errors in time and space. The assimilation of additional types of hydrologic observations such as streamflow, vegetation dynamics, evapotranspiration, and groundwater or total water storage must be developed. As with most current data assimilation efforts, we describe data assimilation procedures that are implemented in uncoupled models. However, it is well known that the high-resolution time and space complexity of hydrologic phenomena have significant interaction with atmospheric, biogeochemical, and oceanic processes. Scale truncation errors, unrealistic physics formulations, and inadequate coupling between hydrology and the overlying atmosphere can feedback to cause serious systematic hydrologic errors. Hydrologic balances cannot be adequately described by current uncoupled hydrologic data systems because large analysis increments that compensate for errors in coupling processes (e.g. precipitation) result in important non-physical contributions to the energy and water budgets. Improved coupled process models with improved feedback processes, better observations, and comprehensive methods for coupled assimilation are needed to achieve the goal of fully coupled data assimilation systems that should produce the best and most physically consistent estimates of the Earth system.

Hydrologic data assimilation 47 REFERENCES Arya, L.M., Richter, J.C., and Paris, J.F. (1983) Estimating profile water storage from surface zone soil moisture measurements under bare field conditions. Water Resources Research, 19(2), 403–412. Bennett, A.F. (1992) Inverse Methods in Physical Oceanography. Cambridge: Cambridge University Press, 346pp. Bernard, R., Vauclin, M., and Vidal-Madjar, D. (1981) Possible use of active microwave remote sensing data for prediction of regional evaporation by numerical simulation of soil water movement in the unsaturated zone. Water Resources Research, 17(6), 1603–1610. Bouttier, F. and Courtier, P. (1999) Data assimilation concepts and methods. ECMWF training course notes. Bouttier, F., Mahfouf, J.-F., and Noilhan, J. (1993) Sequential assimilation of soil moisture from atmospheric low-level parameters. Part I: Sensitivity and calibration studies. Journal of Applied Meteorology, 32(8), 1335–1351. Bratseth, A.M. (1986) Statistical interpolation by means of successive corrections. Tellus, 38A, 439–447. Bruckler, L. and Witono, H. (1989) Use of remotely sensed soil moisture content as boundary conditions in soil-atmosphere water transport modeling: 2. Estimating soil water balance. Water Resources Research, 25(12), 2437–2447. Castelli, F., Entekhabi, D., and Caporali, E. (1999) Estimation of surface heat flux and an index of soil moisture using adjoint-state surface energy balance. Water Resources Research, 35(10), 3115–3125. Charney, J.G., Halem, M., and Jastrow, R. (1969) Use of incomplete historical data to infer the present state of the atmosphere. Journal of Atmospheric Science, 26, 1160–1163. Cosgrove, B.A., Houser, P.R., and Toll, D.L. (2004) Impact of surface forcing biases on snow assimilation (in preparation). Daley, R. (1991) Atmospheric Data Analysis. Cambridge: Cambridge University Press, 460pp. Dee, D.P. and da Silva, A. (1998) Data assimilation in the presence of forecast bias. Quarterly Journal of the Royal Meteorological Society, 124, 269–295. Dee, D.P. and Todling, R. (2000) Data assimilation in the presence of forecast bias: the GEOS moisture analysis. Monthly Weather Review, 128, 3268–3282. Entekhabi, D., Nakamura H., and Njoku, E.G. (1994) Solving the inverse problem for soil moisture and temperature profiles by sequential assimilation of multifrequency remotely sensed observations. IEEE Transactions on Geoscience and Remote Sensing, 32, 438–448. Evensen, G. (1994) Sequential data assimilation with a nonlinear quasi-geostrophic model using Monte Carlo methods to forecast error statistics. Journal of Geophysical Research, 99(C5), 10143–10162. Georgakakos, K.P. and Baumer, O.W. (1996) Measurement and utilization of on-site soil moisture data. Journal of Hydrology, 184, 131–152. Hollingsworth, A. and Lonnberg, P. (1989) The verification of objective analyses: diagnostics of analysis system performance. Meteorology and Atmospheric Physics, 40, 3–27. Houser, P.R. et al. (1998) Integration of soil moisture remote sensing and hydrologic modeling using data assimilation. Water Resources Research, 34(12), 3405–3420. Houtekamer, P.L. and Mitchell, H.L. (1998) Data assimilation using a Ensemble Kalman filter techniques. Monthly Weather Review, 126, 796–811. Jackson, T.J. et al. (1981) Soil moisture updating and microwave remote sensing for hydrological simulation. Hydrological Sciences B., 26(3), 305–319. Kalman, R.E. (1960) A new approach to linear filtering and prediction problems. Transactions of the ASME, Series D, Journal of Basic Engineering, 82, 35–45. Kalnay, E. et al. (1996) The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society, 77, 437–471. Kostov, K.G. and Jackson, T.J. (1993) Estimating profile soil moisture from surface layer measurements – a review. In: Proceedings of the International Society for Optical Engineering, Vol. 1941. Orlando, FL, pp. 125–136. Lorenc, A. (1981) A global three-dimensional multivariate statistical interpolation scheme. Monthly Weather Review, 109, 701–721. Lorenc, A.C., Bell, R.S., and Macpherson, B. (1991) The Meteorological Office analysis correction data assimilation scheme. Quarterly Journal of the Royal Meteorological Society, 117, 59–89.

48 Jeffrey P. Walker and Paul R. Houser Milly, P.C.D. (1986) Integrated remote sensing modelling of soil moisture: sampling frequency, response time, and accuracy of estimates. Integrated Design of Hydrological Networks – Proceedings of the Budapest Symposium, 158, 201–211. Nichols, N.K. (2001) State estimation using measured data in dynamic system models. Lecture notes for the Oxford/RAL Spring School in Quantitative Earth Observation. Ottlé, C. and Vidal-Madjar, D. (1994) Assimilation of soil moisture inferred from infrared remote sensing in a hydrological model over the HAPEX-MOBILHY Region. Journal of Hydrology, 158, 241–264. Parish, D. and Derber, J. (1992) The National Meteorological Center’s spectral statistical interpolation analysis system. Monthly Weather Review, 120, 1747–1763. Prevot, L. et al. (1984) Evaporation from a bare soil evaluated using a soil water transfer model and remotely sensed surface soil moisture data. Water Resource Research, 20(2), 311–316. Radakovich, J.D., Houser, P.R., da Silva, A., and Bosilovich, M.G. (2001) Results from global land-surface data assimilation methods. In: Proceedings AMS 5th Symposium on Integrated Observing Systems. Albuquerque, NM, 14–19 January, pp. 132–134. Reichle, R.H. and McLaughlin, D.B. (2001) Variational data assimilation of microwave radiobrightness observations for land surface hydrologic applications. IEEE Transactions on Geoscience and Remote Sensing, 39(8), 1708–1718. Reichle, R.H., Entekhabi, D., and McLaughlin, D.B. (2001) Downscaling of radiobrightness measurements for soil moisture estimation: a four-dimensional variational data assimilation approach. Water Resources Research, 37(9), 2353–2364. Rodell, M. et al. (2002) Use of MODIS-derived snow fields in the Global Land Data Assimilation System. In: Proceedings GAPP Mississippi River Climate and Hydrology Conference. New Orleans, LA. Rood, R.B., Cohn, S.E., and Coy, L. (1994) Data assimilation for EOS: the value of assimilated data, Part 1. The Earth Observer, 6(1), 23–25. Schuurmans, J.M. et al. (2003) Assimilation of remotely sensed latent heat flux in a distributed hydrological model. Advances in Water Resources, 26(2), 151–159. Stauffer, D.R. and Seaman, N.L. (1990) Use of four-dimensional data assimilation in a limited-area mesoscale model. Part I: experiments with synoptic-scale data. Monthly Weather Review, 118, 1250–1277. Sun, C., Walker, J.P., and Houser, P.R. (2004) A methodology for snow data assimilation in a land surface model. Journal of Geophysical Research Atmospheres, 109. Turner, M.R.J., Walker, J.P., and Oke, P.R. (2004) Ensemble member generation for sequential data assimilation (in preparation). Viterbo, P. and Beljaars, A. (1995) An improved land surface parameterization scheme in the ECMWF model and its validation. Journal of Climate, 8, 2716–2748. Walker, J.P. and Houser, P.R. (2001) A methodology for initialising soil moisture in a global climate model: assimilation of near-surface soil moisture observations. Journal of Geophysical Research Atmospheres, 106(D11), 11761–11774. Walker, J.P. and Houser, P.R. (2002) Soil moisture estimation using remote sensing. In: Proceedings of the 27th Hydrology and Water Resources Symposium. The Institute of Engineers Australia, Melbourne, Australia, 20–23 May. Walker, J.P. and Houser, P.R. (2004) Requirements of a global near-surface soil moisture satellite mission: accuracy, repeat time, and spatial resolution. Advances in Water Resources, 27, 785–801. Walker, J.P., Willgoose, G.R., and Kalma, J.D. (2001a) One-dimensional soil moisture profile retrieval by assimilation of near-surface observations: a comparison of retrieval algorithms. Advances in Water Resources, 24(6), 631–650. Walker, J.P., Willgoose, G.R., and Kalma, J.D. (2001b) One-dimensional soil moisture profile retrieval by assimilation of near-surface measurements: a simplified soil moisture model and field application. Journal of Hydrometeorology, 2(4), 356–373. Walker, J.P., Willgoose, G.R., and Kalma, J.D. (2002) Three-dimensional soil moisture profile retrieval by assimilation of near-surface measurements: simplified Kalman filter covariance forecasting and field application. Water Resources Research, 38(12), 1301. Walker, J.P., Ursino, N., Grayson, R.B., and Houser, P.R. (2003) Australian root zone soil moisture: assimilation of remote sensing observations. D. Post (ed.), In: Proceedings of the International Congress on Modelling and Simulation (MODSIM). Modelling and Simulation Society of Australia and New Zealand, Inc., Townsville, Australia, 14–17 July, 1, 380–385. Zhan, X. et al. (2004) Retrieving medium resolution surface soil moisture from coarse resolution radiometer and fine resolution radar observations using the Kalman filter (in preparation).

CHAPTER 3

Analysis of Remotely Sensed Data R. Krishnan Advanced Data Processing Research Institute (ADRIN), Department of Space, Hyderabad 500 009, India B.L. Deekshatulu ISRO Visiting Professor, University of Hyderabad, Hyderabad 500 046, India

3.1

INTRODUCTION

There has been a significant growth in the volume and quality of remote sensing data over the last 25 years. Spatial and spectral resolutions have improved greatly. Satellites specially designed for dedicated purposes have been launched. High spatial resolution has been achieved at the cost of image quality in some cases. These have called for special image restoration algorithm development. Similarly, this high spatial resolution has brought greater importance to texture as an element and hence texture based classification methods have come into vogue. Likewise, shape also plays a key role and hence shape based Pattern Recognition (PR) tools have become important. The need to get better classification accuracy when the class boundaries are highly nonlinear has prompted the use of models such as Artificial Neural Networks (ANN) (these are mathematical models which mimic the human neurons and are used in many classification/learning applications). In the area of Digital Image Processing, apart from basic tools which manipulate the brightness, contrast or gamma, the main emphasis has been on image restoration, which models the sensor, platform and the atmosphere as well as on image models which lead to image understanding and simulation.

3.2

CHARACTERISTICS OF REMOTE SENSING DATA

In this chapter we will essentially deal with remote sensing data obtained from satellites. These satellites usually have a sun synchronous orbit and gather data at more or less the same solar time. This data is essentially gathered in multiple spectral bands. The variability of the data arises from the following aspects: ● ● ● ● ●

IFOV (Instantaneous Field of View) for Spatial Resolution; swath; number of spectral bands involved; number of bits of quantization; repetitiveness of the data coverage.

Figure 3.1 shows the evolution of data volume versus spatial resolution.

50 R. Krishnan and B.L. Deekshatulu

120

Data volume (MB)

100 80 60 40 20 0 0

5

10

15

20

25

30

35

Spatial resolution (mt)

Figure 3.1.

3.2.1

Data volume evolution (see Color Plate X).

Geometric correction

Traditionally the geometric corrections of satellite data involved earth rotation, attitude and orbit corrections. Both the orbit and attitude were stable over a pass and the correction essentially was simple. Usually the use of a single ground control point for a given path was sufficient. The advent of high-resolution satellites necessitated more complex image acquisition methodologies like Step and Stare and Time Delay Integration (TDI) method, which provide adequate imaging/integration time to the sensors. This calls for more sophisticated modelling. To improve the resolution and for acquiring along-track stereo, satellites should be highly manoeuvrable, that is, scanning should be done in a non-synchronous mode. Non-synchronous imaging implies that the ground scanning velocity is different from the satellite’s ground velocity and that it can be adjusted and optimized to light conditions of the imaged area (Figure 3.2). For some satellites, which allow for scanning in a non-synchronous mode called nimble imaging technology the imaging velocity is much lower than the satellite velocity. Hence the satellite actually bends further backwards as the satellite moves forwards, enabling its detectors to dwell for the necessary time (‘integration time’ or ‘dwell time’) over each imaging area. The low scanning velocity is provided by the backwards movement compensation of the satellites attitude during the imaging process. Such a compensation movement is produced through the use of the reaction wheels, which are commanded by the attitude and control system, according to the specific geometry that is to be obtained according to the tasking from ground station. Thus, for each lighting sun condition the integration time is selected (thus defining the scanning ground velocity). For others, it is achieved using a constant integration time but a variable pitch rate steering mechanism called Step and Stare Imaging. TDI is another approach and is based on the principle of multiple exposure of the same subject. This principle is shown in Figure 3.3 for a three-stage TDI detector. In the practical tests a 96-stage TDI detector was used. For asynchronous imaging modes a higher order polynomial (say 9) is used to model the attitude variation. Also more Ground Control Points (GCPs) (these are matching points corresponding to the same location on the ground and in the image) are required.

Analysis of remotely sensed data 51

Satellite orbit

Satellite ground track

Swath

Figure 3.2.

Photographic time

Asynchronous imaging mode (see Color Plate XI).

Time:

t + ∆t

t

t + 2∆t

t + 3∆t

Flight direction Shift register

Readout

Charge

Figure 3.3. Exposure principle of a TDI detector with three stages. The amount of generated charges is directly proportional to the number of stages (see Color Plate XII).

3.2.2

Radiometric correction

This essentially tries to neutralize the effects of the varying transfer characteristics of the imaging elements (say, Charge Coupled Devices or CCDs.) However, complex imaging geometries like TDI and Step and Stare call for methods like deblurring, MTF (Modulation Transfer Function) improvement, etc. to improve the radiometric quality of the image. The following treatment on classification is largely drawn from Krishnan (2002).

52

R. Krishnan and B.L. Deekshatulu

3.3

PURPOSES OF CLASSIFICATION

Classification is needed to convert a multitude of data into certain meaningful number of labels so that we can make sense of the environment from which the data has come. The methods used for classification depend on factors such as the types of classes that are sought to be identified, the resolution of the data (spectral and spatial), the need for crisp or fuzzy classes, separability between the classes, knowledge about the distribution of the classes and the tolerable degree of penalty or loss associated with misclassification. 3.3.1

What is classification?

Classification in the context of remotely sensed data is to ‘link’ each pixel in the image to one or more user-defined labels so that the radiometric information contained in the image is converted to thematic information such as vegetation, water and built-up. ‘Link’ is a mapping function that constructs a linkage between the raw data and user-defined label set. If the mapping function is a classification technique/algorithm through which each pixel is mapped to a single label, it is ‘one-to-one’ mapping. Classifiers, which perform ‘one-to-one’ mapping are called hard or crisp classifiers. It is also possible to perform ‘one-to-many’ mapping. In this case, each pixel is associated with more than one label, with differing degrees of association between the pixel and each label, and the degree of association is expressed as probabilities of membership. Classifiers that perform ‘one-to-many’ mapping are called soft classifiers. 3.4

CLASSIFICATION METHODS

The process of labelling can be supervised, unsupervised or a combination of both. The supervised labelling method requires the analyst to collect samples to ‘train’ the classifier in order to determine the decision boundaries in feature space. Decision boundaries are significantly affected by the properties and the size of the samples. On the other hand, unsupervised classifiers ‘learn’ the characteristics of each class directly from input data. The classification approaches can be characterized by dichotomies as shown in Table 3.1. 3.4.1

Maximum likelihood classifier

Theoretically the classification problem is that of estimating the a posteriori probability p(i | x), where x is the unknown pixel value and i represents class i. However, in the absence of knowledge of a priori probability the likelihood function p(x | i) itself is used. Hence, the major problem with this classifier is the estimation of the a priori probability. The a priori probability can be estimated either from contextual information or from multi-temporal data. Another problem associated with this classifier is the lack of adequate training samples when a large number of classes and bands are present, the overlap between these classes, and the presence of Table 3.1.

Dichotomies of classification.

Supervised Parametric Fuzzy Assumed probability distribution method Knowledge-based

vs vs vs vs

Unsupervised Non-parametric Crisp Neural Network methods

vs

Purely data oriented

Analysis of remotely sensed data 53 mixed pixels as well as the fact that in real life class boundaries in feature space may be highly complex, which cannot be described as difference of Gaussian probability distributions. 3.4.2

ISODATA

ISODATA algorithm is a migrating means cluster algorithm widely used for automatic image segmentation. This is an unsupervised statistical approach. The analyst has to label the clusters identified by the algorithm. Although widely used, difficulty arises from the fact that the analyst has to estimate the initial number of clusters present in the data. If the initial number of clusters is too small, some significant clusters may go unidentified; if the number is too large, clusters have to be merged. Generally, the latter option is preferred by analysts. 3.4.3

Knowledge-based methods

The methods mentioned above are statistical in nature and depend on users’ inputs in the form of training sets or labelling of clusters. The knowledge of the user is embedded either in the training sets or in the labelling of the clusters. This knowledge is used in conjunction with the statistical measures to perform the classification. The knowledge-based method attempts to incorporate the knowledge of the user in the form of heuristic rules. The hierarchical decision tree method is the most general type of knowledgebased classifier. A hierarchical decision tree classifier is based on the premise that an unknown pattern can be labelled using a sequence of decisions. A decision tree is composed of three basic elements: terminal node or hypothesis represents final classification; interior node or rule representing set of conditions to satisfy the hypothesis; root node or conditions. The advantage of tree classifier lies in the flexibility of defining conditions. The classification methods mentioned here rely solely on spectral characteristics, whereas ‘conditions’ in tree classifier can include ancillary data such as DEM, soil map, etc. along with multispectral data. Figure 3.4 illustrates the decision tree classifier. 3.4.4

Neural network classifiers

The ANN-based classifiers are used to circumvent the complex class boundary problem. The most widely used model is the Back Propagation model, which is a supervised approach. The issues involved in applying ANN models are: ●

●

●

architecture of the ANN that includes elements such as the number of hidden layers and number of neurons in the hidden layer; degree of training – over training may make the network mimic the pattern and not generalize it, whereas under training will cause error in classification; features to be used as inputs – grey levels, auxiliary information, derived contextual information such as texture.

With regard to the number of neurons in the hidden layer, it has been suggested that the number of neurons in the hidden layer should be equal to Np(r(NI ⫹ N0)), where Np is the number of training samples, NI is the number of input features, N0 is the number of output classes and r is related to the noisiness of the data. For unsupervised classification Kohonen and Adaptive Resonance Technique models are employed. For shape classification Hopfield type models are better suited. For classification of highly correlated patterns an iterative variant of Hopfield based on spin glass theory is used. It is to be noted that ISODATA clustering is influenced by the variance of the initial sample distribution to determine class structures, whereas the Kohonen Self organized map cluster structures depend on initial sample distribution.

54

R. Krishnan and B.L. Deekshatulu

Inference flow

Input pixel reflectance

If pixel reflectance in near-IR band >50% True

False

If reflectance in red band <12%

True Vegetation

If pixel reflectance in near-IR band <5%

False

True

?

If reflectance in red band <10% True Water

Figure 3.4.

3.4.5

False ?

False ?

Example of decision tree classifier.

Contextual classifiers

It has been observed that pattern recognition tasks cannot be treated as complete without the satisfactory induction of contextual information either during classification process or as postclassification operation. Context is defined as the local domain from which observations were taken and often includes spatially or temporally related measurements. It is often assumed that contextual relationships decay with distance as it happens in the natural world. The techniques for incorporating contextual knowledge in the classification process include the stacked vector approach which attempts to increase the classification accuracy by including structural measurements such as texture measures as an additional band to a maximum likelihood classifier. The assumption of a normal distribution is often violated by texture measures and may sometimes cause difficulty. 3.4.6

Fuzzy classifiers

The traditional thematic map is used with a presumption that every point on the ground can be labelled as belonging to only one class. Although the discrete categorization is convenient to handle because of its simplicity, it may not be an accurate representation of the real world. The remotely sensed data provides as many as 2kl possible categories of data with k bits per pixel per band and l bands. The crisp classifiers compress nearly continuous measurements into relatively few classes thereby ignoring a certain amount of the information contained in the data to obtain an easy-tohandle simplistic thematic map. The decision-making by crisp classifiers is deterministic. Human language of decision-making is not generally deterministic; it tends to be characterized by a certain level of uncertainty or fuzziness. The same consideration holds good in the classification of imagery. The mislabelling errors of crisp classifiers are due to pixels that show affinity with several information classes. Such pixels are often described as ‘mixed’ pixels.

Analysis of remotely sensed data 55 For example, in an image of agricultural areas there will be some pixels representing more than one crop. The fuzzy classification approach addresses the ‘mixed’ pixel problem by relaxing the discrete membership function of crisp classifiers with the concept of partial membership such that each pixel may simultaneously hold several non-zero membership grades for different labels thereby allowing greater flexibility. The crisp classifiers discussed earlier are modified to incorporate the fuzziness. Commonly used fuzzy algorithms are Fuzzy C-means, Fuzzy maximum likelihood classification, and Fuzzy rule base. The advantage of Fuzzy over crisp classification techniques is that ‘for a given area if the features of interest are agricultural crops, crisp classifier gives acreage estimate as a single number, where as from a fuzzy classified output based on fuzzy membership grade it will be possible to estimate the upper and lower cut off acreage estimates for each crop’. All the techniques mentioned earlier have been used with multispectral data with a reasonable degree of success but their application to hyperspectral data is not straightforward.

3.5

DATA HANDLING METHODS

Whereas data reduction was the focus in the earlier section, the Support Vector Machine (SVM) concept deals with situations where the number of features are small but the class boundaries are complex. As we have seen earlier one way of handling complex boundaries is by using ANN. An alternative approach is to look for simpler boundaries in a higher dimensional space which is created for that specific purpose. SVM, as a concept, was well known in other areas such as character recognition. The original SVM is intended to solve two-class problem and has been extended to handle multi-class problems. Huang et al. (2002) have given a comprehensive comparison of SVM, ANN and DTC and indicated that the performance of SVM improves as number of input bands are increased. The SVM performs better than NNC when seven bands are used (see Figure 3.5). Anthony and Cromp have applied SVM method on AVIRIS data for 4 classes and 16 classes and accuracy rates of 96% and 87% have been reported. 3.5.1

Hyperspectral sensor data handling

The large number of spectral bands complicates their use for classification. The selection of a subset of bands or features is desirable to keep down the volume of data and parameter estimation for classification. New methods and/or modifications to the existing ones are needed to make effective use of the information available in the hyperspectral data sets. 3.5.1.1 Hughes phenomena In the case of Maximum Likelihood Classifier (MLC), it was mentioned that the mean vector and covariance matrix of sample data are used by the classifier (Hoffbeck and Landgrebe, 1996). For n bands the number of elements to be estimated is given by n(n ⫺ 1)/2 ⫹ 2n (Figure 3.6). It is often said that the performance of the classifier can be improved by adding additional features. The performance does improve up to a certain point as additional features (or bands) are added and then deteriorates. This is termed as Hughes phenomena (after its inventor) or peak phenomenon. The explanation for this behaviour (shown in Figure 3.7) is ‘For a fixed sample size, as the number of features are increased, with corresponding increase in number of unknown parameters, even though the separability may increase, the resulting classification accuracy degrades for a fixed sample size.’

(a) 80

Equal sample size, 7 variables

(b)

Equal sample rate, 7 variables 80

75

75

70

70

65

65

60

60 55

55 2

4

6

8

NNC

SVM

10

DTC

MLC

Equal sample size, 3 variables

(c) 80

2

20

SVM

75

70

70

65

65

60

60

55 4 SVM

Figure 3.5.

6

8

NNC

DTC

10

6

8

NNC

10

DTC

20 MLC

Equal sample rate, 3 variables

(d) 80

75

2

4

55

20

4

2

MLC

SVM

6 NNC

8

10 DTC

20 MLC

Classification accuracy (Huang et al., 2002).

Number of elements

30,000 25,000 20,000 15,000 10,000 5000 0 0

50

100

150

200

250

Data dimension

Figure 3.6. Data dimension vs number of elements (number of elements is mean vector and covariance matrix elements) (see Color Plate XIII).

Analysis of remotely sensed data 57

Classification accuracy

N2 > N1 N2 N1

Dimensionality

Figure 3.7.

Accuracy vs dimensionality.

For a linear classifier the number of training samples should be proportional to the number of features for reasonable parameter estimation. For a quadratic classifier, the number of training samples should be proportional to the square of the number of features. 3.5.2

Computer vision-based methods

The overriding importance of shape and context in the interpretation of high resolution imagery has prompted the use of computer vision-based methods. These are broadly in the nature of both pre-processing and special modelling tools. Some of the techniques can be listed as (1) Scale space approaches, (2) Hough Transform and extensions of it and (3) Constraint satisfaction models such as SNAKES. A brief description of these is given in the following paragraphs. 3.5.3

Scale space approach

Scale space approach (Lindeberg, 1994) essentially involves handling the data at different scales. The sealing is achieved by diffusion. This mechanism systematically simplifies the data and removes finer scale details or high frequency information. Diffusion allows a gross image to be created, which will enable a macro understanding of the image. For example, if we are trying to extract ‘roads’ automatically then the trees along the road will hinder this exercise. Scale space allows the creation of an image where the trees are smoothed out and the roads remain. The diffusion can be done isotropically or anisotropically preserving edges. Figure 3.8 shows the main idea behind scale space. The main idea with a scale space representation of a signal is to generate a one-parameter family of derived signals in which the fine-scale information is successively suppressed. Figure 3.8 shows a signal that has been successively smoothed by convolution of Gaussian kernels of increasing width. 3.5.4

Hough transform

A straight line is defined by 2 points A (x1, y1) and B (x2, y2) (Figure 3.8). All straight lines going through the point A are given by the expression y1 ⫽ kx1 ⫹ q for some values of k and q. If we interpret this as a (k, q) space then all the straight lines through A can be represented by the equation q ⫽ y1 ⫺ kx1. Third, straight lines through point B can be represented by q ⫽ y2 ⫺ kx2h

58

R. Krishnan and B.L. Deekshatulu

Coarse scale

Fine scale

Figure 3.8.

Scale space effect achieved through diffusion (Lindeberg, 1994).

q

Y

q = y2 – kx2 B(x2,y2)

q q = –y1 – kx1

A(x1,y1)

X

Figure 3.9.

k

Hough transform.

Initial position

Figure 3.10.

k

Iteration 1

Final position

SNAKES.

The only common point between these two lines in the (k, q) space is the point, which in the original image space represents the only existing straight line connecting points A and B. 3.5.5

Active contour models – SNAKES

SNAKES is defined as an energy minimizing spline – the SNAKES’ energy depends on its shape and location within the image. Local minima of this energy then correspond to desired image objects. ESNAKE ⫽ 兰o ESNAKE (V(s)) ds ⫽ Eint (V(s)) ⫹ Eimage ⫹ Econ

(3.1)

where V(s) is the shape contour, Eint is the internal energy of the spline due to bending, Eimage is a function of edges, lines, etc. and Econ is the external shape constraint. Starting with the initial position (dotted line in the image to the left, in Figure 3.10) iterative energy minimization pulls the contour to its final shape.

Analysis of remotely sensed data 59

Figure 3.11.

Road detection.

A road detection exercise carried out using scale space gives the results as shown in Figure 3.11 (Ramachandran, 2002, personal communication). 3.6

USE OF CLASSIFICATION METHODS IN WATER RESOURCES DEVELOPMENT

Integrated water resources management requires consideration of many disciplines beyond hydrology and watersheds. Space borne multispectral measurements in some cases replaced ground based observations and have provided a quantum jump in our capabilities for water resources management. Remotely sensed data has been used for delineation of water bodies, identification of streams, calculation of flood prone areas, location of suitable water harvesting structures and so on. Classification methods enable these tasks to be automated. Remote sensing classification and expert system methodologies have been used to identify potential ground water sources also using surface expression of geological features. REFERENCES Anthony, G.J. and Cromp, R.F. Support vector machines for hyperspectral remote sensing classification. Available at http://code935.gsfc.nasa.gov/code935/Hyperspectral/Svm_u.pdf Hoffbeck, J.P. and Landgrebe, D.A. (1996) Covariance matrix estimation and classification with limited training data, IEEE Transactions on Pattern Analysis and Machine Intelligence, 18(7), 763–767. Huang, C., Davis, L.S., and Townshend, J.R.G. (2002) An assessment of support vector machines for land cover classification, International Journal of Remote Sensing, 23(4), 725–749. Krishnan, R. (2002) Classification methods. In: Proceedings of International Society for Photogrammetry and Remote Sensing (ISPRS) Commission VII Symposium. Hyderabad, India, December, pp. 48–54. Lindeberg, T. (1994) Scale Space Theory in Computer Vision, Dordrecht: Kluwer Academic Publishers.

CHAPTER 4

Technology Transfer in Remote-Sensing Applications S. Kalluri and P. Gilruth Raytheon, 1616 McCormick Drive, Upper Marlboro, MD 20774, USA

4.1

INTRODUCTION

Remote-sensing data and applications have become an important part of the decision-making process within several US federal, state, and local agencies. Remote-sensing applications such as land use-land cover planning, natural resource management, and disaster management are common among these agencies. The requirements of remote-sensing data such as the spatial, temporal, and spectral resolution differ between applications (Kalluri et al., 2001). With the availability of remote-sensing data from government agencies and private companies around the world, users now have the ability to choose data that suit their specific requirements and budget from a variety of sensors and data providers. Within the United States, the National Aeronautics and Space Administration (NASA) has been working in partnership with several public, private, and academic organizations to implement projects that demonstrate practical uses of NASA sponsored observations from remote-sensing systems and predictions from scientific research (NASA, 2002). One of the goals of NASA’s Earth Science Enterprise Application Division’s Strategy is to expand and accelerate the realization of societal and economic benefits from earth science, information and technology (NASA, 2003). The purpose of this chapter is to share examples of latest developments in the process of moving remote-sensing applications from NASA’s experimental science satellites to mainstream users. One of NASA’s objectives is to explore the use of remote-sensing data collected by its Earth Observation System (EOS) satellites for the development of applications that are relevant to federal, state, local, and tribal agencies. To meet this objective, the Synergy program was started in 2000. The program includes six application areas: agriculture, urban planning, natural resource management, water resource management, disaster management, climate and human health (Kalluri et al., 2003). The Synergy program is a partnership among end users, academia and industry, and has a three-fold objective: ●

● ●

Identify issues at the federal/state/local/tribal governments that can be addressed by using remote-sensing data and determine users’ requirements. Develop sustainable remote-sensing applications using EOS data and demonstrate benefits. Educate and train the end users in using these technologies to promote these applications within their communities.

As a part of the Synergy program, scientists at different US universities have been developing products primarily from NASA’s remote-sensing data, based on the analysis of specific user requirements. These products are then distributed through interactive web-based tools hosted at 11 centers called InfoMarts. During the evolution of the Synergy program, universities explored a variety of strategies to develop, promote, and sustain these applications. This chapter describes the application life

S. Kalluri and P. Gilruth

en t

at io n ge ne r od uc t

al fe as i

de ve lo pm go rit hm Al

e

t da up

n tio ca

System benchmarking

Operational use

d an

i rif ve

Application development and demonstration

e in ef R

d an

g in er ne gi en

Te ch ni c

Production and deployment

n io at lid Va

s em st

Requirements definition

Data characteristics

Figure 4.1.

Prototype development

Sy

bi lit y

User needs and technical feasibility analysis

Pr

62

Distribute and Archive

Remote sensing application development life cycle.

cycle development process, and discusses lessons learned in promoting remote-sensing technologies for application development based on the Synergy program experiences. Since the mandate of the Synergy program is to develop applications using EOS data, the life cycle process and issues discussed here frequently refer to NASA data. Nevertheless, the framework discussed here is broadly applicable to all remote-sensing applications, regardless of who the data provider is. 4.2

APPLICATION DEVELOPMENT LIFE CYCLE

The applications development life cycle can be divided into three stages: user needs and technical feasibility analysis, prototype development, and production and deployment (Figure 4.1). These stages are similar to other commonly used project development life cycles in government and industry (Forsberg et al., 1996; PMI, 2003). 4.2.1

User needs and technical feasibility analysis

Understanding and documenting user requirements is fundamental to developing a successful remote-sensing application. A single application such as a land use-land cover classification could have a variety of users. But the desired classification schema among users could be quite different. Building a common data set that satisfies multiple user requirements is therefore essential for the application to have a broad usage and impact. Tools such as the Quality Function Deployment (QFD) (Akao, 1995; ReVelle et al., 1998) have been shown to be effective in capturing, prioritizing, and defining common requirements among users for remote-sensing applications (see, e.g. http://icrest.missouri.edu/Projects/Infomart/QFDProject/ index.htm). The main requirements to be identified upfront include spatial resolution, frequency of product generation, desired accuracy, geographic extent, and data format. All these factors determine the cost and feasibility of the application. User needs assessments involve understanding not only the technical capabilities of the user organizations but also their institutional and business practices. Understanding organizational structure and management approaches in handling geospatial data by end user agencies is important while building geospatial data clearinghouses. Protocols in decision making and policies related to the use and distribution of geospatial data among different organizations should be considered while developing an application. It is important to understand how different data sets can be integrated seamlessly in an interoperable environment. Establishing common data and metadata standards among partner agencies allows easy sharing of data. For example, city governments map the cities, and county governments in turn incorporate the city maps in their county maps. This hierarchy of geospatial data aggregation is carried out in the state and federal databases (Decker, 2003). Such a seamless vertical and horizontal integration is only possible if all agencies adhere to common data standards.

Technology transfer in remote-sensing applications 63 It is important for the user agency to have adequate infrastructure and capacity to handle remote-sensing and other geospatial data for successful adoption of these technologies. Availability of resources and capital, familiarity with application, familiarity with technology, economic feasibility, and budget constraints have to be identified. Because remote-sensing data sets tend to be large and require complex processing, data processing speed, network connectivity, and archival capacity require detailed systems engineering. Acquiring sophisticated sensors and establishing ground receiving stations is expensive. Before large investments are made into building a comprehensive system that includes data collection, processing, and interpretation, basic questions such as, “Is the application technically feasible? Can we build the system? If we build it, will they use it?” should be answered. 4.2.2

Prototype development

Building a prototype before operational deployment of an application has several advantages. These include: ●

● ● ● ● ● ●

validation and verification of the algorithm – the application should provide accurate and consistent results; benchmark performance – provides sizing information for an operational system; receiving user feedback; enhance communication between application developers and users; identify limitations and problems at an early stage in development; provide a better understanding of detailed requirements; demonstrate cost benefits.

To the extent possible, the prototype application should match the function and performance of the final system. 4.2.3

Production and deployment

Production and deployment is the final step in the life cycle process. An operational application is a result of ultimate convergence of data, algorithms, and end user needs. The system has to be tested and approved by all stakeholders before going operational. A strategy for routine validation and verification of data products should be in place to ensure the integrity of processing algorithms and results. Documenting the complete data processing procedures from ingestion to results and model outputs is the groundwork for future system enhancements. The documentation should include a detailed description of algorithms and assumptions. A mission critical operational application system in general should meet the following requirements: ● ● ● ●

deliver data and products on schedule following predetermined deadlines; robust with redundant hardware; software should be easily maintainable with accurate documentation; continuously monitored for uninterrupted service.

Since computer technologies are rapidly improving, sufficient resources should be allocated for maintenance and upgrade of the hardware and software that run the applications on a periodic basis. 4.3

BARRIERS IN REMOTE-SENSING TECHNOLOGY TRANSFER

The term “Crossing the Valley of Death” is often used to refer to the transition from research and development to applications. Experience has shown that factors such as inadequate

64 S. Kalluri and P. Gilruth planning result in transition breakdown (NRC, 2000). Bottlenecks and hurdles such as lack of awareness by end users of remote-sensing technology, lack of proper feedback mechanisms between application developers and end users, institutional barriers, and unproven cost benefits have to be overcome to cross this valley (Kalluri et al., 2003). These issues are discussed next. 4.3.1

User involvement

Early and continuous involvement of the users is critical for the success of application development, and ultimately for realizing an impact on decision making. The products and data sets that are developed have to be driven by a compelling requirement by the user community. Adequate feedback mechanisms should exist through periodic meetings and workshops between developers (e.g. scientists in academia and government, industry) and product users for continuous innovation and improvement. Workshops with users from multiple agencies help in overcoming institutional barriers by bringing together different user groups with similar requirements and facing similar challenges. Such workshops provide a forum for exchange of ideas among various groups. Developers should ensure progressive alignment of products between meetings with the end users. There is a strong tendency to conduct user surveys through postal and email survey questionnaires. While these mechanisms are useful, they are not as effective as regular face-to-face dialogues between application developers and their users, where the users have the opportunity to engage in a collective discussion. Within the Synergy program, the rangeland monitoring application, “RangeView,” followed an incremental approach where lead users were involved in defining the requirements, testing the prototype application, and providing feedback on the functionality and usability of the application before it was made operational. RangeView (http://rangeview.arizona.edu/) utilizes web-based tools to characterize vegetation and landscape dynamics using the Normalized Difference Vegetation Index (NDVI) time-series animations of current/past conditions of vegetation greenness for rangeland management. Feedback from lead users was gathered during workshops that were designed for the users to interact with the web-based tools and test the functionality. This feedback mechanism enabled the application developers to understand the applicability of the RangeView tools in relation to the users needs. Within a technology adoption life cycle, users can be classified as innovators, early adopters, early majority, late majority, and laggards based on their level of adoption (Moore, 1999). Adopters of remote-sensing applications can also be similarly classified. Winning the support of the early adopters who can showcase the benefits of remote-sensing technology over conventional methods of resource analysis is essential to a wider adoption by a larger user group. Users must therefore be involved as stakeholders within the entire project life cycle, and the application developers should factor in stakeholder’s interests for successful transfer of technology from application to decision making or policy making. 4.3.2

Training and education

At present, remote-sensing technologies are not widely used within state/local/tribal agencies. Not surprisingly, training and educating users is critical for success. User familiarity in remotesensing techniques also helps in building an informed consensus between end users and developers on the product specifications. Training the end users to correctly interpret and apply the results for decision making is as important as developing the application itself (Seelan et al., 2003). Typically, decision makers at state and federal agencies are familiar with Geographic Information System (GIS) data and applications. However, their expertise in handling satellite remote-sensing data is often limited. An end user education strategy should therefore be developed along with the application. If the end users are not aware of the full potential of the application, the entire development may

Technology transfer in remote-sensing applications 65 come to naught. Training sessions and seminars with hands-on tutorials are better than “read the manual” approach. After the development of a technique or a data product routine operational activities should be handed off to the end users or some other intermediary such as a non-profit organization or a private company thereby freeing up the research community to pursue additional improvements. The basic model adopted for the Synergy program was that the scientists from academia would develop and prototype an application by working closely with the end users. After successful demonstration of the benefits of remote-sensing products, these products and methods would then be transferred to the user agencies that would then adopt and maintain them. Training and education are therefore critical to transfer of technology and knowledge from the application developers to the end users within state/local/tribal agencies. Establishing peer learning groups in which an advanced user of a product teaches his peers within the same user community enhances product diffusion (Seelan et al., 2003). For example, a group of farmers is more comfortable adopting a product when a fellow farmer endorses it and shares his positive experiences with them compared to marketing by an outsider. 4.3.3

Data accessibility

Data characteristics and acquisition strategy are driven by user requirements and costs. The cost of remote-sensing data rises exponentially with spatial resolution (Figure 4.2). Data from US government satellites such as Landsat 7 with a spatial resolution of 30 m cost less than $0.02/km2, while acquisition of high-resolution satellite imagery with 2.4 m spatial resolution from private vendors could cost as much as $24/km2. Data policy can be a determining factor for selecting appropriate data for the application. Remote-sensing imagery at a spatial resolution of 1 m or higher with accurate positional accuracy is required for a variety of state and local applications such as the creation of base maps, corridor mapping, and land use-land cover mapping. Imagery from NASA satellites, which are primarily designed for global change research has no copyright restrictions. However, high spatial resolution data from commercial vendors (e.g. Space Imaging, Digital Globe) have a variety of restrictions for use and sharing, and prices vary by the number of users that are

30.00 y = 25.986e–0.2482x R 2=0.9663

Cost ($/sk.km)

25.00 20.00 15.00 10.00 5.00

0

10

20

30

40

50

Spatial resolution (m)

Figure 4.2. Cost of satellite data at different spatial resolutions (Prices of Landsat, SPOT, Ikonos, and QuickBird are shown here. These are the prices published on the Internet by the vendors at the time of publication and are subject to change).

66 S. Kalluri and P. Gilruth allowed to share the data. State and local government agencies typically like to acquire and distribute data to a variety of users without any restrictions and therefore prefer to acquire aerial photography by contracting with aerial photo firms. High-resolution orthoimagery is a key data layer in national geospatial databases such as The National Map (Kelmelis et al., 2003). Remote-sensing imagery often consists of large digital files. Applications that rely on continuous, operational satellite data require real-time acquisitions. Fast network connectivity is therefore essential for efficient transfer of imagery among data providers, developers, users, and decision makers. 4.3.4

Data continuity

In tune with other technologies, the remote-sensing technology is evolving. NASA is continually improving sensors with each generation of satellites. A number of remote-sensing applications require data collected over multiple years for temporal analysis to observe trends. NASA’s EOS satellites are often precursors to operational monitoring satellites managed by NOAA. Since NASA is not an operational agency, there is rarely a redundancy in data collection or backup satellites in case an operating sensor or satellite fails within its planned lifetime. Also, some EOS instruments (e.g. Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), Multi-angle Imaging Spectro-Radiometer (MISR)) are unique and experimental in nature, and data continuity is not always assured. In some cases, transitioning research missions that have demonstrated opportunities for practical applications into operational missions have been either slow or have gone unrealized altogether (NRC, 2003). Users expect consistency in data format and content over time and need to be assured of data longevity before they are willing to switch to remote-sensing data products and techniques for operational decision making. Decision makers and policy makers require data sets and techniques that can provide consistent results each time the tool is used. It is also critical that the remote-sensing data have the right spectral, spatial, and temporal characteristics and be economically viable in terms of cost for successful application development. Because of these factors, EOS data may face difficulties in being accepted universally by all decision makers who rely on remote-sensing data unless data continuity is assured. 4.3.5

Data formats and standards

Remote-sensing data and products are generated by various agencies in a variety of formats. These formats may be suitable for researchers, but end users in non-research organizations require data in a format that can be readily displayed and analyzed using simple software packages. End users at the state/local/tribal level typically do not have access to high-end image processing software and tools. There are a variety of software packages that can process remotesensing and GIS data. While both GIS and image processing packages can handle either data their functionality is more extensive within the specific domain for which they are designed. Certain data formats such as the Hierarchical Data Format (HDF) that are specifically designed for scientific data management are not widely supported by commonly used commercial GIS software packages. The users of these GIS software have to translate satellite imagery in HDF format into a format that their software can handle. Products from remote-sensing imagery should therefore be generated in formats that can be readily ingested into a variety of popular low-cost image visualization and GIS packages and should be compatible with user’s existing systems and infrastructure. Besides data formats, adherence to common, interoperable data standards is also crucial for a seamless integration of remote-sensing and non-remote-sensing data from a variety of sources. The importance of uniform data standards can be illustrated through an example from the Synergy project. When the University of Missouri, a Synergy partner, was developing common

Technology transfer in remote-sensing applications 67 base maps using high-resolution satellite data for city and county agencies in Missouri, a major stumbling block was the integration of existing GIS layers with the imagery. The positional accuracy of the existing line-work within the city government’s vector GIS database was variable, and the vector layers were not in alignment with the more accurate imagery. Different agencies within the county had dissimilar editing procedures that created discrepancies between layers that should share the same geography. Local governments are reluctant to use or show imagery if the imagery displays significant errors in their data. The ability to “correct” these errors in a systematic and repeatable process was critical to the creation of accurate base maps in this project. This correction enabled the county’s investment in these data layers to be maintained into the future. The program was developed and implemented according to a plan developed within an Interagency GIS Policy Board. The key players included the information technology departments, the planning departments, and surveying departments. Achieving interoperability among data sets in this example also resulted in improvements in inter-institutional collaboration. 4.3.6

Proven benefits

One of the major obstacles to date for adoption of remote-sensing data for applications has been the lack of proven benefits, particularly in relation to cost. Adopting remote-sensing imagery requires investment in hardware, software, and training. Before a group of users adopt this technology as an alternate to existing methods they need to be convinced that remote-sensing solutions are economically viable and or enable the realization of other environmental benefits. Using traditional financial cost benefit analysis models, it is feasible to show the monetary savings for applications that directly use remote-sensing techniques for regulation and allocation of resources and where there is direct impact on marketable products such as crops and timber. It is, however, more difficult to assign a monetary value to benefits to the environment and quality of life which resulted from policy changes. Monetizing indirect benefits from using remote sensing data is complex and challenging. Nevertheless, demonstrating tangible benefits of remote sensing through either cost savings or cost avoidance would build a strong case for investing in these technologies by government agencies. Within the Synergy program, projects that had a direct impact on the users’ decision-making practices were more quickly embraced by the user community than the others. For example, the Idaho Department of Water Resources (IDWR) was in need of a system that would allow them to monitor irrigation within the Snake River Basin. The algorithms developed by the University of Idaho to estimate evapotranspiration (ET) from irrigated areas using Landsat data were able to adequately determine consumptive water use, and IDWR adopted this application into their business practices. In another example, the University of Texas at Austin, a Synergy partner university, developed maps showing the extent of invasive saltcedar infestation within the Pecos and Colorado rivers in Texas using Landsat data for eradication. The University of Texas developed detailed cost benefit analysis that showed that the Texas Department of Agriculture could realize $32.10 cost savings per acre of land treated by selective aerial herbicide application through the use of remote-sensing technology. Such an analysis was necessary for the state agency to support these activities. 4.3.7

Incubation period

New users of remote-sensing technology are not willing to invest unless the application has been proven to be a viable alternative to current practices. Experiences from the Synergy program show that 3–5 years are required before end users gain confidence in using remote-sensing data and institutionalize their use. Diffusion and adoption of these technologies is faster in agencies that have prior experience (even if limited). During the initial phases of application diffusion,

68

S. Kalluri and P. Gilruth

the end users need hands-on training and advice, access to free data and processing software, technical support, and sometimes access to equipment as well. The rate of diffusion is directly related to the perceived value that remote-sensing data has by users responsible for policy development or implementation. 4.3.8

Collection of metrics

“One can manage only what can be measured.” This maxim is also true of remote-sensing applications development, although there is no set of universally accepted metrics for all applications due to their specific context. Nevertheless, the collection of metrics for reporting externally to funding sources and to improve internal management of applications development is receiving growing interest since federal agencies such as NASA must demonstrate the value of their investment to the public. Some generalized metrics that could be used to gauge the performance and diffusion of remote-sensing application development include identifying ● ● ● ● ●

changes in policies and impacts of new policies; cost savings or cost avoidance; changes in the numbers of users of a specific data set; the demand for a specific dataset or application; willingness of end users to invest in remote-sensing technology.

Metrics associated with policy impact are most difficult to quantify but would have the most significance. Other metrics such as web transaction logs showing the number and volume of data sets distributed from a data center are easy to quantify and help in tracking the number of users and the data sets that are more popular. Some of the measurable impacts of an application on decision making include factors such as the reliability and quality of decisions, time saved in using these new techniques, and the number of instances a decision has been made as a direct use of the new product compared to existing approaches to decision making. Improvements in the quality of the environment as a result of policy changes due to the adoption of remote-sensing products can be readily measured and quantified. Firm, time-bound goals and project milestones can be difficult to define, implement, and maintain for a science and research project. However, if the goal of application development is to take established remote-sensing algorithms or data sets and turn them into applications, then time-bound goals and milestones help set expectations between application developers and end users, and prevent the project from drifting off course on personal research agendas. The project milestones should be mapped to the project life cycle development steps. Collection of metrics to improve internal project management may appear initially burdensome to applications developers. However, these metrics can help towards long-term sustainability of the project by ensuring that resources stay focused on applications that are most likely to succeed. 4.3.9

Sustainability

A successful remote-sensing application should be sustainable. The Synergy experience suggests that 3–5 years of project life span are required to advance from gathering user requirements to prototyping and developing an application, training users in the appropriate use of products, and winning user confidence. Factors such as proven benefits, ease of use, timeliness of data, and continuity of data contribute to the sustainability of remote-sensing applications. Among various applications of medium and coarse spatial resolution remote-sensing data, numerical weather forecasting, disaster mitigation, and climate prediction have found markets within public and private sectors because of recognizable economic and social benefits

Technology transfer in remote-sensing applications 69 (Williamson et al., 2002). Other applications such as urban planning and precision agriculture are more localized and require imagery at a higher spatial resolution, which are more expensive than medium or coarse resolution data. There is no single model for reaching sustainability because of the uniqueness of each application and its user group. However, within the Synergy program three distinct categories of sustainability models have begun to emerge based on the type of users and sources of funding: Federal/state/local/tribal/non-government organizations. State and federal agencies (e.g. US Department of Agriculture, US Forest Service) that have been benefiting from remote-sensing data for monitoring large areas for environmental quality and compliance policies have invested in medium resolution remote-sensing applications. Some agencies have started to acquire highresolution imagery over selective areas, but the higher cost remains an issue. Private organizations. Commercial entities responsible for managing large landholdings such as timber companies fall within the second group of users. Private companies convinced of the benefits of remote-sensing data for their operations eventually begin to build the infrastructure and technical capability within their organizations to process remote sensing data. They however, would likely depend on universities and NASA for research and development of data processing algorithms. Private individuals and community based organizations. Individual users such as farmers and ranchers have shown great interest in using remote-sensing data for making informed decisions in managing their lands for better productivity. Sustainability of remote-sensing applications within these user groups is more challenging since this group typically does not have the technical and financial resources to readily absorb remote-sensing technologies. This user group tends to rely on well-established remote-sensing data centers in the government, universities, or the private sector companies to receive remote-sensing data and products. Nevertheless, private individuals have the ability to influence public agencies in providing remote-sensing data as they gain awareness of the value of these data for their particular need, which in turn could promote the sustainability of an application. Community based organizations such as farmer cooperatives or local environmental groups also play an active role in influencing federal, state, and local governments in setting funding priorities for remote-sensing activities. As the user base for remote-sensing products is dynamic, application developers should be flexible in their approaches to sustain their operations. For example, working with individuals requires an effective outreach or extension campaign, whereas targeting a state agency might involve a detailed analysis of current organizational processes. Remote-sensing technology is still dominated by academic institutes and government organizations pursing research. Adopting models for achieving sustainability that are based on user interaction at each step in the development process suggests a shift away from the more traditional, technology-driven, research process in applying remote sensing. This paradigm shift requires training/awareness building targeted at each step within the project development cycle among the application developers as well as the users. 4.4

SUMMARY

Incorporating a structured framework for developing an operational remote-sensing application would increase the success of transition from research to applications. A defined life cycle development process would also allow for a planned and systematic allocation of resources and ensure that the user requirements are adequately satisfied. A strategy to gather user requirements, including a good understanding of users’ decision-making processes, organizational structures and data sharing protocols is essential when choosing the appropriate remote-sensing data. Benchmarking the application for performance and collecting metrics that show the

70 S. Kalluri and P. Gilruth benefits of the application are necessary to win users’ confidence and sustain the application. A mechanism to refine the application based on user’s feedback and routine validation and verification is important to maintain the quality of the application and develop new applications. Frequent dialog between application developers and end users, education and training of end users, delivering products and data to end users in formats that best suit them, and collection of metrics and their usage to manage application development ensure the success of application development using remote-sensing data. ACKNOWLEDGMENT This work is funded by NASA Synergy project under the ECS contract NAS5-60000. REFERENCES Akao, Y. (ed.) (1995) Quality Function Deployment: Integrating Customer Requirements into Product Design. University Park, IL: Productivity Press Inc., 387pp. Decker, D. (2003) The national map “of Texas” an example of statewide application. Photogrammetric Engineering and Remote Sensing, 69, 1147–1153. Forsberg, K., Mooz, H., and Cotterman, H. (1996) Visualizing Project Management. New York: John Wiley & Sons. Kalluri, S. et al. (2001) Remote sensing applications for operational decision making at local scales: current status and future opportunities for agriculture and disaster management applications. International Geoscience and Remote Sensing Symposium, Sydney, Australia, 9–13 July. Kalluri, S., Gilruth, P., and Bergman, R. (2003) The potential of remote sensing data for decision makers at the state, local and tribal level: experiences from NASA’s Synergy program. Environmental Science and Policy, 6, 487–500. Kelmelis, J.A. et al. (2003) The national map from geography to mapping and back. Photogrammetric Engineering and Remote Sensing, 69, 1109–1118. Moore, G.A. (1999) Crossing the Chasm. New York: Harper Business, p. 227. National Aeronautics and Space Administration (NASA) (2002) Earth Science Enterprise Application Strategy. Washington DC, 17pp. (http://earth.nasa.gov/visions/appstrat2002.pdf). National Aeronautics and Space Administration (NASA) (2003) Earth Science Enterprise Strategy. Washington DC, 74pp. (http://earth.nasa.gov/visions/ESE_Strategy2003.pdf). National Research Council (NRC) (2000) From Research to Operations in Weather Satellites and Numerical Weather Prediction: Crossing the Valley of Death. Washington DC: National Academy Press, 80pp. National Research Council (NRC) (2003) Satellite Observations of the Earth’s Environment. Washington DC: National Academy Press, 163pp. Project Management Institute (PMI) Inc. (2003) US Department of Defense Extension to: A guide to the Project Management Body of Knowledge, Version 1.0. Fort Belvoir: Defense Acquisition University Press. ReVelle, J.B., Moran, J.W., and Cox, C.A. (1998). The QFD Handbook. New York: John Wiley & Sons, 432pp. Seelan, S.K., Laguette, S., Casady, G.M., and Seielstad, G.A. (2003) Remote sensing applications for precision agriculture: a learning community approach. Remote Sensing of Environment, 88, 157–169. Williamson, R.A., Hertzfeld, H.R., Cordes, J., and Logsdon, J.M. (2002) The socioeconomic benefits of earth science applications research: reducing the risk and costs of natural disasters in the USA. Space Policy, 18, 57–65.

