Use of Satellite and In-Situ Data to Improve Sustainability
Use of Satellite and In-Situ Data to Improve Sustainability
Edited by
Felix Kogan National Oceanic & Atmospheric Administration (NOAA/NESDIS) Center for Satellite Application and Research (STAR) Washington DC, USA
Alfred M. Powell, Jr. National Oceanic & Atmospheric Administration (NOAA/NESDIS) Center for Satellite Application and Research (STAR) Washington DC, USA
Oleg Fedorov Space Research Institute of the National National Space Agency of Ukraine Kiev, Ukraine
Published in Cooperation with NATO Public Diplomacy Division
Proceedings of the NATO Advanced Research Workshop on Using Satellite Data and In-Situ Data to Improve Sustainability Kiev, Ukraine, 9-12 June 2009
ISBN 978-98-481-9620-3 (PB) ISBN 978-90-481-9617-3 (HB) ISBN 978-90-481-9618-0 (e-book)
Published by Springer, P.O. Box 17, 3300 AA Dordrecht, The Netherlands. www.springer.com
Printed on acid-free paper
All Rights Reserved © 2011 Springer Science+Business Media B.V. No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form-or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
PREFACE
The Advanced Research Workshop (ARW) “Using Satellite and In situ Data to Improve Sustainability” was held in Kyiv, UKRAINE during June 9–12, 2009. The ARW was sponsored by the North Atlantic Treaty Organization (NATO) and organized by the National Space Agency of Ukraine (NSAU) in cooperation with the Center for Satellite Application and Research (STAR) of the National Environmental Satellite Data and Information Services (NESDIS), National Oceanic and Atmospheric Administration (NOAA) http://www.star.nesdis.noaa.gov/smcd/emb/ vci/VH. Drs. Powell (NOAA) and Fedorov (NSAU) served as ARW Directors. The ARW was focused on the current issues of changing climate and providing services for sustainable economy, healthy environment and better human life and had the following sessions • • • • • •
Early warning of natural disasters Weather and food security Climate services to enhance national security Land cover/land change and anthropogenic activities Human health and the environment Satellite and in situ data records for trend analysis
Eighty five scientists from North America, Europe and Asia, attended the Workshop, participated in the discussions and gave 53 presentations. The brain-storming discussions at the end of each day session resulted in the ARW Summary. The following important issues were emphasized at the Workshop 1. More than 30-year satellite data time series have already provided sufficient information to be used for monitoring land, ocean and atmosphere, improving sustainable economy, environment and human life; these activities must be expanded, advertised and widely distributed. 2. The gap between research and applications for improving sustainability should be overcome by combining satellite and in situ data for enhancing spatial and temporal coverage of the Earth and expanding the products and services quantity and quality. 3. In order to expand the applications, satellite data sets and products must be presented in a ready-to-use form, easily available and be user friendly. 4. Existing and new satellite data and products must be validated and calibrated to enhance their credibility. 5. Following the ARW goals (climate services and sustainability), the current NOAA satellite data and products can be scaled into three categories: (a) mature science/ready to use (flash-flood, drought, snow, vegetation health); (b) intermedium
v
vi
6. 7.
8.
9. 10.
Preface
maturity science (agriculture, land-water interactions, sensitivity of world ecosystem to ENSO) and (c) prospective science/need development (climate forcing, land cover/land change, anthropogenic activities, loss of bio-productivity, bio-productivity potential, wave analysis and external climate forcing). In situ data must be actively collected for validation and calibration of satellite datasets and products. Create regional polygon (Kherson administrative region, which combines Black Sea open water and costal ecosystems) and collect comprehensive data sub-sets (satellite, in situ, economic) with easy access for scientists in order to develop and test new methods, data, models and products. Create working groups covering (a) climate science (change impacts, prediction from forcing); (b) economic effectiveness of climate services; (c) satellite/in situ data assimilation; (d) bio-productivity potential. Implement NOAA satellite-based technique and products for monitoring droughts, flash-flood rainfall, vegetative health, and snow cover. Approve the first year of Cooperation (MOU) between the National Oceanic and Atmospheric Administration and the National Space Agency of Ukraine. Considering the Cooperation success continue the MOA activities for the next 4 years.
Cooperation between National Oceanic and Atmospheric Administration (NOAA) and the National Space Agency of Ukraine (NSAU) started in June 2008 when NOAA Administrator and Director General of NSAU signed the Memorandum of Agreement (MOA) to promote the application of NOAA operational environmental satellites for climate services in order to achieve sustainability. The MOA was under the umbrella of the Global Earth Observing System of Systems (GEOSS) and Group on Earth Observations (GEO). The most advanced thirty two papers discussing the basic science, new methods, datasets, products and applications were selected to be including in this book. Scientific and application results presented in these papers can be used today for an early detection of large-scale natural disasters, assessments of agricultural production losses, monitoring fires, climate and land surface trend analysis, application of climate forcing for lead-time predictions and others. This book consists of the following five parts and appendix Part I: Large-Scale Weather Disasters: Early Detection and Monitoring from Space and In Situ Data Part II: Environment and Food Security: Diagnosis and Prediction Part III: Climate Change, Environment and Socioeconomics Part IV: Marine Ecosystem, Land Ccover, Atmosphere and Anthropogenic Activities Part V: Satellite and In Situ Long Records for Trend Analysis, Modeling and Monitoring Part I consists of six chapters discussing such large-scale disasters as drought, flood, severe weather and fires. Several methods are presented showing both satellite and in situ data used for monitoring and assessments. Part II consissts of six chapters discussing mostly food security issues in terms of monitoring large-scale
Preface
vii
agricultural production from satellite and in situ data. The authors share their experience in Ukraine, Russia and USA. Part III consists of seven chapters discussing climate change issues and implications for a reduction of glassiers, changes in bio-climatic potential, crop yield variations under different IPCC scenarios etc. In addition, some aspects of climate change origination and climate forcing impact on land surface are also analysed and presented. Part IV consists of seven chapters discussing marine ecosystem issues such as interaction between land and coastal water. Also, some papers are devoted to desertification issue, geomagnetic activities and others. Part V consists of six chapters discussing long-term satellite-based time series records for monitoring changes in land surface, comparing the records produced from the same source but with different processing algorithms. The appendix contains color images for several papers. Felix Kogan Alfred Powell Oleg Fedorov
CONTENTS
PART I
LARGE-SCALE WEATHER DISASTERS: EARLY DETECTION & MONITORING FROM SPACE & IN SITU DATA
Monitoring Droughts and Impacts on Crop Yield in Ukraine from Weather and Satellite Data .................................................................. Tatyana Adamenko and Anatoly Prokopenko
3
Early Detection and Monitoring Droughts From NOAA Environmental Satellites .................................................................... Felix Kogan and Wei Guo
11
Flood Monitoring from SAR Data................................................................. Nataliia Kussul, Andrii Shelestov, and Sergii Skakun Satellite Rainfall Information for Flood Preparedness and Response ................................................................................................... Robert J. Kuligowski
19
31
Monitoring Severe Weather in UKRAINE with Satellite Data .................. Oleksiy Kryvobok, Mykola Kulbida, and Ludmila Savchenko
41
Daily Fire Occurrence in Ukraine from 2002 to 2008 ................................. Wei Min Hao, Shawn P. Urbanski, Bryce Nordgren, and Alex Petkov
49
Satellite-Based Systems for Agro-meteorological Monitoring .................... Alexander Kleschenko, Oleg Virchenko, and Olga Martinenko
61
PART II
ENVIRONMENT AND FOOD SECURITY: DIAGNOSIS AND PREDICTION
Monitoring Droughts and Pastures Productivity in Mongolia Using NOAA-AVHRR Data ...................................................... Leah Orlovsky, Felix Kogan, Eldad Eshed, and Chultem Dugarjav
69
Satellite-Derived Information on Snow Cover for Agriculture Applications in Ukraine ....................................................... Peter Romanov
81
ix
x
Contents
Grain Yield Prediction in the Russian Federation ....................................... Anna Strashnaya, Tamara Maksimenkova, and Olga Chub Satellite-Based Crop Production Monitoring in Ukraine and Regional Food Security ........................................................................... Felix Kogan, Tatiana Adamenko, and Mikola Kulbida
93
99
New Regression Models for Prediction of Grain Yield Anomalies from Satellite-Based Vegetation Health Indices ....................... 105 Gennady Menzhulin, Natalya Shamshurina, Artyom Pavlovsky, and Felix Kogan Phytosanitary Situation of Agrocenosis in Ukraine and New Technologies for Monitoring Harmful Organisms .............................. 113 Vladimir Chayka, Tatiana Neverovska, Nelia Prokopiuk, and Olga Baklanova PART III
CLIMATE CHANGE, ENVIRONMENT AND SOCIOECONOMICS
30-Year Land Surface Trend from AVHRR-Based Global Vegetation Health Data ................................................................................... 119 Felix Kogan Global Warming, Atlantic Multi-decadal Oscillation, Thermohaline Catastrophe and Their Impact on Climate of the North Atlantic Region .......................................................................... 125 Alexander Polonsky Global Warming and Possible Changes in the Recurrences of Grain Crops Anomalies.............................................................................. 145 Gennady Menzhulin and Artyom Pavlovsky Regime Shifts in the Atmosphere and Their Relationship to Abrupt Ocean Changes .............................................................................. 151 Alfred M. Powell, Jr, Jianjun Xu, and Ming Chen Glacier Degradation from GIS and Remote Sensing Data ......................... 159 Azamat Tynybekov ENSO Impact on Vegetation .......................................................................... 165 Felix Kogan Bio-climatic Potential of Russia and Climate Change ................................. 175 Alexander Kleschenko
Contents
xi
PART IV
MARINE ECOSYSTEM, LAND COVER, ATMOSPHERE & ANTHROPOGENIC ACTIVITIES
Consequences of Land and Marine Ecosystems Interaction for the Black Sea Coastal Zone ...................................................................... 181 Vladimir Kushnir, Gennady Korotaev, Felix Kogan, and Alfred M. Powel, Jr Utilizing Satellite Data to Highlight High Ozone Concentration Events During Fire Episodes.......................................................................... 191 Rasa Girgždienė and Steigvilė Byčenkienė Geomagnetic Disturbances and Seismic Events in the Vrancea Zone from in Situ Data.................................................................... 199 Frina Sedova, Vladimir Bakhmutov, and Tamara Mozgovaya First Step Towards Monitoring Surface Ozone Dynamics at Ukrainian Stations ...................................................................................... 209 Oleg Blum, Vira Godunova, Volodymyr Lapchenko, Oleksiy Perekhod, Yaroslav Romanyuk, and Mikhail Sosonkin Satellite Monitoring of Nitrogen Oxide Emissions....................................... 219 Igor Konovalov, Matthias Beekmann, Andreas Richter, and John Burrows Detection of Desertification Zones Using Multi-year Remote Sensing Data ...................................................................................... 235 Lev Spivak, Irina Vitkovskaya, Madina Batyrbayeva, and Alex Terekhov Satellite Desertification Monitoring in Sahara ............................................. 241 Mikhail A. Popov, Sergey A. Stankevich, Alexei I. Sakhatsky, Menny O. El Bah, Daoud Mezzane, and Igor A. Luk’yanchuk PART V SATELLITE & IN SITU LONG RECORDS FOR TREND ANALYSIS, MODELING & MONITORING Global Vegetation Health: Long-Term Data Records ................................. 247 Felix Kogan, Wei Guo, and Aleksandar Jelenak Aero-Space Radar Online Monitoring of Disasters in Ukraine.................. 257 Mariya Belobrova, Dmitry Bychkov, Anatoly Boev, Alexandre Gavrilenko, Valentin Efimov, Alexandre Kabanov, Ivan Kalmykov, Alexandre Matveev, and Valery Tsymbal Comparison of AVHRR-Based Global Data Records .................................. 267 Felix Kogan, Marco Vargas, and Wei Guo
xii
Contents
Merging Remote Sensing and In Situ Data for Estimation of Energy Balance Components Under Climate Change Conditions: Ukrainian Steppe Zone .............................................................. 273 Tatiana Ilienko and Elena Vlasova Atmosphere Aerosol Properties Measured with AERONET/PHOTONS Sun-Photometer over Kyiv During 2008–2009 .......................................................................................... 285 Vassyl Danylevsky, Vassyl Ivchenko, Gennadi Milinevsky, Michail Sosonkin, Philippe Goloub, Zhengqiang Li, and Oleg Dubovik Global Distribution of Magnetic Storm Fields and Relativistic Particles Fluxes ................................................................... 295 Olga Maksimenko and Galyna Melnyk Appendix 1 ....................................................................................................... 305 Appendix 2 ....................................................................................................... 306 Appendix 3 ....................................................................................................... 307 Appendix 4 ....................................................................................................... 308 Appendix 5 ....................................................................................................... 309
Part I
Large-Scale Weather Disasters: Early Detection & Monitoring From Space & In Situ Data
Monitoring Droughts and Impacts on Crop Yield in Ukraine from Weather and Satellite Data Tatyana Adamenko and Anatoly Prokopenko
Abstract The agrarian sector is an important component of the Ukrainian economy. Within this sector about 20% of able-bodied citizens of the country are employed and 12–15% of the gross domestic product is formed. During previous years simultaneous with the tendency of improved agricultural technology, the significant fluctuations of crops productivity owing to weather conditions, first of all owing to droughts, increase. Their frequency and intensity rise. Every 2–3 years the drought covers up to 20–40% of the country territory. Crop losses owing to droughts can run up to 50% and more. There is a dangerous tendency of drought distribution over territory which earlier belonged to a zone of sufficient moisture. In the Ukraine, climate fluctuations affecting crop production account for a 20–50% loss of winter crops and 35–75% loss of summer crops. Keywords Droughts • Agrometeorological monitoring • Yield • Crop forecasting • Modeling • Observing system
Introduction The southern areas of the Ukraine suffer from permanent moisture deficiency under sufficient heat provision. More often, annual precipitation in these areas is limited for sustained agricultural production. The situation becomes very acute in drought years. Optimum soil moisture deficit during the growing season for receiving top yield in the central and southern regions is 50–100 and 150–200 mm, respectively. As the adverse agroclimatic phenomena, droughts are inherent to the climate of the Ukraine. During past 30–50 years they became more frequent and intense, covering up to half of its territory every 10–12 years, and up to 20% every 2– 3 years. Drought-related crop losses in such years can reach 50%. In combination T. Adamenko (*) and A. Prokopenko Ukrainian Hydrometeorological Centre, Kyiv, Ukraine e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_1, © Springer Science+Business Media B.V. 2011
3
4
T. Adamenko and A. Prokopenko
yield t / ha
with anthropogenic factors, increases in drought occurrence and intensity leads to land degradation and desertification. The following are the economic reasons the irrigated areas in Ukraine is reducing. In 2008, droughts started to appear even in the zone of Polesye, an obvious manifestation of climatic change and economic activities. Droughts even occurred in years with near normal precipitation leading to crop shortages. Figure 1 shows 0.5–1.5 t/ha crop reduction in the drought of 2003 compared to the non-drought year of 2008. Figure 2 shows that in the drought years of 1990–2008, winter wheat yield was reduced two to three times compared to non drought years.
5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 Winter wheat
Barley 2008
Corn
Sunflower
2003
Fig. 1 Crop yields in drought (2003) and non drought (2008) years in Ukraine
2008 2006 2004 2002 2000 1998 1996 1994 1992 1990 0
5
10
15
20
Fig. 2 Average Ukrainian winter wheat yield
25
30
35
40
45
Monitoring Droughts and Impacts on Crop Yield in Ukraine from Weather and Satellite Data
5
The Ukrainian Hydrometcenter monitors agrometeorological conditions in Ukraine, regularly observing the state of microclimate, soil moisture, crops, and pastures in a timely and objective manor. The information includes weather observations received from meteorological stations and the satellite data. The data are processed automatically with the aim of simulation of the production process and assessment of crop growth, development, and yield formation.
Methods Ukraine has sufficient experience in the development of various components of drought monitoring, and in carrying out field (ground point and en route) surveys of agricultural crops. The systematic study of droughts and hot winds in Ukraine has a long history. During this time many different criteria of drought, with definition of its types, including atmospheric, soil, and mixed ones, were suggested. The automated program of “drought” was developed on the basis of analysis on the current state of assessment of drought and hot winds, their influence on grain yield by the UkrHMC, in cooperation with the Odessa State Environmental University (OSENU) (A. Polevoy). This program is compatible with automated working place of an agrometeorologist (ARM-Agro), i.e. with the operative agrometeorological information and allows each decade to assess the impact of drought on crop cereals (winter wheat, spring barley) as shown in Fig. 3.
Fig. 3 Estimated reduction in yield of spring barley (%) due to drought in regions of Ukraine. Calculation June 30, 2009
6
T. Adamenko and A. Prokopenko
Study Area and Data Accounting of dry conditions is realized by decades of vegetation by means of analysing of several known methods of evaluation of dry events, which are adapted for different soil-climatic zones of Ukraine. The climatic peculiarities of the territory – moistening of arable (0–20 cm) and 1 m soil layers, the amount of precipitation and of air moisture deficit were assessed. In addition, the UkrHMC uses a well-known complex index of moistening and water-supply of plants – the hydrothermal coefficient – HTC (the ratio of total precipitation to the sum of temperatures above 10°C), which is also calculated in the operational mode. Taking into account the climate change, in particular, the air temperature increase with still persistent amount of precipitation, it is likely that droughts will grow stronger in Ukraine. Ukraine signed the UN Convention “On Combating Desertification”, but today very little attention to assessing and taking measures for mitigation of droughts is paid on the state level. Incorporation of all available know-how in a single monitoring system using the regular data of satellite observations will significantly increase the effectiveness of drought management, primarily by improving the opportunities of providing services and communicating agrometeorological information to end-users online. On ground agrometeorological monitoring is regularly carried out by 188 meteorological stations in Ukraine. In parallel with the meteorological observations, about 145 stations simultaneously monitor growth, development, condition, humidity, and security of crops according to a single method. Observations are carried out for the most common crops in the area of a meteorological station observation. They are winter wheat and spring barley, legumes, oilseeds, and industrial crops. Standard observation of weather elements is carried out every 3 h. The regularity of crop observations is every 2 days, the actual water-supply is defined once in 10 days. Meteorological and agrometeorological information is collected on an operational basis by the regional hydrometeorological centers and the Ukrainian Hydrometeorological Center. Data on the phenological development of crops, height, density, weed infestation, actual water-supply and condition of crops (a points-system) allows to create the general picture on the country. This information is the basis for crop estimating and forecasting. The information on the actual soil water-supply is especially important. Ukraine has accumulated an extensive and reliable data on agrometeorological observations. Its adequacy is proved by estimates obtained in different in degrees of favourability years.
Crop Yield Forecasting The harvest assessment is carried out with a temporal resolution of a decade. Issue (calculation, analysis, adoption of a final figure) of yield prediction and croppage takes into account critical periods of development of a specific culture as for its
Monitoring Droughts and Impacts on Crop Yield in Ukraine from Weather and Satellite Data
7
future harvest. Forecasting is carried out on several levels (weather station, district, region, country) using all the methods developed, tested, and adapted to the territory of Ukraine. A basic software package based on a dynamic model of crops productivity using the actual operational information was created at the UkrHMC. Standard indicators of decadal agrometeorological information – temperature and air humidity deficit, amount of precipitation, sun-shine duration; reserves of productive water-supply in the soil are used. Agro-physical and agrochemical characteristics of soil are taken into account. This information is transformed into a system of input parameters of the model. Calculations of primary characteristics of vegetation cover are realized (the size of the biomass of individual organs of plants – leaves, stems, roots, seeds), initial values of estimates of extreme conditions are given. The model estimates the influence of weather conditions on yield for any period of vegetation (a decade, an interphase period, an entire growing season), a possible reduction of yield due to unfavorable conditions for specific territory (frost, drought, dry winds). as shown in Fig. 4. The level and volume of harvest are calculated directly. Results of surface monitoring and forecasting are used for: • • • • • • •
Operational reporting to leaders of the country, ministries Decision-making on export, import Farmers Insurance companies Recommendations for optimizing of the cultivation of crops Determining of the need for fertilizers and chemicals Justifying the specialization of farms and agricultural cultures zoning
Fig. 4 Assessment of the impact of weather on winter wheat yield (in % of optimum conditions) in administrative regions on May 31, 2009
8
T. Adamenko and A. Prokopenko
In years with particularly unfavourable weather conditions, the meteorological stations carry out ground-route survey of agricultural crops. All the agrometeorological information comes to the regional centres and from there it goes to the UkrHMT.
Receipt and Processing of Data Processing of the agrometeorological observations is performed on an agrometeorologist AWP. The AWP was developed and implemented in the Ukrainian Hydrometeorological Center and regional centers for hydrometeorology over 9 years ago. An agrometeorologist automated working place (an agrometeorologist AWP) is a specialized program for an operative display of operational meteoro logical and agrometeorological information, processing, and systematization it for making the various operational and long term materials by agrometeorologists. An AWP ensures receiving of primary daily and decadal data from weather stations in the special code from the server connection. Then the processing of this information, checking, and editing it on the AWP screen, is carried out. AWP performs the following tasks: • Forming long-term observations databases on all types of information for all types of crops and research plots which were observed • Drawing any agrometeorological information for any date or period on a cartographic basis • Forming new types of maps for selected crops, areas, dates and other information by an agrometeorologist • Getting maps with drawing information from each meteorological station, averaging over areas and soil-climatic zones, etc. • Comparing the factual information with the normals • Generating and printing of various tables, the type and composition of which is formed by an expert agrometeorologist • Automated logging of observational data into introduction tables with the possi bility of sorting and filtering • Comparing of data with the standards and conducting any calculations and averaging • Carrying out the calculations of agrometeorological forecasts of: productivity, the timing of various phenological phases of agricultural cultures onset and drought manifestations or “molting”of grain and others AWP-Agro enables us to assess online the drought phenomena using various methods and evaluate possible losses of grain crops. Evaluation of a drought is done both using the actual data on air temperature, relative humidity and, more importantly, of water-supply. Current agrometeorological information enables us to assess online the impact of drought phenomena on the development and yield of field crops.
Monitoring Droughts and Impacts on Crop Yield in Ukraine from Weather and Satellite Data
9
Today the UkrHMC, having an adequate technical support and staffing, is actively pursuing work on adaptation of satellite data for assessing the status and productivity of crops in Ukraine. Development of monitoring system with regular data from satellite observations significantly will increase the effectiveness of drought management in Ukraine. First of all, it is the use of vegetation index in conjunction with other indicators, adaptation of the European system “Mars” together with the UHMI (Ukrainian Hydrometeorological Institute), as well as the developments of ARRIAM (All-Russia Research Institute of Agricultural Meteorology) (Russia).
Early Detection and Monitoring Droughts From NOAA Environmental Satellites Felix Kogan and Wei Guo
Abstract With nearly 30 years of the accumulated AVHRR data which were collected from NOAA operational polar-orbiting environmental satellites, the area of their applications expanded in the direction of monitoring vegetation condition, modeling agricultural production, analysis of climate and global change, resource management, and early and more efficient monitoring of droughts and their impacts on economy and society. This becomes possible due to the development of Vegetation Health (VH) indices. This paper discusses utility of the AVHRR-based VH focusing on monitoring vegetation with the emphasis on early drought warning and drought features. Keywords Droughts • Environmental satellites • Vegetation Health indices
Introduction Drought is a typical phenomenon of the earth’s climate. The losses from droughts are normally staggering. The average annual cost of drought in the United States of America, a country of the advanced technology, is around $6 billion. However, in extreme drought years such as 1988, the cost of the drought is five to eight times larger. In the recent years, large-scale intensive droughts were reported all over the globe. Developing countries of Africa and Asia were the most affected. For example, the Horn of Africa experienced droughts 6 years in a row which led to serious food shortages. Rangeland in Mongolia have also suffered from a very intensive droughts resulting in a lack of feed for the livestock. Unusual summer
F. Kogan (*) NOAA/NESDIS Center for Satellite Application and Research (STAR), Washington DC, USA e-mail:
[email protected] W. Guo IMSG Inc., Washington D.C., USA F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_2, © Springer Science+Business Media B.V. 2011
11
12
F. Kogan and W. Guo
dryness also affected the new countries (from the former USSR) in the Black Sea and Caspian Sea regions. Western USA experienced drought-induced forest fires for several years. In 2003 and 2007, nearly 20% of the global lands were drought-stricken. Weather data are traditionally used for drought monitoring. However, weatherderived drought-watch system has shortcomings. Weather data represent point locations rather than an area and meteorological stations are not uniformly distributed. Even in the United States, with a well developed weather network, the density of stations is not sufficient to characterize regional drought. For example, in South Dakota one weather station is normally used for monitoring drought on 1–2 million acres of crop land. In Ukraine, each of the 180 weather stations covers 0.8 million acres of land. The problem of insufficient density of weather stations becomes especially acute in the areas with marginal climatic resources such as Africa, Asia, Central and South America. This paper discusses a new satellite-based land-surface observing system used for drought monitoring. Principally, satellite data in the form of vegetation indices have been applied for land monitoring since the mid 1980s (Tarpley et al. 1984). In the last 20 years, NOAA has designed a new AVHRR-based Vegetation Health (VH) theory, system and data set (Global Vegetation Index, GVI-x) that showed success when applied to drought detection and tracking (Kogan 1990, 1997, 2001). Unlike other remote sensing techniques, the new method and system uses multi-spectral radiances and the main ecosystem laws for analysis of vegetation health in response to weather changes. During the last 8 years, this method was tested and validated thoroughly against ground data in all major agricultural countries of the world and proved to be of excellent utility for early drought detection, accurate monitoring of its development, affected area and impacts on agriculture, rangeland and forestry (Salazar et al. 2007). This paper presents the results of using the VH for monitoring droughts.
Satellite Data The GVI-x system was developed based on the NOAA AVHRR Global Area Coverage (GAC) data set. The GAC is produced by sampling and mapping the AVHRR 1-km daily reflectance in the visible (VIS, 0.58–0.68 mm, near infrared (NIR, 0.72–1.1 (mm), and two infrared bands (IR4, 10.3–11.3 and IR5, 11.5– 12.5 mm) to a 4-km map. The VIS and NIR reflectance were pre- and post-launch calibrated and the normalized difference vegetation index (NDVI) was calculated as (NIR-VIS)/(NIR+VIS). The IR4 emission was converted to brightness temperature (BT), which was corrected for non-linear behavior of the AVHRR sensor. Daily NDVI and BT were composited over a 7-day period by saving those values that have the largest NDVI for each map cell. The 1981–2010 NDVI and BT weekly time series were processed to remove high frequency noise, identify seasonal cycle and to calculate climatology.
Early Detection and Monitoring Droughts From NOAA Environmental Satellites
13
Vegetation Health Method A new method is based on the estimation of green canopy stress/no stress from AVHRR-derived indices, characterizing moisture, thermal conditions and total vegetation health (Kogan 1990, 1997, 2001). Unlike the two spectral channels approach (NDVI-based) routinely applied for vegetation monitoring, the new numerical method in addition to NDVI, also uses BT from 10.3–11.3 mm IR4 channel, which estimates the hotness of the vegetation canopy. In dry years, high temperatures, coupled with an insufficient water supply, lead to overheating of the canopy, which intensifies negative effects of moisture deficit impact on vegetation. The three-channel algorithm consists of comprehensive processing of NDVI and BT annual time series, which includes complete removal of high-frequency noise, enhancing seasonal cycle, calculation of climatology and single out medium-to-low frequency fluctuations associated with weather impacts on vegetation (Kogan 1995, 1997,). This procedure was formalized by Eqs. (1)–(3), where climatology was represented by the difference between 22-year absolute maximum and minimum both NDVI and BT values for each pixel and week.
VCI = 100 ∗ (NDVI − NDVI min ) (NDVI max − NDVI min )
(1)
VCI = 100 ∗ (BTmax − BT ) (BTmax − BTmin )
(2)
VHI = a ∗ VCI + b ∗ TCI
(3)
where NDVI, NDVImax, and NDVImin (BT, BTmax, and BTmin ) are the smoothed weekly NDVI (BT), their multi-year absolute maximum and minimum, respectively; a and b = 1 − a are coefficients quantifying a share of VCI and TCI contribution in the total vegetation health. The VCI (Vegetation Condition Index), TCI (Temperature Condition Index) and VHI (Vegetation Health Index) are indices estimating cumulative moisture, temperature and total vegetation health conditions, respectively on a scale from zero (extreme stress) to 100 (favorable condition) with 50 corresponding to the average condition.
Global Droughts The VH system has been used successfully for monitoring vegetation health, including drought-related vegetation stress around the world since the 1990s. These data were presented at http://www.star.nesdis.noaa.gov/smcd/emb/vci/VH/vh_currentImage.php. Examples of vegetation health (condition) in mid 2007 and 2008 are shown in Fig. 1. As seen, intensive summer drought-related vegetation stress were observed in the western USA, southern Ukraine and Russia, Mongolia (Northern Hemisphere, summer) and in Argentina (2008), western Australia, Brazil and Southern Africa (Southern Hemisphere, winter). These estimates are supported by in situ observations (Le Comte 2008, 2009).
14
F. Kogan and W. Guo
−135
45
135
−45
45
135
45
45
−45
− 45
− 45
−15
−15
15
15
2007
45
45
−135
− 45
− 45
−15
−15
15
15
2008 −135 STRESSED 6
− 45
45
135 FAVORABLE
FAIR 12
24
36
48
60
72
84
Vegetation Health in mid-July Fig. 1 Vegetation health index in mid July 2007 and 2008 (Color image is provided in Appendix 1)
It should be emphasized that satellite data in addition to drought start/end, d uration and area, also estimate drought intensity. In the USA, four categories characterize drought intensity (severity): moderate (D1), severe (D2), extreme (D3) and exceptional (D4). In terms of precipitation deficit they are characterized by <70% of normal precipitation for 3 months in a row (D1), <65% for 6 months (D2), < 60% for 6 months (D3) and <65% for 12 months (D4). Satellite data show that these criteria correspond to VHI values below 35, 25, 15 and 5, respectively. Moreover, the repetition of such droughts might be, once per 2–4, 5–10, 11–20 and more than 20 years, respectively. Following these criteria it is easy to estimate area under drought of different intensity. For example, since the new millennium has started, severe-to-exceptional (D2–D4) drought area covered 7–10% of the world land, mostly in Northern Hemisphere (Fig. 2). However, in such years as 2003 and 2007 this area increased to 20%. The area of extreme-to-exceptional droughts (D3–D4) normally occupies 3–5% of global land, increasing to10% in the extreme years (2003 and 2007).
Early Detection and Monitoring Droughts From NOAA Environmental Satellites
15
Fig. 2 Percent of the World and Northern Hemisphere areas under severe-to-exceptional drought
Regional Droughts Ukraine was selected to investigate regional droughts. Having the best for agriculture chernozem soils, Ukraine has some misfortune to be located in a dry zone. Annual shortage of water (based on the difference between precipitation and potential evapotranspiration) is 200–400 mm. Weather data show that droughts affect Ukraine every 2–4 years. Since weather station network is limited for the Ukraine area, VH products were used to assess drought area. Figure 3 shows percent of severe-to-exceptional (D2–D4) droughts since 1991, after Ukraine received independence. First, it should be emphasized that in some periods (1997–2000), the D2–D4 droughts missed Ukraine territory completely. However, in some periods (2001–2004 or 2006–2008), droughts followed every or every other years. Second, in drought years, 5–10% of Ukraine territory is normally affected. But this area might increase up to 20% in less favorable years which occurred 5 years out of 20. In the major crop area (southern part), drought affects 10–40% of the Ukraine territory and in extreme years such as 2007 the area might increase up to 60%. In smaller (state level) regions inside the major crop area (Odessa oblast, located in the southwestern portion of Ukraine) up to 80% of its territory was affected by 2007 drought. The damage to agriculture in such years is enormous, especially if drought is preceded by winter wheat damage by extremely cold weather in winter. Very useful information for planners and decision makers is drought intensity in order to estimate the consequences, especially from the extremely severe droughts. VH data provide such information (Fig. 4). As seen, less intensive droughts occurred in Ukraine every year and even several times during the growing season. When the intensity of drought increases the probability of such drought becomes smaller as well as the affected area. Droughts in the range between moderate (D1) and exceptional (D4) occurred every year covering principally 20–40% of the entire Ukraine; although in very unfavorable years the area might be 50–80%. In the last 20 years, Ukraine experienced such droughts six times (probability 30%). Similar probability with 40% coverage of the entire Ukraine is estimated for the droughts in the
16
F. Kogan and W. Guo
Fig. 3 Percent of Ukraine affected by severe, extreme and exceptional droughts
20 0
Exceptional (D4) 1995
20 0
2000
2005
Extreme-to-D4 1995
2000
2005
Severe-to-D4
40 20 0
1995
80
2000
2005
Moderate-to-D4
60 40 20 0 1991
1995
2000
2005
Fig. 4 Percent of Ukraine affected by droughts of different intensity
2009
Early Detection and Monitoring Droughts From NOAA Environmental Satellites
17
category severe and stronger (D2–D4). Finally, exceptional droughts (D4) might cover 5–10% of Ukraine; from the last 20 years such droughts occurred twice in 2003 and 2007. The VH system was also applied for analysis of drought origin and for early detection of drought start, duration and end. These results are shown in Fig. 5 for an area of 64 km2 in the southwestern part of Ukraine. As seen from the combined NDVI and BT estimates, that area experienced drought-related vegetation stress between May and August, including 2-month period (June–July) with the highest drought severity (VHI was at the level of 5–6 corresponding to D4 category). This drought was triggered by the extreme temperatures, since TCI = 0–10 indicating severe thermal stress (TCI = 0 represent the highest temperature). Following thermal stress, VCI reduced drastically from favorable moisture conditions (80–100) during January–April to unfavorable (below 40) conditions at the end of May. The beginning of drought in Fig. 5 is identified in the late April when VHI crossed down the 40 threshold. However, since VH system reflects cumulative weather impact on vegetation, the drought approach can be observed 4–6 weeks prior to the threshold crossing, specifically, when stable deterioration of the conditions have started (dashed arrows in Fig. 5). Therefore, the 2007 drought-related vegetation stress could be predicted in early March, when VHI (health) indicated normal (41–50) and VCI (moisture) even favorable (80–100) conditions. Considering the early stress detection and the size of affected area, VH data provide considerable lead time warning important for pinpointing the problem, making decisions and implementing measures to mitigate drought consequences.
100 Ion = [30.64, 31.08] lat = [47.49, 48.01]
VCI 80
60
NO
VHI
s
t
r
e
s
s
normal
40
S
T
R
E
S
S
20 TCI 0 Jan
Jun
Fig. 5 Dynamics of vegetation health indices, 2007, Ukraine
Dec
18
F. Kogan and W. Guo
Conclusions The new vegetation health indices, which characterize moisture and thermal conditions and the entire vegetation health, have been used successfully for early drought detection and estimation of crop and pasture production losses in 20 countries around the globe. Unlike other NDVI-applied remote sensing techniques, the new method is based on both NDVI and BT, the three ecosystem’s laws and 22-year climatology. Using the combination of NDVI and 10.3–11.3 mm BT provide additional and more accurate information about drought. This method was tested and validated against ground data in all major agricultural countries of the world and in different climatic zones. With the introduction of the new method, drought can be detected 4–8 weeks earlier than before in any corner of the globe, outlined more accurately and the impact on grain reduction can be diagnosed long in advance of harvest. The drought estimates provide new information such as drought area, intensity and dynamics useful for planners, policy and decision makers. The VHI data set and products are provided operationally to NOAA/NESDIS web site every week (http://www.orbit.nesdis.noaa.gov/smcd/emb/vci/VH).
References Kogan FN (2001) Operational space technology for global vegetation assessments. Bull Am Meteor Soc 82(9):1949–1964 Kogan FN (1997) Global drought watch from space. Bull Am Meteor Soc 78:621–636 Kogan FN (1995) Droughts of the late 1980s in the United States as derived from NOAA polar orbiting satellite data. Bull Am Meteor Soc 76:655–668 Kogan FN (1990) Remote sensing of weather impacts on vegetation in non-homogeneous areas. Int J Remote Sens 11:1405–1419 Le Comte D (2009) Global weather highlights 2008: snow, flood and drought, Weatherwise, vol 62, No 2, March–April, pp 22–27 Le Comte D (2007) Global weather highlights 2008: a mixed bag, Weatherwise, vol 61, No 2, March–April, pp 16–21 Salazar L, Kogan F, Roytman L (2007) Use of remote sensing data for estimation of winter wheat yield in the United States. Int J Remote Sens 28(17):3795–3811 Tarpley JP, Schnieder SR, Money RL (1984) Global vegetation indices from NOAA-7 Meteorological satellite. J Climate Appl Meteor 23:491
Flood Monitoring from SAR Data Nataliia Kussul, Andrii Shelestov, and Sergii Skakun
Abstract This paper presents the intelligent techniques approach for flood onitoring using Synthetic Aperture Radar (SAR) satellite images. We applied m artificial neural networks and Self-Organizing Kohonen Maps (SOMs), to SAR image segmentation and classification. Our approach was used to process data from different SAR satellite instruments (ERS-2/SAR, ENVISAT/ASAR, RADARSAT-1/2) for different flood events: Tisza River, Ukraine and Hungary in 2001; Huaihe River, China in 2007; Mekong River, Thailand and Laos in 2008; Koshi River, India and Nepal in 2008; Norman River, Australia in 2009; Lake Liambezi, Namibia in 2009; Mekong River, Laos in 2009. This approach was implemented using Sensor Web paradigm for integrated system for flood monitoring and management. Keywords Flood • Synthetic Aperture Radar (SAR) • Artificial neural networks • Sensor Web paradigm
Introduction In recent decades the number of hydrological natural disasters has increased considerably. According to Scheuren et al. (2008), we have witnessed during 2000–2007 a strengthening of the upward trend, with an average annual growth rate of 8.4%. Hydrological disasters, such as floods and wet mass movements, represent 55% of the overall disasters reported in 2007, had a tremendously high human impact (177 million victims) and caused high economic damages, accounting for 24.5 billion USD (Scheuren et al. 2008). Earth observation (EO) data from space can provide valuable and timely information when one has to respond to and mitigate emergencies such as floods. Satellite observations enable the acquisition of data for large and hard-to-reach territories, as N. Kussul (), A. Shelestov, and S. Skakun Space Research Institute NASU-NSAU, Kyiv, Ukraine e-mail:
[email protected];
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_3, © Springer Science+Business Media B.V. 2011
19
20
N. Kussul et al.
well as provide continuous measurements. Using satellite data, we can determine flood areas, since it is impractical to provide such information through field observations. The flood extent is very important for the calibration and validation of hydraulic models (Horritt 2006). The flood extent can be also used for damage assessment and risk management, and can be of benefit to rescuers during flooding (Corbley 1999). The use of optical imagery for flood mapping is limited by severe weather conditions, in particular by the presence of clouds. SAR measurements are independent of daytime and weather conditions, providing valuable information for monitoring flood events. This is mainly due to the fact that smooth water surface provides no return to antenna in microwave spectrum and appears black in SAR imagery (Rees 2001). At the same time, a wind-ruffled surface can give larger backscatter signal larger than the surrounding land. This complicates the detection of water surfaces on SAR images for flood applications. Flood mapping procedures from SAR imagery consists of three steps: first, re-construction of satellite imagery taking into account calibration and terrain distortion using the Digital Elevation Model (DEM) and providing exact geographical coordinates; second, image segmentation; and the third, flood extent classification. This paper presents a neural network approach to flood mapping from SAR satellite imagery that is based on the application of Kohonen’s SOMs (Kohonen 1995; Haykin 1999). The advantage of SOMs is in providing an effective software tool for the visualization of high-dimensional data, automatically discovering statistically salient features of pattern vectors in data set, and in finding clusters in training data pattern space which can be used to classify new patterns (Kohonen 1995). We applied our approach to the processing of data acquired from several SAR instruments (ERS-2/SAR, ENVISAT/ASAR, RADARSAT-1/2) for different flood events. This approach is implemented using Sensor Web paradigm for “Namibian Pilot Project on Integrated Flood Management and Water Related Vector Borne Disease Modelling”, a joint effort of UN-SPIDER, NASA/NOAA, DLR and Space Research Institute NASU-NSAU. The project goals are to combine satellite imagery with hydrologic ground data and modelling in order to derive useful flood forecasting tools for the next flood season’s transboundary flood management system for local decision makers.
Related Works Different methods are used for flood mapping using satellite imagery. The European Space Agency (ESA) uses a multi-temporal technique of flood extent extracted from SAR images (ESA Earth Watch, http://earth.esa.int/ew/floods). This technique uses SAR images of the same area taken on different dates (one image is acquired during flooding and the second one in “normal” conditions). The resulting multi-temporal image clearly reveals change in the Earth’s surface by the presence of colour in the image. (Cunjian et al. 2001) applied a threshold segmentation algorithm to flood extent extraction from RADARSAT-1 imagery with the support of
Flood Monitoring from SAR Data
21
digital topographic data using threshold segmentation. (Csornai et al. 2004) used ESA’s ERS-2 SAR images, optical data (Landsat TM, IRS WIFS/LISS, NOAA AVHRR) and change detection technique for flood monitoring in Hungary in 2001. (Horritt 1999) has developed a statistical active contour model for segmenting SAR images into regions of homogeneous speckle statistics. The technique measures both the local tone and texture along the contour so that no smoothing across segment boundaries occurs. (Dellepiane et al. 2004) innovative algorithm discriminate water and land on SAR images using fuzzy connectivity concepts taking into account the coherence measure extracted from an InSAR Interferometric SAR. (Niedermeier et al. 2000) applied an edge-detection method to SAR images to detect all edges above a certain threshold. A blocktracing algorithm then determined the boundary area between land and water. (Martinez and Le Toan 2007) used a time series of 21 SAR images from L-band PALSAR instrument onboard JERS-1 satellite to map the flood temporal dynamics and the spatial distribution of vegetation over a large Amazon floodplain. The mapping method is based on decision rules over two decision variables: the mean backscatter coefficient computed over the whole time series, and the total change computed using an “Absolute Change” estimator.
Data The following SAR Satellite instruments were used: ERS-2/SAR, ENVISAT/ASAR, and RADARSAT-1/2. The flood events covered Tisza River, Ukraine and Hungary, 2001; Huaihe River, China, 2007; Mekong River, Thailand and Laos, 2008 (see Fig. 1); Koshi River, India and Nepal, 2008; Norman River, Australia, 2009 (see Fig. 2); Lake Liambezi, Namibia, 2009; Mekong River, Laos, 2009. European satellite
Fig. 1 ENVISAT SAR image (August 16, 2008) during the flooding on the river Mekong, Thailand and Laos (© ESA 2008)
22
N. Kussul et al.
Fig. 2 RADARSAT-2 SAR image (February 14, 2009) during the flooding on the Norman River, Queensland, Australia (RADARSAT-2 Data and Products ©MacDONALD, DETTWILER AND ASSOCIATES LTD. 2009 – All Rights Reserved. RADARSAT is an official mark of the Canadian Space Agency)
data (ERS-2 and ENVISAT) were provided from ESA Category-1 project “Wide Area Grid Testbed for Flood Monitoring using Spaceborne SAR and Optical Data” (№4181). Data from RADARSAT-1 were provided from the Center of Earth Observation and Digital Earth (China). Data from RADARSAT-2 were provided by the Canadian Space Agency within international initiatives, namely GEOSS and International Charter “Space and Major Disasters”. The pixel size and ground resolution for ERS-2 (in ENVISAT format, SLC ─ Single Look Complex) were 4 and 8 m, respectively; for ENVISAT – 75 and 150 m; and for RADARSAT-1 – 12.5 and 25 m. Ground resolution for RADARSAT-2 was 3 m for UltraFine mode, and 25 m for ScanSAR mode. The following water bodies auxiliary data were derived from Landsat-7/ETM+, European Corine Land Cover (CLC 2000) and SRTM DEM (version 3). Neural networks were built for each SAR instrument. In order to train and test neural networks, we manually selected the ground-truth pixels from auxiliary data sets that corresponded to both territories with the presence of water (denoted as class “Water”) and without water (class “No water”). The number of the groundtruth pixels for each instrument and class are presented in Table 1. For ENVISAT/ ASAR instrument, data from the Chinese flood event was used to construct and train the neural network. This neural network was then used to produce flood maps for India, Nepal, Thailand, and Laos events. The same approach was used for RADARSAT-1/2 data: RADARSAT-1 data from the Chinese flood event was used
Flood Monitoring from SAR Data
23
Table 1 Distribution of ground-truth pixels for ERS-2, ENVISAT and RADARSAT-1 images Number of ground-truth pixels for images Satellite image/region “No water” “Water” Total ERS-2/Ukraine 148,182 153,096 301,278 ENVISAT/China 60,575 34,493 95,068 RADARSAT-1/China 135,263 130,244 265,507
to train the neural network, and RADARSAT-2 data for Australia, Namibia and Laos areas was used for an independent data set. For each image from Table 1, data was randomly divided into the training set (75%) and the testing set (25%). Data from the training set were used to train the neural networks, and data from the testing set were used to verify the generalization ability of the neural networks, i.e. the ability to operate on independent, previously unseen data sets.
Method SAR data processing included pre-processing (calibration, geocoding, (providing exact geographical coordinates), orthorectification), and application of discrimination algorithm, which consists of segmentation and classification on two classes using SOMs. SOM is a type of artificial neural network that is trained using unsupervised learning to produce a low-dimensional (typically two dimensional), discretised representation of the input space of the training samples, called a map (Kohonen 1995; Haykin 1999). The map seeks to preserve the topological properties of the input space. SOM is formed of the neurons located on a regular, usually 1- or 2-dimensional grid. Neurons compete with each other in order to pass to the excited state. The output of the map is a so called, neuron-winner or best-matching unit (BMU) whose weight vector has the greatest similarity with the input sample x. The network is trained in the following way: weight vectors wj from the topological neighbourhood of BMU vector i are updated according to (Kohonen 1995; Haykin 1999) i(x ) = arg min x − w j , j =1, L
w j (n + 1) = w j (n) + h(n)h j ,i ( x ) (n)(x − w j (n)), j = 1, L ,
(1)
where η is learning rate (see Eq. 3), h j ,i (x) (n) is a neighbourhood kernal around the winner unit i, x is an input vector, • means Euclidean metric, L is a number of neurons in the output grid, n denotes a number of iteration in the learning phase. The neighbourhood kernel function h j ,i ( x ) ( n) is taken to be the Gaussian
r −r 2 j i( x) h j ,i ( x ) ( n) = exp 2 2s ( n)
(2)
24
N. Kussul et al.
where rj , ri (x) are the vectorial locations in the display grid of the SOM, σ (n) corresponds to the width of the neighborhood function, which is decreasing monotonically with the regression steps. For learning rate we used the following expression:
−
n
h(n) = h0 ⋅ e t , h0 = 0.1,
(3)
where t is a constant. The initial value of 0.1 for learning rate was found experimentally. Kohonen’s maps are widely applied to the image processing, particularly, image segmentation and classification (Kohonen 1995; Haykin 1999). Prior to neural network training, we selected image features which provided an input to neural network. For this purpose, one can choose original pixel values, various filters, or Fourier transformation etc. In our approach, we used a moving window with backscatter coefficient values for ERS-2 and ENVISAT images and digital numbers (DNs) for RADARSAT-1/2 image as inputs to neural network. The output of neural network, i.e. neuron-winner, corresponds to the central pixel of moving window. In order to choose the appropriate size of the moving window for each satellite sensor, we ran experiments for the following windows size: 3-by-3, 5-by-5, 7-by-7, 9-by-9 and 11-by-11. First, we used SOM to segment each SAR image where each pixel of the output image was assigned a number of the neuron in the map. Then, we used pixels from the training set to assign each neuron one of two classes (“Water” or “No water”) using the following rule: for each neuron, a number of pixels from the training set that activated this neuron were calculated. If maximum number of these pixels belonged to class “Water”, then this neuron was assigned “Water” class as opposed to “No water” class. If, neuron was activated by neither of the training pixels, then it was assigned “No data” class.