Part 2

Remote-sensing data applications

CHAPTER 5

Computing and Mapping of Evapotranspiration Richard G. Allen and Masahiro Tasumi University of Idaho, Kimberly Research Station, Kimberly, ID 83341, USA Anthony Morse and William J. Kramber Idaho Department of Water Resources, Boise, ID 83720-0098, USA Wim Bastiaanssen WaterWatch, General Foulkesweg 28, 6703 BS Wageningen, The Netherlands

5.1

INTRODUCTION

METRIC (Mapping Evapotranspiration at high Resolution and with Internalized Calibration) is an image-processing tool for calculating Evapotranspiration (ET) as a residual of the energy balance at the earth’s surface. METRIC is a variant of the important model SEBAL, an energy balance model developed in the Netherlands and applied worldwide by Bastiaanssen (1995, 2000) and Bastiaanssen et al. (1998a,b, 2004). METRIC has been extended to provide tighter integration with ground-based reference ET and has been applied with Landsat images in southern Idaho to predict monthly and seasonal ET for water rights accounting and for operation of groundwater models. METRIC has also been applied in the Imperial Valley of southern California and along segments of the Middle Rio Grande river of New Mexico. ET “maps” (i.e. images) provide the means to quantify, in terms of both the amount and spatial distribution, the ET on a field-by-field basis. Results from METRIC have been compared and validated using precision-weighing lysimeter measurements from the US Department of Agriculture-Agricultural Research Service (USDA-ARS) at Kimberly, Idaho, and from Utah State University for the Bear River. ETs for periods between satellite overpasses were computed using ratios of ET from METRIC to reference ET computed for ground-based weather stations. ET maps via METRIC provide the means to quantify, in terms of both the amount and spatial distribution, ET from individual fields. The ET images generated by METRIC show the spatial and seasonal changes in ET during the year. Initial application and testing of METRIC indicate substantial promise as an efficient, accurate, and relatively inexpensive procedure to predict the actual evaporation fluxes from irrigated lands throughout a growing season. ET from satellite images may replace current procedures used by Idaho Department of Water Resources (IDWR) and other management entities that rely on ground-based ET equations and generalized crop coefficients that have substantial uncertainty. METRIC and SEBAL represent a maturing technology for deriving a satellite-driven surface energy balance for estimating ET from the earth’s surface. This technology has the potential to become widely adopted and used by the world’s water resources communities. ET maps created using METRIC, SEBAL, or similar remote-sensing-based processing systems will some day be routinely used as input to daily and monthly operational and planning models for reservoir

74

R.G. Allen et al.

operations, ground-water management, irrigation water supply planning, water rights regulation, and hydrologic studies. The reasons why METRIC and SEBAL are attractive to our applications in western United States are as follows: ●

●

●

●

METRIC and SEBAL calculate actual ET rather than potential ET and do not require knowledge of crop type (no satellite-based crop classification is needed). METRIC and SEBAL rely heavily on theoretical and physical relationships but provide for the introduction and automated calibration of empirical coefficients and relationships to make the process operational and accurate. The use of ETr in calibration of METRIC and the use of ETrF in extrapolation to 24-h ET provides general equivalency and congruency with ET as estimated using the traditional Kc ETr (or Kc ETo) approach. This is valuable for use of ET maps generated by METRIC water rights management where water rights are based on previous Kc ETr calculations. METRIC is auto-calibrated for each image using ground-based calculations of ETr (made using weather data) where accuracy of the ETr estimate has been established by lysimetric and other studies in which we have high confidence.

Internal calibrations of the sensible heat computation within SEBAL and METRIC eliminate the need for atmospheric correction of Ts or reflectance (albedo) measurements using radiative transfer models (Tasumi et al., 2004a). The internal calibrations also reduce the impacts of any biases in estimation of aerodynamic stability correction or surface roughness. The IDWR and the University of Idaho have developed a variety of METRIC applications. In Idaho, METRIC has been used to monitor water-right compliance and aquifer depletion as a tool for water resource planning and in hydrologic modeling (Morse et al., 2000, 2004). In the Rio Grande Valley of New Mexico, METRIC has been used to map ET from riparian vegetation. In the Imperial Valley of California, METRIC ET maps are used in irrigation management.

5.2

THEORETICAL CONSIDERATIONS

The theoretical and computational approaches of SEBAL and METRIC are described in Bastiaanssen et al. (1998a), Bastiaanssen (2000), Morse et al. (2000), and Tasumi et al. (2004b). By using an energy balance at the surface, energy consumed by the ET process is calculated as a residual of the surface energy equation: LE ⫽ Rn ⫺ G ⫺ H

(5.1)

where LE is the latent energy consumed by ET, Rn is net radiation (sum of all incoming and outgoing shortwave and longwave radiation at the surface), G is sensible heat flux conducted into the ground, and H is sensible heat flux convected into the air. The utility of using energy balance is that actual ET rather than potential ET (based on amount of vegetation) is computed so that reductions in ET caused by shortage of soil moisture are captured. Nevertheless, the computation of LE is only as accurate as are the values for Rn, G, and H. The algorithms used in METRIC for Rn and G are similar to those described for SEBAL by Bastiaanssen et al. (1998a) and the reader is referred to this and to Tasumi et al. (2004b) for detail. Basically, Rn is computed from satellite-measured broadband reflectances and surface temperature; G is estimated from Rn, surface temperature, and vegetation indices; and H is estimated from surface temperature ranges, surface roughness, and wind speed using buoyancy corrections.

Computing and mapping of evapotranspiration 75 METRIC differs from SEBAL principally in how the “H function” is calibrated for each specific satellite image. In both METRIC and SEBAL, H is predicted from an aerodynamic function where: dT H ⫽ Cp r ah

(5.2)

where  is air density, Cp is specific heat of air at constant pressure, and rah is aerodynamic resistance between two near-surface heights (generally 0.1 and 2 m) computed as a function of estimated aerodynamic roughness of the particular pixel and using wind speed extrapolated to some blending height above the ground surface (typically 100–200 m), with an iterative stability correction scheme based on the Monin–Obhukov functions (Allen et al., 1996). The dT parameter represents the near-surface temperature difference between the two near-surface heights. Because of the difficulties in estimating surface temperature (Ts) accurately from satellite due to uncertainties in atmospheric attenuation and contamination and radiometric calibration of the sensor, dT is estimated as a relatively simple linear function of Ts: dT ⫽ a ⫹ bTs

(5.3)

Bastiaanssen (1995) and Bastiaanssen et al. (2004) provide the rationale and empirical evidence for using the linear relation between dT and Ts. The application of equation 5.3 appears to extend well across a range of surface roughnesses, because as roughness increases and rah reduces, given the same H, dT reduces due to more efficient transfer of H, and Ts reduces for the same reason. In most applications of SEBAL (Bastiaanssen et al., 1998a,b), parameters a and b in equation 5.3 are computed by setting dT ⫽ 0 when Ts is at the surface temperature of a local water body (or in its absence, a well-vegetated field) where H is expected to be zero, and by setting dT ⫽ (H rah)/( Cp) at Ts of a “hot” pixel that is dry enough so that one can assume that LE ⫽ 0. From equations 5.1 and 5.2, dT ⫽ ((Rn ⫺ G) rah)/(  Cp) at the “hot” calibration pixel. In METRIC, the same approach and assumptions are made for the hot pixel as in SEBAL, although a daily surface soil water balance is run for the hot pixel to confirm that ET ⫽ 0 there or to supply a nonzero value for ET for the hot pixel for calibration of equation 5.3. For the lower calibration point of dT in METRIC, a well-vegetated pixel having relatively cool temperature is selected and dT at that pixel is calculated as: dT ⫽

(Rn ⫺ G ⫺ k ETr) rah Cp

(5.4)

The a and b coefficients are determined using the two values for dT paired with the associated values for Ts. With Landsat images, fields of alfalfa or other high leaf area vegetation can generally be identified that are close to or at full cover, so that the ET from these fields can be expected to be near the value of “reference ET” (ETr) computed for an alfalfa reference. In METRIC, we use the standardized ASCE Penman–Monteith equation for alfalfa reference (ASCE-EWRI, 2002), which is typically 20–30% greater than grass reference ET (ETo). The k factor in equation 5.4 is set to 1.05 because we assume that a viewed field having high vegetation and that is colder than average temperature, as compared to other high vegetation fields, will have ET that is about 5% greater than ETr due to higher surface wetness or merely due to its rank within the population of alfalfa fields (or other highly vegetated areas). Generally, METRIC is applied without crop classification, so that specific crop type is generally not known. METRIC and SEBAL, when applied with Landsat images, generally differ somewhat in how ET for the adjoining 24-h period is estimated given the essentially instantaneous ET calculated at the time of the satellite image (generally during late morning). In SEBAL, the evaporative

76 R.G. Allen et al. fraction (EF), defined as the ratio of ET to (Rn ⫺ G), is assumed to be the same at both the observation time and for the 24-h period. The assumption of constant EF can sometimes underpredict 24-h ET in arid climates where afternoon advection or increases in afternoon wind speeds may increase ET in proportion to Rn. In METRIC, the extrapolation from observation time to the 24-h period is done using the fraction of reference ET (ETrF) rather than EF. ETrF is defined as the ratio of ET to ETr (in the case of METRIC, alfalfa reference), and is essentially the same as the well-known crop coefficient, Kc (for an alfalfa reference basis). The assumption of constant ETrF during a day may be better able to capture impacts of advection and changing wind and humidity conditions during the day, as expressed in the ETr calculation (which is done hourly and summed daily). Trezza (2002) and Romero (2004) demonstrated the general validity of constant ETrF during a day using lysimeter data from Kimberly. 5.3

VALIDATION

Precision-weighing lysimeters were used for validation. One set of lysimeters was located in the Bear River Basin, the other at the Kimberly Research Station. 5.3.1

Lysimeters at Montpelier, Idaho

In Phase I (2000) of our study, ET maps were generated monthly for a 500 km ⫻ 150 km area (comprised of 2 Landsat images) encompassing the Bear River Basin. Images were processed for 1985, coinciding with an ET study using lysimeters (Hill et al., 1989) that allowed for comparison to METRIC. Lysimeters near Montepelier, Idaho, just north of Bear Lake, had been planted to an irrigated native sedge forage crop characteristic of the area and local surroundings. The lysimeters were measured weekly. ETs from the three lysimeters were averaged to reduce random error and uncertainty in the ET measurements. Results for four satellite images during the 1985 growing season (July 14, August 15, September 16, October 18) are summarized in Figure 5.1 and Table 5.1. The results compare well to lysimeter data for the last three image dates. The earliest

1.4

Ratio of ET to ETr

1.2 1.0 0.8 0.6 0.4 0.2 0 170

200

230 Julian date

Average ET/ETr by lysimeter

260

290

ET/ETr by METRIC on image date

Figure 5.1. Comparison of ETr fractions (i.e. Kc) derived from 7-day lysimeter measurements near Montpelier, Idaho during 1985 and values from METRIC for 4 Landsat dates (ETc⫽ crop ET and ETr ⫽ alfalfa reference ETr) (see Color Plate XIV).

5.3 3.5 1.9 0.7

2.9

July August September October

July–October

7-day lysimeter ET average for image date (mm d⫺1)

0.73

0.98 0.59 0.57 0.49

METRIC ETr F on image date

3.3

6.8 3.7 2.1 0.6

7-day METRIC ET for image date (mm d⫺1)

15

28 6 10 ⫺14

Difference in 7-day ET (METRIC – Lysimeter) (%)

563

202 201 115 45

Monthly alfalfa ETr (mm)

405

198 119 66 22

METRIC monthly ET (mm)

388

167 145 54 23

Lysimeter monthly ET (mm)

4

19 ⫺18 22 ⫺5

Difference in monthly ET (METRIC – Lysimeter) (%)

Table 5.1. Summary of METRIC- and lysimeter-derived ET for weekly and monthly periods and the associated error for Bear River, 1985.

78 R.G. Allen et al. date, July 14, compares well when examined in context of the impact of precipitation preceding the image date and rapidly growing vegetation during that period (Morse et al., 2000). The Fraction of Reference ET (ETrF) in Table 5.1 is defined as ET/ETr where ETr is reference ET based on an alfalfa-reference basis. ETrF values were computed for each pixel and used to extrapolate ET from the day of the satellite image to days between images. ETrF is synonymous with the well-known crop coefficient Kc when applied to an alfalfa reference as the basis (as opposed to clipped grass ETo). ETr accounts for changes in ET caused by weather variation between satellite image dates. The predicted, monthly ET averaged ⫾16% relative to the lysimeter at Montepelier (Table 5.1). However, seasonal differences between METRIC and lysimeters were only 4% due to impacts of reduction in the random error component present in each estimate. 5.3.2

Lysimeters at Kimberly, Idaho

The validation of METRIC on the Snake River Plain has centered on the use of two precisionweighing lysimeter systems for ET measurement in place near Kimberly, Idaho, from 1968 to 1991. The lysimeter system was installed and operated by James Wright of the USDA-ARS (Wright, 1982, 1996) and measured ET fluxes continuously. ET data are available for a wide range of weather conditions, surface covers, and crop types. Measurements of net radiation, soil heat flux, and plant canopy parameters were frequently made near the lysimeter site. The lysimeter data sets provided valuable information to verify METRIC over various time scales and for various conditions of ground cover. Nineteen Landsat 5 satellite image dates were purchased for Kimberly, Idaho, covering the period between 1986 and 1991. These dates had quality lysimeter and cloud-free micrometeorological data and represent a combination of crop growth stages and times of the year. Eight images from 1989 are discussed here. The lysimeter data for intervening periods between image dates were used to assess the impact of various methods for extending ET maps from a single day to longer periods. They have also been used to assess the variability in ETrF over a day. The success of METRIC is predicated on the assumption that ETrF for a 24-h period can be predicted from the ETrF from the instantaneous satellite image. ETr was calculated for hourly and 24-h periods using the ASCE standardized Penman–Monteith method for an alfalfa reference (ASCE-EWRI, 2002), representing the ET from a well-watered, fully vegetated crop, in this case, full-cover alfalfa 0.5 m in height. The denominator ETr serves as an index representing the maximum energy available for evaporation. Weather data were measured near the lysimeter and included solar radiation, wind speed, air temperature, and vapor pressure. An illustration of ETrF for a day in 1989 is given in Figure 5.2 for clipped grass (alta fescue) and sugar beets. ETrF for many days was even more uniform than shown in the figure. In nearly all cases, the ETrF for the 24-h period was within 5% of the ETrF at 1030. Lysimeter data analyses showed ETrF ⫽ ET/ETr to be preferable to EF parameter used in some applications of SEBAL (Bastiaanssen et al., 1998b; Bastiaanssen, 2000), where EF ⫽ ET/(Rn ⫺ G). The better performance by ETrF was due to its consistency during daytime and agreement between hourly ETrF at satellite overpass time (~1030) and daily average ETrF. Table 5.2 summarizes the error between METRIC and lysimeter measurements during 1989, a year when a significant number (eight) of both lysimeter measurements of ET and Landsat images were available. Absolute error averaged 30% for the eight image days. When April 18 was omitted, the average absolute error was only 14%. April 18 was before planting of the sugar beets and represented a period of drying bare soil following precipitation. The field at this time was non-uniform in wetness due to differential drying, and differences between lysimeter and METRIC computation were only 1 mm. The standard deviation of error between METRIC and lysimeter for dates from May to September was 13%. In comparison, a commonly quoted standard

Computing and mapping of evapotranspiration 79

1.5 Clipped grass

1.2 0.9 0.6

ETrF

ET (mm/h)

Kimberly lysimeters – July 7, 1989 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 –0.10 –0.20

0.3 0

0

400

800

1200

1600

2000

–0.3 2400

Time of day

1.5 Sugar beets

1.2 0.9 0.6

ETrF

ET (mm/h)

Kimberly lysimeters – July 7, 1989 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 –0.10 –0.20

0.3 0

0

400

800

1200

1600

2000

–0.3 2400

Time of day Lysimeter

ETrF

ETrF

Daily average ETrF

Figure 5.2. Hourly measurements of ET, ETr, ETrF, and 24-h ETrF for clipped grass (top) and sugar beets (bottom) at Kimberly, Idaho on July 7, 1989 (see Color Plate XV). (Source: Data from Wright, 1982, 1996.)

error for ET prediction equations that are based on weather data, for example, Penman or Penman–Monteith-types of equations, is about 10% for daily estimates. METRIC was able to obtain close to this level of accuracy for the field surrounding the lysimeter. Results are illustrated in Figure 5.3, where ET is expressed in the form of ETrF. ETrF was used to normalize results for differences in climatic demand (i.e. ETr). The round symbols and horizontal line segments in Figure 5.3 represent ETrF determined from lysimeter on the image date only. These values are those directly comparable with METRIC predictions in Table 5.2. The triangular symbols in the figure represent the ETrF predicted by METRIC for the image date (Figure 5.4). Table 5.2 summarizes the extrapolation of ET by METRIC over the season (April 1– September 30, 1989). Most periods were 16 days, centered on the image date. April 18 was used to represent April 1–April 25, July 23 was used to represent July 16–August 24 and September 25 was used to represent August 25 through September 30. What is surprising is the close

0.73 6.61 1.37 1.73 2.39 7.96 7.64 5.51

Error

4/18/89 5/4/89 5/20/89 6/5/89 6/21/89 7/7/89 7/23/89 9/25/89

4/1/89–9/30/89 Percent

1.74 5.09 1.34 1.78 2.54 5.89 7.17 7.40

METRIC ET on date (mm d⫺1) 139 ⫺23 ⫺2 3 6 ⫺26 ⫺6 34

Error on image date (%) 6.78 7.76 7.27 6.68 6.33 8.44 7.38 8.00

ETr on date (mm d⫺1) 147 94 90 118 127 120 253 201

ETr for period (mm)

16 80 17 30 48 113 262 138 705** ⫺1.8

28 30 22 24 62 116 266 171 718*

Lysimeter ET for period based on image date only (mm)

714*** ⫺0.6

38 62 17 31 51 84 246 186

METRIC ET for period (mm)

**

The sum of daily measurements by lysimeter computed as the sum over all days between April 1 and September 30. The sum of ET computed for each lysimeter period, computed by multiplying summed ETr during the period by the ETrF for the image date. *** The sum of ET predicted by METRIC for the lysimeter 2 field, computed by multiplying the summed ETr during the period by the ETrF computed on the image date by METRIC.

*

Lysimeter ET on date (mm d⫺1)

Image date

Lysimeter ET summed daily for period (mm)

Table 5.2. Summary and computation of ET during periods represented by each satellite image and sums for April 1–September 30, 1989 for lysimeter 2 (sugar beets) at Kimberly, Idaho.

10

10

METRIC ET (mm/day)

6/24/90 8

9/25/89

7/29/91 7/23/89 8/21/88

6

7/7/89 5/4/89

4 4/18/89 6/21/89 6/5/89

2

5/20/89

SEBAL ET(mm/day)–EF method

1:1 line

0

1:1 line

8 7/23/89 6

7/7/89 8/21/88

4

2

6/21/89

5/4/89

6/5/89

9/25/89

5/20/89 4/18/89 0

0

2

4

6

8

10

0

Measured ET (mm/day)

2

4

6

8

10

Measured ET (mm/day)

Figure 5.3. Comparison of daily ET predicted by METRIC using ETrF (left) and SEBAL using EF (right) on satellite image dates for sugar beets (1989), potatoes (1988), peas (1990), and alfalfa (1991) (from Trezza, 2002).

ETrF computed from lysimeter measurements for 1989

Fraction of reference ET (ETrF)

1.2 1.0

Field irrigated 2 days prior, lysimeter only 1 day prior (lysimeter was very wet)

0.8 0.6

Thermal pixels were badly blurred by areas outside Very high winds in lysimeter field AM and missing

0.4

wind speed data (estimated at 6 m/s)

0.2 0.0 18 4 April May

20 5 21 May June June

Five-day average lys. ETrF ETrF from METRIC

7 23 July July

25 September

Lysimeter ETrF for the image date

Figure 5.4. Results by METRIC and ET by lysimeter as ETrF. The thin line is the five-day average ETrF for lysimeter and the thick line is the assumption used in that application to extrapolate between images (see Color Plate XVI).

82

R.G. Allen et al.

agreement for seasonal ET for April 1–September 30. The difference between METRIC (714 mm) and the lysimeter measurement (718 mm) was less than 1% for the sugar beet crop. It appears that much of the error occurring on individual dates was randomly distributed, and tends to cancel, as described in more detail in Allen et al. (2004).

5.4 5.4.1

APPLICATIONS Idaho applications

IDWR developed five separate applications of the METRIC ET model. These applications are (1) to set water budgets for hydrologic modeling, (2) to monitor compliance with water rights, (3) to support water planning, (4) to estimate aquifer depletion, and (5) to estimate water use by irrigated agriculture. 5.4.1.1 Computing a water budget for hydrologic modeling Between 1996 and 2003, IDWR led a $2.5 million, multiagency (state and federal) project to study the water resources of the Lower Boise Valley in Idaho. This area is receiving significant attention because of rapid population growth in the valley. One of the major efforts of the project is the construction of a groundwater model of the valley. The US Bureau of Reclamation has spent the last three years studying irrigation diversions from the Boise River and irrigation return flow into the river in order to better quantify the water balance for the model. The third main component of the water balance is ET. IDWR and USBR have recently agreed to cooperate on a project to compute the ET portion of the water balance using METRIC. The base year for the water balance is 1996. A preliminary comparison was made using results from the METRIC ET analysis (1997 data) and the ET used in the 1996 Water Budget. The comparison was made with ET for floodirrigated lands only as this was the land type that was examined in the water budget. This preliminary comparison was made using three model cells within the Treasure Valley Hydrologic Project groundwater model domain; the selected cells were identified as being the three cells having the greatest flood-irrigated acreage. Each model cell is 2.59 km2 in area. Table 5.3 is an example of the data used in the initial comparison. The weighted average ET used in the 1996 Water Budget was 737 mm. This value was calculated using ET crop coefficients for 11 crop types, percentage of total cropped area for each crop, and the average ETr for the years 1988–94 (Parma Field Station). 5.4.2

Monitoring water rights compliance

5.4.2.1 Introduction IDWR presently has the technical means to identify diversions not having a water right. IDWR has tested and implemented a methodology to accomplish this using water right place-of-use polygons and Landsat TM false-color composite data in GIS. The technical means to identify someone using water “in excess of the elements or conditions of a water right” is more problematic. IDWR tested METRIC as an operational regulatory tool for administering water rights to identify those fields onto which water was applied in violation of some aspect of the water right, in this case the maximum rate of diversion. The test covered part of the eastern Snake River Plain, an area in Landsat path-row 39/30. The test was a comparison of righted pumpage rates with ET for water-right places-of-use during the period of peak water demand in July. The comparison was done for 426 water rights in the study area and required comparing the righted pumpage rate and the minimum possible rate given the volume of ET from each associated water right place of use.

Computing and mapping of evapotranspiration 83 Table 5.3. A comparison of METRIC ET with average ET for three cells of the Treasure Valley hydrologic model (April 15–October 15, 1997). Model row 10 11 28

Model column

Hectares flood irrigated

Water budget ET (mm)

METRIC ET (mm)

17 18 25

255 652 251

737 737 737

731 820 661

ET from METRIC (mm)

250 200 150 100 50 0

0

50

100

150

200

250

Theoretical maximum depth of water based on water right (mm)

Figure 5.5. Comparison of cumulative METRIC ET with maximum water-right ET for 426 water-right polygons in Idaho Department of Water Resources Basin 35 for July 12–28, 2002 period (see Color Plate XVII).

5.4.2.2 Results UI/Kimberly personnel processed two July 2002 Landsat scenes (July 12 and July 28) for waterright analysis and delivered maps of cumulative ET for the 17-day period to IDWR 11 days after the second overpass date. IDWR water rights and GIS personnel compared the METRIC ET data with water rights. The polygons were selected for a straightforward comparison between water consumption and authorized quantities. METRIC was used to determine cumulative ET for the period between the two images. The ET was compared with the volume of water authorized to be diverted based on valid water rights. Authorized diversion volume was calculated based on the allocated rate of flow, continuously diverted over the 17-day period. The comparison results are presented in Figure 5.5 where water-right volume is plotted on the horizontal axis and consumption on the vertical axis. The points lying above the diagonal line indicate consumption exceeding authorized diversions. The line of points at 206 mm on the y-axis is a function of the bounds put on water rights by Idaho Statutes. Some 426 water rights in IDWR Basin 35 could be compared with METRIC-generated ET, and 18 of those were found to have ET greater than the water right could provide. Those 18 water rights were handed off to water-rights personnel for further research. The enforcement process using METRIC offers a significant improvement over the present method that uses power records. METRIC data can be processed for analysis during the irrigation season, which will allow enforcement actions to be brought in a timely manner. Analysis

84 R.G. Allen et al. of power meter records generally cannot be accomplished during the irrigation season due to the reporting protocols and restrictions on personnel time. 5.4.3

Evapotranspiration by land use–land cover class

5.4.3.1 Introduction The purpose of this project was to compute the amount of ET by land use–land cover (LULC). Water planners at IDWR need to understand how the demand for water will be affected during the next 50 years by the transition of land from irrigated agriculture to residential, commercial, and industrial LULC types. The US Bureau of Reclamation and IDWR previously cooperated to generate a land use–land cover (LULC) classification of the Boise River Valley for the year 2000 from 1 : 24,000-scale aerial photographs. The classification consists of 24 LULC classes in a vector format. The availability of detailed LULC classes has enabled IDWR to combine the LULC classification with the METRIC ET data to generate preliminary values for ET by land cover class, data that were not previously available. Preliminary values for ET by LULC class are summarized in Table 5.4. These values are considered preliminary because METRIC parameters for aerodynamic roughness are designed for agricultural canopies rather than for non-agricultural surfaces. IDWR is responsible for comprehensive river basin planning in Idaho. One of the important issues that the planners are contending with is the potential for water availability in a valley that is rapidly changing from agricultural land use to more urban types of land uses. Table 5.4.

Mean seasonal ET by land use–land cover class.

Class name Wetland Water Recreation Perennial Irrigated crops Canal Urban residential Rural residential Farmstead New subdivision Sewage Public Other agriculture Dairy Feedlot Junk yard Abandoned agriculture Transition Idle agriculture Transportation Commercial and industrial Barren Unclassified Rangeland Petroleum tanks

Seasonal ET (mm)

Standard deviation

Area (hectares)

1025 924 826 820 812 731 684 657 609 606 552 548 536 524 479 467 459 437 436 420 380 335 298 242 237

285 165 252 212 189 203 157 192 188 146 256 263 243 182 205 193 211 195 215 222 196 258 239 160 112

5862 5344 2057 2711 141,075 2745 4126 10,164 2243 11,516 232 2120 2853 604 1691 129 1837 2712 3042 2313 5762 1912 12,742 90,647 18

Computing and mapping of evapotranspiration 85 5.4.3.2 Land use–land cover mapping IDWR computed and mapped ET by LULC class. The LULC polygons were mapped from 1 : 24,000-scale, color infrared, aerial photography taken during the summer of 2000. The aerial photographs were scanned to 1.5 m pixels and registered to the Idaho Public Land Survey System base. The registered photographs were mosaicked into tiles that covered an area of approximately 93 km2. IDWR personnel developed comprehensive descriptions of 24 LULC classes for the project. The descriptions were modified from MacConnell (1973) and are available in Kramber et al. (1997). 5.4.3.3 Evapotranspiration computation For this analysis seven dates of Landsat data were processed through METRIC to develop seasonal ET for the period March 15–October 15, 2000. The image dates are March 21, April 30, June 1, June 25, July 27, August 28, and October 2. 5.4.3.4 ET by land use–land cover class The availability of detailed LULC classes has enabled IDWR to overlay the LULC classification with the METRIC ET data to generate ET by land cover class. Figure 5.6(a) is a color infrared image of a portion of the lower Boise River Valley. Figure 5.6(b) is the corresponding area classified to land use and land cover. IDWR personnel overlaid the LULC polygons on the image of seasonal ET (Figure 5.6(c)) and computed the average seasonal ET for each class from all the polygons of each class. The result is summarized in Table 5.4, which shows the mean ET by LULC class with the associated standard deviation and the total area of each class. 5.4.4

Aquifer depletion

The relationship between ET and groundwater pumpage is important to IDWR regulatory processes. Historically, surfacewater diversions have been closely monitored while groundwater diversions have not. There are approximately 300 monitored diversions from the Snake River that irrigate approximately 647,500 hectares on the Eastern Snake River Plain (ESRP). The ESRP also supports approximately 200,000 hectares of groundwater irrigation, but from approximately 5000 wells. From a logistic point alone monitoring ground water pumpage is a large undertaking. IDWR and other associated organizations presently spend approximately $500,000 per year on monitoring groundwater pumpage from the ESRP. The Water Distribution Section of IDWR has visited the approximately 5000 wells on the ESRP over the last 5 years to record the GPS location and to measure the well flow and simultaneous power consumption. These data are stored in the Water Management Information System, which is designed to estimate groundwater pumpage using the power-meter records for its constituent wells. This application hypothesizes that there is a correlation between METRIC ET and groundwater pumpage and that for a given water right the ET for the field or fields covered by that water right can be used to estimate the volume of water pumped. There were 184 POUs for which the METRIC pumpage was made. Figure 5.7 shows the scatter plot of the two variables. No clear relationship is obvious, and a first-order polynomial regression confirms the lack of correlation with an r2 ⫽ 0.14. Nevertheless, a close examination of the two data sets is revealing. Figure 5.8(a) and (b) show the scatter within each individual variable in the data set plotted with AgriMet ET data. The AgriMet data show the ET extremes of alfalfa and peas and were recorded for the year 2000 at the US Bureau of Reclamation AgiMet station in Aberdeen, Idaho. The Aberdeen Station is within 32 km approximately of these fields and is representative of them.

86 R.G. Allen et al.

(a)

(b)

(c)

Figure 5.6. (a) FCC image of T3NR1E of the Boise Valley, (b) land use–land cover polygons in T3NR1E of the Boise Valley, and (c) ET image of T3NR1E of the Boise Valley (see Color Plate XVIII).

The two plots reveal useful information. In Figure 5.8(a), nearly all the METRIC ET observations fall between the extremes of ET, which is the lowest at 365 mm for peas and highest at 890 mm for alfalfa. Furthermore, there is a distinct “floor” at approximately 600 mm of ET, which is an indication of a minimum level of ET from irrigated agriculture. Most of the data fall well above the minimum ET for peas because grains, which have a higher minimum ET, in the range of 556–576 mm, are a more common crop than peas in the Aberdeen area. The reason that the “floor” is approximately 50 mm above the ET level for grains is that the AgriMet figures are crop ET and not field ET. They are adjusted to account for precipitation and do not account for the fact that after harvest many farmers will continue to irrigate bare soil to build soil moisture.

Pumpage from power meter records (mm)

1600 1400 1200 1000 800 600 400 200 0 0

200

400

600

800

1000

1200

METRIC ET (mm)

Figure 5.7. The scatter plot of pumpage vs METRIC ET for the period April–October, 2000 (see Color Plate XIX). (a)

Seasonal ET– METRIC 1600

Millimeters of ET

1400 1200 1000

Alfalfa– seasonal ET Aberdeen AgriMet station

800 600 400

Peas – seasonal ET Aberdeen AgriMet station

200 0 0

(b)

50

100 150 Observation number

200

Seasonal pumpage–power meter data 1600

Millimeters of pumpage

1400 1200 1000 Alfalfa– seasonal ET Aberdeen AgriMet station

800 600 400

Peas – seasonal ET Aberdeen AgriMet station

200 0 0

50

100 150 Observation number

200

Figure 5.8. (a) April–October, 2000 METRIC ET compared with AgriMet ET extremes and (b) April–October, 2000 pumpage compared with AgriMet ET extremes (see Color Plate XX).

88 R.G. Allen et al. Contrast the METRIC ET pattern of Figure 5.8(a) with the pattern for pumpage as illustrated by Figure 5.8(b). The pumpage data are not consistent at either the high end of the chart or at the low end. There is no “floor” evident to show that there is a minimum level of pumping that is a minimum level of irrigation needed to support an irrigated crop. In fact, the pumpage data set indicates that there are fields getting no pumpage at all. The reliability of the data set is called into question by the lack of patterns that reflect irrigation practice on the ESRP and by the abundance of data at the extreme low end of the chart. 5.4.5

Applications in the Imperial Valley

ET maps have been created using METRIC and Landsat 7 images for much of Imperial Valley, California, for the January–March periods of 2002 and 2003 (Allen et al., 2003). The application demonstrated the ability to produce maps of quantitative, spatial distribution of monthly ET in near real time with resolution on the sub-field scale. The high-resolution maps from Landsat were also useful in comparing ET in the “lower” ends of surface irrigated fields with ET in the “higher” ends of fields. Often, ET in lower ends of surface irrigated fields can suffer due to low irrigation uniformity or effects of salinity and inadequate leaching of salts. 5.4.6

Applications in the Middle Rio Grande

METRIC was applied with Landsat 5 and 7 images to irrigated and riparian areas along the Middle Rio Grande river of northern and central New Mexico for year 2002, to spatially and temporally quantify ET by irrigated crops and by riparian vegetation (native and invasive tree species and wetlands) (Allen et al., 2004). The high resolution of Landsat was, again, extremely valuable for assessing ET on a field by field basis and for estimating ET from riparian (tree) systems that were often less than 100 m in width. The Landsat based ET maps, derived for each month of the year, were valuable in showing the amount of evaporation from abandoned agricultural fields that had high water tables. The high water tables precluded farming operations and supplied water to the surface for evaporation. Reducing these evaporation losses by lowering water tables would constitute a real conservation of water in the valley.

5.5

IMPACT

The METRIC work is evolving. Nevertheless, there have been impacts. IDWR found the results of Phase I and II sufficiently compelling to request additional funding from the Idaho Legislature to include METRIC as the ET source for recalibration of the ESRP aquifer model and to generate ET maps to monitor groundwater pumpage. The aquifer model uses 5 km grid cells and aggregating ET up to a 5 km cell is preferable to disaggregating county-averaged data. 5.5.1

Cost savings

ET data derived from METRIC are less expensive to generate than are standard ET data. Since METRIC applications are still in the development stage at IDWR, a rigorous cost-benefit analysis is premature. Nevertheless, it is possible to do a rough cost comparison based on some available figures. Current costs for monitoring water use on the ESRP are estimated to be about $500,000 per year. We estimate costs for remote sensing to be about $100,000 per year. This includes costs for 30 TM scenes representing 8–10 dates for the whole eastern Snake Plain (Landsat scenes cost about $400 each for images). Geo-registration of images costs an additional $400 each, for a total procurement cost of about $24,000, and about three Landsat images

Computing and mapping of evapotranspiration 89 (160 km ⫻ 160 km) are required to cover the full area. Once set up for an area, METRIC processing requires about 8 days per scene (240 days ⫻ 8 h ⫽ 1920 h ⫻ $40.00 per h ⫽ $76,800 for processing for the full year for the full eastern Snake Plain). The total for remote sensing is therefore about $100,000. Set-up and time for aggregation of ET results via GIS results in a total remote-sensing cost of $105,000. Using these figures, the estimated cost ratio of remote sensing to the current measurement program is $105,000/$500,000 ⫽ 0.21, that is, remote sensing costs about 20% of the measurement costs. Measurement costs are for a subset of the total number of wells, not all of which are measured in a single year, whereas METRIC data cover the entire Snake River Plain and all places of use. The use of METRIC ET will not replace the existing measurement program, per se. Pumpage data that can be related to individual water rights will be needed for regression against the METRIC ET data for the same water rights to establish the relationship between volume pumped and volume of ET. That relationship can then be applied to all other non-monitored water rights and their associated wells to estimate both aquifer depletion and water use by individual water rights. 5.6

SUMMARY AND CONCLUSIONS

METRIC and SEBAL use digital image data collected by Landsat and other remote-sensing satellites that record thermal infrared, visible, and near-infrared radiation. ET is computed on a pixel-by-pixel basis for the instantaneous time of the satellite image. The process is based on a complete energy balance for each pixel, where ET is predicted from the residual amount of energy remaining from the classical energy balance. Here ET ⫽ net radiation ⫺ heat to the soil ⫺ heat to the air. In Phase 1 for the Bear River Basin, the difference between METRIC and the lysimeter, total, for the growing season was 4%. For the Phase 2 comparison with precision-weighing lysimeters at Kimberly, differences were less than 2%. These comparisons represent a small sample, but are probably typical. Error as high as 10–20%, if distributed randomly, could probably be tolerated by IDWR and by the water user communities. Comparisons of METRIC predicted ET with precision-weighing lysimeter data at Kimberly, Idaho from the 1980s and early 1990s have provided valuable information on the conditions required to obtain maximum accuracy with METRIC and the best procedure for obtaining ET monthly and annually. ET has been calculated for the entire Snake River Plain of southeastern Idaho and has improved the calibration of groundwater models by providing better information on groundwater recharge as a component of water balances. Groundwater pumpage from over 10,000 wells has been estimated using ET from METRIC by developing correlations between ET and pump discharge at measured wells and then extrapolating over large areas using ET maps from METRIC.

REFERENCES Allen, R.G. et al. (1996) Evaporation and transpiration. In: Wootton et al. (ed.), ASCE Handbook of Hydrology, ASCE, New York, pp. 125–252. Allen, R.G., Tasumi, M., and Lorite Torres, I. (2003) High resolution quantification of evapotranspiration from imperial irrigation district. Research Completion report (phase I) submitted to MWD, December, 130pp. Allen, R.G., Tasumi, M., Morse, A., and Trezza, R. (2004) A Landsat-based energy balance and evapotranspiration model in western US water rights regulation and planning. Journal of Irrigation and Drainage System (in press).

90 R.G. Allen et al. ASCE-EWRI (2002) The ASCE standardized reference evapotranspiration equation. ASCE-EWRI Standardization of Reference Evapotranspiration Task Committee. Report, available at http://www. kimberly.uidaho.edu/water/asceewri Bastiaanssen, W.G.M. (1995) Regionalization of surface flux densities and moisture indicators in composite terrain: a remote sensing approach under clear skies in Mediterranean climates. PhD Dissertation, CIP Data Koninklijke Bibliotheek, Den Haag, The Netherlands, p. 273. Bastiaanssen, W.G.M. (2000) SEBAL-based sensible and latent heat fluxes in the irrigated Gediz Basin, Turkey. Journal of Hydrology, 229, 87–100. Bastiaanssen, W.G.M., Menenti, M., Feddes, R.A., and Holtslag, A.A.M. (1998a) A remote sensing surface energy balance algorithm for land (SEBAL): 1. Formulation. Journal of Hydrology, 212–213, 198–212. Bastiaanssen, W.G.M. et al. (1998b) The surface energy balance algorithm for land (SEBAL): Part 2 validation, Journal of Hydrology, 212–213, 213–229. Bastiaanssen, W.G.M. et al. (2004) SEBAL for spatially distributed ET under actual management and growing conditions. ASCE Journal of Irrigation and Drainage Engineering (in press). Hill, R.W. et al. (1989) Duty of water under the Bear River compact: field verification of empirical methods for estimating depletion. Research report 125. Utah Agricultural Experiment Station, Utah State University, Logan, Utah. Kramber, W.J., Morse, A., Harmon, B., and Anderson, H. (1997) Mapping 80 years of change in irrigated agriculture. In: Proceedings of the 1997 ACSM/ASPRS Annual Convention, Bethesda, MD, pp. 367–372. MacConnell, W.P. (1973) Massachusetts Map Down, Publication 97. Cooperative Extension Service, University of Massachusetts, Amherest, MA. Morse, A., Tasumi, M., Allen, R.G., and Kramber, W.J. (2000) Application of the SEBAL methodology for estimating consumptive use of water and streamflow depletion in the Bear River Basin of Idaho through remote sensing. Synergy Phase 1 Final Report, 108pp., Idaho Department of Water Resources, Boise, Idaho. Morse, A., Kramber, W.J., Allen, R.G., and Tasumi, M. (2004) Use of the METRIC evapotranspiration model to compute water use by irrigated agriculture in Idaho. In: Proceedings of the 2004 IGARSS Symposium, Anchorage, AK. Romero, M.G. (2004) Daily evapotranspiration estimation by means of evaporative fraction and reference evapotranspiration fraction. PhD Dissertation, Utah State University, Logan, Utah. Tasumi, M., Allen, R.G., Trezza, R., and Wright, J.L. (2004a) Use of SEBAL to assess the band width of crop coefficient curves in Idaho. ASCE Journal of Irrigation and Drainage Engineering (in press). Tasumi, M., Trezza, T., Allen, R.G., and Wright, J.L. (2004b) Operational aspects of satellite-based energy balance models for irrigated crops in the semi-arid US. Journal of Irrigation and Drainage Systems (accepted). Trezza, R. (2002) Evapotranspiration using a satellite-based surface energy balance with standardized ground control. PhD Dissertation, Utah State University, Logan, Utah. Wright, J.L. (1982) New evapotranspiration crop coefficients. Journal of Irrigation and Drainage Division (ASCE), 108, 57–74. Wright, J.L. (1996) Derivation of alfalfa and grass reference evapotranspiration. In: C.R. Camp, E.J. Sadler, and R.E. Yoder (eds), Evapotranspiration and Irrigation Scheduling, Proceedings of the International Conference, ASAE, San Antonio, TX, pp. 133–140.

CHAPTER 6

Satellite Remote Sensing of Soil Moisture Thomas J. Jackson USDA ARS Hydrology and Remote Sensing Laboratory, 104 Building 007 BARC-West, Beltsville, MD 20705, USA

6.1

INTRODUCTION

Hydrologic, weather, and climate modeling can be improved through observation of the current status of soil moisture. Better predictions from these models will lead to improved forecasts of floods and other phenomena. Soil moisture products are now feasible using a new generation of microwave remote-sensing satellites. The quality of these products will continue to improve as new sensors are launched over this decade. Microwave sensors are of particular value because they respond to the amount of moisture in the soil. There are several methods that have been shown to be capable of providing soil moisture information. Each has unique capabilities (i.e. temporal coverage, spatial resolution, etc.) that must be matched with specific types of applications. In the past, options have been limited by available satellite systems. Investigators have demonstrated the potential of these data in hydrologic studies using ground and aircraft systems. However, efforts to use the less than optimal available satellite systems have had very limited success. In recent years and continuing over the coming decade, a wide range of new and significantly improved satellites will be launched that will offer new opportunities. 6.2

BASIC PRINCIPLES OF MICROWAVE REMOTE SENSING OF SURFACE SOIL MOISTURE

Microwave remote sensing provides a direct measurement of the surface soil moisture for a range of vegetation cover conditions. Two basic approaches are used, passive and active. In passive methods, the natural thermal emission of the land surface (or brightness temperature) is measured at microwave frequencies using very sensitive detectors. In active methods or radar, a microwave pulse is sent and received. The power of the received signal is compared to that which was sent to determine the backscattering coefficient. Microwave sensors operating at very low microwave frequencies (⬍6 GHz) provide the best soil moisture information. At low frequencies, attenuation and scattering problems associated with the atmosphere and vegetation are less significant, the instruments respond to a deeper soil layer, and have a higher sensitivity to soil water content variations (see Figure 6.1). Most research and applications involving passive microwave remote sensing of soil moisture have emphasized low frequencies (L band). In this range, it is possible to develop soil moisture retrievals based on a single H polarization observation (Jackson, 1993). This approach relies on providing ancillary data on temperature, vegetation, land cover, and soils. Other algorithm approaches are described in Njoku et al. (2000). Estimating soil water content from radar backscatter is easier when the soil is bare. When there is a vegetation cover, establishing soil water under the canopy is much more difficult and requires unraveling the contribution of the soil itself from that of vegetation. The most common approach to estimating soil moisture from backscatter has been linear regression. This of

92 Thomas J. Jackson

Decreasing frequency High

Sensitivity

Bare

Vegetated

Low 1

2

3

5

10

20 30

50

Frequency (GHz)

Figure 6.1.

Brightness temperature–soil moisture sensitivity as a function of microwave frequency.

course does not result in a robust retrieval algorithm. More theoretical approaches have limited applicability and are difficult to implement. A more promising technique involves semi-empirical models. These involve multiple polarization observations and restrictions on the range of applicability. Algorithms incorporating this approach for bare soils are presented in Dubois et al. (1995). A key issue in comparing passive and active microwave methods from satellites is the tradeoff between the high spatial resolution of synthetic aperture radar (SAR) methods and the robust retrieval and frequent temporal coverage provided by passive methods. 6.3 6.3.1

PASSIVE MICROWAVE SATELLITE OBSERVING SYSTEMS AND PRODUCTS Special satellite microwave/imager

Currently, all passive microwave sensors on satellite platforms operate at high frequencies (⬎6 GHz) (Table 6.1). Of particular note, due to the longevity of its data record, is the Special Satellite Microwave/Imager (SSM/I) package on the Defense Meteorological Satellite Platforms. These satellites have been in operation since 1987 and provide high frequencies and two polarizations. Interpreting data from the SSM/I to extract surface information requires accounting for atmospheric effects on the measurement (Drusch et al., 2001). When one considers the atmospheric correction, the significance of vegetation attenuation, and the shallow contributing depth of soil for these high frequencies, it becomes apparent that the data are of limited value for estimating soil water content. Spatial resolution of the SSM/I is very coarse (see Table 6.1). There have been few attempts at generating standard land surface products using SSM/I data. NOAA (Basist et al., 1998) utilizes SSM/I data to produce an experimental data product called the Soil Wetness Index (SWI). This index is intended to provide information on significantly wet soil conditions (areal extent of flooding), which can be more reliably detected than variations at lower levels of soil moisture.

Satellite remote sensing of soil moisture 93 Table 6.1.

Microwave satellite systems.

Satellite

Period of coverage

Passive SSM/I TMI

1987–present 1998–present

AMSR (Aqua) AMSR (ADEOS-II) Windsat

2002–present

SMOS HYDROS

(2007) (2010)

Active ERS Radarsat-1 ASAR

1991–present 1995–present 2002–present

2002 2003–present

Frequency (GHz) 19.4, 22.2, 37.0, 85.5 10.7, 19.4, 21.3, 37, 85.5 6.9, 10.7, 18.7, 23.8, 36.5, 89.0 6.9, 10.7, 18.7, 23.8, 36.5, 89.0 6.8, 10.7, 18.7, 23.8, 37 1.4 1.4

5.3 5.3 5.3

Radarsat-2

(2005)

5.3

PALSAR

(2005)

1.27

Spatial resolution

Repeat frequency (days)

H and V H and V

70–75 km 60–66 km

1–2 1

H and V

75–77 km

2–3

H and V

70–76 km

2–3

H and V (U in some channels) H and V H and V (passive) HH, VV (active)

50–10 km

2–3

50 km 3–40 km

2–3 2–3

30 m 7–100 m

35 24

30–1000 m

35

3–100 m

24

10–100 m

46

Polarization

VV HH HH and VV or HH and HV or VV and VH HH, VV, HV and VH HH or VV and HV or VH

A few studies have attempted to extract actual soil moisture from SSM/I data (Owe et al., 1992; Jackson, 1997; Vinnikov et al., 1999). Jackson (1997) used a single channel/ancillary data approach. For the limited validation data set available the approach performed relatively well. However, part of this may be attributable to the limited conditions and light vegetation conditions evaluated. 6.3.2

Tropical rainfall measurement mission microwave imager

Another current option is the Tropical Rainfall Measurement Mission (TRMM) Microwave Imager (TMI). It is a five-channel, dual-polarized, and passive-microwave radiometer. The lowest TMI frequency is 10 GHz, about half that of the SSM/I. The TMI has a higher spatial resolution (see Table 6.1) as compared to the SSM/I. TRMM only provides coverage of the tropics, which includes latitudes between 38⬚N and 38⬚S for the TMI instrument. However, a unique capability of the TMI is its ability to collect data daily, and in many cases more often, within certain latitude ranges. Jackson and Hsu (2001) and Wen et al. (2003) demonstrated the potential of using these data to retrieve soil moisture. Bindlish et al. (2003) have developed and validated a five-year data set for southern United States based upon the TMI data. 6.3.3 Advanced microwave scanning radiometer Several new multifrequency passive microwave satellite systems were launched in 2002 and 2003. As opposed to the previously available systems these offer a lower frequency channel operating at C band, which should provide a more robust soil moisture measurement.

94

Thomas J. Jackson

These satellites were the NASA Aqua, Japanese ADEOS-II, and the US interagency Coriolis satellite. The ADEOS-II satellite stopped operations in October of 2003 and will not be discussed here. A component of Aqua is the Advanced Microwave Scanning Radiometer (AMSR). This instrument includes a 6.9 GHz (C-band) channel with a nominal 60 km spatial resolution. This instrument should be able to provide information about soil water content in regions of low vegetation cover, less than 1 kg/m2 vegetation water content. Aqua AMSR-E collects data at nominal times of 1.30 am and pm local time. As opposed to previous passive microwave satellite missions, the Aqua project will include soil moisture as a product. The algorithm planned for use by NASA is the multiple channel approach described in Njoku et al. (2003). The Japanese Aerospace Exploration Agency will also produce a soil moisture product using one of four algorithms under consideration (Koike et al., 2000; Njoku et al., 2000). Preliminary studies indicate that there is widespread radio frequency interference (RFI) in the C-band channels. Therefore, it is likely that the most useful channels for soil moisture will be those operating at the slightly higher X band. The Coriolis satellite includes the Windsat instrument, which is a multifrequency passive microwave radiometer system. It is similar to AMSR with some differences in frequencies and more polarization options. It is a prototype of one component of the next generation of operational polar orbiting satellites that the United States will be implementing at the end of this decade. 6.3.4

Projected passive systems

Programs are underway to develop and implement space-based systems with a 1.4 GHz channel that would provide improved global soil moisture information. Toward that goal the European Space Agency (ESA) is developing a sensor system called the Soil Moisture Ocean Salinity (SMOS) mission (Wigneron et al., 2000) and NASA has initiated a mission called Hydros. SMOS will utilize two dimensional synthetic aperture radiometry at L band and provide 40 km H and V polarizations. The ability of SMOS to obtain multiple incidence angle observations of the surface is the key element of the soil moisture retrieval algorithm. The launch of the mission is anticipated in 2007. Hydros would use an active and passive L-band instrument design for mapping at resolutions ranging from 3 to 40 km. Soil moisture will be estimated by Hydros using the radiometer and radar data separately and in combination, taking advantage of the simultaneous, coincident, and complementary nature of the measurements. Hydros is scheduled for 2010.

6.4 6.4.1

ACTIVE MICROWAVE SATELLITE OBSERVING SYSTEMS Current active systems

At present, several radar satellites are in orbit. ESA has operated a satellite SAR series called ERS since 1991, which provides C-band VV data. It includes both an SAR and a scatterometer. Although numerous investigations have been conducted that attempt to utilize ERS SAR data, few results have been reported in the area of soil moisture estimation. This is due to the limitations of using a single mid-frequency channel-single polarization SAR with an exact repeat cycle of 35 days. The most recent satellite in the ERS series is called Envisat (launched in 2002) and it has a C-band Advanced Synthetic Aperture Radar (ASAR) with multiple polarization capabilities. It also offers the option of varying the incidence angle to allow for different viewing angles and more frequent coverage if angle is not a critical parameter in the application. Data from ASAR are just beginning to be made available to investigators. The Canadian Space Agency operates

Satellite remote sensing of soil moisture 95 a C-band satellite SAR called Radarsat, which offers HH polarization and has a variable viewing angle and a wide swath (large range of incidence angles). These choices offer more frequent temporal coverage of a particular region of the Earth if angle is not important. There have been a number of attempts at using the single channel SAR data to retrieve soil moisture. The general consensus from these studies is that there are too many physical variables that have to be known in order to derive soil moisture (Verhoest et al., 1998; Satalino et al., 2002). These variables include the soil moisture, surface roughness, and vegetation. Sitespecific studies and creative selection of data sets have resulted in some degree of success (e.g. Satalino et al., 2002). 6.4.2

Projected active systems

Japan will include an L-band SAR called PALSAR on the Advanced Land Observing System (ALOS) in 2005 or later. PALSAR will have a multipolarization mode as well as varying incidence angles. Canada is developing Radarsat-2, which will be similar to Radarsat-1 but will be fully polarimetric. The satellite will include a mode in which the spatial resolution is 3 m. The expected launch is in 2005. As noted above, progress in SAR soil moisture mapping has been limited by the available data, single channel. The new generation of multipolarization SAR systems will provide at least one additional measurement. There is hope that more robust soil moisture retrieval methods can be developed as these data become available. REFERENCES Basist, A., Grody, N.C., Peterson, T.C., and Williams, C.N. (1998) Using the Special Sensor Microwave/Imager to monitor surface temperatures, wetness, and snow cover. Journal of Applied Meteorology, 37, 888–911. Bindlish, R. et al. (2003) Soil moisture estimates from TRMM Microwave Imager observations over the southern United States. Remote Sensing of Environment, 85, 507–515. Drusch, M., Wood, E.F., and Jackson, T.J. (2001) Vegetative and atmospheric corrections for the soil moisture retrieval from passive microwave remote sensing data: results from the Southern Great Plains hydrology experiment 1997. Journal of Hydrometeorology, 2, 181–192. Dubois, P.C., van Zyl, J., and Engman, E.T. (1995) Measuring soil moisture with imaging radars. IEEE Transactions on Geoscience and Remote Sensing, 33, 915–926. Jackson, T.J. (1993) Measuring surface soil moisture using passive microwave remote sensing. Hydrological Process, 7, 139–152. Jackson, T.J. (1997) Soil moisture estimation using Special Satellite Microwave/Imager satellite data over a grassland region. Water Resources Research, 33, 1475–1484. Jackson, T.J. and Hsu, A.Y. (2001) Soil moisture and TRMM Microwave Imager relationships in the Southern Great Plains 1999 (SGP99) experiment. IEEE Transactions on Geoscience and Remote Sensing, 39, 1632–1642. Koike, T., Njoku, E., Jackson, T.J., and Paloscia, S. (2000) Soil moisture algorithm development and validation for the ADEOS-II/AMSR. Proceedings of the International Geoscience and Remote Sensing Symposium, IEEE Catalog No. 00CH37120, Vol. III, 1253–1255. Njoku, E., Koike, T., Jackson, T., and Paloscia, S. (2000) Retrieval of soil moisture from AMSR data. In: Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere. P. Pampaloni and S. Paloscia, (eds), The Netherlands: VSP Publications, pp. 525–533. Njoku, E.G. et al. (2003) Soil moisture retrieval from AMSR-E. IEEE Transactions on Geoscience and Remote Sensing, 41, 215–229. Owe, M., van de Griend, A.A., and Chang, A.T.C. (1992) Surface moisture and satellite microwave observations in semiarid southern Africa. Water Resources Research, 28, 829–839. Satalino, G. et al. (2002) On current limits of soil moisture retrieval from ERS-SAR data. IEEE Transactions on Geoscience and Remote Sensing, 40, 2438–2447.

96 Thomas J. Jackson Verhoest, N.E.C., Troch, P.A., Paniconi, C., and De Troch, F.P. (1998) Mapping basin scale variable source areas from multitemporal remotely sensed observations of soil moisture behavior. Water Resources Research, 34, 3235–3244. Vinnikov, K.Y. et al. (1999) Satellite remote sensing of soil moisture in Illinois, USA. Journal of Geophysical Research, 104, 4145–4168. Wen, J., Su, Z., and Ma, Y.M. (2003) Determination of land surface temperature and soil moisture from Tropical Rainfall Measuring Mission/ Microwave Imager remote sensing data. Journal of Geophysical Research, 108, 4038. Wigneron, J.P. et al. (2000) Two-dimensional microwave interferometer retrieval capabilities over land surfaces (SMOS mission). Remote Sensing of Environment, 73, 270–282.

CHAPTER 7

Ensemble Streamflow Forecasting: Methods and Applications Balaji Rajagopalan and Satish Regonda Department of Civil, Environmental and Architectural Engineering (CEAE), University of Colorado, Boulder, CO, USA and CIRES, University of Colorado, Boulder, CO, USA Katrina Grantz Department of Civil, Environmental and Architectural Engineering (CEAE), University of Colorado, Boulder, CO, USA and Center for Advanced Decision Support for Water and Environmental Systems (CADSWES)/CEAE, University of Colorado, Boulder, CO, USA Martyn Clark CIRES, University of Colorado, Boulder, CO, USA Edith Zagona Center for Advanced Decision Support for Water and Environmental Systems (CADSWES)/CEAE, University of Colorado, Boulder, CO, USA

7.1

INTRODUCTION

The chapter is organized into several sections. The theme of the chapter is introduced in Section 7.1. Section 7.2 presents a background on large-scale climate and its impacts on the western US hydroclimatology. The basins studied and data used are described in Sections 7.3 and 7.4, respectively. This is followed by the climate diagnostics and identification of predictors for forecasting spring streamflows in Section 7.5. Section 7.6 presents the development of the statistical ensemble forecasting model using the identified predictors. Model application and validation are described in Section 7.7. The last section concludes the presentation with a summary and discussion of the results. Water resources worldwide are faced with increasing stresses due to climate variability, population growth, and competing growth – more so in the western United States (e.g. Piechota et al., 2001; Hamlet et al., 2002). Careful planning is necessary to meet demands on water quality, volume, timing, and flow rates. This is particularly true in the western United States, where it is estimated that 44% of renewable water supplies are consumed annually, as compared with 4% in the rest of the country (el-Ashry and Gibbons, 1988). Consequently, the forecast for the upcoming water year is crucial to the water management planning process involving system outputs such as crop production and the monetary value of hydropower production (e.g. Hamlet et al., 2002) as well as the sustenance of aquatic species.

98

Balaji Rajagopalan et al.

A majority of river basins in the western United States are snowmelt driven in that, snow accumulates in the winter and melts in the spring thus producing a peak in the streamflow. Therefore, it is intuitive to use winter snowpack as a predictor of the runoff in the following spring (Serreze et al., 1999). More recently, information about large-scale climate phenomena such as El Niño Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) pattern has been added to the forecaster’s toolbox. The link between these large-scale phenomena and the hydroclimatology of the western United States has been well documented (e.g. Gershunov, 1998). Clark and Serreze (2001) showed that including large-scale climate information together with Snow Water Equivalent (SWE) improves the overall skill of the streamflow predictions in the western United States. Souza and Lall (2003) showed significant skills at longer lead times in forecasting streamflows in Cearra, Brazil using climate information from the Atlantic and Pacific oceans. Typically, streamflow forecasts are issued by fitting a linear regression with SWE and sometimes with standard indices that describe the ENSO and PDO phenomena. The disadvantages with this approach are (1) the relationship is not always linear, (2) the teleconnection patterns from ENSO and PDO though dominant on a large scale, often fail to provide forecast skill on the individual basin scale. This is so because the surface climate is sensitive to minor shifts in large-scale atmospheric patterns (e.g. Yarnal and Diaz, 1986), and (3) inability to provide realistic ensemble forecasts and thus, the probability of exceedences of various thresholds useful for water resources management. Evidently, there is a need for a generalized framework for ensemble streamflow forecast that utilizes large-scale climate information. We propose such a framework in Figure 7.1. In this, large-scale climate predictors are first identified via climate diagnostics. The identified predictors are then used in a nonparametric framework to generate ensemble of streamflow forecast. The ensembles can then be incorporated in a decision support system for water resources management. In this chapter we focus primarily on the climate diagnostics and ensemble forecast methods, and then demonstrate their utility on the Truckee–Carson River Basin and Gunnison River Basin, both located in the western United States.

Climate diagnostics

Ensemble streamflow Forecasting model

Decision support system (Water resources management)

Figure 7.1.

Flow chart of the forecast framework.

Ensemble streamflow forecasting 99 7.2

LARGE-SCALE CLIMATE AND WESTERN US HYDROCLIMATOLOGY

The tropical ocean-atmospheric phenomenon in the Pacific identified as ENSO (e.g. Allan et al., 1996) is known to impact the climate all over the world and, in particular, the western United States (e.g. Ropelewski and Halpert, 1986). The warmer sea surface temperatures and stronger convection in the tropical Pacific Ocean during El Niño events deepen the Aleutian Low in the North Pacific Ocean, amplify the northward branch of the tropospheric wave train over North America, and strengthen the subtropical jet over southwestern United States (e.g. Rasmussen, 1985). These circulation changes are associated with below-normal precipitation in the Pacific northwest and above-normal precipitation in the desert southwestern United States (e.g. Redmond and Koch, 1991; Cayan and Webb, 1992). Generally opposing signals are evident in La Niña events, but some nonlinearities are present (Hoerling et al., 1997; Clark and Serreze, 2001). Decadal-scale fluctuations in Sea Surface Temperatures (SSTs) and sea levels in the northern Pacific Ocean as manifested by the PDO (Mantua et al., 1997) provide a separate source of variability for the western US hydroclimate. Independence of PDO from ENSO is still in debate (Newman et al., 2003). Regardless, the influence of PDO and ENSO on North American hydroclimate variability has been well documented (e.g. Regonda et al., 2004a). Incorporation of this climate information has been shown to improve forecasts of winter snowpack (McCabe and Dettinger, 2002) and streamflows in the western United States (Clark et al., 2001; Hamlet et al., 2002) while increasing the lead time of the forecasts. Use of climate information enables efficient management of water resources and provides socio-economic benefits (e.g. Pulwarty and Melis, 2001; Hamlet et al., 2002). Often, however, the standard indices of these phenomena (e.g. NINO3, SOI, PDO index, etc.) are not good predictors of hydroclimate in every basin in the western United States – even though these phenomena do impact the western US hydroclimate (as described earlier). Furthermore, certain regions in the western United States (e.g. basins in between the Pacific northwest and the desert southwest) can be impacted by both the northern and southern branches of the subtropical jet, potentially diminishing apparent connections to ENSO and PDO. The Truckee and Carson basins are two such examples, hence, predictors other than the standard indices have to be developed for each basin. 7.3

WATER MANAGEMENT ISSUES IN THE BASINS STUDIED

Our motivation for the development of the ensemble streamflow approaches stems from the need to develop tools for efficient water management on two basins (1) Truckee–Carson River Basins in Nevada (shown in Figure 7.2), western United States and (2) Gunnison River Basin, a tributary of Colorado River, also in the western United States that can be seen in Figure 7.3. On the Truckee–Carson Basin flows at 2 gauging stations are to be forecast, while in the Gunnison streamflow forecasts are required at 6 sites simultaneously. In both the basins, for that matter over much of the western United States, the bulk of the annual streamflow arrives during spring (April–July) from the melting of snowpack accumulated over winter. This is evident in the climatology of precipitation and streamflows for the Truckee River (Figure 7.4) – a similar feature is observed on the Gunnison as well. 7.3.1

Truckee–Carson

The Truckee and Carson Rivers originate high in the California Sierra Nevada Mountains and flow northeastward down through the semiarid desert of western Nevada. The Truckee River

Truckee NEVADA

PYRAMID LAKE

AL O IF

CALIFORNIA NEVADA

C

Carson

N R IA

Nixon

Stillwater NWR Derby TRUCKEE Dam CANAL STAMPEDE Reno/Sparks Fernley Fallon INDEPENDENCE TRUCKEE BOCA Newlands RIVER PROSSER Project Farad Truckee LAHONTAN CARSON MARTISCarson DONNER Ft Churchill LAKE City Tahoe City CARSON RIVER

LAKE TAHOE

Figure 7.2.

Map of the Truckee–Carson Basin (see Color Plate XXI). N 10

0

10

20 Miles E

W

Grand junction

S

n

iso

ve

Ri

th or

Olathe

GUNNISON 1

nn

iso

ahg

eau C

Hotchkiss

Gu

omp

n

Ri

ve

Montrose

iver

re R

Roubid

Crested Butte

Delta Unc

reek

r

N

6 Paonia

r Crystal Reservoir

Blue Mesa Reservoir

MONTROSE

OURAY Ouray

unnis on Fork G

Ridgeway

1 2 3 4 5 6

Gunnison To

4

mic

hi C

ree

k

SAGUACHE

Lake

5

er

Riv

2

3 Morrow Point Reservoir

lor

Tay

Coehetopa Creek

r

Fo

er

n

Orchard City

t Riv

so

MESA

nn

u kG

ni

un

G

DELTA Cedaredge

Eas

Whitewater

HINSDALE

09112500 East River at Almont, CO. 09110000 Taylor River at Almont, CO. 09119000 Tomichi Creek at Gunnison, CO. 09124500 Lake Fork at Gate View, CO. 09147500 Uncompahgre River at Colona, CO. 09132500 North Fork Gunnison River near Somerset, CO.

Figure 7.3. Gunnison River Basin and streamflow locations. USGS gauge locations and river names are mentioned. (Source: Colorado Water Conservation Board.)

Ensemble streamflow forecasting 101

Streamflow volume (kaf) Truckee River at Farad

Streamflow Precipitation

100

3.5 3

80

2.5

60

2 1.5

40

1 20

0.5

Precipitation (in) Sierra Nevada climate division

4

120

0

0 Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Month

Figure 7.4. Climatology of streamflows and precipitation in the Truckee River, at the gauging station Farad (based on data for the 1949–2003 period) (see Color Plate XXII).

originates as outflow from Lake Tahoe in California and terminates approximately 115 miles (185 km) later in Pyramid Lake in Nevada. The Carson River has its headwaters approximately 50 miles (80 km) south of Lake Tahoe, runs almost parallel to the length of the Truckee River and terminates in the Carson Sink area. The areas of the basins are comparable and are approximately 3000 square miles (7770 km2). The Bureau of Reclamation (BOR) Lahontan Basin area office manages operations on the Truckee and Carson Rivers and relies heavily on seasonal (i.e. spring) streamflow forecasts for planning and management. One of the key management issues is the interbasin transfer of water from the Truckee Basin to Lahontan Reservoir in the Carson Basin through the one-way Truckee Canal (Horton, 1995). This transfer augments storage in Lahontan Reservoir for later use by the Newlands Project irrigation district and other water users. If managers divert too much water into the Truckee Canal, they leave insufficient flows in the Truckee River to support other water users, including endangered fish populations, along the last reach of the river. Yet, if managers divert too little water, farmers in the Newlands Project district will have insufficient water in storage to sustain their crops throughout the season. The multiple users with competing objectives coupled with limited canal capacity and the short water season require that managers use seasonal forecasts for planning and management. Recently implemented policies limit diversions through the Truckee Canal and require specific reservoir releases to aid in the protection of the endangered fish populations – adding further constraints to the reservoir operations and management. The accuracy of forecasts has become evermore important to the efficient management of the water-stressed Truckee and Carson River Basins. The BOR currently implements forecasts of the spring runoff (April–July volume) into seasonal planning and basin management. These forecasts are issued on the first of each month starting from January. The January forecast affects flood control operations and is used to estimate the irrigation demand for the coming season and, thus, affects reservoir releases and diversions into the Truckee Canal. Updated forecasts in the ensuing months up to April 1 and throughout the runoff season continue to guide operations throughout the basin. Current forecasting techniques use multiple linear regression analysis based on factors related to the existing snowpack and, hence, long-lead forecast skills are limited. Additionally, the current technique does not provide forecasts prior to January as the snowpack information is only partial. Thus, improvements to the spring forecasts, both in skill and in lead time, are needed to strengthen planning and operations in the Truckee and Carson Basins.

102 7.3.2

Balaji Rajagopalan et al. Gunnison

The Gunnison River Basin (Figure 7.3) resides largely in the southwestern part of the state of Colorado and, is a major tributary of the Colorado River. It consists of six sub-basins, that is, East-Taylor (760 square miles; 1968 km2), Upper Gunnison (2380 square miles; 6164 km2), Tomichi (1090 square miles; 2823 km2), North Fork (959 square miles; 2484 km2), Lower Gunnison (1630 square miles; 4222 km2), and Uncompahagre (1110 square miles; 2875 km2). The basin has a drainage area of approximately 20,534 km2 and basin elevations are extremely variable, ranging from 1387 to 4359 m (McCabe, 1994). It contributes approximately 42% of the streamflow of the Colorado River at the Colorado–Utah Stateline (Ugland et al., 1990). Like Truckee–Carson, almost all of the annual flow in the basin occurs during spring (April–July) due to snowmelt from the higher elevations. The streamflows on the Gunnison impact municipal water supply, power generation, and flow release for endangered species. Therefore, like on the Truckee–Carson skilful forecast of spring seasonal streamflows in the basin is the key to improvement water management. 7.4

DATA

The following data sets for the period 1949–2003 are used in the analysis: 1 Monthly natural streamflow data for Farad and Ft Churchill gaging stations on the Truckee and Carson Rivers, respectively, obtained from USBR. Natural streamflows are computed based on inflows to the seven major storage reservoirs near the top of the basin before any significant depletion has been made (Jeff Rieker, 2003, personal communication). Spring seasonal (April–July) volume is computed from the monthly streamflows that are used in this study. 2 Gunnison Basin streamflows at six locations (Figure 7.3) are selected from the Hydro Climate Data Network (HCDN). This network, HCDN, was developed by USGS (http://water. usgs.gov) to analyze the climate impacts on the rivers and it has more than 1000 streamflow stations across the conterminous United States that is not affected by human activities (Slack and Landwehr, 1992). 3 Monthly SWE data obtained from the NRCS National Water and Climate Center website (http://www.wcc.nrcs.usda.gov). The SWE data is gathered from snow course and Snotel stations in the upper Truckee Basin (17 stations) and upper Carson Basin (7 stations). For Gunnison too we had 13 SWE stations. Basin averages of SWE are calculated using the method employed by the NRCS: the SWE depth from every station in the basin is summed and then divided by the sum of the long-term averages for each of the stations (Tom Pagano, 2003, personal communication). 4 Monthly winter precipitation data for the California Sierra Nevada Mountains region. This is obtained from the US climate division data set from the NOAA-CIRES Climate Diagnostics Center (CDC) website (http://www.cdc.noaa.gov). 5 Monthly values of large-scale ocean atmospheric variables – (SST), Geopotential heights (Z500, Z700), Sea Level Pressure (SLP), wind, etc., from NCEP/NCAR Reanalysis project (Kalnay et al., 1996) also obtained from the CDC website. 7.5

CLIMATE DIAGNOSTICS AND PREDICTOR SELECTION

The first step in the forecasting framework is to identify large-scale climate predictors of spring streamflows in the basin. To this end, we first examined the relationship between SWE and spring runoff in the basins. Next, we correlated spring streamflows with global climate variables from preceeding fall and winter seasons. We chose to examine variables from fall and winter

Ensemble streamflow forecasting 103 because the state of the atmosphere during this time affects the position of the jet stream and consequently, snow deposition, and the resulting spring runoff. Also, predictors from fall and winter allow for potential long lead forecasts. 7.5.1

Truckee–Carson Basin

As expected, there is a high degree of correlation between winter SWE and spring runoff, particularly with April 1 SWE as it provides a more complete representation of the end of winter snowpack in the basins. Correlation values for Truckee spring streamflows are 0.80 and 0.9 with March 1 SWE and April 1 SWE, respectively, and 0.81 and 0.9, respectively, with the Carson flows. High correlations of streamflows with March 1 SWE offers the opportunity for at least a one month-lead forecast. January 1 SWE, however, does not correlate as well with spring streamflows (0.53 for the Truckee and 0.49 for the Carson) and, hence, provides less skill as a predictor of spring runoff. The snow information by January 1 is only partial and hence, the weak correlation with spring flows. Spring streamflows in the Truckee and Carson Basins are likely modulated by ENSO and PDO, but their standard indices of these phenomena did not show significant correlations with spring streamflows (0.22 for the NINO3, ⫺0.13 for the PDO, and ⫺0.21 for the SOI, for the Truckee; results are similar for the Carson). Thus, we correlated the spring streamflows with the standard ocean-atmospheric circulation fields (e.g. 500 mb geopotential height fields, SSTs, SLPs, etc.) to investigate the large-scale climate link and potential predictors. Correlation map of Carson River spring streamflows and the preceding winter SSTs and 500 mb geopotential heights, henceforth, referred to as Z500, are shown in Figure 7.5. Strong negative correlations (approximately ⫺0.7) with Z500 in the region off the coast of Washington can be seen. The SSTs in the northern mid-Pacific Ocean exhibited a strong positive (about 0.5) correlation and to the east of this, a negative correlation. Similar, but slightly weaker correlation patterns were seen with preceding fall (September–November) Z500 and SSTs (Grantz, 2003), suggesting that the physical mechanisms responsible for the correlations are persistent from fall through winter. These correlations offer hopes for a long-lead forecast of spring streamflows – at the least, they can provide significant information about the upcoming spring streamflows in fall, before SWE data is available. (a)

(b)

60N

60N

50N

50N

40N

40N

30N

30N

20N

20N

10N

10N

EQ

EQ

10S

10S

20S

20S

30S 100E 120E 140E 160E 180 160W 140W 120W 100W 80W 60W December – February: 1949–2001: 500 mb geopotential height Seasonal correlation w/April – June Carson spring flow (index lags by 4 months) NCEP/NCAR Reanalysis

30S 100E 120E

–0.7 –0.6 –0.5 –0.4 –0.3 –0.2 –0.1

–0.7 –0.6 –0.5 –0.4 –0.3 –0.2 –0.1

0

0.1 0.2

0.3 0.4 0.5

0.6 0.7

140E 160E 180 160W 140W 120W 100W 80W 60W December – February: 1949 – 2001: surface SST Seasonal correlation w/April – June Carson spring flow (index lags by 4 months) NCEP/NCAR Reanalysis 0

0.1 0.2

0.3 0.4 0.5

0.6 0.7

Figure 7.5. Correlation of Carson River spring streamflows with winter (December–February) climate variables (a) 500 mb geopotential height (Z500) and (b) SST (see Color Plate XXIII).

104 Balaji Rajagopalan et al. To understand the physical mechanisms driving the correlation patterns seen above, a composite analysis was performed. In this, average SST, wind, and Z500 patterns for high and low streamflow years were obtained to identify coherent regions with strong magnitudes of the variables. We chose years with streamflows exceeding the 90th percentile as “high” years and those below the 10th percentile as “low” years. Figure 7.6 shows the composites of vector wind, Z500, and SST anomalies during the winter season preceding the high and low streamflow years. The winds in high streamflow years show a counterclockwise rotation around the low pressure region off the coast of Washington – the region of highest correlation seen in Figure 7.5. This counterclockwise rotation brings southerly winds over the Truckee and Carson Basins. Southerly winds tend to be warm and moist, thus increasing the chances of enhanced winter snow and, consequently, higher streamflows in the following spring. The opposite pattern is seen during low streamflow years when anomalous northeasterlies tend to bring cold dry air and, consequently, less snow, and decreased streamflows. The Z500 patterns and the vector wind anomalies in high and low streamflow years are consistent with each other. The SST patterns in high and low streamflow years (Figure 7.5) are a direct response to the pressure and winds. The winds are generally stronger to the east of a low pressure region – this increases the evaporative cooling and also increases upwelling of deep cold water to the surface. Together, they result in cooler than normal SSTs to the east of the low pressure region. The opposite is true on the west side of the low pressure region. Composite maps for the fall season show similar patterns – indicating that the physical mechanisms are persistent. Results for the Truckee River streamflows are very similar (Grantz, 2003). Thus, based on the correlation and composite analyses we developed predictors of the Truckee and Carson Basins by averaging the ocean-atmospheric variables over the areas of highest correlation (e.g. as in Figure 7.4). These areas were determined by visual inspection of the correlation maps. Specifically, the Z500 index was obtained as the average over the region 225⬚–235⬚E and 42⬚–46⬚N and the SST predictor index as the average over 175⬚–185⬚E and 42⬚–47⬚N. Time series of these indices were obtained to be used as predictors in the forecast model. 7.5.2

Gunnison

In the Gunnison River Basin we have streamflows from six locations (Figure 7.3) that are highly correlated with each other (correlation coefficients of 0.75 and higher). One option is to create a basin streamflow series by averaging the flows across the six locations. The other is to perform 60N

NCAA–CIRES/Climate Diagnostic Center

60N

50N

50N

40N

40N

30N

30N

20N

20N

10N

10N

EQ

EQ

10S

10S

20S

20S

30S 100E 120E 140E 160E 180 160W 140W 120W 100W 80W December – February: 1952, 1967, 1969, 1982, 1983, 1995

1

1.5

2

2.5

3

3.5

4

60W

NCAA–CIRES/Climate Diagnostic Center

30S 100E 120E 140E 160E 180 160W 140W 120W 100W 80W December– February: 1961, 1976, 1977, 1988, 1992, 1994

1

1.5

2

2.5

3

3.5

60W

4

Figure 7.6. Composites of vector winds, SST and Z500 during the winter of high and low streamflow years (see Color Plate XXIV).

Ensemble streamflow forecasting 105 a Principal Component Analysis (PCA) on the streamflow data and retain the first principal component (PC), which is in essence the average of the six streamflows. In fact, the first principal component (PC1) is correlated 0.99 with the basin average streamflows. The PCA is briefly described in the context of ensemble forecast in the following section. The first PC of spring flows is correlated with large-scale climate variables from preceding seasons to identify predictors. Figure 7.7 shows the correlation map of PC1 with winter geopotential height at 700 mb (Z700) and SST. A strong negatively correlated region of Z700 can be seen over southwestern United States also, highly correlated regions of SSTs observed in the Central and Northern Pacific. We also correlated PC1 with other circulation variables such as, zonal and meridional winds (figures not shown). These correlation patterns were persistent in the preceding fall season as well. To understand the physical mechanisms, composite maps of vector winds in the wet and dry years are shown in Figure 7.8. The wet and dry years are defined as the years with PC values greater than the 90th and below the 10th percentiles, respectively. Notice a strong counterclockwise flow around the region with strong negative correlation with Z700 (Figure 7.7(a)) for the wet years and vice-versa in the dry years. This implies, advection of warm moist air from the south to the basin, thus tending to produce more snow and consequently, higher streamflows in the following spring. These are very similar to the findings from the Truckee–Carson streamflows (Figures 7.5 and 7.6). (a) 70N

(b) 70N

60N

60N

50N

50N 40N

40N

30N

30N

20N

20N

10N 10N EQ EQ

10S

10S 140E

160E 180 160W 140W 120W 100W 80W 60W 40W 20S 140E 160E 180 160W 140W 120W 100W 80W 60W 40W November – March: 1949–99: 700 mb geopotential height November – March: 1949 – 99: surface SST Seasonal Correlation w/April –August PC1–AMJJ flows (index lags by 5 months) Seasonal Correlation w/April – August PC1–AMJJ flows (index lags by 5 months) NCEP/NCAR Reanalysis NCEP/NCAR Reanalysis –0.5

–0.4

–0.3

–0.2

–0.1

0

0.1

0.2

0.3

0.4

0.5

–0.6

–0.4

–0.2

0

0.2

0.4

0.6

Figure 7.7. Correlation between PC1 of spring flows and November–March climate indices (a) geopotential height – 700 mb and (b) sea surface temperature (see Color Plate XXV). (a)

NCEP/NCAR Reanalysis 700 mb Vector wind (m/s) composite anomaly 1968– 96 climo

70N

(b)

NOAA–CIRES/Climate Diagnatic Center

NCEP/NCAR Reanalysis 700 mb Vector wind (m/s) composite anomaly 1968 – 96 climo

70N

60N

60N

50N

50N

40N

40N

30N

30N

20N

20N

10N

10N

EQ 10S 140E

0.4

NOAA–CIRES/Climate Diagnatic Center

EQ 160E 180 160W 140W 120W 100W 80W November – March: 1984, 1957, 1995, 1952, 1985, 1997 0.6

0.8

1

1.2

1.4

1.6

1.8

60W

2

10S 40W 140E

0.4

160E 180 160W 140W 120W 100W 80W November– March: 1977, 1981, 1954, 1963, 1990, 1988 0.6

0.8

1

1.2

1.4

1.6

1.8

2

60W

40W

2.2

Figure 7.8. Composite of vector wind at 700 mb for (a) wet years and (b) dry years (see Color Plate XXVI).

106

Balaji Rajagopalan et al.

Based on the correlation maps, we selected four potential predictors of PC1 to be the time series of average values of difference between the positive and negative correlation regions in (1) Z700 (Figure 7.7(a)), (2) SST (Figure 7.7(b)), (3) zonal winds, and (4) meridional winds. Often, the difference between the regions of strong positive and negative correlation constitutes a much better predictor. For example, in the case of the Z700 predictor we took the difference between the average value over the region with negative correlation (32.5⬚–42.5⬚N latitude and 230⬚–250⬚E longitude) and positive correlation (42.5⬚–55⬚N latitude: 275⬚–297.5⬚E longitude) as can be seen from Figure 7.7(a).

7.6

FORECAST MODELS

Statistical forecast models can be represented as: Yt ⫽ f (xt) ⫹ et

(7.1)

where, xt ⫽ (x1t, x2t, x3t, . . . , xpt), t ⫽ 1, 2, . . . , N, f is a function fitted to the predictor variables (x1, x2, . . . , xp), Y is the dependent variable (e.g. spring seasonal streamflow), and et is the errors assumed to be Normally (or Gaussian) distributed with a mean of 0 and variance
. Traditional parametric methods involve fitting a linear function, also known as linear regression. The theory behind the parametric methods, procedures for parameter estimation and hypothesis testing are well developed (e.g. Helsel and Hirsch, 1995; Rao and Toutenburg, 1999). The main drawbacks, however, are: (1) the assumption of a Gaussian distribution of data and errors, (2) the assumption of a linear relationship between the predictors and the dependent variable, (3) higher order fits (e.g. quadratic or cubic) require large amounts of data for fitting, and (4) the models are not portable across data sets, that is, sites. Nonparametric methods, in contrast, estimate the function f “locally.” There are several nonparametric approaches, such as kernel-based (Bowman and Azzalini, 1997), splines, K-nearest neighbor (K-NN) local polynomials (Owosina, 1992; Rajagopalan and Lall, 1999); locally weighted polynomials (Loader, 1999), etc. The K-NN local polynomials and the local weighted polynomial (LOCFIT)1 approaches are very similar. Owosina (1992) performed an extensive comparison of a number of regression methods both parametric and nonparametric on a variety of synthetic data sets and found that the nonparametric methods outperform parametric alternatives. In the following paragraphs we describe a few nonparametric methods for ensemble forecast. 7.6.1 Local methods of ensemble forecast The local polynomial methods obtain the value of the function f at any point ‘x*’ t by fitting a The neighbors can be polynomial to a small set K(⫽ *N,  ⫽ (0, 1]) of neighbors to ‘x*’. t identified based on the Euclidean distance (Lall and Sharma, 1996; Rajagopalan and Lall, 1999) or Mahalanobis distance (Yates et al., 2003). Other approaches include weighting the predictors differently in the distance calculation, such as weighted obtained via coefficients from a linear regression between the dependent variable and predictors (Souza and Lall, 2003). Once the neighbors are identified, there are two main options for generating ensembles: 1 The neighbors can be re-sampled with a weight function that gives more weight to the ‘nearest’ neighbors and less to the farthest, thus generating ensemble (e.g. Souza and Lall, 2003). This has been widely applied for stochastic weather and streamflow simulation in the above mentioned references.

Ensemble streamflow forecasting 107 2 A polynomial or order p can be fit to the neighbors that can be used to estimate the mean of the dependent variable (Rajagopalan and Lall, 1999; Loader, 1999) using LOCFIT and the local variance
le2 of the errors around the mean. The local error variance can be used to generate random normal deviates which, when added to the mean estimate, yield ensembles. Thus, the parameters to be estimated are the size of the neighborhood (K ) and the order of the polynomial ( p), which is obtained using objective criteria such as Generalized Cross Validation (GCV). GCV(K, p) ⫽

兺

N 2 i⫽1 (ei /N

)

(7.2)

(1 ⫺ (m/N ))2

where ei is the error, N is the number of data points, m is the number of parameters. For stability purposes, the minimum neighborhood size should be twice the number parameters to be estimated in the model. The K, p combination to be selected is the one that has the minimum GCV score. This was used in the Thailand summer rainfall forecast (Singhrattna et al., 2004). 3 In (2) above the local errors are assumed to be normally distributed. Often times this may not be the case. To get over this problem, Prairie (2002) and Prairie et al. (in press) proposed an interesting variation and applied it to streamflow and salinity modeling on the Colorado River Basin. This was later applied to the Thailand summer rainfall forecasting (Singhrattna et al., in press). In this variation, the mean value, Yi of the predictor vector ‘x*’ t is obtained from the is selected and the corresteps described in (2) above. Then, one of the neighbors of ‘x*’ t sponding residual e*t is picked up. This residual is then added to the mean forecast Y*t ⫹ e*, t thus obtaining one of the ensemble members; repeating this several times results in an ensemble. This is pictorially shown in Figure 7.9 for the Truckee river spring streamflows and the Z500 index as the predictor. The solid line is the mean fit using LOCFIT, the points are the observations, and the dashed rectangle is the neighborhood size from which the residuals are resampled. The neighbors are obtained using any of the distance metric described in (2).

Truckee spring flow 1989

260 240

et*

200

200

220

yt*

Volume (kaf)

400

280

600

yt*= f (xt*)+et*

xt*

0

Truckee spring volume (kaf)

Spring flow vs winter geopotential height

–100

Figure 7.9.

–50 0 Winter geopotential height anomaly

50

Residual resampling to obtain an ensemble forecast.

108

Balaji Rajagopalan et al.

Furthermore, the selection of one of the neighbors is done using a weight function of the form: W( j) ⫽

兺

j

1

k i⫽1 (1/i)

(7.3)

This weight function gives more weight to the nearest neighbor and less to the farthest neighbors. The number of neighbors to be used to resample the residuals can be the same as (K) that was used in fitting the local polynomial or could be different. In practice, the 兹N neighbors for resampling the residuals seem to work fine. This heuristic rule is justified by theoretical arguments of Fukunaga (1990). Being local estimation scheme, these methods have the ability to capture any arbitrary local features. Furthermore, unlike the parametric alternatives, no prior assumptions need be made regarding the functional form of the relationship (e.g. a linear relationship, Gaussian distribution, etc.). Other variations using nonlinear dynamics-based time series analysis (Regonda et al., 2004b) can also be explored. 7.6.2

Multisite ensemble forecast

Often times forecasts are required at several sites simultaneously that capture the spatial correlation structure. For example, ensemble forecasts will be required at all the streamflow locations in a basin for use in decision support system. The local methods described above can be used in this case but there needs to be a pre-processing step prior to using them. The steps are as follows: 1 A principal component analysis (PCA) (Preisendorfer, 1988; von Storch and Zwiers, 1999) is performed on the seasonal steamflows at all the sites. PCA provides an orthogonal space-time decomposition, with spatial part represented as Eigen Vectors (EVs) and the temporal part as Principal Components (PCs). The theory and implementation of this is widespread in climate analysis and the above references offer a detailed exposition of this and other related approaches. Typically, the leading PC captures almost all of the variance, especially in homogeneous basins. 2 Predictors are identified for the leading PC. 3 For a given predictor vector (i.e. a given year) ensembles of the leading PC are generated from the local methods described in the previous section. 4 The PC ensembles are back transformed to the original flow space by multiplying with the appropriate EV, thus, resulting in ensembles of streamflows at all the locations simultaneously, and preserving the spatial correlation structure. This approach was recently developed and encouraging results from preliminary application to the Gunnison River Basin in the western United States are presented later in this chapter. 7.6.3

Subset selection

As we saw from the previous section several potential predictors are identified for forecasting the streamflows in two basins. The task then is to select the best predictor subset. In the linear regression framework this is done using stepwise regression (e.g. Rao and Toutenburg, 1999; Walpole et al., 2002), wherein the smallest subset that explains most variance in the dependent variable is selected. Other methods use score functions such as Mallow’s Cp, Akaike Information Criteria (AIC) (Rao and Toutenburg, 1999) and so on that favor parsimony. We propose the use of GCV (equation 7.2) as a tool for subset selection. In this, one fits local polynomial for different predictor combinations along with the polynomial order, the neighborhood

Ensemble streamflow forecasting 109 size, and the GCV value computed in each case. The combination that produces the least GCV value is chosen as the best subset. The GCV function is a good surrogate of predictive error (Craven and Whaba, 1979) of the model, unlike least squares which is a measure of goodness of fit and provides no information on the predictive capability. Hence, we feel that the GCV will be a better alternative for subset selection. Applying the GCV criteria we selected Z500 index and SWE as the best set of predictors for the Truckee–Carson streamflows and SST predictor and SWE for the Gunnison. These respective subsets will be used in the ensemble forecast. 7.7

MODEL VALIDATION

The large-scale climate predictors identified for the spring seasonal streamflows are used in the ensemble forecast models. Each year is dropped from the record and ensemble forecasts are made for the dropped year based on the rest of the data. This is repeated for all the years, thus obtaining cross-validated ensemble forecasts. The forecasts are issued at several lead times. 7.7.1

Skill measures for validation

Apart from visual inspection, the ensembles are evaluated on a suite of three performance criteria: 1 Correlation coefficient of the mean of the ensemble forecast and the observed value. This measures the skill in the mean forecast. 2 Ranked Probability Skill Score (RPSS) (Wilks, 1995). 3 Likelihood Function Skill Score (LLH) (Rajagopalan et al., 2002). RPSS and LLH measure the forecast’s ability to capture the probability distribution function (PDF). The RPSS is typically used by climatologists and meteorologists to evaluate a model’s skill in capturing categorical probabilities relative to climatology. We divided the streamflows into three categories, at the tercile boundaries, that is, 33rd percentile and 66th percentile. Values above the 66th percentile are in the “above normal” category, below the 33rd percentile are in the “below normal” category, and the remainder fall in the “normal” category. Of course, one can divide into unequal categories as well. The categorical probability forecast is obtained as the proportion of ensemble members falling in each category. The “climatology” forecast is the proportion of historical observations in each category. For the tercile categories presented here the climatological probability of each category is 1/3. For a categorical probabilistic forecast in a given year, P ⫽ (P1, P2, . . . , Pk) (where k is the number of mutually exclusive and collectively exhaustive categories – here it is 3) the rank probability score (RPS) is defined as: RPS( p, d ) ⫽

1 k⫺1

冤冢 兺 k

i⫽1

Pi ⫺

k

兺

i⫽1

冣冥 2

di

(7.4)

The vector d (d1, d2, . . . , dk) represents the observations, such that dk equals one if the observation falls in the kth category and zero otherwise. The RPSS is then calculated as (Toth, 2002): RPSS ⫽ 1 ⫺

RPS (forecast) RPS (climatology)

(7.5)

The RPSS ranges from positive 1 (perfect forecast) to negative infinity. Negative RPSS values indicate that the forecast has less accuracy than climatology. The RPSS essentially measures how often an ensemble member falls into the category of the observed value and compares that

110

Balaji Rajagopalan et al.

to a climatological forecast. The likelihood function is also used to quantify the skill of ensemble forecasts. This function compares the likelihood of the ensemble forecast falling into the observed category with respect to climatology. The LLH for the ensemble forecast in any given year is calculated as:

L⫽

写 写

ˆ

N P t ⫽ 1 j ,t

(7.6)

N

t ⫽ 1 Pcj,t

Where N is the number of years to be forecasted, j is the category of the observed value in year t, P ˆj,t is the forecast probability for category j in year t, and Pcj,t is the climatological probability for category j in year t. The LLH values range from 0 to number of categories (3 in this study). A score of zero indicates lack of skill; a score of greater than 1 indicates that the forecasts have skill in excess of the climatological forecast, and a score of 3 indicates a perfect forecast. The LLH is a nonlinear measure and is related to information theory (Rajagopalan et al., 2002). 7.7.2

Results

7.7.2.1 Truckee–Carson For the Truckee–Carson Basins we had identified in the previous section, two best predictors – Z500 index and SWE. The SWE, however, is only available from January 1 onwards, as prior to that there is little snow in the basin and hence, of limited use for forecasting spring streamflows. Given that SWE is highly correlated spring streamflows, the utility of Z500 index in the model could be doubted. To investigate this, ensemble forecasts are issued for the Truckee and Carson Rivers on the 1 of each month from November through April from two models – one using Z500 and SWE as predictors and another with only the SWE. The skill scores were computed for the forecasts and are shown in Figure 7.10. The results show that using the Z500 index together with SWE as predictors provides better skills at all lead times. This is a significant outcome in that it clearly demonstrates the importance of incorporating basin specific large-scale climate indices in streamflow forecasts. It is also apparent from Figure 7.10 that the forecast skills are above climatology at all lead times (the RPSS is above zero and the LLH is above 1), indicating the presence of useful information about the spring streamflows from as early as fall. As in most forecasting models, the skills on all the measures improve with decreased lead time. To assess the performance of the model in extreme years we calculated the RPSS and LLH for wet and dry years. We define years with streamflows above the 75th percentile as wet and those below the 25th percentile as dry. Roughly 12 years fall into each category. Median skills for forecasts issued on April 1 and December 1 are shown in Table 7.1. It is apparent that the model has a slightly higher skill in predicting the wet years relative to dry. This asymmetry in the skills is consistent with the nonlinearities seen in the relationship between the predictors and the streamflows (Figure 7.9). Whereas high streamflow years exhibit a strong linear relationship with the Z500 index, this relationship breaks down, that is, flattens out in low streamflow years. The skills for forecast issued on December 1 are relatively poor but there are substantial numbers of these, especially in the extreme years – providing useful long lead forecast for water resources planning. Ensemble forecasts provide the PDF and consequently, they can be used to obtain threshold exceedence probabilities. This information can be very useful for water managers in preparing for extreme events. Figure 7.11 presents the PDF of the ensemble forecasts for 1992 and 1999, below normal and above normal streamflow years, respectively. The climatological PDF, that is,

Truckee RPSS results

Carson RPSS results 1.0 Median RPSS (all years)

Median RPSS (all years)

1.0 0.8 0.6 0.4 GpH and SWE SWE

0.2 0.0 Nov 1

Dec 1

Jan 1

Feb 1

Mar 1

Apr 1

0.8 0.6 0.4

0.0 –0.2

–0.2

GpH and SWE SWE

0.2

Nov 1

Dec 1

Jan 1

Feb 1

Mar 1

Apr 1

Month Month Carson forecasted vs observed correlation coeff.

1

1

0.8

0.8

Correlation coeff.

Correlation coeff.

Truckee forecasted vs observed correlation coeff.

0.6 0.4 GpH and SWE SWE

0.2

0.6 0.4

0

0 Nov 1

Dec 1

Jan 1

Feb 1

Mar 1

GpH and SWE SWE

0.2

Nov 1

Apr 1

Dec 1

2 1.5 1 GpH and SWE SWE Dec 1

Jan 1 Feb 1 Month

Mar 1

Apr 1

2.5 2 1.5 1

Mar 1

GpH and SWE SWE

0.5

0 Nov 1

Feb 1

Carson likelihood results Median likelihood (all years)

Median likelihood (all years)

Truckee likelihood results 2.5

0.5

Jan 1 Month

Month

0 Nov 1

Apr 1

Dec 1

Jan 1

Feb 1

Mar 1

Apr 1

Month

Figure 7.10. Skill scores of forecasts issued from the first of each month November–April for Truckee and Carson rivers (see Color Plate XXVII). Table 7.1. Median skill scores for ensemble forecast issued on April 1, for all years, wet and dry, for the Truckee–Carson Basin. The values in parentheses are for forecast issued on December 1. Median skill score RPSS

All years Wet years Dry years

LLH

Truckee

Carson

Truckee

Carson

1.0 (0.2) 1.0 (0.4) 0.9 (0.0)

0.9 (0.0) 1.0 (0.3) 0.6 (0.0)

2.3 (1.1) 3.0 (1.1) 2.2 (1.1)

2.3 (1.1) 2.6 (1.2) 2.2 (1.1)

112 Balaji Rajagopalan et al.

(b) Ensemble forecast (P= 0.59) Observed value Climatology (P= 0.17)

Ensemble forecast (P= 0.49)

0

PDF 0.010

Observed value

Climatology (P= 0.92)

100 200 300 400 500 600 Truckee spring runoff 1992 (kaf)

0.000

PDF 0.000 0.004 0.008

(a)

0

100 200 300 400 500 600 Truckee spring runoff 1992 (kaf)

Figure 7.11. PDF of the ensemble forecasts in a (a) dry year (1992) and (b) wet year (1999) for the Truckee River. 1

Skill Score (RPSS)

Taylor River, Almont 0.8

East Rover, Almont

0.6

Tomichi Creek, Gunnison Lake fork, Gate view

0.4

North fork Gunnison River, Somerset Uncompahgre River, Colona

0.2 0 January 1 April 1 Date of issuing forecast

Figure 7.12. Median RPSS score for forecasts issued on January 1 and April 1 for the six streamflow sites.

the PDF of the historical data, is overlaid in these plots. Notice that the PDFs of the ensemble forecasts are shifted toward the observed values. In 1992, a dry year, the observed streamflow in the Truckee River was 75 kaf (~93 M m3), much below the historical average. Based on the climatological PDF the exceedence probability of this value is 0.92, while that from the ensemble forecasts is 0.49, indicative of drier conditions. Similarly, for the above average flow of 408 kaf (~504 M m3) in 1999, climatology suggested an exceedence probability of 0.17, whereas the ensemble forecasts show a much higher probability of exceedence (0.59), thereby better capturing the probability of the observed flow value. 7.7.2.2 Gunnison In this basin as described earlier, we generate ensemble forecast of the first PC and then multiply it with the EV to provide the ensemble streamflow forecast at all the six locations. Figure 7.12 shows the RPSS for forecast issued on January 1 and April 1. It can be seen that the skills increase with lead time and they are quite good overall. To evaluate the forecasts in extreme (wet and dry) and average years, the ensemble forecasts for the East River site is shown in Figure 7.13. We loosely defined, for this purpose, the wet years as those with streamflows above the upper tercile, the dry years as those below the lower tercile, and the rest as average. The ensembles are shown as boxplots with the boxes being the interquartile range, the

Ensemble streamflow forecasting 113

Flow (Kaf/season) 50 150 250 350

East River, Almont (09112500)

Dry years 1977 1981 1954 1990 1963 1961 1988 1959 1992 1966 1976 1955 1964 Year

Flow (Kaf/season) 150 250 350

East River, Almont (09112500)

50

Wet years 1957 1995 1984 1952 1965 1997 1986 1993 1979 1980 1982 1985 1996 Year

Flow (Kaf/season) 50 100 200 300

East River, Almont (09112500)

Average years 1972

1989

1953

1991

1956

1987

1950 Year

1951

1982

1949

1958

1975

1978

Figure 7.13. Boxplots of ensemble streamflow forecasts at the station East River, Almont, for the dry, wet, and average years. The wet, dry, and average years are divided based on the terciles – that is, years with streamflows values below the lower tercile are dry, those above the upper tercile are wet, and the rest as average. The horizontal lines are the 25th, 50th, 75th, 90th, and 95th percentiles of the historical data.

whiskers at the 5th, and 95th percentile and the points are outside this range, the observed values are shown as thick solid points. If the observed values fall within the box it implies that the ensembles are better able to capture the PDF of the flows in that year. The horizontal lines are the 25th, 50th, 75th, 90th, and 95th percentile values of the historical data. It can be seen that the boxplots are shifted in the right direction in the extreme years. Notice that the skills in the extreme years are particularly good that can be of great importance in water management in the basin. The median skill scores at all the locations for all years, wet and dry, are shown in Table 7.2.