Results and Discussion In order to choose the best neural network architecture, we ran experiments for each image varying the following parameters: (i) size of the moving window for images that define the number of neurons in the input layer of the neural network, and (ii) number of neurons in the output layer, i.e. the sizes of two-dimensional output grid. Other parameters used during the image processing are: neighborhood topology (hexagonal); neighborhood kernel around the winner (the Gaussian function, see Eq. 2); initial learning rate (set to 0.1) and the number of the training epochs (20). The initial values for the weight vectors are selected as a regular array of vectorial values that lie on the subspace spanned by the eigenvectors corresponding to the two largest principal components of the input data
Flood Monitoring from SAR Data
25
(Kohonen 1995). The results of experiments for the images are presented in Table 2. For the images with higher spatial resolution (i.e. ERS-2 and RADARSAT-1), the best results were achieved for larger moving window 7-by-7. For the ENVISAT/ASAR WSM image, we used the moving window of smaller size 3-by-3. The use of higher dimension of input window for the ENVISAT image led to the coarser resolution of the resulting flood extent image and reduced classification rate. Table 2 Results of SAR images classification using SOMs Satellite image ERS-2 ENVISAT Input dimension Output grid of neurons Classification rate for training set Classification rate for testing set
“No water” “Water” Total “No water” “Water” Total
7-by-7 10-by-10 79.40% 90.99% 85.29% 79.57% 91.06% 85.40%
3-by-3 7-by-5 100.0% 95.64% 98.41% 100.0% 95.90% 98.52%
RADARSAT-1 7-by-7 5-by-5 99.99% 91.93% 96.04% 99.99% 91.89% 95.99%
Fig. 3 Flood extent (black colour) for the Mekong River, Thailand and Laos (© ESA 2008)
26
N. Kussul et al.
Fig. 4 Flood extent (black colour) for Norman River, Australia (RADARSAT-2 Data and Products ©MacDONALD, DETTWILER AND ASSOCIATES LTD. 2009 – All Rights Reserved. RADARSAT is an official mark of the Canadian Space Agency)
The examples of resulting flood extent maps derived from ENVISAT data acquired for the Mekong River, Thailand and Laos (Fig. 1) and RADARSAT-2 data acquired for Norman River, Australia (Fig. 2) are shown in Figs. 3 and 4, respectively.
Sensor Web System for Flood Monitoring Sensor Web is an emerging paradigm and technology stack for the integration of heterogeneous sensors into a common informational infrastructure (Moe et al. 2008; Mandl et al. 2006). The basic functionality required from such infrastructure is remote data access with filtering capabilities, sensors discovery, and triggering of events by sensors conditions. Sensor Web is governed by the set of standards developed by the Open Geospatial Consortium (Botts et al. 2007). At present, the following standards are available and approved by the consortium: OGC Observations and Measurements (http://www. opengeospatial.org/standards/om) – Common terms and definition for Sensor Web domain; Sensor Model Language (http://www.opengeospatial.org/standards/sensorml) – XML-based language for describing different kinds of sensors; Transducer Model Language (http://www.opengeospatial.org/standards/tml) – XML-based language for
Flood Monitoring from SAR Data
27
Fig. 5 Sensor Web perspective of flooding test case
describing the response characteristics of a transducer; Sensor Observations Service (http://www.opengeospatial.org/standards/sos) – an interface for providing remote access to sensors data; Sensor Planning Service (http://www.opengeospatial.org/standards/sps) – an interface for submitting tasks to sensors. The Sensor Web technology was used in the “Namibian Pilot Project on Integrated Flood Management and Water Related Vector Borne Disease Modelling” to integrate heterogeneous data sets: remote sensing satellites data from ASAR, MODIS, MERIS and in-situ observations (water levels, temperature, humidity). Floods forecasting is adding the complexity of physical simulation to the task. The Sensor Web perspective of this test case is depicted in Fig. 5. It shows the collaboration of different OpenGIS specifications of the Sensor Web. The data from different sources (numerical models, remote sensing, in-situ observations) is accessed through the Sensor Observation Service (SOS). Aggregator site is running the Sensor Alert Service to notify interested organization of potential flood events using different communication tools. Aggregator site is also sending orders to satellite receiving facilities using the Sensor Planning Service (SPS) to plan and acquire satellite imagery. Since large amounts of data are generated/acquired, processed, and stored within the Sensor Web, there should be an infrastructure that will allow efficient management and processing of such amounts of information. Grid technology can provide a solution to this problem (Shelestov et al. 2006; Kussul et al. 2009).
28
N. Kussul et al.
Therefore, there are several benefits of integrating Sensor Webs and Grids (Chu et al. 2006): (i) Sensor Web can off-load heavy processing activities to the Grid; Grid can enable real-time sensor data collection and the sharing of computational and storage resources for sensor data processing and management, and (ii) Grid-based sensor applications can provide advance services for smart-sensing by deploying scenario-specific operators at runtime.
Conclusions A neural network approach to SAR-based flood mapping was proposed in this paper. To segment and classify SAR imagery, we applied self-organizing Kohonen’s maps (SOMs) that possess such useful properties as ability to automatically discover statistically salient features of pattern vectors in data set, and to find clusters in training data pattern space which can be used to classify new patterns. As inputs to neural network, we used a moving window of image pixel intensities. We ran experiments to choose the best neural network architecture for different satellite sensors: for ERS-2 and RADARSAT-1/2 the size of input was 7-by-7 and for ENVISAT/ASAR the moving window was 3-by-3. Our approach has the following advantages over existing ones: (i) we apply a moving window to process the image and thus taking into account spatial connection between the pixels of the image; (ii) neural network’s weights are adjusted automatically using ground-truth training data; (iii) to determine flood areas, we need to process a single SAR image. This enables implementation of our approach in automatic services for flood monitoring. Considering the selection of ground-truth pixels to calibrate the neural network, i.e. to assign each neuron one of the classes (“Water” and “No water”), this process can be also automated using geo-referenced information on water bodies for the given region. Three SAR sensors ERS-2/SAR, ENVISAT/ASAR and RADARSAT-1/2 showed the following independently identified classification rates 85.40%, 98.52% and 95.99%, respectively. In this paper SAR images were classified into two classes only (“Water” and “No water”). But our classifier does not provide the estimate of error of belonging to specific class. The future activities will be directed towards the development of classifier that will estimate an uncertainty of a pixel belonging to the specific class. From such a classifier we might also expect that for areas with scarce or no training data we would get larger uncertainty. Therefore, it would indicate that the classifier should be improved with data from that region. Acknowledgments This work is supported by ESA CAT-1 project “Wide Area Grid Testbed for Flood Monitoring using Spaceborne SAR and Optical Data” (No. 4181), and by joint project of the Science and Technology Center in Ukraine (STCU) and the National Academy of Sciences of Ukraine (NASU), “Grid Technologies for Multi-Source Data Integration” (No. 4928).
Flood Monitoring from SAR Data
29
References Botts M, Percivall G, Reed C, Davidson J (2007) OGC Sensor Web Enablement: Overview and High Level Architecture (OGC 07-165). http://portal.opengeospatial.org/files/?artifact_ id=25562. Accessed 10 Sept 2008 Chu X, Kobialka T, Durnota B, Buyya R (2006) Open sensor web architecture: core services. In: Proceedings of the 4th international conference on intelligent sensing and information processing (ICISIP), IEEE Press, Piscataway, NJ, pp 98–103 Corbley KP (1999) Radar imagery proves valuable in managing and analyzing floods red river flood demonstrates operational capabilities. Earth Observation Magazine 8(10) Csornai G, Suba Zs, Nádor G et al (2004) Evaluation of a remote sensing based regional flood/ waterlog and drought monitoring model utilising multi-source satellite data set including ENVISAT data. In: Proceedings of the 2004 ENVISAT & ERS Symposium, Salzburg, Austria Cunjian Y, Yiming W, Siyuan W et al (2001) Extracting the flood extent from satellite SAR image with the support of topographic data. In: Proceedings of International Conference on Info-Tech and Info-Networks, vol 1, pp 87–92 Dellepiane S, De Laurentiis R, Giordano F (2004) Coastline extraction from SAR images and a method for the evaluation of the coastline precision. Pattern Recognit Lett 25:1461–1470 Haykin S (1999) Neural networks: a comprehensive foundation. Prentice Hall, Upper Saddle River, NJ Horritt MS (1999) A statistical active contour model for SAR image segmentation. Image Vis Comput 17:213–224 Horritt MS (2006) A methodology for the validation of uncertain flood inundation models. J Hydrol 326:153–165 Kohonen T (1995) Self-organizing maps. Series in information sciences, vol 30. Springer, Heidelberg Kussul N, Shelestov A, Skakun S (2009) Grid and sensor web technologies for environmental monitoring. Earth Sci Inf 2:37–51 Mandl D, Frye SW, Goldberg MD et al (2006) Sensor webs: where they are today and what are the future needs? In: Proceedings of the second IEEE workshop on dependability and security in sensor networks and systems (DSSNS 2006), Columbia, MD, pp 65–70 Martinez JM, Le Toan T (2007) Mapping of flood dynamics and spatial distribution of vegetation in the Amazon floodplain using multitemporal SAR data. Remote Sens Environ 108:209–223 Moe K, Smith S, Prescott G et al (2008) Sensor web technologies for NASA earth science. In: Proceedings of 2008 IEEE aerospace conference, pp 1–7 Big Sky, MT Digital Object Identifier: 10.1109/AERO.2008.4526458 Niedermeier A, Romaneeßen E, Lenher S (2000) Detection of coastline in SAR images using wavelet methods. IEEE Trans Geosci Remote Sens 38(5):2270–2281 Rees WG (2001) Physical principles of remote sensing. Cambridge University Press, Cambridge Scheuren J-M, de le Polain, Waroux O, Below R et al (2008) Annual disaster statistical review – the number and trends 2007. Center for Research of the Epidemiology of Disasters (CRED). Jacoffsaet Printers, Melin, Belgium Shah-Hosseini H, Safabakhsh RA (2003) TASOM-based algorithm for active contour modelling. Pattern Recognit Lett 24:1361–1373 Shelestov A, Kussul N, Skakun S (2006) Grid technologies in monitoring systems based on satellite data. J Autom Inf Sci 38(3):69–80
Satellite Rainfall Information for Flood Preparedness and Response Robert J. Kuligowski
Abstract Much of the economic and humanitarian toll from flood events is due to a lack of adequate warning and preparation. Information on current and anticipated rainfall from satellite data represents a source of affordable yet useful information for weather forecasters, emergency planners, and other personnel responsible for responding to flood events. This chapter will describe the current state of estimating and nowcasting rainfall at the National Oceanic and Atmospheric Administration (NOAA) National Environmental Satellite, Data, and Information Service (NESDIS), along with plans for the upcoming Advanced Baseline Imager (ABI) onboard Geostationary Operational Environmental Satellites (GOES)-R, which shares many capabilities with the EUMETSAT Spinning Enhanced Visible Infrared Imager (SEVIRI). Examples of these products in actual flood events in Ukraine will be included. Keywords Satellite rainfall estimation • GOES satellite • Flood
Introduction Flooding is one of the most costly natural disasters in the world. According to statistics compiled by the United Nations International Flood Initiative (http:// www.ifi-home.info/), floods affect an average of 520 million people each year with an annual average toll of 25,000 fatalities and US$50–60 billion in damages.
R.J. Kuligowski (*) NOAA/NESDIS Center for Satellite Applications and Research (STAR), Camp Springs, MD, USA e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_4, © Springer Science+Business Media B.V. 2011
31
32
R.J. Kuligowski
Ukraine is no exception: statistics compiled by http://www.preventionweb.net estimated that during the period from 1992 to 2008, a dozen significant flood events affected approximately 2.6 million people, resulting in 76 deaths and US$1.3 billion (UAH 10 billion) in damages. Timely, accurate precipitation information is critical for effective response to natural disasters. Unfortunately, in many parts of the world the current observation network is not well-suited for the demands of flood and flash flood response. Rain gauges are the most widely available source of rainfall information, but the density of such networks generally fails to capture many intense rainfall cells whose spatial dimensions are smaller than the average spacing between gauges. In addition, many gauges do not transmit their data in real-time and hence the information they provide is not available for operational forecasting and decision-making. Radar, when available, provides much more uniform spatial coverage but has its own limitations, particularly in regions of complex terrain, and is not an economically feasible option in many parts of the world. While satellite-based estimates of rainfall are not as accurate as those from gauges and radar, they do provide an excellent supplement to any other available rainfall on information at high spatial (3–4 km) and temporal (15–30) resolution with a lag time of only minutes between the time of the observation and the time when the information can be made available to users. The purpose of this chapter is to briefly describe capabilities for real-time estimation of rainfall from satellite data at the National Oceanic and Atmospheric Administration (NOAA) National Environmental Satellite, Data, and Information Service (NESDIS) in support of flood monitoring and prediction. After briefly reviewing the theory of satellite rainfall estimation, the current and planned NOAA/ NESDIS capabilities will be described.
Theory of Satellite Rainfall Estimation The two primary methods of retrieving rainfall from satellite data involve the use of data from the infrared (IR) portion of the spectrum and from the microwave (MW) portion of the spectrum. In both cases, instruments in space passively measure the amount of radiation leaving Earth’s atmosphere at the wavelengths of interest. (Rainfall retrieval from space can also be performed using active radar – as is done in the Tropical Rainfall Measuring Mission (TRMM) – but this will not be covered in detail here). IR methods relate the cloud-top temperature (as approximated by the brightness temperature in the 11-mm window bands) to the rainfall rate. Specifically, these algorithms assume that the cloud-top temperature is inversely related to the cloudtop height (i.e., cold-topped clouds are higher in the atmosphere and vice versa) and that higher clouds are the result of stronger convective updrafts than lower clouds, which in turn implies a stronger upward transport of moisture into (and
Satellite Rainfall Information for Flood Preparedness and Response
33
thus heavier rainfall from) higher-topped clouds than lower-topped clouds. For convective clouds, this assumption works quite well. However, this assumption can be violated by cirrus clouds (which are cold but do not produce rainfall at the surface) or by nimbostratus clouds associated with cool-season storms in the middle and high latitudes, which have relatively low (warm) tops but can produce significant rainfall. Consequently, IR methods tend to overestimate the spatial extent of heavy rain from convective systems (because of incorrectly assigning rainfall to cirrus shields) and to underestimate rainfall from cool-season synoptic-scale storms. Despite these physical limitations, IR-based algorithms are strongly favored by operational forecasters because they permit nearly continuous monitoring of precipitation (at 15- to 30-min intervals) at high spatial resolution (3–4 km) with a very brief delay from the observation time to when the data become available (often on the order of minutes). A more physically direct estimate of rainfall rate can be obtained using passive MW data. At higher frequencies (89 GHz and up), ice in clouds scatters upwelling terrestrial radiation back downward, creating a depression in the brightness temperature field above a cloud relative to surrounding clear air. Deriving an estimate of the clear-air brightness temperature from other MW bands allows the temperature depression from the cloud to be computed, which in turn is related to the amount of ice in the cloud and thus to the rainfall rate. Over oceans, a different method can be applied which yields more accurate results than the scattering method. At low frequencies (37 GHz and below), water droplets in clouds emit radiation more efficiently than the radiatively cold ocean surface beneath, and thus clouds appear as warm areas in the MW imagery. An estimate of clear-air brightness temperature at these frequencies (based on other MW bands and simple models of ocean-surface emissivity as a function of wind speed) allows the temperature enhancement from the cloud water to be computed, and this can be related to the rainfall rate. However, this technique cannot be applied effectively over land at this time because of the difficulty of estimating the radiative contribution from the land surface, which is dependent on temperature, surface type, soil moisture, vegetation, and other factors. Although both of these techniques generally yield more accurate estimates of rainfall than IR methods, MW-frequency instruments are at this time restricted to low-Earth orbit and thus cannot provide the continuous monitoring available from IR instruments in geostationary orbit. Furthermore, the data latency can be significant because low-Earth orbit satellites can transmit data to Earth only when they are within line-of-sight of a ground receiving facility. Consequently, although MW-based rainfall estimates are preferred for climatological applications, their direct utility for operational forecasting has been limited. However, there are applications that combine the two methods in an effort to achieve the accuracy of the MW estimates with the timeliness and frequency of the IR estimates, and one of these will be discussed in more detail in the “Current NOAA Capabilities” section.
34
R.J. Kuligowski
Current NOAA Capabilities The Hydro-Estimator (H-E) has been the operational NOAA satellite rainfall algorithm supporting real-time flood and flash flood forecasting since 2002. The algorithm is described in more detail in Scofield and Kuligowski (2003) but will be briefly outlined here. As with most IR-based algorithms, the H-E uses the brightness temperature in the IR window (~11 mm) as the foundation for estimates of rainfall rate. However, in response to the issues with cirrus clouds mentioned previously, the H-E considers not only the absolute value of brightness temperature but also its value relative to its surroundings in order to distinguish between active cumulonimbus towers and non-active cirrus areas. If a particular pixel is as warm as or warmer than its surroundings, then it is assumed to not have any convective updrafts underneath and thus is assigned a zero rainfall rate. Conversely, pixels that are colder than the average of the surrounding regions are assumed to be associated with active convective updrafts and rainfall rates are assigned that are related both to the IR brightness temperature and the value relative to its surroundings (i.e., rainfall rates from pixels at the center of convective updrafts are enhanced relative to pixels on the fringes of these updrafts). Since cloud-top temperatures from IR do not contain all of the information related to surface rainfall rates, numerical weather prediction (NWP) model fields are used as a source of supplementary data for adjusting the rainfall rates from the IR data. Regions with higher amounts of low-level moisture (as indicated by the model total column precipitable water – PW) are assumed to have greater upward moisture transport for an updraft of a given strength than regions with less low-level moisture, and so the retrieved rainfall rate is a function of PW as well as IR brightness temperature. Areas where the boundary layer is very dry (as indicated by the mean-layer relative humidity (RH) over the lowest third of the model domain) will likely experience some evaporation of rainfall, and so a rate related to the RH is subtracted from the rainfall rate to account for this. An orographic adjustment to the rainfall rate (described in more detail in Vicente et al. 2002) is used to enhance rainfall in regions where the topography induces upward motion (i.e., where the 850-hPa wind vector is oriented parallel to the elevation gradient). Finally, in some instances the thermodynamic profile precludes the formation of very cold cloud tops but does support strong updrafts beneath relatively warm clouds; the H-E uses temperature and water vapor profile information to adjust the cloud-top temperatures downward in such cases in order to allow heavy rainfall to be depicted in such situations. Additional details are available in Scofield and Kuligowski (2003). The Hydro-Estimator is used to produce real-time estimates of rainfall for the entire globe between 60°S and 60°N using data from five geostationary satellites, which as of this writing are GOES-11 and GOES-13 over the Western Hemisphere, METEOSAT-9 over Europe and Africa, METEOSAT-7 over central Asia and the Indian Ocean, and Multi-Functional Transport Satellite (MTSAT)-1 over eastern Asia, Australia, and the Western Pacific Ocean. These estimates are available in
Satellite Rainfall Information for Flood Preparedness and Response
35
digital and graphic format and can be accessed at http://www.star.nesdis.noaa.gov/ smcd/emb/ff/HydroEst.php. The potential utility of these estimates for flood monitoring and forecasting over Ukraine is illustrated in Fig. 1 for the floods of late July 2008 which resulting in 40 deaths and 170,000 displaced, mainly in Lviv, Zakarpattya, Ternopil, and Chernivtsi oblasts. Total damages from the latter event exceeded UAH 7.6 billion (US$1 billion) in damages.
53N
52N
51N
50N
49N
48N
47N
46N
45N
44N 20E
0
22E
21E
5
23E
10
24E
25E
25
26E
27E
50
100
28E
29E
250
Hydro-Estimator 4 -Day Rainfall (mm) Fig. 1 Hydro-Estimator rainfall accumulation (mm) for the 4 days ending 0000 UTC 26 July 2008
36
R.J. Kuligowski
Planned NOAA Capabilities The launch of GOES-R, currently scheduled for late 2015, will herald a new era in NOAA’s meteorological satellite monitoring capability. The Advanced Baseline Imager (ABI) will improve over the current-generation imager in terms of spatial (4–2 km in the IR; 1–0.5 km in the visible), temporal (full-disk scans in 5 min instead of 30 min), and spectral (16 spectral bands instead of five) resolution (Schmit et al. 2005). The ground processing system for GOES-R will produce dozens of products derived from GOES data, including a next-generation rainfall rate product, a nowcast of rainfall accumulation over the next 3 h, and a nowcast of the probability of measurable rainfall during that same 3-h period. These three products will be described briefly in this section. The rainfall rate algorithm for GOES-R will be based on the Self-Calibrating Multivariate Precipitation Retrieval (SCaMPR), which is described in detail in Kuligowski (2002) but briefly summarized here. SCaMPR is an algorithm that calibrates predictors from GOES against MW rainfall rates and then applies the resulting relationships to independent GOES data in order to provide rainfall estimates that are continuously available but more accurate than estimates based on IR data with a fixed calibration. The SCaMPR calibration takes place in two steps. In the first step, the identification of rainfall is calibrated against the rain/no rain discrimination of the MW data using discriminant analysis. In the second step, the calculation of rainfall rate is calibrated against the MW rain rates of pixels with non-zero values using stepwise forward linear regression. Because the relationship between rainfall rates and IR brightness temperatures tends to be highly nonlinear, each predictor is regressed against the target MW rain rates in log-log space to produce nonlinear transforms of the predictors for use in calibration. The GOES-R version of SCaMPR improves upon the current-generation version (Kuligowski 2002) in several significant ways. First, the number of predictors has been expanded significantly to account for the increase in the number of available IR bands (plus brightness temperature differences). The data are subdivided by cloud type, as indicated by selected brightness temperature differences: water cloud (T7.34 < T11.2 and T8.5–T11.2 < −0.3 K); ice cloud (T7.34 < T11.2 and T8.5–T11.2 > −0.3 K); and cold-top convective (T7.34 ³ T11.2). The data are also subdivided into 30-degree latitude bands (60°−30°S, 30°S-EQ, EQ-30°N, and 30°–60°N) to account for latitudinal differences in the relationship between cloud-top properties and rainfall rate. The training data are in a rolling-value data set: older data are cycled out as newer data are brought in. The amount of data in the file is based on the number of pixels with non-zero rain rates instead of using a fixed period of time to ensure a more consistent calibration. The algorithm has been developed and tested using the METEOSAT Spinning Enhanced Visible InfraRed Imager (SEVIRI) as a proxy. An example of the SCaMPR rain rate output for a routine summer day over Ukraine is presented in Fig. 2 to illustrate.
Satellite Rainfall Information for Flood Preparedness and Response
37
54N 53N 52N 51N 50N 49N 48N 47N 46N 45N 44N 43N 22E 23E 24E 25E 26E 27E 28E 29E 30E 31E 32E 33E 34E 35E 36E 37E 38E 39E 40E 41E 42E
0
10
20
30
40
SCaMPR 24-Hour Rainfall (mm)
50
60
Fig. 2 Accumulated rainfall for the 24 h ending 0000 UTC 6 July 2005 from the planned GOES-R Rainfall Rate algorithm applied to SEVIRI data
In addition to estimates of the current rate of rainfall, the GOES-R ground p rocessing system will produce a nowcast of rainfall accumulation for the next 3 h based on satellite data only. The algorithm is described in detail in Lakshmanan et al. (2003, 2009) and is briefly reviewed here. After filtering the current rainfall rate image to reduce noise, cost minimization is used to organize the pixels in the image into clusters, and smaller clusters are combined into larger ones. The direction of motion for each cluster is determined by overlaying the cluster on the previous rainfall rate image and shifting it to determine the spatial offset that produces the best match. The spatial shifts for each cluster are then objectively analyzed to produce a spatially distributed field and are also smoothed in time using a Kalman filter. These motion vectors are used to move the rainfall rate clusters on the current image forward in time in 15-min increments out to 3 h, and the motion vectors are themselves advected forward in time as well. The resulting 15-min rainfall rates are then accumulated to create a 3-h total. This product is illustrated in Fig. 3 over the Ukraine for part of the time period shown in Fig. 2. A third product related to flood monitoring and forecasting is the probability of rainfall for the 0–3 h time interval. It will provide an indirect measure of confidence in the rainfall potential product – a high rainfall potential accompanied by a
38
R.J. Kuligowski
54N 53N 52N 51N 50N 49N 48N 47N 46N 45N 44N 43N 22E 23E 24E 25E 26E 27E 28E 29E 30E 31E 32E 33E 34E 35E 36E 37E 38E 39E 40E 41E 42E
0
5
10
15
20
25
30
3-hour Rainfall Potential (mm) Fig. 3 Predicted accumulation of rainfall from the planned GOES-R Rainfall Potential algorithm applied to SEVIRI data for the 3-h period beginning 2100 UTC 8 July 2005
relatively low probability of rainfall would imply that the forecast confidence is lower than if the probability of measurable rainfall were 100%. The probability of rainfall algorithm is a statistical calibration of the occurrence of measurable rainfall in the 0–3 h time period against rainfall predicted by the nowcasting algorithm (both in the 3-h rainfall potential and in the 15-min instantaneous rates that are used in it) at both the pixel of interest and at nearby pixels. It should be noted that the algorithm has been calibrated against the occurrence of rain or no rain in the corresponding satellite-retrieved rainfall rate fields rather than against ground validation data from gauges or radar. This was done for two reasons: first, because ground validation data from radar or from gauges with accumulation periods of 3 h or less is difficult to obtain for the vast majority of the SEVIRI coverage area; and second, because any spatially-varying biases in the SCaMPR rain rate detection will tend to produce highly conservative (i.e., low) rainfall probabilities that will have little forecast value.
Satellite Rainfall Information for Flood Preparedness and Response
39
Summary Estimates of rainfall from satellite provide a valuable supplement to other sources of rainfall information (e.g., gauges and radar), particularly in regions where these other sources do not provide sufficiently dense and timely information for flood monitoring and forecasting. IR-based methods provide timely estimates at high spatial and temporal resolution, but represent an indirect estimate based on cloudtop properties. Meanwhile, MW-based methods are more physically direct but have a latency and refresh rate on the order of hours. While the current-generation operational NOAA satellite rainfall algorithm is IR-only, the next-generation algorithm will use MW data as a calibration source to enhance accuracy. NOAA’s nextgeneration satellite rainfall monitoring capabilities will also expand to include short-term (0–3 h) nowcasts of rainfall probability and accumulation.
References Kuligowski RJ (2002) A self-calibrating real-time GOES rainfall algorithm for short-term rainfall estimates. J Hydrometeorol 3:112–130 Lakshmanan V, Hondl K et al (2009) An efficient, general-purpose technique for identifying storm cells in geospatial images. J Ocean Atmos Technol 26:523–537 Lakshmanan V, Rabin R et al (2003) Multiscale storm identification and forecast. J Atmos Res 67:367–380 Schmit TJ, Gunshor MM et al (2005) Introducing the next-generation advanced baseline imager on GOES-R. Bull Am Meteorol Soc 86:1079–1096 Scofield RA, Kuligowski RJ (2003) Status and outlook of operational satellite precipitation algorithms for extreme-precipitation events. Weather Forecast 18:137–1051 Vicente GA, Davenport JC et al (2002) The role of orographic and parallax corrections on real time high resolution satellite rainfall rate distribution. Int J Remote Sens 23:221–230
Monitoring Severe Weather in UKRAINE with Satellite Data Oleksiy Kryvobok, Mykola Kulbida, and Ludmila Savchenko
Abstract Severe weather events associated with deep convection such as flash floods, large hail, damaging winds and even occasional tornadoes are reported every year in Ukraine. There is an increased demand for the assessment of operational strategies for the forecasting and/or nowcasting of severe convective clouds in this country. The forecast of the location of severe convective clouds in advance of their formation is possible with the use of remote sensing techniques such as radar and/or satellite. Ukraine currently has a few EUMETCast stations in the different parts of the country which provide real-time geostationary Meteosat Second Generation satellite images. An experience in the use of Meteosat for detection and tracking of severe convective clouds is discussed. Keywords MSG • Satellite images • RGB images • Severe convective clouds
Introduction Up to the end of 2006 the network of meteorological measurements in Ukraine consisted of nearly 200 meteorological stations, two radars of new generation, which cover only restricted area of the country and 9 radio sounding stations with one sounding per day. It was clear that current meteorological measurements in Ukraine were limited for severe weather monitoring, especially in summer time. The appropriate help might be received from operational satellite data. The geostationary Meteosat Second Generation (MSG) is currently used widely for monitoring severe weather (Kerkmann 2007). The MSG sensors have 12 spectral channels with resolution: spatial 1 or 3 km in sub-satellite point (depending on channels) and temporal O. Kryvobok () Ukrainian Hydrometeorological Research Institute, Kyiv, Ukraine e-mail:
[email protected] M. Kulbida and L. Savchenko Ukrainian Hydrometeorological Centre, Kyiv, Ukraine F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_5, © Springer Science+Business Media B.V. 2011
41
42
O. Kryvobok et al.
5 and 15 min. The reception of real-time MSG data is based on EUMETCast dissemination system, launched in 2002 by the EUMETSAT with coverage of Europe and Africa. The system transmits space-based observations and products to users through a global network of communications satellites, using a multicast, access-controlled, broadband capability. At the technical level, reception is based on off-the-shelf, inexpensive commercial, equipment resulting in relatively low cost reception stations. Ukrainian experience of Meteosat use for detection and tracking of severe convective clouds is discussed in this paper.
eception Facilities at the Ukrainian R Hydrometeorological Services In order to receive real-time MSG satellite data, a EUMETCast receiving station was installed at the Ukrainian Hydrometeorological Research Institute (UHRI) in 2006. Currently, the EUMETCast receiving stations have been installed around Ukraine (Fig. 1) with the following goals: research and training at the Ukrainian Hydrometeorological Research Institute (Kyiv); severe weather products development at the Ukrainian Hydrometeorological Centre (Kyiv); marine satellite products development at the Centre of Azov and Black Seas Hydrometeorology (Odesa); southern local satellite severe weather products development at the Crimea Centre of Hydrometeorology (Simferopol); western local satellite severe weather products development at the L’viv Centre of Hydrometeorology (L’viv). As seen in Fig. 2, the Centres are using standard equipment (PCs, DVB PCI card), satellite offset, antenna and also processing and visualisation software (Kryvobok 2008).
Hotbird-9
Lviv EUMETCAST-2008
Kyiv EUMETCAST-2006
Odesa EUMETCAST-2008 Simferopol EUMETCAST-2008
Fig. 1 Network of EUMETCast stations in Ukraine
Monitoring Severe Weather in UKRAINE with Satellite Data
43
Fig. 2 Configuration of the EUMETCast reception station (computers and antenna system) at the Ukrainian Hydrometeorological Research Institute
Methodology for Interpretation of MSG Images With the launch of MSG satellite with SEVIRI imager the multi-spectral imaging offers enhanced capabilities to look at cloud and air mass in order to evaluate their physical characteristics (Table 1). RGB composites are the main tools for analysis and interpretation of weather events (Final Report 2007). The main features for convective cloud detection on RGB composites are: low temperature and small ice crystals on cloud top, high content of water vapour in the mid level of the atmosphere and enhanced cloud optical thickness. The combination of RGB composites channels are shown in Tables 2–5. One of the important features of MSG IR10.8 image is called “cold-ring and cold U/V” (CRCUV) shape storm. The CRCUV correlate with severe weather or supercells and is using nowcasting. The U or V (resemblance) shape of thunderstorm cloud is the feature is associated with wind shear, scanning geometry and storm maturity. Both cold-ring and cold-U/V shapes are common to the storms exhibiting some form of embedded warm spots at their tops, typically located downwind of overshooting tops or longer-lived elevated “domes,” penetrating the Tropopause into the warmer lower stratosphere. Short lasting Embedded Warm Spots (EWS) are quite common, forming in the lee of individual overshooting tops, and typically disappear after decay of their “parent” overshooting top. Therefore, life cycles of EWS are closely related to the life cycles of overshooting tops, typically lasting 10–20 min, or even less. A storm should be classified as a CRCUV, only if the cold ring and CWS is persist for more than 30 min (excluding cases of transient EWS in the lee of single overshooting tops). The shorter-lived features are less significance for nowcasting. The latest modeling results confirms that wind shear is crucial for the actual shape of the cold feature. Storms with short-lived embedded warm spots are rather not
44
O. Kryvobok et al.
Table 1 Cloud physical properties represented by the MSG channels (Final Report 2007) Channels Physical properties VIS0.6 Optical thickness and amount of cloud water and ice VIS0.8 Optical thickness and amount of cloud water and ice NIR1.6, IR3.9r Particle size and phase WV6.2, WV7.3 Mid- and upper level moisture IR8.7, IR10.8, IR12.0 Top temperature IR8.7–IR10.8 Phase and optical thickness IR12.0–IR10.8 Optical thickness IR3.9–IR10.8 Optical thickness, phase, particle size IR13.4–IR10.8 Top height WV6.2–IR10.8 Top height, overshooting tops HRV Overshooting tops Table 2 RGB composite DAY MICROPHYSICS (Final Report 2007) DAY MICROPHYSICS RGB color Channel plane (difference) Min Max Gamma Prominent features R VIS0.8 0% 100% 1.0 Cloud analysis, convection G IR3.9r 0% 60% 2.5 B IR10.8 203 K 323 K 1.0 Table 3 RGB composite NIGHT MICROPHYSICS (Final Report 2007) NIGHT MICROPHYSICS RGB color Channel plane (difference) Min Max Gamma R IR12.0–IR10.8 −4 K +2 K 1.0 G IR10.8–IR3.9 0K +10 K 2.5 B IR10.8 203 K 293 K 1.0
Prominent features Cloud analysis, convection
Table 4 RGB composite DAY CONVECTIVE STORMS (Final Report 2007) DAY CONVECTIVE STORM RGB color Channel plane (difference) min Max Gamma Prominent features R IR6.2–IR7.3 −35 K +5 K 1.0 Severeconvection, WV in/outflux G IR3.9–IR10.8 −5 K +60 K 0.5 B NIR1.6–VIS0.6 75% 25% 1.0 Table 5 RGB composite SEVERE STORMS (Final Report 2007) SEVERE STORM RGB color Channel plane (difference) Min Max Gamma R HRV 0% 100% 1.7 G HRV 0% 100% 1.7 B IR3.9–IR10.8 203 K 323 K 1.0
Prominent features Severe convection
Monitoring Severe Weather in UKRAINE with Satellite Data
45
severe Storm with long-lived warm spots, resulting in CRCUV or cold-U shape should be monitored carefully since in most cases they are severe (Setvak 2008). One of the RGB’s MSG images advantage is possibility to detect the exact location and intensity of the severe convective clouds, which is very difficult with traditional synoptic charts. Based on 3 years experience of MSG image interpretation at the Ukrainian Hydrometeorological Service, the recommendations satellitebased short-range forecasting and nowcasting of severe convective clouds were developed. The HRV, RGB SEVERE STORM and IR10.8 are the most useful for detection of day-time convective clouds. At night, the IR10.8 images can be used successfully for estimation of cloud top temperature or the cooling rate of cloud tops (by analysis of two successive IR10.8 images). After convective cloud detection, it is most important to estimate the severity of cloud, which can be estimated from the appearance of small ice crystal on the cloud top (the result of strong updrafts and embedded warm spots), cold ring, U/V shape in the field of cloud top temperature. The RGB DAY MICROPHYSICS, RGB NIGHT MICROPHYSICS, RGB DAY CONVECTIVE STORM and RGB SEVERE STORM images clearly demonstrate cloud top microphysics both during day and night. A special software was developed at the Ukrainian Hydrometeorological Centre for automatic interpretation of satellite data in the case of severe weather.
EUMETSAT Operational Products One of the operational satellite products used in severe weather forecasting and nowcasting is the EUMETSAT’s Global Instability Indices (GII). The GII is used for early detection of the unstable air and assessment of potential deep convection (Konig et al. 2001). GII represent a set of indices which describe the stability layer of the atmosphere. The GII are empirical and might be applied in certain geographical regions or under certain circumstances. They can be retrieved in cloud-free conditions. The GII product comprise four classic instability indices: Lifted index LI = T obs − T lifted from surface at 500 hPa
(
)
(
K − index KI = T obs850 − T obs500 + TDobs850 − T obs700 − TDobs700
(
KO − index KO = 0.5 ∗ (Θ eobs500 + Θ eobs700 − Θ eobs850 − Θ eobs1000
)
)
Maximum Buoyancy MB = Θ eobs(maximum between surface and 850) − Θ eobs(maximum between 700 and 300) where T obs is the observed temperature, TDobs is the observed dew point temperature, Θ eobs is the observed equivalent potential temperature. The GII are disseminated hourly by the EUMETCast data stream.
46
O. Kryvobok et al.
Another important satellite product is the Atmospheric Motion Vector (AMV), which shows the direction and speed of moving atmospheric layers or clouds. The AMV is produced from seven channels, every 15 min with special resolution 80 km2 (26 by 26 pixels). With advanced MSG the quality of the AMV will improve due to availability of IR13.4 or CO2 channel (Holmlund 2000). The final vector components (speed, direction, height, temperature and quality) are based on a weighted mean of the intermediate vectors. The vectors are disseminated hourly over the GTS.
Some Applications The example below provides an analysis of severe weather convective storm on 23.06.08 close to L’viv in the western Ukraine. The storm was characterized by squall wind up to 30 m/s, hail, and heavy rains. This severe weather was caused by a mesoscale convective system in the atmospheric occluded front. The NWP model predicted movement of atmospheric front to western Ukraine on the afternoon of June 23, 2008 (Fig. 3). However, only satellite data showed strong convection activity in this region. The severe convective clouds were detected at 8:30 UTC, about 230 km to the west of L’viv (black arrow on Fig. 4). Over the western part of Ukraine there was a field of high values of GII (K index) and severe unstable air mass. The AMVs have a mostly eastern direction on low and mid levels. By 10:30 UTC, the mesoscale convective system came to the Poland/Ukraine border, 60 km west of L’viv. There was a “cold ring” on the IR10.8 (Fig. 5, black arrow) and there were strong updraft, small crystals on the top of clouds on RGB
Fig. 3 Forecast chart 12 UTC on 23.06.2008
Monitoring Severe Weather in UKRAINE with Satellite Data
47
Fig. 4 RGB SEVERE STORM and IR10.8 images at 08:30 UTC on 23.06.08
Fig. 5 RGB SEVERE STORM and IR10.8 images at 10:30 UTC on 23.06.08
Fig. 6 RGB SEVERE STORM and IR10.8 images at 11:30 UTC on 23.06.08
SEVERE STORM (Fig. 5). The direction of wind on AMV product showed that a strong convective storm was moving to the city with the speed of about 80 km/h. At this time the local forecasters warned local population and authories that the approaching storm will come to the city in 30–40 min. One hour later, at 11:30, the severe storm (mature stage) was over the city (Fig. 6). It continued about 1 h and damaged buildings, trees and social infrastructure. Two people were killed by falling trees and many injured by lightning sparks. The advanced satellite and weatherbased warning helped to minimize damages and save human lives.
48
O. Kryvobok et al.
Conclusions Satellite images are extremely important for the detection and monitoring of severe convective clouds, especially when other measurements of atmospheric or cloud state are not available. After the launch of MSG satellite, enhanced spectral resolution, spatial coverage, and time sampling helped forecasters get reliable tools for a short-range forecasting and nowcasting of severe weather. One of the critical points in the monitoring of severe weather is timely image generation, which on MSG is 5 and 15 min that can provide for detection and monitoring of even rapidly developing thunderstorms. Thus, based on the simplicity of reception and a reliable method of interpretation, operational use of satellite data, especially MSG, significantly improved short-range forecasting and nowcasting of severe weather in the Ukrainian Hydrometeorological Service.
References Final Report (2007) RGB composite sattelite imagery workshop. Boulder, CO, 5–6 June 2007 Holmlund K (2000) The atmospheric motion vector retrieval scheme for meteosat second generation. In: Proceedings of the 5th international winds workshop, Lorne, Australia, EUMETSAT, EUM-P28, pp 201–208 Kerkmann J (2007) Applications of meteosat second generation (MSG). Day-time convection. Night-time convection. MSG Interpretation Guide, EUMETSAT König M, Tjemkes S, Kerkmann J (2001) Atmospheric instability parameters derived from MSG SEVIRI observations. Proceedings of the 2001 EUMETSAT meteorological satellite data users’ conference, Darmstadt, Germany, pp 133–140 Kryvobok O (2008) The new capabilities in reception of satellite data based on EUMETCast system. Ukr Hydrometeorol J N3:25–32 Setvak M et al (2008) Cold ring storms in Central Europe. The 2008 EUMETSAT Meteorological Satellite Conference, Darmstadt, Germany, pp 353–358
Daily Fire Occurrence in Ukraine from 2002 to 2008 Wei Min Hao, Shawn P. Urbanski, Bryce Nordgren, and Alex Petkov
Abstract The spatial and temporal extent of daily fire activity in Ukraine at a 1 km × 1 km resolution from 2002 to 2008 is investigated based on active fire detections by the Moderate Resolution Imaging Spectroradiometers (MODIS) on NASA’s Terra and Aqua satellites. During this period about 20,000 fires were detected annually in Ukraine. Ukraine has two distinct fire seasons – spring (March, April, and May) and summer/early fall (July, August, and September). Summer and early fall was the main fire season, accounting for 77% of total active fire detections, while spring detections comprised only 17% of the total. The fire activity was mostly associated with agricultural burning; 91% of active fires were on agricultural land. The agricultural burning was dominated by burning stubble residue following harvest of winter wheat. The summer fire activity was highly correlated with annual wheat production (r = 0.81, p < 0.05). The minimum (2003) and maximum (2008) years of Ukraine fire activity deviated from the 7-year mean by −79% and +114% respectively, and coincided with the extremes of low and high wheat production in Ukraine during the study period (3.6 million tons in 2003 and 25.9 million tons in 2008). Keywords Fire • MODIS • Land cover • Cloud cover • Fire trend
Introduction Fire has always been maintains ecosystem insects and diseases, fire has been widely
an integral part of natural and managed ecosystems. Fire health by clearing understory vegetation, exterminating and facilitating generation of new seeds. For centuries used for agricultural purposes, such as deforestation for
W.M. Hao (*), S.P. Urbanski, B. Nordgren, and A. Petkov Rocky Mountain Research Station, Five Sciences Laboratory, US Forest Service, Missoula, MT, USA e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_6, © Springer Science+Business Media B.V. 2011
49
50
W.M. Hao et al.
ranching, crop production, and clearing of agricultural residues. Agricultural fields are often burned post-harvest or prior to planting to fertilize soil, eliminate residues, facilitate crop rotation, and control insects and diseases. Open biomass burning (e.g., wildfires, agricultural burning, and managed burning of ‘natural’ landscapes) is also a significant source of atmospheric trace gases and aerosols (van der Werf et al. 2006), and these emissions significantly influence the chemical composition of the atmosphere (Simpson et al. 2006) and the Earth’s climate (Naik et al. 2007). Emissions from open biomass burning in Eastern Europe can significantly degrade air quality throughout Europe during spring and late summer. Niemi et al. (2009) reported that between 2001 and 2007 emissions from wildfires and agricultural waste burning in this region caused or contributed to 13 spring and latesummer episodes of high particulate matter pollution in southern Finland. Smoke from agricultural burning in Eastern Europe has also been identified as a source of severe pollution episodes in the Arctic (Stohl et al. 2007). In spring of 2006, monitoring stations in the European Arctic detected record levels of O3 and black carbon concentrations from the intrusion of smoke from agricultural burning in Eastern Europe (Stohl et al. 2007). Ukraine is a nation where cropland is the dominant land cover. Open biomass burning in Ukraine has been identified as a contributor to pollution in Central and Western Europe and the Arctic (Stohl et al. 2007; Niemi et al. 2009). In this paper we use MODIS data for active fire detections, land cover maps, and cloud cover to characterize in Ukraine the seasonality and geographic distribution of fire occurrence, the trends of fire activity from 2002 to 2008, and the causes of the trends.
Method Land Cover Map We used the standard MODIS land cover product in this study. The MODIS instruments, onboard NASA’s Terra and Aqua satellites, have been operational since December 1999 and May 2002, respectively. The land cover product MOD12Q1 is produced at a spatial resolution of 500 m. MODIS data are assembled into “data collections” where all of the data are processed by the same algorithms. We chose collection 5, which is the most current version. Land cover is a level 3 data product which is resampled onto a common global grid. The global dataset is divided into “tiles” with horizontal and vertical coordinates. Expressed as (horizontal, vertical) pairs, the four tiles containing relevant data for Ukraine are: (19, 3), (20, 3), (19, 4), and (20, 4). We assume that land cover does not change in Ukraine for the study period. We chose the land cover dataset for the year 2008 to represent for the 7 years in the study.
Daily Fire Occurrence in Ukraine from 2002 to 2008
51
Weather Station Data and Associated Regions Weather data was retrieved from the NOAA Global Historical Climatology Network, which is the official archive for daily weather data from the Global Climate Observing System (GCOS) Surface Network (GSN). The dataset provided daily maximum temperature, minimum temperature, and precipitation. Weather data from a particular station is applied to all of the area within its associated region. Polygons which define the regions served by the weather stations were generated using the “Create Thiessen Polygons” tool in GIS software Arc/Info.
Fire Detection MODIS observations of fire activity used in this study are the collection 5 “fire and thermal anomaly” level 2 data product. This dataset has a one-to-one correspondence with the on-orbit observation of radiance at a 1 km × 1 km spatial resolution and has not been resampled spatially or temporally. We maintain a global database of these fire detections, which includes the location of the center of the fire pixel, derived parameters calculated by the fire algorithm (e.g., fire radiative power), and the results of statistical tests used by the algorithm itself during the process of classification. A comprehensive database of fire detections within the Ukraine national boundary was compiled by extracting Ukrainian data from each of the annual global databases. The daily weather station database was extended to incorporate the total number of fire detections within the indicated weather station’s polygon on the indicated day.
Cloud Cover One of the major uncertainties of MODIS fire detection is that the instrument cannot detect active fires through cloud cover. Hence, we investigate the percentage of cloud cover in the month of August for each year from 2002 to 2008. We selected August because it was the month of the maximum fire activity during the study period (Table 1). The orthodox process of generating a cloud fraction pertaining to an extended area for a period of time consists of a series of integrations. The basis of these integrations is the initial classification of every 1 km2 MODIS observation into one of four categories of cloudiness: confident cloudy, probably cloudy, probably clear, or confident clear. Each 1 km2 pixel in the original satellite observation is considered to be either 100% cloudy or 0% cloudy based on this classification. Confident and probably cloudy pixels are taken to be 100% cloudy, whereas the other two classes are 0% cloudy. The initial estimation of cloud fraction represents an aggregation of 5 × 5 1 km2 pixels, nominally 25 km2 at nadir. For each 25 km2 pixel, the number of cloudy
52
W.M. Hao et al.
1 km2 pixels within the 5 × 5 cell is divided by 25 to yield the single cloud fraction number for that cell. Approximately six hundred and forty 25 km2 cells are then aggregated both spatially and temporally to yield a single daily number for every 1° × 1° grid cell at 50° north. The 30 or 31 daily cloud fractions in each squaredegree cell are then averaged over the month to produce a single monthly cloud fraction for that cell. Finally, 231 monthly values in the 21 × 11 window over Ukraine are averaged to produce a single measure of cloud cover for Ukraine. Overall, every measure of cloudiness over Ukraine for a 31-day month represents 114.5 million samples of cloud/no-cloud. Uncertainty estimates are produced in the final step by calculating the standard deviation of the 231 monthly cloud fraction values over Ukraine. Our analysis was simplified by the fact that the first four steps of this process were performed by NASA and offered as standard products via its Distributed Active Archive Centers (DAACs). We downloaded monthly cloud fraction data (MXD08_M3) for the month of August, extracted the data over Ukraine, and performed the final spatial average.
Results and Discussion Fire Seasons The land cover map and the geographic distribution of daily active fire locations at a 1 km × 1 km resolution over the Ukrainian landscape in 2008 are shown in Fig. 1. Agricultural land is spread throughout the country and dominates land cover (~81%). The forest areas of mixed forests, evergreen needleleaf forests, and deciduous broadleaf forests constitute the other 11% of the land cover. The forest areas are mostly in the western and northern Ukraine. Approximately 95% of the active fires detected in 2008 are located in cropland and less than 1% in the forest areas. Because the major land cover in Ukraine is agricultural land and forests, we compare the patterns of fire occurrence over two distinct regions: one is dominated by cropland in central Ukraine and the other with a significant forest area in western Ukraine. Cropland accounts for 94% of the land over the region of 29,630 km2 associated with the weather station UP000033711 in central Ukraine (48.48°N, 32.75°E). Figure 2a shows the average daily maximum temperatures between 2002 and 2008 and the total number of integrated daily active fires within the polygon during the 7-year period, based on analysis of the 7-year dataset of MODIS active fire locations. There were two distinct fire seasons, spring and summer/early fall. The spring fire season started in March when the temperatures were above about 5°C and lasted until late April. The majority of the fires occurred in summer and early fall, beginning in July and reaching its peak in August before the daily maximum temperatures decreased. Although the number of active fires in spring and summer/early fall varied from year to year (Fig. 2b), most of the agricultural fires were burned in July, August, and September and to a lesser extent in spring.