114

Balaji Rajagopalan et al. Table 7.2. Median skill scores for ensemble forecast issued on April 1, for all years, wet and dry, for the Gunnison Basin. The values in parentheses are the LLH. RPSS (LLH)

7.8

USGS Station

Total years

Wet years

Dry years

09110000 09112500 09119000 09124500 09132500 09147500

0.72 (1.91) 0.92 (2.58) 0.73 (1.91) 0.73 (1.84) 0.91 (2.35) 0.87 (2.27)

1.00 (3.00) 1.00 (3.00) 0.98 (2.72) 0.93 (2.39) 1.00 (3.00) 1.00 (3.00)

0.92 (2.35) 0.98 (2.68) 0.97 (2.64) 0.91 (2.32) 0.92 (2.35) 0.97 (2.68)

SUMMARY AND DISCUSSION

We presented a framework for ensemble forecast that uses large-scale climate information and demonstrated its utility on providing seasonal streamflow forecasts in two river basins in the western United States. Climate diagnostics is first performed to obtain a suite of potential largescale climate predictors. Local polynomial based nonparametric methods can then be used to identify the best subset of predictors and use them to generate ensemble forecasts. We developed methods for ensemble forecasts at a single site and also for multisite preserving the spatial correlation. The proposed nonparametric methods are data driven and provide a flexible and powerful alternative to traditional parametric (i.e. linear regression) methods in capturing any arbitrary relationship between the predictors and the dependent variable and error structure. Application to Truckee–Carson and Gunnison River Basins show that significant long-lead skill in forecasting seasonal streamflows can be achieved, especially in extreme years. This has tremendous impact on improving the water resources management and planning in these basins. Our preliminary application of these forecasts on the Truckee–Carson Basin to improve the operations of Truckee canal (descried earlier in the chapter) gives encouraging results (Grantz, 2003). The proposed framework can be easily applied to any other basin and other variables – for example, we applied this to Thailand summer rainfall forecast with good success (Singhrattna et al., 2004). ACKNOWLEDGMENTS We thank the Bureau of Reclamation Lahontan Basin area office for funding the Truckee–Carson study. Funding through the CIRES Innovative Research Program at the University of Colorado at Boulder is also thankfully acknowledged. Useful discussions with Tom Scott, Tom Pagano, and Jeff Rieker are very much appreciated. NOTE 1 LOCFIT is the package to perform local polynomial fits developed by Loader (1999) and available freely at http://cm.bell-labs.com/cm/ms/departments/sia/project/locfit/index.html

REFERENCES Allan, R., Lindesay, J., and Parker, D. (1996) El Nino Southern Oscillation and Climatic Variability. National Library of Cataloguing-in-Publication, Collingwood.

Ensemble streamflow forecasting 115 Bowman, A. and Azzalini, A. (1997) Applied Smoothing Techniques for Data Analysis. Oxford: Oxford University Press. Cayan, D. and Webb, R. (1992) El Nino/Southern Oscillation and streamflow in the western Unites States. In: Henry F. Diaz and Vera Markgraf (eds), El Nino, Cambridge: Cambridge University Press, pp. 29–68. Clark, M.P. and Serreze, M.C. (2001) Historical effects of El Nino and La Nino events on the seasonal evolution of the mountains snowpack in the Columbia and Colorado River Basins. Water Resources Research, 37, 741–757. Craven, P. and Whaba, G. (1979) Optimal smoothing of noisy data with spline functions. Numerische Mathematik, 31, 377–403. el-Ashry, M. and Gibbons, D. (1988) Water and Arid Lands of the Western United States. New York: Cambridge University Press. Fukunaga, K. (1990) Introduction to Statistical Pattern Recognition. San Diego, CA: Academic Press. Gershunov, A. (1998) ENSO influence on intraseasonal extreme rainfall and temperature frequencies in the contiguous United States: implications for long-range predictability. Journal of Climate, 11, 3192–3203. Grantz, K. (2003) Using large-scale climate information to forecast seasonal streamflow in the Truckee and Carson rivers. MS Thesis, Colorado, University of Colorado at Boulder. Hamlet, A.F., Huppert, D., and Lettenmaier, D.P. (2002) Economic value of long-lead streamflow forecasts for Columbia river hydropower. Journal of Water Resources Planning and Management, March/April, 91–101. Helsel, D.R. and Hirsch, R.M. (1995) Statistical Methods in Water Resources. Amsterdam: Elsevier Science Publishers B.V. Hoerling, M.P., Kumar, A., and Zhong, M. (1997) El Nino, La Nina, and the nonlinearity of their teleconnections. Journal of Climate, 10, 1769–1786. Horton, G.A. (1995) Truckee River Chronology. Division of Water Planning, Department of Conservation and Natural Resources, Carson City, Nevada. Kalnay, E. et al. (1996) The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society, 77, 437–471. Lall, U. and Sharma, A. (1996) A nearest neighbor bootstrap for resampling hydrologic time series. Water Resources Research, 32(3), 679–693. Loader, C. (1999) Statistics and Computing: Local Regression and Likelihood. New York: Springer. McCabe, G.J. (1994) Relationships between atmospheric circulation and snowpack in the Gunnison River Basin, Colorado. Journal of Hydrology, 157, 157–175. McCabe, G.J. and Dettinger, M.D. (2002) Primary modes and predictability of year-to-year snowpack variation in the western United States from teleconnections with Pacific Ocean climate. Journal of Hydrometeorology, 3, 13–25. Mantua, N.J., Hare, S.R., Wallace, J.M., and Francis, R.C. (1997) A Pacific interdecadal climate oscillation with impacts on salmon production. Bulletin of the American Meteorological Society, 78, 1069–1079. Newman, M., Compo, G.P., and Alexander, M.A. (2003) ENSO-forced variability of the Pacific decadal oscillation. Journal of Climate, 16, 3853–3857. Owosina, A. (1992) Methods for assessing the space and time variability of groundwater data. MS Thesis, Utah State University, Logan, Utah. Pagano, T. (2003) Water supply forecaster, natural resources conservation service. Personal communication, October. Piechota, T.C., Hidalgo, H., and Dracup, J. (2001) Streamflow variability and reconstruction for the Colorado River Basin. In: Proceedings of the EWRI World Water and Environmental Resources Congress, May 20–24 Orlando, FL, American Society of Civil Engineers, Washington, DC. Prairie, J.R. (2002) Long-term salinity prediction with uncertainty analysis: application for Colorado River above Glenwood Springs. MS Thesis, Colorado, University of Colorado at Boulder. Prairie, J.R., Rajagopalan, B., Fulp, T., and Zagona, E. (2005) Statistical nonparametric model for natural salt estimation. ASCE Journal of Environmental Engineering (in press). Preisendorfer, R.W. (1988) Principal Component Analysis in Meteorology and Oceanography. New York: Elsevier, 425pp. Pulwarty, R.S. and Melis, T.S. (2001) Climate extremes and adaptive management on the Colorado River: lessons from the 1997–1998 ENSO event. Journal of Environmental Management, 63, 307–324. Rajagopalan, B. and Lall, U. (1999) A nearest neighbor bootstrap resampling scheme for resampling daily precipitation and other weather variables. Water Resources Research, 35(10), 3089–3101.

116

Balaji Rajagopalan et al.

Rajagopalan, B., Lall, U., and Zebiak, S. (2002) Optimal categorical climate forecasts through multiple GCM ensemble combination and regularization. Monthly Weather Review, 130, 1792–1811. Rao, C.R. and Toutenburg, H. (1999) Linear Models: Least Squares and Alternatives. New York: Springer. Rasmussen, E.M. (1985) El Nino and variations in climate. American Scientist, 73, 168–177. Redmond, K.T. and Koch, R.W. (1991) Surface climate and streamflow variability in the western United States and their relationship to large scale circulation indices. Water Resources Research, 27, 2381–2399. Regonda, S., Rajagopalan, B., Clark, M., and Pitlick, J. (2004a) Seasonal cycle shifts in hydroclimatology over the western US. Journal of Climate (in press). Regonda, S., Rajagopalan, B., Lall, U., Clark, M., and Moon, Y. (2004b) Local polynomial method for ensemble forecast of time series. Nonlinear Processes in Geophysics, Special issue on “Nonlinear Deterministic Dynamics in Hydrologic Systems: Present Activities and Future Challenges” (in press). Rieker, J. (2003) Water Resources Engineer, US Bureau of Reclamation. Personal communication. Ropelewski, C.F. and Halpert, M.S. (1986) North American precipitation and temperature patterns associated with El Nino-Southern Oscillation (ENSO). Monthly Weather Review, 114, 2352–2362. Serreze, M.C., Clark, M.P., Armstrong, R.L., McGinnis, D.A., and Pulwarty, R.S. (1999) Characteristics of the western United States snowpack from snowpack telemetry (SNOTEL) data. Water Resources Research, 35, 2145–2160. Singhrattna, N., Rajagopalan, B., Clark, M., and Krishna Kumar, K. (2004) Forecasting Thailand summer monsoon rainfall. International Journal of Climatology (in press). Slack, J.R. and Landwehr, J.M. (1992) Hydro-climatic data network (HCDN): a U.S. geological survey streamflow data set for the United States for the study of climate variations, 1874–1988. US Geological Survey Open-File Report, 92–129, 200pp. Souza, F.A. and Lall, U. (2003) Seasonal to interannual ensemble streamflow forecasts for Ceara, Brazil: applications of a multivariate, semi-parametric algorithm. Water Resources Research, 39, 1307–1320. Toth, Z. (2002) Assessing the value of probabilistic forecasts from a scientific perspective. Validation of Probabilistic Forecasts, Predictability Seminar, ECMWF, September 9–13. Ugland, R.C., Cochran, B.J., Hiner, M.M., Kretschman, R.G., Wilson, E.A., and Bennett, J.D. (1990) Water Resources Data for Colorado, Water Year 1990, Vol. 2, Colorado River Basin. US Geological Survey Water-Data Report CO-90-2, 344pp. Von Storch, H. and Zwiers, F.W. (1999) Statistical Analysis in Climate Research. New York: Cambridge University Press, 484pp. Walpole, R.E., Myers, R.H., Myers, S.L., and Ye, K. (2002) Probability and Statistics for Engineers and Scientist. Upper Saddle River, NJ: Prentice Hall. Wilks, D. (1995) Statistical Methods in the Atmospheric Sciences. San Diego, CA: Academic Press. Yarnal, B. and Diaz, H.F. (1986) Relationships between extremes of the Southern Oscillation and the winter climate of the Anglo-American Pacific coast. Journal of Climatology, 6, 197–219. Yates, D.S., Gangopadhyay, S., Rajagopalan, B., and Strzepek, K. (2003) A technique for generating regional climate scenarios using a nearest neighbor bootstrap. Water Resources Research, 39(7), SWC 7(1–15).

CHAPTER 8

Regional Climatic Variability and its Impacts on Flood and Drought Hazards B. Gozzini, M. Baldi, G. Maracchi, F. Meneguzzo, M. Pasqui, and F. Piani Institute of Biometeorology – National Research Council (IBIMET–CNR), Via Caproni 8, I-50145 Firenze, Italy

8.1

INTRODUCTION

The long term management and planning of catchments for flood protection and water resource conservation, agricultural practices (irrigation, crop types and yields), civil and industrial water supply, hydroelectric power, etc. need the accurate knowledge of heavy rainfall and drought regimes which are connected to the climate variability and trends. According to the Third Assessment Report (TAR) of the Intergovernmental Panel on Climate Change (IPCC) (Houghton et al., 2001), the climate system is forced by natural factors (solar energy output, volcanic ash and aerosols, internal dynamics, and feedbacks) and anthropogenic forcings (emission of greenhouse gases and aerosols, land use changes). There is a general understanding and agreement on the sign (positive) of the thermal response of the climate system, the surface, and the atmosphere to the current anthropogenic forcings, although large uncertainties concerning the warming rate remain (e.g. Houghton et al., 2001). The chance of abrupt warming resulting from temporary heat storage in large natural reservoirs, especially the deep oceans (Pielke, 2003), the effects of the land use, vegetation and carbon cycle feedbacks (Jones et al., 2003; Rial et al., 2003), and of the water vapor feedbacks (Del Genio, 2002) are still not completely understood. On the other hand, uncertainty is left even on sign of the changes in the intensity of the hydrological cycles in the warming climate. Yang et al. (2003) demonstrated that the average global annual precipitation change is almost linearly dependent upon surface warming and in the case of small surface warming, is connected to small sea surface temperature (SST) variations, that is, with low sensitivity to CO2 increases, the changes in precipitation are small or even negative. The key factor affecting this behavior is the leading mechanism responsible for the preservation of the equilibrium atmospheric temperature, that is, the balance between radiative cooling and condensational heating. Yang et al. (2003) state: for the case of a temperature perturbation initiated by a reduction in radiative cooling (e.g. due to an increase in CO2), a possible pathway for the atmosphere to adjust toward a new steady state of higher temperature is by a decrease in condensational heating and a corresponding reduction in precipitation. On the other hand, for the case of a temperature perturbation initiated by an enhancement in condensational heating, for instance, due to increased sea surface temperatures (SSTs), the atmosphere can adjust itself to a new steady state through an increase in radiative cooling. For both cases, although the new steady-state atmospheric temperature is higher, the changes in the hydrological cycle are in opposite directions. Rial et al. (2003) give a comprehensive picture of the interrelations, nonlinearities, and feedbacks in the climate system, which can lead to abrupt and sudden transitions between very

118 B. Gozzini et al. different near-equilibrium states. This can happen even in the absence of significant external forcings, and can be triggered by such perturbations, which currently appear to have attained their maximum over several hundred thousands years. The global surface temperature, the heat content in large natural reservoirs, and the global hydrological cycle are not the only quantities and features affected by climate change and variability. Recent observations show significant impacts also due to other natural, physical, and ecological systems. Some of these, which directly affect the regional hydrological regimes are briefly described in the following paragraph. According to Chen et al. (2002), the Hadley meridional and Walker zonal overturning tropical circulations, which are associated with stronger equatorial and monsoonal convection patterns (upwelling) and stronger downwelling (drying) in the subtropics (along with higher upward longwave fluxes), have strengthened in the 1990s. This agrees with predictions of enhanced monsoon precipitations over some areas (e.g. over West Africa, see Maynard et al., 2002). Chang and Fu (2002), on the basis of the global (NCEP–NCAR) (National Centres for Environmental Prediction–National Centre for Atmospheric Research) atmospheric and surface reanalyses, showed that the northern hemisphere winter storm track intensity has increased since the early 1970s in both the Pacific and the Atlantic branches. On the basis of the comparison of the NCEP–NCAR reanalyses with unassimilated radiosonde data, Harnik and Chang (2003) confirmed this result, yet limited to the North Atlantic branch. They showed that decadal timescale variations were most effective in decreasing the correlation between the Atlantic and Pacific storm tracks. On the basis of the linear trends’ analysis of the intensity of synoptic-scale processes in the North Atlantic at various time scales of variation (6 h–30 days), Gulev et al. (2002) found that the winter synoptic variability patterns increased from 1958 to 1998 over the northern North Atlantic and diminished over lower latitudes and the western Mediterranean. This closely followed the trends in the low level atmospheric baroclinicity and the Northern Hemisphere Annular (NAM) mode or the North Atlantic Oscillation (NAO, e.g. Hurrell, 1995). On the basis of a mass of data, Hoerling et al. (2001) delineated for the first time, the link between the gradual warming of the tropical oceans, mainly in the Indo-Pacific area, and the North Atlantic winter climate change since 1950. The statistical and numerical analyses demonstrated that (1) the forcing produced by the increasing SST in the global tropical oceans forced the long term increase of the phase of the winter NAO, (2) the extratropical SST does not show a direct feedback to the NAO-like circulations, and (3) the tropical Atlantic SST exert only a marginal impact. Since the warming of the sea surface in the tropics is likely to be mainly a result of the changes in atmospheric composition (Hoerling et al., 2001), the changes of the North Atlantic climate can be considered to be a clear anthropogenic signal. On the other hand, Hurrell and Folland (2002) found a significant increase in summertime sea level pressure over the northeastern North Atlantic and a corresponding northward shift of the mean storm track. Such variations, affecting mostly the latitude belt 45°–70°N and impacting mostly central and northern Europe and the Arctic latitudes of the Atlantic, could be linked to the patterns of organized atmospheric convection over the tropical Atlantic and West Africa. This appears to take place mainly through an atmospheric bridge mechanism similar to that operating over the Pacific sector during El Niño Southern Oscillation (ENSO). A recent study (Gillett et al., 2003) shows evidence for anthropogenic impact on the sea level pressure trends over large portions of both hemispheres (mostly in the northern hemisphere), with large increases at low to mid-latitudes and decreases at higher latitudes. This evidence matches well with the wintertime rising trend of the NAO. Santer et al. (2003) examined the contributions of anthropogenic and natural forcings to the recent increase of the tropopause height. They found that the anthropogenic contribution to the ongoing warming of the troposphere and cooling of the stratosphere predominates over natural

Regional climatic variability 119 forcing. A rising tropopause is likely to impact the chemical, dynamical, and convective processes in the underlying troposphere. 8.2

GLOBAL AND REGIONAL VARIABILITY OF THE HYDROLOGICAL CYCLE

Yang et al. (2003) examine how the global hydrological cycle itself undergoes significant changes over a variety of time and spatial scales as a result of various forcings (the anthropogenic ones likely to prevail in the last few decades), internal feedbacks, and inherent nonlinearities, all affecting the climate system. Groisman et al. (1999) analyzed the trends of the summer daily rainfall regimes in eight countries in the world (Canada, USA, Mexico, former Soviet Union, China, Australia, Norway, and Poland), which together cover about 40% of the global land mass. The observations show that the average monthly precipitation during the twentieth century increased everywhere by about 5% except over China. As the frequency of rainy days did not increase significantly, it follows that the average precipitation increase was due mostly to increased precipitation intensity. The heaviest rainfall events (daily rainfall exceeding area-specific thresholds) increased by nearly 20%, much more than the average precipitation. This increase of the intense summer precipitation is associated with the relative increase of the average tropospheric water vapor content and low to mid-tropospheric temperature, thus suggesting an intensification of the hydrologic cycle. The increase in the extreme precipitation will most likely be much higher than the change of the average seasonal precipitation. In relation to Italy, Brunetti et al. (2001) considered 67 sites during 46 years (1951–96) where daily precipitation data were available. They analyzed the frequency of rainy days and precipitation intensity, both at single stations and over large areas, and reached the following conclusions: ●

● ●

●

The annual number of rainy days decreases significantly, with most of contribution from winter. The daily average rainfall intensity increases significantly in all seasons but winter. In northern Italy, the increase of the precipitation intensity is mainly contributed by the extreme rainstorms, while in the south it is due to the global increase of the average daily rainfall. The precipitation events over high thresholds show a definite increase since 1970s, the opposite of events below low thresholds, getting the highest and the lowest frequencies, respectively, in about 120 years.

The reduction of the frequency of rainy days agrees with the shifts of the atmospheric winter circulation regimes during the last 50 years, while the increase of the precipitation intensity agrees with other observational and modeling studies (e.g. Groisman et al., 1999). In a more recent study, Brunetti et al. (2002) extended the Italian daily precipitation series to year 2000. They showed that while the extreme annual or seasonal events do not exhibit significant trends, probably due to their rarity, an increasing trend becomes evident when more data are included, particularly those related to the precipitation events over high thresholds. Several recent studies evaluated the likely changes of the meteorological and climatic extremes associated with future climate change. On the basis of the integration of multiple Atmosphere-Ocean coupled General Circulation Models (AOGCMs) forced by increasing levels of CO2 and sulphate aerosols, Palmer and Ralsanen (2002) found that while the winter average precipitation increases moderately over central-northern Europe and decreases over southern Europe there is a greater probability of extremely wet winters over most of Europe when CO2 doubling occurs. Christensen and Christensen (2003) showed that along with a likely

120

B. Gozzini et al.

decreasing trend of summer precipitation over central Europe and part of southern Europe (with significant exceptions over northern Italy and eastern Spain) during the twenty-first century, the probability of extreme events (precipitation in five consecutive days above the 99th percentile) could increase significantly more than the local increase of the average seasonal precipitation. Hegerl et al. (2003) compared the global distributions of the changes of precipitation at the time of CO2 doubling as percentages with regard to present climate, as produced by the models CGCM1 of the Canadian Centre for Climate Modelling and Analysis (CCCma) and the Hadley Centre for Climate Prediction and Research (HadCM3), taking into account the annual average and the wettest day of the year. The largest change concerned the wettest day of the year, with greater agreement at the middle and higher latitudes than at the lower latitudes. The two models are in agreement in predicting a lower annual average precipitation and lower peak precipitation over the Iberian Peninsula, a moderate increase of extreme events over central and northern Europe, but no significant agreement over Italy. The features of the hydrological cycle are vital to many communities living in marginal areas of the world such as western tropical Africa, north of the Guinea Gulf, and in particular the Sahel-Sudan area. The Sahel countries receive the largest part of total precipitation during the period May–October, when the interannual and decadal variability is relatively high. Moreover, seasonal distribution of the rainfall is critical to the crop yields. Figure 8.1(a) shows the time series of the average seasonal rainfall (May–October) from 1898 to 2000 (Janowiak, 1988), based on monthly data collected at a maximum of 14 stations in the area 20⬚–8⬚N, 20⬚W–10⬚E (Figure 8.1(b)). The average seasonal rainfall is around 630 mm, with a standard deviation around 127 mm, a minimum value of 336 mm in 1916, and a peak value of 898 mm in 1962. The mechanisms leading to the interannual and decadal variability of the precipitation over the Sahel have been extensively studied, mainly covering interhemispheric SST gradients, ENSO and the Indian monsoon (Ward, 1998), the SST anomalies over the Gulf of Guinea and the eastern North Atlantic (Vizy and Cook, 2002), the SST anomalies over the Mediterranean (Rowell, 2003), the soil moisture (SM) anomalies over North Africa, and even remotely, for example, over Europe (Douville, 2001). The intra-seasonal variability of the monsoon precipitations has been studied in the context of the onset of the intense convective precipitations linked to the abrupt northward shift of the intertropical convergence zone (ITCZ) to sub-Saharan latitudes. The meridional overturning circulation (regional Hadley cell, see e.g. Dima and Wallace, 2003) appears to be linked to the occurrence of a westward-traveling monsoon depression over the Sahel with characteristic periodicities of 20–40 days, able to modulate a part of the subsequent monsoon precipitation evolution (Sultan and Janicot, 2003). The West African monsoon system is a perfect example of a critical quasi-regular periodic feature of the global hydrological cycle subject to numerous teleconnections to other regional, remote, and global boundary conditions, and large scale persistent events. This is also why its prediction has received considerable attention and has gained some skill at seasonal lead times, while its forecast at longer time scales is particularly relevant for the future of those regions. Maynard et al. (2002) performed a numerical simulation of transient climate change, assuming a scenario of moderate emissions of greenhouse gases and the direct and indirect effects of sulphate aerosols. They found that an enhancement of the hydrological cycle and the monsoon precipitations over West Africa could be associated with the increasing climate forcings, along with the intensification and northward displacement of the regional Hadley circulation (hence the potential for wider impacts to the Atlantic and European mid-latitudes). Recently, using the NCEP–NCAR reanalysis and the GPCP (Global Precipitation Climatology Project) rainfall data, Dalu et al. (2005) analyzed the behavior of the West African monsoon in the region confined between 10⬚W–10⬚E and 0⬚–20⬚N and showed that the ITCZ moves northward from its position at 3⬚N in spring to reach 10⬚N in August, where it stays for a month, and retreats southward in early fall.

Regional climatic variability 121

Sahel monsoon rainfall (20°–8°N; 20°W–10°E) – May–October

(a) 1000 900

Precipitation (mm)

800 700 600 500 400

1898 1901 1904 1907 1910 1913 1916 1919 1922 1925 1928 1931 1934 1937 1940 1943 1946 1949 1952 1955 1958 1961 1964 1967 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000

300

(b)

40 35 30 25 20 15 10 5 0 –30 –25 –20 –15 –10

–5

0

5

10

15

20

25

30

Figure 8.1. (a) Seasonal precipitation time series over the Sahel area (data from JISAO, http:// tao.atmos.washington.edu/data_sets/sahel/) and (b) boundaries of the Sahel region.

South of 10⬚N the rainfall is mainly due to the deep convection induced by the ITCZ, whereas in the Sahel (bounded by the ITCZ to the south and by the ITF to the north, 10⬚–20⬚N), the rainfall is induced by the combined action of barotropic–baroclinic instabilities. The moisture advection into the WAM region is related to the SST in the Gulf of Guinea: the monsoon is strong when the water in the gulf is warmer than its climate summer temperature, whereas the monsoon is weak when the water in the gulf is colder. Furthermore, they developed a hydrological index for the onset and the withdrawal of WAM, HOWI, and showed that HOWI precedes the half-value of the rainfall of 2–6 weeks, with the

122 B. Gozzini et al. length of the pre-warning decreasing for increasing latitude. The withdrawal of WAM is more rapid than the onset with the HOWI returning negative at the half-value of the normalized rainfall. Therefore, the HOWI index can be considered as a valuable aid in recognizing bogus onsets in the infancy of WAM, whereas the change of the positive sign of the zonal and of the meridional wind component indicates that the monsoon has arrived. The impact over central-southern Europe (with special focus on the Mediterranean hot summer climate) of the summer diabatic processes over the tropical Atlantic and of the monsoon circulation over West Africa have been investigated by Baldi et al. (2003a,b). They found that the sea surface temperature anomalies (SSTAs) over the Gulf of Guinea, which are among the most effective drivers of the West African monsoon and in turn of the African easterly waves, also impact significantly the sea level pressure, mid-troposphere geopotential heights, Atlantic storm tracks at their exit regions in central Europe and western Mediterranean, and, finally, the precipitation patterns in those areas. Most notably, cold SSTAs over the Gulf of Guinea produce relevant geopotential height anomalies over the western and central Mediterranean and weaker baroclinic storms in late summer. In the normally semi-arid summer climate of the central and western Mediterranean, the occasional storms which penetrate from the North Atlantic bring widespread precipitation and relief from local and regional droughts; the identification and the climatic-scale prediction of mechanisms driving the North Atlantic storm track are thus critically important. By diagnosing the storm track strength by means of the 300-hPa meridional wind variance (computed using a 24-h difference filter; e.g. Harnik and Chang, 2003) and using the NCEP–NCAR reanalysis data, we have computed the composite difference of the storm track strength over North Atlantic, Europe, and the Mediterranean, between years with strong West African monsoon and years with weak monsoon. The results concerning the month of August are shown in Figure 8.2. To obtain this result, the Sahel rainfall data were standardized and the West African monsoon was classified every year since 1950, during July, August, September, July–August,

Storm track change (August, 12 UTC) Positive monsoon anomaly – Negative monsoon anomaly 75°N

–10

10

10

–10

70°N

–10

–10

65°N

–50 –75

60°N 55°N

10 30 –50 10 –30 –75

–100

–75

50°N

–150

–30 –100

50 75 100 150

–10

50

10 30

–50

150 200 100 75 –10 98 50 –30 –100 –150 –50 –75

10 –30

–100

–10

–10

–150 –50

30

50 30 –10

30

–10

–75

–200 –150

–100

–50 –100

–75

–30 –50

–75

35°N 30°N 80°W

30

75 100 75

–75

45°N 40°N

–30 –10 10 –10

10 30 50

–100

70°W

60°W

CNR– IBIMET Climate observatory

50°W

40°W

30°W

–75

20°W

10°W

0

10°E

20°E

30°E 40°E

Processing of NCEP–NCAR reanalyses

Figure 8.2. Composite difference of the storm track strength between strong and weak West African monsoon events during August.

Regional climatic variability 123 August–September, and July–September. The variation is deemed to be strong if the transformed rainfall data exceeded one standard deviation and weak if the same data is lower than one negative standard deviation. The relationship between the strong monsoon and the suppression of the storm track over the Atlantic mid-latitudes and Mediterranean is apparent, along with the reinforcement of the storms at higher latitudes, around 55⬚–65⬚N. This result is suggestive of dramatic consequences for the summer precipitation regime over the Mediterranean and over part of central western Europe, in case the West African monsoon should intensify, as some climate scenarios predict (e.g. Maynard et al., 2002). 8.3

ARNO RIVER BASIN, ITALY – A CASE STUDY

The project “Climate reanalysis and prediction over the Arno River Basin,” funded by the Arno River Basin Authority (Italy), aims at providing quantitative information concerning the past,

(a)

0

2

4

6

8

10

12

14

16

18

48

48

46

46

44

44

42

42

40

40

38

38

36

36 0

(b)

7

2

8

4

6

9

8

10

10

11

12

12

14

13

16

18

14

45

45

44

44

2000 m 1500 m 1000 m

43

43

42

42

500 m 250 m

7

8

9

10

11

12

13

14

50 m

Figure 8.3. (a) Location of the Arno River Basin and (b) orography of the Arno River Basin and location of the rain gauges (see Color Plate XXVIII).

124

B. Gozzini et al.

current, and future variability and trends of heavy rainfall and drought occurring over the Arno River Basin, Italy (about 9200 km2; Figure 8.3(a) and (b)). The general purpose of the Project is to support the periodic updating of the distributed constraints system around the basin, design and management of flood protection works, assisting the local weather forecasting system, and the formulation of the water quality and conservation policies.

(a) 1800 1600

60

1400 50

1200

40

1000

30

800 600

20

Precipitation (mm)

Frequency of rainy days (%)

70

400 10

200 1999

1997

1995

1993

1991

1985

1987

1985

1983

1981

1979

1977

1975

1973

1971

1969

1967

1965

1963

1961

1959

1957

1955

1953

0 1951

0

Year Frequency of rainy days (> = 0.2 mm)

Precipitation

14 12

50

10 40 8 30 6 20

4

1999

1997

1995

1993

1991

1989

1987

1985

1983

1981

1979

1977

1975

1973

1971

1969

1967

1965

1963

1961

1959

0 1957

0 1955

2

1953

10

1951

Frequency of rainy days (%)

60

Average daily precipitation intensity (mm)

(b)

Year Frequency of rainy days (> = 0.2 mm)

Precipitation intensity

Figure 8.4. Annual precipitation time series over the Arno River Basin: (a) annual frequency of rainy days and total precipitation and (b) annual frequency of rainy days and average daily precipitation intensity over the lower portion of the Arno River Basin (see Color Plate XXIX).

Regional climatic variability 125 For the ongoing analysis authors use the historical series of in situ rainfall data, the GPCC (Global Precipitation Climatology Centre) (Xie and Arkin, 1996) and GPCP (Xie et al., 2003) gridded precipitation products, the NCEP– NCAR global atmospheric and surface reanalyses (Kistler et al., 2001), and the climate scenarios produced by the CCCma (Flato and Boer, 2001). The in situ data, covering a period of 50 years till the year 2000, has been analyzed to delineate the seasonal interannual variability trends in the frequency of rainy days, precipitation totals and average daily intensity, and frequency of local and basin-scale rainfall events above few thresholds. The local and basin-scale depth-duration-frequency curves have been drawn. The significance of a given trend has been evaluated using the Mann-Whitney test (Kendall and Ord, 1990). The same analyses concerning the historical in situ data were performed on the climate scenarios provided by the CCCma, on the (coarse) grid cell covering the Arno River Basin. Our study has led to the following conclusions: 1 The total annual precipitation has not changed significantly since 1950, but the frequency of rainy days has decreased until early 1980s (Figure 8.4(a)), whereas the year-round average daily rainfall intensity increased significantly in the last 30 years (Figure 8.4(b)). 2 In winter, the total precipitation and the frequency of rainy days have decreased significantly until mid-1990s (Figure 8.5). 3 In the case of the spring season, the total precipitation has increased as a result of the increasing average daily rainfall intensity, whereas in summer the total precipitation is decreasing on the upper mountainous portion of the basin. The average daily rainfall intensity is significantly increasing elsewhere, with a pronounced interannual variability. In the fall season, statistically robust increasing trends are evident for the total precipitation, the frequency of rainy days, and precipitation intensity in the last 30 years, superimposed to a relevant interannual variability.

600

80

500

60 400

50

300

40 30

200

Precipitation (mm)

Frequency of rainy days (%)

70

20 100

10 1999

1997

1995

1993

1991

1989

1987

1985

1983

1981

1979

1977

1975

1973

1971

1969

1967

1965

1963

1961

1959

1957

1955

1953

0 1951

0

Year Frequency of rainy days (> = 0.2 mm)

Precipitation

Figure 8.5. Winter season precipitation time series over the Arno River Basin: annual frequency of rainy days and total precipitation (see Color Plate XXX).

1860–89 1862–71 1864–73 1866–75 1868–77 1870–79 1872–81 1874–83 1876–85 1878–87 1880–89 1882–91 1884–93 1886–95 1888–97 1890–99 1892–1901 1894–1903 1896–1905 1898–1907 1900–09 1902–11 1904–13 1906–15 1908–17 1910–19 1912–21 1914–23 1916–25 1918–27 1920–29 1922–31 1924–33 1926–35 1928–37 1930–39 1932–41 1934–43 1936–45 1938–47 1940–49 1942–51 1944–53 1946–55 1948–57 1950–59 1952–61 1954–63 1956–65 1958–67 1960–69 1962–71 1964–73 1966–75 1968–77 1970–79 1972–81 1974–83 1976–85 1978–87 1980–89 1982–91 1984–93 1986–95 1988–97 1990–99 1992–2001

Frequency (%)

(c) 1993–2002

1991–2000

1989–98

1987–96

1985–94

1983–92

1981–90

1979–88

1977–86

1975–84

1973–82

1971–80

1969–78

1967–76

1965–74

1963–72

1961–70

1959–68

1957–66

1955–64

8 25

6 20

4 15

2 10

0

4.5

Frequency Frequency over the rainy days

3.5 8

3 7

2.5 6

2 5

1.5 4

1 3

0.5 1

0 0

Frequency over the rainy days (%)

10

6

Frequency Frequency over the rainy days

4 14

12

3

10

2

8

6

0

Frequency over the rainy days (%)

4 1953–62

1951–60

Frequency (%)

Frequency Frequency over the rainy days

Frequency over the rainy days (%)

(b)

1951–60 1952–61 1953–62 1954–63 1955– 64 1956–65 1957–66 1958–67 1959–68 1960–69 1961–70 1962–71 1963–72 1964–73 1965–74 1966–75 1967–76 1968–77 1969–78 1970–79 1971–80 1972–81 1973–82 1974–83 1975–84 1976–85 1977–86 1978–87 1979–88 1980–89 1981–90 1982–91 1983–92 1984–93 1985–94 1986–95 1987–96 1988–97 1989–98 1990–99 1991–2000

Frequency (%)

(a) 12 Frequency of daily precipitation >= 40 mm/day 35

30

5

0

Decade

Frequency of daily precipitation >= 75 mm/day 10

9

2

Decade

Frequency of daily precipitation >= 50 mm/day 20

5 18

16

1

4

2

0

Decade

Figure 8.6. Frequency of rainy days over high thresholds: (a) lower Arno River Basin, (b) rain gauge in the upper Arno River Basin, and (c) rain gauge in the medium Arno River Basin (see Color Plate XXXI).

Regional climatic variability 127 4 The daily precipitation events exceeding given thresholds, which could trigger local flash floods and floods are growing in frequency both locally and averaged over the sub-basins. Thus they are today more frequent than ever, at least over the 150 years. When evaluated during subsequent decades each sharing nine years with the previous, such frequency is found to have increased from 20% in the upper portion of the basin to about 150% in the lower portion (Figure 8.6). 5 The extreme annual precipitations have increased, but only at very short duration (1 h and 3 h, not shown). 6 The CCCma climate scenario outputs produced by CGCM2-A2 (Canadian Global Coupled Model-emission scenario A2) (Flato and Boer, 2001) were analyzed over the grid cell of size 3.75⬚ Long–Lat around the Arno River Basin (Figure 8.7). A very close agreement (not

(a)

–150 –120

–90

–60

–30

0

30

60

90

120

150 4000

60

60

30

30

0

0

3000 1500 1000

–30

–30

500

–60

–60

250 0

–150 –120

(b)

9.5

–90

10

–60

10.5

–30

0

11

30

11.5

60

12

90

12.5

120

150

13

44.5

44.5

44

44 4000

43.5

43.5 3000

43

43 1500

42.5

42.5

42

41.5

1000

42

500

41.5

250

41

41 9.5

10

10.5

11

11.5

12

12.5

13

Figure 8.7. (a) Global topography of the underlying surface in the CGCM2 climate scenario (CCCma) and (b) 1 km resolution topography and boundaries of the Arno River Basin in the grid cell of the CGCM2 climate scenario (see Color Plate XXXII).

128

7

8

9 10

11

B. Gozzini et al.

shown) has been observed when the simulated trends of surface air temperature in the recent past (1961–2000) were compared with the local observations. Also, the simulated rainfall trends in the last 40 years agree with observations, when the phase differences are taken into account (not shown). These CCCma climate scenarios suggest that beyond insignificant changes of the total precipitation and of the frequency of rainy days, a further increase of the frequency of excessive daily rainfalls is likely to occur until at least 2015. Beyond this period, it should remain constant, about 30% higher than in the current climate (Figure 8.8). The summer rainfall regime appears most sensitive to future climate change. The CGCM2A2 scenario suggests increasing summer daily rainfall intensities in the next few decades (Figure 8.9(a)), a sudden collapse of the summer rainfall, and the frequency of rainy days in about 30 years (Figure 8.9(b)). The winter precipitations should gradually increase (not shown), at least on an annual basis, so as to completely offset the future dry summers. A significant acceleration is predicted for the regional surface warming, in summer more than in winter (Figure 8.10). Hence the estimated deficit in summer-time precipitation, together with much higher temperatures, could result in frequent extreme droughts. As a result of the changes of the local precipitation regimes (less precipitation in winter, decreased frequency of rainy days) and the regional warming, the Arno River Basin has lost about 30% of its streamflow in the last 40 years on an annual basis (Figure 8.11), with most of the contribution from winter, followed by spring and summer; while the precipitation contribution to the loss of streamflow dominates, the warming contribution is significant, especially in spring. An interpretation of few seasonal precipitation trends predicted by the CGCM2-A2 scenario is offered in terms of changes of the simulated seasonal storm track strength. Figure 8.12 shows the simulated changes of the winter storm track strength in the period 2001–20 relative to 1981–2000. It may be noted that there is a remarkable increase along most of the path, from the North American coasts to western Mediterranean and eastern Europe, which

10 9 8

Frequency (%)

7 6 5 4 3 2 1

1961–70 1963–72 1965–74 1967–76 1969–78 1971–80 1973–82 1975–84 1977–86 1979–88 1981–90 1983–92 1985–94 1987–96 1989–98 1991–2000 1993–2002 1995–2004 1997–2006 1999–2008 2001–10 2003–12 2005–14 2007–16 2009–18 2011–20 2013–22 2015–24 2017–26 2019–28 2021–30 2023–32 2025–34 2027–36 2029–38 2031–40 2033–42 2035–44 2037–46 2039–48 2041–50 2043–52 2045–54 2047–56 2049–58 2051–60 2053–62 2055–64 2057–66 2059–68 2061–70 2063–72 2065–74 2067–76 2069–78 2071–80 2073–82 2075–84 2077–86 2079–88 2081–90 2083–92 2085–94 2087–96 2089–98 2091–2100

0

Decade

Figure 8.8. Annual time series of the frequency of rainy days over the threshold 20 mm around the Arno River Basin (1961–2100), simulated by means of the CGCM2-A2 scenario (CCCma) (see Color Plate XXXIII).

Regional climatic variability 129

(a) 4.5

Average daily precipitation intensity (mm)

4

3.5

3

2.5

2

1.5

1

0.5

1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025 2027 2029 2031 2033 2035 2037 2039 2041 2043 2045 2047 2049 2051 2053 2055 2057 2059 2061 2063 2065 2067 2069 2071 2073 2075 2077 2079 2081 2083 2085 2087 2089 2091 2093 2095 2097 2099

0

Year (b) 70

180 Frequency of rainy days (> = 0.2 mm) Precipitation

160

60 140

120 40

100

80

30

Precipitation (mm)

Frequency of rainy days (%)

50

60 20 40 10 20

0 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025 2027 2029 2031 2033 2035 2037 2039 2041 2043 2045 2047 2049 2051 2053 2055 2057 2059 2061 2063 2065 2067 2069 2071 2073 2075 2077 2079 2081 2083 2085 2087 2089 2091 2093 2095 2097 2099

0

Year

Figure 8.9. (a) Annual time series of the summer average rainfall intensity around the Arno River Basin (1961–2100) and (b) annual time series of the summer total precipitation and frequency of rainy days around the Arno River Basin (1961–2100), simulated by means of the CGCM2-A2 scenario (CCCma) (see Color Plate XXXIV).

is apparently linked to the predicted gradual increase of the winter precipitation over the Arno River Basin. 12 Figure 8.13 shows the simulated changes of the summer storm track strength in the period 2041–60 relative to 1981–2000. The strong weakening of the North Atlantic storm activity, extending over Western Europe, central Europe, and western Mediterranean, accounts, at least partly, for the abrupt and sudden decrease of the summer precipitation over the Arno River Basin.

Winter Spring Summer Autumn

26 24

Temperature (°C)

22 20 18 16 14 12

1900 1905 1910 1915 1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050 2055 2060 2065 2070 2075 2080 2085 2090 2095 2100

10

Year

Figure 8.10. Annual time series of the seasonal average surface air temperature around the Arno River Basin (1900–2100), simulated by means of the CGCM2-A2 scenario (CCCma) (see Color Plate XXXV). (a)

35

Streamflow (m3/s)

30 25 20 15 10 5 1970

1974

1978

1982

1986

1990

1994

1998

2002

1970

1974

1978

1982

1986

1990

1994

1998

2002

1966

1962

1958

1954

1950

1946

1942

1938

1934

1930

0

Year (b)

70

Streamflow (m3/s)

60 50 40 30 20 10 1966

1962

1958

1954

1950

1946

1942

1938

1934

1930

0

Year

Figure 8.11. Arno River discharge at the Subbiano section, during 1930–2003 (units m3 s⫺1): (a) daily values averaged over the year, (b) winter, (c) spring, (d) summer, and (e) fall. The streamflow scale is logarithmic.

Figure 8.11. Continued.

Year 1990 1994 1998 2002

1990

1994

1998

2002

1966

1962

1958

1954

1950

1946

1942

1938

1986

0

1986

10 1982

20

1982

30 1978

40

1978

50 1974

60

1974

70 1970

Year

1970

1966

1962

1958

1954

1950

1946

1942

1938

(e) 1934

1930

Streamflow (m3/s)

(d)

1934

1930

Streamflow (m3/s)

2002

1998

1994

1990

1986

1982

1978

1974

1970

1966

1962

1958

1954

1950

1946

1942

1938

1934

1930

Streamflow (m3/s)

(c) 40

35

30

25

20

15

10

5

0

Year

10

9

8

7 6

5

4

3

2 1 0

132

B. Gozzini et al.

70°N 2

2

5

2

5

65°N

5

2

10

60°N

15

15

15

10

10

20

5 55°N

25

20

20 50°N

10

10

45°N

20

35°N

15

10

2

15

5 –2 –5

5

40°N

20

25

2

10

2

–2

5 70°W

5

–10

5 60°W

50°W

40°W

30°W

20°W

10°W

0

10°E

20°E

30°E

Figure 8.12. Changes of the winter storm track strength (2001–20 minus 1981–2000), simulated by means of the CGCM2-A2 scenario (CCCma).

70°N

–2

65°N –5 60°N

–2

–10 –5

–2 –5

–2

55°N

2

–5

–10 –15

50°N –10

–5

45°N

–10

–10

–5

–2

–5

–10

40°N

–5 –2

2 –5

35°N 70°W

60°W

50°W

40°W

30°W

20°W

10°W

0

10°E

20°E

30°E

Figure 8.13. Changes of the summer storm track strength (2041–60 minus 1981–2000), simulated by means of the CGCM2-A2 scenario (CCCma).

13 Both flood and drought hazards are predicted to undergo significant increases over the Arno River Basin, in the course of next years or decades. While the flood hazard is already significantly increasing, and is projected to increase rapidly in the next several years or few decades, the drought risk is just starting to show up, and may lead to a potentially dramatic challenge for water quality and availability in the coming decades. 14 From a synoptic point of view, the fall season is going to behave similar to current and past summer seasons (fewer organized storms, occasional local and regional intense rainstorms). The winter season will look like the current fall season in terms of peak storms frequency. The summers will become generally drier, punctuated by occasional very heavy and short duration rainstorms, eventually becoming very dry and hot. 15 The climate of late spring and summer 2003 has been a very large outlier with regard to precipitation (in the lower 5th percentile of the last 150 years) and temperature (May–August anomaly about 3.5⬚C higher than the 1961–90 climatological average). They

Regional climatic variability 133 are much above any past record and largely exceed the interannual variability simulated by the CCCma climate scenario. ACKNOWLEDGMENTS The authors gratefully acknowledge the assistance of the Project “Improvement of the management of water resources and crop yield in hazardous areas with regard to drought and desertification” of the Italian Presidency of the Council of Ministries and the Project “Climate analysis and prediction over the Arno River Basin” of the Arno River Basin Authority. Discussions with Prof. Giovanni Menduni, Dr Marcello Brugioni, and Dr Bernardo Mazzanti of the Arno River Basin Authority (Firenze, Italy) have been very useful. The assistance of Dr Francesca Marrese of the Applied Meteorology Foundation (Firenze, Italy), who has processed most of the data over the Arno River Basin, and Dr Luca Fibbi of the Laboratory for Meteorology and Environmental Modelling (Firenze, Italy), who provided the rain gauge data over the Arno River Basin, is gratefully acknowledged. M. Baldi acknowledges the National Science Foundation under Grant No. ATM-9910857. REFERENCES Baldi, M. et al. (2003a) Mediterranean summer climate and its relationship to regional and global processes. In: Proceedings of the Sixth European Conference on Applications of Meteorology, Rome, 15–19 September. Baldi, M. et al. (2003b) Numerical analysis of the teleconnection of the West Africa monsoon with the Mediterranean summer climate. Environmental Fluid Mechanics, submitted. Brunetti, M., Colacino, M., Maugeri, M., and Nanni, T. (2001) Trends in the daily intensity of precipitation in Italy from 1951 to 1996. International Journal of Climatology, 21, 299–316. Brunetti, M., Maugeri, M., Nanni, T., and Navarra, A. (2002) Droughts and extreme events in regional daily Italian precipitation series. International Journal of Climatology, 22, 543–558. Chang, E.K. and Fu, Y. (2002) Interdecadal variations in northern hemisphere winter storm track intensity. Journal of Climate, 15, 642–658. Chen, J., Carlson, B.E., and Del Genio, A.D. (2002) Evidence for strengthening of the tropical general circulation in the 1990s. Science, 295, 838–841. Christensen, J.H. and Christensen, O.B. (2003) Severe summertime flooding in Europe. Nature, 421, 805–806. Dalu, G.A., Gaetani, M., Meneguzzo, F., Crisci, A., Guarnieri, F., and Capecchi, V. (2005) The hydrological onset and withdrawal index (HOWI) for the West Africa Monsoon. Accepted for presentation at 85th AMS Conference, San Diego, CA, January 2005. Extended abstract available at www.ametsoc.org, paper number: P5.4. Del Genio, A.D. (2002) The dust settles on water vapor feedbacks. Science, 296, 665–666. Dima, I.M. and Wallace, J.M. (2003) On the seasonality of the Hadley Cell. Journal of Atmospheric Science, 60, 1522–1527. Douville, H. (2001) Influence of soil moisture on the Asian and African monsoons. Part II: interannual variability. Journal of Climate, 15, 701–720. Flato, G.M. and Boer, G.J. (2001) Warming asymmetry in climate change simulations. Geophysics Research Letters, 28, 195–198. Gillet, N.P., Zwiers, F.W., Weaver, A.J., and Stott, P.A. (2003) Detection of human influence on sea-level pressure. Nature, 422, 292–294. Groisman, P.Ya. et al. (1999) Changes in the probability of heavy precipitation: important indicators of climatic change. Climate Change, 42, 243–283. Gulev, S.K., Jung, T., and Ruprecht, E. (2002) Climatology and interannual variability in the intensity of synoptic-scale processes in the North Atlantic from the NCEP–NCAR reanalysis data. Journal of Climate, 15, 809–828.

134

B. Gozzini et al.

Harnik, N. and Chang, E.K. (2003) Storm track variations as seen in radiosonde observations and reanalysis data. Journal of Climate, 16, 480–495. Hegerl, G.C., Zwiers, F.W., Stott, P.A., and Kharin, V.V. (2003) Detectability of anthropogenic changes in temperature and precipitation extremes. Journal of Climate, submitted. Hoerling, M.P., Hurrell, J.W., and Xu, T. (2001) Tropical origins for recent North Atlantic climate change. Science, 292, 90–92. Houghton, J.T. et al. (eds) (2001) Summary for policymakers. A Report of Working Group I of the Intergovernmental Panel on Climate Change, 20pp. Available at http://www.ipcc.ch Hurrell, J.W. (1995) Decadal trends in the North Atlantic Oscillation: regional temperatures and precipitation. Science, 269, 676–679. Hurrell, J.W. and Folland, C.K. (2002) A change in the summer atmospheric circulation over the North Atlantic. CLIVAR Exchanges, 25, 52–54. Available at http://www.clivar.ucar.edu/publications/ exchanges/ex.25/pdf Janowiak, J.E. (1988) An investigation of interannual rainfall variability in Africa. Journal of Climate, 1, 240–255. Jones, C.D. et al. (2003) Strong carbon cycle feedbacks in a climate model with interactive CO2 and sulphate aerosols. Geophysics Research Letters, 30, 1479–1482. Kendall, M. and Ord, J.K. (1990) Time Series, 3rd edn, London: I Edward Arnold. Kistler, R. et al. (2001) The NCEP–NCAR 50-year reanalysis: monthly means CD-ROM and documentation. Bulletin of the American Meteorological Society, 82, 247–268. Maynard, K., Royer, J.-F., and Chauvin, F. (2002) Impact of greenhouse warming on the West Africa summer monsoon. Climate Dynamics, 19, 499–514. Meneguzzo, F. et al. (2003) Sensitivity of meteorological high-resolution numerical simulations of the biggest floods occurred over the Arno river basin, Italy, in the 20th century. Journal of Hydrology (in press). Palmer, T.N. and Ralsanen, J. (2002) Quantifying the risk of extreme seasonal precipitation events in a changing climate. Nature, 415, 512–514. Pielke, R.A. (2003) Heat storage within the climate system. Bulletin of the American Meteorological Society, 84, 331–335. Rial, J.A. et al. (2003) Nonlinearities, feedbacks, and critical thresholds within the earth’s climate system. Climatic Change (in press). Rowell, D.P. (2003) The impact of Mediterranean SSTs on the Sahelian rainfall season. Journal of Climate, 16, 849–862. Santer, B.D. et al. (2003) Contributions of anthropogenic and natural forcing to recent tropopause height changes. Science, 301, 479–483. Sultan, B. and Janicot, S. (2003) The West African monsoon dynamics. Part II. The pre-onset and the onset of the summer monsoon. Journal of Climate (in press). Vizy, E.K. and Cook, K.H. (2002) Development and application of a mesoscale climate model for the tropics: influence of sea surface temperature anomalies on the West African monsoon. Journal of Geophysics Research Atmosphere, 107(D3). Ward, M.N. (1998) Diagnosis and short-lead time prediction of summer rainfall in tropical North Africa at interannual and multidecadal timescales. Journal of Climate, 14, 795–821. Xie, P. and Arkin, P.A. (1996) Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions. Journal of Climate, 9, 840–858. Xie, P. et al. (2003) GPCP pentad precipitation analyses: an experimental dataset based on gauge observations and satellite estimates. Journal of Climate, 16, 2197–2214. Yang, F., Kumar, A., Schlesinger, M.E., and Wang, W. (2003) Intensity of hydrological cycles in warmer climates. Journal of Climate, 16, 2419–2423.

CHAPTER 9

Climate Drivers, Streamflow Forecasting, and Flood Risk Management Gonzalo Pizarro and Upmanu Lall Department of Earth and Environmental Engineering, Columbia University, 842 Mudd Building, 5400 W 120th Street, New York, NY 10027, USA

9.1

INTRODUCTION

Floods are the most significant natural hazard. In the decade of 1988–97, floods accounted for over half of the 390,000 recorded fatalities and a third of the damages from all the natural catastrophes worldwide (Kunreuther and Linnerooth-Bayer, 2003). There has been an increasing trend in flood related damages in the United States in the 1932–97 periods (Pielke and Downton, 2000). Worldwide, the economic losses due to weather related events have had a dramatic increase in losses post 1980, as seen in Figure 9.1 (Berz, 1999; Munich Re, 2001), which is suggestive of a change in the behavior of the system rather than a chance occurrence. While looking at this data, an immediate question that comes to mind is: what is the source of this nonstationarity. Is this a passing phase owing to a change in the behavior of the hydrologic system, is it a permanent change in the hydrologic system, is it a change in the valuation method of the losses, or are these changes in the conditions of the affected basins? This question is difficult to answer since we typically make inferences from less than a 100 years of data 40,000 35,000 30,000 25,000 20,000 15,000 10,000 5000 0 1950

1955

1960

1965

1970

1975 Year

1980

1985

1990

1995

2000

Figure 9.1. Economic losses due to major weather-related natural disasters in the 1950–2000 period, 2001 prices, USD millions (see Color Plate XXXVI). (Source: Berz, 1999; Munich Re, 2001.)

136

Gonzalo Pizarro and Upmanu Lall

about an event that may happen once in a 100 years, and it has been observed that the natural climate has temporal structure that admits clustering of extremes at a variety of time scales. The impact of changes in the urbanized surface of a basin has long been known to alter the response of the basin to extreme events (Chow et al., 1994). But this is not the only factor. In order to thoroughly analyze the values presented in Figure 9.1, several factors must be considered, namely: 1 2 3 4 5 6 7

environmental change; climate variability and change; economic growth; urban growth; land use and infrastructure change; channel modification; warning response.

Thus, socioeconomic factors such as population growth, economic development, and land use policies also have a direct impact on the existence and magnitude of a damaging flood, and to do proper flood management, all such factors must be considered. These factors do not follow steady trends, and some of them are very hard to predict. As such, designs and flood risk management measures being carried out today, using a set of assumptions and design constraints based on today’s reality, and expectations on the evolution of such may not be valid before the end of the useful life of the project. While these factors are important, they are hard to address. Traditional approaches to flood risk management commonly involving the construction of dams and river protection are being socially resisted, and in the existing communities land use changes or moving people out of the floodplain are not realistic options. The increasing development of the floodplain, due to population increase and economic growth, is also a trend that is politically difficult to address. Thus, understanding the sources of variability of the hydrological cycle and using that knowledge to better manage flood risk is the first approach to tackle the problem. The recognition that increased weather volatility and catastrophic losses are difficult to address using traditional insurance practices (Chichilnisky and Heal, 1998), and the growth of societal impacts related to climate, have necessitated the search for innovative solutions to manage this risk. Flood risk management has traditionally been done via structural measures, such as dams, bypasses and levees, and post event mitigation. The design and operation of such structures is subject to the assumption that the frequency and magnitude of flooding events are independent and stationary. This assumption allows the process to be modeled by an appropriate probability distribution using a large enough data set. Under these assumptions, the risk of a flood being larger than a predetermined threshold (i.e. the height of the levee) is the same in any given year regardless of the climatic conditions or the extent of the designed life of the project. This is referred to as the static risk paradigm. The understanding of the connection between ocean-atmospheric climatic states and flood risk and the identification of regional responses to such states has helped to identify organized systems which respond to ocean states that have recognizable cycles, such as the El Niño Southern Oscillation (ENSO), or the Pacific Decadal Oscillation (PDO) phenomena. In other words, the risk of a flood exceeding a predetermined threshold will depend on the state of the climate, and can be estimated and forecasted. This is what is referred to as a dynamic risk paradigm. The recognition of this reality has allowed a shift to adaptive measures and the exploration of mixed public–private partnerships to manage flood risk such as financial options that were traditionally not available. We will explore how to include climate information in the design opted for in an operation of flood risk management and how the increased knowledge on the functioning of the ocean-atmosphere system is allowing the exploration of new avenues of flood risk management, particularly in the insurance world.

Climate drivers and flood risk management 137 9.2 9.2.1

FLOOD FREQUENCY AND CLIMATE VARIABILITY Introduction

Only recently has it been understood that the structured interannual, interdecadal, and longer time planetary climate imparts temporal structure to the flood frequency analysis (Jain and Lall, 2000). Possibly, the changes in atmospheric composition have steadily changed the climate forcing and thus perhaps also the hydroclimatic response in recent decades. Estimated flood exceedance probabilities can increase quite rapidly with time even in the presence of rather mild rising climatic trends (Porporato and Ridolfi, 1998). There is compelling evidence that, at least in some regions, “normal” climatic variability may yield significant non-randomness in flood series (Hidalgo and Dracup, 2003). Under the reality of climate change and variability, decision making under uncertainty must be formally addressed in any economic discussion on climate change policy (Heal and Kristrom, 2002). Such uncertainties are present not only in the natural sciences but also in the economic system – examples being discount rates, growth rates, and rate of technological advance (Heal and Kristrom, 2002). Thus, uncertainties on both sides of the problem must be considered for better decisions to be made. Since the discovery of regime-like or quasi-periodic behavior of climate and systematic trends such as the Quasi-Biennial Oscillation, ENSO, Quasi-Decadal Oscillations, and the century scale purported global warming trend, increasing interest has developed in linking these climatic states to local and/or regional variability of the climate. Evidence of these features has been reported in the western United States (Cayan et al., 1999). Connections between ENSO and floods in the southwest have also been indicated (Gutzler et al., 2002). However, to date only limited work has been done to systematically relate low frequency climate signals to flood potential in the region. 9.2.2

Sources of variability: teleconnections of SSTs with flood variability

9.2.2.1 El Niño Southern Oscillation ENSO is a phenomenon characterized by cyclical departures of sea surface temperatures (SSTs) in the tropical Pacific from their normal state (Figure 9.2). This cycle has a periodicity of 3–7 years and is coupled with anomalous sea level pressures in the Eastern Pacific. ENSO was not understood until recently (Cane and Zebiak, 1985) but it got its name from Peruvian and Ecuadorian fishermen more than a century ago, who observed a warming of the ocean water during Christmas time every few years (hence the term El Niño, which is Spanish for child Christ). The entire ENSO consists of three phases: ●

●

●

Normal conditions (Figure 9.2). The Pacific Ocean is warmer in the western shore, where the highest convection is also happening, and therefore also the greatest amount of rainfall. The sea level pressure has a big dipole with a high pressure center in front of the North American continent, and a low pressure center in the western tropical Pacific. This feature effectively blocks frontal systems and extratropical cyclones from reaching the North American continent. As a consequence, the easterlies are formed, that is, the winds near the surface travel from east to west across the Pacific. Warm phase or El Niño (Figure 9.3(a) and (c)). The central and eastern tropical Pacific is warmer than normal years, debilitating the easterlies wind, extending the convection center, and weakening the high pressure center in front of the North American continent. Consequently, enhanced precipitations are common during El Niño episodes in the western United States. Cool phase or La Niña (Figure 9.3(b) and (d)). The central and eastern tropical Pacific is cooler than normal years, effectively strengthening the easterlies. The high pressure center in front of the North American continent is also strengthened, which results in drier than usual conditions for the western United States.

138 Gonzalo Pizarro and Upmanu Lall

NCEP/NCAR reanalysis (a)

90N 80N 70N 60N SST

50N 40N 30N 20N 10N EQ 10S 20S 60E

90E

120E

150E

180

150W

120W

90W

60W

30W

0

March–May: 1945–2004

–20

–15

–10

–5

–0

5

10

15

20

25

30

NCEP/NCAR reanalysis (b)

90N 80N 70N 60N SLP

50N 40N 30N 20N 10N EQ 10S 20S 60E

90E

120E

150E

180

150W

120W

90W

60W

30W

0

March–May: 1948–2004

1005

1008

1010

1012

1014

1016

1018

1020

1022

Figure 9.2. Normal state of (a) sea surface temperature (SST) and (b) sea level pressure (SLP). Normal conditions determined by taking the long-term average over the 1948–2004 period, March through May composite (see Color Plate XXXVII).

Once developed, El Niño and La Niña events typically persist for about a year, and so the shifted rainfall patterns associated with them typically persist for several seasons as well. El Niño events recur on average every 3–7 years. However, there are periods of decades over which the recurrence frequency of El Niño events can be substantially different. As is illustrated in Figure 9.4, the ENSO phenomenon does not only affect the tropical Pacific, but has a global impact. 9.2.2.2 Pacific Decadal Oscillation The “Pacific Decadal Oscillation” (PDO, Mantua et al., 1997) is a 20–30 year long ENSO like phenomenon that occurs in the Northern Pacific. The climatic fingerprint of the PDO is most

Climate drivers and flood risk management 139

SSTa

SLPa

(a)

(c)

(b)

(d)

Figure 9.3. Typical extent of (a) the warming, (b) cooling in the equatorial Pacific, (c) weakened Pacific high-pressure center, and (d) enhanced Pacific high-pressure center during developed El Niño (a and c) and La Niña (b and d) events. March through May composites done with 10 greatest Niño and Niña years in the 1950–95 period (see Color Plate XXXVIII).

visible in the North Pacific/North American sector, whereas secondary signatures exist in the tropics. The cool phase is characterized by a wedge of lower than normal SSTs in the eastern equatorial Pacific and a warm horseshoe pattern of higher than normal SSTs connecting the north, west, and southern Pacific. In the warm or “positive” phase, the west Pacific Ocean becomes cool and the wedge in the east warms (Figure 9.5). There is evidence of two full PDO cycles in the twentieth century, with cool PDO phases in the 1890–1924 period and in the 1947–76 period. Warm PDO was present during the 1925–46 period and during the 1977–99 period. The twentieth century PDO fluctuations were most energetic in two general periodicities, one from 15 to 25 years, and the other from 50 to 70 years (Minobe, 1999). 9.2.3

Spatial structure of flood occurrence in the western United States

While much of the work on relating climatic indices to hydroclimatic impacts has focused on the use of seasonal attributes, it is clear that the nature of intra-seasonal variability of storms and their modulation by the seasonal or longer term climate state is of interest for understanding the causative factors of floods. The weather phenomena that cause floods include intense convective thunderstorms, tropical storms and hurricanes, cyclones and frontal passages, and rapid snowmelt (Hirschboeck, 1991). These processes are part of a climatic framework that determines the large-scale delivery pathways of atmospheric moisture, their seasonal variations, typical locations, degree of persistence, and seasonal variation of climate-related land-surface conditions such as antecedent soil moisture or snow cover.

140 Gonzalo Pizarro and Upmanu Lall

(a)

(b)

Figure 9.4. Typical climatic anomalies attributable to fully developed El Niño or La Niña events. (a) Warm episode relationships (June–August) and (b) cold episode relationships (June–August) (see Color Plate XXXIX). (Source: International Research Institute for Climate Prediction (IRI) website, http://iri.columbia.edu/climate/ENSO/globalimpact/ temp_precip/region_elnino.html)

For example, correlations for the annual maximum flood series with ENSO and PDO are statistically significant for the entire western United States region, despite differences in mechanism, drainage area, and season of occurrence, as seen in Figure 9.6 (Pizarro and Lall, 2002). Past analyses (Cayan et al., 1999) of extreme precipitation have identified some of these features but not so clearly over the entire region and not over the entire year. This observation suggests that ocean-atmosphere land-hydrologic processes associated with the annual maximum may have considerable predictability conditional on climate state. The sources of moisture for the western United States are the Pacific Ocean and the Gulf of Mexico (Hirschboeck, 1991). The moisture delivery pathways shift seasonally, effectively controlling the timing of the flood season in the subregions of this area (Hirschboeck, 1991). The combined effect of the NINO3 and PDO conditions is to modulate the location and strength of westerly flows and the eddies coupled to it in such a way that it impacts the potential for flooding. Extratropical, cyclonic storms and fronts are brought to the US west coast by strong westerly winds. The extratropical cyclone tracks have a seasonal shift. They reach their southernmost

(a)

(b)

Figure 9.5. Sea surface temperature anomalies during the (a) positive and (b) negative phase of the Pacific Decadal Oscillation. March through May composites of the Reynolds reconstructed SSTa, CDC/NOAA (see Color Plate XL).

142 Gonzalo Pizarro and Upmanu Lall

(a)

(b)

Correlation >0.37 Correlation >0.23

Correlation >0.37 Correlation >0.23

Figure 9.6. Relationship of the highest annual daily flow at 137 locations for streams in the western United States to two slowly varying modes of global climate. The daily peak flood is seen to relate well to the seasonal average of two climate indices (NINO3, which is a measure of the ENSO that has an average recurrence interval of 4–7 years, and PDO, which is a measure of the Pacific Decadal Oscillation, that has an average recurrence interval of 16–22 years). (a) Illustrates the partial correlation of the annual peak flood with the January–April average of the NINO3 given that the effect of PDO is taken out, and (b) Illustrates the partial correlation of the annual peak flood with the January–April average of the PDO given that the effect of NINO3 is taken out. The two sizes of circles correspond to the 95% and 99% confidence levels, with the magnitude of the correlation indicates in the legend. The solid circles represent negative correlations and the open circles represent positive correlation. Thus, knowing both the NINO3 and the PDO in this season would allow us to say something about the annual flood. The flood season varies by location, being January–March in the Pacific northwest (upper left corner of the picture), and August–October for the southwest (bottom right corner). Thus, for the concurrent season one can understand the long-term variability and for the leading season we can issue a forecast. Similar analyses can be developed using climate data from other seasons.

position, which make storms hit California in the winter, and their northernmost position, which makes storms hit Canada in the summer. During ENSO periods there is an active subtropical jet stream, which allows extratropical cyclonic storms to reach the southwest (Hirschboeck, 1991). On the northwest, extratropical cyclonic storms deliver precipitation in the form of rain and snow, which will contribute to the peak flows due to melting during spring. Since the primary source of atmospheric moisture are the oceans, several studies have investigated links between ocean structured phenomena and precipitations and floods. The relationship between ENSO signal and precipitation, snow accumulation and streamflow in western North America has been well established and anomalies of SST are also involved in the decadal fluctuations of precipitation events (Barlow et al., 2001). Statistically significant relationships between wet conditions in California and dry conditions in the Pacific northwest and the warm phase of ENSO, and vice versa, have also been found. Also, rainfall over California has been found to have a significant correlation with ENSO for 2–3 months composites (Livezey et al., 1997). It has also been shown that ENSO modifies the duration, frequency, and magnitude of the precipitation events. In the southwest a warm phase ENSO event produces longer duration precipitation events and a cool phase ENSO event produces shorter duration cases. Conversely, in the northwest, the opposite behavior is observed. Taken together these changes in the duration of heavy precipitation help to explain the amplification of the ENSO streamflow signal over

Climate drivers and flood risk management 143 its precipitation signal, as well as its pattern over the west (Cayan et al., 1999). In the case of the peak flows, a difference is observed regarding the southwest, where the influence of summer cyclonic events, enhanced by ENSO events dominates the signal. The influence of a positive PDO has been shown to be mainly in the form of a wet climate in southwestern states of the United States, mixed results in southern California, and dry years over the rest of the region (Minobe and Mantua, 1999; Pizarro and Lall, 2002). 9.2.4

Predictability of climate

9.2.4.1 ENSO/PDO seasonal prediction Given the teleconnection of equatorial SSTs and rainfall, the persistence of ENSO type tropical SST patterns allows making seasonal (three-month) climate forecasts possible. To predict the likelihood of an El Niño or La Niña, models that predict SST states in the tropical Pacific are needed. There are two main types of models for predicting the state of the SSTs, namely, the dynamical and statistical models. Causes for the PDO are not currently known, and therefore it is not robustly predictable. Even so, because of its strong tendency for multi-season and multi-year persistence, PDO climate information improves season-to-season and year-to-year climate forecasts for North America. 9.2.4.1.1 Dynamical models The dynamical models are based on a series of mathematical expressions that represent the governing physical laws of the ocean-atmosphere system. The forecasting process for this type of model consists of identifying the current conditions of the ocean and atmosphere, and using the mathematical expression of the system contained in the model to determine future conditions. Such models are used to make forecasts up to six month in advance. 9.2.4.1.2 Statistical models Statistical models use past observations to predict the future. Statistical models try to identify key features of the ocean-atmosphere interaction that preceded changes in tropical SSTs that lead to El Niño or La Niña states. To do so, a large set of observations are needed, at least 30 years. Some ENSO drivers that have been identified so far are variations in the total heat content in the western Pacific, variations of ocean temperatures at certain depths in the western Pacific, and variations in the strength of easterly trade winds in the central equatorial Pacific. These models are calibrated using past observations of the state of such drivers and the consequent state of the SSTs in the tropical Pacific, so that when they are given the current observations of the drivers, they can predict the likelihood of various possible ENSO conditions for the upcoming seasons. Statistical models can range from linear regressions to more complex nonlinear models such as nonlinear canonical correlation analysis or neural networks. 9.2.4.2 Anthropogenic climate change and floods The global climate is a dynamic system driven by the ocean-atmosphere-biosphere-lithosphere interaction, the cycles of intensity of solar radiation and the variations in the earth’s orbit and the changes in its axis of rotation. As a matter of fact, the natural climate system seems to oscillate between glacial and warm eras, but as seen in Figure 9.7, the current trend is beyond any previous observed values (IPCC, 2001). There is plenty of evidence suggesting an anthropogenic alteration of the normal cycle of the earth’s climate. Emissions of greenhouse gases and changes in the vegetation cover since the onset of the industrial revolution have nearly doubled the CO2 concentration in the atmosphere, and as a result, the average global temperature is increasing to levels that go beyond what would be expected from natural variation (IPCC, 2001). One of the expected consequences of a global warming trend is the emergence of more intense and frequent clusters of floods (Milly et al., 2002). It has been shown that under the presence of a mild warming trend, the risk of flooding in the mid-latitudes gets severely enhanced

Gonzalo Pizarro and Upmanu Lall

340

340

300

300

93 /1

5/

86 5/ /1

03

79 03

/1

5/

72

58

150

5/

100 Age (kya)

/1

50

03

0

5/

180

03

180

/1

220

03

220

65

260

5/

260

/1

CO2 max (ppmv)

(b) 380

CO2 max (ppmv)

(a) 380

03

144

Date

Figure 9.7. CO2 concentrations at different epochs. (a) CO2 concentration in the Vostok Antarctic ice core. Age in 1000-years BP; and (b) present-day concentrations as measured at the Mauna Loa observatory in Hawaii. (Source: Data extracted from IRI data library.)

(Porporato and Ridolfi, 1998). Such a change would make water resource management more difficult, generating situations of scarcity even in the presence of an increase in the total precipitation for some areas (IPCC, 2001). To forecast the consequences of anthropogenic climate change, statistical models cannot be used, since there are no past observations that will be repeated in the future. To predict, there has to be an understanding of how the system works, and what the governing physical laws are. This means that dynamical models must be used. To model global climate change, these models must include representations of the interactions of the ocean-atmosphere-biosphere-lithosphere system for the entire planet, and represent complex phenomena, such as cloud formation at different altitudes, and the interaction of incoming and outgoing radiation with aerosol particles in the atmosphere, the thermohaline circulation in the ocean, etc. That is, they must include representations and interactions of phenomena at very different scales. These models are known as General Circulation Models (GCMs). 9.2.4.2.1 GCM predictions and concerns The outcomes of the GCMs have a wide uncertainty due to several facts. On one hand the physics of many of the processes such as the formation of clouds, or the feedback of aerosols in the atmosphere is not well understood at the scales of interest. Others are not modeled properly or are not included at all such as the dynamics of the biosphere, or a consequence of problems of scale and dynamics at the cloud level. Even though such issues are being addressed and some of them are solved as knowledge and computing power increases, there are some uncertainties that

Climate drivers and flood risk management 145 are intrinsic to a long term prediction. There are uncertainties such as the rate of technological innovation, population growth, etc. which have an impact on the behavior of the population and its impact on the environment so that, even though one can simulate some of these changes by building appropriate scenarios, the uncertainty in the likelihood of such an outcome, even if the science in the model was perfect, is not avoidable, and would not be easy to estimate. 9.2.5

Estimating dynamic risk

Using both dynamical and statistical models, seasonal predictions (3–6 month lead) of flood risk can be made in certain places, where the teleconnection of floods with climate drivers is significant. Such estimates can be made at particular stations, where an estimate of a peak flow associated to a risk level can be obtained. They can also be obtained as a regional likelihood of higher/lower than normal risk level. These risk level estimates are achieved by identifying coherent regions of response of the maximum annual, or n-day, flow to appropriately identified climate drivers. In the case of the western United States, such coherent regions have been identified (Pizarro and Lall, 2002), allowing the opportunity of regional risk estimates, and also, the work on individual stations. The climate of the western United States is sensitive to both ENSO and PDO, and work done in the case of floods suggests that such interaction is nonlinear. Suitable models need to be developed to estimate risk in a dynamic way. The first step is to identify the appropriate set of drivers of flood risk variation. To do so, the flood record must be investigated and its connections to low frequency climate variability must be established. This provides the context to understand the flood risk variations, such as the ones seen for Sacramento in the American River Basin, and the need to alter the traditional flood risk management decision-making process by including climate information (NRC, 1999). Once the drivers have been identified, the predictability of the system must be analyzed, in order to facilitate the decision-making process. If the system is not predictable, and the adaptive measures are not feasible, the static risk framework must be used. Even so, studying the climate variability might allow better estimates of the uncertainty of the design parameters. 9.3 9.3.1

FLOOD RISK MANAGEMENT Introduction

There are many alternatives for the management of flood risk, whose main purpose is to minimize disaster losses in hazard-prone areas. These management alternatives can be broadly classified as: 1 Structural. This refers to actions taken that lead to reduction in the exposure to floods through physical changes in the basin. Examples are levees, bypasses, dams, and changes in building codes. Such actions are taken before or in advance of the event (ex ante), and are, therefore, proactive measures. Incorporation of climate information in the design process gives an opportunity for optimizing design through the reduction in the bias and uncertainty of the risk estimate. 2 Nonstructural. This refers to actions taken to reduce the losses via means that do not physically alter the basin. Such measures usually have a time frame smaller than the climatic regime residence time. Examples are insurance schemes, water release from dam contracts, and changes in land-use policies. Many nonstructural options are either of limited application or are not available for flood risks. Climate information in the form of both its temporal and spatial structures opens the door to the opportunity of implementing financial options for flood risk management. These options, in the form of catastrophe bonds (Cat Bonds), make flood insurance more attractive for the industry via reducing their exposure to catastrophic losses.

146

Gonzalo Pizarro and Upmanu Lall

A comprehensive flood risk management program is usually a combination of structural, nonstructural, and reactive actions. Reactive actions are those taken once the event has occurred, such as evacuation and relief. The extent to which the program is developed will depend on the societal exposure to natural disasters, economic development, and political willingness. 9.3.2

Traditional flood risk management methods

The traditional approach has been to use structural solutions to prevent the occurrence of highprobability low-volume floods (a 1/100 probability is required by the National Flood Insurance Program, administered by the US Federal Emergency Management Agency (FEMA) to qualify for insurability), and then use insurance for the high volume-low-probability ones, in addition to flood relief programs for post event management. Other options that have been commonly used involve restricting some land uses, such as allowing the high flood probability areas to be used as parking lots or public parks only. Also, public–private partnerships have been proposed (Kleindorfer and Kunreuther, 1999), where flood insurance is combined with enforcement and certification of flood-proofing guidelines in building codes, premium reductions are linked with long-term loans directed at risk-mitigating measures, or the offering of lower premiums is tied to the adoption of risk-mitigating measures by the policyholder. 9.3.2.1 Management of the financial risk of disaster losses The management of the financial risk of any kind of disaster loss is performed through three general mechanisms (Stipple, 1998): 1 governmental benefit programs; 2 tort liability; 3 private insurance. The tort system plays a less important role in natural disasters, since no individual or institution can be held liable for earthquakes or changing frequency and intensity of extreme weather events. However, if industrialized nations are allocated responsibility for altering the climate through the release of greenhouse gases, or in the case of “predictable” natural catastrophes, liability for not taking effective mitigative measures may become a relevant factor. To choose the best options for spreading the financial burden of disaster losses, they must be analyzed in terms of their efficiency and equitability. These are the dominant points of discussion on the policy arena while determining mechanisms for catastrophic risk management (Freeman and Kunreuther, 1997). There are several mechanisms for spreading the financial costs. The most common are Governmental Relief Programs, since, as it will be shown in the next section, insurance premiums have been difficult to estimate in advance due to high spatial concentration and correlation, and uncertainty in projecting the expected losses in future. In the realm of flood risk management projects are generally evaluated on two criteria, namely, the maximization of net benefits, and/or its cost-effectiveness. Here, net benefits are damages not incurred due to the actions taken, relative to the base situation minus the cost of taking such actions. The cost-effectiveness criterion refers to the minimization of aggregate costs to achieve some risk management objective. Governmental flood relief programs are generally inefficient because they rely on ex post remedies. This type of approach gives little incentive to preventive measures that could reduce costs to society should a flood happen. A government can increase incentives for mitigative measures by adopting a policy where no disaster assistance will be provided to uninsured victims of natural disasters. Issues of equity are raised with such programs, since usually the poorest members of society tend to live on high risk areas, and they are the ones with less access to private insurance. Nevertheless, private insurance can serve as a mechanism for increasing

Climate drivers and flood risk management 147 mitigation measures only if premiums are linked to mitigation actions on the part of the insured. Such schemes would also help to combat “moral hazard”; this is the tendency to have a false sense of security due to the fact that there is insurance, therefore increasing the exposure rather than decreasing it (Doherty, 1997). Faced with a catastrophic event, a government is politically obligated to provide relief, but also to minimize the costs of doing so. Options for raising money for disaster relief are: 1 2 3 4

budget transfers; borrowing money; raise taxes; offering subsidized loans.

Whatever option is chosen, the government will have less funds available to invest on development, since funds will be transferred whether to directly pay for the relief, or for the payments of the newly acquired debt. This problem is particularly acute in developing countries, where studies by the World Bank have shown that natural catastrophes have a direct impact on the rate of growth of nations, and depending on the level of access to post-disaster financing, they can offset poverty reduction policies (Freeman et al., 2002). Finally, raising taxes is never popular, and in the case of floods, where there is a geographical concentration of the affected, those outside the floodplain are usually not willing to pay for someone else’s problem. 9.3.3

Traditional flood insurance

To be insurable, a risk must comply with the following criteria: 1 2 3 4

Mutuality. A large number of people that are at risk combine to form a risk community. Need. When the event occurs it must place the insured in a position of financial need. Assessability. It must be possible to calculate the expected burden of loss. Randomness. The time at which the event occurs must not be predictable, and the occurrence must be independent of the will of the insured. 5 Economic viability. The community must be capable to cover its future loss-related financial needs on a planned basis. 6 Similarity of threat. The community must be exposed to the same threat level, and the occurrence must give rise to the need for funds in the same way for all concerned. The perception in the insurance industry is that floods do not meet the mutuality requirement, since frequently affected areas are the only ones insured, and such a community is usually too small to carry an economically sound solution for both the insurer and the insured. Assessability and economic viability are present only to a certain degree, particularly for large catastrophic floods, where damages are difficult to foresee and most of the community is affected at once. In other words, insurance premiums have been difficult to estimate in advance due to highly local spatial concentration and correlation, and uncertain future expected losses. To overcome such a problem, some actors in the insurance industry have proposed grades systems, where the price of the premium is weighted to the exposure (Swiss Re, 2002), as shown in Figure 9.8. Even though it is widely assumed that flood risk has large amounts of inherent unpredictability and uncertainty in the magnitude and frequency of the extreme event, flood risk level for the following season can be dynamically estimated, and therefore mitigating actions can be performed ex ante, thereby reducing the costs of flooding events. 9.3.4

Securitization

Securitization means that the risk is packaged on a standardized form (e.g. a bond) and sold on the capital market, thus securing the risk in the capital market. The investor who buys the bond

148

Gonzalo Pizarro and Upmanu Lall

92% 1% Property distribution

3%

Hazard: affected every

100 – 200 years

Risk premium per zone

3.5%

Risk premium when all property in relevant zone is insured for the same premium

1%

1%

2%

200 – 300 years

1.6%

300 – 400 years

400 – 500 years

0.8%

0.5%

500 –1000 >1000 years years

0.2%

0.05%

3.5% 2.8% 2.4% 2.2% 1.9% 0.2%

Compulsory insurance

Figure 9.8. Risk adequate premiums as a function of the size of the risk community (see Color Plate XLI). (Source: Menzinger and Brauner, 2002; Swiss Re, 2002.)

issued by the insurer will lose his or her interest, or even the entire investment (depending on the structure of the contract) if there is a catastrophe of a specified magnitude of loss, within a specified maturing date. If there is no catastrophe, the investor will get the money back with a high rate of interest. In theory, securitization can spread the financial costs beyond national borders and place the financial burden voluntarily on investors over the world. Further, securitization may involve low transaction costs compared to traditional reinsurance. Finally, securitization can reduce the financial unpredictability of disaster relief (Stipple, 1998). Insurance Linked Securities (ILS) in its most common form of Cat Bonds, has emerged as an alternative to traditional reinsurance. To issue an ILS, the government, or insurance company forms a figure known as “Special Purpose Financing Vehicle” (SPFV) or “Special Purpose Reinsurance Vehicle” ( SPRV) if it is an insurance company issuing the Cat Bond. The setting is shown in Figure 9.9. The SPFV acts as both a financial partner to the government and to the issuer of the bond. The SPFV sells the government a catastrophe reimbursement contract, the price of which depends on the price of the bond. To offset the reimbursement obligation undertaken, the SPFV issues the Cat Bond. The bond can be “principal at risk” or “interest at risk.” That is, the investors will lose their principal (the invested capital), or part of it, or only the interest if a catastrophe occur. If there is no catastrophe, the investors will get their money back with interest. The SPFV invests the money received for the bond in a trust fund and receives the interest on the trust fund. The rate earned at the trust fund level must be supplemented by the rate on line payment to offset the interest cost of the bond. This means that the actual risk premium paid for the catastrophe reinsurance is the difference between the interest cost of the bond and the rate earned on the trust fund. The SPFV invests the money received for the bond in a trust fund and receives the interest on the trust fund. When it is a government issuing the Cat Bond the trust fund can be managed more aggressively since the capital management in the trust fund can take on larger risks (Stipple, 1998). The difference when it is an insurance company issuing the Cat Bond is that the SPRV acts as both reinsurer to the ceding company and issuer of the bond. The SPRV sells the insurer a catastrophe reinsurance contract and receives a payment calculated on the rate on line.

Climate drivers and flood risk management 149

Proceeds

Reimbursement contract

Special Purpose Finance Vehicle (SPFV) or Special Purpose Reinsurance Vehicle (SPRV)

Catastrophe bond Interest and principal

Investors

Proceeds

Interest

Government or corporation

Principal

Supplemental interest expense

Trust fund

Figure 9.9.

Institutional structure for a catastrophe bond (Cat Bond). (Source: Stipple, 1998.)

To date, Cat Bonds have been issued for earthquakes in California and Japan, hurricanes in the coast of the United States, typhoons in Japan, and windstorms in Europe. No flood triggered Cat Bond has been issued mainly due to low assets to be securitized, high spatial concentration of the losses, and lack of good estimates of the assets at risk (i.e. combination of the economic value of the assets exposed, and the probability of being flooded). In the case of Sacramento, the potential losses are similar to those estimated for some of the Cat Bonds being issued and the climate understanding of the predictability of the risk overcomes the second objection; therefore, this option becomes a possibility for the basin. ILS was first introduced in 1996, and since then there has been an issuance of more than 8.1 billion dollars, 75% being Cat Bonds. Cat Bonds have been successful owing to their attractiveness to both the sponsor and the investor. For the sponsors (insurers, reinsurers, governments, and corporations), Cat Bonds offer multi-year protection against natural catastrophes without the credit risk present in traditional reinsurance. To the investors, Cat Bonds offer attractive returns and the potential of reducing portfolio risk, as Cat Bonds are largely uncorrelated with most other securities.

9.4 9.4.1

CLIMATE VARIABILITY AND FLOOD RISK MANAGEMENT General considerations

For the structural solutions to be effective their design parameters must accurately reflect the risk that the system will endure during the useful lifetime of the project. As was illustrated in the previous section, the presence of nonstationarities in the hydrologic time series cannot be ignored when estimating design values for future time horizons the presence of nonstationarities in the flood record may result in the design parameters being fictitious if the regime cycles are not known for the basin. Thus, incorporation of climate information in the estimation of design parameters can lower the risk of failure of the system. It has been shown that ignoring a weakly significant nonstationarity in the flood record may seriously bias the quantile predicted for time horizons as near as 0–20 years (Cunderlik and Burn, 2004). The clustering of extreme events also has a nonlinear aggregate effect on economic development. This becomes particularly significant when limited access exists to post-disaster financing, which is the case in most low-income developing countries (Freeman et al., 2002). Funds of the

150

Gonzalo Pizarro and Upmanu Lall

magnitude required to rebuild damaged infrastructure investments may not be available to emerging economies in advance, particularly in the case of severe catastrophes. Also, funds are required immediately after the catastrophe to implement urgent relief programs, which divert funds from other development related programs. Studies of the World Bank have shown that the impact of natural disasters on developing countries can stall or defeat completely the poverty reduction measures. By knowing the drivers of the climate system a more thorough analysis of the exposure to flood risk can be performed, and a better decision can be obtained regarding the optimal options to manage the risk according to the available budget, thus allowing appropriate development and poverty reduction projections. In terms of infrastructure, reservoirs are one of the key tools for water resource management in general and flood control in particular. Reservoirs are frequently multipurpose, that is, they have more than one goal in their operation. Multipurpose reservoirs usually have conflicting optimal operating policies, which make their operation complex. One example of conflicting objectives is water supply and flood control goals. For water supply the goal is to keep the reservoir as full as possible so as to minimize the amount and length of failures of supply. For flood control, on the other hand, the optimal situation is to have as much volume available as possible, so that in the event of a flood, all or as large a portion as possible, will be contained there by minimizing downstream damage. These policies are usually derived from the historical ensemble of flows, and using the driest season available, and, in the case of flood control allocated volume, regulation mandated volumes that must be kept available. This is done on the basis of the value that has been derived as the longterm peak flow that represents a desired residual risk such as the 100-year or 500-year flow. In the past 20 years, there has been an increasing understanding of the ocean-atmosphere teleconnections and their impact on total and peak flows for a particular stream. Consequently, the incoming streamflow and FC volume needed for the upcoming season are becoming increasingly predictable. To date, few studies have investigated the impact on the management of reservoir systems ahead-of-season forecasts of streamflow and FC volume needs (Yao and Georgakakos, 2001). Incorporating climate information and forecasting allows the operation of the reservoir based in a policy regimen of adapting the allocated volumes to those that have the desired residual risk for that particular year. Another issue is the timing of investment; that is, when should a particular piece of infrastructure be built in light of changing climate? In the case of the American River Basin this is crucial. Whether the peaks remain as they are today, continue increasing, or switch to a low flood setting for the next 50 years – these imply completely different answers to the need of building a new dam, as was recommended by the USACE. As discounting really impacts benefit-cost analysis if there are identifiable climate regimes, the questions are as to how the timing of investment in the dam needs to be determined, and whether we have accounted for all possible management/operational changes. 9.4.2

Regional considerations and reinsurance

As previously mentioned, the assumption of randomness and independence of extreme weather events have led the insurance industry to consider each floodplain as an independent unit to be insured, exposed to a static risk, with high level of uncertainty. Climate variability, manifested as nonstationarities in the flood record, shows that the flood risk is not static in a temporal dimension, and it can be estimated more accurately than previously thought. Such advancements in the understanding of the nature of the mechanisms that produce floods and its precursors allow for temporal hedging and the adjustment of the prices of insurance to a more accurate risk level. Another feature that has been identified in the literature is the existence of coherent regions that respond in a statistically significant way to common precursors, and in some cases, respond

Climate drivers and flood risk management 151 in opposite ways. Thus, in response to a particular state of the SSTs in the Pacific Ocean, such as an El Niño, certain regions may have an enhanced risk of high floods, while others will have a diminished risk of high floods (Pizarro and Lall, 2002). These characteristics of the hydrological system open the possibility of spatial hedging, also breaking the limitation of economic viability as now the community at risk can be seen not as the floodplain, but a group of floodplains with similar risk levels, but opposite responses to climate conditions. It is important to bear in mind that the public interest and the reinsurance interest on weather related extreme events are not exactly the same. This arises from the fact that the reinsurance industry is interested in events that cause losses in heavily insured areas, while, as it has been pointed out earlier, floods often hit regions of little insurance penetration and relatively limited infrastructure for emergency response (Murnane, 2004). Further research is needed in order to design instruments that allow appropriate financial coverage for developing nations, while being of interest to investors and institutional players in the industry. A simple way of understanding reinsurance is to consider it an insurance for the insurance industry; that is, they are contracts used to spread the risk of catastrophic losses for insurance companies rather than individuals. These contracts are for portfolios, not for individual premiums. Also, reinsurance contracts are generally renewed on an annual basis with most contracts starting in January 1. Skillful seasonal forecasts issued before January 1 can, therefore, affect the pricing of the contract or make the reinsurers adjust their exposure through retrocession, that is, reinsurance of reinsurance. There is typically a one-time mandatory reinstatement clause on reinsurance contracts; that is, if an event occurs and a company accesses a reinsurance layer then it is required to “top up” its cover by purchasing additional reinsurance (Murnane, 2004). In other words, the insurance company has to obtain extra coverage after the occurrence of a second flood, imposing a higher financial burden. Therefore, clustering of flooding events that cause large losses could pose a significant problem for an insurer. As of today, reinsurers do not use forecasts as a primary factor in business decisions (Murnane, 2004). Incorporating climate information, and developing suitable forecasts with good skills, related to the risk level of the asset, could lead to a better management of contracts. Also, it could lead to incentives to create multi-year contracts that cover entire hydrologic cycles, rather than year-by-year ones. 9.4.3

General approach for a dynamic flood risk management

It has been shown in the previous sections that (1) understanding climate variations is the key to understanding changes in flood risk, and (2) climate and hence flood risk may be predictable at seasonal time scales, thus providing unprecedented opportunities for warning and planning. Additionally, remote sensing, mapping, and modeling tools provide the opportunity to monitor rapidly changing conditions over a large area very effectively and provide the opportunity for targeted actions. The general approach is illustrated in Figure 9.10. 1 Long range planning and analysis. Using planetary and local hydroclimate data, develop an understanding of how regional climate variations lead to changed risk of flood events in the dominant flood season. Diagnose whether the local patterns of changing flood risk we see are a part of a hemispheric pattern; using 100 years or more of global climate data provide an assessment of the regimes of climate that lead to persistent changes in flood probabilities as well as estimates of the recurrence intervals of these regimes to better define long-term flood risk in the region and its multi-year variation. Are their spatial patterns associated with flood recurrence at the national scale? Use this information to inform decisions on land use, insurability, etc. 2 Season ahead preparation. Develop an understanding of the specific predictable aspects of flood risk in the region, a season to one year ahead, using climate models as well as statistical predictors. Use design interventions (short-term structural measures, insurance programs)

152

Gonzalo Pizarro and Upmanu Lall

Planning Climatological and physical spatial risk characterization and socioeconomic vulnerability analysis Evaluation of risk management options Structural – dams/levees/drainage Non-structural – land use/relief supplies and distribution Financial – insurance/relief Design of risk management program Seasonal – potential hot spots/insurance changes/preparation Event – monitoring, quick response

Season ahead preparation Forecast of changing risk implementation and operation of risk management processes

Event warning and response Warning system operation Models Imagery Operation of response and relief actions

Figure 9.10.

Concept diagram for climate-driven risk management approach for flood hazard.

and monitoring programs that lead to specific actions that reduce losses and displacement of populations. Identify which parts of the country/region have high vulnerability in the upcoming season and what processes including education/communication can be put forward. Use quantitative methods (as in Jain and Lall, 2000, 2001; Sankarasubramanian and Lall, 2003) to provide a medium-term forecast of flood risk using statistical methods. 3 Event warning and response. Develop the strategy for rapidly highlighting where the problems are evolving and where response is immediately needed. Besides the existing network of streamflow and precipitation (revisable regarding specific key locations), a system for remote sensing of soil moisture, with the goal of estimating landslide risk, must be put in place.

Climate drivers and flood risk management 153 4 Integrated approach to institutional hazard response and preparation. Use all the above analyses as part of a formal framework to inform appropriate institutions and design new policy measures and financial instruments to increase the resilience of the local population to hazards and to better quantify the impact of climate-driven shocks on the economy at local and aggregate levels. 9.5

SUMMARY AND CONCLUSIONS

Nonstationarities in the flood time series can be related to low frequency climatic variability and present potential for prediction. It has been shown, also, that such nonstationarities imply that the risk of flooding, or the risk of exceeding an appropriate threshold level, changes dynamically, contingent on climate state. The need for a framework that formally addresses the estimation of potentially changing flood frequency distributions and their uncertainty has been identified (NRC, 1999). For the case of Sacramento, a combination of actions seem to be best suited for managing a flood risk that changes dynamically. Such actions should include: 1 Optimizing the operating rules of the existing Folsom dam to use climate information to allocate flood control capacity according to the flood risk of that year. 2 A combination of traditional flood insurance, such as the NFIP program, and flood-triggered Cat Bonds. 3 Conditioning access to flood insurance coverage to the inclusion of mitigative actions to be included in building codes. 4 Land-use restrictions for areas to be developed, to limit exposure. Whether a new dam will ever be built is more a political question than a technical one in this case, but, from the technical point of view, further research related to the optimal point of investment is needed owing to the nature of the nonstationarity. Also, the question of what is the optimal discount rate to be used for economic analysis in situations where larger damages are expected in the distant future needs to be addressed (Heal and Kristrom, 2002). Under such circumstances, long-term flood protection management should be dealt with as a dynamic process, with short-lived projects that have 15–20 years of expected life, for example, and by reevaluating future needs on a regular basis as the basin changes its behavior, or the nature of the nonstationarity and its consequences, both socioeconomic and hydrologic, are better understood. There is a need for a new approach to flood risk management, an approach that understands that flood risks are dynamic and dependent on the climate, and have the institutional setting necessary to have the proper feedbacks in place. REFERENCES Barlow, M., Nigam, S., and Berbery, E.H. (2001) ENSO, Pacific decadal variability, and US summertime precipitation, drought, and stream flow. Journal of Climate, 14(9), 2105–2128. Berz, G.A. (1999) Catastrophes and climate change: concerns and possible countermeasures of the insurance industry. Mitigation and Adaptation Strategies for Global Change, 4(3/4), 283–293. Cane, M.A. and Zebiak, S.E. (1985) A theory for El Niño and the Southern Oscillation. Science, 228(4703), 1085–1087. Cayan, D.R., Redmond, K.T., and Riddle, L.G. (1999) ENSO and hydrologic extremes in the Western United States. Journal of Climate, 12(9), 2881–2893. Chichilnisky, G. and Heal, G. (1998) Managing unknown risks: the future of global reinsurance. Journal of Portfolio Management, 24(4), 85–91.

154

Gonzalo Pizarro and Upmanu Lall

Chow, V.T., Maidment, D.R., and Mays, L.W. (1994) Applied Hydrology. Bogota, Colombia: McGraw-Hill. Cunderlik, J.M. and Burn, D.H. (2004) Linkages between regional trends in monthly maximum flows and selected climatic variables. Journal of Hydrologic Engineering, 9(4), 246–256. Doherty, N. (1997) Financial innovation for financing and hedging catastrophe risk. Paper presented at the Conference on Financial Risk Management for Natural Catastrophes. Freeman, P.K. and Kunreuther, H.C. (1997) Managing Environmental Risk through Insurance. Boston, MA: Kluwer Academic. Freeman, P.K. et al. (2002) Wharton School, environmental risk management for developing countries. The Geneva Papers on Risk and Insurance: Issues and Practice, 27(2), 196–214. Gutzler, D.S., Kann, D.M., and Thornbrugh, C. (2002) Modulation of ENSO-based long-lead outlooks of southwestern US Winter precipitation by the Pacific Decadal Oscillation. Weather and Forecasting, 17(6), 1163–1172. Heal, G. and Kristrom, B. (2002) Uncertainty and climate change. Environmental and Resource Economics, 22(1), 3–39. Hidalgo, H.G. and Dracup, J.A. (2003) ENSO and PDO effects on hydroclimatic variations of the Upper Colorado River Basin. Journal of Hydrometeorology, 4(1), 5–23. Hirschboeck, K.K. (1991) Climate and floods. In: R.W. Paulson, E.B. Chase, R.S. Roberts, and D.W. Moody (eds) National Water Summary 1988–89 – Hydrologic Events and Floods and Droughts. Water-Supply Paper 2375, US Geological Survey, pp. 67–88. IPCC (2001) Summary for Policymakers: A Report of Working Group I of the Intergovernmental Panel on Climate. Geneva: The Panel. Jain, S. and Lall, U. (2000) Surface water and climate – magnitude and timing of annual maximum floods: trends and large-scale climatic associations for the Blacksmith Fork River, Utah (Paper 2000WR900183). Water Resources Research, 36(12), 3641–3652. Jain, S. and Lall, U. (2001) Surface water and climate – floods in a changing climate: does the past represent the future? (Paper 2001WR000495). Water Resources Research, 37(12), 3193–3207. Kleindorfer, P.R. and Kunreuther, H. (1999) Annual meeting of the society for risk analysis: invited Nugget papers – the complementary roles of mitigation and insurance in managing catastrophic risks. Risk Analysis, 19(4), 727–739. Kunreuther, H.C. and Linnerooth-Bayer, J. (2003) The financial management of catastrophic flood risks in emerging-economy countries. Risk Analysis, 23(3), 627–639. Livezey, R.E., Masutani, M., Leetmaa, A., Rui, H., Ji, M., and Kumar, A. (1997) Teleconnective response of the Pacific–North American region atmosphere to large Central Equatorial Pacific SST anomalies. Journal of Climate, 10(8), 1787–1820. Mantua, N.J., Hare, S.R., Zhang, Y., Wallace, J.M., and Francis, R.C. (1997) A Pacific interdecadal climate oscillation with impacts on salmon production. Bulletin of the American Meteorological Society, 78, 1069–1079. Menzinger, I. and Brauner, C. (2002) Floods are Insurable! Zurich: SwissRe. Milly, P.C.D., Wetherald, R.T., Dunne, K.A., and Delworth, T.L. (2002) Increasing risk of great floods in a changing climate. Nature, 415(6871), 514–516. Minobe, S. (1999) Climate and land surface studies – resonance in bidecadal and pentadecadal climate oscillations over the North Pacific: role in climatic regime shifts (Paper 1999GL900119). Geophysical Research Letters, 26(7), 855–858. Minobe, S. and Mantua, N. (1999) Interdecadal modulation of interannual atmospheric and oceanic variability over the North Pacific. Progress in Oceanography, 43(2), 163–192. Munich Re (2001) Topics – Annual Review. Natural Catastrophes. Murnane, R.J. (2004) Climate research and reinsurance. Bulletin of the American Meteorological Society, 85(5), 697–707. NRC (1999) Improving American River Flood Frequency Analyses Corp Author(s): National Research Council (US); Committee on American River Flood Frequencies. Washington, DC: National Academy Press. Pielke, R.A. Jr and Downton, M. (2000) Societal impacts group, precipitation and damaging floods: trends in the United States, 1932–97. Journal of Climate, 13(20), 3625–3637. Pizarro, G. and Lall, U. (2002) El Niño and Floods in the US west: what can we expect? Eos, 83(32), 349, 352.

Climate drivers and flood risk management 155 Porporato, A. and Ridolfi, L. (1998) Influence of weak trends on exceedance probability. Stochastic Hydrology and Hydraulics Research Journal, 12(1), 1–14. Sankarasubramanian, A. and Lall, U. (2003) Flood quantiles in a changing climate: seasonal events and causal relations. Water Resources Research, 39(5), 1134–1145. Stipple, J. (1998) Securitizing the Risks of Climate Change. Laxenburg, Austria: IIASA. Swiss Re (2002) Floods are Insurable! Zurich: Swiss Re. Yao, H. and Georgakakos, A. (2001) Assessment of Folsom Lake response to historical and potential future climate scenarios – 2. Reservoir management. Journal of Hydrology, 249(1), 176–196.

CHAPTER 10

Remote Sensing in Water Resource Management D.P. Rao Former Director, National Remote Sensing Agency, Department of Space, Government of India, Hyderabad 500 037, India

10.1

INTRODUCTION

This chapter is geoscience oriented. On the basis of the Indian experience, it describes the benefits of using satellite imagery in different areas of water resources management. Water resources management is essentially a linkage between availability from various sources and sectoral demand in which both quality and quantity need to be considered, and where both conservation and control measures need to be addressed. Conjunctive use of surface and ground water is an integral part of water resources management. Both remote sensing and ground measurements have to be integrated to provide an overview of large area as well as details of local availability. Conventional hydrologic measurements on the ground suffer from the limitations of reliability, time effectiveness and adequacy. These measurements are also discrete in space, necessitating areal averaging methods. Repetitive ground measurements many times are scarce due to the constraints of manpower and funds. Measurements over inaccessible areas and inhospitable terrain are also a limitation. It is in this context that remote sensing technique can complement and supplement ground measurements, to enable collection of data needed for sound water resources management. Remote sensing has been used as an important tool in the inventory, survey, monitoring, planning and management of natural resources including water resources. The steady flow of data from the constellation of Indian Remote Sensing Satellites (IRS-1A, -1B, -P2, -1C, -P3, -1D and P4) and international missions, namely SPOT3 and 4 and LANDSAT5 and 7, ADEOS, ERS1 and 2, JERS and RADARSAT, have facilitated operational use of this technology in the management of water resources. Sustainable water resources management calls for optimal utilization of available water resources to meet the domestic, agricultural and industrial demands. A thorough understanding of hydrological phenomena and fresh water availability and its utilization is a prerequisite to achieve the goal. Identification and quantification of key hydrological variables across a range of scales, and the development or modification of hydrologic process need to be taken up for furthering our understanding in hydrological phenomena. Furthermore, estimation of water balance at the land surface for which precise information on precipitation, runoff, evapo transpiration (ET), soil water and groundwater storage is also required. Currently, only point observations of precipitation are available. The Earth Observing System (EOS) sensors, namely Tropical Rainfall Mapping Mission (TRMM) – Microwave Imager (TMI), Visible Infrared Scanner (VIRS), Advanced Microwave Sounding Unit (AMSU) and Advanced Microwave Scanning Radiometer (AMSR) may provide data relating to precipitation and clouds. In addition, Atmospheric Infrared Sounder (AIRS) and microwave sounders (AMSU and HSB – Humidity Sounder

158

D.P. Rao

Brazil) will together provide information regarding precipitation volumes, cloud thickness and cloud water content (Anonymous, 1999). The near-surface soil moisture has been estimated through passive radiometry and active radar. It has, however, met with limited success. Direct satellite observation of near-surface soil moisture is currently under experimental study. Conventional and experimental (such as inflatable antenna) L-band radiometer under NASA’s Small Satellites and New Millennium, development of algorithm and multi-sensor signals, and improved representation of soil moisture in Land Surface Models of sub-humid regions across a variety of scales may help mapping soil moisture.