Daily Fire Occurrence in Ukraine from 2002 to 2008 25°E
53
30°E
35°E
40°E
55°N
50°N
45°N Black Sea
2008 Active Fires
Forest
Cropland
Other
0 55 110
220
Kilometers
330
N
Fig. 1 Daily MODIS-detected active fires at a 1 km × 1 km resolution in 2008 overlaid on the Ukrainian land cover map of MOD12Q1
Agricultural fires are often caused by burning agricultural residues before plowing soil in spring or after harvest in summer and fall. In contrast to agricultural burning, more forest fires occurred in spring than during summer and early fall in western Ukraine. An example is the region of 15,373 km2 associated with the weather station UP000033393 (49.82°N, 23.95E) where 71% of the land cover is agricultural land and 22% is the forest area (Fig. 3). More than 90% of the forest fires were burned in spring. The forest fires are used to clear understory vegetation after snow is melted in spring or before snowfall in November, which is a common practice of prescribed burning in Ukraine and the United States. In this region, about 57% of agricultural fires were burned in spring, and 43% in summer and early fall. Almost all the fires from July to September were caused by agricultural burning. We have investigated the roles of precipitation on fire occurrence in these two regions, in addition to the temperature effects. However, no conclusion can be made because of limited precipitation data available. The pattern of the seasonality of fires for the entire Ukraine is similar to that at the regions associated with the weather stations of UP000033711 and UP000033393. The 7-year dataset of MODIS active fire locations reveals two distinct fire seasons in Ukraine, spring (March, April, May) and summer/early fall (July, August, September) (Fig. 4, Table 1). Between 2002 and 2008, 77% of the fires in Ukraine occurred in summer and early fall. In a typical year, the summer
54
W.M. Hao et al.
Fig. 2 (a) Average daily maximum temperature at the weather station 000033711 from 2002 to 2008 and the total number of integrated daily active fires for the corresponding region during the same time period; (b) the total number of active fires in spring and summer/early fall for the same region in each year from 2002 to 2008
fire activity increased rapidly beginning in July, reached a maximum in August, and then declined swiftly in September. In the spring, the fire activity began in March, peaked in April, and then declined briskly in May, and reached a minimum in June. The spring fire season accounted for 17% of all active fire detections during the study period.
Daily Fire Occurrence in Ukraine from 2002 to 2008
55
Fig. 3 (a) Average daily maximum temperature at the weather station 000033393 from 2002 to 2008 and the total number of integrated daily active fires for the corresponding region during the same time period; (b) the total number of active fires in spring and summer/early fall for the same region in each year from 2002 to 2008
Fire Occurrence Trends Table 1 summarizes the number of MODIS active fires at a 1 km × 1 km resolution each month from 2000 to 2008 in Ukraine. On average, about 20,000 active fires were detected each year; however, there was significant inter-annual variability. The number of detected active fires in the extreme years (2003 and 2008) differed by an order of magnitude.
56
W.M. Hao et al. 3500
Active Fires
3000
2500
2000
1500
1000
500
0
J
F
M
A
M
J
J
A
S
O
N
D
Month Fig. 4 Integrated daily active fires during the period of 2002–2008
Table 1 Total number of active fires in Ukraine for each month from 2002 to 2008 2002 2003 2004 2005 2006 2007 January 9 4 0 59 9 145 February 108 1 24 4 28 598 March 1,521 225 626 347 271 4,905 April 314 1,799 1,934 4,114 2,797 953 May 254 473 206 227 845 430 June 170 175 135 101 394 478 July 6,961 161 1082 3,227 4,141 8,326 August 9,386 488 5,929 4,118 8,856 4,356 September 2,310 592 2,068 2,120 2,114 917 October 66 133 287 1795 511 766 November 85 29 102 169 29 34 December 70 45 22 8 30 1 Total 21,254 4,125 12,415 16,289 20,025 21,908
2008 99 623 671 319 181 355 5,215 28,747 5,085 761 302 32 42,390
Note that because the MODIS sensor on the Aqua satellite did not provide active fire detections prior to June 2002, data for January – May 2002 is not comparable to the following years
One of the major uncertainties of fire detection by MODIS is that the instrument cannot detect active fires through clouds. Figure 5 shows the percentage of cloud cover and the number of detected active fires in August, the month of peak burning activity, in Ukraine during the period of 2002–2008. Overall, about 15–38% of Ukraine was covered by clouds, with an average of about 20%, in August for each
Daily Fire Occurrence in Ukraine from 2002 to 2008
57
Fig. 5 Percentage of cloud cover and the number of active fires in August 2002–2008
Fig. 6 Active fires in 2002–2008 aggregated by months and land cover
year between 2002 and 2008. The cloud cover detected by the Terra satellite was always slightly lower than that detected by the Aqua satellite. There is no apparent correlation between the percentage of cloud cover and the number of detected active fires (Terra: r = −0.14, p = 0.76; Aqua: r = −0.25, p = 0.59). Hence, cloud cover was not the primary factor causing the inter-annual variability of active fires in Ukraine. Ninety-one percent of the detected active fires in Ukraine occurred in cropland. The prevalence of agricultural related burning is not surprising given that cropland is the dominant land cover in Ukraine (see Fig. 1). The dominance of agricultural fires spanned all seasons (Fig. 6). Agricultural fires are often caused by burning crop residues before plowing soil in spring or after harvest in summer and fall. The dominant crop in Ukraine is winter wheat, which is planted in the fall and harvested in July and August
58
W.M. Hao et al.
of the following year. The burning of wheat stubble following harvest is common practice throughout the wheat growing regions in the world. Annual detected active fires in summer/early fall (July–September) and spring (March– May) are plotted with the annual Ukrainian wheat production (Table 2) in Fig. 7. The summer and early fall fire activity in Ukraine is highly correlated Table 2 Wheat production in Ukraine from 2002 to 2008 Area harvested Production Yield Yeara (×106 ha) (Mt) (t/ha) 2002 6.8 20.6 3.0 2003 2.5 3.6 1.4 2004 5.9 17.5 3.0 2005 6.8 18.7 2.8 2006 5.5 14.0 2.5 2007 6.0 13.9 2.3 2008 7.1 25.9 3.6 Mean 2002–2008 5.8 16.3 2.7 Data is from the USDA Foreign Agricultural Service (USDA 2009) a The year refers to the year of the Ukrainian winter wheat harvest which occurs in July and August; ha: hectares; Mt: million tons (t)
Fig. 7 Annual spring and summer/early fall active fires and wheat production for Ukraine in 2002–2008. The inset is a figure of annual summer/early fall active fire detections versus wheat production for Ukraine in 2002–2008
Daily Fire Occurrence in Ukraine from 2002 to 2008
59
with annual wheat production from 2002 to 2008 (r = 0.81, p < 0.05, see inset in Fig. 7). This correlation strongly suggests that post-harvest burning of winter wheat stubble is the source of the majority of active fires that were detected during July–September in central and eastern Ukraine. The minimum number of active fires (4,125) in 2003 and the maximum number of active fires (42,390) in 2008 deviated from the 7-year mean by −79% and +114% respectively, and coincided with the extremes of Ukraine wheat production during the study period (3.6 million tons in 2003 and 25.9 million tons in 2008, see Table 2). In 2003, low temperatures in December and persistent ice crusting during February and March resulted in extensive damage to wheat crops in southern and central Ukraine (USDA 2003a, b). The unfavorable weather conditions resulted in widespread loss of winter wheat and dramatically reduced the area harvested, the yield, and total production (Table 2). In 2008, the combinations of increased sown area of winter wheat (up 15% from 2007) (USDA 2008, Table 2) and favorable growing conditions (USDA 2008) resulted in the maximum of area harvested, yield, and production.
Conclusion The spatial and temporal extent of daily active fires in Ukraine at a 1 km × 1 km resolution from 2002 to 2008 is investigated based on active fire detections by the MODIS instruments on NASA’s Terra and Aqua satellites. Ukraine has two distinct fire seasons – spring (March, April, and May) and summer/early fall (July, August, and September). Summer and early fall was the main fire season, accounting for 77% of total detected active fires, while spring fires comprised only 17% of the total. During this 7-year period about 20,000 fires were detected annually in Ukraine. The fire activity was mostly caused by agricultural burning; 91% of the active fires were detected on agricultural land. Overall, Ukraine fire activity was dominated by burning stubble residue following harvest of winter wheat in summer and early fall. Despite uncertainties in the effectiveness of MODIS fire detection due to the variability in cloud cover, the summer/early fall fire activity was highly correlated with annual wheat production (r = 0.81, p < 0.05). The minimum (2003) and maximum (2008) years of Ukraine fire activity deviated from the 7-year mean by −79% and +114% respectively, and coincided with the extremes of Ukraine wheat production during the study period (3.6 million tons in 2003 and 25.9 million tons in 2008). The meager 2003 wheat harvest resulted from adverse winter weather, low temperatures in December and persistent ice crusting in February and March (USDA 2003a, b). An increase in wheat sown area and favorable weather were responsible for the peak wheat production of 2008 (USDA 2008). The increase in wheat sown area in 2008 may have been driven by the high wheat prices of 2007. This study concludes that the large inter-annual variability in agricultural burning in Ukraine between 2002 and 2008 is attributable to weather conditions in the growing seasons and fluctuations in agricultural intensity (i.e. area sown).
60
W.M. Hao et al.
Our findings suggest that the efforts to develop scenarios of future fire emission inventories should consider the likelihood of significant inter-annual variability in agricultural burning on a regional scale. Long-term changes and large inter-annual fluctuations in both weather conditions in growing seasons and agricultural intensity should be represented in the development of future emission scenarios.
References Naik V, Mauzerall DL, Horowitz LW, Schwarzkopf MD, Ramaswamy V, Oppenheimer M (2007) On the sensitivity of radiative forcing from biomass burning aerosols and ozone to emission location. Geophys Res Lett 34:L03818. doi:10.1029/2006GL028149 Niemi JV, Saarikoski S, Aurela M, Tervahattu H, Hillamo R, Westphal DL, Aarnio P, Koskentalo T, Makkonen U, Vehkamäki H, Kulmala M (2009) Long-range transport episodes of fine particles in southern Finland during 1999–2007. Atmos Environ 43:1255–1264 Simpson IJ, Rowland FS, Meinardi S, Blake DR (2006) Influence of biomass burning during recent fluctuations in the slow growth of global tropospheric methane. Geophys Res Lett 33:L22808. doi:10.1029/2006GL027330 Stohl A, Berg T, Burkhart JF, Fjæraa AM, Forster C, Herber A, Hov Ø, Lunder C, McMillan WW, Oltmans S, Shiobara M, Simpson D, Solberg S, Stebel K, Ström J, Tørseth K, Treffeisen R, Virkkunen K, Yttri KE (2007) Arctic smoke – record high air pollution levels in the European Arctic due to agricultural fires in Eastern Europe in spring 2006. Atmos Chem Phys 7:511–534 United States Department of Agriculture, Foreign Agriculture Service, Production, Supply, and Distribution online database, downloadable datasets (2009) URL: http://www.fas.usda.gov/ psdonline/psdDownload.aspx United States Department of Agriculture, Foreign Agriculture Service (2008), World Agricultural Production, Circular Series, WAP 08–08, June 2008, Page 4. URL: http://www.pecad.fas.usda. gov/search.cfm United States Department of Agriculture, Foreign Agriculture Service (2003a), World Agricultural Production, Circular Series, WAP 06–03, June 2003, Page 3. URL: http://www.pecad.fas.usda. gov/search.cfm United States Department of Agriculture, Foreign Agriculture Service, Production Estimates and Crop Assessment Division (2003b), Ukraine: Extensive damage to winter wheat, May 23, 2003. URL: http://www.pecad.fas.usda.gov/search.cfm van der Werf GR, Randerson JT, Giglio L, Collatz GJ, Kasibhatla PS, Arellano A Jr (2006) Interannual variability in global biomass burning emissions from 1997 to 2004. Atmos Chem Phys 6:3423–3441
Satellite-Based Systems for Agro-meteorological Monitoring Alexander Kleschenko, Oleg Virchenko, and Olga Martinenko
Abstract Discussed in this report are the concepts of remote sensing observations and the principal characteristics of satellite systems used for crop monitoring. The main functions of the systems and procedures for data processing and interpreting are proposed. Some example are presented and validated. Keywords Remote sensing system • Agrometeorological monitoring
Introduction Agriculture was one for the first branches of economy where remote sensing provided considerable contribution. In the beginning qualitative methods were used. At present, a numerical system for satellite-based crop conditions and productivity monitoring has been developed and is being used in the operational mode. This paper presents the fundamental concept of the remote sensing system as well as specific features used in Russia. Also, future tasks were formulated.
Remote Sensing and Systems Remote Sensing of Crops The term “remote sensing” is used to describe satellite observations that are produced, retrieved, processed, and interpreted. “Remote sensing” is an instrumentbased technique employed in the acquisition and measurement of spatially organized
A. Kleschenko (), O. Virchenko, and O. Martinenko National Institute for Agricultural Meteorology, Roshydromet, Obninsk, Russian e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_7, © Springer Science+Business Media B.V. 2011
61
62
A. Kleschenko et al. 0.5 R (l,nm)
(1)
0.4
0.3 (2)
0.2
0.1
0 500
600
700
800
900
1,000
Fig. 1 Spectral reflectance of winter wheat at heading (1) and soil (2)
(most commonly, geographically) data on an array of target points (pixels) within the sensed scene that correspond to features, objects, and materials. The visible (VIS), near infrared (NIR) and infrared (IR) bands of solar spectrum are normally used to sense the Earth surface. Figure 1 shows the reflection of a typical soil and healthy crop in VIS and NIR. As seen, the soil curve shows slightly increase, while the crop curve shows minimum reflectance within VIS (670–680 nm) and maximum in NIR. The difference between the NIR and VIS are widely used in agricultural monitoring. It should be noted that many factors can change the appearance of these curves. They include biomass amount, illumination condition, crop and soil type, water content and others. Figure 2 illustrates how different water content in winter wheat biomass changes the shape of these curves. The operational meteorological satellites such as Russian Meteor, American NOAA, and Chinese FY can be used for crop monitoring. Radar sensors are very promising for agricultural applications, especially for water content estimation.
Russian Systems for Crop Monitoring Russia has several operational systems consisting of several sub-systems (modules) for monitoring crop development and productivity (Report 2008). Figure 3 shows the main component of the system. The first module takes care of noise removal based on the available models. From these tasks, the atmospheric correction is difficult, since it requires information (water content, temperature and aerosol) that is not readily available. The vertical profiles are taken from the atmosphere sounding measurements and aerosol would be proportional to the horizontal visibility, although the total error could be up to 20%. The next module is cloud
Satellite-Based Systems for Agro-meteorological Monitoring
63
R, % 70 60 50 (1)
40 30
(2)
20 (3) (4)
10 0 0.54
0.8
1.2
1.5
1.8
2.05
2.3
l, mm
Fig. 2 Spectral reflectance of winter wheat under different water content in plant tissue (1 – <40%, 2 – 41–54%, 3 – 55–66%, 4 – >66%)
Fig. 3 The main components of the system
detection, consisting of two steps: first, negative NDVI values indicate clouds; second, thermal index was applied (Becker and Li, 1990) but only for the central part of the European Russia, and only during the vegetation period. The third module addresses crop selection for high spatial resolution images, which show separate fields; this procedure takes into consideration crop calendar. In the fourth and fifth modules, vegetation indices are calculated; ground data such as phenology, soil type, and soil moisture are used. The sixth module includes the calibration of satellite measurements from in-situ, data such as crop density. Since this parameter is measured frequently at agrometeorological stations,
64
A. Kleschenko et al.
Fig. 4 Maps of different states with estimated yield
it correlates with crop yield; the yield is estimated as “bad” if the harvest is below 60% annual yield detected from trend, “satisfactory”– 61–80%, “good” – 81–120% and “excellent” – exceeds 120%. The last module maps the different parameters for visualization purposes. An example is shown in Fig. 4 for the different regions of Russia. Is should mention that there could be two satisfactory gradations: the one for low density and the second for high density. The software used by the operational system was developed by the Russian Institute for Space Research. The 16-day assessment maps are transmitted to the server of the Russian Ministry of Agricultural. These maps are overlaid with the Landsat-based (for some parameters MODIS information is used) cultivated lands, lands occupied by winter crops, forests land, etc. Figure 5 shows that the correlation between winter wheat yield and NDVI in mid-May for the four major winter crop areas is quite strong. NDVI below 0.35 corresponds to winter wheat yield below 25 q/ha, which is a large reduction normally associated with drought.
Another Systems for Crop Monitoring The oldest satellite-based system was developed at the Food and Agricultural Organization (FAO) and presented at Fig. 6. This system is used in Africa and is
Satellite-Based Systems for Agro-meteorological Monitoring
Fig. 5 The relationship between winter wheat yield and NDVI for mid-May
Fig. 6 General structure of Artemis system
65
66
A. Kleschenko et al.
based on the Cold Cloud Duration day method used for estimation of precipitation. Another system is the European system, entitled MARS (Monitoring Agriculture with Remote Sensing) used from the late 1980s. MARS uses several crop growth and development models and high spatial resolution images to estimate crop area. Similar systems are available in the USA, China, Argentina, Brazil, and other small countries such as Morocco, Hungary, etc. The descriptions of the current satellite systems can be found in some Proceedings (Workshop, 2006). Most of the systems are based on digital images obtained from NOAA/AVHRR, MODIS (Aqua and Terra) and SPOT/Vegetation.
References Report 2008: Conducting scientific investigations and developing technology of crop state monitoring, expected productivity and rational placement crops on the base of economic, hydro-meteorological and satellite information with bioclimatic potential and effects of climate changes. Report for the Russian Ministry of Agricultural, No 1055 / 13; July 16, 261 p Becker F, Li ZL (1990) Temperature-independent spectral indices in thermal infrared bands. J Rem Sens Environ 132:17–33 Workshop 2006: Remote sensing support to crop yield forecast and area estimates at the International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XXXVI, W48 Volume, ISPRS WG VIII/10, Stresa, Italy, 155 p
Part II
Environment And Food Security: Diagnosis And Prediction
Monitoring Droughts and Pastures Productivity in Mongolia Using NOAA-AVHRR Data Leah Orlovsky, Felix Kogan, Eldad Eshed, and Chultem Dugarjav
Abstract The significant part of Mongolian economy is pastoral agriculture, which is a traditional scope of activity and main source of income for the rural population. Study of the natural vegetation dynamics is of essential interest both for decision-makers and herdsmen. During the last decades, Mongolia has suffered from prolonged droughts in combination with extensive grazing in many areas. This situation requires frequent monitoring environmental conditions and the state of pastures. This is an important and challenging security task for Mongolia since weather station network is limited for effective special monitoring and providing services and advises to decision-makers and herdsmen. During 1985–2004, the NOAA-AVHRR Global Vegetation Index (GVI) data set and its Vegetation Health (VH) products have been studied and used for analysis of pastoral changes in Mongolia. This paper discusses application of VH for early drought detection (one of the leading environmental disasters), monitoring drought impacts on pasture conditions and estimation of biomass production. Keywords Biomass production • Drought • Vegetation Health • Mongolia
Introduction The steppe grassland of Mongolia comprises one of the largest grassland ecosystem complexes of the world. About 78% of the country is devoted to nomadic pastoralism (Fernandez 2001). According to the FAO (2010), in the past 10 years, livestock number has increased by 32% and this trend is expected to continue. Mongolia’s livestock L. Orlovsky () and E. Eshed Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Israel e-mail:
[email protected] F. Kogan National Oceanic and Atmospheric Administration, USA C. Dugarjav Institute of Botany, Mongolian Academy of Sciences, Ulan Bator, Mongolia F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_8, © Springer Science+Business Media B.V. 2011
69
70
L. Orlovsky et al.
production depends much on the productivity of natural grassland. The major obstacle limiting grassland productivity is dry climate with frequent droughts, which suppress the development of grass biomass (Suttie 1999). Over 70% of the total pastureland of Mongolia has already been degraded due to prolonged drought conditions and pasture degradation due to overgrazing (Fernandez 2001; Adyasuren 1996; Manzano and Návar 2000). Pasture grazing by the livestock is the major form of land use which is the traditional way of life in Mongolia for Instead of 1,000 years – for centuries. Pastoral production in Mongolia is highly dependent on climate and weather. Four different ecosystems, representing the climatic intersection of Mongolia from south to north were addressed: forest steppe; typical steppe; dry steppe and desert steppe. Most Mongolian pastureland is located in arid, semi-arid and sub-humid zones with a limited amount of precipitation that fluctuates widely from year to year. This often results in drought conditions during the growing seasons. Drought is a typical phenomenon for the Mongolian climate occurring every 2–3 years. In the last 25 years, intensive droughts were observed in 1984, 1988–1989, 1996 and 2001–2002, causing major changes in biomass productivity (Orlovsky et al. 2005). Meteorological stations data are the main sources of information used in Mongolia for monitoring weather and consequently droughts. One of the greatest shortcomings of this source is limited density of this network: there are 35 meteorological stations for Mongolia’s area of over 1.5 million square kilometers. Another shortcoming is area representation: the station data describes weather conditions (especially rainfall) inside 3–5 km around the station. In addition, weather data are quite often not available in real time (economic, communication and other problems). Along with those problems, the current economical situation in Mongolia puts additional constraints on weather station network and operation. Therefore, scientists are looking into the application of satellite data for the assessment of pasture biomass in Mongolia (Kogan 2004). The socioeconomic problems in Mongolia dealing with pasture, such as ineffective land management, inadequate treatment of water wells, poor utilization of pasture lands and inadequate grazing practices are addressed through different programs. Many case studies claim that these problems are due to the geopolitical changes Mongolia experienced since the beginning of the 1990s, starting with the collapse of the Soviet Union. In this respect, the research of environmental change is of high importance because the Mongolian economy is based primarily on livestock production, which in turn, depends on pasture productivity. The environmental study are seeking ways to improve availability of pasture biomass for livestock and at the same time prevent further degradation of pasture lands through balanced distribution of grazing activity throughout the rangelands of Mongolia. During the 1985–2004, the NOAA-AVHRR Global Vegetation Index (GVI) data set and its Vegetation Health (VH) products were studied and used for analysis of pastoral changes in Mongolia in response to weather conditions. As the result of these efforts, the AVHRR-based system was developed for real time monitoring Mongolia’s pasture condition. This paper discusses and summarizes the application of VH for early drought detection (one of the leading environmental disasters), monitoring drought impacts and pasture conditions and estimation of biomass production. The real
Monitoring Droughts and Pastures Productivity in Mongolia Using NOAA-AVHRR Data
71
time assessments are expected to provide the following: (a) Monitoring drought start development and the impacts; (b) identify drought area and provide the nomads with the areas not affected by drought and consequently, pasture availability.
Data and Methods In situ measurements used in this study include data from field geobotanical observations and weather stations information. Remote sensing data were represented by the Advanced Very High Resolution Radiometer measurements flown on NOAA polar-orbiting operational satellites.
Experimental Research sites The geobotanical observations were performed at the four long-term monitoring sites (Fig. 1) located inside the main natural zones: forest steppe (Partizan), typical steppe (Tumentsogt), dry steppe (Bayanunjuul) and desert steppe (Bulgan). The sites had the following geographic and climate characteristics: • Bulgan (44° 00¢ 46² N/103° 33¢ 34² E; 1442MSL): 580 km SW of Ulaanbaatar; desert or dry steppe with loamy brown soils; annual precipitation 121 mm, January temperature −13.9°C, July 22.3°C; vegetation community: Stipa gobica, St. gloreosa, Allium polyrrhizum; season average biomass 212.34 kg/ha, minimum 22.80 kg/ha, maximum 696 kg/ha
Fig. 1 Location of the research sites in Mongolia
72
L. Orlovsky et al.
• Bayanunjuul (46° 49¢ 45² N/105° 46¢ 35² E; 1369MSL): 180 km SW of Ulaanbaatar; typical steppe with dark chestnut loamy, sandy and stony soils; annual precipitation 240 mm, January temperature −19°C, July 17–19°C; vegetation community: Cleistogenesis, Agropyron, Caragana, season average biomass 618.08 kg/ha, minimum 84 kg/ha, maximum 1,535 kg/ha • Tumensogt (47° 40¢ 31² N/112° 23¢ 55² E; 1000MSL): 520 km E of Ulaanbaatar; steppe zone; annual precipitation 280 mm, January temperature −17.3°C, July 22°C; vegetation community: Stipa grandis, St. sibirica; season average biomass 539.20 kg/ha, minimum 480 kg/ha, maximum 2,000 kg/ha • Partizan (48° 05¢ 34² N/106° 42¢ 14² E; 1320MSL): 53 km NW of Ulaanbaatar; forest steppe with dark chestnut soils; annual precipitation 250 mm, January temperature −23°C, July 16°C; vegetation community: Poa attenuata, Stipa krilovii; season average biomass 375.77 kg/ha, minimum 22.80 kg/ha, maximum 2,000 kg/ha At each location an area of 1 ha (100 by 100m) was fenced. Field observations from mid-June through September of 2002–2004 included (A) every 10-day measurements of vegetation weight, inside (control) and outside (grazing site) of the fenced area; (B) vegetation composition (density and fraction); (C) rainfall; and (D) soil moisture. At the Tumentsogt research site similar data also exist for 1982–1997 (Kogan 2004). The vegetation productivity was measured by cutting and weighing above-ground green biomass from four 1 m2 plots. The 4-plot average weight was expressed in kg per ha (kg ha−1). For the following 10-day measurement, the four plots were moved to the neighboring locations. Meteorological variables included monthly (in some cases 10-day) total precipitation for 1982–2004 (Bulgan) and 1989–2004 (Tumentsogt and Bayanunjuul collected from meteorological stations adjacent to the research sites.
Remote Sensing Data The radiances measured by the AVHRR sensor on-board NOAA 9, 11, 14, and 16 afternoon polar orbiting satellites were used in this study. The data were collected from the Global Vegetation Index (GVI) data set. The GVI includes radiance in the visible (Ch1), near infrared (Ch2) and two thermal bands (Ch4 and Ch5), NDVI, solar and satellite angles. The GVI data were sampled from one to 16 km spatial resolutions and from daily to 1-week temporal composite resolution (Unganai and Kogan 1998) for the 19-year period (1985–2004). Pasture conditions were estimated from a numerical analysis of weather-related components of the processed NDVI and brightness temperature (BT). These components were calculated by normalizing the weekly NDVI and BT values relative to the amplitude of their change during the years with favorable (moist) and unfavorable (dry) weather, which were selected based on maximum and minimum NDVI and BT during 1982–2005 for each pixel. Thus, weekly NDVI- and BT-based weather signal was amplified by ranking it on a linear scale from the absolute
Monitoring Droughts and Pastures Productivity in Mongolia Using NOAA-AVHRR Data
73
inimum to the absolute maximum of these parameters. These thresholds were m expressed as 0 and 100, respectively. The indices are called the Vegetation Condition Index (VCI), Temperature Condition Index (TCI) and Vegetation Health Index (Kogan 2001, 2004):
VCI = 100 × ( NDVI i − NDV I min ) / ( NDV I max − NDV I min )
(1)
TCI = 100 × ( BTmax − BTi ) / ( BTmax − BTmin )
(2)
VHI = a × VCI + (1 − a ) × TCI (3) where NDVIi, NDVImax, NDVImin, are processed weekly Normalized Difference Vegetation Index, its multiyear maximum and minimum respectively for the week (i). Notations are similar for the BT (Kogan 2001); a is a coefficient quantifying a share of VCI and TCI in VHI. The VCI, TCI and VHI provide numerical approximation of weather impacts on vegetation and characterize moisture, thermal and vegetation conditions, respectively. The indices values below 40 indicate drought conditions and reduction in biomass, while values higher than 60 describe the opposite conditions; the indices values between 40 and 60 characterize normal conditions (Kogan 2004).
Biomass Anomaly Index The biomass measurements for each area (inside and outside the fence) and time interval were expressed as the biomass anomaly index (B) the following way:
B Banomaly = i *100 Bmean
(4)
where Bi and Bmean are measured and mean biomass, for each decade (Kogan 2004). The Bmean was calculated for the available time series in each of the sites. The B index was use to compare with the remote sensing indices (VHI, TCI and VCI) for (1) early drought detection, identifying drought area, duration and intensity and for assessment of pastoral conditions in Mongolia; (2) modeling of pasture production.
Results and Discussion Monitoring Overgrazing Overgrazing effects on the plant community and soils are very destructive to pastures because of the reduction of canopy cover, compaction of soils (as a result of the physical effect of trampling), consequent soil crusting, reduced soil infiltration,
74
L. Orlovsky et al.
Table 1 Biomass productivity in the open for grazing and protected areas Fractional vegetation cover ratio between open and Total biomass in open/protected areas (kg/ha) protected areas (%) Area/year 2002 2003 2004 2002 2003 2004 Bulgan 1,490/1,730 4,610/5,410 580/970 86 85 59 Tumentsogt 5,680/7,371 9,901/10,740 10,930/14,221 77 92 77 Bayanunjuul 4,240/6,591 8,761/13,650 3,341/4,510 64 64 74 Partizan 7,108/7,529 9,685/9,522 6,636/7,661 94 102 87
enhanced surface runoff and diminishing soil water availability for plants (Manzano and Návar 2000). Field experiments helped to measure soil degradation in each site by comparing biomass productivity in the fenced and open grazing environments. The differences between grazing and non-grazing all sites are shown in Table 1, we identified a clear distinction. The open area (excluding 1 year at Partizan site) consistently presents lower levels of biomass than the protected area. We attribute this difference to the anthropogenic activity, which is present in each of the sites in different levels. We also assume that most of the anthropogenic pressure is mostly linked to grazing activity. According to our measurements the most affected site during observation period was Bayaunjuul. The least affected was Partizan and highest difference during observation period between the open and protected area was registered in Bulgan, during 2004 growing season. The open area in Bulgan at that year performed only 59% of total biomass in the protected area. The least difference for all sites between the two areas was during 2003 season where grazing impacts had the least effect, probably due to favorable climatic conditions and extensive growth of perennials.
Relationship Between Biomass Anomaly and Satellite Indices Figure 2 shows correlation between biomass anomaly (BA) and VH for four experimental sites (a–d). The R2 is very high 0.68–0.84. It is clearly seen that for three sites, below normal (<100%) BA is associated with VH below 40–50 due to drought-related vegetation stress. In the fourth site, Partizan, biomass reduction is initiated at the VH below 60. On the opposite side of the scale, above normal (<100) biomass is associated with VH > 60, indicating favorable conditions. For near normal (100%) biomass, association between two parameters is mixed since at that biomass level a combination of other conditions (wind, air humidity etc) might contribute to the biomass value. We should note that the experimental sites have some specific features contributing to the BA–VH relationship. For Tumentsogt, the best-fit was obtained for temperature index (TCI), with R2 = 0.74, compared to the R2 = 0.68 for VH. For Bayanunjuul and Bulgan, both VCI and VH are well correlated with BA. It is important to emphasize that these two are the most southern sites with typical desert
Monitoring Droughts and Pastures Productivity in Mongolia Using NOAA-AVHRR Data Bulgan. Biomass Anomaly
Biomass Anomaly
300 250 200 150 100 50 0
0
20
40
VHI
60
80
100
180 160 140 120 100 80 60 40 20 0
0
20
40
60
80
100
Partizan 140 Biomass Anomaly
Biomass Anomaly
Tumentsogt
VHI
Bayanunjuul 200 180 160 140 120 100 80 60 40 20 0
75
0
20
40
VHI
60
80
100
120 100 80 60 40 20 0
0
20
40 VHI
60
80
Fig. 2 Correlation of 10-day biomass anomaly (y) with VH (x), 2002–2004. (a): y = 3x − 37.6, R2 = 0.84, n = 30, SE = 38, P < 1.05E−11; (b): y = 1.6x ± 26SE, R2 = 0.68, n = 105, P < 3.29E−26; (c): y = 1.55x + 31.3 ± 20.7, R2 = 0.80, n = 22, P < 1.86E−08; (d): y = 1.9x − 24 ± 12.6SE, R2 = 0.71, n = 17, P < 2.09E–5
and dry steppe ecosystems, respectively, where precipitation deficit is the most important factor regulating the amount of biomass. The BA–VH relationship explains 71% of variance for Partizan and 80% for the other three sites. Lower reliability for Partizan is explained by a heterogenic environment represented by a mixture of grass and forest lands.
Monitoring Drought After finding best-fitted VH index the following VH-based classification scheme of weather-created conditions in Mongolia was developed: (a) extreme drought (0–25) and (b) moderate drought (26–50); favorable conditions (51–65), very favorable (66–85) and extremely favorable (86–100). Using this classification it is possible to estimate pasture conditions and possible biomass. Figure 3 shows the end-of-August VH-estimated weather conditions in Mongolia during the experimental phase of the project (2002–2004). From 3 years, the 2002 growing season was extremely dry. Drought affected most part of the country and was the most severe in the central and northern Mongolia (Fig. 3a). Moreover, it was the longest (not shown here). The 2003 growing season was the most favorable year during the observation period for the entire Mongolia and also in all four experimental sites. However, in the north and northeastern Mongolia where Partizan
76
L. Orlovsky et al. 511779.
511779.
511779.
97
102
107
97511779.
102511779.
107 511779.
112
511779.
117
511779.
38
645759.
41
645759.
44
645759.
47
645759.
50
645759.
53
38645759. 41645759. 44645759. 47645759. 50645759. 53645759.
511779.
645759.
92
92 511779.
0-25 511779.
511779.
117 511779.
VH 51-65
66-85 511779.
86-100 511779.
97
102
107
97511779.
102511779.
107 511779.
112
511779.
117
511779.
38
645759.
41
645759.
44
645759.
47
645759.
50
645759.
53
645759.
38645759. 41645759. 44645759. 47645759. 50645759. 53645759.
92
26-50
112 511779.
92 511779.
112 511779.
117 511779.
VH
0-25
26-50
51-65
66-85
86-100
Fig. 3 Classified VH conditions at the end of August for growing seasons 2002 (a), 2003 (b), 2004 (c) in Mongolia
Monitoring Droughts and Pastures Productivity in Mongolia Using NOAA-AVHRR Data 511779.
511779.
511779.
97
102
107
97511779.
102511779.
107 511779.
112
511779.
117
511779.
38
645759.
41
645759.
44
645759.
47
645759.
50
645759.
53
38645759. 41645759. 44645759. 47645759. 50645759. 53645759.
511779.
645759.
92
77
92 511779.
112 511779.
117 511779.
VH
0-25
26-50
51-65
66-85
86-100
Fig. 3 (continued)
and Tumentsogt sites were located, drought conditions during the growing season were shifting between light and moderate (Fig. 3b). The 2004 growing season was characterized by extreme drought in the northeastern Mongolia affecting the area of Tumentsogt. The central and southern parts of the country, where Bayanunjuul and Partizan sites were located, suffered from moderate drought; light drought affected the central regions of Mongolia, where Bulgan site was located (Fig. 3c). The presented classification and analysis can be used in real time to provide herders with spatial dynamics of vegetation state helping them to determine better pasture condition in order to move their herds during the growing season. Long term VH-based drought information is also important to combat pasture degradation.
Monitoring Grazing in the Central Mongolia (Tov Aimag) The region of Tov aimag (Mongolian: Töv, center) occupies 74,000 km2 with about 200,000 rural population. This number though doesn’t include the Mongolia’s capital, Ulaanbaatar, with 770,000 inhabitants, 25% of which are living in the outskirts of the city, and continue traditional livestock-breeding mode of life. One of the outcomes of the geopolitical changes in Mongolia after collapse of the socialist system in the early 1990s, was stable migration of herders to the Tov aimag region
L. Orlovsky et al. 100
180
90
160
80
140
70
120
60
100
50
80
40
60
30
40
20
20
10
0
0
Vegetation Health Index
200
19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04
Preipitation anomaly
78
Year Precipitation
VH
Linear (VH)
Linear (Precipitation)
Fig. 4 Comparing of Precipitation Anomaly with seasonal average VH during 1982–2004, Tov Aimag, Mongolia
and closer to Ulaanbaatar. As strong migration intensified, it increased grazing activity, pasture degradation, soil destruction and other negative anthropogenic effects. Therefore, we examined 1982–2004 VH dynamics at the 19,200 km2 (10*10 16 km2 GVI pixels) Tov aimag area and compared these results with the dynamics of precipitation anomaly calculated as: P Panomaly = i *100, Pmaen where Pi is a given yearly total precipitation (mm) and Pmean – mean annual precipitation for 1960–1990 (mm) (Monthly Climatic Data for the World, 2010). The 1982–2004 time series of P and VH parameters are shown in Fig 4. Although, the precipitation trend line shows slight upward trend the VH shows a clear downward trend indicating a reduction of biomass conditions from 60 (moderately favorable) to 40 (moderately dry). This finding is likely to be a result of the large anthropogenic activity in this area, induced also during the 1990s.
Conclusions In the past, weather station data were used for estimation of climate impacts on vegetation and specifically pasture condition. Currently, satellite data has shown great potential for monitoring purposes. The application of 30-year NOAA-AVHRR vegetation health data showed them to be very efficient due to excellent simulation results, appropriate temporal and special resolution and longevity. Furthermore,
Monitoring Droughts and Pastures Productivity in Mongolia Using NOAA-AVHRR Data
79
spectral resolution, wide swath and large area coverage provide more advantages for their use advantage in Mongolia where most of the country massively covered by pasture land. The main part of this work included correlation between the BA and the three VH indices. Besides providing estimation of pasture conditions and reduction in biomass, the VH technology might be also used for providing advises on better pasture state and finding areas for better grazing. Also, the VH might provide information on grazing pressure. The AVHRR-based VH data could be used as a sole source of information for monitoring pasture conditions throughout the entire Mongolia. Considering Mongolia’s vast land area, harsh climate, limited weather network, spatial distribution of herds and lack of communication, these results can be used to provide advice to farmers on the quantity of available biomass when they search for better grazing areas. These results are also useful for monitoring pasture productivity and for a real-time assessment of pasture conditions and biomass production. These assessments are particularly beneficial in areas where weather data are not available and/or non-representative due to the sparseness of the weather-observing network.
References Adyasuren T (1996) Economic reform policies and sustainable development in a transitional economy: the case of Mongolia. In: Kumssa A, Khan HA (eds) Transnational economies and regional economic development strategies: lessons from five low-income developing countries, UNCRD Research Report Series 19:91–104 FAO (2010) Production: Live animals. http://faostat.fao.org. Accessed 26 July 2010 Fernandez GME (2001) Vegetation change along gradients from water sources in three grazed Mongolian ecosystems. Plant Ecol 157:101–118 Kogan FN (2001) Operational space technology for global vegetation assessment. Bull Am Meteor Soc 82(9):1949–1964 Kogan FN (2004) Derivation of pasture biomass in Mongolia from AVHRR-based vegetation health indices. Int J Remote Sens 25:2889–2896 Manzano MG, Návar J (2000) Processes of desertification by goats overgrazing in the Tamaulipan thornscrub (matorral) in north-eastern Mexico. J Arid Environ 44:1–17 Monthly Climatic Data for the World (2010). http://www7.ncdc.noaa.gov/IPS/. Accessed 10 Feb 2010 Orlovsky L, Kogan FN et al (2005) Estimation of seasonal dynamics of arid and semi arid zone pasture productivity in the Mongolian Gobi using NOAA-AVHRR data. Final Report submitted to the U.S. Agency for International Development Suttie JM (1999) Grassland and pasture crops: Country Pasture/Forage Resource Profile-Mongolia: http://www.fao.org/ag/AGP/AGPC/doc/Counprof/ Mongolia. Accessed 11 Jan 2005 Unganai LS, Kogan FN (1998) Drought monitoring and corn yield estimation in Southern Africa from AVHRR data. Remote Sens Environ 63:219–232
Satellite-Derived Information on Snow Cover for Agriculture Applications in Ukraine Peter Romanov
Abstract This paper demonstrates how NOAA interactive satellite-based maps of snow cover can be used in the assessment of unfavorable agricultural conditions in Ukraine. The focus was on two events, the extensive winterkill in winter of 2002–2003 and the drought in the early 2007. Both events had a strong adverse effect on the yield and on the production of major grain crops. The analysis of NOAA daily snow maps has revealed an extremely short duration of snow in Ukraine in winter of 2006–2007. This is indicative of lower winter-time precipitation that contributed to the soil dryness in spring 2007. To identify potential crop freeze damage we have estimated the minimum temperature of snow-free land surface from snow charts combined with satellite land surface temperature retrievals. Temperatures below –18°C indicating potential winterkill were observed in the Central and Eastern Ukraine in December 2002. Keywords Satellite-based snow cover maps • minimum temperature • winterkill
Introduction For over 35 years satellite observations have been used to monitor the global snow cover distribution and snow pack properties. In the last decade a substantial increase in the number of satellite sensors for snow monitoring, as well as enhanced sensors capabilities and image analysis techniques have led to a noticeable improvement in the derived snow cover products. This concerns the accuracy of the maps, their spatial resolution, coverage and the update frequency. The urge to fully utilize satellite observing capabilities, in particular, their multispectral measurements and high spatial resolution, stimulated the development of automated algorithms to identify snow cover in satellite imagery and to generate maps of snow cover distribution. Current P. Romanov (*) University of Maryland, College Park, MD, USA e-mail:
[email protected]
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_9, © Springer Science+Business Media B.V. 2011
81
82
P. Romanov
operational automated systems derive snow cover from satellite observations in the visible and infrared spectral bands (e.g., Hall et al. 2002; Romanov et al. 2003), from observations in the microwave (Kongoli et al. 2004; Derksen et al. 2003), or use synergy of the two techniques (Foster et al. 2008; Romanov et al. 2000). The accuracy of snow identification and mapping by automated snow detection algorithms is high, often exceeding 90%, however the algorithm performance depends on the topography, vegetation cover, weather conditions, snow pack properties and other environmental factors (e.g. Simic et al. 2004). As a result the accuracy of automated snow maps changes with the surface type and varies throughout the year. Relatively short lifespan of satellite sensors, especially back in 1980s and 1990s, differences in their spectral bands and spatial resolution, degradation of sensitivity of some sensors with time and satellite orbital drift complicate developing consistent long-term snow datasets from these data. Identification of snow in satellite imagery by visual inspection of satellite imagery is the oldest snow cover mapping technique. Since 1972 this approach has been routinely used at NOAA to generate weekly maps of snow and ice distribution in the Northern Hemisphere. In 1999 a computer-based Interactive Multisensor Snow and Ice Mapping System (IMS) was implemented to facilitate the image analysis by human analysts (Ramsay 1998). This allowed the spatial resolution of the maps to improve from 180 to 24 km and to start daily snow mapping. In 2004 the IMS system was upgraded and spatial resolution of the product was further increased to 4 km (Helfrich et al. 2007). IMS maps of snow and ice cover are considered as the primary NOAA snow cover product and are incorporated in all global and mesoscale operational numerical weather prediction models run by NOAA National Centers for Environmental Prediction (NCEP). With over 35 years of continuous snow cover monitoring, NOAA interactive snow maps present a unique source of information for global climate change studies (Frei and Robinson 1999). High spatial resolution and daily updates of IMS maps also make them potentially useful for various environmental and practical applications at regional and local scale. In the Ukraine information on the snow cover distribution and extent is of major importance for agriculture. The length of snow season ranges from several weeks in the very south of the country to several months in the north. Water accumulated in the snow pack and released through the snowmelt is critical for the winter crops development in early spring. Snow pack is also an important factor preventing the frost and freeze damage of winter crops. Most information on the snow cover distribution in Ukraine is obtained from ground-based meteorological stations. Satellite observations of snow can complement surface observations by providing spatially detailed and frequent in time information on the snow cover distribution and seasonal change. In this paper we examine daily IMS snow cover maps over Ukraine for ten winter seasons, from 1999–2000 to 2008–2009. The objective of the work was to demonstrate the efficacy of these maps for agriculture applications and in particular their usefulness in the analysis of spring-time drought conditions and for identification of areas of potential freeze damage of winter grains. The primary focus was on the winter season of 2002–2003 when a severe winterkill destroyed about 55% of winter wheat in Ukraine and on the 2006–2007 winter season, which preceded the intensive drought in spring and summer 2007.
Satellite-Derived Information on Snow Cover for Agriculture Applications in Ukraine
83
Data In this study we have used NOAA daily snow and ice cover maps for the Northern Hemisphere generated within the IMS system. Original IMS maps have polar stereographic projection with every pixel in the map corresponding to the land surface labeled as “snow-free land” or “snow”. Pixels over water are classified as “ice-free water” or “ice”. In contrast to automated snow products, IMS maps do not contain “undetermined” pixels. If persistent cloud cover prevents from reliably delineating snow-covered areas the analyst makes an intelligent guess regarding the possible change in the snow cover distribution beneath the clouds or retains the snow cover distribution from the previous day product. IMS snow and ice charts for ten winter seasons (November to March) from 1999– 2000 to 2008–2009 were acquired from the National Snow and Ice Data Center (NSIDC). For convenience all daily maps were regridded to a latitude-longitude projection with a grid cell size of 0.05° or about 5km. The portion of the map within 43–53°N and 19–41°E completely covering Ukraine was extracted and saved. Figure 1 presents an example of the original IMS daily snow cover map for the Northern Hemisphere and a portion of this map over the study area in the latitude–longitude projection.
Application of IMS Snow Maps Snow Cover Duration and Drought Seasonal snow cover is an important hydrological and climate feature of mid and highlatitude areas. Longer duration of continuous snow cover is most often associated with larger wintertime precipitation amounts and corresponds to larger moisture accumulation in the snow pack at the end of the cold season. Shorter snow cover duration means less moisture accumulation in the snow pack at the end of the cold season and, hence, more probable shortage of water supply for crops in spring.
90
Northern Hemisphere Snow & Ice Chart
52
135
51 45
50 49
180
48
90 60 45 −135
0
47
30
46 15
45
−45 snow ice
Jan 31,
−90 Sat Jan 31
2009
22
24
26
28
30
2009 32
44 34
36
38
40
Fig. 1 IMS snow and ice cover map for January 31, 2009 (left) and its portion over the study area regridded to the latitude-longitude projection (right). Snow is shown in white, ice is light gray, snow-free land surface is medium-gray and ice free water is dark gray
84
P. Romanov
In this study we processed all daily IMS snow cover maps over Ukraine to determine the duration of snow cover for every winter season since 1999 and average duration of snow cover for the last 10-year long period. The duration of snow cover for every grid cell was estimated by calculating the number of days the grid cell was labeled by IMS analysts as “snow covered”. Maps of the yearly and multiyear-mean snow cover duration are presented in Fig. 2. Overall they demonstrate a general increase in the length of the snow season in the northern and north-eastern direction and it substantial variation from year to year. The yearly duration of snow cover ranges from 3–4 weeks in the south of Ukraine to 3–4 months in the north. As it follows from the maps in Fig. 2, the winter season of 2006–2007 was characterized by unusually short duration of snow cover. Over most of Ukraine snow cover remained on the ground for about twice less time than normal. This reduction in the snow cover duration and associated drop in the available melt water may have contributed to the shortage of water supply in spring and to severe drought in early summer 2007. According to the data of United States Department of Agriculture (USDA) the drought has affected Ukraine’s winter grain crops by noticeably reducing the yield of Ukraine’s two major winter grain crops, wheat and barley (see Fig. 3). It is interesting that the decrease in the yield of winter barley in 2007 was much more pronounced than the corresponding decrease in the yield of winter wheat. This difference may be explained by the fact that drought conditions were more severe in the east of Ukraine, where most barley is produced. As it is seen from Fig. 2, the winter season of 2000–2001 presents another example of shorter than average snow cover duration. In this year the largest reduction in the snow cover duration was observed in the south of the central part of Ukraine. This anomaly, however, did not develop into a wide-spread drought similar to the one in 2007. Lower than usual soil moisture content in agricultural areas in Ukraine was indeed observed throughout most of the winter months (USDA 2001), however, above-normal early spring rainfall compensated for low winter precipitation. As a result, winter-wheat and winter barley yields in 2001 were close record levels of the decade.
Extent of Winterkill Winterkill is one of primary reasons for the winter crop loss in Ukraine. While in the winter-dormant stage, winter wheat typically survives temperatures down to −17°C to −18°C (USDA 2006). A combination of lower temperature and shallow or absent snow cover presents a potentially damaging condition that may result in the injury to the crop. The most recent extensive winterkill occurred in Ukraine in December 2002 causing the country’s wheat production in 2003 to decrease by more than four times from the previous year and by more than three times from the annual average level. In this study we have used daily IMS snow cover maps along with satellite-derived information on the land surface temperature (LST) to identify areas of possible winterkill.