10.2

GROUNDWATER

10.2.1 General considerations The importance of groundwater in water resources management arises out of the following considerations: (1) The quantity of freshwater within the drillable depth is about 70 times greater than all the surface waters (rivers, lakes, reservoirs, etc.) in the world put together, (2) quality of groundwater is generally superior to surface water, particularly in terms of biological quality and (3) groundwater is the main source of potable water for domestic purposes in most parts of the world. Being a hidden resource, groundwater is not directly amenable to remote sensing. However, remote-sensing data provide integrated information on several factors, which directly or indirectly control the movement and occurrence of water below the ground. The synoptic view of the satellite data helps in mapping different lithologic, structural and morphological units in their correct spatial relationship, and the multi-spectral data provide certain additional information which is not otherwise easily observable on the ground. Groundwater regime is a dynamic system wherein the water absorbed at the earth’s surface passes through geological strata and is recycled back to the surface. In this process, apart from geological framework, various elements like relief, slope, ruggedness, depth of weathering, nature of weathered material, thickness of deposition and nature of deposited material, distribution of surface water bodies, river/stream network, canal commands, the extent and intensity of precipitation, etc. also exercise significant control on the groundwater regime. Thus, the framework in which the groundwater occurs is as varied as that of rock types, as intricate as their structural deformation and geomorphic history and as complex as that of the balance amongst the lithological, structural, geomorphic and hydrologic parameters. The possible combinations and their intricacy are virtually infinite leading to the unavoidable conclusion that the ground water conditions at a given site are unique and are not completely amenable to scientific understanding. Some of the conditions are often obscured and are not clear even during the field observations. However, factor-wise analysis, systematic mapping, data integration and interpretation based on conceptual understanding are helpful in overcoming this problem to some extent. 10.2.2 Assessment of groundwater resource In India, more than 90% of rural and nearly 30% of urban population depend on groundwater for meeting their drinking and domestic requirements. In addition, groundwater provides for nearly 60% of the irrigation potential existing in the country. The distribution of groundwater is not uniform in all the regions. The spatial and temporal variations in rainfall and regional/local differences in geology and geomorphology have led to an uneven distribution of ground water in different regions across the country. This uneven distribution (poor prospects) and indiscriminate tapping (over-exploitation) in certain zones are the main reasons leading to scarcity of drinking water in many parts of the country. In view of this, a large number of habitations in the country have remained as problem villages not having sustainable drinking water sources. It is in this

Remote sensing in water resource management 159 context, that the identification of groundwater sources for planning sustainable drinking water schemes assumes great significance. In many parts of India, groundwater levels have started declining and in some parts, the existing wells have dried up due to lowering of water table. In recent years, satellite data have proved to be highly useful in mapping and monitoring the groundwater in over exploited areas in conjunction with the ground hydro-geological information and providing additional inputs for more realistic estimation of groundwater resource. A modified approach has been evolved by Reddy (1999) by integrating lithology, landform, structure, slope and soil and land use–land cover information for estimating the individual unit-wise groundwater recharge. This approach not only helps in reducing the error of averaging of aquifer parameters over large areas but also facilitates more systematic planning and management of groundwater resource. As irrigation accounts for more than 90% of groundwater utilization in the rural areas, irrigated area statistics form one of the key inputs for groundwater draft estimation. Satellite data helps in accurate identification, mapping of the area irrigated by groundwater, that is, other than canal and tank commands. Thus, it helps in approximately arriving at the amount of groundwater that is being used during different seasons of the year in different zones. In the past, the groundwater studies were conducted by conventional hydro-geological surveys. In this approach the geological and hydrological information was collected in the field along traverse lines and representative points. With the development of remote-sensing technology, the mapping procedures have undergone significant changes. 10.2.3

Factors controlling the groundwater regime

Varying degrees of uncertainty and inconsistency are inherent in the existing methodology (conventional hydro-geological mapping). Hence, a new approach which is systematic and simpler with well-defined units has been evolved. In this approach, all the variables controlling groundwater regime have been grouped into the following four parameters: 1 2 3 4

geology/lithology; geomorphology/landforms; geological structures; recharge conditions.

If the information on these four parameters is precisely known, it is possible to understand the groundwater regime better, and visualize the gross aquifer conditions, like nature of the aquifer material, type of aquifer, the type of wells suitable, the depth range, yield range, success rate, area of influence and sustainability in each unit. The influence of each of these parameters on the groundwater regime and the information that can be derived with respect to these four parameters using the satellite imagery are briefly discussed in the following paragraphs. 10.2.3.1 Geological mapping Remote sensing provides the basis for discrimination and differentiation of rock types. Though direct identification is limited to a few contrasting rock types, many of the rocks can be discriminated based on their spectral and morphological characteristics (Table 10.1). Once rock types are identified, the physical continuity of individual rock units can be easily traced and the exact shape, size, and geometry can be identified and mapped with minimum ground survey and more accurately by using satellite imagery. 10.2.3.2 Mapping geological structures The most obvious structural features that are important from the point of groundwater are lineaments. They are seen on the satellite imagery as linear alignments of structures, lithology,

Table 10.1. The morphological and spectral characteristics of different rock groups as expressed on the satellite imagery. Sl. no.

Rock type

1

Plutonic igneous rocks (e.g. granite, syenite, diorite, gabbro, dunite, pyroxenite, amphibolite, etc.)

2

Dykes and other intrusives (e.g. dolerite/peridote dykes and sills, pegmatite bodies, quartz reefs)

3

Volcanic rocks (e.g. basalt, andesite, dacite, rhyolite)

4

Sedimentary rocks (e.g. sandstone, quartzite, shale, limestone, conglomerate)

5

Metamorphic rocks (e.g. quartzite, phyllite, schist, gneiss)

6

Calcareous rocks (e.g. limestone, Dolomite, marble, gypsum, chalk)

7

Duricrusts (e.g. laterite and bauxite)

Physical characteristic Occurs as positive landforms, boldly standing out as hills and inselbergs, in the form of exfoliation domes, monadnacks, bornhardts, tors, koppies, etc. with radial, concentric, annular and mixed complex drainage patterns and sparse vegetation Display linear trends and appear as wall-like features, resistant ridges, hogback ridges, etc. If the surrounding rocks are more resistant, they form linear depressions and furrows Volcanic cones, lava plains and tongues with branching pattern, flow lines, pyroclastics, ropy structures, etc. are common. They show crude stratification, with successive lava flows varying in thickness, composition, and sometimes interbedded with tuff and ash beds. Occur as mesas, buttes and flat valley floors with mixed complicated drainage pattern due to columnar jointing Horizontal to sub-horizontal beds, often with contrasting lithologies, forming, hogback homoclinal ridges, cuestas and stratiplains. Ridge and valley topography with modified trellis and parallel drainage patterns. Stratification, lensing, etc. can be seen on large-scale aerial photographs. Arenaceous rocks have light tone and low drainage density. Argillaceous rocks have medium tone and high drainage density Appear as elongate parallel belts of valleys and hills, often display bedding, foliation, schistosity as thin parallel lineations; folded, crenulated, wavy or crinky pattern, characterised by trellis, rectangular and parallel drainage patterns; quartzites usually occur as strike ridges Indicated by sinkholes, solution-cavities, depressions, blind-valleys etc. and karst topography, often marked by pitted appearance. Deranged type of drainage pattern is common Occur as plateaux, with bright tone on the imagery with dendritic drainage and sparse vegetation

Source: Reddy (1987). Note: Based on their spectral characteristics manifested on the imagery as different tones (colours), further differentiation within the above rock groups can be made into acidic rocks (light tone), intermediate rocks (medium tone), basic and ultrabasic rocks (dark tones), etc. The moist and thin soil covered areas appear as dark toned areas on the image. Vegetation appear as bright red patches in the standard FCC images and water bodies appear as light blue to black depending on their depth and turbidity. Generally, the morphology (landforms) indicates the textural and structural variations and tone on the image indicates the compositional variations in the rock types. By combining both morphological and spectral (tonal) characteristics, different rock groups/types can be mapped.

Remote sensing in water resource management 161 topography, vegetation or drainage anomalies, either as straight lines or curvilinear features. They can be further classified into faults, fractures, joints, bedding traces, contacts of rocks, shear zones, etc. based on the image characteristics and association, after corroborating with the ground information. In hard rock areas, faults and fractures mainly act as conduits for groundwater movement and form prospective groundwater zones. 10.2.3.3 Geomorphological mapping Geomorphology exercises a significant control over groundwater regime. The relief, slope, depth of weathering, type of weathered material, thickness of deposition, nature of the deposited material and the assemblage of different landforms play an important role in defining the groundwater regime, especially in the hard rocks and the unconsolidated sediments. Satellite imagery, due to its synoptic view, facilitates better appreciation of geomorphology and helps in mapping different land forms and understanding their origin, sequence of evolution, material content and other characteristics. Palaeo-river course, buried channel, alluvial plain, bajada deposit, closed valley, deeply weathered pediplain, glacial till, fracture/fault line valley and karst landscape, which form prospective groundwater zones, are most evident from satellite data. 10.2.3.4 Recharge to ground water Recharge is the most important factor in evaluating the groundwater prospects. Recharge to groundwater depends on precipitation, surface water bodies like reservoirs, lakes, tanks, streams, canals, canals commands, favourable rocks, landforms and structures, etc. Satellite imagery provides information on surface water bodies, canals, canal commands, and irrigated fields, which directly contribute to groundwater recharge. Based on geomorphic and hydrologic analysis, the areas can be classified into runoff, recharge, storage and discharge zones, and the amount of recharge to groundwater from different sources can be estimated. 10.2.3.5 Classification system For mapping the four parameters mentioned before, that is, lithology (geology), landforms (geomorphology), geological structures and recharge conditions, well defined classification systems have been evolved as furnished in Tables 10.2, 10.3 and 10.4, respectively. Based on the visual interpretation of satellite imagery in conjunction with the existing geological/hydro-geological data and limited field checks, all the four parameters have to be separately mapped. The rock (lithologic) units have to be indicated with numerical code numbers, landforms with alphabetic codes and geological structures and hydrological data with line symbols. 10.2.4

Preparation of groundwater prospect maps

The groundwater prospect map is prepared by integrating the information on the four parameters mentioned earlier. For this purpose, by combining the rock type and landform information, integrated rock-cum-landform units (hydro-geomorphic units) are evolved. These hydrogeomorphic units will have less heterogeneity and exercise better control on the groundwater regime. While mapping these hydro-geomorphic units, the boundaries between rock type and landform have to be made co-terminus, wherever within a rock unit two or more landforms and vice-versa are possible. The structural information as per Table 10.4 has to be represented on the map with appropriate line symbols. Similarly, the hydrological information like depth to water table, rainfall, observation wells with their yield range, etc. have to be represented on the map with contours and point symbols in different colours. Thus, the groundwater prospective zones map is prepared, which provides much better information than the conventional hydro-geological map.

Table 10.2. Code no.

Hydro-geological (lithological) classification system. Rock group

1

Unconsolidated sediments

2

Residual cappings

3

Deccan traps and inter-trapeans

4

Other volcanics/ metavolcanics

5

Semi-consolidated sediments

6

Consolidated sediments

7

Intrusive rocks

8

Crystalline rocks

Code no. 11 12 13 14 15 16 17 21 22 23 24 25 26 31 32 33 34 35 36 37 41 42 43 44 45 51 52 53 54 55 56 57 58 61 62 63 64 65 66 67 68 71 72 73 74 75 81 82

Lithologic unit Gravel, sand, silt Clayey sand Sandy clay Clay Layers of sand, silt and clay Colluvium Others Laterite (ferricrete) Bauxite (alucrete) Kankar (calcrete) Chert (silcrete) Secondary/detrital laterite Others Massive basalt Vesicular basalt Tuffaceous basalt Rhyolite Red/green bole Intertrappeans Others Basalt Rhyolite Dacite Andesite Others Sandstone and conglomerate Shaly sandstone Sandy shale Shale, clay, coal/lignite Shell-limestone Sandstone with shale/coal partings Shale with sandstone partings Others Shaly limestone Thin-bedded limestone Shale with limestone bands/lenses Thick-bedded limestone Compact shale Thick-bedded sandstone/quartzite Thin-bedded sandstone/quartzite Others Quartz reef Basic dyke/pipe rock Basic sill Quartz/pegmatite/aplite intrusion Others Massive granite and plutonic rocks Granite and gnessic complex

Remote sensing in water resource management 163 Table 10.2. Code no.

Continued. Rock group

Code no. 83 84 85 86 87 88 89

Lithologic unit Phyllite/schist/slate with bands of quartzite/quartz reefs Quartzite with shale/phyllite bands Phyllite, schist and slate Gneiss Schist and gneiss mixed Meta-basics/metavolcanics Others

Source: Reddy (1991). Note: Additional lithological units which are not covered in the above classification may be added in each group.

10.2.4.1 Legend and colour scheme An innovative legend has been developed for preparing the groundwater prospective zones map. In this legend, individual hydro-geomorphic unit-wise rock type, landform and recharge conditions are listed out in tabular format, and taking these factors into consideration the groundwater prospects of each unit have been evaluated in terms of (1) nature of the aquifer material, (2) type of aquifer, (3) type of wells suitable, (4) depth range of wells, (5) yield range of wells, (6) success rate of wells, etc. Further, to make the hydro-geomorphic units less heterogeneous, the faults, fractures and lineaments occurring in different hydro-geomorphic units can also be separated and mapped as fracture zones and the groundwater prospect for these fracture zones can be indicated separately. Thus, the heterogeneity arising out of fracture system in hard rocks can be reduced to a great extent. In the maps, each hydro-geomorphic unit’s groundwater prospects are indicated by different colours. For this purpose, a colour scheme has been devised using 7 colours of VIBGYOR colour spectrum for representing seven categories of groundwater prospects, that is, excellent, very good, good, moderate, poor, limited and nil, based mainly on the yield range of wells in different hydro-geomorphic units. The excellent prospects can be indicated with violet colour followed by very good-indigo, good-blue, moderate-green, poor-yellow, limited-orange and nil-with red colours. Wherever, the quality of water is not good/not potable it is indicated with grey colour. Based on this colour scheme it is easier for the user to identify the target zone around a specific point using the groundwater prospect map. In India, nearly half of the current irrigated area of about 90 million ha, and large urban and rural populations draw water from ground water resources. Studies have shown how integration of geological, geophysical and remote sensing data in geo-statistical models can help improve the success rates of high yielding irrigation wells.

10.3 10.3.1

OTHER APPLICATIONS OF REMOTE SENSING IN WATER RESOURCES MANAGEMENT Rainfall, snow and glacier studies

Rainfall is one of the most important processes in the hydrological cycle and is also one of the most difficult to monitor. Since late 1960s, many researchers have attempted to derive techniques for the estimation of rainfall from the visible and infrared imagery provided by

Table 10.3.

Hydro-geomorphological (landform) classification system.

Physiography

Code

Geomorphic unit/landform

Hills

CH CHR CHH CHC CHB CHD CHM RH RHR RHH RHC RHB RHD RHM — LR FR C M D I PIC PRC PCC PMC UP UPS UPD MP MPU MPD LP LPU LPD OPF — PS — PA PF — PD PP PPS PPM PPD PPG PPL SP SPS SPM

Composite hills Ridge type Homoclinal type Cuesta type Butte/mesa type Dome type Massive type Residual hills Ridge type Homoclinal type Cuesta type Butte/mesa type Dome type Massive type Inselbergs Linear ridge Folded/accurate ridge Cuesta Mesa/butte Dome Inselberg Pediment Inselberg Ridge complex Cuesta complex Mesa/butte complex Upper plateau Undissected Dissected Middle plateau Undissected Dissected Lower plateau Undissected Dissected Outer fringe of plateau Erosional Piedmont slope Depositional Piedmont alluvial plain Piedmont alluvial fan Erosional Pediment Pediplain Shallow weathered Moderately weathered Deeply weathered Gullied Lateritic Stripped plain Shallow basement Moderate basement

Plateaus

Piedmont zone

Plains

Table 10.3.

Continued.

Physiography

Valleys

Other Landforms

Code

Geomorphic unit/landform

AP APS APM APD APG FP FPS FPM FPD FPG DP DPS DPM DPD DPG CP CPS CPM CPD CPG EP EPS EPM EPD — SV CV — FV IV — BC MC PB CB NL — DF DC PY AR — B BR BRC PBC SW

Alluvial plain Shallow Moderate Deep Gullied Flood plain Shallow Moderate Deep Gullied Deltaic plain Shallow Moderate Deep Gullied Coastal plain Shallow Moderate Deep Gullied Eolian plain Shallow Moderate Deep Erosional Structural valley Closed valley Depositional Filled-in-valley Intermontane valley Fluvial landforms Buried channel Migrated channel Point bar Channel bar Natural levee Eolian landforms Dune-field Dune-complex Playa Absorbed river/stream Coastal landforms Beach Beach–ridge Beach–ridge complex Plain–beach ridge complex Swale

Source: Reddy (1991). Note: Additional landforms if any depending upon the areas may be added in each group.

166

D.P. Rao

Table 10.4. Excellent Very good Good Moderate Limited Poor

Classification of recharge conditions. Recharge from continuously irrigated canal commands, permanent water bodies, perennial streams, etc. (continuous recharge throughout the year with shallow water table conditions) Recharge from temporarily/seasonally irrigated canal commands, seasonal/ephemeral water bodies/streams (high recharge for part of the year) Large recharge area with high rainfall, facilitating good recharge Limited recharge area and moderate rainfall or seepage from other sources or limited by other factors High relief areas occupying large areas with heavy runoff, very limited recharge only along narrow valleys Very limited recharge area/poor rainfall/shadowed by ground water barriers like linear ridges, dolerite dykes, etc.

Source: Reddy (1991). Note: The recharge condition for each unit may be generalized and indicated in the legend.

meteorological satellites. Manual, interactive and automatic methods have been developed for the estimation of rainfall at a number of temporal and spatial scales, and these have been applied in many different areas and situations with varying degree of success. Snow acts as a water reservoir. Hence, an assessment of its possible effect on snowmelt runoff is essential for planning and management of various multi-purpose projects on snow-fed rivers. Unlike in many other countries, where snowfields stretch along flat terrain and mountains that are easily approachable, the snowbound land in India is mostly in difficult and inaccessible mountainous terrain. Remote sensing, therefore, continues to be the only practical way of obtaining information on snow cover. The economical hydroelectric potential of Indian subcontinent has been tentatively assessed at 85,550 MW at 60% load factor (Central Electricity Authority, Government of India). Seventy five percent of the potential lies in the rivers of Himalaya, which are fed by snow, and which offer large hydropower potential on account of topography and perennial nature of flow. Snowmelt runoff estimation is thus a critical item for planning and management. Seasonal and short-term (weekly) forecasts of snowmelt runoff, with less than 5% error, are being provided for some basins in western Himalaya by the National Remote Sensing Agency since 1970s, for use by the State River Water Management Boards and the State Electricity Boards. Inventory of glaciers in the Himalaya have also been conducted at different scales using satellite data, and the information on their numbers and areal extent has been generated as a bench mark study for use in future monitoring.

10.3.2 Irrigation water management In a number of developed countries, the extent of irrigated land has now stabilized or even diminished (Shiklomanov, 1998). At present, about 15% of all cultivated land is being irrigated. In India, the ultimate irrigation potential has been estimated at 113 million ha. While enormous irrigation potential has been created at huge cost, the gap between created potential and utilization is significantly large (around 9 million ha). The technology should, thus, shift from irrigation development to more efficient irrigation management. Prolonged irrigation without providing for adequate drainage has led to water logging and subsequent soil salinity and/or alkalinity in some areas.

Remote sensing in water resource management 167 Space-borne multi-spectral measurements at regular intervals have helped in evaluating the performance in many irrigation projects across the country. The anticipated increase in irrigation area, equitable distribution and crop productivity under sponsored development programmes have been studied in some of the major irrigation command projects. The temporal and spatial analysis of satellite data helped in mapping problem pockets of poor performance. Diagnostic analysis supported by farmer surveys has enabled identification of causative factors for taking up corrective measures. The analysis has also helped in identifying cropping pattern, extent of unauthorized irrigation and poor recovery of water rates, thus, providing inputs for changes in policies and operational plans. Furthermore, spatial analysis of crop sowing periods and crop condition assessment have thrown up policy issues of relevance to irrigation scheduling, canal maintenance and agricultural productivity. Apart from performance evaluation of irrigation systems, multi-temporal satellite data have also been used to map current status and to monitor the spatial extent of water logging and soil salinity and/or alkalinity through the years in most of the irrigation projects. Such exercise has also helped in the evaluation of the progress and effectiveness of reclamation programmes by monitoring the extent and magnitude of the problem. 10.3.3

Reservoir sedimentation

Many reservoirs built at a huge investment are undergoing rapid silting and loss of storage capacity and consequent reduction in the economic life of reservoirs. The analysis of sedimentation data of Indian reservoirs show that the annual silt rate has been generally 1.5–3 times more than the designed rate and the reservoirs are generally losing capacity at the rate of 0.30–0.92% annually (Dhruva Narayan and Ram Babu, 1983). Conventional hydrographic surveys to reassess reservoir capacity are both costly and time-consuming. Multi-temporal satellite data have been used as an aid to capacity survey of many reservoirs. In addition, spaceborne multi-spectral measurements have also been related to suspended sediment load in many reservoirs, which provide information on the sediment distribution, circulation pattern and active silting zones. A comparison of turbidity levels in irrigation tanks derived from satellite data can thus assist in initiating desilting operations or catchment area treatment. 10.3.4

Watershed management

Inappropriate land use practices in the catchment area lead to accelerated soil erosion and consequent silting up of reservoirs. Watershed management is thus an integral part of any water resource project. Since the treatment of a large number of watersheds is not feasible simultaneously, watersheds requiring immediate attention are prioritized. The prioritization of watershed is based on sediment yield potential so that the treatment would result in minimizing sediment load into the reservoir. Satellite data have been extensively used in many watersheds, for deriving the parameters of the Sediment Yield Index (SYI) model developed by All India Soil and Land Use Survey, Ministry of Agriculture, Government of India. Sediment yield prediction models have been used to provide quantitative silt load estimates in watersheds. Space-borne multi-spectral data have been used to generate baseline information on various natural resources, namely, soils, forest cover, surface water, groundwater and land use–land cover and integration of such information along with slope and socio-economic data in a Geographic Information System (GIS) to generate local-specific prescription for sustainable development of land and water resources development on a watershed basis. The study covering around 84 million ha and spread over 175 districts has been taken up by the Department of Space, Government of India, under a national level project entitled ‘Integrated Mission for Sustainable Development (IMSD)’. Implementation of appropriate rainwater harvesting structures in selected watersheds under this programme has demonstrated the significant benefits by

168 D.P. Rao observing rise in groundwater levels through increased groundwater recharge and agricultural development of what was once a barren area. Multi-year satellite data is also used to monitor the impact of the implementation of watershed management programmes. 10.3.5 Flood management Floods have been causing severe damages to agriculture and human settlements and have been claiming several human lives throughout history. India is the worst flood-hit country in the world after Bangladesh. India accounts for one-fifth of global death count due to floods. About 40 million ha. or nearly one-eighth of the country is flood-prone and one-fifth of this area experiences floods in any one year. The chronic flood-prone river basins in India are the Ganges and the Brahmaputra. These Himalayan rivers flowing down the hills cause flood problems in the states of UP, Bihar, West Bengal and Assam due to high discharges concentrated during monsoon months (June–September) and large volumes of sediment are carried to the plains. Space-borne multi-spectral measurements have been used for mapping flooded areas and the assessment of flood damages by the National Remote Sensing Agency since 1970s. However, the methodology for flood mapping has become operational only in 1986. Soon after the occurrence of a significant flood event, the spatial extent of inundated area is mapped and damage statistics generated and sent to concerned State and Central Government authorities for providing relief. Active microwave sensor data with cloud penetration capability have been used since 1993 for flood mapping and damage assessment. Geographic Information System (GIS) has been used to generate Digital Elevation Models (DEM) of a flood-prone area in one of the southern states (Andhra Pradesh) with the help of in situ measurements made by Global Positioning System (GPS) to enable assessment of spatial inundation at different water levels in the river. When the satellite-derived land cover–land use and ancillary ground-based socio-economic data is draped over DEM, flood vulnerability can be assessed to provide locale-specific flood warning. 10.3.6 Drought management Drought is one of the worst natural disasters affecting the social and economic life of millions of people every year. Developing countries have suffered large losses from drought compared to developed countries. The World Meteorological Organization estimated that in the quarter of a century (1967–91) droughts have affected 50% of the 2.8 billion people who suffered from weather-related disasters. Moreover, 1.3 million of the 3.5 million people were killed owing to direct or indirect causes of drought (Obasi, 1994), with the highest loss of about 8.16 million dead and 1314 millions affected respectively in Asia, and 2.07 and 245 million respectively in Africa. The study indicated that more than 500 million people live in the drought prone areas of the world and 30% of the entire continental surface is affected by droughts or desertification process. Owing to the abnormalities in monsoon precipitation in terms of both spatial and temporal distribution, drought is a frequent phenomenon over many parts of India. Out of net sown area of 140 million ha, about 68% of the area is reported to be vulnerable for drought conditions and about 50% of the drought prone area is classified as severe where frequency of drought is regular. Based on coefficients of variation of rainfall derived from historic data, occurrence of drought is a reality once in 3–4 years in major part of the country (Rao, 1999). Timely and reliable information about the onset of drought, its extent, intensity, duration and impacts can limit the drought related losses of life, minimize the human suffering and reduce damage to the economy and environment. Space technology has made substantial contributions in every aspect of drought management such as preparedness, early warning, monitoring and mitigation. Remote-sensing data from geo-stationary and polar orbiting weather satellites such

Remote sensing in water resource management 169 as INSAT, NOAA, METSAT and other global data is used as major inputs to all the three types of rainfall predictions, namely, long-term seasonal predictions, medium-range predictions and short-term predictions, while communication satellites have significant potential for real-time dissemination of information. Normalized Difference Vegetation Index (NDVI) derived from NOAA AVHRR (1.1 km) and/or IRS WiFS (188 m) imagery is now being continuously used to monitor drought conditions on a real time basis often helping the decision makers to initiate appropriate strategies for recovery by changing cropping pattern and practices. 10.3.7

Water quality

Nearly one million children in India die of diarrhoeal diseases each year directly as a result of drinking unsafe water and living in unhygienic conditions. Some 45 million people are affected by water quality problems caused by pollution, by excess fluoride, arsenic, iron or by the ingress of salt water. Millions of people do not have adequate quantities of safe water, particularly during summer months. Increasing environmental concerns on deteriorating water quality are not well supported by ground monitoring mechanisms. Remote sensing of water quality can complement ground efforts in mapping and monitoring point and non-point pollution sources, the influx and dispersal of pollutants in the aquatic environment and consequent impact such as algal blooms and weed growth (UNICEF, 1998). Point source identification calls for high spatial resolution satellite data. Regional models of non-point source pollution loading will be benefited from remote-sensing inputs on land use–land cover, supported by sample ground data collection. Growth of aquatic weeds and algal blooms, an indication of eutrophication, have been mapped by satellite and aircraft data. Ground truth requirements are also more stringent than in land remote sensing. The GIS technology provides enhanced capability for water quality modeling.

10.4

CONCLUSIONS

Water resource management need to be planned and implemented within the framework of integrated resource management which requires consideration of a range of impacts, sometimes extending far beyond the immediate hydrological system, and over considerable time periods. Space-borne multi-spectral measurements have in some cases replaced ground-based observations and in others complemented at varying levels. Improved spatial, spectral and temporal resolution data from planned missions, namely IRS-P5 (Cartosat-1)/IRS-P6 (RESOURCESAT), ORBIMAGE, Resource-21 and currently operating Ikonos-2 will lead to a quantum jump in our capabilities for remote-sensing application towards water resources management. To complement these developments in the space segments, significant improvements in the methodologies of data processing, analysis, interpretation and integration are necessary for water resources management.

REFERENCES Anonymous (1999) EOS Science Plan. In: M.D. Kind, R. Greenstone and W. Bandeen (eds), The State of the Science in the EOS Programme. NRSA Publication No. NP-1998-12-069-GSFC. Dhruva Narayan, V.V. and Ram Babu (1983) Estimation of soil erosion in India. Journal of Drainage Division (ASCE), 109(4), 419–434. Obasi, G.O.P. (1994) WMO’s role in the International decade for natural disaster reduction. Bulletin of the American Meterological Society, 75, 1655–1661. Rao, D.P. (1999) Space and drought management in Asia – Pacific region. Space Forum, 4, 223–247.

170

D.P. Rao

Reddy, P.R. (1987) Geological and geomorphological studies through remote sensing. Proceedings of the Workshop on Remote Sensing Technologies, Malaysia, 21–26 June. Reddy, P.R. (1991) New concepts and approach for ground water evaluation with special reference to remote sensing. PhD Thesis, Osmania University, Hyderabad, India. Reddy, P.R. (1999) Remote sensing in ground water studies. In: D.P. Rao (ed.), Remote Sensing for Earth Resources. Hyderabad, India: AEG. Shiklomanov, I.A. (1998) Water use in the world, present situation and future needs. Prepared in the FrameWork of International Hydrological Program, UNESCO. UNICEF (1998) Emerging Fresh Water Crises in India.

CHAPTER 11

Geospatial Information Technology in Watershed Management I.V. Murali Krishna Centre for Spatial Information Technology, Jawaharlal Nehru Technological University, Hyderabad 500 028, India

11.1

INTRODUCTION

This chapter describes how geospatial information technology (which is based on integrated application of remote sensing, GIS, GPS, etc. technologies) could be used in watershed management, using the Veligonda Irrigation project in India, as a case. The geospatial information technology covers a wide range of topics, disciplines, subjects, and technological tools. These are concerned with surveying, Geographical Information Systems (GIS), Global Positioning Systems (GPS), Computer Aided Drafting (CAD), cartography, transportation management, construction management, water supply and sanitation, photogrammetry, remote sensing, terrain mapping and visualization, surface and groundwater resources management, urban and rural planning, coastal zone management, database management, software engineering, web technology, computer graphics, pattern recognition, expert systems, and image processing (Gregory and Walling, 1973; Pimentel, 1993; Burrough, 1998).

11.2

WATERSHED MANAGEMENT

Watershed can be defined as the area drained by a stream or a system of streams such that all the surface runoff originating in this area leaves the area in a concentrated flow through a single outlet. A watershed, which is a manageable independent hydrological unit, can be taken as the basic unit of development planning. In hydrological terms, watershed is an independent unit where runoff from the area drains off from a common outlet. The surface runoff follows the general slope and comes out through a single point. Watershed gives an idea about the available water in an area. Another advantage with watershed as a unit is that even a geographically large watershed in a much-damaged state can be rectified part-by-part by dividing the watershed into subwatersheds. Watershed is the ideal unit for management of the natural resources for sustainable development. The concept of watershed as a basic unit does not mean that the administrative boundaries should be on watershed basis, but it calls for planning natural resource management on a watershed basis. Ideally, the implementation could be on watershed basis. Each watershed has its own characteristic features and problems, thus making it unique. As each watershed or sub-watershed is an independent hydrological unit, any modification of the land used in the watershed or sub-watershed will reflect on the runoff as well as sediment yield of the watershed. The main objective of the watershed management is proper utilization of land as per its capacity and limitations, including prevention of soil erosion and designing methods for groundwater recharging. This helps in maximizing productivity per unit area of land under the given constraints through integrated approach and better farming systems.

172

I.V. Murali Krishna

11.3

REMOTE SENSING

During the last three decades, there has been a phenomenal improvement in the resolution of satellite data – for instance, as against the spatial resolution of about 80 m in multispectral mode for Landsat in 1972, the IKONOS satellite operated by Space Imaging, provides data at 2.5 m resolution in multispectral mode and 1 m resolution in panchromatic mode. Quick Bird satellite operated by Digital Globe achieved a resolution of 0.6 m. The Indian satellite, Cartosat, has a resolution of 2.5 m. High-resolution satellite data is making it possible to generate high-resolution ortho photo mosaics which are required for water resources planning. The cost-effectiveness of using satellite imagery could be judged from the comparative analysis of costs between conventional and remote sensing methods (Table 11.1; Perumal, 1994). Though the actual costs are applicable to India, and may have changed during the last 10 years, there is little doubt that the cost differential remains valid virtually in any part of the world. The high-resolution data helps in the preparation of large-scale maps which are useful for irrigation projects planning and management, urban applications, land degradation studies, agriculture inventory, etc. For example, identification of existing or potential degraded lands or erosion prone lands helps in planning, reclamation, or preventive measures. Methods of erosion detection and assessment using remote-sensing techniques are based on the recognition of tone, texture, and physiography of the features. Drainage, precipitation, vegetation, elevation, and relief are the factors to be considered in estimating the water-induced erosion. Soil erosion features may occur in a regular sequence of types and intensities along the topographic relief in a certain area. The remote-sensing imagery helps in the delineation of large-scale degraded lands. The image interpretation takes into consideration the direct and indirect features that are identifiable at a given scale. The remote-sensing image helps in identification of erosion features, which are qualitatively apparent, say, as changes in soil type, color, vegetation cover, etc. Seasonal changes in vegetation cover and moisture levels, as indicated by tonal changes, can be evaluated to increase the accuracy of interpretation of terrain conditions. In addition, any changes that occurred in the watershed during the year such as clear cutting of forest, forest fire or developmental works could be delineated. Interpretive elements of different terrain features on the multispectral satellite images of different available satellite sensor data facilitate the mapping of erosion prone areas. Land capability analysis is a very important component of watershed management. Land capability is an expression of the effect of physical land conditions, including climate, on the total suitability for use without damage to crops that require Table 11.1. A comparative analysis of the cost in INR (and equivalent US cents) per hectare for carrying out integrated resources mapping (1 USD is roughly equivalent to INR 50).

Theme Hydro-geomorphology Land use–land cover Soil Transport network Drainage Meteorological data Socio-economic data Action plan Total

Conventional survey/mapping INR (US cents) 1.30 (2.6) 1.00 (2) 2.00 (4) 0.05 (0.1) 0.05 (0.1) 0.01 (0.02) 0.07 (0. 14) 0.54 (1.08) 5.02 (10.04)

Based on remote-sensing data INR (US cents) 0.31(0.62) 0.22 (0.44) 1.10 (2.2) 0.05 (0.1) 0.05 (0.1) 0.01 (0. 02) 0.07 (0.14) 0.54 (1.08) 2.35 (4.70)

Geospatial information technology in watershed management 173 tillage, for grazing, for woodland and for wildlife. In short, land capability is a measure of suitability of land for use without damage. Remote sensing data when used with conventional data can be an effective tool to arrive at land capability classes. The land capability classification proposed by USDA is an extremely useful method and is widely used with some local variations. 11.4

GEOGRAPHICAL INFORMATION SYSTEMS

GIS happens to be the prime component of spatial information technology. It is a computerized information system with unique facilities for inputting, analyzing, querying, and managing the spatial and non-spatial data. GIS helps to handle digital maps, analyze them, and suggest management strategies. Thematic maps prepared on the basis of remote-sensing data are integrated with non-spatial information under a GIS. A GIS can generate two- or three-dimensional images of an area, showing such natural features as hills and rivers besides artificial features such as roads and power lines. Scientists use GIS images as models, making precise measurements, gathering data, and testing ideas with the help of the computer. Many GIS databases consist of sets of information called layers. Each layer represents a particular type of geographic data. For example, one layer may include information on the water bodies or cropping pattern in an area. Another layer may contain information on the soil in that area, whereas another records elevation. The GIS can combine these layers into one image, showing how the water bodies or cropping pattern and elevation relate to one another. Watershed managers might use this image to determine whether a particular part of a water body or bund is more likely to deteriorate. A GIS database can include as many as 100 layers. A GIS is designed to accept geographic data from a variety of sources including maps, satellite photographs printed text, and statistics. GIS sensors can scan some of this data directly – for example, a computer operator may feed a map or photograph into the scanner, and the computer ‘reads’ the information it contains. The GIS converts all geographical data into a digital code, which it arranges in its database. Operators program the GIS to process the information and produce the images or information they need. Most of the spatial information tools have no facilities for database development. The conventional database packages provide access only for data, which is in a tabular form. A comparison of related information systems in various conventional packages like CAD, DBMS, MIS, and GIS is given in Figure 11.1. This provides a basis for understanding the extent of respective spatial and non-spatial capabilities.

Spatial data

Related information systems

Low

Figure 11.1.

AM/FM CAD

Integrated GIS

Transaction processing systems

Database management systems

High

Attribute data

A comparative study of spatial and attribute data in related information systems.

174

I.V. Murali Krishna

In a GIS the spatial analysis and mapping components are very high. The topographic maps prepared by Survey of India on different scales from 1 : 25,000 to 1 : 250,000 are used for base map preparation and the thematic information is superimposed on these maps. The remotesensing data facilitate preparation of these thematic maps considering the satellite sensor resolutions as available today. The thematic information, which is in geospatial form is integrated with non-spatial or attribute data. This sort of integration is the prime scope and intent of development of any geospatial information system. The concept behind such an integration is shown schematically in Figure 11.2. A classification of GIS based on its mapping and analysis capabilities can be seen in Figure 11.3. Land information systems (LIS) primarily deal with cadastral data and provide information about land ownership and related details. Automated cartography has a high component of mapping capabilities, whereas spatial analysis could provide only statistical significance of any

Geographical information system

Base map for georeferencing

Geospatial and non-spatial data – thematic maps text data

Figure 11.2.

The relationship between GIS and base map/data.

Mapping

Classification of GIS

Automated cartography

Integrated GIS

Land information systems

Spatial analysis

High

Low Analysis

Figure 11.3.

Classification of GIS with reference to mapping and analysis capabilities.

Geospatial information technology in watershed management 175

Data analysis software Vector data (maps) Hardware

Raster datasatellite images or scanned images or photographs/ maps

Data output

Input

Data storage (database)

Planning and implementation/ monitoring

Tabular non-spatial data

Figure 11.4.

Concept of GIS.

attribute variation. A very special blend of spatial analysis capabilities for automated cartography leads to the development of an integrated GIS. The GIS basically consists of four groups of functions, namely, maintenance and analysis of spatial data, maintenance and analysis of non-spatial data, integrated analysis, and output generation. The scanning and/or digitization are the prime means of data input to a GIS. This involves what is currently called as data conversion or raster to vector conversion (R2V). The remote-sensing data in raw or classified mode and photogrammetry form the automatic inputs. The GIS package must have specific analysis capabilities for map overlay and intersection, proximity analysis, network analysis, attribute analysis and merging, map projections, data clipping, data updating, and data aggregation. In the context of discussion of the salient features of GIS, the basic concept of geographic information system is presented schematically in Figure 11.4. In the initial stages, all applications along with database are built with GIS occupying a center position. But the experience of GIS development has shown that GIS along with applications revolve around a database, which occupies a central position. In the networking era the web-based developments are expected to occupy a very important place. The developments in information technology have significantly altered the cost scenario. A schematic representation of variation of the cost of various GIS components over the period of late 1960s to the present time is shown in Figure 11.5. The ability for spatial and non-spatial querying is an essential component of GIS. The display and reporting capabilities in the form of tables, maps, and charts provide the required user interface. The data which is at the heart of GIS must be of high quality. The components of GIS data quality consist of micro-level components, macro-level components, and usage components. Table 11.2 gives details about the elements of these data quality components. 11.4.1

GIS packages

There are a large number of GIS packages available in the international market – mainly from North America, Europe, and Australia. Some of the packages are as follows and the list is only illustrative and not exhaustive: ARC/INFO and ARC GIS, PAMAP, SPANS, IDRISSI, ILWIS Map Info, Microstation, Geo-media, Geoconcept, and so on.

176

I.V. Murali Krishna

Relative cost

GIS database development

Data management GIS software

Hardware

1960s

Figure 11.5.

2003

GIS project cost scenario.

Table 11.2.

Data quality components.

Micro-level components Location accuracy Attribute accuracy Logical compatibility Resolution

Macro-level components Completeness Time Lineage

Usage components Accessibility Direct and indirect costs

All the packages have proved their utility and depending on the technical requirements, vendor support and funds available, one can decide upon the package for procurement. As a simple example, some features of ARC INFO GIS package are given and details of any of the packages can be obtained from user manuals and by browsing the respective websites. ARC/INFO (ARC GIS) has been developed by Environmental Systems Research Institute (ESRI), Redlands, CA, USA (ESRI, 2002). ARC/INFO is a vector-based GIS package capable of handling both spatial and non-spatial data. It organizes geographical data using vector topological models (non-spatial data using relational models in a DBMS). Each vector is either a point feature or a vertex of an arc, each arc is either a line feature or one line for a polygon feature. The arc–node and polygon topology are organized to identify point, line, and polygon relations. The cartographic data are then linked to the attribute data through a link-item. ARC/INFO has a wide range of functionality which has been developed based on a tool-box concept where each function can be visualized as a tool and having a specific utility. Thus, based on the user requirement, a specific tool or function could be utilized. Database creation in ARC/INFO is possible through the process of digitization using the ARC Digitizing System (ADS) and the ARCEDIT module. The ADS is a menu-driven module for digitizing and to perform editing on spatial features. ARCEDIT is a powerful editing utility having capabilities for feature-based editing. These modules include the functions to coordinate entry using different devices such as, digitizers, screen cursors, etc.

Geospatial information technology in watershed management 177

Vector data model






Figure 11.6.

Points . . . Lines Polygons

Vector data model.

INFO is a complete relational database manager for the tabular data associated with geographic features in map coverage. ARC/ INFO keeps track of and updates map feature attribute tables which are stored as INFO data files. INFO can be used to manipulate and update each feature’s attribute by performing logical and arithmetic operations on the rows and columns of the table. INFO provides facilities for data definition of data files, use of existing data files data entry, and update, sort, and query. ARC/INFO offers spatial overlay capabilities based on topological overlay concepts. Overlays, buffer generation, proximity analysis, feature aggregation, feature extraction, transformation, nearness functions, and other integration utilities are available. ARCPLOT module has got capabilities for generating cartographic quality outputs from the database. This includes utilities for interactive map composition, editing map compositions, plotting and printing, etc. The map composition functionalities include the incorporation of coverage features as per required scale, generalization, symbolization, transformation, etc. Placements of non-coverage features like legends, free text, logos, graphic shapes, and so on can also be done. Triangulated Irregular Network (TIN) module of ARC/INFO can be used to create, store, manage, and perform analysis pertaining to the three-dimensional data. The modeling capabilities include calculation of slope, aspect, iso-lines or contouring, range estimation, perspectives, volumes, etc. Additional functions for determining spatial visibility zones and line of sight are also provided. Network module of ARC/INFO performs two general categories of functions – network analysis and address geocoding. Network analysis for optimal path determination and resource allocation analysis is possible. Geocoding module allows for associating addresses to line networks and determining the spatial framework of addresses in an application. There are several other modules with advanced concepts of data base, web application, and open GIS, the details of which are not covered here. The data models to represent the real world scenario describe various features of a GIS package. In practice, the data models are vector data models, raster data models, and TIN. The GIS software packages like ARC/INFO uses the vector data model as shown in Figure 11.6. The details of these data models of ESRI are given in the following text. A vector data model consists of points, lines, and polygons to represent any feature on the surface of the earth. The method of representation of vector features in terms of x, y coordinates is given in Figure 11.7. The vector representation is nearer to real life situation. The topology, which explains the neighborhood of any point, line, or polygon, is shown as an example in Figure 11.8. The raster or grid representation for a typical earth surface is shown in Figure 11.9. These examples are described in detail in ARC/INFO ESRI publications. 11.5

GIS AND WATERSHED MANAGEMENT

The watershed has emerged as a convenient planning unit for development and management programs. Maintaining the soil fertility is one of the aims of watershed management programs.

10 8, 8 8

Line

5, 7 Road

6

1, 5

9, 5

6, 5

3, 5

Polygon

4

Lake Building

2

11, 3

8, 2

.

Point 2

Figure 11.7.

4

6

8

10

Vector data structure.

Points +2

x, y coordinates

Point number

+3

1 2 3 4

+4 +1

2, 2 3, 6 5, 5 6, 3

Lines (arcs)

2

Polygon number 1 2

1 2

A

1

B

8

1, 5 3, 6 6, 5 1, 1 3, 3 6, 2

6

C 9

E

1, 4 1, 5 2, 7 5, 7 4, 4 1, 4 1, 2 2, 3 4, 3 5, 4 7, 5 7, 3 6, 1 3, 1 1, 2 Polygon-arc topology

Polygon

ARC list

B

1, 6, 8, 5

C D E F

2, 4, 9, 6 3, 5, 10, 4 7 8, 9, 10, 0, 7

F D

10

4

3

Figure 11.8.

7, 6 7, 3

x, y coordinates

2

7 5

x, y coordinates

Line number 1 2

1

Topology for point, line, and polygon (see Color Plate XLII).

Geospatial information technology in watershed management 179

Point features represented in a grid

Linear features represented in a grid

Discrete area features represented in a grid

Figure 11.9.

Grid/raster representation of point, line, and area features (see Color Plate XLIII).

Therefore, the amount of soil loss and the causes of soil erosion must be monitored. Geographic information system is a useful tool for estimating soil erosion effectively. In the present study universal soil loss equation has been incorporated into GIS to calculate soil erosion. The workflow for GIS application based on satellite data and field observations and empirical model development is given schematically in the Figure 11.10. These efforts require generation of micro-level land use and geomorphic features as layers. The existing data on land and water features are obtained from related records and converted to digital form to be used in conjunction with thematic maps of resource data. The micro-level land use features and other infrastructure are then verified on the ground and marked on the map to generate a comprehensive information source. The GIS has unique features to relate the point, linear, and area features in terms of the topology as well as connectivity. For example, there could be two microwatershed areas that are adjacent to each other or there could be a feature that requires separate delineation for identification or exclusion. In the context of this kind of situation, the area–area relationships as given by ARC-INFO GIS software would be of immense use. These kinds of relationships are shown in Figures 11.11 and 11.12. These figures include all possible combinations such as watershed boundaries along with linear features like drainage or road and point features such as water outflow/inflow points.

Remote sensing imagery

Ground control survey using high accuracy geodetic GPS receivers and post processing

Geometric correction of imagery

Planimetric data capture using feature collection software Universal soil loss equation

Field verification data collection

Runoff model Vegetation index

Incorporation of field data

Final map preparation

Figure 11.10.

Workflow diagram of the GIS layer creation approach.

Area–area relationships

(a)

(b)

(a) Two watersheds adjacent to each other (b) A microwatershed feature within a higher order watershed Area–line relationships

(a)

(b)

(c)

(d)

(e)

(f)

(a) A linear feature within a watershed polygon. (b) A linear feature ends at a watershed polygon. (c) A linear feature ends in a watershed polygon. (d) A linear feature touches a watershed polygon. (e) A linear feature intersects a watershed polygon. (f) A linear feature acts as a border of a polygon.

Figure 11.11. Typical examples of GIS topology build up for watershed features for area–area and area–line features.

Geospatial information technology in watershed management 181

Point–line relationships

(a)

(b)

Point–area relationships

(c)

(d)

(a) A point feature on a stream like feature. (b) A point feature surrounding/beside a stream like feature. (c) A point feature within a watershed polygon. (d) A point feature on border of a polygon.

Figure 11.12. Typical examples of GIS topology build up for watershed features for point–line and point–area features.

11.6

WATERSHED CHARACTERISTICS AND GIS DATABASE CREATION

The important watershed characteristics identified for the purpose of defining and managing are as follows. 11.6.1

Size

A small watershed behaves differently from a larger one in terms of the relative importance of various phases of the runoff phenomenon. In small catchments the overland flow phase is predominant over the channel flow. Hence the land use and intensity of rainfall have important role on the peak flood. In a large watershed the channel flow phase is more predominant. Further, the larger the area, the greater will be the heterogeneity in the soils, vegetation, slope, land use, precipitation characteristics, etc. The size of watershed can be determined through GIS. First of all, watershed should be delineated by using contour and drainage map. 11.6.2

Shape

A watershed may have many shapes, for example, it could be rectangular, square, palm shape, oval, etc. The shape controls the length : width ratio which affects the runoff characteristics. For example, the longer the watershed, the greater is the time of concentration. The longer the time of concentration, the greater would be the time available for the water to infiltrate, evaporate, and get utilized by vegetation. The shape of a watershed may be described by a shape index, Sw: Sw ⫽

L L2 ⫽ W A

(11.1)

where Sw is the watershed index, L the length of the watershed along the main stream from the watershed outlet to the most distant ridge, W the average width of watershed, and A the area of watershed. Shape of watershed is the output of delineated watershed boundary map by using contour and drainage as input. 11.6.3 Slope This is a very important characteristic of a watershed. It affects the time of concentration, infiltration opportunity time, runoff, and soil loss. The average watershed slope in percent may be

182 I.V. Murali Krishna determined from a map by the following formula: S⫽

MN ⫻ 100 A

(11.2)

where M is the total length of contours within the watershed, N the contour interval and A the watershed area. For very small watersheds the average slope can be taken as the ratio of elevation difference between outlet and most distant ridge to average length of watershed. Slope map can be derived from Digital Elevation Model.

11.7

SOIL COVER COMPLEX

A combination of a specific soil and a specific vegetative cover is referred to as a soil cover complex and a measure of this complex can be used as a watershed parameter in estimating runoff. The hydrologic properties of a soil or a group of soils are an essential factor in the hydrologic analysis of watershed data. Soils can be classified according to their hydrologic properties if considered independently of watershed slope and cover. Four major soil groups are recognized for the primary classification of watershed soils. These are essentially based on infiltration rates (low, moderate, and high), drainage characteristics, soil depth, texture (fine, coarse, etc.), and rate of water transmission. Hydrologic soil group map can be created by reclassifying soil map. The soil map can be derived by using physiographic association method or by using remote sensing in conjunction with ground truth. In physiographic association method, aspect map, land use map, and elevation map are crossed and then the composite map is reclassified as soil map.

11.7.1

Vegetation cover

Vegetation influences water movement behavior significantly by affecting the infiltration rate, soil erosion, evapotranspiration, sediment production, etc. Remote sensing is a very effective tool to get this information. FCC can be divided into various land-use–land-cover situations. NDVI gives the vegetation intensity, which can also be used to assess the vegetal status of land.

11.7.2

Drainage

This is another important factor which influences the watershed behavior. The drainage density is defined as length of drainage channels per unit area. A large drainage density creates situation conducive for quick disposal of runoff down the channels. In the watershed where drainage density is small, overland flow will be predominant.

11.7.3

Climate

Climate parameters such as precipitation, humidity, temperature, wind, etc. affect the functioning of watershed. Their intensity, duration, and frequency greatly affect the watershed hydrology. These parameters are the input for hydrological models in GIS environment.

Geospatial information technology in watershed management 183 11.7.4

Time of concentration

This is defined as time taken for water to travel from the most distant point of a watershed to the watershed outlet. The following formula is generally used to determine the time of concentration. Tc ⫽

L1.15 7700H 0.38

(11.3)

where Tc is the time of concentration (h), L is the length of the watershed along the mainstream. This can be calculated through distance calculation in GIS environment and H is the difference in elevation between the watershed outlet and the most distant ridge (ft). H can be calculated through DEM. However, differences in elevation due to falls, rapids, or other sudden drops should be subtracted from H. Various watershed characteristics can be obtained from remote sensing toposheets and base map. Accurate prediction of runoff is difficult as it depends upon several factors. There are several methods such as rational method, SCS method, etc., developed after field observation, for estimating the maximum rate of runoff that could occur from a particular catchment. In the rational method, the peak rate of runoff is estimated through a model which is dependent on the intensity of rainfall (mm/h) for a duration equal to the time of concentration and for the given frequency, runoff coefficient, area of the catchment in hectares or acres. Runoff coefficient C is defined as the ratio of the peak runoff rate to the rainfall intensity. Values of C for different slopes and land-use conditions are determined from field observations. The value of the intensity of rainfall to be used in the empirical model should be calculated for the period equal to the time of concentration of the catchment. When the duration of rainfall equals the time of concentration, all parts of the catchment will be able to contribute to the discharge at the outlet and as such, the discharge will be maximum. The peak rate of runoff is to be determined with the help of GIS. The catchment characteristics may vary and this can be considered by assigning weightages for different parts of the catchment area. 11.7.5 Estimation of runoff The method that Soil Conservation Services (SCS) proposed in 1986 is perhaps the most commonly used method all over the world for estimation of runoff resulting from rainfall over small watersheds. To estimate runoff using SCS model following input GIS layers are required: ● ● ● ● ● ●

landuse map; soil series map; DEM for terrain information; watershed boundary map; rainfall map; curve number (CN) map.

The SCS approach involves the use of simple empirical formulae and readily available tables and curves. The empirical equation requires the rainfall and a watershed coefficient as inputs. The watershed coefficient called CN is an index that represents the combinations of hydrologic, geomorphic, and land-use categories. The thematic map/information derived from remote sensing is used for estimating the land-cover distributions; hence these provide useful input support for the SCS model. During a given rainfall, water is continually being abstracted to saturate the upper levels of the soil surface; however, this saturation or infiltration is only one of many continuous abstractions. Rainfall is also intercepted by trees, plants, and roof surfaces, and at the

184

I.V. Murali Krishna

same time it also gets evaporated. Once the rain falls and fulfills initial requirements of infiltration, natural depressions collect falling rain to form small puddles creating depression storage. In addition, numerous pools of water forming detention storage build up on permeable and impermeable surfaces within the watershed. This stored water gathers in small rivulets, which carry the water originating as overland flow into small channels, then into larger channels, and finally as channel flow to the watershed outlet. Remote sensing and GIS play a very important role in watershed management. With the help of remote-sensing data, change in the land use/land cover can be detected. The high-resolution data which is available in panchromatic mode can be used to increase the accuracy of land-use/land-cover classification. Digital elevation model provides the perspective view of watershed. Slope and aspect map can be created by taking DEM as input in a GIS environment. The overlaying of drainage map over DEM and land-use maps helps in the identification of sites for the location of water harvesting structures. 11.7.6 Estimation of soil loss The most commonly used method for the estimation of soil loss is Universal Soil Loss Equation (USLE). The essence of the USLE is to isolate each variable and reduce its effect to a number so that when the numbers are multiplied together the result is the amount of soil loss. The equation takes into account the influence of the total rainfall energy for a specific area rather than rainfall amount. Total rainfall energy can be readily computed for localized areas from existing weather bureau data. The USLE predicts the longtime average soil losses from a specified land in a specified cropping and management system. The equation predicts only the losses from sheet and rill erosion under specified conditions. With appropriate selection of numerical values for various soil erosion variables, the equation will compute the average soil loss for a cropping system, for a particular crop year in a rotation, or a particular crop stage period within a crop year. It computes the soil loss for a given site as a product of six major factors whose most likely values at a particular location can be expressed numerically. The USLE can be presented as: A ⫽ R ⫻ K ⫻ LS ⫻ S ⫻ C ⫻ P

(11.4)

where A is the average annual soil loss in tons/acre, R is the rainfall factor, which is a measure of the erosion force of the rain, K is a soil erodibility factor; the erosion rate per unit of erosion index for a specific soil, LS is slope length factor, that is, the ratio of soil loss from the field slope length to that from a 22.1 m length on the same soil type, S is slope gradient factor, that is, the ratio of soil loss from the field gradient to that from a 9% slope on the same soil type and slope length, C is a cropping and management factor, and P is the supporting conservation practice, such as terracing, strip cropping, and contouring. The rainfall erosion potential, R-factor, can be determined by using the following formula, R ⫽ ⫺8.12 ⫹ 0.562 ⫻ Annual rainfall

(11.5)

Rain gauges installed at various meteorological observatories give depth of rainfall at that place. This point information can be converted to spatial distribution by Theisean polygon method in GIS environment. Once this Theisean polygon map is derived, then by using the above formula, the R-map can be drawn. The soil erodibility factor, K, defines the inherent erosion potential of the soil. It is expressed as the soil loss in tons per acre for each unit of rainfall erosion index for the locality and for continuous fallow tillage on a 9% slope, 22 m (73 ft) in length. Standard K values have been established for only a few soil types in the country where actual soil loss measurements have been made. Research now in progress may eventually provide criteria for estimating erodibility.

Geospatial information technology in watershed management 185 Numerous factors influence erosion of cohesive soils including but not limited to, texture, grain-size distribution, nature of clay minerals, thickness, and permeability of strata and organic content present. Estimates of soil erodibility must be made on the basis of know erosion characteristics of the soil. K ⫽ M1.4 (10⫺4) (12 ⫺ a) ⫹ 0.25(b ⫺ 2) ⫹ 2.5(c ⫺ 3)

(11.6)

where M is the particle size parameter, a is the percent organic matter, b is the soil structural code used in soil classification, and c is the profile permeability class. The slope length steepness factor or soil loss ratio, LS, is determined by dividing the existing length and steepness by the standard 9% slope, 22 m (73 ft) in length. The cropping management factor, C, is the expected ratio of soil loss from land cropped under specified conditions of soil loss relative to clean tilled fallow, on identical soil and slope and under the same rainfall. This item reflects the combined effect of crop sequence, productivity level, length of growing season, tillage practices, residue management, and the expected time distribution of erosive rain storm with respect to seeding and harvesting parameters in the locality. This factor is the most complicated because there is an almost infinite number of different ways of managing the growing of crops. In the early system, such as the slope practice equation, a single value for this crop factor was used to give the average effect over the whole season. The product of RKLS in the USLE computes the soil loss that would occur under fallow conditions. Actual loss from the cropped field is usually much less than this amount. The reduction depends on the particular combination of land cover, crop sequence and management practices. It also depends upon the particular stage of growth and development of the vegetal cover at the time of rain. C adjusts the soil loss estimate to suit these conditions. Based on the data from runoff plots of various sizes established in different agro climatic areas of the country, a preliminary evaluation of the factor C has been made. Although a number of crops have been studied but the detailed data of soil loss for various crop stages is rarely available in the published record. In the absence of crop stage growth period data, average value of C of the crop has been determined in a number of cases based on the total seasonal soil loss data.

11.8

WATERSHED CHARACTERIZATION OF UNITS UNDER VELIGONDA IRRIGATION PROJECT – A CASE HISTORY

11.8.1 Biophysical setting The upland areas of Prakasam, Cuddapah, and Nellore districts in Andhra Pradesh, India, have been identified as chronically drought affected areas. The Government of Andhra Pradesh has conceived and planned a project to provide relief against drought and famine in the semi-arid, famine-affected areas of these three districts by supplying 43.5 TMC of floodwaters, from river Krishna, which are expected to be the surplus flows of the Srisailam reservoir. This project is called Veligonda Irrigation Project. The project can be constructed as a network of canals, tunnels, and reservoirs feeding each other and the command area. The Veligonda Project proposes to provide irrigation facility to 1.772 lakh ha. (4.38 lakh acres) and drinking water to about 15 lakh people in the districts of Prakasam, Cuddapah, and Nellore districts (one lakh ⫽ 100,000). The estimated gross command area under the project is 4.06 lakh ha. and the cultivable command area is 2.40 lakh ha. The ayacut proposed is 1.772 lakh ha. The ayacut is spread over 29 mandals (county) in the three districts. The construction of reservoirs may lead to submergence of settlements, agricultural lands, and forest areas and it is important to ascertain the ecological impact of this situation. The life span of the reservoir depends on efficient management of the

186

I.V. Murali Krishna

vegetation cover and soil cover in the catchment. In this context it is necessary to make a precise estimation of these effects and plan for compensatory afforestation and socio-economic rehabilitation. 11.8.2 Preparation of maps The remote-sensing technology due to its synoptic, repetitive, and multispectral coverage facilitates mapping of natural resources information on vegetation, land cover, land use, geology, hydro-geomorphology, and soils. The specific objectives of the study are as follows: ● ●

● ● ●

●

to map surface water bodies and drainage on 1 : 50,000 scale for the catchment areas; to map land use/land cover, hydro-geomorphology, and slope on 1 : 50,000 scale for the catchment areas; to estimate sediment yield and submersible areas under reservoirs; to prepare catchment area treatment plans on 1 : 50,000 scale; to map land use–land cover, groundwater potential, and soils maps on 1 : 250,000 scale for the command area; to integrate these thematic maps as layers under a GIS environment.

The following data were used for thematic mapping on 1 : 50,000 and 1 : 250,000 scales: ●

●

Geocoded Imagery of Indian Remote Sensing Satellite (IRS-1B/1C-LISS II/III) for February–March 1999 and October 1992. Survey of India topographical maps on 1 : 50,000 scale and on 1 : 250,000 scale.

The streams drained into the reservoirs have been demarcated based on Survey of India toposheets on 1 : 50,000 scale. The base maps for the study area are prepared showing major roads, settlements, rivers/tanks, and forest boundaries on 1 : 50,000 scale. The base maps are superimposed on IRS-1B LISS-geocoded (FCC) products at 1 : 50,000 scale and various land use/land cover categories, namely deciduous forest, open forest, degraded/scrub forest, wastelands such as land with scrub, land without scrub, irrigated, unirrigated, double crop, and water bodies were delineated and mapped using visual interpretation technique. The areas of various categories were calculated and statistics are presented. The submersible areas under each reservoir were mapped and unit wise land-use/land-cover areas are estimated. The same procedure is adopted for command area mapping on 1 : 250,000 scale. The hydro-geomorphology mapping was made on both 1 : 50,000 and 1 : 250,000 scales. The land use/land cover analysis for the five reservoir catchments was carried out with a view to assess baseline status of land use–land cover of the catchments and submersible areas to account for the ecological conditions. These maps have been used to study the sediment yield in the catchments and to prepare catchment area treatment plans. The prime aim of catchment area treatment plans is to suggest suitable treatment measures that will reduce the rate of erosion in the directly drained catchment area and to retard the rate of siltation. The plan also includes areas identified for afforestation of the under-stocked areas and increasing the density of existing vegetation cover. This would help the maintenance of environmental balance, preservation of bio-diversity and maintenance of the gene pool in the existing area and bringing about eco-development of the area. 11.8.3

Land-use/land-cover study

The land use/land cover, slope, soil, and drainage maps along with rainfall data were taken into account for the preparation of treatment plan on 1 : 50,000 scale. The area of each unit, which is proposed for the treatment, has been estimated and presented. The treatment plan comprises essentially of two components such as biotic treatment with soil and moisture conservation measures and engineering and gully control treatment works. The areas identified for the biotic

Geospatial information technology in watershed management 187 treatment with soil conservation measures primarily fall under the land without scrub, land with scrub, open forest, degraded/scrub forest and kharif (first crop in the agricultural season) unirrigated lands. Any biotic treatment for land with scrub and land without scrub is mainly to conserve soil and water, and minimize soil losses from erosion using plants, shrubs, and legumes/grasses, and afforestation in the gap areas with social forestry or horticultural crops. All the wastelands that do not qualify for placement under any of the wasteland categories outside the forest area are included under this class. Besides scrub lands, other lands with soils that are too shallow, skeletal, or otherwise chemically degraded, lands with extremes of slopes, severely eroded lands, and lands subject to excessive aridity, etc., are placed in this group, earmarked for biotic treatment for land with and without scrub. Biotic treatment for deciduous open and scrub forest/degraded forest is for rejuvenation and regeneration of degraded forests and afforestation in the gap areas for eco-restoration. To retard siltation from directly drained areas into the reservoir, the lands situated below 0.6 m (2 ft) and above 0.6 m of the FRL have been delineated from satellite imageries and recommended for green belt plantation. 11.8.4

Gully control measures

The gully control works/engineering treatment measurements are recommended in the treatment areas mainly to control the sediment discharge from the catchments as well as to increase the ground water recharge. Engineering measures include the construction of hydraulic structures like check dams and water harvesting ponds, gully plugs, bank protection, etc. Five types of such works, namely brush wood dam, gully plugs, rock fill dams, check dams, minor tanks, and stream bank stabilization are proposed in catchments based on the details inferred from satellite imageries and field observation. The soil map prepared using satellite imageries for the project area on 1 : 250,000 scales provide information on soils of the area. The study provides baseline data and action plan that would help the field officer to implement schemes for reduction of rate of siltation of reservoir under appropriate management strategies. The study also demonstrates the utility of remote sensing technology for command area development and catchment area treatment. The remote-sensing data can also be used to derive engineering parameters. As an example, calculation of the vegetative cover is given which will be required for estimation of some of the parameters like vegetative cover to estimate the sediment yield of the catchments. The details of land use/land cover derived from remote-sensing data are given below as examples of a typical case of application: Kharif unirrigated Kharif irrigated Deciduous forest (dense) Deciduous forest (open) Degraded/scrub forest Land with scrub

: : : : : :

123.975 km2 1.150 km2 176.000 km2 15.225 km2 29.850 km2 22.425 km2

The above parameters are regrouped to calculate the vegetative cover factor (Fc ) Fc ⫽ Vegetative cover ⫽ 0.2F1 ⫹ 0.2F2 ⫹ 0.6F3 ⫹ 0.8F4 ⫹ F5 The values of F1 to F5 are calculated from the corresponding thematic maps. F1 ⫽ Area under reserve and protected forest ⫽ Deciduous forest (dense ⫹ open) ⫽ 176 ⫹ 15.225 ⫽ 191.225 km2 F2 ⫽ Unclassified forest area ⫽ 0 (zero) F3 ⫽ Cultivated area ⫽ Kharif unirrigated ⫹ Kharif irrigated ⫽ 123.975 ⫹1.15 ⫹ 0.25 ⫽ 125.375 km2

(11.7)

188

I.V. Murali Krishna

F4 ⫽ Grass and pasture land ⫽ 0 (zero) F5 ⫽ Wasteland ⫽ land with scrub ⫹ land without scrub ⫹ scrub forest ⫽ 29.85 ⫹ 22.425 ⫹ 5.9 ⫽ 58.175 km2 The digital thematic layers of land use help in the calculation of parameters, F1 to F5. Thus, the extent of vegetative cover can be estimated by integration of the values of F1 to F5. The performance of an irrigation project is judged on the basis of actual utilization of irrigation potential created and increase in agricultural production. The area commanded by the irrigation canal network has to be scientifically developed to receive the water. Basically, the prime task for field officers and engineers concerned with the command area development, envisages land grading and shaping. The land surface to be irrigated is expected to provide for controlled and safe flow over it. The land is to be graded to the required longitudinal and cross slopes. The land grading for irrigation should meet both grade criteria and the grading criteria. The former aims at efficient application of water in the field such that the deep percolation (beyond the root zone) and the surface runoff are minimized and uniform wetting of the soil profile is attained. The grading criteria on the other hand is concerned with minimum earthwork, appropriate cut fill ratio, and limited cuts so as not to expose the subsoil. The land grading is a costly affair. In order to minimize the cost, existing topography should be exploited to the maximum extent. The basic approaches in land grading and field layout on gently sloping lands, with the fields laid perpendicular to the contour lines, will provide the necessary down field gradient for irrigation. On strongly sloping lands the benches are constructed across the slopes on which the graded fields are laid out. The farm irrigation system in the outlet commands should consist of field channels, lined or unlined, along with control structures to convey the water to the fields. A farm irrigation system should have counterpart farm drainage system for removal of excess water to prevent water logging and salinity. The farm drainage system can be either a surface drainage system comprising open drains or sub surface drainage system comprising tile drains or open drains. The surface drainage removes the excess rain or irrigation water from land surface, while the latter keeps the water table below the root zone. Farm irrigation structures are used for water control from the outlets. They are needed to divide flow of water (division structures), control velocity (drop structures), deliver water without erosion (check structures), measure water (weirs, flumes), and deliver correct amount of water (turnouts, siphons, valves, and gated pipes). Inadequate distribution of water can be corrected through the use of irrigation structures to improve water use efficiency. The benefits derived from properly installed irrigation structures would be in the form of reduced erosion on sloping lands, control of amount of water being delivered to fields, reduction in water course deterioration, and improvement in equitable water distribution among the farmers. It helps control of waste and seepage water and reduction in water logging and salinity. The area commanded in the present study is gently sloping. The on-farm development warrants systematic land development and construction of bunds, provision of surface drains, improvement of existing drains, construction of irrigation structures, training of farmers in irrigated agriculture, and provision of credit facilities through banks to enable the farmers to purchase inputs. 11.8.5

Conclusion

The basic issue for any application related to GIS essentially is concerned with input data. The input data is in both spatial as well as non-spatial form. Maintenance and analysis of spatial data, maintenance and analysis of non-spatial data, and integration of spatial and non-spatial data are the three major components of any effort related to development of GIS for watershed management.

Geospatial information technology in watershed management 189 The issues related to irrigation and command area of the development of the Veligonda Irrigation Project are described primarily to draw attention to those measures that can be handled by remote-sensing data and integration with collateral data so as to develop a GIS for command/catchment area development. REFERENCES Burrough, P.A. (1998) Principles of Geographic Information Systems (2nd edn), Oxford University Press: Oxford. Environmental Systems Research Institute (ESRI) (2002) ARC/ INFO/ARC/GIS Manual, Redlands, CA. Gregory, K.J. and Walling, D.E. (1973) Drainage Basin Form and Process: Geomorphological Approach. London: Edward Arnold. Perumal, A. (1994) Watershed characterization. Integrated Mission for Sustainable Development (IMSD), Report of the National Remote Sensing Agency, Hyderabad, India. Pimentel, D. (ed.) (1993) World Soil Erosion and Conservation. Cambridge: Cambridge University Press.

Part 3

Water resource management Case histories

CHAPTER 12

Runoff Agroforestry P.R. Berliner Wyler Department of Dryland Agriculture, Blaustein Institute for Desert Research, Ben Gurion University of the Negev, Israel

12.1

INTRODUCTION

This chapter deals with ways and means of increasing and stabilizing yields through the use of water harvesting techniques, illustrating them with a case in Kenya. Arid and semi-arid lands (ASAL) occupy approximately one-third of the land surface of our planet and are settled by around 800 million inhabitants. These lands are characterized by low and highly variable rainfall, and the annual precipitation is much lower than the annual evaporative demand of the atmosphere (Lovenstein et al., 1991). The latter is met only during extremely brief periods that occur immediately after rain events. The interannual variability in the total annual precipitation is typically very large (Bruins and Berliner, 1998). In Figure 12.1 these features are presented for a farm located at the border of the arid zone in Israel. This area is characterized by Mediterranean climate with precipitation occurring during the period October–March. In addition to the large interannual variability there is an additional complicating feature, which is typical of this region, namely, the date of the onset of the rainy season. It varies as well from year to year, as a result of which the length of the “dry season” varies in an unpredictable fashion. Owing to the factors mentioned above (low rainfall being the main one) primary productivity is extremely low in these areas. These areas which can only sustain a very low population density have been traditionally exploited by pastoralists. During the recent decades, improvement in health care and a decrease in mobility due to the establishment of national boundaries resulted in an increase in population and a mounting pressure on the local resources (Droppelmann and Berliner, 2003). Almost inevitably, overgrazing and the indiscriminate felling of trees and shrubs occur. Both trees and shrubs constitute an essential resource, as firewood is usually the main fuel source for arid-zone dwellers (Sauerhaft et al., 1998). As a consequence of overgrazing and/or tree felling, the soil surface is left bare and during the first rains the impact of the rain drops destroys the surface aggregates and a dense crust is formed (Agassi et al., 1985). The presence of the crust decreases the infiltration of water into the soil and runoff is produced. If the water is left to flow unchecked its speed increases and the flow becomes turbulent, entraining the detached soil surface particles. If the area on which the runoff is generated is large and the slope moderate to high, a flood develops. After the flood subsides the eroded material covers the fertile upper layers of the soil in the flooded areas and agricultural activities are impaired. In the area generating runoff water does not infiltrate into the soil, dramatically decreasing the water available to plants and limiting therefore the regrowth of annuals, trees, and shrubs, in addition to the loss of fertile soil. In an agronomic sense unchecked floods are therefore completely negative events. The conventional approach to overcome the generation of floods is to ensure the existence of a perennial green cover, which minimizes runoff and erosion. To this effect the cover crops need to be irrigated using water whose source may be the local aquifer (if available), or water

194

P.R. Berliner

200 180

Total seasonal rainfall

160 140 120 100 80 60 40 20 0 1960

1965

1970

1975

1980

1985

1990

1995

2000

Year

Figure 12.1. Total seasonal rainfall (in mm) for the years 1962–2000 recorded at the Evenari Farm at Avdat (31° 08⬘ N, 34° 53⬘ E; elevation 40 m a.m.s.l.). Absolute maximum, minimum, and average (83.5 mm) are presented.

imported from other regions. Both approaches are capital and energy intensive and require a well-developed infrastructure. In the few cases in which the governments of developing countries decided to develop their arid zones a policy of monumental projects has been implemented. Importing water from far away regions through very long channel systems has been one of the favourite options (e.g. Indira Gandhi Canal in Rajasthan, India; Karylkum Canal in Turkmenistan). The effectiveness of this type of projects has to be assessed in the context of the benefits to the local population, their cost, sustainability, and potential damage to the environment their misuse may cause. Usually the mammoth development projects fail in one or more of the categories mentioned above. In developing countries, arid areas are usually located far away from the centers of commerce and industry and have, due to the low population density and scarcity of resources, very limited political influence. The sustainable development of these regions has to be based therefore on locally available resources and cheap technology. It is therefore worthwhile to look back in time and see how ancient civilizations, which did not possess our technical knowledge, coped with similar situations. 12.2

HARVESTING OF RUNOFF WATER IN ANCIENT CULTURES

Archeological findings indicate that much of the arid zones of today were settled in the past. This is particularly true in the southern and eastern shores of the Mediterranean, but evidence of settlements has also been found in the southwestern United States and Central Asia, to name just a few. A view has been put forward that these settlements were possible probably due to higher precipitation during historical time, but even if this were to be the case, these areas would still be considered as arid and the natural precipitation would not have been enough for agricultural production. The common feature of these historic settlements is that they supplemented

Runoff agroforestry 195 the meager rain by collecting the runoff water in structures that were adapted to the geomorphology of the area (Evenari et al., 1982). The principle of this system is extremely simple and ingenious: concentrate the naturally generated runoff from a large area to a smaller one and trap it. The result is that not only will the equivalent depth of water in the receiving area be much higher than the rainfall depth but also sufficient water may be stored to allow the normal development of a crop, albeit on a reduced area. Even though the basic idea is simple, many technical problems are involved (Ben-Asher and Berliner, 1994). In the Middle East, the Nabateans are considered to be the pioneers of this technique. The Romans used all their considerable engineering skills to develop and improve the water harvesting systems that during their heyday were widespread throughout the southern and eastern shores of the Mediterranean. They conveyed the runoff by gravity from the area where it was generated (usually slopes of hills) to the plot in which the crop was to be planted. The latter was frequently located in the flatter areas in which soil depth was sufficient. Dams surrounded these plots, and the water was thus “trapped” in the plot. It infiltrated into the soil and was available to the existing trees or to the crop that was planted thereafter. The ponding depth was adjusted to the local soil depth and it was thus sufficient to stow the collected runoff “in situ.” The ratio of contributing to receiving area found for the loessial soils prevalent in the area is typically 20–30 : 1 (Ben-Asher and Berliner, 1994). This ratio may however vary considerably as it is affected by the geomorphological characteristics of the catchment in general, and by the length of the overland flow path in particular. There is no generalized model that allows its prediction from basic soil and catchment characteristics, and the runoff ratio is determined empirically for each set-up. Even though this fact is a serious drawback during the planning stages, it is offset by the advantages of the system (considerable increase in water availability to plants ease and cheapness of implement), which make it a prime candidate for the development of arid zones. During the last two decades the possibility of using an improved version of this technique to produce firewood and fodder has been explored in a series of large-scale field trials (Zohar et al., 1988). In particular, attention has been paid to the possibility of increasing the productivity and the variety of crops by planting an intercrop in between rows of trees (Lovenstein et al., 1991). The systems in which trees and intercrop are grown simultaneously or successively on the same patch of land are known collectively as Agroforestry Systems (Sanchez, 1995). They are used in the higher rainfall areas and to a lesser extent in the drier ones. Due to the fact that the systems that have been developed for arid zones are based on the irrigation with runoff water they have been termed Runoff Agroforestry Systems (RAS). In the following pages the salient features of this approach will be discussed. 12.3

RUNOFF AGROFORESTRY SYSTEM (RAS)

An RAS system is schematically depicted in Figure 12.2. The runoff is generated in an area, which for convenience, is located close to the area in which we intend to establish our dykesurrounded plot. The area of this contributing surface is adjusted in such a way that the volumes of runoff produced when spread over the cropped area, result in wetting the soil to a predetermined depth. The volumes of runoff produced are a result of the interaction between soil surface characteristics and rainfall properties. The latter cannot be modified but the surface of the soil can be treated in such a way that a stable crust is formed and water infiltration minimized. Weeding, removing stones that are not embedded in the soil and smoothing and compacting the soil surface are among the simple activities that can considerably enhance the volumes of runoff produced on a certain area (Evenari et al., 1982). An additional factor that affects the efficiency by which the rainfall is transformed into runoff is the overland path length. The larger this distance, the lower the efficiency and the higher the threshold rainfall necessary to produce

196 P.R. Berliner

P

E, T R W S

B

D

Figure 12.2. Schematic description of a runoff agroforestry system with P: precipitation, R: runoff water, B: cropping area, E, T: evaporation and transpiration, D: deep drainage, W: walls surrounding the cropping area, and S: spillway (adapted from Lovenstein et al., 1991).

runoff (i.e. to allow the collection of runoff water at the lower end of the plot) (Ben-Asher and Berliner, 1994). Splitting the runoff generating area into sub-catchments drained by secondary channels is a good approach to increase the overall runoff efficiency. The optimum size of the runoff-receiving plot is approximately 1 hectare (ha). This size is on the one hand the maximum that can be satisfactorily leveled by hand (necessary in order to ensure homogeneous water spreading) and allows, on the other hand, the rational cultivation of the land. Smaller plots spread over a larger area would force the farmer to move across long distances and spend therefore more time on all agricultural activities (sowing, weeding, lopping, etc.). The height of the spillway above the bottom of the dyke surrounded plot (Figure 12.2) determines the equivalent depth of water that may be ponded, and hence the wetted soil depth. The width of the spillway has to be such that during peak events the inflow of runoff water into the plot equals the flow through it without exceeding a critical height that is well below the crest of the dyke. Once the runoff event ceases, the water that remains trapped in the plot continues to infiltrate into the soil (the infiltration process commenced at the onset of the inflow of water into the plot). The rate of infiltration depends on the hydraulic properties of the soil and the antecedent moisture. In the loess soils of Israel’s Negev, the first flood is typically absorbed within a day. For subsequent floods that follow closely, the rate of absorption of water by the soil is considerably slower and may continue for various days. The depth of water penetration depends, in addition to the ponding depth, on the soil characteristics. Water can be lost from the wetted soil slab by three pathways: uptake by the crop (present or to be planted), direct evaporation into the atmosphere through the soil surface, and flow to the deeper soil layers and/or aquifer. It is of the utmost importance to reduce the last two, as they do not contribute to crop development, and can be viewed as unproductive losses (Lovenstein et al., 1991).

Runoff agroforestry 197 One option is to increase plant density in order to reduce the radiation flux at the soil surface thereby decreasing the evaporation from the soil surface. This increase in density will probably express itself below ground as an increase in the competition for water, and result in enhanced rooting depth. If trees are the plants of interest, there is a limit to the density to which they can be planted, and part of the soil is always bare. This distinct disadvantage can be overcome by introducing a crop in between the rows of trees (Lovenstein et al., 1991). Such a system is worthwhile to implement if the production of both crops on the same patch of land is higher than the one that would be attained by both crops growing independently. In order to achieve this, the crops should not compete for resources, which in practical terms means that for arid zones: ●

●

●

the source of the water absorbed by the intercrop should be from the upper soil layers, and would, in the absence of the intercrop, evaporate directly to the atmosphere through the soil surface; the water that percolates deep into the soil profile should be exploited by trees or shrubs that have deep root systems, without seriously affecting their biomass production; during the growth period of the intercrop, solar radiation should not be a limiting factor (i.e. not intercepted by shrubs or trees).

12.4

SUGGESTED SEQUENCE OF STEPS

A schematic description of an ideal sequence of events for such a hypothetical system is presented in Figure 12.3. After a flood event the soil profile is completely wet. Trees are planted, which take up water and produce biomass. When the addition of biomass is negligible (usually due to water shortage) a lopping of the tree (leaving only the trunk and a few stems) should be carried out. Immediately after the following flood the intercrop is planted. It commences its development as a normal crop, because there is neither competition for solar radiation nor for water owing to the fact that the trees are at this stage leafless. Thereafter trees start to develop leaves very slowly, and the rate of water uptake is therefore higher than the one the intercrop would have if a crop is solely grown. Once the intercrop reaches its peak and leaf shedding commences (as would be the case for a grain crop), the rate of biomass production of the tree should accelerate. At this point in time water is taken up from deeper in the profile than at the same stage in period I and more water has been taken up. 12.5

CASE HISTORY IN KENYA

A system based on the principles mentioned earlier was tested in the field by growing simultaneously and on stored water (collected during runoff events) shallow rooting annuals and deep rooting perennials. The field trial was carried out in the Turkana district of Northern Kenya using Acacia saligna as the tree component due to its resistance to drought (Nativ et al., 1998) and Sorghum bicolor and Vigna unguiculata as intercrops. The area is arid and the average annual rainfall is 330 mm. Runoff water was collected and stored in the soil profile. The effects of two different tree planting densities (2500 and 833 trees per ha), tree pruning (no pruning versus pruning), annual intercrops on total biomass production, and their interactions were tested (Droppelmann et al., 2000a). High levels of biomass production were attained by trees (more than 13 t.ha⫺1 over a two-year period), irrespective of intercropping and pruning. The high land equivalent ratios (Vandermeer, 1992) computed for pruned and intercropped systems (1.36 and 1.47 for low- and high-tree density, respectively) indicate that there was no competition for water and solar radiation, and probably no competition for the major nutrients as well (Droppelmann et al., 2000a,b).

198

P.R. Berliner

Flood events

Trees

Intercrop

Relative biomass production

100% I

Fraction of available water taken up

0

II

Time

Depth

Maximum rooting depth of intercrop

100%

Maximum rooting depth of trees

Figure 12.3. Schematic description of an ideal sequence of events for a runoff agroforestry system. I and II denote growth cycles of trees (for detailed description of scheme see text) (see Color Plate XLIV).

The key issue that ensured that the tested system performed satisfactorily is the fact that large volumes of water could be collected and stored in the soil profile. If soils are shallow, and not enough water can be collected, competition between trees and intercrop will ensue and no benefit from their intercropping will be obtained. Runoff agroforestry has proved itself a highly flexible and suitable production system for arid environments that allows to increase and stabilize yields and provides the farmer with a variety of essential products.

REFERENCES Agassi, M., Morin, J., and Shainberg, I. (1985) Effect of raindrop impact and water salinity on infiltration of sodic soils. Soil Science of America Journal, 49, l86–190. Ben-Asher, J. and Berliner, P.R. (1994) Runoff irrigation. In: K.K. Tanji and B. Yaron (eds), Management of Water Use in Agriculture. Berlin: Springer Verlag. Bruins, H.J. and Berliner, P.R. (1998) Managing bioclimatic aridity, climatic variability and desertification. In: H. Bruins and H. Lithwick (eds), The Arid Frontier. Interactive Management of Environment and Development. Dordrecht, The Netherlands: Kluwer Academic Publishers. Droppelmann, K.J. and Berliner, P.R. (2003) Runoff agroforestry – a technique to secure the livelihood of pastoralists in the Middle East. Journal of Arid Environments, 54, 571–577.

Runoff agroforestry 199 Droppelmann, K.J., Ephrath, J.E., and Berliner, P.R. (2000a) Tree crop complementarity in a runoff agroforestry system. Agroforestry Systems, 50, 1–16. Droppelmann, K.J., Lehmann, J., Ephrath, J.E., and Berliner, P.R. (2000b) Water use efficiency and uptake patterns in a runoff agroforestry system in an arid environment. Agroforestry Systems, 49, 223–243. Evenari, M., Shanan, L., and Tadmor, N. (1982) The Negev: The challenge of a desert (2nd edn). Cambridge, MA: Harvard University Press, 437pp. Lovenstein, H., Berliner, P.R., and van Keulen, H. (1991) Runoff agroforestry in arid lands. Forest Ecology and Management, 45, 59–70. Nativ, R., Ephhrath, J.E., Berliner, P.R., and Saranga, Y. (1998) Drought resistance and water use efficiency in Acacia saligna. Australian Journal of Botany, 47, 577–586. Sanchez, P.A. (1985) Science in agroforestry. In: F.L. Sinclair (ed.), Agroforestry: Science, Policy and Practice. Dordrecht, The Netherlands: Kluwer Academic Publishers, 287pp. Sauerhaft, B., Berliner, P.R., and Thurrow, T. (1998) The fuelwood crisis and runoff agriculture as a viable solution. In: H. Bruins and H. Lithwick (eds), The Arid Frontier. Interactive Management of Environment and Development. Dordrecht, The Netherlands: Kluwer Academic Publishers. Vandermeer, J. (1992) The Ecology of Intercropping. Cambridge: Cambridge University Press, 237pp. Zohar, Y., Aaronson, J.A., and Lowenstein, H. (1998) Cultivation of multipurpose trees in rain water harvesting systems in the arid zones of Israel. Commonwealth Forest Review, 67(4), 339–349.

CHAPTER 13

Water Pollution and its Numerical Modeling in Coastal Watersheds A. Ghosh Bobba Environment Canada, National Water Research Institute, Burlington, Ontario, Canada, L7R 4A6 Vijay P. Singh Department of Civil and Environmental Engineering, Louisiana State University, Baton Rouge, LA, USA

13.1

INTRODUCTION

A coastal zone, in general, includes parts of the continent under direct maritime influence. Its weather, vegetation, wildlife, and soils are different from those of non-coastal areas. The coastal water and its biota are totally different in appearance. The water is salty and rises and falls with the tide. A biophysical boundary can be defined in terms of natural features: biological, geological, physical, or a combination thereof. These features can include drainage basins, flood plains, dune formations, ecosystems, and ridges of coastal mountain ranges. The coastal zone is logically separated into two major systems, which are distinct, but interlocking. The major connection is provided by the water flow. Although shorelines are ecologically complex and have high resource values, we are concerned here only with their interaction with coastal water systems and their influence on the carrying capacities of coastal ecosystems. Because this influence is exercised mostly through the discharge of runoff water, coastal water concerns are specifically directed toward coastal watersheds, and lands that drain directly into coastal waters; these lands are defined and referred to as shorelines. Environmental pollution is becoming a common occurrence in many coastal areas (Bobba, 2002b). In the pollution literature, pollution is regarded as occurring in such high dosages or concentrations that it renders the polluted medium very hazardous or highly deleterious to biota. Many urban and rural areas of the industrialized and industrializing world have been adversely affected by large-scale pollution resulting in losses of human, material, and financial resources. Volumes of these pollutants are produced yearly through natural and anthropogenic activities such as industrial activities, agricultural practices, waste disposal systems, etc. High-level, medium-level, and low-level wastes in solid, liquid, or gaseous forms are released into the environment at discrete intervals or on a continuous basis. These pollutants may be physical, chemical, biochemical, biological, or microbiological in nature. They may have short or long half-lives in the environment. They have continued to damage the environment of many industrialized countries, having defied many painstaking control programs. Many coastal cities of developing countries are now also similarly threatened. The natural processes and anthropogenic activities that generate pollutants are many and varied, and so are their sources. The natural processes include products of soil and gully erosion, physicochemical weathering and mass wasting, sediment transport, floods, volcanic eruptions,

202

A. Ghosh Bobba and Vijay P. Singh

seawater intrusions, etc. The manmade ones include industrial, agricultural, sewage wastes and lagoons, garbage dumps and barnyards, and mining wastes along the coastal areas. These pollutants reach the subsurface water systems via the hydrologic cycle and pollute them. Through circulation of water within the hydrologic cycle, pollutants on the surface are transferred through the soil zone into the subsurface layers where they damage potable water supplies. The present control techniques with regard to pollution hazards, particularly in developing countries, need to be greatly improved. Government authorities do not show priority and concern adequately, and hence appropriate planning and management strategies to check pollution are generally absent. The expertise or requisite manpower may be lacking. Funds for basic research may not have been provided. Environmental protection laws or effects may be non-existent and where available are rarely enforced. These have exacerbated the spreading phenomenon of many pollutants in many countries. In this study, necessary suggestions for improvement of this situation would be given to control environmental problems along coastal areas. Parts of the environment currently being polluted include atmosphere, pedosphere, hydrosphere, lithosphere, and biosphere. This chapter focuses on pollution of the hydrosphere, with particular emphasis on the subsurface water regime, and pollution incidences in coastal areas of developed and developing countries. The scope would embrace sources and types of pollution, processes generating them, implications of geology/subsurface hydrology, and pollution dynamics and mechanisms. Potentials of subsurface water pollution of coastal areas in developing countries vis-à-vis the developed ones would be outlined, highlighting their health hazards. Relevant suggestions for combating pollution more effectively shall be made. The primary objective is to review the general incidences of environmental problems in relation to the effects of pedology and geology in close association with the dynamics of the hydrologic cycle in coastal areas. Proper understanding of sources and types of pollutants and their genesis, hydrodynamics, and hydrogeochemistry would help determine appropriate control measures. Thus, the goal is to contribute to better control methods. It is believed that present control methods in many parts of the world lack the depth of understanding required. In addition, many of these control efforts seem to be uncoordinated. Developing countries, still at the threshold of widespread pollution, can learn from the costly mistakes of the industrialized countries and hence take the necessary actions to protect their coastal environments. 13.2

TYPES OF POLLUTANTS

The two main sources of pollutants are point sources (Table 13.1) and non-point sources (Table 13.2). Pollutants from the two sources may be released continuously or at discrete intervals (Figure 13.1). Point sources of pollution can be geometrically defined and the dimensions are amenable to mathematical analysis in assessing pollution loads and rates of discharge determined. Point sources of pollution may assume any geometrical shape such as circular, triangular, Table 13.1.

Point sources of pollution in a coastal watershed.

Type of pollution Sewage disposal systems Surface waste disposal sites Underground waste disposal sites Spills, washings, and intrusions Mining sources Natural mineral/ore deposits

Examples Sewage lagoons, septic systems, cesspools, barnyards/feed lots Landfills, garbage dumps, surface waste dumps Storage tanks (low-, medium-, high-level wastes) Oil, gas, waste spills: auto workshop washings, research laboratory washings, seawater or saltwater intrusions Acid mine drainage: mine waste dumps, seepages gas explosions Saline springs, hot spring waters, anhydrite, pyrite deposits, etc.

Water pollution and its numerical modeling 203 Table 13.2.

Non-point sources of pollution in a coastal watershed.

Source

Examples

Agriculture Silviculture Construction Mining Utility maintenance Urban runoff

Cropland, irrigated land, woodland, feed lots Growing stock, logging, road building Urban development, highway construction Surface, underground Highways, streets and de-icing Floods and snowmelt

Atmospheric depth Fertilizer

Wastewater

Soil Vadose zone

Seepage face

Aquifer Salt water

Estuary Salt water

Salt water

Figure 13.1.

Different sources of pollutants in a coastal watershed.

spherical, etc. The areal sources of pollutants or leachates are comparatively smaller, easily mapable, and readily distinguishable. However, where the input/output load functions from point sources into the subsurface environment are continuous, the polluted area may eventually become widespread. Distributed sources of pollutants are much more widespread and can rarely be geometrically defined as precisely as a point source. Hence, it is more difficult to subject the input/output source to precise mathematical analysis. Rather a measured and intelligent assumption of the affected area is made for use in modeling and analysis. In heavily polluted areas, both point sources and distributed sources may occur together or may be independent of one another. Successful control methods or mathematical modeling of the affected/polluted area must recognize this situation in order for the control program to be effective. 13.3

POINT SOURCES OF POLLUTION IN COASTAL AREAS

In the list of point sources given in Table 13.1, the pollutants are generated in zones or areas of known and definable boundaries that are easily amenable to mathematical analysis and modeling. The pollution loads can be controlled at the point of input before they can spread into the surrounding environment in a time-discrete or continuous manner. Point sources include sewage lagoons (solid, gaseous, and liquid), industrial wastes, landfills/garbage dumps/barnyards, liquid/gaseous spills (oil, chemicals, etc.), mining (pits, holes, excavations, wastes, and gangue minerals), saline lakes, aquaculture ponds, deposits, evaporate sequences, etc. Through the complex interplay of various soil and geologic factors and rainwater events of the hydrologic cycle, pollution substances reach the subsurface systems and pollute them. For example, buried

204

A. Ghosh Bobba and Vijay P. Singh

refuse or garbage undergoes biodegradation in the soil zone. The resulting leachates are released into the subsurface water flow system where dissolved geochemical constituents are transported for various distances and in different directions. Piled up animal wastes in barnyards or liquid wastes in lagoons are similarly leached out and transported causing pollution of surface waters, beaches, and subsurface systems. In many coastal rural and urban areas today, industrial and domestic wastes are indiscriminately dumped into rivers, lakes, streams, dry valleys, wetlands, mangroves, etc. This was the case in many coastal areas, and such practices still persist in some of them today. These wastes damage surface waters and eventually destroy parts of the subsurface water regime. Mining wastes and gangue products and other point sources of pollutants produce a similar havoc to the environment. 13.4

POLLUTION DUE TO NON-POINT SOURCES IN COASTAL AREAS

Non-point sources of pollutants given in Table 13.2 are those in which the pollutants are spread through a large area of subsurface environment and in which they extend over the entire source area. Non-point sources tend to be widespread and the pollutants may be introduced from various sources and directions. The spreading of the pollution may be enhanced by wind, rain, and snowfall activities through atmospheric circulation and precipitation. The areal extent or boundary conditions for the pollutants are difficult to define because of the regional nature of sources, thereby posing problems for mathematical analysis. The sources include acid-alkaline rain, floods, erosion, agricultural fertilizer applications, generated agricultural wastes, sea sprays and intrusions, volcanoes, etc. Surface waters and atmospheric fallout pollute shallow subsurface waters. In urban, suburban, and rural areas of many coastal areas, particularly in the tropics, soil and gully erosion produce heavy sediment loads carried by floods that pollute surface water and subsurface systems and coastal areas. Waste products in urban coastal areas are transported away by runoff. In mining areas, gangue/waste materials dumped about recklessly on surface decay and liquid wastes are leached from them, becoming components of the subsurface environment. In regions of intense geomorphic degradation and mass wasting, physicochemical and biological weathering disintegrate pedologic and geologic materials to produce sediments that provide great quantities and varieties of pollutants. Fallouts from industrial areas or atmospheric tests in one area may spread into other regions of the world; wind, wave action, sea spray or saltwater intrusion may drive pollutants inland and road salt application for de-icing during winter and widespread fertiliser usage, particularly in coastal areas, are also major distributed sources of pollution. 13.5

BIOLOGICAL POLLUTANTS IN SUBSURFACE WATERS

Biological pollutants of subsurface waters include dissolved organic constituents and microorganisms that seep or leach into subsurface waters from polluted surface waters. Microorganisms may contribute to pollution in many ways, namely they may themselves be pathogenic; aesthetically they may produce undesirable biomass, or they may generate toxic metabolites in the subsurface water. The microorganisms may be either pathogenic or nonpathogenic. In both cases, they produce undesirable effects in the subsurface water itself and in the distribution network and the populations using it. 13.5.1 Pathogenic microorganisms Pathogenic microorganisms are present in subsurface waters, especially in the vicinity of facilities that discharge sewage effluents or polluted surface waters, and septic tanks, agricultural

Water pollution and its numerical modeling 205 Table 13.3.

Major water- and excreta-related pathogens.

Pathogen

Type

Disease

Rotavirus Cryptosporidium parvum Legionella pneumophila Campylobacter jejuni Escherichia coli

Virus Protozoan Bacterium Bacterium Bacterium

Helicobacter pylori Enterocytozoon bienusi Cyclospora cayetanenis Hepatitis E Encephalitozoon hellum Vibrio cholerae Hepatitis F

Bacterium Protozoan Protozoan Virus Protozoan Bacterium Virus

Diarrhea Acute enterocolitis Legionnaires’ disease Diarrhea Hemorrhagic colitis, hemolytic uremic syndrome Gastric ulcers, stomach cancer Diarrhea Diarrhea Enteric hepatitis Conjunctivitis Cholera Enteric hepatitis

wastes, and refuse dumps in coastal areas. Microorganisms, however, must survive the tortuous task of passing through the soil cover, which constitutes an excellent natural process for water filtration and treatment. Even with this barrier, it follows that the nearer these sources of pollution are to subsurface water sources the greater the chance of successful seepage of these microorganisms. Shallow wells and some deep boreholes are prone to pollution by these pathogens. The isolation of pathogenic microorganisms from subsurface waters is difficult but, when achieved, it serves as obvious proof of potential danger to the users, regardless of the number of pathogens present. Generally, however, the majority of waterborne pathogenic microorganisms enter water supplies as a result of fecal pollution. Therefore, the ability to detect fecal pollution at low levels is the main safeguard in preserving the potability of water supplies. Pathogenic microorganisms normally associated with water supplies are shown in Table 13.3. All of these have been isolated from polluted shallow wells and deep bore holes in coastal areas. In addition, Dracunculus medinesis (Guineaworm) was reported from wells in parts of India. These parasites are widespread in many parts of India, sometimes occurring in epidemic proportions. Fecal pollution in water is usually demonstrated by the detection of specific bacteria that are present in large numbers in the intestines. The test normally employed is the presumptive Coliform test, which involves the most probable number (MPN) counts using liquid media. Coliform organisms include Escherichia coli, Citrobacter, and Klebsiella, which are members of the family Enterobacteriaceae. They are gram-negative, oxidize-negative, non-spores forming rods that can grow aerobically in a medium containing salts. They are able to ferment lactose within 48 hours, producing acid and gas at 37⬚C. A presumptive coliform test with a very high count is usually followed by a confirmatory test, which is specific for E. coli. 13.5.2

Non-pathogenic microorganisms

Many non-pathogenic bacteria are as important as the pathogenic ones in the pollution of surface water and subsurface water supplies. These include sulphur and iron bacteria. Among the sulphur bacteria are the sulphate reducers, such as Desulfovibrio, Desulfomonas, and Desulfotomaculatum, which produce elemental sulphur from sulphates. On the other hand, some of the sulphur bacteria oxidize elemental sulphur to sulphates, all of which involve

206

A. Ghosh Bobba and Vijay P. Singh

complex oxidation–reduction reactions. These include the ubiquitous chemolithotrophic Thiobacillus and the filamentous gliding bacteria Beggiatoa and Achromatium. Iron bacteria are frequently present in subsurface waters and, in particular, those subject to a degree of organic pollution. They obtain energy for their metabolism by oxidation of ferrous and/or manganese ions. These include the gliding bacteria, Toxothrix; the sheathed bacteria, Spaethilus, Lepothrix, Crenothrix, and Clonothrix; the budding and/or appendage bacteria, Pedomicrobium, Gallionella, Metallogenium, and Kusnezovia and the gram-negative chemolithotrophic bacteria, Thiobacillus ( T. ferrooxidans ), Siderocapsa, Naumaniella, Ochrobium, and Siderococcus. Pathogenic and non-pathogenic microorganisms (bacteria, fungi, and viruses) are thus hazardous environmental pollutants to the subsurface environment. They enter this environment from waste disposal and treatment areas, sewage lagoons, barnyards, landfills, and mine areas. They occur in varying degrees in both oxidizing and reducing environments. In the process of complex redox activities that break down organic and inorganic materials to release energy for metabolic activities, poisonous substances are generated that may be fatal to the hosts that ingest them. These redox microbial activities also degrade their habitats, rendering them unusable. Such degraded states may remain so for a long time. 13.6

ORGANIC POLLUTANTS IN SUBSURFACE WATER

Organic pollutants that may be found in subsurface water through shallow wells and deep bore holes include dissolved organic carbon (DOC) and particulate organic carbon (POC). They, in association with microorganisms, cause destructive pollution in subsurface environments. They may serve as nutrient/energy sources for microorganisms. Where they are heavily loaded into subsurface, DOC and POC enhance microbial multiplication and growth, thereby rendering the habitat anoxic. In such environments, denitrification, desulphurization, etc., are rampant, engendering an abundant growth of bacteria, fungi, and viruses that may be highly pathogenic. Hence, serious organic pollution signals a potential heavy microbial pollution of a subsurface water system. 13.7

THERMAL POLLUTION

Industries are the main source of thermal pollution. In industrialized countries and in some developing ones, heat generated by industries is discharged through wastewater into the environment. High temperature waters eventually reach shallow aquifers and adversely affect subsurface water. Hot waters, discharged into ponds, that are influent, may form high temperature haloes that extend into the aquifers underlying coastal areas. Unchecked thermal pollution not only negatively affects the life in coastal areas but also that of the subsurface system associated with the shoreline. Other problems that can arise are changes in physical, chemical, and biological characteristics of the subsurface system, thereby rendering the surface and subsurface waters unusable. 13.8

PROCESSES AND ACTIVITIES THAT GENERATE POLLUTANTS

Various processes, some of which may be caused by man or are anthropogenic, generate pollutants that enter subsurface flow systems. These processes include physicochemical weathering, mass wasting, erosion, sediment transport, and deposition; agricultural activities; mining, mine-waste disposal, and acid mine drainage problems; oil exploration, exploitation, and gas

Water pollution and its numerical modeling 207 flaring; other industrial activities such as manufacturing, distribution of manufactured products, etc.; sewage treatment, disposal, and management; runoff, floods, and snow melt; biological pollution of wetlands and impounded reservoirs; saline lakes, ponds, and evaporate deposits; atmospheric fallout and rainout; burial grounds, garbage dumps, landfills, etc. Some pollution sources in coastal rural environments that are usually ignored, even though they may be hazardous, include pit latrines, open-space communal toilets, wide scale and indiscriminate uses of the bush for defecation, personal hygienic uses of water for washings, etc., microbiological activities (bacteria, virus, fungi, worms, etc.), radioactive material, thermal products, heavy metals, trace elements, ions, etc.

13.9

CHEMICAL POLLUTANTS OF SUBSURFACE WATER

Coastal watersheds are witnessing a stage of development where subsurface waters from shallow wells and boreholes are gradually supplementing the original source of drinking water, namely, the surface water. The preference for subsurface water over surface water is based on the belief that when surface water has been distributed as tap water it must always be subjected to some purification prior to distribution. Although surface waters are easily accessible where they exist in lakes, rivers, streams, and springs, many people believe that wells produce water of good quality. Thus, subsurface water is not treated before use and is believed to be free from pollution. Any subsurface water system may be naturally polluted to a certain degree at all times. The concern of many water resource engineers and managers is whether the amount of measured pollutants is within the acceptable limits of water quality. The number of chemical pollutants and the degree of chemical pollution of subsurface water depend on the geology, pedology, and the mineral composition of the soil and rock through which water flows before reaching the aquifer. Subsurface waters may have pollutants that not only depend on the pedology, geology, and mineralogy of the formations it flows through but also on the constituent pollutants in the water that recharges the subsurface water. The three components of water quality are bacteriological quality, physical quality, and chemical quality. Filtration and sedimentation processes take care of the physical quality. In practice, natural processes filter subsurface waters as they pass through columns of soils, sands, strata, or sedimentary layers of rocks. Subsurface waters are usually clear of solid materials as they come from the aquifer, particularly if they are deep-seated ones. The intricate pore spaces or water passageways of the aquifer materials act as a fine filter and remove small particles of clay or any other fines. Organic materials decay, or are destroyed in transit. Thus, the dirtiest and most polluted sewage water may become clear of suspended/particulate solid materials once it has gone through a thick bed of sand or geologic and pedologic units. As a result of this natural self-cleansing of polluted water by deep-seated aquifers, physical and some biological aspects of pollution may not pose serious problems in subsurface waters. The most undesirable trace elements – pollutants in subsurface water – are mercury (Hg), lead (Pb), cadmium (Cd), arsenic (As), barium (Ba), boron (B), cyanide (CN), selenium (Se), chromium (Cr), uranium (U), sulphur (S), and nitrogen (N). Most subsurface waters may also contain many major inorganic elements, compounds, and ions in excess of acceptable standards, such as iron oxide (Fe2O3), manganese (Mn), calcium (Ca), magnesium (Mg), chloride (Cl), aluminium (Al), and silica (SiO2). Anions and cations can be found in their dissolved states. Deep subsurface waters contain more dissolved chemical elements than shallow waters. Subsurface waters in limestone or Karstic terrains may contain dissolved calcium and magnesium salts.