25
2008 –2009
40
60
1999 –2009 mean
80
100
46 44
48
46
25
50
48
44
52
50
25
2007 –2008
25
2004 –2005
25
2001 –2002
52
Duration of snow cover (days)
20
25
46
46 44
48
48
2006 –2007
50
50
44
52
52
25
46
46 44
48
48
2003 –2004
50
44
52
50
44
52
25
46
46 44
48
48
2000 –2001
50
50
44
46
48
50
52
44
46
48
50
52
44
46
48
50
52
Fig. 2 Snow cover duration over Ukraine for winter seasons 1999–2000 to 2008–2009 and the mean snow cover duration for the 10-year period calculated from daily IMS snow cover maps. Areas outside Ukraine are shown in white, water is black and snow cover duration is presented in shades of gray
25
2005 –2006
25
2002 –2003
25
1999 –2000
52
52
Satellite-Derived Information on Snow Cover for Agriculture Applications in Ukraine 85
86
P. Romanov 4
Yield, MT/HA
3
2
1 Barley, Yield (MT/HA) Wheat, Yield (MT/HA)
0 1998
2000
2002
2004
2006
2008
2010
Year Fig. 3 Yearly yield of wheat and barley in Ukraine in 1999–2009 (US Department of Agriculture). For the year 2009 projected yield values are given
Information on the land surface temperature was obtained from LST maps derived from observations in the infrared spectral band of Moderate Resolution Imaging Spectroradiometer (MODIS) onboard Aqua and Terra satellites. Daily daytime and nighttime LST maps from MODIS Aqua and Terra (MOD11C1 and MYD11C1 products, respectively) were acquired from NASA Distributed Active Archive Center (DAAC). Since clouds are opaque in the infrared, information on the LST is available only in clear sky conditions. The projection of MOD11C1 and MYD11C1 products is latitude–longitude with a grid cell size of 5 km, i.e. the same as the projection of regridded IMS snow cover maps over Ukraine. This feature of MODIS LST maps has allowed for an easy pixel-by-pixel spatial matching of the two products. MODIS-based daily maps of the land surface temperature and corresponding maps of snow cover distribution were used to generate maps of the monthly minimum land surface temperature for snow-free land surface. Both daytime and nighttime LST maps have been processed. The minimum temperature maps for three winter seasons from 1999–2000 to 2001–2002 were derived from MODIS Terra LST products, whereas for later years MODIS Aqua retrievals have been used. The reason for the use of MODIS Aqua is that Aqua nighttime overpass time is later than Terra (1 AM vs 10 PM) and therefore its observations have a better chance to capture the nighttime minimum temperature. Figure 4 presents maps of the minimum temperature of snow-free land surface for the month of December for 9 years from 1999 to 2008. Three shades of gray correspond to surface temperatures above −17.5°C, within −17.5°C to −18.5°C and below −18.5°C. The map for December 2002 clearly shows a large area of very low minimum temperatures of snow-free land surface in the central and eastern Ukraine indicating potential freeze damage. There is much similarity between the spatial
BELOW
−18
−17
Min Surface Temperature, C
−19
ABOVE
44
46
46
25
48
48
44
50
50
December 2007
52
52
44
46
46
25
48
48
44
50
50
December 2004
52
44
52
25
46
46 44
48
48
December 2001
50
50
25
December 2008
25
December 2004
25
December 2002
44
46
48
50
52
44
46
48
50
52
44
46
48
50
52
Fig. 4 Maps of minimum temperature of snow-free land surface in the month of December of the years 2000–2008 in Ukraine. Areas of potential winter crop freeze damage (minimum temperatures below −18.5°C without snow cover on the ground) are shown in dark gray
25
December 2006
25
December 2003
25
December 2000
52
52
Satellite-Derived Information on Snow Cover for Agriculture Applications in Ukraine 87
88
P. Romanov
Fig. 5 Percent of winterkill damage of winter wheat in winter 2002–2003 based on in-situ assessment in Ukraine (US Department of Agriculture)
distribution of these areas and ground-based estimates of winter wheat damage as reported by USDA (see Fig. 5). According to in situ data most affected were also the eastern and especially the central part of Ukraine where the reported loss due to the winterkill ranged from 61% to 88%. The analysis of maps for other years in Fig. 4 show little freeze damage to winter crops. Isolated small areas of potential winterkill are most probably the result of occasional misclassification of cloud as the cloud-clear land surface. Most often these misclassifications occur within a large area covered by clouds. The algorithm retains an observation with the minimum estimated surface temperature in every pixel of the map over a month-long period and therefore tends to accumulate these errors.
Discussion In Ukraine snow cover determines to a large extent the condition of winter crops in the end of winter and their development in early spring. This explains the importance of information on the spatial distribution of snow cover and its seasonal changes for agricultural applications. In the same time snow cover is by no means the only factor that has to be accounted for in the winter crop condition monitoring and yield forecasting. Short duration of seasonal snow cover should be viewed only as an indirect evidence of reduced wintertime precipitation that may contribute to the development of drought conditions later in spring. Other factors, such as soil moisture and precipitation in the preceding fall season and, most notably, the amount of liquid precipitation in early spring also affect availability of soil moisture for the crop development at the beginning of the growth season.
Satellite-Derived Information on Snow Cover for Agriculture Applications in Ukraine
89
The lack of snow cover protecting crops from low wintertime temperatures and associated freeze damage is one of the primary causes for the winter crop loss in Ukraine. Still this is not the only threat to the successful survival of crops through the winter season. For example, in winter of 2002–2003 the freeze damage of crops in December 2002 was further exacerbated by persistent ice crusting in February and March 2003 (USDA 2003). In situ inspection of winter grain crops in spring focuses at the assessment of the damage from all adverse factors combined but can hardly provide estimates of their individual contribution. In this study we presented estimates of the crop freeze damage only in the beginning of the winter season when winterkill from freezing is most probable. A more comprehensive approach to identify potential frost damage should involve the analysis of snow cover and surface temperatures expanded to the whole period of the year when freezing conditions are possible. This analysis should also account for the grain development stage. Winter crops are much less freeze resistant in the post-dormant stage therefore injury during these periods may occur at temperatures only slightly below freezing (Shroyer et al. 1995). Although IMS snow cover maps are one of the most reliable sources of information on the snow cover distribution, there are several limitations associated with these maps which the user has to be aware of. These limitations may not be critical for the analysis of seasonal changes of hemisphere or continental snow extent, but should considered if maps are used for smaller scale applications. First, due to extensive workload, IMS analysts may not be able to review and update the state of every pixel in the Northern Hemisphere map on a daily basis. Specifically delays are most probable during fast large-scale changes in the snow cover distribution, in particular during active snowmelt in spring. In some cases changes in the snow extent since the day before may be too small for analysts to identify them. Second, since analysts primarily rely on the satellite imagery in the visible spectral band, clouds sometimes hamper accurate and timely reproduction of changes in the snow extent. Clouds also present a serious problem for surface temperature retrievals when trying to identify areas of potential winterkill. Although the coldest surface temperatures are typically associated with cloud-clear skies, cloud-caused gaps in satellite-derived maps of surface temperature may still result in misses of freeze damage cases. Confusion of clouds with cloud-clear land surface by satellite image classification algorithms is another important issue. In clear sky conditions the accuracy of surface temperature estimates from satellite observations in the infrared is about 2– to 3 K. However misinterpretation of a cloudy pixel as clear most often causes a severe underestimation of the land surface temperature and hence may lead to false freeze-damaged area identification. In Fig. 4 spurious identifications of potential freeze damage in the form of isolated dark pixels or small clusters of dark pixels are clearly seen in all maps. The fact that cloud misclassifications and hence, freeze damage false identifications rarely occur over large areas may help to filter at least part of these errors out by examining spatial variations of the derived land surface temperature.
90
P. Romanov
Conclusions In this work NOAA daily interactive snow cover maps have been used to estimate the duration of seasonal snow cover over Ukraine and to determine the monthly minimum temperature of the snow-free land surface. The analysis of information on the snow cover duration and on the minimum temperature of snow-free land surface for the last 10 years has shown that these products indeed present an additional piece of information complimentary to available in situ agricultural reports. This information may contribute to the assessment of potential crop damage due to winterkill and to the prediction of drought conditions in spring. The next step in the improvement of the two products presented in the paper consists in the use of daily snow maps derived with automated snow cover mapping algorithms. The most promising approach involves application of algorithms that combine satellite-based snow retrievals in the visible/infrared spectral bands and in the microwave. This combination allows for generation of daily continuous (i.e., gap-free) snow cover maps at a spatial resolution of several kilometers similar to IMS maps. As compared to IMS analysts, automated algorithms can detect smaller changes in the snow cover distribution and therefore have a potential to more accurately reproduce daily changes of the snow extent. Some information on the snow pack properties, particularly on the snow depth can be obtained from satellite observations in the microwave. This data can also help in delineating areas affected by winterkill.
References Derksen CA, Walker A, LeDrew E, Goodison B (2003) Combining SMMR and SMM/I data for time series analysis of central North American snow water equivalent. J Hydrometeorol 4: 304–316 Foster JL, Hall DK, Eylander J, Kim EJ, Riggs GA, Tedesco M, Nghiem S, Kelly REJ, Choudhury B, Reichle R (2008) A new blended global snow product using visible, microwave and scatterometer satellite data. 88th Annual Meeting of American Meteorological Society, 20–24 January 2008, New Orleans, LA. http://ams.confex.com/ams/pdfpapers/130069.pdf Frei A, Robinson DA (1999) Northern Hemisphere snow extent: regional variability 1972 to 1994. Int J Climatol 19:1535–1560 Hall DK, Riggs GA, Salomonson VV, DiGirolamo NE, Bayr KJ (2002) MODIS snow-cover products. Remote Sens Environ 83:181–194 Helfrich SR, McNamara D, Ramsay BH, Baldwin T, Kasheta T (2007) Enhancements to and forthcoming developments in the Interactive Multisensor Snow and Ice Mapping System (IMS). Hydrol Process 21:1576–1586 Kongoli C, Grody NC, Ferraro RR (2004) Interpretation of AMSU microwave measurements for the retrievals of snow water equivalent and snow depth. J Geophys Res-Atmos 109 (D24), Art. No. D24111 Dec 29, 2004 Ramsay B (1998) The interactive multisensor snow and ice mapping system. Hydrol Process 12:1537–1546 Romanov P, Tarpley D, Gutman G, Carroll TR (2003) Mapping and monitoring of the snow cover fraction over North America. J Geophys Res 108(D16):8619. doi:10.1029/2002JD003142
Satellite-Derived Information on Snow Cover for Agriculture Applications in Ukraine
91
Romanov P, Gutman G, Csiszar I (2000) Automated monitoring of snow cover over North America with multispectral satellite data. J Appl Meteorol 39:1866–1880 Shroyer JP, Mikesell ME, Paulsen GM (1995) Spring freeze injury to Kansas wheat. Cooperative Extension Service, Kansas State University, Manhattan. http://sanangelo.tamu.edu/agronomy/ pdf/spgfrze.pdf Simic A, Fernandes R, Brown R, Romanov P, Park W (2004) Validation of VEGETATION, MODIS, and GOES plus SSM/I snow-cover products over Canada based on surface snow depth observations. Hydrol Process 18:1089–1104 USDA (2001) Russia and Ukraine: soil moisture reserves dwindle as dryness continues United States Department of Agriculture, Foreign Agricultural Service. Production, Market and Trade Feature Reports-2001, February 7, 2001. http://www.fas.usda.gov/pecad2/highlights/2001/02/ 01feb07.htm USDA (2003) Ukraine: Winter Wheat Situation. United States Department of Agriculture, Foreign Agricultural Service. Production, Market and Trade Feature Reports-2003, March 18, 2003. http://www.fas.usda.gov/pecad2/highlights/2003/03/ukr_march03/index.htm USDA (2006) Ukraine: Frost Damage to Winter Wheat in Eastern Region. United States Department of Agriculture, Foreign Agricultural Service. Commodity Intelligence Report February 10, 2006. http://www.pecad.fas.usda.gov/highlights/2006/02/ukr_09feb2006/
Grain Yield Prediction in the Russian Federation Anna Strashnaya, Tamara Maksimenkova, and Olga Chub
Abstract Some of the peculiarities of forecasting of grain-crops yield and leguminous plants in the Russian Federation are considered. The basis for a probabilistic empirical statistical method for step-by-step forecasting of yield productivity and croppage for the regions and the Russian Federation as whole, based on the agrometeorological factors and implemented on IBM personal computers is stated. The dynamic-statistical model presented is in the framework of the Information-Prognostic System (IPS). A possibility of forecasting yield productivity at the stage of harvesting during years of unfavorable conditions is shown. The results of skill scores forecasts are given. Keywords Grain yield • Agrometeorological factors • Dynamical–statistical models
Results and Discussion Timely grain crop forecasting is one of the most important tasks in the Russian Space Agency, Roshydrome. Accurate forecasting helps to plan grain distribution, purchases, and harvest planning. Regional grain yield forecasting in Russia is a very complex problem because of the considerable variations in weather and climate conditions and many crop varieties. Figure 1 demonstrates the contribution of each crop to the total grain yields. As seen, winter wheat, spring wheat, and spring barley are the major contributors (40%, 21% and 19%, respectively) to the total grain yeild. The primary winter wheat growing area is located in a very fertile Chernozem soil zone, but with a dry climate and very frequent droughts. The main area under spring wheat is Povolzhie, Ural, and Siberia, which is also characterized by dry climate with frequent droughts, especially during May–June. In droughty years, productivity of grain crops decreases from 25 to 45% or more. A. Strashnaya (*), T. Maksimenkova, and O. Chub Hydrometeorological Research Centre, Roshydromet, Ministry of Natural Resources and Ecology of the Russian Federation, Moscow, Russia e-mail:
[email protected];
[email protected]
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_10, © Springer Science +Business Media B.V. 2011
93
94
A. Strashnaya et al. 6
6
2
40
6
19 21
Winter wheat Barley winter and summer Rye Leguminous cultures
Spring wheat Oats Late grain crops
Fig. 1 Percent of different crops’ contribution to the total Russia grain production
The non-Chernozem zone has much better moisture conditions but inferior soil. Adverse grain harvest conditions are characterized by heavy, long-lasting rains and high air humidity, resulting in waterlogging, crop lodging, and prolonged harvest time, which results in reduced yield, especially in the northern European Russia and in crop-producing areas of the central Siberia. By the example, the Central federal district (Fig. 2) shows the probability (% years) of a high reduction in the yields (by 25% and more relative to a trend) due to droughts and adverse harvest conditions. Large-area crop yield (Yп) time series are normally described by a function consisting of: deterministic component, which approximate yield tendency (Yтр,) and is controlled by agricultural technology, economic and political factors, and random component, expressed as yield deviation from a trend (DY), which is controlled by weather conditions.
YΠ = YTp + ∆Y
(1)
Figure 3 shows the mean Russian grain yield with two components: deterministic, shown in the form of straight line and random, shown as a deviation from the trend. The approximated trends indicate major changes in deterministic component since the 1980s. Three trend periods related to political, economic, and agricultural technology were detected: before disintegration of the USSR (prior to 1990), immediately after (1991–1999), and the most recent (after 1999). The piecewise-linear trends were estimated by a polynomial yт = at + b, where t is year number and a and b are slope and intercept. As seen, during the early stable period (1980–1990) a very intensive technology-related yield growth was observed; during 1990–1999, annual grain yield reduction was 0.36 t/ha due to structural changes in agriculture, problems
Grain Yield Prediction in the Russian Federation
95
Fig. 2 Probability (%) of yield reduction (25% below technological yield level) due to droughts and adverse harvest conditions in the Central Federal District of Russia
1,90
Productivity, t/ha
1,70
1,50
1,30
1,10
2006
2004
2002
2000
1998
1996
1994
1992
1990
1988
1986
1984
1982
1980
0,90 Years
Fig. 3 Russian mean grain yield (t/ha) dynamics and deterministic linear trend, 1980–2007
with material and technical base of farms, reduced fertilizer, and inadequate crop protection, and also due to changes in the accounting and reporting technique. Starting from 2000, the grain yield gradually increases again due to some improvements in agricultural technology, increased investments, and development of farming.
96
A. Strashnaya et al.
The DY is normally controlled by weather and climate factors. In the crop yield anomaly models (regression equations) they were represented by the meteorological parameters (temperature, precipitation, etc.) and agrometeorological indices (soil moisture and combination of rainfall and temperature). Prognostic models were obtained for each of eight regions federal districts. Due to collinearity between predictors, the principal components analysis method was used to pass from the original (correlated) variables Y1, …, Y8 to the non-correlated major components Z1, … , Z8. In this case, each component is a linear combination of the original variables (Z = a1Y1 + a2 Y2 + + a8 Y8). This regression equation obtained from major components and the transition to original variables allowed us to find the required equation in which the regression coefficients are not correlated Yp = −0,022 + 0.08Y1 + 0.089Y2 + 0.087Y3 + 0.20Y4 + 0.12Y5 + 0.19Y6 + 0.15Y7 + 0.105Y8 R = 0.99; SYp = 0.09 t / ha ,
(2)
Equation (2) represent a model to forecast average Russian grain yield (t/ha), where Y1 … Y8 are the average crop yield in the Central, Volga-Vyatka, Central Chernozem, Povolzhie, North Caucasus, Urals, West Siberian and East Siberian administrative regions (t/ha); R is the multiple correlation coefficient; SYp is the equation error (Fig. 4). The “PROGNOZ” program, developed by the Russian Hydrometcenter, provides IBM PC-based 1 month-step forecasts of grain yield. In addition to the growing season conditions, a technique was developed to predict yield reduction during harvesting time. The data shown indicates that during the period from grain maturity 120,0
Total gathering, mln t.
110,0 100,0 90,0 80,0 70,0 60,0
Actual data
Forecasting data
Fig. 4 Russia’s average forecasted and actual grain yield, 1992–2007
2006
2004
2002
2000
1998
1996
1994
40,0
1992
50,0
Years
Grain Yield Prediction in the Russian Federation
97
to harvesting time, 20–90% areas of grain yield might be reduced by up to 0.4 t/ha due to unfavorable weather conditions. Longer harvest time is associated with grain yield losses. Expressing the increment in harvested (thrashed) area (DS%, harvesting rates) and the changes in total grain yield (DW, million tons, thrashing dynamics) quantitative relationships of (DS and DW) with precipitation sum (SОс), the number of days with precipitation of 1 mm and more (rдн,), and the air temperature (t) in August (the month of the massive grain harvesting in most regions) was developed. Correlation coefficients for such relationships ranged within 0.50–0.71. This relationship allows for estimating harvest losses due to unfavorable weather. Simultaneously, a relationship between the final grain yield and thrashed area increment (%) for different harvesting was found. Correlation coefficients were 0.75–0.95. A method was developed for the adjustment of final yield (Y1) for adverse weather conditions during harvest time. The following equation approximates this relationship during August–October: Y1 = aX + bZ1 − c where X (%) is the threshed area (expressed as a function of precipitation and temperature).
Conclusion Empirical methods are used for forecasting crop yeild for all of Russia, large economic regions, and medium administrative (oblast) regions based on agrometeorological factors and implemented on personal computers. The dynamical–statistical models are presented in the framework of the Information-Prognostic System (IPS).
Satellite-Based Crop Production Monitoring in Ukraine and Regional Food Security Felix Kogan, Tatiana Adamenko, and Mikola Kulbida
Abstract Every year weather vagaries have caused shortfalls of agricultural production regionally and every 3–4 years these shortfalls occurred globally. Therefore, early assessment of crop losses in response to weather fluctuations is an important task for the estimation of global, regional and countries food supply/demand, donor’s decision to assists the nations in need and to those receiving the assistance. The new satellite-based technology has been recently developed to provide timely and accurate crops’ monitoring and assessments. This technology includes the theory, algorithm, data base and operational implementation of vegetation health (VH) assessments from observations provided by the Advanced Very High Resolution Radiometer (AVHRR) flown on NOAA operational polar-orbiting satellites. Several AVHRR-based VH indices were developed and used to provide weekly cumulative estimation of moisture, thermal and health conditions of vegetation canopy throughout the growing season. The indices were calculated for the entire 1981–2010 period of the AVHRR sensor in space and were compared with regional crop yields in the two dozens of countries. Strong correlation between wheat (both winter and spring), corn, soybeans and sorghum yield and VH indices was found during the critical period of the tested crops. The test results showed that VH indices can be used as proxy for early (2–5 months in advance of harvest) assessment of crop yield with the errors of estimation less than 10%. This paper discusses utility of space observations for early forecasting regional crop yield in Ukraine, with specific emphasis on 2–5 months warning of weather-related losses in agricultural production and their impact on agricultural supply/demand and food security. Keywords Food security • Operational satellites • Vegetation health • Modeling crop losses
F. Kogan () NOAA/NESDIS Center for Satellite Application and Research (STAR), Washington DC, USA e-mail:
[email protected] T. Adamenko and M. Kulbida Ukrainian Hydrometeorological Centre, Kyiv, Ukraine F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_11, © Springer Science+Business Media B.V. 2011
99
100
F. Kogan et al.
Introduction Weather-related crop production assessments, especially losses, have always been a concern for farmers, governments, traders and policy makers for the purpose of balanced food supplies/demands, trade, food aid to the nations in need and food security in general. This is also a very important issue for Ukraine which is a country where agriculture is the major sector of the economy. Every 2–3 years, weather vagaries in Ukraine have caused shortfalls of agricultural production. In some years such as 2003 or 2007, the losses might be staggering. Therefore, assessment of early crop losses in response to weather fluctuations is an important issue for the estimation of global, regional and countries food supply/demand, donor’s decision to assists the nations in need and to those receiving the assistance. Ukraine has the best in the world chernozem soils to produce excellent harvest if the weather is supportive. Unfortunately, Ukraine’s misfortune is to be located in a dry climate zone, where annual shortage of water due to lack of precipitation and excessive thermal resources (leading to elevated evapotranspiration) accounts for 200–400 mm. Droughts affect Ukraine every 2–4 years. The largest agricultural losses occur when drought is preceded by a very cold winter with a lack of snow to cover winter wheat crop. Therefore, monitoring weather impacts on agriculture in Ukraine is a very important component for assessment of agricultural production, food supply/demand, potential trade and food security in general. Weather data are traditionally used in Ukraine for agricultural assessments. One hundred and eighty operational weather stations is a very good source of information about precipitation, temperature, snow and other weather parameters used for the assessments. However, for the 223,000 square miles of Ukrainian territory this number is not sufficient for state and county level analysis since each station covers nearly 1 million acres of land. Therefore, an attempt was made to test the NOAA/ NESDIS Vegetation Health (VH) technology which has spatial coverage of every 4 km2. This paper discusses application of VH indices for monitoring crops and early assessment of yield in Ukraine.
Satellite Data and Method Satellite data where retrieved from the NOAA/NESDIS global archive. The observations from the Advanced Very High Resolution Radiometer (AVHRR) flown on NOAA polar-orbiting satellites created the basis for this data set. The AVHRRbased Global Area Coverage (GAC) data set were produced by sampling and mapping the AVHRR 1-km daily reflectance/emission in the visible (VIS, 0.58–068 mm), near infrared (NIR, 0.72–1.1 mm), and two infrared bands (IR4, 10.3–11.3 and IR5, 11.5–12.5 mm) to a 4-km map. The daily GAC data were aggregated to 7-day composite saving those pixels, which have the highest (during 7-day period) Normalized difference vegetation index (NDVI = (NIR − VIS)/(NIR + VIS)). The 4 km and 7-day VIS and NIR reflectance were pre- and post-launch calibrated and the IR4
Satellite-Based Crop Production Monitoring in Ukraine and Regional Food Security
101
counts were converted to brightness temperature (BT), which was corrected for non-linear behavior of the AVHRR sensor (Kidwell 1997). The 1981–2010 NDVI and BT weekly time series were processed to remove high frequency noise, identify seasonal cycle, calculate climatology and extract medium-to-low frequency variations associated with weather impacts during the growing season. The new method is based on estimation of green canopy stress/no stress from AVHRR-derived indices, characterizing moisture, thermal conditions and total vegetation health (Kogan 1990, 1995, 1997, 2001). Unlike the two spectral channel NDVI-based approach applied for vegetation monitoring, the new numerical method in addition to NDVI, uses also BT from 10.3 to 11.3 mm IR4 channel, which estimate the hotness of the vegetation canopy. In dry years, high temperatures, at the background of insufficient water supply, lead to overheating the canopy, which intensifies negative effects of moisture deficit impact on vegetation. The VH procedure was formalized by equations 1–3, where climatology was represented by the difference between 22-year absolute maximum and minimum both NDVI and BT values for each pixel and week.
VCI = 100 ∗ (NDVI − NDVI min )
TCI = 100 ∗ (BTmax − BTmin )
VHI = a ∗ VCI + b ∗ TCI
(NDVI − NDVImin )
(BTmax − BTmin )
(1) (2) (3)
where NDVI, NDVImax, and NDVImin (BT, BTmax, and BTmin ) are the smoothed weekly NDVI (BT) their multi-year absolute maximum and minimum, respectively; a and b = 1 − a are coefficients quantifying a share of VCI and TCI contribution in the total vegetation health. The VCI (Vegetation Condition Index), TCI (Temperature Condition Index) and VHI (Vegetation Health Index) are indices estimating cumulative moisture, temperature and total vegetation health conditions, respectively on a scale from 0 (extreme stress) to 100 (favorable condition) with 50 corresponding to the average condition.
Results and Discussion In order to understand if VH carries the information about regional crop production in Ukraine the values of the indices were compared with crop yield. Figure 1 shows 2002–2004 VH for mid-May and also Ukraine’s average yield of cereals (winter and spring wheat, barley, oats and corn). The 2002 and 2004 images taken by NOAA-16 polar-orbiting satellite identified favorable vegetation condition (green) resulted in high yields, 2.75 and 2.84 t/ha, respectively. Opposite to these 2 years, the 2003 indicates a very severe vegetation stress (red color) in the principal agricultural area due to severe drought, which caused 35% losses in cereal yield (1.84 t/ha) versus the years before and after.
51
F. Kogan et al.
51
49
102
49
Cereal Yield 27.5
45
45
47
47
2002 34
38
51
30
51
49
26
49 45
45
47
47
2003 26
34
38
28.4
45
45
47
47
49
49
51
51
30
18.4
2004 26 6
12
30 24
36
34 48
60
38 72
84
Fig. 1 Mid-May VH and Ukraine’s average cereal yield, 2002–2004 (Color figure is in Appendix 2)
Statistical Modeling of Regional Yield Following these encouraging results, correlation and regression analysis were employed to investigate numerically regional yield dependence on VH. Average winter wheat yield for Odessa oblast during 1980–2006, shown in Fig 2, were collected from Ukrainian Statistical Administration. According to Obuhov (1949), the long-term yield time series are normally separated into two components: technological (TEC), driven by agricultural technology (fertilizers, genetics, plant protection, irrigation etc.) and weather (WETH), controlled by variations in meteorological parameters from year to year. The TEC component is approximated by trend (straight line in Fig. 2) and WETH by deviation from the TEC trend.
Satellite-Based Crop Production Monitoring in Ukraine and Regional Food Security
103
T/Ha 4
3
2
1
0 1980
Winter Wheat Yield Odessa Obl. UKRAINE
1985
1990
1995
2000
2005
Fig. 2 Winter wheat yield and trend, 1980–2006, Odessa oblast, Ukraine
The equations 4 and 5 represent the approximation procedure, respectively.
ˆ = a +a Y Υ t 0 1 t
(4)
ˆ dYt = Yt Υ t
(5)
where Yt is measured yield, Ŷt is the trend yield – the deterministic component regulated by the agricultural technology, dY is a random component regulated by weather fluctuations and t is year number. The random component can be approximated by either a difference or ratio of actual and trend (estimated from equation 5) yields. In this paper the ration was used (Obuhov 1949; Salazar et al. 2007). Figure 3 shows the dynamics of correlation coefficients for the end-of-season winter wheat dY with every week from the first week in January to the last week in June. As seen, (a) during January and February the correlation is close to zero; (b) thereafter, it is increasing; (c) reaches maximum (0.5–0.6) during the critical period (2–3 weeks before and after winter wheat heading (April–May); and (d) drops thereafter, almost to zero at the grain filling and beyond. Based on the results presented at Fig. 3, VH variables for several weeks of critical period were selected to build regression model. The equation was written as
dY = 0.286 − 0.057VH5 + 0.067VH6 − 0.041VH18 + 0.044VH19,
(6)
where dY is winter wheat yield deviation from trend and VH is vegetation health index for the weeks indicated by the attached number. This equation indicates that 4 weeks were selected as independent variables: 2 weeks in February reflecting moisture (VCI) contribution and 2 weeks in May reflecting thermal contribution. This equation was validated independently using “Jack-Knife” or “one-in-one-out” techniques. Following this technique (a) each year
104
F. Kogan et al. VH 0.5
VCI 0.0
− 0.3
Jan
TCI 10
Mar
20
May
Jun
30
Week
Fig. 3 Correlation dynamics of winter wheat dY versus VH indices (VCI, TCI and VHI)
Fig. 4 VH-simulated and actual winter wheat yield, Odessa oblast, Ukraine
yield data are removed from the time series one by one, (b) new model is developed without removed year and (c) the model is applied to the removed year. This procedure is repeated until all years were independently tested. Comparison of measured and independently simulated winter wheat yield are shown in Fig. 4. The two time series are well correlated (R = 0.86) with the error of estimation 0.2 t/ha.
References Kidwell KB (1997) Global Vegetation Index user’s guide (Camp Springs MD: US Department of Commerce, NOAA, National Environmental Satellite Data and Information Service, National Climatic Data Center, Satellite Data Services Division) Kogan FN (2001) Operational space technology for global vegetation assessments. Bull Am Meteor Soc 82(9):1949–1964 Kogan FN (1997) Global drought watch from space. Bull Am Meteor Soc 78:621–636 Kogan FN (1995) Droughts of the late 1980s in the United States as derived from NOAA polar orbiting satellite data. Bull Am Meteorol Soc 76:655–668 Kogan FN (1990) Remote sensing of weather impacts on vegetation in non-homogeneous areas. Int J Remote Sens 11:1405–1419 Obuhov VM (1949) Urozhainost i Meteorologicheskie Factoru (in Russian): Yield and Meteorological Factors. Gosplanisdat, Moscow, 314 pp Salazar L, Kogan F, Roytman L (2007) Use of remote sensing data for estimation of winter wheat yield in the United States. Int J Remote Sens Vol 28(Nos 17–18):3795–3811
New Regression Models for Prediction of Grain Yield Anomalies from Satellite-Based Vegetation Health Indices Gennady Menzhulin, Natalya Shamshurina, Artyom Pavlovsky, and Felix Kogan
Abstract In the late 1970s, the first operational weather satellite system had been launched, which showed utility for monitoring land greenness, vigor and vegetation productivity. Currently, 30-year satellite data from the Advanced Very High Resolution Radiometer (AVHRR) are available for monitoring land surface, atmosphere near the ground, natural disasters, and socioeconomic activities. Statistical modeling of agricultural crop yield and production was one of the applications. This paper discusses the topic, how design the new regression models of yield anomaly based on multivariate algorithms and selection of best-fit ensemble of predictors. Keywords Crop yields anomaly • Vegetation health indices • Precipitation • Temperature • Models
Introduction Assessment and prediction of crops production have always been one of the most important agrometeorological tasks. In crop model development, surface meteorological parameters, primarily precipitation and temperatures, were traditionally used as the predictors. Unfortunately, the surface weather network is not dense enough to represent adequately changing weather conditions. Therefore, scientists put a lot of efforts in using satellite data for this purpose. The first operational weather satellites of NOAA series have been launched in the late 1970s with the AVHRR instrument on board for monitoring weather. G. Menzhulin (*), N. Shamshurina, and A. Pavlovsky Research Center for Interdisciplinary Environmental Cooperation, Russian Academy of Sciences, St. Petersburg, Russia e-mail:
[email protected] F. Kogan NOAA, NESDIS, Washington D.C., USA F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_12, © Springer Science+Business Media B.V. 2011
105
106
G. Menzhulin et al.
During the 30-year period in service, AVHRR data have been successfully used for monitoring land surface, atmosphere near the ground, natural disasters, and socioeconomic activities. Such a long period is sufficient for statistical modeling of crop yield and production. Many researchers, using simplified hypotheses applied satellite data for designing linear regression models to forecast regionalscale crop yield (Dabrowska-Zielinska et al. 2002; Liu and Kogan 2002; Manjunath et al. 2002; Kogan et al 2003; Leon et al 2003; Savin and Negre 2003; Domenikiotis et al 2006). In this paper the authors made an attempt to use multivariate linear regressions is devoted to estimation of efficiency of satellite remote sensing data using when designing of the for designing crop models, without prior limitations.
Satellite and In situ Data and Method The modern satellite remote sensing is using vegetation indices for characterization of land surface. The Normalized Difference Vegetation Index (NDVI) is the most known and widely used (Deering 1978). The NDVI is a ratio of solar radiation fluxes (F), reflected from the surface in the visible (VIS) and near infrared (NIR) solar band range (NDVI = (FNIR − FVIS)/(FVIS+FNIR)). Vegetation indices have some advantages before in situ observations, such as weather. They have very dense special and temporal coverage; have a “memory” or showing cumulative condition (example NDVI). However, on a weak side, satellite indices are proxy, do not have a deep biophysical explanation and have some errors related to satellite/sensor problems and atmosphere interference. In this research, satellite data were presented by the Vegetation Health (VH) indices which in numerous researches showed considerable improvement before previous generation of indices in characterization of weather- and climate-related conditions (Dabrowska-Zielinska et al. 2002; Liu and Kogan 2002; Manjunath et al. 2002; Kogan et al 2003; Leon et al 2003; Savin and Negre 2003; Domenikiotis et al 2006). The two VH indices were used: Vegetation Condition Index (VCI) developed from NDVI and Temperature Condition Index (TCI) developed from infrared-based brightness temperature (BT). The VCI and TCI were normalized for each pixel and time interval. The absolute multi-year maximum and minimum values were used as the normalization criteria. As the result, the following expressions were used: VCI = (NDVI − NDVI min)/(NDVI max − NDVImin) and TCI= (BTmax − BT)/(BTmax − BTmin). Both indices change from zero representing the extreme vegetation stress to 100 characterizing no-stress conditions. The VH indices data were received from the Global Vegetation Index dataset (Kogan 2002). For selection of crop yield data, the main criteria was availability of long time series, quality and high territorial resolution. These data were collected the historical archives of United States Department of Agriculture’s National Agricultural Statistical Services (http://www.nass.usda.gov). In this archive the
New Regression Models for Prediction of Grain Yield Anomalies
107
information for main crops productions and areas for each world county and each US state is presented. The duration of the continuous series for basic crops for the majority of US counties is around 60–70 years (the last 40-year series have the most complete records). At designing of weather-related wheat yield anomaly statistical model it is important first, to remove from yield series the long-term deterministic component related to the impact of agricultural technology (fertilizers, tillage, weeds, pests and diseases control and others). Its figures calculated by the formula h i = (yi − Yi)/Yi , where yi and Yi are the real and trend-line values, respectively in the year i.
Statistical Algorithms We designed several statistical algorithms for yield anomaly versus VH indices, which take into consideration many predictors to chose from (52 of vci and 52 weeks of tci) and colinearity between the predictors. The first one named phenomenological (P-method) requires sequential inclusions all vci and tci predictors (vci and tci) and leave only those having the highest correlation with predictand (hi ). Besides, the predictors were selected with minimal colinearity (correlation coefficient <0.75). In the final equation statistically non-significant predictors were excluded. The second algorithm is “the step-by-step inclusion” (S-method). The first predictor in a model is that having the highest correlation with predictand, after inclusion the second predictor, the model has to improve statistically based on Student t test. This algorithm requires two runs: forward and backward. The third algorithm (E) examines models with all two to six predictors’ combination (Menzhulin et al. 2008).
Study Area The study area was USA’s Kansas state and winter wheat was selected as a crop, which is cultivated in all 104 counties. As the growing season of winter wheat falls into 2 successive years starting from September of year before harvest and ending in June–July of the harvest year, satellite data were collected from the third week of the previous year July through the second week of the next year July. Based on the values of winter wheat production in Kansas, we divided its agricultural territory in eastern and western parts. In the western part (west of 97°W), the counties are characterized larger sowing areas and larger production compared to eastern counties. Regression models for winter wheat yield anomaly (YA) against vegetation health (VH) indices were developed for all Kansas counties, however, the regression models are presented for two counties Woodson in eastern Kansas and Thomas in western Kansas.
108
G. Menzhulin et al.
Results and Discussion The regression analysis presented in Table 1, show that the least accurate models were obtained when P method for predictors’ selection was used. For Thomas county the coefficient of determination was 0.79 but for Woodson only 0.41. The results with the S and E (except for E-2 and E-3) algorithms are better (D = 0.94 − 0.93). The E algorithm for six-factor models (except for one) in Thomas county had the best fit at the 95% significance level. These models were based on both thermal (tci) and moisture (vci) conditions. The vci is important as a predictor in the early September, end of December–early January and during the end of spring through early summer. The tci, predictors were selected for the last week of June, which corresponds to grain filling phase of winter wheat. This fact shows convincingly that the algorithms of predictor selection basing on the simplified a priori methods are hardly capable to foresee such selection of absolute direct exhausting techniques. The E algorithm increases considerably coefficient of determination: in Thomas county by up to 0.97 with 99% of significance level for all selected predictors. In Woodson county, the winter wheat regression models are less accurate and the selected predictors are quite different from the other county. According to Table 1, the models with six-predictors (all with 99% significance level) have 4 vci weeks during the end of August–end of October and 2 tci weeks in spring of the next year. When an E algorithm is used the determination coefficient reaches 0.93. The Woodson’s S linear regression model also includes six predictors and all of them statistically significant according to Student t test. Five of six predictors are for tci weeks (three in fall and three in spring) and one for vci (the last week in May). With the purpose to show it additionally we designed the for each Kansas county the regression wheat productivity anomaly models using the data on dimensionless monthly minimal and maximal surface. Figure 1 shows CD’s changes following an increase in the number of predictors’ from two to six. As seen, the CD is increasing in both counties but Woodson experience significant increase from CD = 0.45 to 0.93, while Thomas – only slight increase between 0.78 and 0.95. Figure 2 shows attenuation of the adjusted CD for the first 100 best sixfactor (E-6) models for Woodson county. It is clearly seen that the model improvement (based on values of the adjusted CD) reduced considerably after the first 15 iterations. Finally, we compare CD-based performance of the satellite- and weather-based models. The weather-based models were developed using similar to satellite data regression algorithm. Mean monthly air temperature and total monthly precipitation averaged for each Kansas state county during the past 30 years were collected and processed. Since the weather data were monthly, the number of tested predictors might be up to eight. These models’ performance was tested using adjusted CD and Student t-criterion. The searching process continues wile the best multifactor model was discovered. These results are presented in Table 2. Analysis of the table indicates that satellite-based models have significant higher adjusted CD compared to the weather models. In some counties (Elk, Doniphan, Crawford etc.) the difference might be 1.5–2 times.
1
v45 v45 v18 v18 v13 v33 v13
Model
P S E-2 E-3 E-4 E-5 E-6
Woodson
−1.5 −7.1 3.8 4.1 3.7 4.8 8.3
v36 t07 v22 v22 v15 t27 v14
2
2.3 −5.7 −3.5 −3.8 −4.9 7.2 −8.4
t49 t12 v44 v16 t30 v17
3 1.9 6.1 −2.2 5.7 −6.9 10.9
t36 v21 t34 v22
4 −3.4 −6.0 7.8 −10.9
v04 t36 v42
5 2.7 −7.1 −7.8
t05 t43
6 2.6 −5.4
Table 1 Statistical characteristics of winter wheat yield anomalies models for Thomas and Woodson counties, Kansas, USA Thomas Model 1 2 3 4 5 6 P t45 5.1 v42 −1.3 v50 1.3 t31 1.7 S t45 7.4 t28 4.5 t22 −4.0 t09 2.9 v48 1.9 t40 −1.5 E-2 t36 3.3 t47 6.7 E-3 t03 −3.9 t05 4.4 t46 9.6 E-4 t08 3.5 t22 −3.5 t27 3.9 t46 8.7 E-5 t02 −5.7 t03 5.4 t36 5.4 t38 −5.6 t44 10.8 E-6 v16 4.3 t02 −8.7 t03 8.3 t36 9.9 t38 −10.2 t43 14.0
0.41 0.85 0.46 0.59 0.74 0.86 0.93
D
D 0.79 0.92 0.78 0.87 0.89 0.93 0.97
0.34 0.78 0.45 0.59 0.75 0.86 0.93
Da
Da 0.75 0.88 0.76 0.85 0.86 0.91 0.95
New Regression Models for Prediction of Grain Yield Anomalies 109
G. Menzhulin et al. 1
Coefficient of determination
Coefficient of determination
110
0.8 0.6 0.4 0.2 0
2 3 4 5 Amount of predictors
1 0.8 0.6 0.4 0.2
6
0 2
3 4 5 Amount of predictors
6
Fig. 1 Changes in the adjusted coefficient of determination of winter wheat yield anomaly using E-6 models: left Thomas and right Woodson counties, Kansas state, USA
Adjusted coefficient of determination
0.94
0.92
0.90
0.88
0.86
0.84
0.82
0.80 1
10
20
30
40
50
60
70
80
90
100
Number of the 100 best models Fig. 2 Reduction in the adjusted coefficient of determination of the 100 best winter wheat anomaly six-factor models developed with E-6 algorithm, Woodson county, Kansas, USA
Conclusion This paper shows, that even medium resolution (16 × 16 km) Vegetation Health (VH) data provide sufficient information for designing very accurate and high reliable regression models to predict and monitor crop yield. These models
Allen Anderson Atchison Barber Barton Bourbon Brown Butler Chase Chautauqua Cherokee Cheyenne Clark Clay Cloud Coffey Comanche Cowley Crawford Decatur Dickinson Doniphan Douglas Edwards Elk Ellis
0.79 0.72 0.86 0.88 0.92 0.85 0.87 0.82 0.39 0.88 0.84 0.92 0.84 0.85 0.91 0.79 0.92 0.57 0.83 0.95 0.81 0.86 0.82 0.83 0.92 0.94
0.54 0.6 0.52 0.72 0.78 0.62 0.51 0.56 0.43 0.43 0.64 0.84 0.77 0.67 0.75 0.68 0.82 0.48 0.44 0.69 0.63 0.58 0.70 0.71 0.48 0.72
Ellsworth Finney Ford Franklin Geary Gove Graham Grant Gray Greeley Greenwood Hamilton Harper Harvey Haskell Hodgeman Jackson Jefferson Jewell Johnson Kearny Kingman Kiowa Labette Lane Leavenworth
0.86 0.92 0.94 0.71 0.79 0.89 0.87 0.87 0.92 0.96 0.83 0.95 0.88 0.79 0.89 0.89 0.74 0.87 0.94 0.78 0.95 0.87 0.98 0.76 0.91 0.89
0.65 0.84 0.82 0.46 0.57 0.75 0.75 0.69 0.8 0.87 0.59 0.84 0.67 0.61 0.82 0.82 0.47 0.59 0.76 0.61 0.83 0.74 0.80 0.47 0.78 0.46
Lincoln Linn Logan Lyon Marion Marshall McPherson Meade Miami Mitchell Montgomery Morris Morton Nemaha Neosho Ness Norton Osage Osborne Ottawa Pawnee Phillips Pottawatomie Pratt Rawlins Reno
0.86 0.81 0.91 0.89 0.79 0.88 0.82 0.94 0.79 0.91 0.84 0.81 0.93 0.79 0.81 0.88 0.93 0.84 0.84 0.91 0.92 0.84 0.54 0.89 0.81 0.88
0.66 0.48 0.75 0.68 0.54 0.65 0.70 0.85 0.63 0.79 0.73 0.59 0.85 0.56 0.47 0.70 0.65 0.70 0.60 0.72 0.87 0.67 0.32 0.73 0.69 0.75
Republic Rice Riley Rooks Rush Russell Saline Scott Sedgwick Seward Shawnee Sheridan Sherman Smith Stafford Stanton Stevens Sumner Thomas Trego Wabaunsee Wallace Washington Wichita Wilson Woodson
0.95 0.92 0.81 0.83 0.91 0.85 0.79 0.95 0.78 0.88 0.93 0.85 0.88 0.91 0.92 0.92 0.95 0.68 0.93 0.89 0.83 0.91 0.94 0.92 0.89 0.91
0.78 0.74 0.51 0.56 0.77 0.68 0.60 0.84 0.53 0.73 0.65 0.67 0.82 0.86 0.75 0.82 0.79 0.58 0.85 0.75 0.57 0.84 0.75 0.80 0.73 0.66
Table 2 Adjusted coefficients of determinations for six-factor (E-6 algorithm) regression models of winter wheat yield anomalies versus satellite (D1) and weather (D2) data for 104 Kansas state counties USA Counties D1 D2 Counties D1 D2 Counties D1 D2 Counties D1 D2
New Regression Models for Prediction of Grain Yield Anomalies 111
112
G. Menzhulin et al.
s urpass the accuracy of the statistical crop-weather models. We believe that the higher resolution of VH data will allows to develop more accurate models with 3–5 months of lead time of prediction.
References Dabrowska-Zielinska K, Kogan F, Ciolkosz K, Gruszczynska M, Kowalik W (2002) Modeling of crop conditions and yield in Poland using AVHRR-based indices. Int J Remote Sens 23:1109–1123 Deering D (1978) Rangeland reflectance characteristics measured by aircraft and spacecraft sensors. Ph.D. Dissertation, Texas A&M University, 198 Domenikiotis C, Spiliotopoulos M, Tsiros V, Dalezios N (2006) Early cotton yield assessment by the use of the NOAA/AVHRR derived vegetation condition index (VCI) in Greece. FAO Crop Production 133–145 Kogan F, Gitelson A, Zakarin A, Spivak L, Lebed V (2003) AVHRR-based spectral vegetation indices for quantitative assessment of vegetation state and productivity: calibration and validation. Photogram Remote Sens 69:899–906 Kogan F (2002) World droughts in the new millennium from AVHRR-based vegetation health indices. EOS 32:557–564 Leon C, Shaw D, Cox M, Abshire M, Ward B, Wardlaw M (2003) Utility of remote sensing in crop production and soil characteristics. Precision Agric 4:359–384 Liu W, Kogan F (2002) Monitoring Brazilian soybean production using NOAA/AVHRR based vegetation condition indices. Int J Remote Sens 23:1161–1179 Menzhulin G, Kogan F, Peterson G, Shamshurina N (2008) Advanced techniques of crop productivity regression modeling based on vegetation indices satellite information. Bulletin of St. Petersburg State University 7 (Geology and Geography), 4:86–98 (in Russian) Manjunath K, Patdar M, Purohit N (2002) Large area operational wheat yield model development and validation based on spectral and meteorological data. Int J Remote Sens 23:3023–3038 Savin I, Negre T (2003) On the new approach to using NDVI Index for crop state monitoring. Earth Space Res 4:91–96 (in Russian)
Phytosanitary Situation of Agrocenosis in Ukraine and New Technologies for Monitoring Harmful Organisms Vladimir Chayka, Tatiana Neverovska, Nelia Prokopiuk, and Olga Baklanova
Abstract In light of present and future climates change, it is possible to expect an increase in the general diversity of entomofauna that might distort the ecological stability of agrolandscapes. This paper shows some decrease in the level of ecological stability of agroecosystems which generally deteriorates phytosanitary situation. Keywords Agrocenosis • Phytosanitary situation • Entomofauna • Global warming • Geographic informative systems (GIS)
Introduction Due to global warming, the current climate of Ukraine is experiencing certain changes, such as an increase in the annual average temperature and a sum of effective temperatures. This warming led to changes in the duration of vegetation seasons, which is affecting the development and productivity of agricultural crops, insect pests, and diseases (Chayka 2004; Kingsolver 1989). Recent decades have shown outbreaks of locusts, an increase in the population density of the cutworm Agrotis segetum and Margarita sticticalis, the expansion of desert conditions in the Southern region and changes in crop productivity (Chayka 2004). Regarding the future dynamics of the phytosanitary situation, it is likely to expect an increase in insect and pests, their migration activity, and harmfulness (Stubbles 2001). The aim of this paper is to forecast probable change in the phytosanitary situation of agrocenosises in Ukraine owing to global warming.
V. Chayka (*), T. Neverovska, N. Prokopiuk and O. Baklanova Institute for Plant Protection, Kyiv, Ukraine e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_13, © Springer Science+Business Media B.V. 2011
113
114
V. Chayka et al.
Data and Methods Two data types were used in this research: the size of the main insects and pest population affecting agricultural crops, taken from the Central State Inspection for Plant Protection in Ukraine, and meteorological data taken from the Hydrometheocentre of Ukraine. The insect/pest data were compared analytically with meteorological data.
Climate Analysis The most intensive climate warming in Ukraine occurred in late 1988, destabilizing phytosanitary situation of agrocenoses. As the result, the number of insect pests increased up to two times compared with the preceding period (Chayka 2003; Baklanova et al. 2003). This occurred due to (1) some changes in land tenure in Ukraine: beginning in 1990, up to 8.5 million hectares of cultivated lands were withdrawn from agricultural practice, resulting in the formation of a wide ecological niche for a numbers of polyphages and some species of the insects; (2) three to five times reduction in the application of plant protection. During the 20-year period, annual temperature increased by 0.4°C; in addition, in 1995–1996, a transition from the 22nd to the 23rd cycle of the Sun activity occurred. As known, the period with a minimum of the Sun activity is accompanied with the outbreaks of phytophages.
Results and Discussion Until 2003, the number of main pests (expressed in the form of indices) feeding on agricultural crops increased from year to year. Two outbreaks of Acridodae spp. (1996–1997 and 2003) were recorded in the southern Ukraine. A partial stabilization of the average index-population occurred for the sunn bug, the beetroot weevil, and others after the extremely cold winter of 2003–2004. Meanwhile, the number of larvae of such pests as click-beetles (Elateridae), darking beetles (Tenebrionidae), Zabrus tenebrioides G and other insects-geobionts species continuee to increase in spite of plant protection measures. This is shown in Fig.1 for the Elateridae and Tenebrionidae larvae populations in Ukraine. Linear trend analysis shown in Fig 1, indicates a three to five times increase in the insect population as compared to 1989. As was mentioned, this occurred with temperature increase and a reduction in plant protection measures and crop area (Chayka 2007). In the last 10 years, temperature warming has impacted the entomocomplex of the winter crops in the Ukrainian forest-steppe zone. There is a tendency towards an increase in the numbers of such flies as Opomiza florum F and Phorbia securis Tiens., Flover thrips Frankliniella tritici, cereal Scarabaeidae
Phytosanitary Situation of Agrocenosis in Ukraine and New Technologies
115
Fig. 1 Dynamics of the Elateridae and Tenebrionidae larvae populations in Ukraine (from the Central State Inspection of Plant Protection in Ukraine)
(Anisoplia spp.: during the planting-tillering period their harmfulness increased three times (Kozak et al. 2004). Contemporary estimations of a level of insects diversity in Ukraine is estimated at 25,000 species by 2000. The share of insects among species of biota increased from 53% to 75%, and their total weight exceeds biomass of other animals (Lesovoy 2007). The Geographic Information Systems (GIS) technology is currently helping to evaluate past (Fig. 2) and plan protection measures (Chayka et al. 2009).
Conclusions Climate worming in Ukraine will lead to: (1) Some changes in the zones of harmfulness of insects-phytophages. The zones of ecological optimum of diverse species will be spreading to the north, leading to reconstruction of the species structure of entomocomplexes and increasing potential yield losses. (2) An increase in the number of generations of the polyvoltine species of insects such as aphids (Aphididae), cutworms (Noctuidae), leaf-rollingmothes (Fortricidae), the European Corn Borer and others, with an increase of their harmfulness. (3) Soildwelling insect pests will adapt to new agroclimatic conditions; during the last 20 years, the area of their colonization increased. (4) The harmfulness of field crops’ phytophages will grow, especially in drought years. (5) Polyphagous insect pests such as Margarita sticticalis, Locusts spp. and others will increase their harmfulness and spread to the north of Ukraine.
116
V. Chayka et al.
Fig. 2 Spread and numbers of the turnip moth (Agrotis segetum Schiff.) in Ukraine (2008)
References Stubbles J (2001) Global warming – fact or fiction? Pt. II. Iron & Steelmaker Vol. 28,12: 98–99 Chayka V (2004) Theoretical bases of the entomological prognosis. Quarantine and plant protection. Kiev 50:3–20 Chayka V (2003) Conditions of the phytosanitary situation. Plant protection. Kiev 4:1–3 Baklanova O, Kravchenko V, Chayka V (2003) Population dynamics of the main polyphagous insect pests in Ukraine. Plant protection. Kiev 10:8–10 Kingsolver J (1989) Weather and the population dynamics of insect: integrating physiological and population ecology. Physiol Zool 62(2):314–334 Chayka V (2007) Long-term dynamics of numbers of Elateridae in Ukraine. Quarantine and plant protection. Kiev 6:7–9 Kozak G, Syadrista O, Chayka V (2004) Harmfulness of phytophages on a winter wheat in forest-steppe of Ukraine in the conditions of global warming of climate. Quarantine and plant protection. Interdepartmental Temat Sci Digest 50:21–28 Lesovoy M, Chayka V (2007) Entomological diversity and its ecological and economical significance. Agroecol J 4:18–24 Chayka V (2009) Analisis of phytosanitary situation of agrocenoses in Ukraine and place of new technologies in pest organisms’ monitoring. Inform Bull Kishinev 40:349–350
Part III
Climate Change, Environment And Socioeconomics
30-Year Land Surface Trend from AVHRR-Based Global Vegetation Health Data Felix Kogan
Abstractâ•… The past 30 years of environmental observations showed considerable global temperature increase and global changes in snow and ice cover, sea level, biological systems timing (plants, birds etc.) and others. It was also shown with 20-year satellite records that Earth vegetation has an early greening, especially in the northern latitudes. Currently, 10 more years were added to the satellite records requiring re-evaluation of vegetation trends. NOAA/NESDIS has recently updated long-term satellite records produced from AVHRR data. These innovations permitted to develop the new 30-year Global Vegetation Health (GVH) dataset and products. The GVH were processed comprehensively to remove noise even those which had not been removed before. This paper investigates the 30-year no-noise NDVI time series for the purpose of trend detection. Data showed that the 30-year trend both global and latitudinal is very negligible. Keywordsâ•… Global Vegetation Health dataset • Green-up trend • NDVI change
Introduction According to IPCC (2007) report, the average global temperature over the past 100 years (from 1906) increased 0.74°C. The past 20-year environmental observations also showed global changes in snow and ice areas, sea level, biological systems (plants, birds etc.) and others. Regarding Earth vegetation, it showed an early greening, especially in the northern latitudes (Lucht et€ al. 2002; Myneni et€ al. 1997; Nemani et€al. 2003; Zhou et al. 2001). These results were obtained from the analysis of the nearly 2 decades of the Normalized Difference Vegetation Index (NDVI) calculated from the Advanced Very High Resolution Radiometer (AVHRR) measurements on board NOAA polar-orbiting satellites. Currently, F. Kogan (*) NOAA/NESDIS Center for Satellite Application and Research (STAR), Washington DC, USA e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_14, © Springer Science+Business Media B.V. 2011
119
120
F. Kogan
more than 10 years have been added to the AVHRR-based NDVI time series, the processing improved considerably through the noise correction, the data were studied comprehensively and the most important, were validated against in situ biological observations. Moreover, following biophysical and ecosystem laws the new theory of vegetation health derivation was introduced permitted the development of AVHRR-based products and new applications in agriculture, forestry, human health, climate forcing and others. All of these innovations permitted to develop the new 30-year Global Vegetation Health (GVH) dataset and products (Kogan 2001; Kogan et€al. 2010). The purpose of this paper is to investigate trend in the 30-year NDVI time series.
Global Vegetation Health Data Set The GVH data records were developed from the measurements made by the AVHRR instrument flying on board NOAA afternoon (NOAA-7, 9, 11, 14, 16, 18 and 19) polar-orbiting operational satellites (Kidwell 1997). The AVHRR instrument scans the earth surface at 1.1-km resolution in four wavelengths of the solar spectrum: visible (VIS, 0.58–0.68 mm), near infrared (NIR, 0.725–1.1 mm) and two infrared (IR, 10.3–11.3 mm and 11.5–12.5 mm). These measurements were aggregated to the 4-km global resolution and archived as the NOAA’s Global Area Coverage (GAC) dataset (Kidwell 1997; Cracknell 1997). The archive of daily GAC data was used to develop AVHRR-based 4-km resolution CLAVR-x processing system and dataset which were well navigated, geolocated and mapped to Plate Carre projection with nominal grid cell length of 4-km (Jacobowitz et€al. 2003; Heidinger and Pavolonis 2009). The GVH system development started form data extraction from the CLAVR-x processing system reflectance/emission (in digital counts) in the visible (VIS, 0.58–0.68 mm), near infrared (NIR, 0.725–1.1 mm) and two infrared (IR4, 10.3–11.3 mm, channel 4 (Ch4) and IR5, 11.5–12.5 mm) wavelength of solar spectrum. Following the CLAVR-x system, the GVH global dataset has 4-km special resolution and spans from 75.024° (north edge) to −55.152° (south edge) in the latitudinal and from −180° (west) to 180° (east) in longitude directions. Daily data were composited over 7-day period saving that day which has the highest NDVI. Using pre-launch calibration, the 7-day composite VIS and NIR counts were converted into reflectance and the Normalized Difference Vegetation Index was calculated as. (NDVI = (NIR − VIS)/(VIS + NIR) (Kidwell 1997; Cracknell 1997). In addition to pre-launch, post-launch calibration was applied following Rao and Chen (1999). The IR counts were converted to brightness temperature (BT) and corrected for non-linear behavior of the instrument (Cracknell 1997). Since noise in AVHRR data creates fundamental constraints to the remote sensing of the Earth, NDVI and BT were massaged considerably to remove all long/ medium-term and high frequency noise (Kogan et€ al. 2010; Kogan 1997; Rao and Chen 1999; Cracknell 1997; Kidwell 1997). The following noise removal
30-Year Land Surface Trend from AVHRR-Based Global Vegetation Health Data
121
� procedures were applied: complete removal of high frequency noise (clouds, �aerosol, water vapor etc.), noise due to satellite orbital drift, sensor degradation, satellite change, difference in response functions between AVHRR-3 and previous versions of the instruments, inconsistencies between the satellites in equator crossing time, troposphere aerosols from volcanoes. The algorithms for noise removal eliminated completely the high frequency outliers, including random long-term noise, approximated accurately the annual NDVI and BT cycles, and singled out medium-to-low frequency weather-related fluctuations (Kogan et€al. 2010). This processing was sufficient for 30-year NDVI and BT trend analysis to investigate vegetation green up during the period of sharp global temperature increase of the last 20 years.
Trend Analysis During 1982–2007 Figure€ 1 presents the average no-noise NDVI for the central part of the globe (between 40°N–40°S). This part was selected because it includes major agricultural areas (USA, Europe, China, India, former Soviet Union, Brazil, Argentina and Australia) and major ecosystems (forest-steppe, grassland, tropical forest, desert). As seen, the trend slope is negligible since NDVI is 0.2180 at the beginning and 0.2184 at the end, the changes are 1.8% for the entire 27-year period. We also investigated trend of NDVI averaged for 1° latitude of global circle every 10° latitude between 70°N and 40°S The results for the annual time series and separately average for winter and summer are shown in Table€1. For the annual time series trend is negligible (below 0.001) between 30°N and 30°S. However, north of 30°N, the trend is elevated, although by the statistical standards, the trend is very small. High slope value for the 40°S was not taken into consideration since the land sample for that are is very small represented by the southern tip of South America. At the general background the seasonal NDVI trend are negligible: interesting that summer NDVI trend in general is smaller than winter in Northern Hemisphere and slightly higher in Southern Hemisphere. The 27-year NDVI changes (right part of Table€1) showed similar feature to the slope analysis. For the northern latitudes, annual NDVI increased 10–12% by 2007
Fig.€1â•… Average NDVI and trend for a box (Lat 40°N–40°S, Long 180°W–180°E)
122
F. Kogan
Table€1â•… 1982–2007 linear trend slope (*10−3) and NDVI change (%) during 26-year period Slope NDVI change (%) from beginning Latitude circles Year Winter Summer Year Winter Summer 70N–69°N 6.34 −4.57 13.47 11.3929 −7.65 11.70 60N–59°N 13.54 10.43 11.98 12.0163 16.90 4.91 50N–49°N 14.13 8.12 6.95 11.2558 15.04 2.76 40N–39°N 11.41 12.91 7.04 10.3706 17.62 4.26 30N–29°N 6.52 12.05 3.16 7.4496 14.24 3.10 20N–19°N −1.14 −0.25 −3.21 −1.1100 −0.20 −3.41 10N–09°N −0.80 5.18 −11.24 −0.4559 3.31 −5.10 000–01°S −4.92 1.55 −6.79 −2.2707 0.69 −2.87 10S–11°S −6.36 −3.66 −6.61 −2.7184 −1.50 −2.44 20S–21°S 0.16 1.96 8.05 0.0968 1.13 4.73 30S–31°S −6.39 −3.47 −5.80 −4.0864 −2.10 −3.42 40S–41°S 16.61 3.61 35.27 12.1710 1.96 34.42
relative to the beginning level in 1982. However, south of 30°N, changes in NDVI were negligible, between −4 and 7%. The largest NDVI changes occurred in winter (14–17%) in northern latitudes (north of 29°N). South of 29°N no statistically significant trends were observed. It is also important to emphasize that in summer, all latitude show very small (−3% to 5%) NDVI changes by 2007. Only near 70°N latitude shows 12% NDVI growth at the end of 27-year period. We also ignored 34% NDVI increase for latitude 40°S due to the reasons indicated above.
Conclusions A thoroughly processed 30-year GVH dataset analyzed for long-term trend detection indicated that there is no statistically meaningful NDVI trend in the major agricultural regions (average for latitudes 40°N–40°S). Similar results were obtained for the individual latitudes in the annual cycle and for winter and summer. Interestingly, if winter shows 10–12% increase in NDVI at the end of 2007 in the northern latitudes, southern latitudes and summer do not show such increase. These results disagree with previous studies done on a shorter NDVI time series indicating considerable green-up trend. Another paper presented in this issue discussed comparison of the GVH dataset time series with the other long-term data records (see Kogan et€al. 2010). The two of the possible causes might be longer period of observation for the new data set and differences in the processing algorithm.
References Cracknell AP (1997) The advanced very high resolution radiometer. Taylor & Francis, USA, 534 p. GVH (2010) http://www.star.nesdis.noaa.gov/smcd/emb/vci/VH/vh_browse.php
30-Year Land Surface Trend from AVHRR-Based Global Vegetation Health Data
123
Heidinger, Andrew K Pavolonis, Michael J (2009) Global daytime distribution of overlapping cirrus cloud from NOAA’s advanced very high resolution radiometer. Journal of Climate, 18(22):4772–4784 IPCC (2007) http://www.ipcc.ch/publications_and_data/ar4/wg2/en/contents.html Jacobowitz H, Stow LL, Ohring G, Heidinger A, Knapp K, Nalli N (2003) The advanced very high resolution radiometer PATHFINDER Atmosphere (PATMOS) climate data set: A Resource for Climate Research. Bull. American Meteorological Society, June, 785–793. Kidwell KB (ed) (1997) Global vegetation index user’s guide. National Oceanic and Atmospheric Administration, National Environmental Satellite Data and Information Services, National Climatic Data Center, Camp Springs, MD Kogan F, Guo W, Jelenak A (2010) In: Kogan F, Powell A, Fedorov O (eds) Global vegetation health: long-term data records. Use satellite and in situ data to improve sustainability. Springer, New York, in this book Kogan FN (2001) Operational space technology for global vegetation assessments. Bull Am Meteorol Soc 82(9):1949–1964 Kogan FN (1997) Global drought watch from space. Bull Am Meteorol Soc 78:621–636 Lucht W, Prentice IC, Myneni RB, Sitch S, Friedlingstein P, Cramer W, Bousquet P, Buermann W, Smith B (2002) Climate control of the high-latitude vegetation greening trend and Pinatubo effect. Science 296:1687–1688 Myneni RB et€al (1997) Increased plant growth in the northern high latitudes from 1981–1991. Nature 386:698–702 Nemani RR, Keeling CD, Hashimoto H, Jolly WM, Piper SC, Tucker CJ, Myneni RB, Running SW (2003) Climate-driven increases in global terrestrial net primary production from 1982 to 1999. Science 300(5625):1560–1563 Rao CRN, Chen J (1999) Revised post-launch calibration of the visible and near-infrared channels of the advanced very high resolution radiometer on the NOAA-14 spacecraft. Int J Remote Sens 20:pp 3485 Zhou L, Tucker CJ, Kaufmann RK, Slayback D, Shabanov NV, Myneni RB (2001) Variation in northern vegetation activity inferred from satellite data of vegetation index during 1981 to 1999. J Geophys Res 106(D17):20069–20083
Global Warming, Atlantic Multi-decadal Oscillation, Thermohaline Catastrophe and Their Impact on Climate of the North Atlantic Region Alexander Polonsky
Abstract This paper presents investigation of regional and global warming, Atlantic Multidecadal Oscillation (quasiperiodic natural variations of the ocean-atmosphere system in the North Atlantic with typical time scales of 50–100 years) and thermohaline catastrophe (blocking of thermohaline circulation in the North Atlantic). The typical scale of the Atlantic Multidecadal Oscillation (AMO) is determined by the intensity of the meridional oceanic circulation in the North Atlantic. The analyzed oscillation affects various climatic characteristics: air temperature, river discharge in the European and North-American regions, the number and intensity of tropical cyclones in the Atlantic Ocean, and the parameters of mid-latitude cyclones and anticyclones in the Atlantic–European region. The main mechanism by which the AMO affects the climatic characteristics of the regions neighboring with the North Atlantic is the atmospheric response to the thermal anomalies in the ocean leading to a shift of the centers of atmospheric action and to the changes in the intensity and predominant directions of propagation of atmospheric cyclones and anticyclones. By using the results of long-term instrumental observations carried out in Eastern Europe and the data array of reconstructed temperature in the Alpine region, it is shown that the AMO is responsible for a significant part of low-frequency variations of temperature in Europe. This fact confirms the potential predictability of the regional atmospheric AMO on the decadal-scale. The rate of quasi-periodical regional warming/cooling of surface air temperature due to AMO can exceed the regional temperature rising due to global warming. So, the fast warming of the North Atlantic region during the last 3–4 decades of the twentieth century is due to coincidence of human-induced temperature increase and transition from negative to positive phase of the AMO. Realization of thermohaline catastrophe for the recent climatic epoch is unlikely. Keywords Global warming • Atlantic Multidecadal Oscillation • Thermohaline catastrophe A. Polonsky () Marine Climate Research, Marine Hydrophisical Institute of National, Academy of Sciences of Ukraine, 2 Kapitanskaya St., Sevastopol, Ukraine e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_15, © Springer Science+Business Media B.V. 2011
125
126
A. Polonsky
Introduction: General Characteristics of Global Warming, Atlantic Multidecadal Oscillation, Thermohaline Catastrophe
0.0 1900
1950 Year
1900
Temperature anomaly (�C)
0.0 2000
South America 1.0
1950 Year
2000
Africa 1.0 0.5
Asia 1.0 0.5 0.0
1900
1950 Year
2000
0.0 1950 Year
1.0 0.5 0.0 1950 Year
2000
2000
Global Land 1.0 0.5 0.0 1900
1950 Year
models using only natural forcings
2000
Temperature anomaly (8C)
1900
2000
1950 Year
1900
0.0
0.5
Global
1900
0.5
Temperature anomaly (�C)
0.5
Europe 1.0
Temperature anomaly (�C)
1.0
Temperature anomaly (8C)
Temperature anomaly (8C)
Temperature anomaly (�C)
North America
Temperature anomaly (�C)
Temperature anomaly (�C)
Global warming. Global warming of the low troposphere due to anthropogenic emission is assessed as 0.74°C per 100 years. Greenhouse effect is due mostly to the emission of CO2, CH4, N2O, and SF6. Human-induced global warming is assessed for 2005 as about 1.6 W/m2 (IPCC4 Assessment 2007). However, its regional manifestations differ for various continents and world ocean. Figure 1 shows that there is the superposition of century-scale linear trend of the surface air temperature (SAT) and quasi-periodical (period is ~60 years) high-amplitude signal over the North America and Europe. The relative role of human-induced trend and quasi-periodical natural climatic signal in the recent warming period is under discussion. In fact, quasiperiodical interdecadal warming and cooling of the North Atlantic is of the same
Australia 1.0 0.5 0.0 1900
1950 Year
2000
Global Ocean 1.0 0.5 0.0 1900
1950 Year
2000
observations
models using both natural and anthropogenic forcings
Fig. 1 Comparison of observed continental- and global-scale changes in surface temperature with results simulated by climate models using natural and anthropogenic forcing. Decadal averages of observations are shown for the period 1906–2005 (black line) plotted against the centre of the decade and relative to the corresponding average for 1901–1950. Lines are dashed where spatial coverage is less than 50%. Dark shaded bands show the 5–95% range for 19 simulations from five climate models using only the natural forcing due to solar activity and volcanoes. Light shaded bands show the 5–95% range for 58 simulations from 14 climate models using both natural and anthropogenic forcing (After IPCC4 Assessment 2007)
Global Warming, Atlantic Multi-decadal Oscillation
127
order and exceeds human-induced warming (Polonskii 2001; Polonskii and Voskresenskaya 1996, 2004). Therefore, the discussion of the relative role of anthropogenic and natural climate changes is one of the principal goals of this paper. Atlantic Multidecadal Oscillation. The quasi-periodic thermal variations in the North Atlantic with typical time scales of 50–100 years are now called the Atlantic Multidecadal Oscillation (AMO). The natural variations of the ocean–atmosphere system on these time scales are described in numerous works (Polonskii 2001; Polonskii and Voskresenskaya 1996, 2004; Delworth et al. 1996; Delworth and Greatbatch 2000; Eden and Jung 2001; Enfield and Mestas-Nunez 1999; Hatun et al. 2005; Knight et al. 2005; Kushnir 1994; Raa et al. 2004; Schlesinger and Ramankutty 1994). For the first time, this problem was analyzed in (Kushnir 1994) using the results of century instrumental observations of temperature and atmospheric pressure for the North Atlantic region. Almost simultaneously, Schlesinger and Ramankutty (1994), by using the one-dimensional spectral analysis of the global surface air and sea-surface temperature data averaged over several large regions, showed that their ~65-year oscillations are presented mainly by the North-Atlantic mode because it manifests itself just in this region. Polonskii and Voskresenskaya (2004) described in details the lowfrequency variability of hydrometeorological fields and turbulent heat fluxes in the North Atlantic with the typical time scale of ~65 years, which have been firstly revealed by them in 1994. Later, the following important fact was established (Enfield and Mestas-Nunez 1999): The interdecadal mode of the global sea-surface temperature is one of the main signals in the temperature field of the world ocean on the scales from interannual to multidecadal, which are not connected with the El-Niño–SouthernOscillation and it induces an atmospheric response of the pressure field. The maximum amplitude of this signal is in the North Atlantic spread to the southeast of Greenland. Therefore, in some research (see, e.g., Polonskii 2001; Polonskii and Voskresenskaya 1996), it was called the interdecadal mode of the North-Atlantic Oscillation (NAO). The term AMO has become generally accepted for this mode only in recent years. It emphasizes not only the Atlantic origin of this oscillation but also its low-frequency character. Somewhat later, the AMO was detected in modeling of the global climatic system by using the coupled models of the ocean–atmosphere system (Delworth and Greatbatch 2000; Knight et al. 2005; Raa et al. 2004). As a quantitative characteristic of the AMO, one uses an index reflecting the annual average anomaly of the sea-surface temperature (SST) in the North Atlantic (as a rule, averaged between the equator and 60°N). The variations of this index are quasi-periodic (Fig. 2a). As an additional characteristic of the AMO, one can use the NAO index smoothed by a low-frequency filter. The analysis of the long-term series of the NAO index reconstructed from the paleodata since 1675 shows that the 54–68-year variations of the NAO index are significant on a 1% level (Luterbacher et al. 1999). Later it was shown (Eden and Jung 2001) that there is a significant interdecadal variability of the heat fluxes on the ocean–atmosphere boundary connected with the low-frequency mode of the NAO (i.e., in fact, with the AMO). The quasi-periodic oscillations of the meridional heat transport (MHT) in the North Atlantic are usually regarded as the main mechanism for formation of the low-frequency variations SST and heat fluxes on the ocean–atmosphere boundary in the North Atlantic (i.e., of the AMO). However, there are different viewpoints
128
A. Polonsky
SST anomaly (�C)
a
0.3 0.2 0.1 0.0 −0.1 −0.2 −0.3
b
1880
1900
1920
1940 Year
1960
1980
2000
60N 30N 0 30S 180 −0.15
90W
0
−0.1 −0.05 0 0.05 0.1 Standard Temperature Anomaly (�C)
90E 0.15
Fig. 2 Variations of the AMO index (upper panel) and the anomalies of the surface air temperature over the globe for positive values of the AMO index equal to the standard deviation s (lower panel) (After Knight et al. 2005)
concerning the main cause of generation of the AMO. Some researchers believe that the quasi-periodic low-frequency oscillations of AMO in the North Atlantic appear as a consequence of the phase shift between the thermal and haline modes. The other researchers insist on the predominantly thermal origin of the AMO. In their opinion, the AMO is supported by the phase shift between the variations of SST and heat fluxes on the ocean–atmosphere boundary (for the discussion of this problem, see IPCC4 Assessment (2007), Polonskii (2001), Delworth et al. (1996), Delworth and Greatbatch (2000), Knight et al. (2005), Raa et al. (2004)). The importance of the last mechanism is confirmed by the fact that a phase shift of 16–17 years is observed between the low-frequency variations of absolute humidity in the boundary layer of the atmosphere (leading to the variations of latent heat fluxes through the sea–air boundary) and SST in the northwest part of the North Atlantic (Polonskii 2001). Regional North American and European manifestations of AMO will be described and discussed below. Thermohaline catastrophe. One of the possible consequences of the global warming is thermohaline catastrophe. This catastrophe is resulted from the fast warming of high-latitudes, ice and snow melting and associated critical freshening of the upper ocean layers in the North Atlantic, which causes the blocking of thermohaline convection. As a result, the meridional thermohaline circulation in the
Global Warming, Atlantic Multi-decadal Oscillation
129
North Atlantic will be shut down. Associated climate changes may be tremendous for the North America and Europe. Henry Stommel was the first who pointed to this (Stommel 1961). However, there are strong discussions concerning the likelihood of catastrophic scenario for the recent climate situation. One group of researchers argues that thermohaline catastrophe is a reality, while the other one disagrees (Polonskii 2001). Critical discussion of likelihood of thermohaline catastrophe in the recent climatic conditions is one of the focuses of this paper.
Manifestations of the AMO in the Large-Scale Atmospheric Characteristics and the Statistics of Cyclones in the Atlantic–European Region Although the AMO represents mainly an oceanic signal, it leads to a significant atmospheric response, which was, apparently, first noticed by Kushnir (1994). By using the data on atmospheric pressure accumulated since the middle of the nineteenth century, Enfield and Mestas-Nunez (1999) showed that there is a significant global atmospheric response to the anomalies of SST in the North Atlantic whose manifestations are detected in the entire lower and middle troposphere. As a result, the general atmospheric circulation is characterized by different modes for high and low AMO indices (i.e., for the positive AMO indices exceeding the standard deviation and their negative values lower than -s). Furthermore, the AMO index correlates with various climatic characteristics: air temperature, river discharge in the European and North-American regions, the number and intensity of tropical cyclones in the Atlantic Ocean, and the parameters of mid-latitude cyclones and anticyclones in the Atlantic–European region (Polonskii 2001; Kerr 2005; Polonskii et al. 2004, 2007). As an illustration, let’s consider three important manifestations of the AMO in the regional atmospheric characteristics. First, the AMO generates large-scale thermal anomalies in the lower troposphere of the North-Atlantic region. If the AMO index is high, then almost the entire North Atlantic, North America, and Western Europe are characterized by positive temperature anomalies (Fig. 2). Despite their relatively small values (~0.2°C), these anomalies are stable and their time scale is sufficiently large to exert a significant influence on the climatic variability in this region. Between the early 1960s and mid 1970s, the North Atlantic SST has been decreased by about 0.4°C. This tendency essentially exceeded the anthropogenic warming. As a result, the long-term linear trend of SST in the Northern North Atlantic was a negative till the early 1990s (Polonskii and Voskresenskaya 2004). Second, the AMO affects the number and intensity of the Atlantic tropical cyclones characterized by the destructive force upon the southeast states of the USA. For high SST (positive phase of the AMO), the number of tropical cyclones (Fig. 3) and their intensities significantly increase, which confirms the well-known fact of SST influence on the tropical cyclones. Third, the AMO affects the number of cyclones in the European region. This influence is caused by the shifts of the North-Atlantic centers of atmospheric action to the different phases of the AMO (Fig. 4). If the
1903 -1925, 1971-1994 47 Years
1928 -1970, 1995-2000 51 Years
Fig. 3 Trajectories of tropical cyclones in the northwest part of the Subtropical Atlantic at the period of negative (a) and positive (b) phases of the AMO (After Kerr 2005)
Fig. 4 Changes in (1) the surface pressure difference between the Azores High and Icelandic Low in winter and (2) in the difference between the latitudes of these centers of atmospheric action; also the third-order polynomial approximations of (1) and (2) are shown (the linear trend was removed preliminarily, and the series were filtered by a 5-year sliding filter) (After Polonskii et al. 2004)
Global Warming, Atlantic Multi-decadal Oscillation
131
Fig. 5 Time variation of the frequency of cyclones in the Black Sea region for four seasons of 1952–2000; thin solid lines correspond to data on all cyclones without filtration, bold solid curves correspond to 5-year sliding averages for all cyclones, straight lines show the linear trends of the frequency of cyclones from 1962, and dashed curves correspond to data for 25% of the most intense cyclones (After Polonskii et al. 2007)
AMO index is high, then the North-Atlantic cyclones propagate mainly to Central and Eastern Europe, which leads to the increase in the number of cyclones in the Black-Sea region (Fig. 5). It is important to emphasize the following two facts. (1) For the regional EastEuropean manifestations of the AMO, the displacement of the centers of action is much more important than the changes in their depth. The character of displacements of the centers of atmospheric action for the positive phases of the AMO (on multidecadal scale) and NAO (on the interannual scale) is different. Thus, the positive phase of the NAO is accompanied by a shift of the centers of action in the northeast direction but, for the positive AMO indices, these centers are shifted to the southwest (Polonskii and Semiletova 2002). This feature is responsible for the difference in the characters (and even in the signs) of correlations between the NAO
132
A. Polonsky
and AMO indices and the parameters of cyclones in Europe. (2) In Fig. 5, the seasonal difference between the low-frequency variations of the Black-Sea cyclones is well visible. This difference is caused by the fact the outlined mechanism of the AMO impact on the Black-Sea cyclones and is realized mainly in the winter–spring period. At the same time, in summer, the influence of the AMO on the Black-Sea region is at a minimum.
tatistical Significance of the Detection of Different S Characteristics of the AMO and Its Predictability Since the typical period of the AMO is 50–100 years, the possibility of the statistically significant detection of this oscillation from ~100–200 years of instrumental data is problematic, indicated that, which provoked a long discussion in the literature. Avoiding the principal arguments, we should accept the major authors’ idea (Elsner and Tsonis 1994) that for the statistically significant detection of the AMO, it is necessary to have substantially longer series of observations than the available observations. This is also right for long time series containing intermittent low-frequency quasi-periodic oscillations, and the AMO, which belongs exactly to the oscillations of this type. To illustrate this fact, the long-term series of instrumental observations of January temperature in Warsaw (from 1779 to 1998) and its wavelet decomposition were selected (Fig. 6a and b). In both diagrams the time dependence of temperature itself and the time variation of the coefficient D6 of the wavelet decomposition, one can see pronounced 60–80-year temperature oscillations imposed on the trend or on the lower-frequency variations of significant intensity (see the curve for the coefficient A6). It should be emphasized that the characteristic amplitude of these oscillations changes by a factor greater than 2, increasing at the end of the nineteenth and twentieth centuries as compared with the end of the eighteenth century and the first three quarters of the nineteenth century. For the entire period of observations, at most three periods of the AMO were involved, which makes it difficult to reliably evaluate the statistical significance of the obtained results. For the evaluation of the statistical significance of the AMO characteristics calculations, it is necessary to use much longer series, e.g., reconstructions of hydrometeorological parameters based on different data (including paleodata). The 2000-year (from 90 bc to 1935 ad) annual average air temperature in the Alpine cave was reconstructed based on the isotopic composition of oxygen in the stalagmites (Mangini et al. 2005). The data demonstrates significant natural temperature oscillations in the range of periods from 50 years to several hundreds of years (Fig. 7a). It should be especially noted that there was a sharp temperature increasing in the early Middle Ages (before the fast cooling in the Little Glacier Period), which substantially exceeds the rate of the present-day warming (for the Alpine region), and the presence of quasi-periodic 60–70-year oscillations. However it should be emphasized that the reconstructed data underestimate the natural lowfrequency variability, as was clearly shown in Storch et al. (2004).
Global Warming, Atlantic Multi-decadal Oscillation
a T
b A6
133
T, �C 2 0 −2 −4 −6 −8
−1.5 −2 −2.5 −3 −3.5 1
D6
0 −1 1
D5
0 −1 1
D4
0 −1 5
D3
0 −5 5
D2
0 −5 5
D1
0 −5
20
40
60
80
100
120
140
160
180
200
yr
Fig. 6 Variation in (a) January temperature in Warsaw during 1779–1998 and (b) the coefficients of its wavelet decomposition (in the wavelet decomposition, the Meyer mother wavelet is used)
For the analysis of 60–70-year oscillations, temperature fluctuations with periods exceeding 200 years were filtered out. Then the spectrum over the entire realization was calculated by using the Parzen filter. This filter was chosen because its characteristics maximally increase the number of degrees of freedom (decrease the variance of an estimate). However, at the same time, the bias of an frequency estimate increases, but, in our case, the precise localization of the frequency (period) of the predominant oscillation was not of the most important priority. Our goal was to determine the periodicity of the 60–70-year oscillations evaluating their significance. Figure 7b shows a significant peak in temperature spectrum on a period of about 67 years. The energy (amplitude) of the corresponding oscillations is at a maximum over the entire range of the periods from 2 to 200 years, and these oscillations
134
A. Polonsky
a 3.5
T, �C
3 2.5 2 1.5 0.5 0 −0.5
100 BC
b
ST (�), deg2
3.0
400
900
1400
1900 AD
Period of 66.7 yr n~100
2.0
Period of 20 yr 1.0
0.0 0.00
0.04
0.08
0.12
0.16
0.20
ω, cycle/yr Fig. 7 Variation in (a) the reconstructed temperature (Mangini et al. 2005) from the data on the isotopic composition of oxygen in the stalagmites of one of Alpine caves and (b) temperature spectrum from which the periods exceeding 200 years were filtered out ( n is the number of degrees of freedom)
themselves are responsible for almost 40% of the total variance of temperature fluctuations in the indicated frequency range. Note that the spectrum presented in Fig. 7b is characterized by red shift (i.e., the energy of spectral components decreases from low to high frequencies). The fraction of temperature fluctuations with the periods exceeding 20 year in the total variance is greater than two thirds. This may seem strange because, as a rule, the more significant proportion of fluctuations of air temperature is concentrated in the higher-frequency part of the spectrum, i.e., on the interannual scale (in another words, the spectrum should be more white and this is confirmed, in particular, by the time variation of air temperature in Warsaw shown in Fig. 6). However, one should not give much attention to the last fact because the method of reconstruction itself decreases the amplitude of the highest-frequency components of temperature fluctuations. The 60–70-year temperature oscillations in the lower troposphere caused by the AMO and of their fairly high intensity (corresponding amplitude of temperature
Global Warming, Atlantic Multi-decadal Oscillation
135
Streamfunction (Sv)
fluctuations is about 1°C) are statistically significant. This provides a strong ground for predictability of the AMO and its atmospheric manifestations on at least the decadal scale. Indeed, as a result of the high inertia of the ocean–atmosphere system (mainly due to oceanic component, which determines the typical oscillation period of 60–70 years), the predictability of the AMO and the related repeatability of various hydrometeorological events (including catastrophic) in the North Atlantic region on the decadal scale is feasible (Griffies and Bryan 1997). However, this topic is still debated because: (1) Some researchers are not completely sure that the meridional thermohaline circulation in the present climatic epoch is stable, although in recent years, this opinion is gradually revised (Polonskii 2001; Manabe and Stouffer 1999); (2) the signal-to-noise ratio of the atmospheric response to the AMO is fairly low, which is caused by both the relatively small absolute value of the SST anomalies at different phases of the AMO signal and the significant amplitude of the natural internal noise; (3) recent best coupled models are able to simulate the AMO (Fig. 8). However, the period and magnitude of oscillation are not the same as in the observation because of poor parameterization of small-scale processes (interaction of ocean and atmosphere, etc.).
18
a
17
16
15
Period (Years)
1000
b
c
300 100 30 10 0
200
400
600
800
1000
1200
Time (Years) 2
4
6
8
0.1
1.0
Power (Sv2years)
10.0
10
Power (Sv2)
Fig. 8 Row of meridional streamfuntion in the North Atlantic (a, Sv), its wavelet decomposition (b) and average spectrum (c). Results of 1,400 years simulation of coupled system evolution (From Knight et al. 2005)
136
A. Polonsky
Likelihood of Thermohaline Catastrophe in Different Climatic Epochs Stommel (1961) considered the thermohaline catastrophe (TC) as hypothetical haline regime. However, now, there are numerous evidences that such regime was realized in the past (see e.g., Broecker 2006; Ellison et al. 2006). Some researchers argued that TC can occur in the nearest future as a result of the human-induced climate warming (Rumstorf 1995). However, the most recent research indicated that likelihood of TC at the present climatic tendencies is very low (Polonskii 2001; Manabe and Stouffer 1999; Latif et al. 2000). Here, we provide a physical ground for the TC using simplified box model and the results published in (Polonskii 2001, 2002). Considering that the oceanic basin is given by D + dD depth and consists of three homogeneous boxes (simplified Stommel’s four-box model) shown in Fig. 9, we simulated environmental parameters in the North Atlantic. The heat/salt equations integrated over the boxes yield the following system of ordinary differential equations for temperature and salinity (Ti: and Si): d T1 = R1T (T3 − T1 ) + QT1 dt
(1)
d V2 T2 = R2T (T1 − T2 ) − QT2 dt
(2)
d V3 T3 = R3T (T2 − T3 ) dt
(3)
d V1 S1 = R1S ( S3 − S1 ) + QS2 dt
(4)
V1
Fig. 9 Schematic diagram of the three-box model of the North Atlantic: (1) surface tropical box, (2) surface extratropical box, (3) abyssal box (the notation used here is clarified in what follows)
Global Warming, Atlantic Multi-decadal Oscillation
137
V2
d S2 = R2S ( S1 − S2 ) − QS2 dt
(5)
V3
d S3 = R3S (S2 − S3 ) dt
(6)
where the Vi are the volumes of the boxes, the terms RT and RS are the total heat and salt advection (index T and S denote the heat and salt terms, respectively) between the boxes equal to the average advectively redistributed part of the heat/fresh balance of the ocean, QS. / QT – fresh /heat fluxes at the ocean surface; the following dimensionless geometric parameters of the model were introduced here: the ratio of the depth of the upper boxes to the depth of box 3 (d), 0 < d < 1 (d << 1), and the ratio of the volume of each of the upper boxes to their total volume ai = Vi/Vh, where Vh= V1 + V2, a1 + a2 = 1. It is easy to see that, for system (1)–(3), the average temperature is preserved. Hence, T3 + d(a1 T1 + a2 T2) = 0, where Ti are regarded as deviations from the average temperature. Therefore, we can reduce the order of the original system by one, preserving, Eqs. (1) and (2) and replacing T3 with – d(a1 T1 + a2 T2). Since the average temperature is conserved, we arrive at the following stationary solution of the Eqs. (1)–(3): T1 = ∆
1 + α2δ 1+ δ
T2 = −∆α1
δ , 1+ δ
QT . The same relationship can be written for the salinity. RT For the solution of the nonstationary problem the initial conditions were determined from the archive data for the boxes (Fig. 10). Using these data and standard equation of the water state one can calculated the initial state of above boxes for the recent climate parameters (temperature, salinity and density) as: Т1 = 21.6°С, S1 = 36.4‰, r1 = 1.02541 g/cm3, and Т2 = 7.0°С, S2 = 34.6‰, r2 = 1.02712 g/cm3. In order that the thermohaline catastrophe started, the high-magnitude haline anomaly must be imposed at the initial salinity of the box 2. This anomaly must decrease the density of the box 2 to the density of the box 1. Achievement of such condition under the present climatic is practically impossible. In fact, to decrease the density of the second box to about 1.02541 g/cm3 it is necessary to reduce its salinity by about 2‰. This claims additional inflow of about 1 × 106 m3/s of the fresh water into the second box per decade (taken into account the size of the box). Such rate of freshening of the Northern North Atlantic exceeds the observed value by the order of ~2 (Polonskii 2001). This rate of freshening for century is difficult to achieve because first, in case of quasi-periodical AMO the sign of high-latitude temperature trend must change after a few decades; and second, there are not enough ice and snow in the present climate to support a very high rate of melting for a long time (century-scale). where ∆ =
138
A. Polonsky −60
−30
0
−100
−75
50
25
−90
0
−5
50
0
25
25
−25
75
75
25
50
0 25 −25
0 W
−90
25
−60
0
25 −25 0
25
−30
0
B. . E
Fig. 10 Average (climatic) heat fluxes (W/m2) at the ocean surface and schematic boundary of the boxes used in the simplified model. Integral heat gain in the tropical box is equal to the negative heat flux over the extratropical box. The similar condition is imposed on the average fresh balance of these two boxes
The other climatic conditions took place during transition from glacier to interglacial periods (in particular, during last glacier/interglacial cycle). There was extended glacier over the North America and Europe at the peak of glacier period (in contrary to the climatic optimum, see Fig. 11). During transition to the interglacial period the high rate of warming and melting occurred for a long time. It is likely that both events led to critical freshening of the North Atlantic and development of thermohaline catastrophe (Broecker 2006; Ellison et al. 2006).
How Reliable the Ocean Observational Network to Detect the Multidecadal Variability? Problem with Long-Term Deep-Ocean Observations Ocean variability is a crucial factor in regulating the interdecadal-to-multidecadal changes in the coupled ocean-atmosphere system and, hence, causing the low-frequency climate variability. Therefore, the low-frequency ocean variability is crucial factor to study the long-term (natural and human-induced) climate changes.
Global Warming, Atlantic Multi-decadal Oscillation
139
Fig. 11 Reconstruction of climatic zones for the last climate optimum (~6,000 years ago, upper panel – a) and glacier maximum (~21,000 years ago, lower panel – b) (After IGBP Science series 2003)
Lack of quantity and inferior quality of oceanographic observations, creates some problem with the detection of the climatic signals in deep-ocean layers. A brief discussion of this problem is provided below (for comprehensive discussion see Polonskii 2001; Polonsky 2001). Deep-sea observations have been performed for more than a century. Before the 1970s, Nansen profiles were the main source of data, which were replaced later
140
A. Polonsky
Number of observations
60000
40000
Number of the NODC deep-see observations over the globe 1 - XBT data 2 - CTD/STD data 3 - Nansen bottles
1
20000 3
2
0 1940
1960
1980
YEARS
Fig. 12 Total number of sub-surface observations in the World Ocean available from the NODC archive (After Polonsky 2001)
with the sounding observations. However, only a small portion of the profiles reached the deep layers (Fig. 12). The maximum depth of XBT sounding is usually 800 m and this activity decreased with the beginning of satellite era. This means that even in the regions with the most intense observational activity only decadalscale (not longer-term) variations in the upper (~1 km) ocean layer may be estimated with reasonable accuracy (signal-to-noise ratio > 1). The unique Ocean Weather Station (OWS) network was established after the WWII (Fig. 13). Unfortunately, after the World economic crisis of 1972–73 the number of OWS vessels has been reduced by three times and after satellite era beginning the OWS network interrupted its activity (except OWS “Mike” which is now the national Norwegian property). Recent ocean observational programs (such as WOCE and ARGO in the World Ocean, TOGA-TAO in the Pacific Ocean, PIRATA and RAPID in the Atlantic, etc.) are very important. However, the observations should continue for decades for detection of muiltidecadal climate variations (Polonsky 2001; Cunningham et al. 2007). As a result there are two problems to address the issue: (a) detection of long-term variations in the deep-ocean layers in the presence of high-magnitude noise (Fig. 14) and (b) the separation of trend-like and low-frequency quasi-periodical signals. So, the recent and historical deep-ocean observational datasets are too short, sparse and noisy to detect the multidecadal oceanic variability with the reasonable statistical significance. This is absolutely clear that the cessation of the OWS observations was rash resolve and the similar faults should not be repeated in the future.
Global Warming, Atlantic Multi-decadal Oscillation
141
Fig. 13 OWS network in the North Atlantic. Station “Mike” (in light square) is only one operating OWS now (After 60 years of Station Mike 2000)
Fig. 14 Variability of meridional heat transport in the Subtropical Atlantic Ocean (between 26 and 32°N, PW), assessed in (Polonsky and Krasheninnikova 2007) using direct method (Hall and Bryden 1982) and all available zonal hydrographic sections. Vertical lines show the standard deviations, bold curve represents the sixth order polynomials
142
A. Polonsky
Conclusions Most recent publications argue that observed global warming and its manifestations over the North Atlantic region (especially during the last 30–40 years) are unprecedented for the climatic history of the Earth and they are due to greenhouse effect of anthropogenic origin. In fact, fast increasing of SST in the North Atlantic region during the last 3–4 decades of the twentieth century is due to coincidence of human-induced trend and transition from negative to positive phase of the AMO. The AMO is the quasi-periodic variations of the coupled ocean-atmosphere system in the North Atlantic with typical time scale of 50–100 years, which is determined by the rate of the meridional oceanic circulation in the North Atlantic. The AMO manifests itself in the following climatic characteristics: air temperature and river discharge in the European and North-American regions, the number and intensity of tropical cyclones in the Atlantic Ocean, and the parameters of the mid-latitude cyclones and anticyclones in the Atlantic–European region. The main mechanism by which the AMO affects the climatic characteristics of the regions neighboring with the North Atlantic is the response of heat fluxes on the ocean–atmosphere interface and atmospheric pressure to the thermal anomalies in the ocean. This leads to a shift of the centers of atmospheric action and to changes in the intensity and predominant directions of propagation of atmospheric cyclones and anticyclones. The high inertia of the ocean–atmosphere system (due to mostly oceanic component, which determines the oscillation period) provide a basis for AMO’s predictability and the related repeatability of various hydrometeorological events in the North Atlantic region on the decadal scale. Lack of the long-term deep-sea observations creates a problem for detection of the multidecadal climate changes. However, the available data permit to conclude that realization of thermohaline catastrophe for the recent climatic epoch is unlikely.