208

A. Ghosh Bobba and Vijay P. Singh

13.10 13.10.1

GEOLOGIC AND SUBSURFACE HYDROLOGY OF A COASTAL WATERSHED Geologic setting

Natural geological processes are primary contributors to subsurface water pollution. In this regard, the rocks of the earth’s crust are the major contributors of subsurface water pollutants. Pollutants enter subsurface waters from the recharge areas of aquifers. It is, therefore, necessary to know the geology and mineral distribution in recharge areas of aquifers. Subsurface water that is recharged from mining areas, fertilized agricultural farmlands, and industrial areas may not be safe for human consumption unless the recharge area is protected. Rain and surface water can leach pollutants from mine dumps, ore deposits, city dumps, and fertilizer applied to farmlands into the recharge areas for subsurface water supplies. 13.10.2

Hydrologic system setting

A watershed is topographically defined as an area in which water enters through precipitation and leaves as evapotranspiration, surface runoff, and subsurface water discharge. In the case of a coastal watershed, runoff, and subsurface water discharge enter the sea (Figure 13.2). Rainfall and potential rate of evapotranspiration are determined by climate. Actual evapotranspiration is limited by the climatically controlled potential rate, vegetation, and the wetness of the soil. Soil properties, topography, and the history of rainfall and evapotranspiration determine runoff and subsurface water discharge. For the purpose of this discussion, surface runoff includes direct runoff, which occurs as streamflow immediately following a rainstorm, and drainage of subsurface water into creeks, which accounts for water flow between streams. The amount of direct runoff generated by a storm depends on the amount of rainfall and on the moisture condition of the soil. In general, more runoff occurs when the soil is initially wet. The subsurface water discharge can be calculated if rainfall, evapotranspiration, runoff, and the change in the amount of water stored on the watershed are known. A link between rising sea level and changes in the water balance is suggested by the general description of the hydraulics of subsurface water discharge at the coast. Fresh subsurface water rides up over denser saltwater in the subsurface system on its way to the sea (Figure 13.3) and subsurface water discharge is focused into a narrow zone that overlaps with the intertidal zone. The ET

P

OF R

WT

Sea level

I

GD sea

Fresh water

h Fres e n i l Sa

Salt water

WT – Water table; P – Precipitation; ET – Evapotranspiration, OF – Runoff

Figure 13.2.

Hydrological pathways in a coastal watershed.

Water pollution and its numerical modeling 209

India Mumbai

Godavari delta Chennai

N

Uppada

Rajahmundry

Kakinada Ga uta

V a si

aG

od

ava

– ri R

Yanam

od

N i l a r ev u

ar

mi

av

i R–

Vain a Razole

teyam

Narasapur

Amalapuram

Bay

Figure 13.3.

l enga m 30 m 0 20

ta

s ht

m

iG

G au

of B

8 km

Location of the Godavari Delta.

width of the zone of subsurface water discharge measured perpendicular to the coast is indirectly proportional to the discharge rate. The shape of the water table and the depth to the fresh/saline interface are controlled by the difference in density between freshwater and saltwater, the rate of freshwater discharge, and the hydraulic properties of the subsurface system. The elevation of the water table is controlled by mean sea level through hydrostatic equilibrium at the shore. Because pollutants are transported in large part by the bulk motion of subsurface water, the parameters of subsurface water flow are of major importance in the understanding of pollution processes. The various aspects of the subsurface water environments as well as stratigraphic factors that control or could influence subsurface water motion are also of major consideration. The subsurface hydrology environment is shown schematically in Figure 13.2. It consists mainly of saturated and unsaturated zone. The unsaturated zone occurs above the capillary fringe where the soil pores are partially saturated with water. This zone is important in waste management because in most cases, it is the burial zones for wastes. Consequently, a thick unsaturated zone may sometimes be preferred for waste disposal since it would take a much longer time for pollutants to reach the water table. In the saturated zone, the pores are saturated with water. When this zone is capable of transmitting significant quantities of water for economic use it is referred to as an aquifer. In most field situations, two or more aquifers occur, separated by impermeable strata or aquitards. In the situation illustrated in Figure 13.2, the upper or unconfined aquifer is much more prone to pollution than the lower confined aquifer. 13.10.3 Geochemical processes The migration of pollutants in subsurface water flow systems is due mainly to subsurface water motion. Transport rates, however, are moderated by a variety of geochemical and biochemical

210 A. Ghosh Bobba and Vijay P. Singh processes, acid–base reactions, oxidation–reduction processes, precipitation–desorption reactions, and microbial reactions. The precipitation–dissolution, adsorption–desorption, and micro-reactions can lead to the removal of pollutants from solution, whereas other processes affect the availability of the pollutant for adsorption or precipitation. Appropriate instrumentation, sampling, and monitoring would make these subsurface geochemical species as veritable environmental tracers. There are a large number of physicochemical properties or processes that are responsible for determining the speciation, fate, and transport of inorganic chemicals in the environment. The fundamental processes, which are important in the area of inorganic contaminants, are (1) hydrolysis, (2) precipitation/dissolution reactions, (3) oxidation/reduction reactions, (4) complexation, (5) adsorption reactions, (6) ion exchange reactions, and (7) volatilization. The details of hydro-geochemical processes are explained earlier (Bobba et al., 1995; Aswathanarayana, 2001). 13.11

ANTHROPOGENIC IMPACT IN COASTAL WATERSHEDS

Pollutants can be introduced into the surface and subsurface water system as a result of human activities. This category of pollution of subsurface water systems differs from the ones described earlier in that such human activities do not introduce wastes into the subsurface. The major human activities that eventually end up polluting subsurface water systems include agricultural activities, storage of gasoline tanks in the subsurface, pipe lines, road de-icing, mining, and pumpage of aquifers, etc. (Figure 13.1). 13.11.1

Sanitary landfills and garbage dumps

Much of the solid waste that is now disposed of on land is placed in sanitary landfill sites along coastal areas. In humid areas, in particular, buried waste in sanitary landfills and dumps is subject to leaching by the percolating rainwater. The leaching process is accompanied by chemical reactions that tend to consume all available oxygen, while releasing carbon dioxide, methane, ammonium, bicarbonate, chloride, sulphate, and heavy metals. The liquid mix of these constituents is referred to as leachates. The total number and chemical concentrations of these constituents can be variable, depending on the initial composition of the waste and climatic conditions. Leachates contain large numbers of inorganic and organic pollutants and also have high total dissolved solids. Leachates emanating from landfills contain pollutants and toxic constituents derived from solid wastes, as well as from liquid, industrial wastes placed in the landfill. 13.11.2

Septic tanks and cesspools

Septic tanks are designed to settle solids, reduce biochemical oxygen demand, eliminate microorganisms before (the treated) sewage is released through a drain field into the ground. The septic tanks and cesspools are the largest contributors of wastewater to the ground and are the most frequently reported sources of subsurface water pollution in coastal areas. Apart from the effluent that is directly released into the subsurface, there arise large volumes of solid residual materials known as sewage sludge. In many parts of the world, this sludge, which contains a large number of potential pollutants, is applied on agricultural lands to enhance crop nutrients, such as nitrogen, phosphorus, and heavy metals that are needed for plant growth. Although this practice actually improves soil fertility, it has been observed that one of the potential negative impacts of this type of sewage disposal is degradation of subsurface water quality. Pollutants in the sewage sludge/effluent reach and pollute subsurface water through infiltrating water from rain or snow.

Water pollution and its numerical modeling 211 13.11.3 Aquaculture pollution The rapid expansion of aquaculture is often considered as a measure of success, but this expansion and commonly practised rearing methods have caused environmental problems. Most of the solid waste (uneaten food and fecal material) settles onto the sediment and can have an impact on the benthic ecosystem of inland and marine waters. The changes which take place include the formation of anoxic sediments, an increase in the flux of nutrients such as ammonium and dissolved reactive phosphate from sediment into the overlying water, and changes in the population structure of the benthic macrofauna. The most severe impacts are generally restricted to the immediate vicinity of the aquaculture sites. In many cases, enhanced activity of sulphate reducing and methanogenic bacteria within the sediment results in outgassing of carbon dioxide and hydrogen sulphide at some marine cage farms which has been held responsible for loss of appetite, gill damage and increased mortalities of fish. A major problem noted is the pollution and salinization of drinking water in many villages. Excessive pumping of groundwater along the coast for the purpose of salinity control in the ponds has resulted in intrusion of seawater. Also, storing of seawater in ponds continuously for many months tends to turn the groundwater saline through seepage. Seepage of saline water from ponds has led to salinization of soil of nearby agricultural lands. Thus, there is a danger that most of the fertile lands near the coast would become barren after few years. It is found that effluents which contain chemical fertilizers, antibiotics, and toxic elements are discharged from ponds polluting nearby estuaries, canals, and tanks. Thus, the prosperity of downstream aquaculture has created adversity in agriculture in the upstream side. 13.11.4 Oil and gas pollution due to gas stations and refineries Nitrate loading of shallow subsurface water systems arising from fertilizer application occurs mainly through leaching. Where there is significant downward flow, deep-seated aquifers can become affected. Numerous case studies in various parts of the watershed have shown that the nitrate derived by oxidation and leaching of natural organic nitrogen in the soil is often responsible for extensive nitrate pollution of shallow subsurface waters. The pathway down to the water table of the pollution generated through mining activities and road salt application is similar to that for nitrates, that is, they reach the water table through leaching and flushing through the unsaturated zone by infiltration of percolating water from rain and snow melt. Oil refineries tend to be located near harbors and coastal areas. Petroleum leakage from underground storage tanks and oil pipelines, as well as spills from oil-producing wells, constitute an increasing threat to subsurface water quality. A simple subsurface hydrology condition is assumed. In the initial migration stage (seepage stage), the oil moves primarily in a downward direction under the influence of gravitational forces. On reaching the water table, the oil zone spreads laterally, first under the influence of gravity-related gradients and subsequently in response to capillary forces. Capillary spreading is very slow, and eventually a relatively stable condition is attained. 13.12

SALTWATER INTRUSION ALONG COASTAL AREAS

Saltwater intrusion usually refers to the movement of salty water into aquifers or to the encroachment of saline water into freshwater reaches of estuaries (Figure 13.2). Bobba (1993a) discusses the relationship between fresh and saline groundwaters in coastal areas. Under natural conditions an “interface,” that is, a narrow transition zone, commonly separates circulating freshwater from seawater in coastal aquifers. Any pumpage from a coastal aquifer reduces the seaward discharge of fresh groundwater and decreases fluid pressure along the interface,

212

A. Ghosh Bobba and Vijay P. Singh

causing it to shift landward and upward. However, if a net flow of fresh groundwater toward the sea is maintained, the movement of interface represents only an adjustment toward a new equilibrium, in which the saline water again becomes stationary, and a part of the original fresh water is diverted to well fields. In this process of adjustment toward a new equilibrium position, individual wells which are either too deep or too close to the shoreline may begin to draw saline water; regionally, however, the situation is one in which most of the wells in the area are simply intercepting freshwater that formerly discharged to the sea. On the other hand, if pumpage exceeds the regional flow to the sea, which existed naturally, the saline water cannot become stationary in a new equilibrium position. Rather, saltwater will continue to move inland until it reaches the major pumping centers; in this case a general seawater intrusion problem emerges. In such situations, various strategies (such as the injection of treated water near the coast line) may be employed to retard saltwater movement while freshwater is taken from the storage in the aquifer. Ultimately, however, pumpage must be reduced to match the seaward flow of freshwater, which can be intercepted by well fields. Saltwater intrusion in estuaries occurs where the freshwater flow into an estuary is reduced either by upstream diversions of water or by drought conditions in the areas drained by inflowing streams. Intrusion may also be caused by engineering activities in the estuary itself, such as channel dredging. Dams that prevent the inland (upstream) movement of saltwater can control saltwater intrusion into freshwater reaches of estuaries. Saltwater also may be of concern to water users as a result of point and non-point sources of dissolved solids where there is extensive irrigation or where natural sources of saline water exist. 13.13

NUMERICAL WATER QUALITY MODEL AND ITS APPLICATION TO COASTAL WATERSHEDS

Bobba et al. (2000a) presented a detailed description of available model characteristics and recent trends in subsurface water flow and transport modeling. Mathematical groundwater models can be classified into four distinct categories according to their purposes: (1) prediction models that simulate the behavior of subsurface water systems and their response to stresses, (2) management models that integrate prediction models and simulation/optimization models for evolving alternative management decisions, (3) identification models that are used for parameter estimates, and (4) models used for storage, retrieval, manipulation, and management of data bases. It should be noted that the distinction between these categories is artificial and hazy. For example, a prediction model can be used iteratively for estimation of parameters, or a simulation model can be used to examine various alternative planning scenarios (Bobba and Joshi, 1989). In this chapter, a finite element model is applied to two cases. This model simulates fluid movement and the transport of either dissolved substances or energy in the subsurface system. The model can be applied areally or in cross section. It uses a two-dimensional, combined finite-element and integrated-finite difference method to approximate the equations that describe the two interdependent processes being simulated. When used to simulate saltwater movement in the subsurface system in cross section, the two interdependent processes are the density-dependent saturated subsurface water flow and the transport of dissolved solids in the subsurface water. Either local – or regional – scale sections having dispersed or relatively sharp transition zones between saltwater and freshwater may be simulated. The results of numerical simulation of saltwater movement show distributions of fluid pressures and dissolved solids concentrations as they vary with time and also show the magnitude and direction of fluid velocities as they vary with time. Most of the subsurface properties that are entered into the model may vary in value throughout the simulated section. Sources and boundary conditions may vary with time. The finite-element method using quadrilateral elements allows the simulation of

Water pollution and its numerical modeling 213 irregular areas with irregular mesh spacing. The model has been applied to real field data and observed to give favorable results (Bobba, 1993b, 1998; Bobba et al., 2000a,b). 13.13.1 Case study 1: Godavari Delta, India The Godavari Delta is located in the east coast of India (Figure 13.3). The details of geology and environmental problems have been explained earlier (Bobba, 2002a). The Godavari Delta lies between the sea level and 12 m contour. The delta has a projection of about 35 km into the sea from the adjoining coast. The Godavari Delta consists of an alluvial plain. It has a very gentle land slope of about 1 m per km. The coastal line along the study area measures to about 40 km and the general elevation varies from about 2 m near the sea to about 13 m at the upper reaches. Texturally, a major part of the study area consists of sandy loams and sandy clay loams. The silty soils, which are very deep, medium textured with fine loamy soils is located all along the Godavari River as recent river deposits. The very deep, coarse-textured soils with sandy subsoils representing the coastal sand are also found along the coast near the sea. The delta has a network of canal systems. Increased use of new crops along with chemical fertilizers and pesticides has brought about rapid growth in the agricultural output. The crops are irrigated with river water throughout the year except between the last week of April and the second week of June. As ample surface water is available to irrigate the delta, there has been no effort to use groundwater. Potassium fertilizers are extensively used for increasing the crop yield. The variation in potassium concentrations in groundwater in the study area is shown in Figure 13.4. The peak

100

Upper limit of K for drinking water is 12 PPM (EEC)

90

(1) The peak values are generally in November (2) Up to 1994 no change in K at Tallarevu (3) Potassium seems to have derived from fertilizers

Concentration of K in PPM

80 70 60 50 40 30 20

Karapa Kajuluru 10 Tallarevu 99 19

98 M ay

19

19

97 M ay

19

96 M ay

95 M ay

19

19

94 M ay

19

93 M ay

92 M ay

19

91 M ay

19

19

90 M ay

M ay

M ay

19

89

0

Period of analysis of groundwater

Figure 13.4. Variation of potassium concentration in groundwater in the Godavari Delta (Chachadi and Teresa, 2002).

214 A. Ghosh Bobba and Vijay P. Singh

9 8 7 6 5 Karapa

4 3 2 1

99 19

98 M

ay

19

97 M

ay

19

96 M

ay

19

95 M

ay

19

94 M

ay

19

93 M

ay

19

92 M

ay

19

91 ay

19 M

ay

19

89 ay M

M

ay

19

90

Tallarevu Kajuluru

0

M

Ratio of [CI/HCO3 + Co3] in EPM

10

Period of analysis of groundwater

Figure 13.5. Variation of [Cl/HCO3 ⫹ CO3] ratio in groundwater in the Godavari Delta (Chachadi and Teresa, 2002).

values are generally observed in November. Higher concentrations are found due to the application of potash fertilizer besides contribution from soils. The temporal variation of chloride and bicarbonate from groundwater samples is shown in Figure 13.5. It is quite clear that the bicarbonate concentrations in groundwater increase with time. The ratio of chloride/ bicarbonate ⫹ carbonate can be used as criteria to evaluate seawater intrusion. Chloride is the dominant ion in seawater and it is only available in small quantities in groundwater, whereas bicarbonate which is available in large quantities in groundwater, occurs only in very small quantities in seawater. The details of the model and application to the Godavari Delta Basin have been explained earlier by Bobba (2000, 2002a). The prediction of the water table depth due to irrigation and saltwater intrusion have been reported earlier (Bobba, 2000, 2002a). Figures 13.6(b), and 13.7(b) depict the influence of sea level rise variations in irrigation (rainy) and non-irrigation seasons. During high tide level and irrigation (rainy) season periods, the water table rises. The distance between surface soil and water table in the coastal area is very small, and the material is generally composed of sands, which do not retain significant amounts of moisture under unsaturated conditions. Hence, the water that overflows the soil directly recharges the groundwater. The distance between the water table and surface soil is at a minimum in the central portion of the delta. It has been observed that areas of minimum depth from the ground level to the water table have high freshwater potential whereas lowering of the water table from the ground surface reduces the freshwater potential substantially. The water table elevation varies from 0.5 to 1 m from MSL and decreases gradually towards the coastal side. Patches of freshwater zones are also present along coastal areas (Figures 13.6(b) and 13.7(b)). The areas of coastal aquifer contaminated by saltwater are delineated in Figure 13.6(b). The sea level raise (SLR) and non-irrigation season may cause an upward movement of saline water in coastal aquifers. The freshwater potential in the aquifer is depicted in Figures 13.6(b) and 13.7(b). The aquifer likely to be saline is more along the eastern side than the southeastern side. Saline water contamination due to SLR and non-irrigation may be critical to the southern tapering segment of the island.

(a) 80

70

60

50

40

30

20

10

0 0 10 20 30 40 50 60 70 80

Kakinada

Bay of Bengal

(b)

80

70

60

50

40

30

20

10

0

Rajahmundry Kakinada Amalapuram

Bay of Bengal

0 10 20 30 40 50 60 70 80

Figure 13.6. (a) Simulated hydraulic heads and (b) freshwater depth of Godavari Delta in non-irrigation months (Bobba, 2002a).

(a) 80

70

60 Rajahmundry

50

40

30

20

10

0

Kakinada

Bay of Bengal

(b) 80

70

60

50

40

30

20

10

Rajahmundry Kakinada

Amalapuram

Bay of Bengal

0

0 10 20 30 40 50 60 70 80

0 10 20 30 40 50 60 70 80

Figure 13.7. (a) Simulated hydraulic heads and (b) freshwater depth of Godavari Delta in irrigation season months (Bobba, 2002a).

216

A. Ghosh Bobba and Vijay P. Singh

80

Rajahmundry 70 60

50

40

30

20

10 Kakinada

0

0 10 20 30 40 50 60 70

Bay of Bengal

80

Figure 13.8. Simulated hydraulic heads of Godavari Delta in different seasons (solid line in heavy rainy season – drought conditions) (Bobba, 2002a).

Figure 13.8 shows the results in different environmental conditions due to heavy and long rainy season and drought conditions due to high temperature and less rainfall (climate change). Higher water table conditions are observed due to more rain and irrigated water is recharged to the aquifer. The saltwater was flushed out and seawater intrusion to the aquifer got stopped. However, if the severe drought conditions (higher temperature, lesser rainfall) occur in the delta, the water table is reduced due to higher evapotranspiration and over pumping the groundwater for irrigation and domestic purposes. The saltwater intruded the aquifer and freshwater thickness reduced in the delta. 13.13.2

Case study 2: Port Granby radioactive disposal site

Canada has a uranium refinery at Port Hope, Ontario. The waste from the refinery is disposed at Port Granby waste management site located on the north shore of Lake Ontario (Figure 13.9). In recognition of concern over the possible contamination of surface lake waters, the concentrations of radium and uranium are measured in water samples collected off Lake Ontario coastal zone near waste site. These data show the leachate infiltration and seepage to coastal zone of Lake Ontario. The plume moves parallel to the shoreline in the direction of the prevailing wind (Figure 13.9). The finite-element model was applied to calculate hydraulic head and contamination discharge to lakeshore. The predicted Ra-226 contamination concentration from the waste disposal site to the beach is shown in Figure 13.10. The details of the application of model and interpretation of results were presented earlier (Bobba and Joshi, 1988). 13.14

EPILOGUE

According to Fano et al. (1985), it may be expected that over the next decade the management of water quality problems will be one of the outstanding issues relating to the protection and conservation of the national stock of water in each country. In order to protect their environments, coastal provinces in countries like India should enforce environmental pollution control laws. Strong emphasis should be placed on public health, education programs, and enlightenment. Coordinated sampling and monitoring programs are required by zones, nations, and

Water pollution and its numerical modeling 217

Water shed boundary Direction of natural drainage

G

ra

nby

Cr

ee

k

N

rt Po

Crysler point 0

1 km Lake Ontario

W es

Port Hope Toronto Lake Ontario

1⬘ .04 1

k ree tC Waste disposal site

.06

.06

.05

105 m

.07 .04

.05

2⬘ N

East C

Bouchette point

reek

Port Granby radioactive waste disposal site (approx. 43°54⬘N 78°27⬘W)

75 m 1⬘ .04 1

3⬘ .04 2

.04

2⬘

.04

.04

.04

4⬘ 5⬘ .04 3

.12

3⬘ .04 2 4⬘

.04

.04

.04

6⬘

.04 .04

.04 4

6⬘

.04

.04

5⬘ 3

.06 .04 4

.10

.04 5

6

.04 F

0

.10

.06

.08 .07 .04 E

.08

.09 .04 D

C

.07 .06 .05

.04 B

50 100 150 200 m Scale

Shore Marker Marker Buoy Dam

.04

May 31, June 7 Surveys June14, June 21 Surveys

.06 5

.04

6

A

June 7, 1977 Wind from NW (20) Waves from W (70)

Figure 13.9.

Location of waste disposal site and Ra-226 concentration in Lake Ontario, Canada.

regions to check widespread/regional pollution. Hence, local, international, and regional pollution events should be traced and monitored on a continuous basis and warning signals against hazards issued to areas affected. Information exchanges should be encouraged between nations, among experts/professionals, governments, and aid-giving agencies.

A. Ghosh Bobba and Vijay P. Singh

1.3

1.7

6.7

370

Lake Ontario

1.2 1.0

0.9 0.9

73.2

0.4

37 37 94.7 Waste disposal site

925

268.0 185

0 1850

85.3

74

55 5 00 0

Ra (mBq/L)

37

Meters above sea level

226

1.2

0 925 00 74

97.5

0.9

14,800

105 m

103.6

100 200 m

218

0

100

200

300

1.0

N

61.0

0.9

0.4

1

0.8

74

400

Meters

Figure 13.10. Computed Ra-226 concentration in waste site, beach and observed concentration in coast (Bobba and Joshi, 1988).

Research programs at both central and provincial levels must pay attention to sources and types of pollution, modes of occurrence and spread, dynamics of transport and dispersion, lifeexpectancy of pollutants, and means of disposal of wastes. Development of effective control technology should be continuously and adequately funded. REFERENCES Aswathanarayana, U. (2001) Water Resources Management and Environment. Rotterdam, The Netherlands: A.A. Balkema Publishers. Bobba, A.G. (1993a) Mathematical models for saltwater intrusion in coastal aquifers – literature review. Water Resources Management, 7, 3–37. Bobba, A.G. (1993b) Field validation of “SUTRA” groundwater flow model to Lambton County, Ontario, Canada. Water Resources Management, 7, 289–310. Bobba, A.G. (1996) Environmental modelling of hydrological systems. Report No. 1014a and 1014b, Department of Water Resources Engineering, Lund Institute of Technology, Lund University, Lund, Sweden. Bobba, A.G. (1998) Application of numerical model to predict freshwater depth in islands due to climate change effect: Agati island, India. Journal of Environmental Geology, 6, 1–13. Bobba, A.G. (2000) Numerical simulation of saltwater intrusion into coastal basin of Indian sub-continent due to anthropogenic effects. ICIWRM. In: Proceedings of International Conference on Integrated Water Resources Management for Sustainable Development, Vol. 1. Published by National Institute of Hydrology, Roorkee, India, pp. 323–340. Bobba, A.G. (2002a) Numerical modelling of saltwater intrusion due to human activities and sea level change in the Godavari Delta. Hydrological Sciences Journal, 47(S), S67–S80. Bobba, A.G. (2002b) Environmental problems in subsurface water systems in coastal watersheds due to anthropogenic activities. In: Proceedings of International Conference on Hydrology and Watershed Management, 1, 375–389. Bobba, A.G. and Joshi, S.R. (1988) Groundwater transport of Radium-226 and uranium from Port Granby waste management site to Lake Ontario. Nuclear and Chemical Waste Management, 8, 199–209. Bobba, A.G. and Joshi, S.R. (1989) Application of an inverse approach to a Canadian radioactive waste disposal site. Journal of Ecological Modelling, 46, 195–211. Bobba, A.G. and Singh, V.P. (1995) Groundwater contamination modelling. In: V.P. Singh (ed.), Environmental Hydrology. Dordrecht, The Netherlands: Kluwer Academic Publishers, pp. 225–319.

Water pollution and its numerical modeling 219 Bobba, A.G., Jeffries, D.S., Booty, W.G., and Singh, V.P. (1995) Watershed acidification modelling. In: V.P. Singh (ed.), Environmental Hydrology, Dordrecht, The Netherlands: Kluwer Academic Publishers, pp. 13–68. Bobba, A.G., Singh, V.P., and Bengtsson, L. (2000a) Application of environmental models to different hydrological systems. Journal of Ecological Modelling, 125,15–49. Bobba, A.G., Singh, V.P., Berndtsson, R., and Bengtsson, L. (2000b) Numerical simulation of saltwater intrusion into Laccadive island aquifers due to climate change. Journal of Geological Society of India, 55, 589–612. Chachadi, A.G. and Teresa, L. (2002) Health of the groundwater regime in a coastal delta of east Godavari, Andhra Pradesh. Coastin, A Coastal Policy Research Newsletter, Published by TERI, 9607/13, Multani Dhanda, Paharganj, New Delhi-110 055, India, pp.5–8. Fano, E., Brewster, M., and Thompson, T. (1985) Water quality management in developing countries, Part 1. In: Proceedings of the 5th World Congress on Water Resources, Brussels, Belgium, pp. 641–654.

Author index

Aaronson, J.A. 199 Abdulla, F.A. 17 Adam, J.C. 23 Agassi, M. 193, 198 Agbu, P.A. 8, 22 Akao, Y. 62, 70 Alexander, M.A. 115 Allan, R. 99, 114 Allen, R.G. 73, 75, 82, 88, 89, 90 Anderson, H. 90 Anonymous 158, 169 Anthony, G.J. 55, 59 Arkin, P.A. 125, 134 Armstrong, R.L. 116 Arya, L.M. 29, 47 Aswathanarayana, U. 210, 218 Azzalini, A. 106, 115 Baldi, M. 117, 122, 133 Bandeen, W. 169 Barlow, M. 142, 153 Basist, A. 92, 95 Bastiaanssen, W.G.M. 73, 74, 75, 78, 90 Baumer, O.W. 30, 47 Becker, F. 8, 22 Beljaars, A. 29, 48 Bell, R.S. 47 Ben-Asher, J. 195, 196, 198 Bengtsson, L. 219 Bennett, A.F. 29, 47 Bennett, J.D. 116 Berbery, E.H. 153 Bergman, R. 70 Berliner, P.R. 193, 195, 196, 198, 199 Bernard, R. 29, 47 Berndtsson, R. 219 Berz, G.A. 135, 153 Bindlish, R. 93, 95 Bobba, A.G. 201, 210–216, 218, 219 Boer, G.J. 125, 127, 133 Booty, W.G. 219 Bosilovich, M.G. 48 Bouttier, F. 29, 36, 41, 47

Bowman, A. 106, 114 Brakensiek, D.L. 23 Bratseth, A.M. 37, 47 Brauner, C. 148, 154 Brewster, M. 219 Bruckler, L. 29, 47 Bruins, H.J. 193, 198, 199 Brunetti, M. 119, 133 Burges, S.J. 22 Burn, D.H. 149, 154 Burrough, P.A. 171, 189 Camp, C.R. 90 Cane, M.A. 137, 153 Capecchi, V. 133 Caporali, E. 47 Carlson, B.E. 133 Casady, G.M. 70 Castelli, F. 41, 47 Cayan, D.R. 99, 114, 137, 140, 143, 153 Chachadi, A.G. 213, 214, 219 Chang, A.T.C. 95, 134 Chang, E.K. 118, 122, 133 Charney, J.G. 28, 47 Chase, E.B. 154 Chauvin, F. 134 Chen, J. 118, 133 Cherkauer, K.A. 22 Chichilnisky, G. 136, 153 Chow, V.T. 136, 154 Christensen, J.H. 119, 133 Christensen, O.B. 119, 133 Clark, M.P. 97, 98, 99, 115, 116 Cochran, B.J. 116 Cohn, S.E. 44 Colacino, M. 133 Compo, G.P. 115 Cook, K.H. 120, 134 Cosgrove, B.A. 45, 47 Cotterman, H. 70 Courtier, P. 36, 41, 47 Cox, C.A. 70 Coy, L. 48

222

Author index

Craven, P. 109, 115 Crisci, A. 133 Cromp, R.F. 55, 59 Cunderlik, J.M. 149, 154 Czajkowski, K.P. 22 Daley, R. 29, 47 Dalu, G.A. 120, 133 da Silva, A. 46, 48 Davis, L.S. 59 Decker, D. 62, 70 Dee, D.P. 46, 47 Deekshatulu, B.L. 49 Del Genio, A.D. 117, 133 Delworth, T.L. 154 De Troch, F.P. 96 Derber, J. 48 Dettinger, M.D. 99, 115 Dhruva Narayan, V.V. 167, 169 Diaz, H.F. 98, 115, 116 Dima, I.M. 120, 133 Doherty, N. 147, 154 Douville, H. 120, 133 Downton, M. 135, 154 Dracup, J.A. 115, 137, 154 Droppelmann, K.J. 193, 197, 198, 199 Drusch, M. 92, 95 Dubayah, R.O. 22, 23 Dubois, P.C. 92, 95 Dunne, K.A. 154 el-Ashry, M. 97, 115 Engman, E.T. 95 Entekhabi, D. 29, 47, 48 Ephrath, J.E. 199 Evenari, M. 195, 199 Evensen, G. 40, 47 Fano, E. 216, 219 Feddes, R.A. 90 Flato, G.M. 125, 127, 133 Folland, C.K. 118 Forsberg, K. 62, 70 Francis, R.C. 115, 154 Freeman, P.K. 146, 147, 149, 154 Fu, Y. 118, 133 Fukunaga, K. 108, 115 Fulp, T. 115 Gaetani, M. 133 Gangopadhyay, S. 116 Georgakakos, A. 150, 155 Georgakakos, K.P. 30, 47 Gershunov, A. 98, 115 Gibbons, D. 97, 115 Gillet, N.P. 118, 133

Gilruth, P. 61, 70 Goward, S.N. 8, 23 Gozzini, B. 117 Grantz, K. 97, 103, 104, 114, 115 Grayson, R.B. 48 Greenstone, R. 169 Gregory, K.J. 171, 189 Grody, N.C. 95 Groisman, P. Ya. 119, 133 Guarnieri, F. 133 Gulev, S.K. 118, 133 Gupta, S.K. 8, 22, 23 Gutzler, D.S. 137, 154 Haddeland, I. 21, 22 Halem, M. 47 Halpert, M.S. 99, 116 Hamlet, A.F. 97, 99, 115 Hare, S.R. 115, 154 Harmon, B. 90 Harnik, N. 118, 122, 134 Heal, G. 136, 137, 153, 154 Hegerl, G.C. 120, 134 Helsel, D.R. 106, 115 Hidalgo, H.G. 115, 137, 154 Hill, R.W. 76, 90 Hiner, M.M. 116 Hirsch, R.M. 106, 115 Hirschboeck, K.K. 139, 140, 142, 154 Hoerling, M.P. 99, 115, 118, 134 Hoffbeck, J.P. 55, 59 Hollingsworth, A. 35, 47 Holtslag, A.A.M. 90 Horton, G.A. 101, 115 Houghton, J.T. 117, 134 Houser, P.R. 25, 30, 33, 36, 37, 39, 41, 42, 47, 48 Houtekamer, P.L. 40, 47 Hsu, A.Y. 93, 95 Huang, C. 55, 56, 59 Huppert, D. 115 Hurrell, J.W. 118, 134 Iredell, L. 23 Jackson, T.J. 29, 47, 91, 93, 95 Jain, S. 137, 152, 154 James, M.E. 8, 22 Janicot, S. 120, 134 Janowiak, J.E. 120, 134 Jastrow, R. 47 Jeffries, D.S. 219 Ji, M. 154 Jones, C.D. 117, 134 Joshi, S.R. 212, 216, 218 Jung, T. 133

Author index 223 Kalluri, S. 61, 64, 70 Kalma, J.D. 48 Kalman, R.E. 36, 47 Kalnay, E. 46, 47, 102, 115 Kann, D.M. 154 Kelmelis, J.A. 66, 70 Kendall, M. 125, 134 Kharin, V.V. 134 Kind, M.D. 169 Kistler, R. 125, 134 Kleindorfer, P.R. 146, 154 Koch, R.W. 99, 116 Koike, T. 94, 95 Kostov, K.G. 29, 47 Kramber, W.J. 73, 85, 90 Kratz, D.P. 22 Kretschman, R.G. 116 Krishna Kumar, K. 116 Krishnan, R. 49, 51, 59 Kristrom, B. 137, 153, 154 Kumar, A. 115, 134, 154 Kunreuther, H.C. 135, 146, 154 Laguette, S. 70, Lakshmi, V. 3, 7, 14, 15, 19, 22, 23 Lall, U. 98, 106, 107, 115, 116, 135, 137, 140, 143, 145, 151, 152, 154, 155 Landgrebe, D.A. 55, 59 Landwehr, J.M. 102, 116 Leetmaa, A. 154 Lehmann, J. 199 Lettenmaier, D.P. 17, 22, 23, 115 Li, Z. 8, 22 Liang, X. 16, 17, 18, 22 Lindeberg, T. 57, 58, 59 Lindesay, J. 114 Linnerooth-Bayer, J. 135, 154 Lithwick, H. 198, 199 Livezey, R.E. 142, 154 Loader, C. 106, 107, 114, 115 Logsdon, S.D. 23 Lohmann, D. 17, 18, 22 Lonnberg, P. 35, 47 Lorenc, A.C. 37, 38, 47 Lorite Torres, I. 89 Lovenstein, H. 193, 195, 196, 197, 199 Ma, Y.M. 96 McCabe, G.J. 99, 102, 115 MacConnell, W.P. 85, 90 McGinnis, D.A. 116 McLaughlin, D.B. 48 Macpherson, B. 47 Mahfouf, J.-F. 47 Maidment, D.R. 154 Mantua, N.J. 99, 115, 138, 143, 154

Maracchi, G. 117 Markgraf, V. 115 Masutani, M. 154 Matheussen, B.V. 22 Maugeri, M. 133 Maurer, E.P. 17, 18, 21, 22, 23 Maynard, K. 118, 120, 123, 134 Mays, L.W. 154 Mehta, A. 23 Melis, T.S. 99, 115 Meneguzzo, F. 117, 133, 134 Menenti, M. 90 Menzinger, I. 148, 154 Miller, D.A. 18, 23 Milly, P.C.D. 29, 48, 143, 154 Minobe, S. 139, 143, 154 Mitchell, H.L. 40, 47 Moody, D.W. 154 Moon, Y. 116 Moore, G.A. 64, 70 Mooz, H. 70 Moran, J.W. 70 Morin, J. 198 Morse, A. 73, 74, 75, 78, 89, 90 Munich Re 135, 154 Murali Krishna, I.V. 171 Murnane, R.J. 151, 154 Myers, R.H. 116 Myers, S.L. 116 Nakamura, H. 47 Nanni, T. 133 Nativ, R. 197, 199 Navarra, A. 133 Newman, M. 99, 115 Nichols, N.K. 37, 38, 48 Nigam, S. 153 Nijssen, B. 17, 22, 23 Njoku, E.G. 47, 91, 94, 95 Noilhan, J. 47 Obasi, G.O.P. 168, 169 O’Donnell, G. 23 Oke, P.R. 48 Ord, J.K. 125, 134 Ottlé, C. 29, 48 Owe, M. 93, 95 Owosina, A. 106, 115 Pagano, T. 102, 115 Palmer, T.N. 119, 134 Paloscia, S. 95 Paniconi, C. 96 Paris, J.F. 47 Parker, D. 114 Parrish, D. 38

224

Author index

Pasqui, M. 117 Paulson, R.W. 154 Perumal, A. 172, 189 Peterson, T.C. 95 Pfaendtner, J. 23 Piani, F. 117 Piechota, T.C. 97, 115 Pielke, R.A. 117, 134 Pielke, R.A. Jr 135, 154 Pimentel, D. 171, 189 Piraino, P. 23 Pitlick, J. 116 Pizarro, G. 135, 140, 143, 145, 151, 154 Porporato, A. 137, 144, 155 Prairie, J.R. 107, 115 Preisendorfer, R.W. 108, 115 Prevot, L. 29, 48 Prihodko, L. 8, 23 Prince, S.D. 7, 8, 14, 15, 23 Pulwarty, R.S. 99, 115, 116 Radakovich, J.D. 45, 48 Rajagopalan, B. 97, 106, 109, 110, 115, 116 Ralsanen, J. 119, 134 Ram Babu 167, 169 Rao, C.R. 106, 108, 118 Rao, D.P. 157, 168, 169 Raschke, E. 22 Rasmussen, E.M. 99, 116 Rawls, W.J. 18, 23 Reddy, P.R. 159, 160, 163, 165, 166, 170 Redmond, K.T. 99, 116, 153 Regonda, S. 97, 99, 108, 116 Reichle, R.H. 40, 41, 42, 43, 48 Reuter, D. 23 ReVelle, J.B. 62, 70 Rhoads, J. 22, 23 Rial, J.A. 117, 134 Richter, J.C. 47 Riddle, L.G. 153 Ridolfi, L. 137, 144, 155 Rieker, J. 102, 116 Roberts, R.S. 154 Rodell, M. 45, 48 Rokke, L. 23 Romero, M.G. 76, 90 Rood, R.B. 23, 33, 48 Ropelewski, C.F. 99, 116 Rosenfield, J. 23 Rowell, D.P. 120, 134 Royer, J.-F. 134 Rui, H. 154 Ruprecht, E. 133 Sadler, E.J. 90 Sanchez, P.A. 195, 199

Sankarasubramanian, A. 152, 155 Santer, B.D. 118, 134 Saranga, Y. 199 Satalino, G. 95 Sauerhaft, B. 193, 199 Schlesinger, M.E. 134 Schnur, R. 17, 23 Schubert, S. 8, 23 Schuurmans, J.M. 29, 48 Seaman, N.L. 38, 48 Seelan, S.K. 64, 65, 70 Serreze, M.C. 98, 99, 115, 116 Shainberg, I. 198 Shanan, L. 199 Sharma, A. 106, 115 Shiklomanov, I.A. 166, 170 Sinclair, F.L. 199 Singh, V.P. 201, 218, 219 Singhrattna, N. 107, 114, 116 Slack, J.R. 102, 116 Smith, E.A. 7, 23 Souza, F.A. 98, 106, 116 Stauffer, D.R. 38, 48 Stipple, J. 146, 148, 149, 155 Stott, P.A. 133, 134 Strzepek, K. 116 Su, Z. 96 Sultan, B. 120, 134 Sun, C. 43, 44, 45, 48 Susskind, J. 7, 8, 14, 15, 19, 22, 23 Swiss Re 147, 148, 155 Tadmor, N. 199 Tasumi, M. 73, 74, 89, 90 Teresa, L. 213, 214, 219 Thompson, T. 219 Thornbrugh, C. 154 Thurrow, T. 199 Todling, R. 47 Toll, D.L. 47 Toth, Z. 109, 116 Toutenburg, H. 106, 108, 116 Townshend, J.R.G. 59 Trezza, R. 76, 81, 89, 90 Troch, P.A. 96 Tucker, C.J. 8, 23 Turner, M.R.J. 39, 48 Ugland, R.C. 102, 116 Ursino, N. 48 van de Griend, A.A. 95 Vandermeer, J. 197, 199 van Zyl, J. 95 Vauclin, M. 47 Verhoest, N.E.C. 95, 96

Author index 225 Vidal-Madjar, D. 29, 47, 48 Vinnikov, K.Y. 93, 96 Viterbo, P. 29, 48 Vizy, E.K. 120, 134 von Storch, H. 108, 116

Witono, H. 29, 47 Wood, A.W. 17, 22, 23 Wood, E.F. 22, 23, 95 Wootton 89 Wright, J.L. 78, 79, 90

Walker, J.P. 25, 26, 29, 30, 33, 36, 39, 41, 42, 48 Wallace, J.M. 115, 120, 133, 154 Walling, D.E. 171, 189 Walpole, R.E. 108, 116 Wang, W. 134 Ward, M.N. 120, 134 Weaver, A.J. 133 Webb, R. 99, 115 Wen, J. 93, 96 Wetherald, R.T. 154 Whaba, G. 109, 115 White, R.A. 18, 23 Wigneron, J.P. 94, 96 Wilber, A.C. 22 Wilks, D. 109, 116 Willgoose, G.R. 48 Williams, C.N. 95 Williamson, R.A. 69 Wilson, E.A. 116

Xiang, X. 7, 23 Xie, P. 125, 134 Xu, T. 134 Yang, J. 117, 119, 134 Yao, H. 150, 155 Yarnal, B. 98, 116 Yates, D.S. 106, 116 Ye, K. 116 Yoder, R.E. 90 Zagona, E. 97, 115 Zartarian, V.G. 23 Zebiak, S.E. 116, 137, 153 Zhan, X. 42, 48 Zhang, Y. 154 Zhong, M. 115 Zohar, Y. 195, 199 Zwiers, F.W. 108, 116, 133, 134

Subject index

active microwave satellite observing system current active systems 94–95 projected active systems 93, 95 ADJOINT xvii ADS 176 AIRS/AMSU xvii, 5, 6 Akaike Information Criteria (AIC) 108 ANN xvii, 49 anthropogenic impacts on coastal watersheds aquatic pollution 211 cesspools 210 garbage dumps 210 Godavari Delta (India) 213–216 oil and gas pollution 211 Port Granby (Canada) 216 salt water intrusion 211–212 sanitary landfill 210 septic tanks 210 AOGCMs 119 application development life cycle production and deployment 63 prototype development 63 technical feasibility 62–63 user needs 62–63 applications of evapotranspiration methodology Idaho (USA) 82 Imperial Valley (USA) 88 lysimeters 76, 78 Middle Rio Grande (USA) 88 water budget 82 water rights 82 aquifer depletion monitoring 85–87 ARCEDIT 176 ARC GIS 176 ARC/INFO 176 Arno River Basin (Italy) 123–133 ASCE-EWRI 75, 78 ASTER xvii, 5, 6 automated cartography 175 AVHRR xvii, 4, 5, 8, 15 AVIRIS xvii

barriers in remote-sensing technology transfer collection of metrics 68 data accessibility 65 data continuity 66 data formats and standards 66–67 incubation period 67–68 proven benefits 67 sustainability 68–69 training and education 64–65 user involvement 64 basin studies Gunnison (USA) 104–106 Truckee–Carson (USA) 103–104 Bureau of Reclamation 101 case studies Arno River (Italy) 123–133 Godavari Delta (India) 213–216 Gunnison (USA) 102, 104, 105, 112, 113 Idaho (USA) 78, 80 Imperial Valley (USA) 88 Middle Rio Grande (USA) 88 Port Granby (Canada) 216 Truckee–Carson (USA) 99–101, 103, 104, 110, 111 Turkana district (Kenya) 197, 198 Upper Mississippi River (USA) 21, 22 Veligonda Irrigation Project (India) 185–188 Cat Bonds 148, 149 CCCma 120 CEOP xvii CGCM2-A2 127, 128 climate change 143–144 climate diagnostics 102–106 climate predictability anthropogenic climate change 143–144 ENSO/PDO seasonal prediction 143 flood frequency 137–145 streamflow forecasting 98 climate variability sources of variability 137–138 teleconnections with SSTs 137–138

228

Subject index

coastal watersheds geochemical processes 209–210 geologic setting 208 hydrological pathways 208 hydrologic system setting 208–209 contaminant transport 209 cost-effectiveness 172 cost savings 88 covariance matrix 56 DAAC xvii, 3 DAO xvii data assimilation methodologies direct observer assimilation 31 dynamic observer assimilation 32 quality control in assimilation 34 validation using data assimilation 34–35 database 173 data classification methods contextual classifier 54 fuzzy classifier 54–55 ISODATA 53 knowledge-based methods 53 maximum likelihood classifier 52–53 neural network classifiers 53 data handling methods active contour models 58–59 computer vision-based methods 57 Hough transform 57–58 hyperspectral sensor data handling 55 scale space approach 57 DBMS xvii Decision Tree Classifier (DTC) 54 DIAGNOSTIC xvii Digital Image Processing 49 direct observer assimilation methods 3D variational 38 analysis correction 38 direct insertion 36 Kalman filter 38–40 nudging 38 optimal interpolation 37 statistical correction 37 successive correction 37 droughts 168–169 dynamic flood risk management event warning and response 152 hazard preparation 153 hazard response 153 long range planning and analysis 151 season ahead preparation 151–152 dynamic observer assimilation methods – case studies downscaling with data assimilation 42 skin temperature assimilation 45–46

snow assimilation 42–45 soil moisture assimilation 41–42 El Niño 137 ensemble streamflow forecasts local methods 106–108 multisite ensemble forecast 108 subset selection 108–109 ENSO 98, 120, 142 ESRI 176 evapotranspiration computing 73, 84, 85 mapping 73–89 extreme events 20–21 financial risk of disaster losses 135 flood frequency 137–145 flood insurance 147 flood occurrence 139–143 flood risk management 145–149 forecast models 106–109 gain matrix 31 GAPP xvii GCMs xvii, 5, 44 Generalized Cross Validation (GCV) 107 GEOS xvii geospatial information technology 171–189 GEWEX xvii, 6 GIS database creation 174 GIS for watershed management 177–181 GIS layer creation 180 GIS packages 175 GIS project cost scenarios 176 GIS topology 180 global analysis 9 Godavari Delta (India) groundwater geochemistry 213, 214 location 213 simulated hydraulic head 215, 216 GOES xviii, 5, 6 GPCC xviii, 125 GPCP 125 groundwater assessment of groundwater resources 158–159 factors controlling groundwater regime 159 groundwater prospect maps 161–163 mapping geological structures 159, 160 recharge to groundwater 161 GSFC xviii HadCM3 120 HCDN 102 HIRS2-MSU 8 HOWI 121

Subject index 229 hydrological cycle global variability 119–123 regional variability 119–123 hydrological modeling applications 18 forcing 18 model parameters 18 observations 18 hydrologic data assimilation 28–30 IDRISSI 175 integrated GIS 175 IPCC 143, 144 ITCZ 120, 121 Kalman filter 29 LAI xviii land information systems 174 Landsat 157 land use–land cover 84 La Niña 137 LDAS xviii LOCFIT 107, 114 lysimeters Kimberly, Idaho 78–80, 82 Montpelier, Idaho 76, 78 METRIC 73, 74 microwave remote sensing 91–92 MLC xviii modeling studies streamflow and soil moisture 18 surface temperature 19–20 model validation likelihood function Skill Score (LLH) 109 ranked probability Skill Score (RPSS) 109 skill measures for validation 109–110 MODIS xviii, 4 monitoring of water rights 82–84 monsoons 122 MTF xviii, 51 NAM 118 NAO 118 NASA 61 NCEP–NCAR 118, 122, 125 NDVI xviii, 4 NNC xviii NRC 64, 66, 145, 153 numerical water quality modeling 212–216 Pacific Decadal Oscillation (PDO) 98, 138, 139 PAMAP 175

passive microwave satellite observing system advanced microwave scanning radiometer 93–94 projected passive system 94–95 special satellite microwave imager 92–93 TRMM xviii, 93 Pattern Recogntion (PR) xviii, 49 PMI 62 pollutants in coastal watersheds biological pollutants 204 chemical pollutants in surface waters 207 non-point source pollution 204 organic pollutants 206 point source pollution 203–204 processes and activities 206–207 sources of pollutants 203 thermal pollution 206 types of pollutants 202–203 Port Granby (Canada) disposal site 216 Lake Ontario 216 Ra-226 concentrations 217, 218 Principal Component Analysis (PCA) 105 PROGNOSTIC xviii raster data 177 regional regression analysis 12, 14 reinsurance 148 remote-sensing applications drought management 168–169 flood management 168 groundwater studies 158–163 irrigation water management 166–167 rainfall, snow, and glacier studies 163, 166 reservoir sedimentation 167 surface water resources 157 water quality 169 watershed management 167–168 remote-sensing data characteristics geometric correction 50 radiometric correction 51 remote-sensing datasets 4 runoff agroforestry system 195–197 runoff harvesting in ancient times 194–195 salt water intrusion 211–212 satellite data validation 7 scaling 6 SEBAL 73, 74 securitization 147–149 skin temperature 28 snow 28 soil cover complex climate 182 drainage 182

230

Subject index

soil cover complex (Continued) estimation of runoff 183–184 estimation of soil loss 184–185 soil loss equation 184 time of concentration 183 vegetation cover 182 soil moisture 18, 28 SPANS 125 spatial analysis 174 spatial and atribute data 173 SSMI 5, 92 SSTAs 105, 120, 122 SVM xviii

UNICEF 169 Upper Mississippi River 21, 22

tangent linear model 31 TIN 177 TOVS xviii, 5, 6, 8, 15 traditional flood insurance 147 TRMM xviii, 5 Turkana district, Northern Kenya 197, 198

WAM 121 water quality models 212–216 watershed characteristics 181–182 watershed management 171 WCRP xviii western US hydroclimatology 99

validation 76 VCL xviii, 5 vector data model 177 Veligonda Irrigation Project (India) biophysical setting 185–186 gully control measures 187–188 land-use/land-cover studies 186–187 preparation of maps 186 VIC xviii