References Broecker WS (2006) Was the Younger Dryas triggered by a flood? Science 312(5777):1146–1148 Cunningham SA et al (2007) Temporal variability of the Atlantic meridional overturning circulation at 26, 5N. Science 317:935 Delworth T, Greatbatch RJ (2000) Multidecadal thermohaline circulation variability driven by atmospheric surface flux. J Climate 13(9):1489–1495 Delworth T, Manabe S, Stouffer RJ (1996) Interdecadal variability of the thermohaline circulation in a coupled ocean–atmosphere model. J Climate 6(11):1993–2011 Eden C, Jung T (2001) North Atlantic interdecadal variability: oceanic response to the North Atlantic Oscillation (1865–1997). J Climate 14(5):676–691 Ellison CRW, Chapman MR, Hall IR (2006) Surface and deep ocean interactions during the cold climate event 8200 years ago. Science 312(5783):1929–1932 Elsner JB, Tsonis AA (1994) Low-frequency oscillation. Nature 372:507–508 Enfield D, Mestas-Nunez AM (1999) Multiscale variabilities in global SST and their relationships with tropospheric climate patterns. J Climate 12(9):2719–2733 Griffies A, Bryan K (1997) Decadal predictability of the North Atlantic variability. Science 275(5695):181–184
Global Warming, Atlantic Multi-decadal Oscillation
143
Hall MM, Bryden HL (1982) Direct estimates and mechanisms of ocean heat transport. Deep-Sea Res 29(3A):872–881 Hatun H, Drange H, Hansen B et al (2005) Influence of the Atlantic Subpolar Gyre on the thermohaline circulation. Science 309(5742):1841–1844 IGBP Science series (2003) No.3:18 IPCC4 Assessment (2007) (Topics 1 and 2): 1–14 Kerr RA (2005) Atlantic climate pacemaker for millennia past, decades hence? Science 309(5731): 41–42 Knight J, Allan R, Folland C, Vellinga M, and Mann M, (2005) The atlantic multidecadal oscillation: a signature of thermohaline circulation cycles in observed climate. CRCES Workshop on Decadal Climate Variability, 19 October 2005 Kushnir Y (1994) Interdecadal variations in North Atlantic Sea surface temperature and associated atmospheric conditions. J Climate 7(1):141–157 Latif M, Roeckuer E, Mikolajewicz U, Voss R (2000) Tropical stabilization of the thermohaline circulation in a greenhouse warming simulation. J Climate 13(11):1809–1813 Luterbacher J, Gyalistras D, Schmitz C et al (1999) Reconstruction of monthly NAO and EU indices back to AD 1675. Geophys Res Lett 26(17):2745–2748 Manabe S, Stouffer RJ (1999) Are two modes of thermohaline circulation stable? Tellus 51A(3): 400–411 Mangini A, Spütl C, Verdes P (2005) Reconstruction of temperature in the Central Alps during the past 2000 years from a d18O stalagmite record. Earth Planet Sci Lett 235(3–4):741–751 Polonskii AB (2001) Role of the ocean in the present-day climatic changes. Morsk Gidrofiz Zh 6: 32–58 Polonskii AB (2002) On the mechanism of decadal oscillations in the ocean–atmosphere system. Morsk Gidrofiz Zh 1:25–34 Polonskii AB, Semiletova EP (2002) On the statistical characteristics of the North Atlantic Oscillation. Morsk Gidrofiz Zh 3:28–42 Polonskii AB, Voskresenskaya EN (1996) Low-frequency variability of meridional drift transfers in the North Atlantic. Meteorol Gidrol 7:89–99 Polonskii AB, Voskresenskaya EN (2004) On the statistical structure of hydrometeoro-logycal fields in the North Atlantic. Morsk Gidrofiz Zh 1:14–25 Polonskii AB, Yu BM, Voskresenskaya EN (2007) Variability of Black Sea cyclones in the second half of the 20th century. Morsk Gidrofiz Zh 6:47–58 Polonskii AB, Basharin DV, Voskresenskaya EN (2004) North Atlantic Oscillation: description, mechanisms, and influence on the climate of Europe. Morsk Gidrofiz Zh 2:42–59 Polonsky AB (2001) Are we seeing human-induced warming of the deep layers in the North subtropical Atlantic. CLIVAR Exchanges 6(1):17–19 Polonsky AB, Krasheninnikova SB (2007) Meridional heat transport in the North Atlantic and its tendencies in the second half of XX century. Morsk Gidrofiz Zh 1:39–52 Raa L, Dijkstra HA, Gerrits J (2004) Identification of the mechanism of interdecadal variability in the North Atlantic Ocean. J Phys Oceanogr 34(12):2792–2807 Rumstorf S (1995) Bifurcations of the Atlantic thermohaline circulation in response to changes in the hydrological cycle. Nature 376:145–149 Schlesinger ME, Ramankutty N (1994) An oscillation in the global climate system of period 65–70 years. Nature 367:161–164 Stommel H (1961) Thermohaline convection with two stable regimes of flow. Tellus 13(2):224–230 Storch H et al (2004) Reconstructing past climate from noisy data. Science 306(5696):679–682. 60 years of Station Mike (2000). Norway
Global Warming and Possible Changes in the Recurrences of Grain Crops Anomalies Gennady Menzhulin and Artyom Pavlovsky
Abstract The recurrence of droughts caused by global warming for the period 2011–2050 in the Russian northwest region and some European countries is discussed in this paper. The selected regions include: Petersburg, Novgorod, Pskov and Kaliningrad oblasts, Sweden, Norway, Denmark, Lithuania, Latvia, Estonia, part of Great Britain, Finland and Ireland. Winter and summer wheat yields anomalies were used. To forecast yield anomalies recurrences the ensemble approach technique was used when yield was approximated by two components agrotechnological and weather-related. Following the ECHAM5 MPI-OM climatic scenario, the negative yield anomaly, especially will be increasing especially for winter wheat during 2010–2030. After 2030, the recurrence of the negative anomalies will decrease. The difference between winter and spring wheat will be negligible. Keywords Crop productivity anomalies • Droughts • Climate change • Statistical modeling
Introduction Scientific research indicates that climate changes impacts on the society might be considerable in the temperate latitudes of Northern Hemisphere. Thereby, it is extremely necessary to estimate possible consequences of global warming for such human activity as agriculture. Climate change (due to carbon dioxide increasing in the atmosphere) impacts on crop productivity are currently studying in terms of changes in average crop productivity triggered by changes in hydrothermal regime. This work is reflected in the international project entitled “Climate change and agriculture: Analysis of the possible consequences” sponsored by the US NSF and the US G. Menzhulin (*) and A. Pavlovsky Research Center for Interdisciplinary Environmental Cooperation, Russian Academy of Sciences, St. Petersburg, Russia e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_16, © Springer Science+Business Media B.V. 2011
145
146
G. Menzhulin and A. Pavlovsky
Agronomy Society. However, the task of global warming impact on the recurrence of drought years and crop losses on regional basis is far from the solution (Menzhulin and Savvateev 2002; Parry and Rosenzweig et al. 2002; IPCC 2008). This research is focused on the forecasting of winter and spring wheat yield anomalies in the Russia’s Northwest and Northern Europe based on the ensemble approach to the prediction of extreme yield. The yield was split into technological and weather components. (Pavlovsky 2006, 2007, 2008; Menzhulin et al. 2009). In Russia the following regions (oblast) were investigated St. Petersburg, Novgorod, Pskov and Kaliningrad. Other countries were represented by Sweden, Norway, Denmark Finland, Latvia, Lithuania, Estonia, a part of Great Britain, Scotland, Northern Ireland and Wales. As the initial information in our research the long and continuous series of crops yield and meteorological parameters for administrative regions of each country were used.
Input Data Wheat yield and production for the countries have been available since 1960 (http:// faostat.fao.org; http://eurostat.ec). For Russian northwest yield was obtained for administrative regions (oblast). Russian data series have 42 years; Swedish series on wheat cover the period from 1965; Denmark – from 1990. Beside country statistics they also contain administrative data. Others have only country statistics: Lithuania, Latvia and Estonia from 1945 for cereals; for Norway, Finland, Great Britain and Ireland – since 1960 for wheat but both spring and winter. The main source of historical meteorological surface data was USA’s National Climate Data Center (NCDC). We also used the series of 223 stations of the former USSR, prepared at the All-Russian Research Institute of Hydrometeorological Information. The basic requirement to meteorological data was the longest duration with no mission data.
Methodology Since the yield time series were different longevity and we did not have a priory information on trend change we used the 16-components “ensemble” of trend lines consisting of eight polynomials (linear, quadratic, cubic and so on) and eight exponential functions of the series. Yield deviation from trend was expressed hg ( j) = [y( j) − Yg ( j)]/Yg ( j), (g = 1,2,3,…,16), where y( j) is the productivity in the year j, Yg ( j) – the agrotechnology level of this crop productivity in the same year, g is the type of the trend line. For meteorological parameter series (tempera-ture and precipitation Ta , Pb ), the five components “ensembles” of trend lines were used. Meteorological anomalies of temperature (taj ) and precipitations (pj ) were calculated as: ta(i, j) = [T(i, j) − Ta (i, j)]/Ta(i, j), and pb(i, j) = [P(i, j) − Pb (i, j)]/Pb (i, j), (a,b = 1,…,5). In these two formulae T(i, j) and P(i, j) are the monthly mean air surface temperature and atmospheric precipitation in the month i of vegetative season of the year j.
Global Warming and Possible Changes in the Recurrences of Grain Crops Anomalies
147
After deriving yield trends for each series, yield anomalies (deviation from trend) were calculated. Each meteorological series were used for calculation of the 5-components ensembles (predictors) for each regression model. For example, for winter wheat regression models, 16-set ensemble of the predictant series and 5-set ensemble of 12 monthly temperature and 12 monthly precipitation predictors were developed; for spring wheat models the 16-set ensemble for predictant values and 5-set for eight temperature and eight precipitation predictors. The goal was to find the best statistical models from all options. When selecting the best model, the adjusted coefficient of determination (Da.) was used as the main criterion of the linear regression exactness. Also, all selected predictors must be statistically significant (according to the Student t-criterion) on the reassigned level. To estimate an absence of autocorrelation in the series of the predictant the Durbin–Watson test was used.
Discussion For the northwest European Russian the best winter wheat yield anomaly model was for Pskov oblast (Da= 0.73) and spring wheat model - for Novgorod oblast (Da= 0.67). Mean (Da) for all oblast of northwestern Russia was −0.57 for winter wheat and −0.50 for spring wheat. The highest (Da) for the rest of European countries were obtained for winter wheat models in Denmark (0.90) and Sweden (0.71), for the spring wheat models 0.86 and 0.55, accordingly. In Baltic countries, the highest (Da) for winter wheat were 0.52 in Latvia, 0.64 in Lithuania and 0.42 in Estonia, for spring wheat −0.64, 0.46 and 0.58, respectively. In the countries where wheat was not divided in winter and spring the largest (Da) were in Finland (0.63), Ireland (0.60) and Norway (0.54). In the Northern Ireland and Wales, Da values were 0.90 and 0.85, respectively. Figure 1 shows actual and modeled anomalies of the wheat of winter wheat in Pskov oblast (Russia, Da = 0.71), winter wheat in Kalmar county (Sweden, Da = 0.78), spring wheat in Funen county (Denmark, Da = 0.97) and spring wheat in Wales (Da = 0.90). These models can be used to estimate future impact of global warming on yield. In the fourth IPCC Report, the ECHAM5 MPI-OM model was recommended for the impact studies. In order to estimate global warming impacts on wheat we first, extrapolated the yield trend using five polynomial lines. For selection of “the best” we used the Kolmogorov’s criterion and the empirical indicator d, which calculated as d = min[∑(si-swn,i)2] where si is a fraction (in the series of deviations from the trend line) of the total mean square deviation with the frequencies higher than i, swn,I is a parameter used for the “white noise” characterization. For the future anomalies of weather parameters two criteria were identified: first, the integral periodogram line of basic meteorological parameter should not fall outside the 99% confidential intervals limit (“white noise”); second, the d value for the most “correct” trend line should be minimal. Tables 1 and 2 provide assessments of yield anomalies during 2011–2050 based on scenarios of the projections of the ECHAM5 MPI-OM climate model. Gray
Fig. 1 Actual and modeled yield anomalies for winter wheat in (a) Pskov (Russia), (b) Kalmar county (Sweden) and spring wheat in Funen county (Denmark) and Wales Table 1 Forecast of winter wheat yield anomaly (deviation from trend) for 2011–2050 based on the ECHAM5 MPI-OM climate scenario (white cells – positive anomalies, gray cells – negative anomalies) Country Russia/oblast Vologda Leningrad Great Novgorod Pskov Denmark/county Arhus Bornholm Copenhagen Funen North Jutland Ribe Ringköbing South Jutland Storström Vejle Viborg West Zealand Baltic/country Latvia Lithuania Estonia Sweden/county Älvsborg Blekinge Gotland Kristianstad Malmöhus Skaraborg Kalmar Örebro Östergötland Södermanland Stockholm Uppsala Vamland Västmanland
Decades 2011-2020 2021-2030 2031-2040 2041-2050 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0
Global Warming and Possible Changes in the Recurrences of Grain Crops Anomalies
149
Table 2 Forecast of spring wheat yield anomaly (deviation from trend) for 2011–2050 based on the ECHAM5 MPI-OM climate scenario (white cells – positive anomalies, gray cells – negative anomalies) Decades 2011-2020 2021-2030 2031-2040 2041-2050 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0
Country Russia/oblast Vologda Kaliningrad Leningrad Great Novgorod Pskov Denmark/county Arhus Bornholm Copenhagen Funen North Jutland Ribe Ringköbing South Jutland Storström Vejle Viborg West Zealand Baltic/country Latvia Lithuania Estonia Sweden/country Älvsborg Blekinge Gotland Kristianstad Malmöhus Skaraborg Kalmar Örebro Östergötland Södermanland Stockholm Uppsala
Table 3 Forecast of spring wheat yield anomaly (deviation from trend) for 2011–2050 based on the ECHAM5 MPI-OM climate scenario (white cells – positive anomalies, gray cells – negative anomalies) Country
Decades 2011-2020 2021-2030 2031-2040 2041-2050 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0
Ireland Norway North Ireland Wales Finland
cells represent the years with negative (below trend) yield anomalies and white cells with positive anomalies. In general, during the projected period, favorable hydrothermal conditions will be resulted in more years with positive yield anomalies. About 12% of the years will be with negative yield anomalies (both winter and spring wheat). Especially high difference in these recurrences will fall on the nearest decade 2011–2020. In the northwestern Russia more negative yield anomalies are observed due to higher frequency of drought especially during the winter wheat
150
G. Menzhulin and A. Pavlovsky
growing season. In contrast, the period of more favorable hydrothermal conditions for winter wheat will fall on 2041–2050. The period of 2031–2050 will be favorable for winter wheat due to improvement of weather conditions. The recurrences of favorable and unfavorable conditions for winter and spring wheat are almost equal. The first decade of the period is characterized by higher recurrence of the negative (below trend) wheat anomalies due to anticipated unfavorable hydrothermal conditions. For the rest of the period these anomalies will decrease. The optimal conditions in Norway and Finland will be during 2021–2030 and in Ireland, Northern Ireland and the Wales during 2031–2040.
Conclusion Over the entire investigated region, the first half of the 2011–2050 will be less favorable for wheat productivity based on the ECHAM5 MPI-OM global climate change scenario; after the 2030, the situation will improve. Since in this study only one climate change scenario was used this work should be expanded to the other scenarios recommended by IPCC Report.
References IPCC (Intergovernmental Panel on Climate Change) (2008) Climate Change 2007 – Impacts, Adaptation, and Vulnerability. Cambridge University Press, Cambridge, UK Menzhulin GV, Savvateev SP (2002) Global food production problem and contemporary global warming. Climate changes and their consequences. Nauka, St. Petersburg, pp 122–152, in Russian Menzhulin GV, Pavlovsky AA, Shamshurina NV (2009) Statistical modeling of crop productivity anomalies based on ensemble approach. Bulletin of St. Petersburg State University, Series 7, Issue 3, pp 76–85 (in Russian) Parry ML, Rosenzweig C et al (2002) Climate change and world food security: a new assessment. Global Environ Change 9 Pavlovsky AA (2006) Climate changes and recurrence of extreme hydrothermal phenomena. Bulletin of St. Petersburg State University, Series 7, Issue 3, pp 88–94 (in Russian) Pavlovsky AA (2007) Statistical modeling of wheat productivity dynamics and values of its changing under the impact of global warming. Ph.D. Thesis, Main geophysical observatory, St. Petersburg, 16 p (in Russian) Pavlovsky AA (2008) Statistical modeling of the wheat productivity dynamics and assessment of its changes under global warming. In: Materials of 61st Ghertzen Memory Conference. Section of Geography and Interdisciplinary Sciences. 24–25 April 2008. TESSA, St. Petersburg, pp 141–143, in Russian
Regime Shifts in the Atmosphere and Their Relationship to Abrupt Ocean Changes Alfred M. Powell, Jr, Jianjun Xu, and Ming Chen
Abstract The ocean community has monitored abrupt climate changes or regime shifts in various fish species around the Pacific and Atlantic basins via fish stock and fish catch statistics. These regime shifts occur over relatively short intervals of 1–3 years, and appear to represent basin wide as well as ecosystem level changes that can last for many years. While regime shifts have been observed in the ocean through changes in physical and biological responses, their primary source has not been attributed to either atmospheric or oceanic forcing. Research results are discussed that make the case for the atmosphere being a key forcing for the abrupt regime shift changes. Also, a set of the independently identified ocean regime shifts are linked with abrupt changes in the atmosphere. Keywords Abrupt ocean change • Fish stock • El Nino/La Nina • Global waves
Introduction Regime shifts or abrupt climate changes have been identified in the literature for a number of years. The current basis for many of the regime shifts comes from the study of marine life primarily through the use of fish catch and fish stock statistics. The purpose of this short research paper was to identify key regime shift periods in the literature, and compare them to major environmental events and abrupt changes in fish populations thought to be due to a regime shift. Once the most likely regime shift periods have been identified, an analysis of the NCEP-NCAR and ERA-40 Reanalysis data sets will be performed looking for abrupt changes. The key question is whether the atmosphere plays a role in the regime shifts or whether the A.M. Powell, Jr (*) NOAA/NESDIS/STAR, Washington D.C, USA e-mail:
[email protected] J. Xu and M. Chen IM Systems Group Inc, Washington D.C, USA
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_17, © Springer Science+Business Media B.V. 2011
151
152
A.M. Powell, Jr et al.
abrupt shifts are a result of ocean processes. A comparison of the regime shifts identified in the fisheries literature and the atmospheric analysis will try to determine whether the atmosphere is a likely forcing factor in the regime shifts.
Data Regime Shift Data. The data consists of a compilation of regime shifts identified in the literature. Table 1 shows the compilation of the literature search and the years in which regime shifts are identified. Table 1 includes a grouped subjective summary of Overton’s et al. (2008) results (column 1) where his mathematically probable regime shift periods were grouped according to their similarity from all the results shown in the tables of his paper. This summary provided a guide for determining likely regime shift periods and became a benchmark for aiding the more formal literature search. Column 2 of Table 1 shows the specific climate regime shift years associated with abrupt changes in the fisheries and environmental data. Source references for each identified regime shift are provided. Since a number of references have suggested that the abrupt shifts were simply the result of El Nino and La Nina years column 3 contains the list of El Nino years from the Stormfax. com web site. The web site provides open access to the public for the El Nino and La Nina years for anyone that would like them and covers the period back to 1900. These years are consistent with the information on NOAA’s Climate Prediction Center (CPC) website. The information from the CPC is more detailed and does not list the El Nino years specifically but does show additional information at the monthly level. Also, a number of the CPC parameters used to assess El Nino periods only go back approximately 50 years on their website. For comparative purposes and simplicity, the Stormfax website was chosen as a source for the El Nino years for the following reasons (1) consistency in yearly dates to compare with yearly timeframes of the regime shifts, (2) the Stormfax data spans the period back to 1900, and (3) the yearly El Nino dates are openly available on a public web site. Using the yearly dates also simplifies the comparisons with regime shifts identified in the marine and environmental literature. Reanalysis Data. The monthly NCEP-NCAR (referred to as NCEP in the paper) reanalysis (Kalnay et al. 1996) is employed for the same period as the ERA-40 reanalysis data set. It should be noted that the reanalysis period of 1958–1978 has no satellite data. After 1978, satellite data was used in both the NCEP and ERA-40 reanalyses. The satellite data includes the Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS), the Microwave Sounding Unit (MSU), the High Resolution Infrared Radiation Sounder (HIRS) and the Stratospheric Sounding Unit (SSU) data. The Special Sensor Microwave/Imager (SSM/I) data was assimilated in this system from 1993. The NCEP reanalysis has 17 vertical layers that range from 10 hPa to the surface (1,000 hPa). The ERA-40 reanalysis extends to 1 hPa and is shorter in length than the NCEP reanalysis. The analysis of the two data sets will be
Regime Shifts in the Atmosphere and Their Relationship to Abrupt Ocean Changes
153
Table 1 Regime shift periods identified in the literature and their comparison with El Nino years. Shaded cells correspond with Minobe’s (1997) 50–70 year climate oscillation Regime shift periods and their comparison with El Nino periods Regime shift periods based Literature identified on Overtona analysis regime shift periods El Nino yearsm 1910–1912 1911–1912 1916–1917 1914–1915, 1918–1919 1924–1928 1925b 1923–1924, 1925–1926 1924–1925c,d,f 1934–1935 1930–1931, 1932–1933 1943–1943 1940–1941, 1941–1942 1946–1949 1947b 1946–1947 1947–1948c,f 1953–1955 1951–1952, 1953–1954 1957–1960 1957–1958f 1957–1958 1961–1966 1966–1968l 1963–1964,1965–1966 1969–1974 1969–1970, 1972–1973 1975–1980 1976a 1976–1977, 1977–1978 1977b,e,g 1976/1977c,d,f 1977e 1981–1985 1983–1984k 1982–1983 1987–1990 1989g 1986–1987 1989 8 1991–1995 1991–1992, 1992– 1993, 1994–1995 1996–1999 1998g 1997–1998 1998h,i,j 2000–2004 2004j 2002–2003, 2004–2005 2006–2006 2006–2007 Overland et al. 2008 Ebbesmeyer et al. 1991 c Mantua et al. 1997 d McGowan et al. 1998 e Minobe and Mantua 1999 f Chao et al. 2000 g Hare and Mantua 2000 h McFarlane et al. 2000 i McPhaden and Zhang 2004 j Behrenfeld et al. 2006 k Inque and Matsumoto 2007 l Baines and Folland 2007 m El Nino years from Stormfax.com Website (http://www.el-nino.com) a
b
p erformed over the matching periods and vertical domains between the ERA-40 and NCEP reanalyses. It should be noted that the regime shift identified near 1977–1978 has both supporters and detractors. The detractors claim the shift in this period was due to the addition of the satellite data in the reanalysis beginning in Nov 1978. The supporters claim there are ancillary independent data sources,
154
A.M. Powell, Jr et al.
such as the fish catch and stock numbers, which verify the regime shift is real and not simply caused by the addition of the satellite data alone.
Method According to Overland et al (2008), the definition of a regime shift is that it must occur in a short period of time and be stable for a period of years greater in length than the regime shift itself. For the regime shifts identified in Table 1, they appear to satisfy this criteria although the number of regime shifts identified (nine) tend to show a period of approximately 5–9 years between shifts with a regime shift typically occurring in 1–2 years. In addition, the approximate 50–70 year cycle identified by Minobe is highlighted by shading cells light gray. As can be seen, Minobe’s cycle is consistent with the regime shift observations. The regime shifts identified by Overton’s mathematical approach seems to correspond well with the El Nino years. This suggests that El Nino may play a strong role in global dynamics with its impacts extending beyond the Pacific region. It also suggests that the Overton analysis may have identified changes associated with El Nino more than with regime shifts. It should be noted that the regime shifts identified in the literature from biological, marine and environmental data do not always occur with each El Nino period or shift. This could be due to an incomplete literature search, a lack of published research in the potential regime shift gap periods or it may represent an actual physical phenomena with no changes in the periods in question. When the marine regime shifts occur El Nino seems to also to be a factor. For this analysis, we will use the regime shifts specifically identified in the marine and environmental literature as independently observed events. Even though the marine shifts are often thought of as occurring at the ecosystem level and have differing results in the eastern and western Pacific, our premise is that all of the marine and environmental shifts result from a more general global change in the atmosphere where it plays a key role in regime shifts in general. Our primary comparison will be with the regime shifts identified from biological, environmental and associated marine changes. Following the global premise, the NCEP and ERA-40 reanalyses were analyzed for indications of abrupt regime shifts. Our assumption was that the only likely source of abrupt climate regime shifts caused by the atmosphere would be a global change in the hemispheric wave pattern. If true, this could explain the differences in marine and environmental impacts from region to region yet be consistent with a global atmospheric change. Winds blowing on-shore versus off-shore could change the ocean pattern from upwelling to downwelling which could impact the feeding pattern of the marine life similar to the impacts El Nino/La Nina off the coast of South America for example. From El Nino/La Nina shifts, it has been documented that the fish populations change off the coast of South America and can lead to devasting effects on the region’s fishing industry. To search for global
Regime Shifts in the Atmosphere and Their Relationship to Abrupt Ocean Changes
155
changes, the two reanalyses were analyzed for global wave energy variations. The wave energy analysis was based on the temperature fields since the authors thought that any changes in the global wave energy would likely result from changes in the radiative balance which would be most easily observed as changes in the temperature fields. A Fourier analysis of the global wave amplitudes by wavenumber was performed by latitude band. Each band was 30 degrees in latitude representing the tropics, mid-latitudes and polar regions respectively. Figure 1 shows the results of the reconstructed global wave amplitude pattern for wavenumbers 1–5 for each latitude band with altitude.
Description of the Results Our analysis shows that the wave energy in the tropics and mid-latitudes is generally small at all attitudes. Figure 1 shows the pattern in the northern hemisphere for both the NCEP and ERA-40 reanalyses. The northern hemisphere has the most land based observations and is thought to best represent changes throughout the 40 year period. The biggest contributor in all three northern hemisphere latitude bands appears to be in the Earth’s boundary layer where waves are generated that propagate upward in the atmosphere. The boundary layer is very shallow and visible in each latitude band below about 800 hPa according to this analysis. Outside of the boundary layer effects, there are few noticeable global wave energy changes in the tropics and mid-latitudes. The key surprising factor from this analysis is the strength of the change in the stratospheric global wave energy in the polar region. Figure 1 for both the NCEP and ERA-40 reanalyses shows essentially the same features even though the models are different, the individual physics modules are different, and two different approaches for assimilating the satellite data were used (NCEP assimilated the derived temperature profiles and the ERA-40 assimilated the satellite radiances directly). This implies that both reanalyzes have captured a real physical change in the data and it can be described in terms of our fundamental knowledge of how our models and the atmosphere work. Even though the details of the wave energy analysis vary to some degree between the NCEP and ERA analysis, the core features are very clear. The periods of strong and weak wave energies (amplitudes) are essentially the same. The strong and weak wave amplitudes are shown above approximately 100 hPa in the polar zone with the higher energy waves shown as darker and the weaker waves shown as lighter in shade. If one draws a line down the graph in the periods of the lowest wave energy, a very distinct pattern arises and the pattern is very similar to the abrupt climate shifts from the marine and environmental data and Overton’s analysis in Table 1. The low global wave amplitude periods seem to correspond with the periods where regime shifts have been identified. Also, it is clear that the El Nino years also seem to correspond with the global wave energy shifts in the stratosphere. This raises the question of whether wave energy changes may also impact El Nino events.
Polar Zone
Mid-Lat Zone
Eq Zone
Fig. 1 Global Wave Energy Power Spectra Using Wavenumbers 1–5 by Altitude (pressure level in hPa). NCEP Reanalysis (left) and the ERA-40 reanalysis (right) show strong global wave energy changes in polar zone above 100 hPa with comparatively small changes elsewhere. The dashed lines represent identified regime shift periods that correspond with periods or weak global wave energy
156 A.M. Powell, Jr et al.
Regime Shifts in the Atmosphere and Their Relationship to Abrupt Ocean Changes
157
Discussion The periods of the lowest wave energy are approximately 1962/1963, 1971/1972, 1978/1979, 1989, 1994 and 1998 (on the edge of the plot). These match the abrupt climate shifts from the marine/environmental literature, Overton’s mathematical analysis and also corresponds with the El Nino periods remarkably well. What could this mean? The authors were expecting to find wave energy changes in the troposphere and in the boundary layer where the changing global temperature gradients would reflect wind changes that would drive changes in the ocean circulation. However, it is hard to distinguish any change in the boundary layer, even in the polar regions, from this analysis. In fact, there does not appear to be a significant change in wave energy below 100 hPa. In addition, there is little wave energy effect in the mid-latitudes where most meteorologists would have thought that frontal boundaries and the tracking of high and low pressure systems might influence, in a cumulative fashion, the stress on the ocean surface that could drive large scale changes. These surprising results suggest that the scientific community should look at the polar regions as a major influence on the global wave energy and possibly the ocean forcing pattern. This analysis raises new questions. How do the wave energy changes in the lower polar stratosphere impact the ocean surface or provide forcing (or the lack of forcing) to generate the regime shifts? The answer to this and related questions will be part of the follow-on research that will attempt to identify how the global stratospheric wave amplitude changes can cause or influence abrupt climate shifts. This initial cursory analysis still leaves open the question as to whether the atmosphere plays a role in the regime shifts since a likely mechanism for stratospheric forcing cannot be provided at this time.
Conclusions In summary, this analysis demonstrates a tantalizing and unexpected result that the wave energy changes in the lower stratosphere (between 10 and 100 hPa in this analysis) appear to be associated with the abrupt climate shifts identified in the marine literature, and mathematical analyses for potentially determining climate shifts. The lowest global wave amplitude periods also appear to correspond with the El Nino phases. This raises the question as to how this interaction may occur between the lower stratosphere and the surface. Also, the unexpected result that the tropics and mid-latitudes do not seem to have a significant role in the wave energy changes is also perplexing given the mid-latitudes are where the temperature gradients and consequently the winds are traditionally thought of having the greatest variability. This surprising result compounds the mystery since the lower atmosphere (troposphere) does not seem to show any signs of a significant change in wave amplitude or interaction with the surface. Given that the abrupt regime shifts are associated with the periods of lowest stratospheric wave energy, it suggests that the regime shifts may not be caused by
158
A.M. Powell, Jr et al.
a positive change in the forcing but by a lack of forcing instead. Could the El Nino phases and marine biological shifts be caused by turning down the volume on the stratospheric wave amplitude? This analysis implies that may be the case. Our future research will attempt to address how the wave energy may propagate between the stratosphere and the surface of the earth. A mechanism to explain this peculiar result must be found in order to understand how regime shifts could be created that in turn will effect marine and ocean activity.
References Baines PG, Folland CK (2007) Evidence for a Rapid Global Climate Shift across the Late 1960s. Journal of Climate 20:2721–2744. doi:10.1175/JCLI4177.1 Behrenfeld, M.J., R.T. O’Malley, D.A. Siegel, C.R. McClain, J.L. Sarmiento, G.C. Feldman, A.J. Milligan, P.G. Falkowski, R.M. Letelier and E.M. Boss, 7 December 2006: Climate-driven trends in contemporary ocean productivity. Nature, 144. doi:10.1038/nature05317 Chao Y, Ghil M, McWilliams JC (2000) Pacific interdecadal variability in this century’s sea surface temperatures. Geophys Res Letters 27:2261–2264 Ebbesmeyer, C.C., Cayan, D.R. McLain, F.H. Nichols, D.H. Peterson and K.T. Redmond, 1991: 1976 step in the Pacific climate: forty environmental changes between 1968-75 and 1977-1984. In: Proc. 7th Ann. Pacific Climate Workshop, California Dept of Water Resources, Interagency Ecol. Stud. Prog. Report 26. Hare SR, Mantua NJ (2000) Empirical evidence for North Pacific regime shifts in 1977 and 1989. Progress in Oceanography 47:103–145 Inque T, Matsumoto J (2007) Abrupt Climate Changes Observed in Late August over Central Japan between 1983 and 1984. Journal of Climate 20:4957–4967. doi:10.1175/JCLI4217.1 Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M, Saha S, White G, Woollen J, Zhu Y, Leetmaa A, Reynolds B, Chelliah M, Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C, Wang J, Jenne R, Joseph D (1996) The NCEP/NCAR 40-Year Reanalysis Project. Bull Amer Meteor Soc 77:437–472 Mantua N, Hare S, Zhang Y, Wallace J, Francis R (1997) A Pacific Interdecadal Climate Oscillation with impacts on salmon production. Bull Amer Meteor Soc 78:1069–1079 McFarlane, G.A., King J.R., Beamish R.J. (2000) Have there been recent changes in climate? Ask the Fish. Progress in Oceanography, 47:147–169 McGowan JA, Cayan DR, Dorman LR (1998) Climate-Ocean variability and ecosystem response in the Northeast Pacific. Science 281:210–217. doi:10.1126/science.281.5374.210 McPhaden MJ, Zhang D (2004) Pacific Ocean circulation rebounds. Geophys Res Letters 31:L18301. doi:10.1029/2004GL020727 Minobe S (1997) A 50-70 year climatic oscillation over the North Pacific and North America. Geophys Res Letters, 24 No. 6:683–686 Minobe S, Mantua N (1999) Interdecadal modulation of interannual atmospheric and oceanic variability over the North Pacific. Progress in Oceanography 43:163–192 Overland J., S. Rodionov, S. Minobe, and N. Bond, 2008: North Pacific regime shifts: Definitions, issues and recent transitions. Progress in Oceanography
Glacier Degradation from GIS and Remote Sensing Data Azamat Tynybekov
Abstract Application of the multilevel geoinformation system is focused currently on estimation of danger from mudflow and monitoring the mudflow processes. Within the limits of southern coast of Lake Issyk-Kul (from Ton in the west up to Kyzyl-Suu in the east) the analysis of glaciers and the potential for forming mudflow risk has been performed using remote sensing and GIS technologies. Features of freshet and mudflow dangers for the Tone valley differs other regions. Also quantitative assessment of linier and aerial changing is conducted. Sustainable reduction of glaciation and the rate of degradation in various volleys are estimated. Keywords Glaciers area • Mudflow danger • GIS • Remote sensing • Issyk-Kul
Introduction Changes in Tien-Shan glaciers always drew attention of researchers. In the majority of cases glaciers’ contraction or expansion were analyzed for volleys. Comprehensive analysis of glaciers fluctuations has been executed recently for glaciers zone of eastern Terskei Ala-Too ridge which is characterized by various orogenic-climatic conditions That was a survey of five rivers basins glaciers in relation to both axial line of Terskei Ala-Too ridge and to its center (map scale 1:25,000). Once a glacier map was determined, further displacement of the glacier front the map was monitored. Naturally, this method is not accurate. In such case the condition of a glacier recession was defined on glaciers-geomorphologic attributes (flat or abrupt glaciers tongue, a condition of a hydronetwork, character of moraine accumulation at the end of ice and color of a moraine). This paper is addressing the received results.
A. Tynybekov (*) Kyrghyz Russia Slavonic University, Bishkek, Kyrgyz Republic e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_18, © Springer Science+Business Media B.V. 2011
159
160
A. Tynybekov
Area of Investigation A valley of Chon-Kyzyl-Suu cut through the Terskei Ala-Too slopes of ridge in its central part with 43.3 km2 congelation area. The glaciers are confined by the main ridge and occupy 3.5 km2. The river Chichkan forms the drain on a slope of the spur departing from the main watershed. There is only one steadily reduced glacier here, but traces of the former congelation are presented everywhere. River basin of Akterek is located down on the advanced ridge with the 4,600–4,700 m maximum elevation. The length of the basic glacier is 3.9 km, and the congelation area is 6.7 km2. A valley of river Tosor begins from the back side on the western Terskei Ala-Too ridge. It widely ramified and reaches an axial part. Glaciers cover area of 20.1 km2 but their length seldom reaches 2.0 km. Congelation has been degrading, but because of high bedding rates recession is insignificant, although the big excess of lateral moraines over a surface of clean ice is seen. Quantitative assessment of linear and aerial changes is presented in Table 1 which indicates a steady reduction in the glaciation. The rates of degradation in various valleys are different. From 22 glacier’s surveys during 1981–1995, all of them receded: seven by more than 90 m, six between 60 and 89 m and nine by between 25 and 59 m. Minimal recession occurred in the Valley of Tosor River (compare to 1966). Therefore their conditions are assessed as stationary. The recession rate is reduced with the height of the location. The Kara-Batkak glacier recedes with average annual speed 7.8 m. The surface thinning for the period of survey was 14.6 m. Such loss of ice for a glacier is more destructive process, than the recession. For the analysis of glaciations we used 1:250,000 topographical map from the aerial photography in 1963 and June, 2001 space picture from NASA. The layer on the map was digitized using ArcGIS-8.3 software and ENVI 3.5 software for grid space pictures. Figure 1 shows the layer of 23 studied glaciers. Scientists have been reporting reduction of glaciers in the world since the 1970s and sharp reduction since the 1980s. The example of such research has been carried out by Kyrgyz–Switzerland project of river basin Sokuluk on the northern slope of ridge Kyrgyz Ala-Too. Figure 2 shows changes in those glaciers’ area since 1963. Speed of glaciers’ reduction increased twice, from 0.6% during 1963–1986 to 1.3% during 1987–2000. Similar dynamics was observed on glaciers of the ridge Terskei Ala-Too. The current map of glaciers congelation area for the river basin Tone (northern slope of Terskai-Ala-Too ridge) is shown in Fig. 3. The area of a glaciation for the last 38 years has decreased nearly 30%. Between 1963 and 1986, the glaciers area decreased 13.3%, and between 1987 and 2000 – 17.1%. Eight glaciers have completely disappeared during 1963–2000. They belonged to the I class (<0.5 km2). The speed of the I class glaciers disappearance increased considerably from 9.1% in 1963–1986 to 41.5%, in 1987–2000.
Glacier Degradation from GIS and Remote Sensing Data
161
Table 1 Glaciers’ linear and aerial change Glaciers No. Change Length, m on catalog area, km2 River basin Chon–Kyzyl-Suu 244 90–100
Glaciers end altitude on catalog
Size of image
0.05
3,620
Image year–1981 249 250 251 254 255 256 257 260 262 263 265 267 269 270 271
0.07 0.05 0.05 0.03 0.03 0.03 0.03 0.04 0.02 0.05 0.12 0.34 0.03 0.06 0.08
3,340 3,780 3,780 3,450 3,510 3,410 3,460 3,460 3,710 3,680 3,390 3,230 3,720 3,800 3,740
0.09
3,650
50–60
0.06
3,720
70
0.03
3,520
70–80
0.02
3,750
20–30
0.01
3,750
20–30
0.05
3,650
90 60–80 60–80 50–60 50–60 50 80 50–60 40–50 40–50 100–120 90–100 80–110 80–110 80–110
River basin Kichine–Kyzyl-Suu 233 70–80 1981 River basin Chychkan 178 River basin Ak-Terek 175 1981 174 River basin Tosor 77 1966 78
Therefore, glaciers with area less than 0.5 km melted more intensively, than the glaciers of other classes. The GIS technology was also used for the analysis of mudflow risk of southern coast of Issyk-Kul lake. This area was divided into areas by risk intensity. Following Tynybekov et al. 2001, there are seven levels of risk category’s intensity: Comprehensible (1–2 risk category); Partially comprehensible (3–4 risk category) and Not comprehensible (5–7 risk category). Figure 4 shows a map with flood and mudflow dangers. The moraine glacial lakes Tujuk-Ter, Koltor, Korumdu in the upper land belong to the first danger category but in the lowlands as well as in the coastal zone, the category one is absent. Based on these assessments some risk reduction recommendations were developed.
Area, km2
Fig. 1 Layers of the glaciation of the Tone river basin overlay on 2001 NASA space picture 20 18 16 14 12 10 8 6 4 2 0
18,293 15,435
7,652 5,779
1963
14,265
6,953 4,067 5,102
4,465
1986
2000
Year
Fig. 2 Glaciers area dynamics (1963–2000), northern slope of ridge Kyrgyz Ala-Too
Fig. 3 Congelation areas of river basin Tone glaciers in the northern slope of Terskai-Ala-Too ridge (relative to 1963)
Glacier Degradation from GIS and Remote Sensing Data
163
Fig. 4 The synthetic map of danger resulted for the considered area
References Tynybekov AK, Torgoev IA, Alyoshin JUG (2001) Assessment of risks of natural processes of southern coast of lake Issyk-Kul. The International conference on human health and strategy of the environment: The Program of a new millennium, Bishkek, May, pp 14–16 Tynybekov AK, Kulenbekov ZhE, Aliev MS (2006) Use GIS technologies for risk assessment of mudflow dangers. The Bulletin of the Kyrgyz State Technical University, Bishkek, Vol. 2, pp 419–423
ENSO Impact on Vegetation Felix Kogan
Abstract This paper examines the 1981–1997 association between monthly SST anomalies in the 3.4 tropical Pacific and vegetation health (VH) indices for every 16 km2 pixel of the world. The VH indices are represented by the Vegetation condition (VCI), Temperature condition (TCI), and Vegetation Health (VHI) indices. VCI determines moisture conditions, TCI – thermal conditions and VHI – the total vegetation health. Two types of responses were identified for boreal winter: ecosystems of northern South America, southern Africa, and Southeast Asia experienced severe moisture and thermal stress during El Niño and favorable conditions during La Niña years. In Argentina and the Horn of Africa the response was opposite. One of the most interesting results this paper shows are related to an advanced warnings of ENSO impacts. The eastern Brazil is sensitive to ENSO as early as in the spring (March–May) of the year the ENSO is starting its development. Keywords Land ecosystems • Vegetation Health • Drought • El Niño • La Niña • AVHRR data • Lag correlation
Introduction ENSO is one of the principal climate forcings affecting the weather around the world with the period 3–7 years (IPCC 2001). During the ENSO events some world areas experience a combination of either dry and hot or wet and cool, or only wet/dry or hot/cool conditions. Traditionally, precipitation and temperature measurements were used to identify areas and intensity of ENSO impacts (Ropelewski and Halpert 1989). Since weather station network is not dense enough and stations are not distributed uniformly, especially in marginal climate zones, satellite data in form of the Normalized Difference Vegetation Index (NDVI) were used to estimate ENSO
F. Kogan () NOAA/NESDIS Center for Satellite Application and Research (STAR), Washington DC, USA e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_19, © Springer Science+Business Media B.V. 2011
165
166
F. Kogan
impacts on vegetation (Myneni et al 1996; Kogan 1998). However, using NDVI alone is not sufficient since development and productivity of ecosystems are highly temperature-dependent (Kogan 1997, 2000). This paper presents application of both NDVI and thermal index for characterization of vegetation response to ENSO. The new satellite-based approach assessing vegetation health was used for this goal.
Data and Method Measurements of the Advanced Very High Resolution Radiometer (AVHRR) on board NOAA afternoon polar-orbiting satellites were used in this study. The data were collected from the NOAA’s Global Vegetation Index (GVI) data set during 1981 through 2004. Data resolution was: spatial – 4 km, sampled to 16 km and temporal 1 day sampled to 7-day composite (Kidwell 1997). The GVI counts in the visible (VIS, 0.58–0.68 mm), near infrared (NIR, 0.72–1.1 mm), and infrared (IR, 10.3–11.3 mm, Ch4) spectral bands were processed (applying pre- and post-launch calibration coefficients) to convert VIS and NIR counts to reflectances and to calculate the NDVI = ((NIR − VIS)/(VIS + NIR)). The Ch4 counts were converted to brightness (radiative) temperature (BT) and corrected for non-linear behavior of AVHRR sensor (Kidwell 1997). NDVI and BT were used to produce Vegetation Health (VH) indices. First, interannual and intra-annual noise was removed from the data by smoothing the time series with median filter; second, the seasonal cycle was approximated; third, the 22-year climatology of the smoothed NDVI and BT was estimated and fourth, three VH indices (Kogan 1997), Vegetation Condition (VCI), Temperature Condition (TCI) and Vegetation Health (VHI) were approximated as: VCI = 100 ∗ (NDVI − NDVI min ) (NDVI max − NDVI min ) TCI = 100 ∗ (BTmax − BT ) (BTmax − BTmin ) VHI = a ∗ VCI + (1 − a ) ∗ TCI, where NDVI, NDVImax and NDVImin (BT, BTmax and BTtmin) are the smoothed weekly NDVI (BT) and their 1981–2004 absolute maximum and minimum (climatology), respectively; a is a coefficient quantifying a share of VCI and TCI contribution into the VHI. Since this share is not known for a specific location and time of the year it was assumed that the share is equal and a = 0.5. The VH indices change from zero quantifying severe vegetation stress to 100 quantifying favorable conditions; VH between 40 and 60 represents average or normal conditions. Decrease in VH from 40 to 0 signals about intensification of vegetation stress and on the opposite side of the scale, VH increase from 60 to 100 indicates improvement of vegetation condition/health. Monthly VH indices were further calculated to match them with sea surface temperature (SST) data. SST for the 3.4 ENSO area in the tropical Pacific were collected from the improved SST analysis data set (Reynolds and Smith 1994). The new technique of data preparation blends both in situ measurements (ship and buoy) and satellite-derived
ENSO Impact on Vegetation
167
SST. The later was spatially corrected based on in situ measurements. The blended SST was optimally interpolated on a 1° latitude/longitude spatial grid. The period of observation matched with vegetation data and included monthly SST values and anomalies deviation from 1981–2004 mean) for each 1° by 1°of the global ocean. Sensitivity of regional vegetation health to ENSO was studied by correlating VH indices for each 16 km pixel of the world with regional (the 3.4 tropical Pacific) average monthly SST anomaly. Since both parameters were expressed as a deviation from climatology (monthly mean for SST and weekly MAX–MIN for VH) the results showed the influence of intensity and duration of ENSO on changes in vegetation conditions due to switch from one type of the ENSO cycle to the opposite one. We correlated every pixel’s monthly VCI, TCI and VHI with the monthly SST anomaly for ENSO years. Since ENSO appearance and impacts are seasons specific (Ropelewski and Halpert 1989), the VT/SST correlations were investigated during boreal winter (December–February). The period of observation was 1982–1997 excluding 7-month (August 1994 through February 1995) when NOAA-11 afternoon satellite was inoperative.
Results and Discussion The correlation of three VH indices (VCI, TCI and VHI) for each 16 km pixel with SST anomaly is presented on Fig. 1. These results: (a) delineate the land area of the ENSO impacts, (b) identify critical periods vegetation sensitivity to ENSO and (c) investigate the contribution of moisture and thermal components into the total vegetation health response of land ecosystems to ENSO. Two types of VT/SST correlation patterns are identified: positive (dark green/blue) and negative (dark red/ brown/grey). If the correlation is positive, it means that cooler water (below multiyear mean or normal) in the tropical Pacific (La Niña case) triggers vegetation stress (VT below 40), and warmer water (El Niño) stimulates favorable vegetation conditions (VT larger than 60). Oppositely, for the negative VT/SST correlation, healthy/unhealthy vegetation conditions are associated with below normal SST (La Niña case) and above normal SST (El Niño case), respectively. The VT/SST correlation coefficients (R) change between −0.58 and +0.56. They were calculated for 40 pairs of observations and are significant at 1% level when R ³ |0.228|, 5% when |0.174| < R < |0.228|, and <5% when R£|0.174| (Snedecor 1965). As seen in Fig. 1, there are several sensitive to ENSO land areas during December–February and most of them are located in the Southern Hemisphere. High sensitivity in winter is in agreement with the highest phase of ENSO development in the tropical Pacific (maximum SST anomalies) and major impacts in the tropics (Ropelewski and Halpert 1989). Both moisture (VCI) and thermal (TCI) condition indices showed similar sensitive areas in southern Africa and western Australia with relatively strong Index/SST negative correlation (0.40–0.58, grey, brown and red color). This indicates that the areas experience vegetation stress
168
F. Kogan
Fig. 1 Correlation of VH indices (VCI, TCI, VHI) with SST anomaly in 3.4 tropical Pacific during the ENSO years in 1982–1997 (Color image is in Appendix 3)
d uring El Niño (SST above normal) and favorable conditions during La Niña (SST below normal). Another two areas where the ENSO impact is clearly identified but with positive correlation for both: moisture (VCI) and thermal (TCI) conditions are northern Argentina (R = 0.48–0.56) and the Horn of Africa (0.38–0.47). The positive correlation (green and blue color) indicates that El Niño (SST above normal), opposite to the previous two areas, triggers favorable vegetation condition and La Niña (below normal SST) triggers vegetation stress. There are also a few other areas, as seen in Fig. 1, where only one of the indices shows the ENSO impacts. For example, the eastern portion of the USA is sensitive to ENSO-triggered vegetation stress in boreal winter. During El Niño (warm phase), favorable conditions are developed due to wetter weather and during La Niña, dryer winter conditions are developed. The indicated area is less sensitive to thermal (TCI) conditions although the area north of Great Lakes shows some sensitivity, which is in agreement with Wang et al. 2010, which showed that La Niña
ENSO Impact on Vegetation
169
stimulates cooler weather and more ice on the surface of the Lakes. Another area is northern Brazil, where thermal vegetation conditions (TCI) show stronger negative correlation (0.40–0.58, grey, brown and red color) with SST anomaly: La Niña stimulates cooler land conditions and El Niño – dryer conditions. Southeast Asia has similar sensitivity although the correlation is less consistent. The VHI produced from a linear combination of VCI and TCI (see equations) adopts the features of their relationship with SST anomaly. All regions discussed above preserve their sensitivity. Positive correlation for northern Argentina and the Horn of Africa (R = 0.36–0.58) and negative correlation (0.37–0.57) for northern Brazil, eastern Australia and southeastern Asia. Moreover, in some regions the correlation becomes stronger, larger and more concentrated (Brazil, Australia and Southeast Asia) than for individual indices. Another discussion we are presenting in this paper is regarding possibility of an advance warning for ENSO impacts on land ecosystems. Figure 2 shows VHI’s lag correlation up to nine months. The lag0 repeats the VHI diagram from Fig. 1 (for convenience of comparison), the lags 3, 6 and 9 investigate the impact of SST anomaly in the 3.4 tropical Pacific during September–November, June–August and March–May, respectively, on sensitivity of VHI from the current year December through next year’s February. It is known that ENSO starts much earlier than their maximum manifestation during boreal winter (IPCC 2001; Suplee 1999; Gershunov et al. 1999; Trenberth 1997). Therefore, investigation of the advance warning of vegetation stress/no stress would be very important for making important decisions on food security, forest fires, vector deceases, agricultural losses and others. As seen in Fig 2, some sensitive to ENSO areas discussed above (Fig. 1) preserve relatively strong correlation between VH indices and SST anomaly in the 3.4 tropical Pacific. However, for most of them the correlation either disappears or attenuated
Fig. 2 Lag correlation of VHI for December–February with SST anomaly for December–February (lag0), September–November (lag3), June–August (lag6) and March–May (lag9). Red box indicates 3.4 area where SST data were collected. (Color image is in Appendix 4)
170
F. Kogan
gradually or changes the location. The detailed analysis indicates that December–February VHI correlates with lag3 (September–November) ENSO similar (to some extend) way like lag0 and differently compared to lag6 and lag9. We should emphasize strong correlation in the northern Argentina although the area becomes smaller. However, there are important exceptions, which are discussed further. At the background of better match of correlation with lag3 (compared to lag6 and lag9), we should indicate that such area as eastern Australia (including southeast) and the Horn of Africa became no sensitive to SST anomaly during September–November. Also, southern Africa becomes less sensitive, especially in the far south (had R = 0.58 for lag0). Smaller area became sensitive in the northern Brazil, although VHI in the eastern part still correlates strongly with September–November SST anomaly. Also, the Southeast Asia still indicates strong sensitivity. We avoid discussion of the correlation with the desert surfaces (Sahara, Arabia, southeastern USA etc.) since NDVI and VHI data are not reliable, especially in winter when solar angle is quite low. Sensitive ENSO areas in lag0 and lag3 become less sensitive for lag6 and lag9. Meanwhile, Brazil continued to show high sensitivity of vegetation health (VH index) to ENSO. Both lag6 and lag9 of VHI and SST relationship show strong correlation (<−0.55, grey area). However, a few specific feature can be observed distinguishing these two lags from previous: (a) the area of the impact shifted from northeast to the east-central location and far east for lag9, (b) the area of strong ENSO impact enlarged compared to lag3 and became more consistent (stable color).
Conclusions For the first time, sensitivities of all global land ecosystems toward ENSO events have been evaluated numerically based on vegetation response. The response was based on a combined contribution of moisture-and thermal-derived conditions. The sensitivity was estimated for ENSO years only. In general, sensitive to ENSO areas identified from satellite showed similar to weather-based (precipitation and temperature) features (Ropelewski and Halpert 1989; Halpert and Ropelewski 1992). However, satellite data delineate more precisely the affected areas and the period of the highest sensitivity. Besides, compared to weather data, satellite-based indices provide a combined contribution of moisture and thermal conditions based on cumulative vegetation response during the growing season. Basically, it was emphasized that in boreal winter, ecosystems of northern South America, southern Africa, Southeast Asia and a part of Australia experienced dry and hot weather during El Niño and wet and cool during La Niña years; in Argentina and the Horn of Africa the moisture and thermal regime was opposite. The advantage of this study is also in the determining temperature contribution. One of the most interesting results this paper shows are related to an advanced warnings of ENSO impacts. Not many areas shows the advanced impacts but such area as eastern Brazil is sensitive to ENSO as early as the spring (March–May) of the year the ENSO is starting its development.
ENSO Impact on Vegetation
171
References Gershunov A, Barnet TP, Cayan DR (1999) North Pacific interdecadal oscillation seen as factor in ENSO-related North American climate anomalies. Eos, Trans Am Geophys Union 80(3):25–30 Halpert MS, Ropelewski CF (1992) Surface temperature patterns associated with the Southern Oscillation. J Climate 5:577–593 IPCC (2001) Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Houghton JT, Ding Y, Griggs DJ, Noguer M, van der Linden PJ, Dai X, Maskell K, Johnson CA (eds) Cambridge University Press, Cambridge, United Kingdom and New York, USA, 881 pp Kidwell KB (1997) Global vegetation index user’s guide. Department of Commerce, NOAA/ NESDIS, National Climate Data Center, Washington D.C., 52 Kogan FN (1997) Global drought watch from space. Bull Am Meteorol Soc 78:621–636 Kogan FN (1998) A typical pattern of vegetation conditions in southern Africa during El Niño years detected from AVHRR data using three-channel numerical index. Int J Remote Sens 18:3689–3695 Kogan FN (2000) Satellite-observed sensitivity of world land ecosystems to El Nino/La Nina. Remote Sens Environ 74:445–462 Kogan FN (2001) World droughts in the new millennium from AVHRR-based Vegetation Health Indices. Eos, Trans Am Geophys Union 83(48, 26 November): 557–564 Myneni RB, Los SO, Tucker CJ (1996) Satellite-based identification of linked vegetation index and sea surface temperature anomaly areas from 1982–1990 for Africa, Australia and South America. Geophys Res Lett 23:729–732 Reynolds RW, Smith TM (1994) Improved global sea surface temperature analysis using optimal interpolation. J Climate 6:929–948 Ropelewski CF, Halpert MS (1989) Precipitation patterns associated with the high index phase of the southern oscillation. J Climate 2:268–284 Snedecor GW (1965) Statistical methods. The Iowa State University Press, Ames, IA, 534 p Suplee C (1999) El Niño/La Niña. National Geogr 195(March) 73–95 Trenberth KE (1997) Short-term climate variations: recent accomplishments and issues for future progress. Bull Am Meteorol Soc 78:1081–1096 Wang J, Bai X, Leshkevich G, Colton M, Glites A, Lofgren B (2010) Severe ice cover on great lakes during winter 2008–2009. Eos, Trans Am Geophys Union 91(5) 2 Feb, 41–42
Part IV
Marine Ecosystem, Land Cover, Atmosphere & Anthropogenic Activities
Bio-climatic Potential of Russia and Climate Change Alexander Kleschenko
Abstract The concept of bio-climatic potential of Russia and some neighboring countries has been developed. Components of the bio-climatic potential and their transformations under the climate change scenario are presented in this paper. A few adaptation measures for some regions of Russia are proposed. Keywords Bio-climatic potential • Climate change • Adaptation measures
Introduction Russia considers agriculture as the main source of freedom. Therefore, the investigation of bio-climatic potential (BCP) of Russian agriculture has a very important meaning. The BCP studies began in the pre-revolutionary Russia and continued through the USSR time and current eras. In the present climate changes scenarios, the interest to this problem become more acute. The BCP project was requested by the Russian Ministry of Agriculture in 2005 and during the next 3 years, three books were issued presenting the main results. These results and recommended adaptation measures are presented in this paper.
The Concept of BCP In Russia and USSR, the BCP was studied for the past 100 years. It have been shown that agricultural productivity depended on soil fertility, climate, weather, cultivated crops, agricultural technology and other factors. Timirjazev paid attention to solar radiation, A. Kleschenko () National Institute for Agricultural Meteorology, Roshydromet, Obninsk, Russia e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_20, © Springer Science+Business Media B.V. 2011
175
176
A. Kleschenko
indicating that plants use only 2% of the incoming energy; if this amount could be increased by only 1% then plant productivity might increase considerably. Dokuchaev explained some features of the BCB combining soil water and heat parameters. Visotsky gave scientific foundation of crop reaction to weather and climate. Koloskov was the first who introduced the BCB in the 1930s . His map shown in Fig. 1, provides BCB assessments based on crops, soils, radiation and weather parameters. Figure 2 shows one of the parameters (growing season precipitation) used for BCB. As seen, growing season precipitation is extremely variable. Similar results were obtained for other environmental parameters (radiation, temperature, etc.). For example, for temperature, different thresholds were used (above 5°C, 10°C and other) to characterize crop growth and development during warm season and cold conditions for winter crops. Table 1 shows estimated losses of agricultural production due to hazardous weather (drought, frost, flood, etc.) and decreases. Dynamic model approach was further used for simulation of crop growth and development. Dry biomass during the vegetation period was used as the BCB criterion. The formed biomass in a period characterizes that period. Table 2 shows estimated BCP for different economic regions of Russia. Indices in the heading of the table have the following meaning: 0 – BCB for climate- provided moisture and current nitrogen application, W –sufficient moisture, N – sufficient fertilizers, WN – sufficient both moisture and fertilizer. As seen, all economic regions of Russia suffer from a lack both moisture and chemical elements.
Fig. 1 The agricultural productivity of the country according to Dr. Koloskov
Bio-climatic Potential of Russia and Climate Change
<150
151-200
201-250
251-300
401- 450
451-500
501-550
>550
177
301-350
351-400
Fig. 2 Total April to October precipitation (mm) Table 1 Probability (P%) of deceases and crop losses (C, ton per hectare) Crop Winter Winter Spring wheat rye wheat Oats P C P C P C P Name of region Moscow 61 0.34 59 0.34 44 0.59 47 Vironezh 71 0.57 85 0.46 76 0.54 82 Kursk 85 0.51 79 0.33 79 0.54 85 Saratov 82 0.63 79 0.45 91 0.68 85 Rostov 78 0.41 81 0.37 78 0.47 75 Krasnodar 69 0.31 75 0.24 66 0.28 63
C 0.57 0.71 0.54 0.73 0.56 0.29
Barley P C 47 0.46 91 0.83 88 0.70 88 0.80 75 0.48 72 0.33
In Russia, weather affects considerably the harvest. Table 3 compares the contribution of different components into harvest. Following the presented numbers, one should say that agricultural technology and weather are the most important BCP components. In some ecosystems, weather contributions account for up to 50% of harvest variation. Table 4 shows climate and crop contribution into the BCP in Russia and in several European countries. As seen in Table 4, for the majority of countries, BCP parameters exceed the corresponding numbers typical for Russia. The following relationship provides a qualitative estimation of resources utilization (P) P = Y/bcp * 100%, where Y – quantifies harvest.
178
A. Kleschenko Table 2 BCB for the present climate, economic regions of Russia Ton per hectare Economic region BCP0 BCPw BCPN BCPWN North 4.7 10.6 4.8 10.6 North-West 5.5 11.8 5.5 11.8 Kaliningrad 6.0 14.9 6.1 11.8 Central 5.2 12.9 5.7 13.2 Volga – Vjatka 4.6 11.8 5.3 12.4 Central – Black Soil 6.2 12.5 9.2 14.9 Povolzhje, North 5.0 11.3 8.0 14.0 Povolzhje, South 3.7 5.8 9.3 15.9 North Caucasus 3.7 10.1 12.2 17.2 Ural 5.5 10.6 7.1 12.2 West Siberian 5.8 10.7 7.3 11.8 East Siberian 5.4 10.4 6.2 10.6 Far East 5.7 12.2 6.0 12.3 Russia in general 5.3 10.8 7.3 13.2
Table 3 Influence of different factors on agricultural crops (%) Level of Weather Zone Fertilizer improvement conditions North of boreal forest 34.4 21.3 44.3 Middle of boreal forest 36.7 23.5 39.8 South of boreal forest 30.5 20.1 49.4 Forest and prairie 23.3 35.8 40.9 Prairie 12.6 44.3 43.1 Dry prairie 10.4 30.5 59.1
Following the results presented in Tables 4 and 5, it is possible to emphasize that Russian agriculture is hard to conduct. Table 5 also shows that some countries such as Great Britain, Germany and France used 35–45% of the available resources, while Russia use only 15% having a considerable potential. Based on Fig. 3, we estimated that Russian agriculture will be affected by the climate changes. First, the duration of the vegetation period in Russia will increase by 16 days; in some regions, the length will be less intensive (5–10 days), and in some will not change. Second, the frost free period will not change much. Third, southern Russia has already experienced considerable dryness with frequent droughts; in the future, the dryness will persist and droughts will be more frequent and intensive. Table 6 provides an assessment of climate change-related crop productivity change based on GFDL (USA) scenarios. As seen, Russian agriculture would be favored by the anticipated climate change scenarios. Similar results were obtained using other scenarios recommended by the IPCC.
Bio-climatic Potential of Russia and Climate Change
179
Table 4 The BCP in Russia and selected European countries Moisture coefficient Sum of Geographic Veg. Precept. temperature, KW region period day mm T > 10° GTK Finland 150 280 1,369 1.13 1.57 Great Britain 288 662 2,208 1.22 1.56 Germany 239 509 2,614 0.90 1.43 France 301 600 3,237 0.84 1.22 Italy 335 781 4,761 0.73 1.26 Poland 208 432 2,377 0.93 1.53 Hungary 245 470 3,336 0.61 1.04 Russia 176 302 2,270 0.66 1.20
Table 5 Utilization of resources (%)
BCP, ton per hectare 8.5 13.5 14.5 16.5 14.6 14.4 14.6 10.3
Country Great Britain Germany France Hungary Poland Russia
Fig. 3 Dynamic of dryness index based on 1975–2004 data
Wheat productivity, ton per hectare 6.10 5.52 5.51 2.85 3.24 4.19 1.53
Estimation 45 38 33 29 22 15
180
A. Kleschenko Table 6 Crop productivity change due to climate change Scenario realization period, years 10–15 30–40 60–70 10–15 30–40 Economic region Cereal crops Forage crops North 12 26 24 9 22 North – West 10 22 12 8 21 Kaliningrad 16 34 25 9 22 Central 12 27 25 7 19 Volga – Vjatka 9 20 26 8 21 Central – Black Soil 6 15 15 8 20 Povolzhje, North 7 16 19 11 24 Povolzhje, South 3 7 30 2 5 North Caucasus −1 −6 −7 1 2 Ural 5 11 16 6 14 West – Siberian −3 −7 −1 3 6 East – Siberian −5 −12 −18 0 0 Far East 4 10 12 5 2 Russia 5 11 14 6 13
60–70 32 24 22 24 30 24 30 14 3 28 19 0 6 21
Conclusion The discussed in this paper climate changes scenarios are suppose to benefit Russian agriculture on 85% of its territory since it has considerable bio-climatic reserves, especially in the non-chernozem zone, to increase agriculture production. However, it would require considerable investments to promote some adaptation measures in order to mitigate negative influence of climate changes. One of such beneficial measure would be the expansion of areas under late season and more productive crops; planting winter crops immediately after harvesting summer crops; increase in application of fertilizers, herbicides and other measure of crops’ protection. The adaptive measure for southern Russia (North Caucuses and Lower Volga) would require to increase the cultivation of drought resistant crops such as sunflower and millet, stimulate viniculture, gardening, cultivation of tea, citrus and even cotton; and finally improve crop irrigation.
Consequences of Land and Marine Ecosystems Interaction for the Black Sea Coastal Zone Vladimir Kushnir, Gennady Korotaev, Felix Kogan, and Alfred M. Powel, Jr
Abstractâ•… The Normalized Difference Vegetation Index (NDVI) was used for the analysis of ecological characteristics of the Northwest Black Sea region, which includes significant coastal territories and sea water areas between the Crimean and northwest coast of the Black Sea. The data used in this study were 1997–2008 monthly NDVI from NOAA’s GVI2 dataset and visible channels from 1998 to 2008 from MODIS (Aqua and Terra) and SeaWiF scanners. It is shown, that the NDVI can be used as the characteristic of the integrated “land–sea” ecosystem, including internal links in this system. There is a direct-proportional relationship between NDVI in the land coastal zone and near-coastal water optical–biological characteristics. Although the NDVI values for water are small compared to the atmospheric signals (aerosol and molecular scattering) the resulting estimations have shown, that the possible relative error of the NDVI measurement for the northwest area of the Black Sea is within 12–25%. Keywordsâ•… Land–sea ecosystem • NDVI • Eutrophication • Chlorophyll • Seaweeds
Introduction The coastal northwestern Black Sea (43–47.5°N and 27.9–34.8°E) is an example of the typical, integrated “land–sea” ecosystem with the following features: the extensive shelf with depths up to 200 m which occupies approximately 25% of the Black Sea area, strong influence of drain of the large rivers (Danube, Dnepr, Dnestr, etc.), the large population density, intensive industry and agriculture. The rivers bring in the sea large quantities of nutrients containing phosphorus and nitrogen which create V. Kushnirâ•›() and G. Korotaev Marine Hydrophysical Institute, National Academy of Sciences of Ukraine, Sevastopol, Ukraine e-mail:
[email protected] F. Kogan and A.M. Powel, Jr NOAA/NESDIS Centre for Satellite Applications and Research, USA F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_21, © Springer Science+Business Media B.V. 2011
181
182
V. Kushnir et al.
favorable conditions for development of micro-seaweed and bacteria causing thereby eutrophication or “flowering”. In 1950th, about 14,000 t phosphates, 150,000 t nitrates and 2,350,000 t other organic substances annually drained from Danube, Dnepr and Dnestr rivers to the Sea, now, these values have increased to 55,000, 340,000 and 10,700,000 t accordingly. Currently, the following factors disturb the marine ecosystem balance in this region: large pollution from industrial, household and agricultural waste (containing nitrogen, ammonium, phosphorus and silicon compounds) draining from the rivers, sewer and storm drains; and of nitrates in the coastal water (ten times more, than in the open sea). The classical relation of Â�mineral forms of nitrogen and phosphorus ( N ⁄ P ) can be used to characterize the “biogenic pollution”. For example, in 2005, this ratio was 56.7 considerably exceeding the equilibrium level equal to 16. Strong correlation (−0.72) exists between salinity and ( N ⁄ P ) which indicates a leading role for river drainage for biogenic pollution. One of the negative events in marine ecology is eutrophication resulting in intensive production of a organic substance and reduction of water transparency. Traditionally it is considered, that the water transparency directly influences bio-productive of the sea environment. Oxidation of the superfluous organic material and the connected increased consumption of the oxygen dissolved in water also increase sea water turbidity. Another problem is development of the local layers of benthonic hypoxia (with thickness from several meters up to 20 m at the depth of 45–55 m) lead to destruction of benthonic biological community, including the molluscs (the main clearing sea water), killing large fish and hydrosulphuric infection. In the heaviest cases hypoxia covered the area exceeding of 20,000 km2, leading to death of ground fauna from 3 to 8 million tons per year. The average losses of seafood during 1972–1990 were about 60 million tons. Favorable conditions stimulate development in turbid water among decaying organic material such dangerous bacteria as intestinal stick, cholera vibrio, etc. (Vinogradov et€al. 1992; Zaichev et€al. 1992; Mankovsky et€al. 1996). The major marine local natural disasters in the coastal northwestern Black Sea are triggered pollution from the rivers drain, storms and sewer drains. In order to forecast these disasters objective analysis of interrelation land and sea ecosystems is required. This research investigates ecological characteristics of the “land–sea” ecosystem and their internal links.
Data and Methods Satellite data included weekly Vegetation Health (VH) indices from the NOAA’s Global Vegetation Index (GVI) dataset (spatial resolution 16 km) during 1997–2008 for southern Ukraine (Kogan 2004); MODIS optical scanners data for southwestern Ukraine (spatial resolution 0.5 km), covering land and water (http://ladsweb.nascom. nasa.gov/data/search.html); and SeaWiFS weekly data (spatial resolution 11 km) from 29 August 1997 for the northwestern Black Sea (http://oceancolor.gsfc.NASA.gov).
Consequences of Land and Marine Ecosystems Interaction for the Black Sea Coastal Zone
183
MODIS data were pre-processed by soft of Beam VISAT 2,3 and included extraction of VIS (0.645 mm) and NIR (0.8585 mm) channels counts, converting them to brightness in mW cm−2 sr−1 nm−1, and locating each pixel to geographic coordinates. Using these channels the Normalized Difference Vegetation Index (NDVI) was calculated (Rouse et€al. 1973).
NDVI =
Lt (0.8585) − Lt (0.645) · 10 2 , Lt (0.8585) + Lt (0.645)
(1)
where Lt (0.8585), Lt (0.645) are the brightness measured onboard photometer. SeaWiFS data (third level of processing) were used to calculate spatial distributions of the chlorophyll-a concentration. The hydrological and bio-optical characteristics of the northwestern Black Sea water environment strongly depend on Danube and Dnepr drains. Long-term flow variability of these rivers (Kirilenko et€al. 2009) were used to characterize “land–sea” ecosystems. Physical and biological processes of the “land–sea” interactions zone depend on the coastal trapped waves which are affected by the bottom relief. Therefore, for quantitative comparison of land (NDVI_L) and water (NDVI_W) NDVI, the distance values Ld on which the amplitude of the trapped waves is reduced in e times, was defined. For the northwester Black Sea this value was determined as 60–90 km and average NDVI was calculated for this zone of marine and land. Average NDVI_L was calculated for the area of 100–150 km from the shore line.
Results First, some links between land and marine ecosystems were explored by comparing images of vegetation health index (VHI) in the coastal zone with the amount of chlorophyll-a concentrationn in the water of the northwestern Black Sea. Four Â�conditions were selected: two with healthy vegetation (1999 and 2000) and two with stressed (2001 and 2003). During the years with drought-related stressed conditions, average chlorophyll-a concentration in the coastal water was 2.51 (2001) and 1.61 (2003) mg m−3. In case of the favorable conditions average chlorophyll-a concentration was 3.85 (1999) and 3.76 (2001) mg m−3. Larger chlorophyll concentration in the northwestern Black Sea coastal zone for healthy land vegetation years (wet conditions) was associated with much larger drains from Danube and Dnepr rivers and heavier contamination of marine ecosystem. In dry years (land vegetation is stressed) the pollution was much smaller. Figure€1 presents correlation between average near coastal zone land VHI and average chlorophyll concentration in the entire 60–90 km water. Low chlorophyll concentration is clearly associated with low VHI (drought-related vegetation stress). This relationship confirms a reduction of rivers and storm drains into the sea in dry years. Reduction of drains is accompanied by a reduction of the mineral suspension and nutrients in the sea environment and the corresponding reduction of the chlorophyll-a concentration and seaweed biomass. Opposite situation is Â�developed during
184 R =0.77 80
60
VHI
Fig.€1â•… Correlation between average land VHI and average water chlorophyll-a concentration, northwestern Black Sea, mid-May and end of July, 1998–2008; VHI = 39.29 ln(Chl) + 12.731 R − correlation coefficient
V. Kushnir et al.
40
20
0
1
2 3 Chlorophyll-a, mg/m3
4
5
the healthy vegetation years associated with wet conditions and above normal rainfall. This leads to rivers and storm drains increase as well as an increase in the quantity of the mineral suspension and nutrients in the sea, which is resulted in seaweed biomass growth and chlorophyll-a concentration increase. The result shown in Fig.€1 is supported by the annual analysis of Danube river flow shown in Fig.€2. Variation of the flow (QD) is quite coherent with vegetation health (VHI) during spring (mid-May) especially in extreme years (favorable vegetation conditions – 1999, 2001, 2005; stressed conditions – 2003, 2007). Moreover, both QD and VHI variations match well with chlorophyll-a concentration, especially in the close to the shore zone (60–90 km, Chl_2). NDVI from MODIS was used for the analysis of the transition zone between land and marine ecosystems of the northwestern Black Sea. The 85 cross-sectional NDVI measurements (Fig.€ 3) were taken from land and water for favorable and stressed vegetation conditions identified by VHI on the ground. The NDVI time series clearly indicate considerable changes from large to small NDVI while crossing land-water line, significant reduction of NDVI variability in water and what is the most important, much larger NDVI values for both land (favorable, wet conditions) and water (larger contamination due to land drainage). The features presented in Fig.€3 are summarized in Fig.€4, which shows correlation between NDVI_W and NDVI_L. Two groups are clearly formed for favorable (grey) and stressed (black) ecologically situations. During stressed vegetation conditions (mainly due to droughts) the drains of the rivers and storm into the sea considerably decreases resulting in a reduction of mineral suspension, nutrient and correspondingly chlorophyll-a and seaweed biomass in the sea environment. Sea water becomes more transparent and homogeneous, resulted in NDVI_W decrease.
Consequences of Land and Marine Ecosystems Interaction for the Black Sea Coastal Zone
50
40
Flow of Danube, QD, m3/s
VHI
60
10
10000
30
8 8000 Chl_2
6
VHI 6000 4
QD
4000 1996
Chl_1
Chlorophyll-aconcentration, mg/m3
70
185
2 2000
Years
2004
2008
Fig.€ 2â•… Inter-annual variability of Danube flow (QD), average chlorophyll-a concentration: in (Chl_1) the northwestern Black Sea (44–47°N, 28.3–34°E) and (Chl_2) 60–90 km marine coastal zone (44.7–46.5)°N, (29.0–31.5)°E, and VHI mid-May (week 18), 1997–2008
0.8
NDVI 10−2
0.4
0
-0.4
Land
-0.8 −80
−40
Sea 0
40
80
Distance from the sea coast, km Fig.€3â•… NDVI profiles at crossing of the land–sea line (l = 0); the black and grey colors are stressed and favorable conditions, respectively
Healthy vegetation is developing in moist environment created by an abundant rains resulted in an increase in rivers and storm drains and more mineral suspension and nutrient deposition in the coastal water, which stimulates formation of chlorophylla n and an increase in seaweed biomass. Spatial and time structure of the “impurity” is determined by complex dynamics of the water environment, i.e. the trapped
186
V. Kushnir et al. 0
R= 0.634
−0.2
NDVI_W 10−2
Fig.€4╅ Correlation between NDVI_W and NDVI_L for stressed (black circles) and favorable (grey circles) weather-related vegetation health conditions
−0.4
−0.6
−0.8 0
0.2
0.4
0.6
0.8
NDVI_L 10−2
coastal waves, topographical eddies, meso-scale distortions and other dynamic processes. Large dispersion of the points is probably due to these features. The data of the third level processing of the color scanners from http://reason. gsfc.nasa.gov/OPS/Giovanni/ocean.seawifs.2.shtml were used for the further analysis. An empirical relationship between chlorophyll-a concentration and the color index I wn (443 ⁄ 555) = 1.08C −0.44 where I wn (443 / 555) = Lwn (443g \ mm) / Lwn (555 \ mm) â•›was developed from the SeaWiFS data. This dependence is similar to the known algorithms such as OC2 for 0.443 and 0.555 µ m waves (Kushnir and Stanichny 2007; Urdenko and Shimerman 1987; O’Reilly et€al. 1998). The empirical ratios between color index I wn (443 ⁄ 555) and depth of visibility of the white disk (Urdenko and Shimerman 1987)
Z d = 14.95 · ln I wn (443 / 555) + 10.285,
(2)
and also for weighed substance concentration Cv as a function on the depth of the white disk visibility (Vituk 1983):
Cv = 4.59 · Z d − 0.85
(3)
are used for further analysis. The spatial distributions of the chlorophyll-a concentration, depths of the white disk visibility and suspension concentration for July 2007 (week 30, stressed conditions) and for May 2008 (week 18, favorable conditions) were calculated using Eqs.€2 and 3. Figure€5 presents relationship between these parameters and NDVI_W. As seen, the relationship is strong and provides some way estimation bio-optical parameters from the NDVI_W data in the 60–90 km coastal zone of the Â�northwestern Black Sea.
Consequences of Land and Marine Ecosystems Interaction for the Black Sea Coastal Zone 10
2
12
6
Zd, m
Chl, mg/m3
8
4
2
R=0.8
16
1
187
1
8 4
2 R=0.73
0 20
30
0 40
50
20
60
30
−(NDVI_W)
40
50
60
70
−NDVI_W 20
1
Cv, g / m3
16 12 8
2
4 0
R=0.81
20
30
40
50
60
−(NDVI_W) Fig.€5â•… Correlation between NDVI_W and chlorophyll-a concentration (Chl = 2509.6 • |NDVI_W|-1.8574), depth of the white disk visibility (transparency), (Zd = 0.0011 • |NDVI_W|2.2741), concentration of a mineral suspension (Cv = 302198 • |NDVI_W|-3.0742); (1) filled points, May 2008 and (2) open points July 2007. R is correlation coefficient
Influence of noise factors in 0.645 mm and 0.8585 mm channels on NDVI_W estimations Since the NDVI_W is a very important parameter this part investigate statistical characteristics of the NDVI_W relative to noise level. First, it is necessary to present Lt (0.8585) and Lt (0.645) values as Lt (0.8585) = x1 + x1 and Lt (0.645) = x2 + x2, where x1, x2 are the average values, x1, x2 are the noise components of the brightness signals on the specified waves. We also assume that the noise is random, have normal distribution, no collinearity, and standard deviations are equal to s1 + s2. At NDVI_W = Lt (0.8585) - Lt (0.645)/Lt (0.8585) - Lt (0.645) calculation, average value and dispersion of numerator and a denominator are equal accordingly to x1 - x2 = xc; x1 + x2 = xz; á(x1 - x2)2ñ = s12 + s22 = sc2; á(x1 + x2)2ñ = s12 + s22 = sz2. Coefficient of correlation rcz of numerator and denominator of the NDVI_W estimation equals
188
V. Kushnir et al.
ρcz =
(x 1 − x 2 )(x 1+ x 2 ) s cs z
=
s 12 − s 2 2 = 0. s 12 + s 2 2
(4)
In case (rcz = 0), the two-dimensional density of the probabilities distribution for numerator and denominator of the NDVI_W estimation can be presented as (based on Levin 1960): u 2 + u2 2 1 exp − 1 Wcz (xc , xz ) = , (5) 2ps cs z 2 xc − x c
xz − x z . sc sz The density of probabilities distribution of the NDVI_W estimation is determined following random variables transformation (Levin1960) as: where u1 =
WNDVI ( y) =
; u2 =
∞
∞
∞
−∞
0
0
∫ Wcz (uy, u) u du = ∫ Wcz (−uy, −u)udu + ∫ Wcz (uy, u) udu.
(6)
After substitution of Wcz (xc, xz) in (6) and performance of transformations, expression for WNDVI (y) can be written as: WNDVI ( y) = =
α exp[ −0.5 ε c 2 + ε z 2 π
(
∞
)∫ z exp (−β z )· ch (γ z )dz 2
0
γ 2 γ C0 γ exp Φ 1 + π , β β 2 β 2 β
гдe x 2 x 2 c + z ε c2 + ε z2 σc σz α α C0 = exp − = exp − π π 2 2 β=
σ y2 + α2 ; α = c = 1; γ = yε c + αε z ; σz 2 F (r) =
2 p
∞
∫ exp(−t
2
) dt.
(7)
0
Results of WNDVI_W (y) calculations at various ratio of the standard deviation component of a signal to its average value (useful signal) are presented in Fig.€6. As a result ∞
of these calculations the values of dispersion σ NDVI _ W 2 = ∫ ( y − y ) WNDVI _ W ( y) dy 2
0
and dependence of σ NDVI _ W / NDVI _ W = f (σ x / NDVI _ W ), σ x 2 = σ12 + σ 2 2 are determined. The relation of the noise/useful signal for photometer channels and their influence on accuracy of the NDVI_W calculations is estimated using these ratios (with the level of approximation reliability R2 = 0.996).
Consequences of Land and Marine Ecosystems Interaction for the Black Sea Coastal Zone
189
5
4
3 1 2
1
2 3
0 −2
−1
0
1 y
Fig.€ 6â•… Density of probability distribution for NDVI_W estimations for the three ratios of the standard deviation of noise to the average value of the useful signal. (1) −0.1; (2) −0.2; (3) −0.3
2
σx σx (8) = −0.4355 + 0.7309 − 0.0045 NDVI _ W NDVI _ W NDVI _ W σ NDVI _ W
The relation signal/noise for channels of the MODIS photometer is equal 128 for 0.645 mm and 201 for 0.8585 mm. For favorable ecological conditions (low transparency of sea water), áLt (0.645) = 25â•›\â•›m Wcm-2sr-1nm-1, áLt = 0.8585)ñ = 12â•›\â•›m Wcm-2 sr-1nm-1. Thus sx = 0.2 \mu Wcm-2sr-1nm-1 and sNDVI_Wâ•›/áNDVI_Wñ = 0.25. For stressed ecological conditions (high transparency of sea water), áLt (0.645) = 13â•›\â•›m Wcm-2sr-1nm-1, sx = 0.107â•›\â•›m Wcm-2sr-1nm-1, áLt = 0.8585)ñ = 3.4â•›\â•›m Wcm-2 sr-1nm-1 , and sNDVI_Wâ•›/áNDVI_Wñ = 0.12.
WNDVI_W(y)
Thus, the influence of the photometer noise results in the NDVI_W relative error of 12 and up to 25% depending on conditions.
Conclusions 1. The NDVI can be used as ecological indicator of both land (NDVI_L) and 60–90 km marine near-coastal zone (NDVI_W) in the northwestern Black Sea. 2. There is relationship between NDVI_L and NDVI_W in the land–sea coastal area. 3. Relative error of the NDVI_W estimation in the northwestern Black Sea is 12–25%.
190
V. Kushnir et al.
References Kirilenko NF, Kushnir VM, Lemeshko EM (2009) Influence of a river drain on ecological conditions in Northwestern area of the Black Sea according to contact measurements and remote sensing data. Geoinformatic 4 (in Russian) Kogan F, Stark R, Gitelson A, Adar E, Jargalsaikhan L, Dugrajav C, Tsooj S (2004) Derivation of pasture biomass in Mongolia from AVHRR-based vegetation health indices. Int J Remote Sens 25(14):2889–2896 Kushnir VM, Stanichny SV (2007) Color index in the Northwestern Black Sea derived from scanner MODIS data. Res Earth Space 4:62–73 (in Russian) Levin BR (1960) Theoretical basis of the statistical radio engineering. Second book. Soviet radio, Moscow, 503 p (in Russian) Mankovsky VI, Vladimirov VL, Afonin EI, Mishonov AV, Solov’ev MV, Anninsky BE, Georgieva LV, Unev OA (1996) The long-term variability of the transparency of water in the Black Sea and the factors which have caused its strong reduction at the end of 80 beginning of 90-th years. Preprint /NAS of Ukraine. Marine Hydrophysical Inst., Sevastopol, 32 p O’Reilly JE, Maritorena S, Mitchell BG, Siegel DA, Carder kL, Garver SA, Kahru M, McClain C (1998) Ocean color chlorophyll algorithms for SeaWiFS. J Geophys Res 103(C11):24937–24953 Rouse JW Jr, Haas RH, Schell JA, Deering DW (1973) Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation. Progress Rep. RSC 1978-1, Remote Sensing Center, Texas A&M University, College Station, 93 p Urdenko VA, Shimerman G (eds) (1987) Optical remote sensing of the sea and the influence of the atmosphere, vol 2, Part 2. Inst. for Space Research Academy of Sciences of GDR, Berlin, Moscow, Sevastopol, p 197, in Russian Vinogradov ME, Sapozhnikov VV, Shychina EA (1992) Ecosystem of the Black Sea. Moscow, Science, 112 p (in Russian) Vituk DM (1983) Weigh substance and its biogenic components. Naykova Dymka, Kiev, 210 p (in Russian) Zaichev UP (1992) Ecological condition of the shelf zone of the Black Sea at coast of Ukraine (review). Hydrobiol J 28(4):3–19 (in Russian)
Utilizing Satellite Data to Highlight High Ozone Concentration Events During Fire Episodes Rasa Girgždienė and Steigvilė Byčenkienė
Abstract The long-term ground-level ozone concentration measurements at the Preila station in Lithuania showed no significant trend of ozone peak values over 1982–2008. The ozone episodes that can be characterized by unusually high concentration during spring were analyzed. The air mass backward trajectories, satellite fire location data, air pollutant dispersion modeling and measured concentration of other pollutants were used in the study. The investigations have shown that pollution released during the fires in southern Russia and Ukraine is a factor influencing the ozone concentration in Lithuania, at distances of over 1,000 km from the fire areas. The ozone concentration increase up to 20 mg/m3 was established as a result of fires in the neighboring regions. Keywords Ozone concentration • Fires • Air quality • Satellites
Introduction The ambient air quality is one of the main environmental issues, which determines human health, comfort and well being. Many efforts to reduce the air pollution were made by different institutions: the acceptance of new directives and regulations, setting of new health-based ambient standards, the increase of funding for reducing pollutant emissions, etc. For example, the Clean Air Act requires Environmental Protection Agencies to set National Ambient Air Quality Standards for widespread pollutants from numerous and diverse sources considered harmful to public health and the environment. Monitoring and predicting ozone concentrations are a matter of special concern because ozone is one of the most significant air pollutants damaging human health and vegetation (Madden and Hogsett 2001). Since the 1980s, rural ozone concentrations have increased in many areas (The Royal Society 2008) with quite different rates of change at different locations. R. Girgždienė (*) and S. Byčenkienė Institute of Physics, Vilnius, Lithuania e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_22, © Springer Science+Business Media B.V. 2011
191
192
R. Girgždienė and S. Byčenkienė
Measurements have shown that ozone background in the mid-latitudes of the Northern Hemisphere has increased from 4 mg/m3 per decade before to 70–80 mg/m3 presently. However, the trends in ground-level ozone are highly variable and depend on the region. The annual cycle of ozone at the background sites in the Northern Hemisphere is characterized by a spring maximum. The sites, which are affected, to some extent, by local ozone production, exhibit a broad summer maximum or even spring–summer maximum (The Royal Society 2008; Sánchez et al. 2008; Girgzdiene and Girgzdys 2003). Ozone level, especially its peak concentrations, depends on the precursor level and it is generally higher downwind of ozone precursor sources, at distances of hundreds or even thousands of kilometers. These precursors are transported and dispersed by wind from the cities, and in the presence of sunlight ozone is formed (Syri et al. 2001). The fires can increase concentration of the ground-level ozone and contribute to many elevated ozone events by releasing nitrogen oxides and hydrocarbons, which can form ozone near the fire or far downwind as a result of chemical reactions in sunlight. Regional background ozone concentrations can be elevated by 20–40 mg/m3 during forest fire episodes (Wotawa and Trainer 2000). The monitoring and scientific researches have recognized the importance of satellites in detecting active fires and applications of these data in the study of the environment quality. The aim of this investigation is to examine the causes of ozone episodes with high concentrations using satellite application facility products during spring periods in Lithuania.
Location and Methods The Preila environmental pollution research station (55°22¢N and 21°00¢E at an elevation of 5 m above see level) is located in the western Lithuania on a coast of the Baltic Sea, on the Curonian Spit. It is a narrow sandy strip that separates the Baltic Sea and the Curonian Bay. There are no large sources of anthropogenic pollution of the atmosphere close to the monitoring site. The station is part of the EMEP (The European Monitoring and Evaluation Programme) network. The ground-level ozone concentration was measured with the UV absorption ozone analyzer O341M, with a detection limit of 2 mg/m3. The data of other pollutants and meteorological parameters from the Lithuanian State monitoring network (www.gamta.lt) were used in the analysis as additional parameters. The 48-h backward air masses trajectories arriving at heights of 20, 500 and 1,000 m above ground level were produced using HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) model accessed via NOAA ARL READY website (http://www.arl.noaa.gov/ready/hysplit4. html). For detection of fire areas the Web Fire Mapper (http://firefly.geog.umd.edu/ firemap), which is a part of the MODerate-resolution Imaging Spectroradiometer (MODIS) Rapid Response System, was used. The Navy Aerosol Analysis and Prediction System (NAAPS) model results were used to determine the distribution
Utilizing Satellite Data to Highlight High Ozone Concentration Events During Fire Episodes 193
of smoke, dust and sulphate aerosols. Details of the model and its results are available at the Internet site (http://www.nrlmry.navy.mil/aerosol) of the Naval Research Laboratory, Monterey, California.
Results and Discussion The long-term ozone measurement data at the rural Preila station show an increasing trend (0.84 ± 0.08 mg/m3 per year, p < 0.0001) in the annual mean during the period of 1982–2008. Nevertheless a clear decrease of the ozone level was established during 2007–2008. The annual ozone peak level is mostly caused by human activities, which can be different in nature as well as in intensity; therefore it is more difficult to estimate the ozone peak trend. The search for the peak value trend in Preila was performed and no statistically significant trend (0.2 mg/m3 per year, p = 0.546) was found in the ozone peak values during 1982–2008; although since 1992 the tendency of a decreasing ozone peak is observed. The average seasonal cycle of ozone at the station is distinguished by broad spring–summer maximum (the peak values are observed mostly during summer months). Hourly ground-level ozone concentration values vary from 20 to 140 mg/ m3, depending on temperature, sunlight, wind pattern and NOx concentrations. However, in some years, an elevated ozone concentration was observed during early spring (March–April). The hourly data for 2008 showed the traditionally seasonal dynamics of ozone monthly mean values, (except ozone values higher than 95th percentile) was in spring (Fig. 1).
Fig. 1 Box – whisker analysis of hourly ozone concentrations for 2008. The box and whiskers denote the 5th, 25th, 50th, 75th and 95th percentiles and the square is the mean, triangle – minimum and star – maximum values of ozone concentration, full circle – 1st and empty circle – 99th percentiles
194
R. Girgždienė and S. Byčenkienė
Fig. 2 Variation of the hourly ozone concentration at two stations in Preila and Klaipeda during March–April, 2008. Ozone episodes under investigation are marked by circles
Two episodes with high ozone concentration were observed during 30 March–01 April and after 25 April. Ozone data from the other monitoring station on the outskirts of Klaipeda city (www.gamta.lt), which is located about 40 km from the Preila station, showed the same episodes (Fig. 2). The peak values reached 113.9 and 113.2 mg/m3 in March and 123.9 and 119.4 mg/m3 in April in Preila and Klaipeda, respectively, and these levels are not characteristic of Lithuanian climatic condition during this period, i.e. the concentrations exceeded usual level by up to 20 mg/m3. The origin of these events was ascertained by means of synoptic and meteorological analysis, determination of backward air mass trajectories, satellite fire location data and the examination of behavior of other pollutants. The meteorological data showed that after cold March, when air temperature did not exceed 5°C, the warm period from March 29 started and air temperature increased to +15°C in Lithuania. Similar situation was observed in the neighboring regions from where air masses came, i.e., the 48-h backward air mass trajectory analyses demonstrated that air masses were transported from the southern regions (Fig. 3). The satellite-based fire data from MODIS sensor on Terra (morning) and Aqua (afternoon) satellites (presented on Web Fire Mapper), revealed that these air masses passed the region of the Ukraine, Belarus and Kaliningrad area (Russia) with intensive fires. People burning last year dry vegetation caused most of these fires. Such biomass burning activity is the most intensive in spring resulted in large amount of ozone-components (NOx and CO) emitted during these fires. The NAAPS model results indicated that the strongest smoke plumes reached Lithuania on 31 March (Fig. 4). It is known (Niemi et al. 2005) that seasonal forest, grass and peat-bog fires are the main source of smoke aerosol in Eastern Europe. The fire is a source of particulate matter with particle sizes <2.5 mm (PM 2.5). They can be transported hundreds or thousands of kilometers from the source. The mass concentration of PM2.5 was measured using the beta adsorption method by the ambient suspended particulates monitor model MP 101M, Environnement S.A.
Utilizing Satellite Data to Highlight High Ozone Concentration Events During Fire Episodes 195
Fig. 3 The 48-h backward air mass trajectories and fire locations, March 31, 2008
NAAPS Surface Concentration (ug/m**3) for 12:00Z 31 Mar 2008 Smoke 70N
65N
60N
55N
50N
45N
40N
35N
30N 30W
25W
20W
1
15W
2
10W
5W
4
0
5E
8
10E
16
15E
32
20E
25E
64
30E
35E
40E
128
Fig. 4 NAAPS model results showing surface smoke concentration for the strongest stage in Lithuania, March 31, 2008
196
R. Girgždienė and S. Byčenkienė
Fig. 5 The PM 2.5 concentration at Klaipeda station during March–April, 2008
at the Klaipeda outskirts station. The data indicated large concentration (77 mg/m3) of particles during this episode (Fig. 5), when the average concentration during March–April was only 12 mg/m3. Similar situation was observed at Preila station. It is known that carbon isotope measurements provide a powerful tool to identify different types of carbonaceous aerosols (Marley et al. 2008). The 13C/12C ratio of the aerosol samples is good tracer of biomass burning, i.e. fires. The 13C/12C value (−30.9 ± 0.2‰) of the aerosol samples collected at Preila during March 31–April 1, 2008 (Garbaras et al. 2008) indicated that air masses from fire regions have reached the station. The high-pressure ridge brought dry air from the southern Russia into the region of Lithuania during this period. The second high concentration ozone episode was observed in April. Longterm ozone measurement data demonstrate that the elevated ozone level during the end of April is often observed in Lithuania. If monthly average ozone in April accounts for 68.3 ± 10.2 mg/m3, then the largest ozone concentration can reach 130 ± 10 mg/m3. The backward air masses trajectory analyses showed that on April 25, 2008, when ozone monthly peak was 123.9 mg/m3, the air masses moved slowly over the regions where no fire sources were detected by satellites (Fig. 6). The weather was determined by slow moving anticyclone and the air temperature increased to 20–22°C, the wind speed did not exceed 2 m/s. The hourly concentration of PM2.5 at the monitoring station on the outskirts of Klaipeda did not exceed usual level –20 mg/m3, whereas the hourly PM10 concentration increased to
Utilizing Satellite Data to Highlight High Ozone Concentration Events During Fire Episodes 197
Fig. 6 The 48-h backward air mass trajectories and fire locations, April 25, 2008
120 mg/m3 (normally, PM10 monthly average in April is only 30 mg/m3). In addition to ozone, very high hourly concentration of NO2 (148 mg/m3) was registered as well. These results suggest that synoptic situation was the cause of the high ozone concentration measured at the Preila and Klaipeda stations.
Conclusions The average seasonal cycle of ozone at Preila station in Lithuania was identified to have a broad spring–summer maximum. The long-term ground-level ozone concentration measurements at the station showed an increasing trend (0.84 ± 0.08 mg/m3 per year) of monthly ozone values. However, no significant trend of ozone peak values was identified during 1982–2008. High ozone episodes in Lithuania were shown to be stimulated by fires in the neighboring region (ozone concentration increased up to 20 mg/m3), which occurred during March 30–April 1, 2008 over large areas of southern Russia and Ukraine. Backward air mass trajectories, satellite detections of fire areas, air pollutant dispersion modeling results and pollutant concentration indicated that emissions from fires arrived to Lithuania and influenced the ozone level during this episode. The ozone concentration increase observed on April 25, 2008 can be attributed to local pollution conditions determined by meteorological and synoptic situation when the high-pressure system was slowly moving to the east allowing high pressure to prevail over the region. Acknowledgement This research was supported by the Lithuanian State Science and Studies Foundation and the Environmental Protection Agency.
198
R. Girgždienė and S. Byčenkienė
References Garbaras A, Andriejauskiene J, Bariseviciute R et al (2008) Tracing of atmospheric aerosol sources using stable carbon isotopes. Lith J Phys 48:259–264 Girgzdiene R, Girgzdys A (2003) Variations of the seasonal ozone cycles in the Preila station over the 1988–2001 period. Environ Chem Phys 25:11–16 Madden MC, Hogsett WE (2001) A historical overview of the ozone exposure problem. Hum Ecol Risk Assess 7:1121–1131 Marley NA, Gaffney JS, Tackett MJ et al (2008) The impact of biogenic carbon emissions on aerosol absorption in Mexico City. Atmos Chem Phys Discuss 8:18499–18530 Niemi JV, Tervahattu H, Vehkamäki H et al (2005) Characterization of aerosol particle episodes in Finland caused by wildfires in Eastern Europe. Atmos Chem Phys 5:2299–2310 Sánchez ML, García MA, Pérez IA et al (2008) Evaluation of surface ozone measurements during 2000–2005 at a rural area in the upper Spanish plateau. J Atmos Chem 60:137–152 Syri S, Amann M, Schöpp W et al (2001) Estimating long-term population exposure to ozone in urban areas of Europe. Environ Pollut 113:59–69 The Royal Society (2008) Science policy report 15/08. Ground-level ozone in the 21st century: future trends,impacts and policy implications. http://royalsociety.org/displaypagedoc. asp?id=31506. Accessed 9 Feb 2009 Wotawa G, Trainer M (2000) The influence of Canadian forest fires on pollutant concentrations in the United States. Science 288:324–328
Geomagnetic Disturbances and Seismic Events in the Vrancea Zone from in Situ Data Frina Sedova, Vladimir Bakhmutov, and Tamara Mozgovaya
Abstract Based on example from the Vrancea’s concentrated seismicity zone the connection between definite type of geomagnetic disturbances and 200 earthquakes during the period of 1988–1996 is shown. The energy classes of earthquakes are related to abrupt changes (“gradients”) in H-component of geomagnetic field. Taking into account the temporal interval from maximum of sub-storm up to the shock, the relationship between geomagnetic disturbances and earthquakes is always positive. The time interval between maximum of polar sub-storm and earthquake is correlated with the depth of the earthquake. Differences in the duration of polar substorms before crustal (shallow) and deep earthquakes are revealed. Four main types of morphological features of geomagnetic variations preceding the seismic event are established. Keywords Geomagnetic disturbances • Earthquakes • Seismicity zone • Polar substorm
Introduction Today, the Sun, interplanetary space–magnetosphere, ionosphere, atmosphere and Earth’s tectonosphere the majority of scientists are considered as a complicated open dynamic nonlinear system with high-energy, which affects complex processes. In this context the seismic phenomena of the Earth should be considered as a part of the whole “Sun - Earth” chain. In the numerous publications devoted to the analysis of the relation of the system’s physical processes to seismicity, scientists agree that such relation does exist. The study of seismicity (in regions) in relation to the solar activity, earth tide amplitudes, and Earth’s rotation velocity gives F. Sedova, V. Bakhmutov, and T. Mozgovaya (*) Institute of Geophysics, Ukrainian National Academy of Sciences, Kiev, Ukraine e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_23, © Springer Science+Business Media B.V. 2011
199
200
F. Sedova et al.
grounds to state that these processes affect the stress–strain state of the crust (Sobolev et al., 2001). Several authors argue that the process of seismic activation involves changes in the groundwater level, parameters of the ionosphere, electromagnetic and geomagnetic fields, electrical conductivity of rocks, radon emanation, and so others; eventually, these changes can be considered as precursors of a seismic event (Rikitake, 1979; Sobolev, 1993, 2003). Since we examine here the dynamics of geomagnetic variations in relation to seismicity in a local concentrated seismicity zone, the question is “How the geomagnetic field can promote seismic energy release?”. The influences of magnetic storm with sudden commencements on seismicity have been discussed in (Sobolev et al. 2001). The studies of technogenic impulse electrical signals when the number of the seismic events tends to increase after 3–4 days impulse passing confirmed that (Tarasov et al. 1999). The duration of activation was several days and affected area was hundreds kilometers. As far as the energy pumped into the Earth by several orders less than the released seismic energy, the trigger mechanism of the influence have been taken into consideration. An assumption on the trigger mechanism of the effect of the magnetic disturbance (magnetic storms) on the seismicity in Kirghizia, Kazakhstan and Caucasus (with the maximum effect in 2–7 days after disturbance) was discussed by (Sobolev et al. 2001, Zakrzhevskaya and Sobolev 2002). However the correlation was not evident that the magnetic storm could be considered as the cause of tectonic events. The positive or negative effects were explained by different geological structures of the regions. One of the possible mechanisms might be electro-osmotic phenomena in the rocks (Kormiltsev et al. 2002). The electro-osmotic fluid flow induced by magnetic storms generates anomalous porous pressure which could trigger tectonic event. Since this effect is small, the correlation between the magnetic field variations and tectonic processes is weak. In this paper we present correlation analysis between geomagnetic field disturbances and earthquakes in the concentrated seismicity zone of Vrancea in South Carpathian region. We should mention that do not have intention to search for earthquake precursors, which requires clear understanding of the physical mechanisms and origins of various electromagnetic phenomena accompanying earthquakes.
Initial Data Geomagnetic situation was studied for more then 200 earthquakes with intermediatedepth hypocenters beneath Vrancea zone during 1988–1996. We have analyzed more than 150 deep-foci and ca 60 crust shocks including strongest earthquakes of 1977, 1986, and 1990. The energy class of the shocks (K) and the foci depth (h, km) were taken for the seismic characteristics according to the Seismological bulletins of Ukraine [Seismological Bulletin of Ukraine, 1977, 1988–1996]. The Vrancea zone is regarded as a classical example of concentrated seismicity (Martin et al., 2005). The area of earthquakes here is a narrow band 90 km long and 25 km wide. Earthquake
Geomagnetic Disturbances and Seismic Events in the Vrancea Zone from in Situ Data
201
sources are located in both the crust and upper mantle (at depths of 100–180 km). Deep shocks are characterized by high seismic energies; the strongest earthquakes have magnitudes M > 7. The geomagnetic situation was examined during the same period, but we did it separately for deep-foci earthquakes (sub-crust) and crust ones (shallow). According to the classification scheme of magnetic disturbances, their fields are caused by various sources that lie at the boundary of or in the magnetosphere or are, at least partially, of ionospheric origin. We examined changes in the amplitude of the H-component of the geomagnetic field. The amplitude of the variations is used to determine the field perturbation, estimation of which is usually based on various indices of geomagnetic activity; in particular, the most used three-hour index Kp or its diurnal sum ∑Kp. The different oscillations in the structure of geomagnetic disturbances have been well studied. We stress attention to the substorms, produced by a complex of magnetic and ionospheric phenomena among which the main is a sharp and appreciable increase in the auroral electrojet. The latter leads to a sharp increase in the values of the horizontal component of the geomagnetic field reaching a few hundreds nT and more in the high latitudes. Each of the disturbances is distinguished by individual features because of the dynamic nature of auroral electrojets. Injection of particles into a ring current and a change in magnetic fields and currents of the geomagnetic tail and the magnetopause leads to a close relation of a substorm to the phenomenon of a planetary magnetic storm. Substorms occur both during a magnetic storm and against a comparatively quiet background. In the auroral zone, two types of disturbances are distinguished: polar substorms 0.5 to 1.5 hour’s long and long-term disturbances lasting for 2 to 5 hours. At middle latitudes, oscillations longer than two hours are a comparatively rare phenomenon, particularly against the generally quiet background of the magnetic field. They are related to spreading currents of the polar electrojet. We analyzed geomagnetic variations in relation to Vrancea zone earthquakes by the Yastrebovka (45.5°N; 34.1°E) and Korets (50.6°N; 27.2°E) magnetic observatories. The morphological features of the variations were studied by the changes of the amplitude of the H-component of the geomagnetic field from daily magnetograms.
Methods and Results Sedova et al., 2001 presented that noticeable changes in the geomagnetic field precede deep earthquakes in the Vrancea zone. The concept of the “main gradient” (the greatest change in the horizontal H-component for more than 2 days before shock) and the “gradients” (changes in the horizontal H-component for 1–2 days before shock) were described by (Sedova et al. 2001, Bakhmutov et al. 2007). The connection between deep-foci earthquakes in Vrancea and preceding geomagnetic disturbances was determined, while the seismic events themselves occur mainly at
202
F. Sedova et al.
a
b
n, % 30
n, % Shallow earthquakes Sub-crust earthquakes
40
25 20
30
15
20
10 10
5 0
4
8
12 16 20 24 28 32 36 40 days
0
0
1
2
3
4 5 days
6
7
Fig. 1 Time interval t from (a) the “main gradients” of the magnetic field to deep earthquakes and (b) the “gradients” preceding the sub-crust and the shallow shocks in Vrancea. The relative number of seismic events n is plotted on the ordinate axis
weakly disturbed and quiet days (85%), and only 15% of the events took place in the geomagnetic disturbed days (∑Kp > 30). The geomagnetic activity by Kp-indices with relation to seismicity in the Vrancea zone is always positive when considering the time intervals from the “gradient” to the shock (Bakhmutov et al. 2006). The time interval t from the “main gradients” in the geomagnetic field to the deep-foci earthquakes varies from several days to one month and more (Fig. 1a) while t from “gradients” to the shallow shocks is more often for several hours and rarely exceeds two days (Fig. 1b). Bakhmutov et al. 2006 showed that more than 90% of shocks recorded in 1977 and 1988–1996 in Vrancea zone were related to mid-latitude manifestations of midnight polar substorms. Figure 2 presents the distribution of substorms (H-component of the geomagnetic field) depending on the time before shallow and sub-crust earthquakes. An example of the mid-latitudinal manifestation of polar substorm during the quiet background of the geomagnetic field is given in Fig. 3. The shock of April 18, 1993 at 02:03 UT (K = 9.9, h = 150km) is preceded by the substorm om March 17 with the maximum in the H-component about 6 pm. The amplitude of substorm was digitized from a base line conventionally taken as a normal field (zero level). For the whole period of studies the earthquakes intensity in Vrancea were stable, within the range of K = 9−11 for both the deep-foci earthquakes and for the crust shocks. The weaker shocks (K < 9) were 5.3%, and the shocks with K > 11 attain 7.6%. Mainly distribution of the energy class K of sub-crust earthquakes for 1988–1996 and the mean statistical values of the amplitudes of the H-component at the maximum of the magnetic polar substorms development preceding the shock are shown on the Fig. 4. In spite of a narrow limits of the earthquakes intensity changes their linear dependence on the substorms amplitude ∆H is clearly seen, especially if the main seismic events on March 4, 1997 (K = 15.8) and May 30, 1990 (K = 16) are included in the diagram.
Geomagnetic Disturbances and Seismic Events in the Vrancea Zone from in Situ Data n, % 16
203
Shallow earthquakes Sub-crust earthquakes
12 8 4 0
12
16
20
24
4
8
UT
Fig. 2 Distribution of daily substorms before the shallow and sub-crust earthquakes in the Vrancea zone for the period of 1988–1996
Fig. 3 Magnetograms from April 17–18, 1993 at the Yastrebovka magnetic station
Before the shallow earthquakes the D H amplitudes are 50–80 nT which are comparable with amplitudes of substorms occurring before the deep-foci earthquakes (Fig. 5a). Before the deep-foci shocks the amplitude range is wider and ∆ H = 90-110 nT in 10% of events. An analysis of the duration (T, min) of the midnight polar substorms preceding the shallow and sub-crust earthquakes shows that they are different (Fig. 5b). In most cases sub-crust earthquakes preceded the substorms with T > 60min while before shallow shocks it is T £ 60min. Figure 6a shows a linear relationship between the earthquake foci depth and the duration of midnight polar substorms for both deep and shallow (crustal) earthquakes. This result agrees with the geodynamic model of the Vrancea zone which is a localized narrowing downwards crater-shaped structure for which the notion of crust-mantle discontinuity (the estimated depth of 50–60 km) may be considered as
204
F. Sedova et al.
Fig. 4 The distribution of the energy class K of sub-crust earthquakes for 1988–1996 and the mean values of the amplitudes of the H-component at the maximum of the polar substorms development. The number of events is indicated in parentheses
Fig. 5 The amplitudes (a) and the duration (b) of polar substorm preceding the shallow (shaded columns) and sub-crust (open columns) earthquakes in Vrancea zone
n, % 20
a
15 10 5 0
n, % 25
10
30
50
70
90
110
130 DH,nT
b
20
Shallow earthquakes Sub-crust earthquakes
15 10 5 0
40
80
120
160
200
240
280
320 T, min
a conventional (Martin et al. 2005). The dependence of the interval t (Fig. 2) from the maximum of substorm development versus the earthquake foci depth h of the subsequent seismic event reveals the following pattern: the deeper the earthquake source, the larger the delay time (Fig. 6b). But the interval t also depends on the degree of perturbation of the geomagnetic field. During world magnetic storms the t can be substantially shorten (sometimes to 4–7 hours), which disturbs the quasi-linear relationship between t and foci depth h.
Geomagnetic Disturbances and Seismic Events in the Vrancea Zone from in Situ Data
205
a T, min 200
(2) (14)
160 120 (1)
80
(5)
(24)
(2)
(12)
(22) (16)
(15)
40 0 40
b
60
80
100
120
140
160
180 h, km
τ, h 70 60
(22) (16)
(14)
50 40
(2)
(12)
(1) (24) (2)
30
(15)
(5)
20 10 0 60
80
100
120
140
160
180 h, km
Fig. 6 Relationship of (a) the earthquake depth h with the substorm length T (the T average over all shallow (<60 km) shocks is shown by triangles); (b) time intervals t with the substorm development maximum to the occurrence times of deep earthquakes in the Vrancea zone as a function of the source depth (the t was averaged over shallow shocks shown by triangles). The number of events is given in parentheses. The vertical bars are standard deviations
Discussion and Conclusions As mentioned above more than 90% of shocks recorded in the Vrancea zone are related to the mid-latitudinal manifestation of the near-midnight polar substorms. The substorms occur against both relatively calm magnetic field background and magnetic storms. We systematize the changes (“gradients”) in the magnetic field related to polar substorms in accordance with each seismic event. Four main types are distinguished (Fig. 7): First (I) : midnight polar substorm against relatively calm background; Second (II) : substorm precedes the main phase of a magnetic storm, i.e., before the maximum of the Dst variation; Third (III): substorm occurs at the decay stage of the Dst variation (the recovery phase); Fourth (IV): storm with sudden commencement serving as background for polar substorm. The other cases are rare, amounting to 8–10% of the total number, and not connected with substorms. In periods when substorms were absent at the magnetograms, deep earthquakes in the Vrancea zone were not observed at all.
206
F. Sedova et al.
Fig. 7 Four main types of “gradients” in the H-component of geomagnetic field caused by the development of midnight polar substorms (shown at the magnetograms of the Yastrebovka magnetic station)
Regarding the probable mechanism, we adhere to the hypothesis of trigger-like effect of magnetic storms on seismicity (Sobolev et al., 2001; Zakrzhevskaya and Sobolev, 2002). Although regular features of the relation of substorms to seismicity were inferred here only for one region, they should probably be taken into account in the study of the relation between the geomagnetically disturbed state and seismicity. In summary we emphasize that the seismicity in the Vrancea zone is connected with a definite type of geomagnetic disturbances, such as mid-latitude and nearmidnight polar substorm in the horizontal H-component of the geomagnetic field. This is confirmed by the following regular features inferred from the present study: (a) the geomagnetic disturbance is related to seismic events if the time interval from “gradient” of the magnetic field before a shock is taken into account; (b) an abrupt change (“gradient”) in the geomagnetic field (H-component) associated with the maximum of the polar substorm development precedes the seismic energy release: a seismic event occurs after a definite time interval t after substorm; (c) the time interval t from the maximum of the substorm development to an earthquake depends on the earthquake focal depth; (d) the energy class of the earthquake is
Geomagnetic Disturbances and Seismic Events in the Vrancea Zone from in Situ Data
207
connected with the amplitude of the H-component changes (during near-midnight polar substorm manifestation); (e) specific morphological features in the geomagnetic variation preceding the seismic energy release are revealed: four main types of variations are embraced more than 90% of related seismic events. These results do not contradict the assumption of a triggering mechanism of the magnetic storm effect on the seismicity but suggest a definite contribution of the substorms which are manifested at middle latitudes.
References Bakhmutov VG, Sedova FI, Mozgovaya TA (2006) Morphological features in the structure of geomagnetic variations in relation to earthquake in Vrancea zone. Geophys J 28(1):42–50 (in Russian) Bakhmutov VG, Sedova FI, Mozgovaya TA (2007) Geomagnetic disturbances and earthquakes in the Vrancea zone. Fizika Zemli 11:30–36 Kormiltsev VV, Kostrov NP, Ratushnyak AN, Shapiro VA (2002) The influence of electro-osmotic pressure generating by geomagnetic disturbances on the evolution of seismotectonic process. In: Nayakawa M, Molchanov OA (eds) Electromagnetic: lithosphere–atmosphere–ionosphere coupling. Terrapub, Tokyo, pp 203–207 Martin M, Wenzel F, the CALIXTO working group (2005) High-resolution teleseismic body wave tomography beneath SE Romania – II. Imaging of a slab detachment sce-nario. Geophys J Int 164(3):579–595 Rikitaky T (1979) Earthquakes prediction. Mir Press, Moscow, p 388 (in Russian) Sedova FI, Mozgovaya TA, Bakhmutov VG (2001) On morphological signs in the structure of geomagnetic variations on the eve and at the moment of earthquake in the Crimea-Black Sea and the Carpathian regions. Geophys J 23(4):61–68, in Russian Seismological Bulletin of Ukraine (IGF NAN Ukrainy, Simferopol, 1977, 1988–1996) (in Russian) Sobolev GA (1993) Bases of the forecast of the earthquakes. Nauka Press, Moscow, p 313 (in Russian) Sobolev GA (2003) Earthquake physics and precursors. (Nauka, Moskow (in Russian) Sobolev GA, Zakrzhevskaya NA, Kharin EP (2001) On the relation between seismicity and magnetic storms. Phys Solid Earth Russ Acad Sci 11:62 (in Russian) Tarasov NT, Tarasova NV, Avagimov AA, Zeigarnik VA (1999) The effect of high-power electromagnetic pulses on the seismicity of the Central Asia and Kazakhstan. Volcan Seismol Russ Acad Sci 4/5:152–160 (in Russian) Zakrzhevskaya NA, Sobolev GA (2002) On the seismicity effect of magnetic storms. Phys Solid Earth Russ Acad Sci 4:3–15 (in Russian)
First Steps Towards Monitoring Surface Ozone Dynamics at Ukrainian Stations Oleg Blum, Vira Godunova, Volodymyr Lapchenko, Oleksiy Perekhod, Yaroslav Romanyuk, and Mikhail Sosonkin
Abstract The atmospheric composition changes have a great influence on the global environment. To obtain a wide picture of the present-day state of the atmosphere, it is necessary to provide adequate monitoring of its gaseous and other components. The National Academy of Sciences of Ukraine continues to exert its efforts to develop an air monitoring network in the southeastern part of Europe – a hitherto poorly observed region. Since 2006, there have been three sampling points in Ukraine and a high-mountain station located at Terskol Peak in the Northern Caucasus. Currently, the continuous surface ozone measurements are continuing at this site. In this report, the description of monitoring sites and some results of observations are presented. Keywords Ozone • Air monitoring • Monitoring stations
Introduction Nowadays, countries, business, research institutions and communities work together to achieve the best possible results in understanding of climate variability and global change. In this connection, the monitoring of de Earth’s atmosphere from space has been a powerful approach that accounts for rapid progress made today. However, experience O. Blum National Botanical Garden, National Academy of Sciences of Ukraine, Kiev, Ukraine V. Godunova () International Center for Astronomical, Medical and Ecological Research, National Academy of Sciences, Kiev, Ukraine e-mail:
[email protected] V. Lapchenko Kara-Dag Natural Reserve, the Crimea, Ukraine O. Perekhod, Y. Romanyuk, and M. Sosonkin Main Astronomical Observatory of the National Academy of Sciences, Kiev, Ukraine F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_24, © Springer Science+Business Media B.V. 2011
209
210
O. Blum et al.
has already shown that many interpretations and theoretical models based on space observations require verification with in situ and surface measurements. Thus, even with the continuous improvement in space instruments and techniques, ground-level observations remain an integrated part of research on atmospheric composition change. The maintenance of sustainable development of European countries demands the exact knowledge of physical and chemical transformations of substances, which pollute the atmosphere, to increase the efficiency of the air quality management. Air pollutants have an adverse impact on both the environment and health. Ozone is the third largest contributor to the radiative forcing, which is the physical driving force of anthropogenic climate change. Of the two types of ozone – stratospheric and tropospheric – the latest one, which occurs at low atmospheric levels, is a pollutant of the highest category of toxicity; which concentrations are regulated by the legislations of all developed countries. The surface ozone increase may cause human mortality and morbidity, as well as corrosion to materials and large losses of crops. Thus, the quantity of ozone in the ground boundary layer was approximated as a significant air quality index. The continuous growth of surface ozone has been observed from the very beginning of the regular measurements of its concentrations in the air. It is caused by anthropogenic effects, namely operation of transport, agricultural activities, and combustion of petroleum products and natural gas. Ozone is not directly emitted in the atmosphere from these sources; it is a secondary pollutant, i.e. a product of photochemical reactions between primary pollutants (NOx, VOC, CH4, CO and others) of both biogenic and anthropogenic origins. The relationships between primary pollutants and ozone are, non-linear. Surface ozone is detected not only in urban and industrial areas but also in suburban and rural areas as well. The high emission density of reactive precursors in urban areas might lead to high ozone concentrations within the city or at short distances downwind. But ozone precursors may also be transported over distances of hundreds to thousands kilometers, resulting in ozone formation far from the sources (Gregg et al, 2003). Therefore, ground-level ozone is an important issue of European scale that requires a continuous control of ambient air. In 2002, the European Parliament and the EU council adopted guidelines (Directive 2002/3/EC), which set concentration limit values for ozone to protect human health and the environment (Table 1). In Europe and North America, the networks of monitoring stations have been developed, which track the content of both air ozone and its chemical precursors. There are 1,500 observing stations in Europe (Air quality in Europe 2003).
Study Area and Observation Sites During the last decade, environmental research has been the subject of systematic and intensive improvement in Ukraine. Thirty three substances are currently being monitored at 160 observation stations operated by the Ministry of the Environmental Protection of Ukraine. However, no surface ozone measurements are performed at
First Steps Towards Monitoring Surface Ozone Dynamics at Ukrainian Stations
211
Table 1 Definitions, target values and long-term objectives for ozone (from: Directive 2002/3/EC of the European Parliament and of the Council of February 12, 2002) Target value for the protection of human health 120 mg/m3 (maximum daily 8-h mean) Information threshold 180 mg/m3 (1 h average) Alert threshold 240 mg/m3 (1 h average) Warning threshold 360 mg/m3 (1 h average) Target value for the protection of vegetation 18,000 mg/m3 · h averaged over 5 years (parameter AOT40: the sum of the difference between hourly concentrations greater than 80 mg/m3 and 80 mg/m3 over the period from May to July) Forest protection 20,000 mg/m3 · h (parameter AOT40: the sum of the difference between hourly concentrations greater than 80 mg/m3 and 80 mg/m3 over the period from April to September) Materials protection 40 mg/m3 · h (1 h average) Long-term objective for the protection of vegetation 6000 mg/m3 · h (parameter AOT40: the sum of the difference between hourly concentrations greater than 80 mg/m3 and 80 mg/m3 over the period from May to July) Target value means a level fixed with the aim, in the long term, of avoiding harmful effects on human health and/or the environment as a whole, to be attained where possible over a given period. Alert threshold means a level beyond which there is a risk to human health from brief exposure for the general population and at which immediate steps shall be taken. Information threshold means a level beyond which there is a risk to human health from brief exposure for particularly sensitive sections of the population and at which up-to-date information is necessary. Long-term objective means an ozone concentration in the ambient air below which, according to current scientific knowledge, direct adverse effects on human health and/or the environment as a whole are unlikely. This objective is to be attained in the long term, save where not achievable through proportionate measures, with the aim of providing effective protection of human health and the environment.
these sites. Since the end of the 1990s, scientists from the National Academy of Sciences have been making efforts to bridge this gap. Their activities include the development and installation of observational stations for monitoring of ozone, its precursors and other gases in the atmosphere in order to develop a regular network in southeastern Europe. Since 2006, there are four sampling stations performing measurements of groundlevel ozone concentrations (Fig. 1): (1) National Botanical Garden (NBG) in Kiev City set up in 1995 (Blum et al. 1998); (2) Golosiiv Forest (Golosiiv) at the Main Astronomical Observatory of the National Academy of Sciences in the south of Kiev set up in 2006; (3) Kara-Dag Natural Reserve (Kara-Dag) in the Crimea located at the northeastern slope of Mount Svyataya (at the 270 m elevation), 1.5 km from the
212
O. Blum et al.
Fig. 1 Location of ozone monitoring sites
Black Sea coast. Moreover, the distance to the nearby industrial centers is more than100 km and to the health resort Koktebel is – 2.5 km. Some ozone observations were intermittently carried out at Kara-Dag in the 1990s but regular measurements started in 2006. The fourth site, Terskol Station, is located at Terskol Peak in the Northern Caucasus, Russia (43°16¢29²N, 42°30¢03²E, 3120 m) and operated by the International Center for Astronomical, Medical and Ecological Research (ICAMER). Due to remote location, this mountain site can serve as a background station providing monitoring of greenhouse gases and ozone deplleting substances.
Results and Discussion The results presented in this paper show the analysis of surface ozone concentration during 2004–2008 at four Ukrainian stations. Figure 2 indicates that a significant part of the variability of ozone in 2006 at Kara-Dag was linked to its seasonality. The site is characterized by the wide spring-summer ozone maximum with mean monthly concentrations of about 45 ppbv (90 mg/m3) and more; minimum concentrations are found in late autumn (20 ppbv). The amount of ozone depends mainly on meteorological factors (visibility, relative humidity and wind direction). Kiev’s data (NBG) showed ozone summer maximum during May–August (70 ± 10 ppbv) and winter minimum in November–December (20 ± 5 ppbv). Ground-level ozone concentrations at NBG in 2007 are presented in Table 2. During 2007 and 2008, the highest mean annual ozone concentrations observed were 51.3 mg/m3 in 2007 at NBG, 46.2 mg/m3 in 2008 at MAO, and 52.0 mg/m3 in 2007 at Kara-Dag. High ozone concentrations occur mainly during summer days with low air humidity (35–60%) and temperature exceeding 20°C in the afternoon.
First Steps Towards Monitoring Surface Ozone Dynamics at Ukrainian Stations
213
mg/m3
180 160 140 120 100 80 60 40 20 01.01.07
01.12.06
01.11.06
01.10.06
01.09.06
01.08.06
01.07.06
01.06.06
01.05.06
01.04.06
01.03.06
01.02.06
01.01.06
0
Fig. 2 Time series of ozone concentration measured at Kara-Dag in 2006
Table 2 Ground-level ozone concentrations in Kiev (the National Botanical Garden) in 2007 Mean 24-h concentrations Mean 1-h concentrations Month Mean monthly of the year concentrations (mg/m3) max (mg/m3) min (mg/m3) max (mg/m3) min (mg/m3) January 42.48 49.42 35.16 66.52 15.32 February 32.98 60.82 4.86 77.42 2.12 March 56.75 78.28 26.80 119.96 5.66 April 67.37 93.96 47.36 126.60 4.06 May 76.06 99.28 39.46 148.88 3.14 June 76.68 99.76 46.18 153.24 5.88 July 79.37 111.42 48.74 201.40 2.74 August 66.58 96.50 41.34 173.20 2.20 September 44.21 79.74 21.94 119.30 2.06 October 29.47 50.48 2.14 103.54 2.08 November 24.24 47.48 8.58 62.60 2.02 December 19.06 35.08 4.34 47.26 2.72
The number of exceedances of the 1-h threshold limit at NBG in the period 2007– 2008 is shown in Fig. 3. Figure 4 shows diurnal variations of surface ozone for different seasons of the year, which were observed for Kiev (NBG). For the typical urban environment, the highest ground-level ozone concentrations are observed between 12.00 a.m. and 3.00 p.m., when solar radiation is at its maximum and when, due to vehicle emissions, the ratio NO2/NO increases.
214
O. Blum et al. Year
Cover, % of measurements
>60
>80
>100
>120
>140
>160
>180
Max. 1-hour (mg/m3)
2007
90
3051
1544
636
185
46
12
5
197,4
2008
87
2309
1180
477
149
58
21
2
183,9
10000 2007
Number of exceedences
2008
1000
100
10
1 > 60
> 80
> 100 > 120 > 140 > 160 1-hour ozone concentration, mg/m3
> 180
Fig. 3 Total number of events in 2007 and 2008 when hourly ground-level ozone concentrations exceeded the critical levels of 60–180 mg/m3, NBG station
Fig. 4 Daily variability of ground-level ozone concentration in Kiev (NBG)
First Steps Towards Monitoring Surface Ozone Dynamics at Ukrainian Stations
215
50 45 40 35
ppbV
30 25 20 15 10 5 0
22 23 24 25 26 27 28 29 30 31 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 July-August 2006
Fig. 5 Comparison of ozone concentrations measured at two sites in Kiev: at NBG (half-tone line) and at Golosiiv (black line)
Comparison of July–August 2006 ozone measurements at two sites in Kiev indicates that: (1) the two curves have similar trend, and (2) ozone concentrations at Golosiiv are significantly higher (due to its location downwind from industrialized region) than those at NBG, which is located in the urban core (see Fig. 5). Continuous high-frequency automatic measurements of surface ozone at the high-altitude station at Terskol in Caucasus indicate that there was no increase in surface ozone since the summer of 2003. Ozone concentrations show a clear seasonal cycle with maximum values from May to August (up to 85 ppbv) and minimal values in November–December (down to 20 PPbv) (Godunova et al. 2006). Daily variability of ozone concentrations is mainly related to the following factors (in addition to usual photochemical processes): variations in meteorological conditions, horizontal advection and vertical exchange processes over mountain terrain. However, other natural events, such as intrusion of stratospheric ozone, could induce somewhat larger fluctuations of the ozone mixing ratio and cause the average daily ratio to vary by about 10 ppbv around its monthly average. Figure 6 shows a daily ozone concentration profile, which is typical during a sunny windless day for the period from late spring to mid-autumn. There is a slight increase during the first half of the day and a decrease in the afternoon. Mean daily values typically do not exceed 50 ppbv between September and February and are, on average, 40 ppbv in this period and 60 ppbv and higher between April and August.
216
O. Blum et al.
11.09.06 60 50
ppbv
40 30 20 10 0 0:00
6:00
12:00
18:00
0:00
Fig. 6 Daily variability of ground-level ozone concentration at Terskol Peak during a sunny windless day July-August 2006 80
July-August 2007
70
ppbV
60 50 40 30 20 10 7-Jul
17-Jul
27-Jul
6-Aug
16-Aug
Fig. 7 Mean daily concentrations of ground-level ozone at Terskol Peak in summer 2006 and 2007
Figure 7 illustrates little difference in daily averages of ozone concentrations in summer at Terskol.
Conclusions Ground-level ozone variability at the four sites in southeastern Europe has been studied based on ozone records covering the period from 2004 to 2008. Three of these sites are situated at different locations in Ukraine; the fourth station is set up at an altitude of 3125 m in the Northern Caucasus, Russia. Since air quality in this part of Europe still needs more investigation, it is imperative to develop here a monitoring network, which would include these stations, as well as additional sites to be set up in Ukraine and in Russia. Data sets from this region could significantly contribute to an effective control of atmospheric composition over continent. Other tasks to be performed in the near future might include: the development of technical
First Steps Towards Monitoring Surface Ozone Dynamics at Ukrainian Stations
217
cooperation in air quality monitoring; distribution of ozone information to the public; and enhancement of educational attainments in atmospheric sciences.
References Air quality in Europe: state and trends 1990–1999 (2003) (Topics report No.4/2002). EEA, Copenhagen, 2003 Blum O, Butnerowicz A, Maning W, Popovicheva L (1998) Ambient tropospheric ozone in the Ukrainian Carpathian mountains and Kiev Region: detection with passive samplers and bioindicator plants. Environ Pollut 98(3):299–304 Directive 2002/3/EC of the European Parliament and of the Council of 12 February 2002 relating to ozone in ambient air. OJ L 67/14-30 (2002) Godunova V, Sosonkin M (2006) Terskol observatory as a regional station for monitoring air quality in eastern europe. In: Granier C et al (ed) Air quality in eastern europe. A review of measurement and modelling practices and needs. Report 8.2006 of the ACCENT/JRC Expert Workshop Gregg J, Jones C, Dawson TE (2003) Urbanization effects on tree growth in the vicinity of New York City. Nature 424:183–187
Satellite Monitoring of Nitrogen Oxide Emissions Igor Konovalov, Matthias Beekmann, Andreas Richter, and John Burrows
Abstract The efficient control of air pollutant emissions into the atmosphere is important for sustainable development. Remarkable recent progress in satellite measurements of the composition of the troposphere has opened new prospects for monitoring of air pollution and emissions of pollutants. This paper presents examples of using satellite measurements of atmospheric composition for estimation of long-term changes of emissions in nitrogen oxides which are important air pollutants playing a major role in photochemical smog formation. The estimations are based on the use of inverse modeling methods enabling combination of the data for tropospheric NO2 column amounts derived from the long-term (1996–2008) GOME and SCIAMACHY satellite measurements with simulations performed by the CHIMERE chemistry transport model. Keywords Air quality • Pollutants • Nitrogen oxides • Chemistry transport model • Satellite measurements
Introduction The emissions of pollutants into the atmosphere have strong impact on both air quality and climate. Accordingly, the efficient control of these emissions into the atmosphere is one of the important conditions for a sustainable development. Usually, emissions are estimated by compiling the available information about emission I. Konovalov () Institute of Applied Physics, Russian Academy of Sciences, Nizhniy Novgorod, Russia e-mail:
[email protected] M. Beekmann Université Paris-Est and Université Paris 7, Créteil, France A. Richter and J. Burrows University of Bremen, Bremen, Germany
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_25, © Springer Science+Business Media B.V. 2011
219
220
I. Konovalov et al.
sources (e.g., industry, power plants, transportation etc.) and by applying pre- estimated emission factors to individual sources. However, this procedure frequently involves incomplete or uncertain information, as well as many assumptions that are difficult to verify. An alternative way for emission estimation is provided by inverse modeling approach (see, e.g., Enting 2002). The idea of this approach is to derive the information about emissions directly from atmospheric measurements by using a chemistry transport model. Remarkable recent progress in satellite observations of the troposphere composition has opened new prospects for monitoring air pollution and emissions of pollutants. In particular, it has been demonstrated that satellite measurements of nitrogen dioxide (NO2) allow estimating anthropogenic emissions of nitrogen oxides and their multi-annual changes (e.g., Martin et al. 2003; Richter et al. 2005; Konovalov et al. 2006, 2008; Konovalov 2007). Oxides of nitrogen (NOx) are important air pollutants affecting the oxidising properties of the atmosphere and playing the major role in the photochemical smog formation. In this paper, we provide an overview of our recent inverse studies aimed at estimation of multi-annual changes of NOx emissions from satellite measurements of tropospheric column amounts of nitrogen dioxide. Accurate independent estimates of past changes in NOx emissions provide new ways for the evaluation of the efficiency of air pollution control measures, for the assessment of uncertainties in emission cadastres, and for testing the ability of chemistry transport models to reproduce past and predict future changes in atmospheric composition.
Measurement and Model Data Satellite Data Tropospheric NO2 columns derived from satellite measurements by IUP, University of Bremen, have been used. Seven years (1996–2002) of GOME measurements (Burrows et al. 1999) were complemented with 6 years (2003–2008) of SCIAMACHY measurements (Bovensmann et al. 1999). The GOME and SCIAMACHY instruments provide measurements of NO2 columns at the horizontal resolution of 320 × 40 km2 and 60 × 30 km2, of global coverage at the equator achieved in 3 and 6 days, respectively. We use the same data-products for tropospheric NO2 columns derived from satellite measurements and analysed earlier in Richter et al. (2005), where a general description of the retrieval method can be found. We considered only summertime measurements because of prevailing cloudy conditions in eastern Europe during the cold season. A pre-processing stage (specific for this study) includes, the projection of daily data for tropospheric NO2 columns onto a 1° × 1° grid and averaging of all the data for 3 summer months (June–August) of each year. In order to avoid systematic “jumps” in the time series of the gridded NO2 columns between 2002 and 2003 when linking the lower resolution GOME data and the higher spatial resolution SCIAMACHY data, an additional transformation of data was performed using two
Satellite Monitoring of Nitrogen Oxide Emissions
221
different methods. The first method involves deconvolution of the GOME NO2 data. This method is very similar to the one suggested in our earlier study (Konovalov et al. 2006). The basis for this approach is to superimpose the spatial structure of the NO2 columns derived from the SCIAMACHY measurements over the spatial structure of the NO2 columns derived from the GOME measurements. The second method is used to obtain consistent time series for NO2 columns over megacities and is used only in the framework of the nonlinear version of our inversion algorithm (see Section 3.2). The idea is to simulate the smoothing of the spatial structure of NO2 columns, introduced by the GOME measurements. The method involves the convolution of the NO2 columns from SCIAMACHY:
2m [ j − m]2 cos(f )2 cs(conv c exp ≈ ∑ s (i − m + j ) i) − 2 wc 2 j=0
(1)
where cs(i) are the original NO2 column amounts in the grid cell i, m is the number of grid cells on the longitudinal plane within 320 km (the typical resolution of the GOME measurements), f is the latitude, and wc is a scaling factor. Here, we approximate the smoothing introduced by the GOME window on the longitudinal plane into the seasonally averaged NO2 columns by the Gauss function with an efficient width (wc/cos(f))2. The parameter wc was estimated to be equal 0.88 by minimizing the mean squared differences between the convoluted NO2 columns for 2003 and the original NO2 columns (from GOME) for 2002 over 14 megacities considered in Section 4.2. These convoluted NO2 columns were used in our analysis instead of the original NO2 columns from SCIAMACHY.
Simulated Data To generate the simulated NO2 columns, the CHIMERE CTM was used (http://www. lmd.polytechnique.fr/chimere/). The model takes into account all important processes that determine the evolution of nitrogen oxides released into the atmosphere, such as gas-phase reactions which define the chemical balance between NOx species and their transformation to nitric acid, dinitrogen pentoxide and organic nitrogen compounds. Dry deposition and wet scavenging which are responsible for the removal of the reactive nitrogen compounds from the atmosphere; advective transport, eddy diffusion and deep convection. The simulated NO2 columns were sampled consistently in space and time with the measurement-based daily NO2 columns. In this study we use a spatial domain that covers all of Europe, the Mediterranean area and the Middle East with a horizontal resolution of 1° × 1°. The model runs were performed with 12 layers defined as hybrid coordinates. The top of the upper level was fixed at 200 hPa pressure level. Meteorological input data were calculated off-line with horizontal resolution of 100 × 100 km2 using the MM5 non-hydrostatic meso-scale model (http://www.mmm.ucar.edu/mm5/). MM5 was initialized with NCEP Reanalysis-2 data (http://www.cpc.ncep.noaa.gov/products/wesley/ncep_data/).
222
I. Konovalov et al.
The anthropogenic emission data used in this study are based on the so-called “expert” annual data of the EMEP emission inventory (Vestreng et al. 2005) for the years 1996–2004. We have used the two versions of the expert data that were available at the EMEP web site (http://webdab.emep.int/) on a 0.5° × 0.5° grid in autumn 2006 (before 30 November) and in spring 2007. The comparison of the older and the newer versions of the emission data gives some idea about uncertainties in the bottom-up inventories which is helpful for validation of emission estimates obtained in this study, particularly because more recent emission data are expected to be more accurate than the older ones. As it is shown below, the differences in trends between the two versions of the EMEP data are, in some cases, rather large. CHIMERE was run independently for each summer season starting on 24 May with the same initial and boundary conditions. A “base” decadal run of CHIMERE was performed with the EMEP emission data for the year 2001. Only the results of the base run are used in the inversion algorithm. Additional control runs were performed with emission estimates obtained in this study.
Method Linear Approach Following the standard Bayesian approach (see e.g., Enting 2002), assuming that uncertainties in inter-annual variations of NO2 columns satisfy the normal probability distribution and linearizing the modeled relationship between perturbations of the NO2 columns and NOx emissions, we get the following probability distribution for the interannual changes of NOx emissions constrained by observations:
p( ∆E | ∆Co ) ∝ 2 n +1 n ∂(Cmi + Cmi ) 1 N N exp − ∑ ∑ ∆ E j |E = E0 − ∆ (Cio − Cim |E = E 0 ) s ci−2 pa ( ∆E) (2) 2 ∂E j 2 i =1 j =1
where E is a vector of emission estimates (a subscript “0” denotes a base case defined above), Co and Cm are the observed and modeled NO2 columns, sc is the standard deviation for the uncertainties in the NO2 columns, i and j are the indexes of a grid cell, n is the number of a year, D is an operator of an interannual variation (e.g., DE(n) = En+1 − En) and pa is the a priori probability distribution (specified below) for interannual changes in emissions. The idea behind the distribution (2) is very simple: if we have some differences between inter-annual changes in NO2 columns from observations and the model with constant emissions, we assume that this difference is probably due to corresponding changes in NOx emissions. As common in inversion methods, we are looking for the maximum likelihood a posteriori estimates of DE that yield a maximum of p. Note that the vector E here represents
Satellite Monitoring of Nitrogen Oxide Emissions
223
only anthropogenic NOx emissions, because available data on the consumption of fertilizers suggest that changes in biogenic emissions in the concerned regions were rather small during the considered decade. In order to ensure that the a posteriori estimates of emissions are always positive, we seek the solution in terms of natural logarithms of emissions, denoted below as e (that is, we replace E with exp(e)). As it is common in geophysical studies, we characterize the long-term changes in the considered atmospheric characteristics by the linear trend and the year-to-year variability superimposed over the trend:
e n = e + (n − n )∆ t e + ∆ d e n
(3)
where the subscripts “t” and “d” denote the trend (independent on the year) with the deviation from the trend, respectively. Accordingly, our inversion procedure consists of two major steps. First, we estimate Dte by finding values of Den which provide a maximum of the distribution (2) in which D(C0 − Cm) is replaced by the linear trend in the difference (C0 − Cm), that is, Dt(C0 − Cm); the estimated Den for each pair of neighboring years are averaged. At this step we have also to define the probability distribution function for a priori emission estimates, pa. As we do not dispose of any specific information on uncertainty in the EMEP emission data which could provide a priori estimates for emission trends, we put pa as a constant inside realistic limits (lmin < Dt e < lmax) and zero outside. Limits are chosen as the minimum and maximum values of linear trends in the EMEP NOx emissions data considered on a 1° ´ 1° grid of our model. This gives values of lmin = −0.07 and lmax = 0.1. This procedure gives an estimate of the spatial distribution of the NOx emission trends which is practically independent of the corresponding distribution based on the “bottom-up” inventory, while avoiding unrealistically high magnitudes of emission trends. Second, we estimate deviations from the trend, Dde, that provide the maximum of the distribution (2) where D(C0 − Cm) is replaced by Dd(C0 − Cm) and pa is a Gaussian distribution for (De)d with the standard deviation equal 0.14. This value is chosen to adjust the variance of the a posteriori deviations from the trends in Great Britain to the variance calculated for the deviation from the trends in the “new” EMEP data for the same country (about 4.2%). Note that typical magnitudes of interannual variations in our a posteriori emission estimates are determined not only by parameters pa, but also by the values of sc. Although the choice of Great Britain is rather arbitrary, it seems reasonable taking into account that we consider ground based measurement of NOx in this country for validation of our results. Values of the standard deviation for uncertainties in the NO2 columns, sc, are estimated from above as a function of magnitude of the measured NO2 columns by calculating the mean squared differences between the inter-annual changes of the measured and modeled NO2 columns within a “moving window” (note that only the variation of sc from one grid cell to another is of interest in Eq. (2)). This approach is described in Konovalov et al. (2005). The such defined sc is found to increase monotonically from about 0.25 to 2.5 as Co increases from about zero to 15 (always in molecules · 1015 cm−2).
224
I. Konovalov et al.
Finding an accurate mathematically solution of our inversion problem would be computationally too expensive. Instead, we use approximate methods that were developed in our earlier studies. The main idea is to substitute the original model for a set of linear statistical models describing the relationships between perturbations of NOx emissions and NO2 columns approximately (Konovalov et al. 2006). Such an approximation is possible because the lifetime of freshly emitted NOx is rather limited in summer. Our method allows us to take into account the transport of NOx on the scale of three neighboring grid cells (that is, 200–300 km). The statistical models were created by performing 100 model runs with randomly perturbed NOx emissions for each year independently. The optimal estimates of Dte and Dde for each pair of years are obtained by means of the iterative steepest descent method using zero and the obtained estimates of Dte as initial guesses, respectively. Although a sufficiently accurate estimation of the uncertainties in our results is hardly possible because of the lack of knowledge about numerous factors that can contribute to these uncertainties, we tried to get rough estimates of uncertainties as follows. First, we considered how deviations from the trend in the columns, Dd(C0 − Cm), can influence the estimates of Dte. Although such deviations are not necessarily due to uncertainties, this approach allows us to estimate, at least, the upper limit of the respective uncertainties in Dte. Technically, we performed a Monte-Carlo experiment based on the bootstrapping method: each vector Dd(C0n − Cmn) corresponding to some pair of years (n, n + 1) is attributed to another pair (e.g., l, l + 1), where l is a random number. The experiment included 100 inversions with such randomly mixed deviations. In order to estimate uncertainties in the emission trends associated with the numerical method and approximations we (i) replace the observed NO2 columns values with the those calculated by the model with the a posteriori emissions (serving as a substitute for an exact solution), (ii) perform the inversion, and (iii) calculate the differences between the results and the “exact” solution. These differences are summed up with the “random” uncertainties in the trends. To take into account uncertainties related to systematic biases in the modelled and measured NO2 columns, we scaled the measured NO2 columns by the mean ratio of the simulated (for the base case) and measured NO2 columns in a given grid cell and repeated the inversion procedure with such modified inputs. If the biases were caused by errors in NOx emissions specified for the base case, then such scaling would probably yield most accurate estimates of relative changes in emissions. But since we do not know the actual reason for the disagreements between the model and measurements, we attribute the differences between the results obtained with the original and modified input data to the uncertainty in our estimates of emission trends found with the original data for measured NO2 columns. Contributions from all different sources of uncertainties discussed above are summed up quadratically to provide an estimate of the overall uncertainty in the NOx emission trend for a given grid cell or country. Note, finally, that we do not attempt here to estimate a part of uncertainties in our results due to essentially unknown systematic uncertainties in the modeled or measured NO2 columns.
Satellite Monitoring of Nitrogen Oxide Emissions
225
However, it seems very unlikely that systematic errors in the NO2 columns exceed the random uncertainties estimated as it is mentioned above.
onlinear Approach (the Case of Emissions N from Megacities) Formulation of the problem. In a general case, our task can be formulated as follows. Let’s assume that we want to evaluate temporal changes of the value x, while in our disposal are only noisy inaccurate measurements y: y i = x i + e i , i = 1, …, n, where i is the temporal index (e.g. the number of year) and e are the errors. And let us make three assumptions. First, the noise is uncorrelated (white): áeiejñ = 0 for any i ¹ j, where the angled brackets denote averaging over a statistical ensemble. Second, the changes of the real value x are small in comparison with the level of noise (xi+1 − xi)2 « áe2ñ for any i. And third, e satisfies the normal distribution. The goal of the method is to filter out the noise without constraining the type of the temporal evolution of x. That is, our aim is to obtain a series of values xe, such that (xei − xi)2 « áe2ñ for any i. In principle, the second condition can be further relaxed; however it is not necessary in the framework of the given study. The special requirements are the self consistent estimation of uncertainties of the results and their applicability to short time series. General description of the algorithm. The basic ideas of our algorithm are the following. First, we use a neural network (of the perceptron type) for approximation of the unknown nonlinear trend: N
xei (w) = w0 + wL i + ∑ wk gk , k =1
gk =
1 , 1 + exp (wˆ k i + wˆ 0 k )
(4)
where w are weight coefficients, and N is the number of neurons. Indeed, it is well known that a neural network is a universal approximator; that is, given a sufficient number of neurons, it can approximate any smooth function with any given accuracy. Second, we follow the probabilistic approach applied here for estimation of weight coefficients of the neural network. Specifically, using Bayes’s theorem, we get the following a posteriori probability distribution function (pdf) for the weight coefficients.
n (y − x (w))2 i ei pa ( w), pa = const | w < wmax , p (w | y ) ∝ exp − ∑ 2s ε2 i =1 pa = 0 | w > wmax
(5)
where se2 = áe2ñ. Note that we a priori constrain only the maximum magnitude of w. In principle, we could specify the a priori pdf in different ways. The rectangular structure of the a priori pdf was chosen as the result of preliminary experiments with both artificial and real data. It was found that this simple structure enables both
226
I. Konovalov et al.
an efficient filtering of noise and high sensitivity of the algorithm to actual nonlinearities in input data. The maximum likelihood estimates of w can then be found as follows: n
wˆ (wmax , N ) = arg min ∑ (yi − xei (w )) , w ∈[− wmax ; wmax ] 2
(6)
i =1
Third, we use the leave-one-out method of the cross validation in order to filter out the noise. The idea is to minimize the difference between a given measurement and its approximation which was built without using this measurement. Specifically, following the Bayesian approach, we can define the probability distribution for the parameters involved in (5, 6):
n (y − x (wˆ ))2 i ei p (wmax , N | y ) ∝ exp − ∑ i =1 2s ε2
(7)
Here, the estimate xei is obtained without using corresponding measurement yi. We assume that the difference between yi and xei is due to noise. By finding the maximum of this distribution, we define the optimal constraints for weight coefficients, wmaxх. Technically, this optimization is carried out by means of the onedimensional golden search method. In principle, a similar procedure could also be used to estimate the optimal number of neurons. However, in practice, it is also necessary to take into account that the uncertainties of estimates xei obtained with a larger number of neurons are larger. Besides, the differences between estimates obtained with different number of neurons are frequently too insignificant. Thus we define the optimal number of neurons in a different way. Specifically, we find the estimates of the trends consecutively with N = 0, N = 1, and so on, and each time we check the difference between xe(N) and xe(N + 1). When this difference becomes statistically insignificant (in terms of 68.3 percentile), the procedure is stopped. The corresponding value of N is considered optimal. We can also estimate the level of noise as follows: 2 1 n (8) ∑ (yi − xei (wˆ )) n i =1 This estimate is further used to assess the uncertainties in results by means of the Monte Carlo method. Specifically, we sample the errors, ei, from the normal distribution with se defined by Eq. (8) and add these errors to the xei (obtained using all data points). Then the whole estimation procedure is repeated (with the fixed optimal N) many times (in this study, 300) and the 68.3 percentile of the statistical distribution of the obtained estimates xe is evaluated. Because of the random errors e in the original dataset, there is a probability that our algorithm will detect a nonlinear trend even when the real values x demonstrate only a linear trend or no trend at all. In order to assess this probability, we performed an additional Monte Carlo experiment. Namely, we sampled an entirely random time series from the normal distribution having the same standard deviation as the original input data, and applied our algorithm to these random
σ ε2 ≈
Satellite Monitoring of Nitrogen Oxide Emissions
227
data. The 68.3 percentile of the distribution of xei obtained this way was then added to the uncertainty range of the optimal estimate of the trend. Before application to real data, the algorithm was tested with artificial time series representing both “ideal” and “noisy” cases. In the latter, the level of noise corresponded to that in the real data sets which we used in this study. In all the cases considered, the algorithm has managed to retrieve the character of the actual trend and to estimate the uncertainties properly. Combining measured and simulated NO2 columns over megacities. We assume that the dependence of NO2 columns on NOx emissions can be approximated as C(ti) = Cb(ti) + a(ti)E(ti), where C(ti) is the seasonally averaged NO2 column amount over a given megacity for a year i, Cb(t) is the “background” NO2 column amount which is not related to emissions from the given megacity, E(ti) are the seasonally average NOx emission rate and a(ti) is the sensitivity of the NO2 columns to changes of the NOx emissions. In special tests this linear approximation was found to be sufficiently accurate. Having in mind this approximation, we estimate the normalized annual emission rates as follows:
Co (ti ) − Cbm (ti ) E( t i ) / E 0 ≅ C m (ti ) |E = E0 −Cbm (ti )
(9)
where the indexes “o” and “m” denote the observed and modeled data, respectively, and E0 are emissions for the reference year (2001). By employing the model in the context of Eq. (9) we attempt to account for those variations in NO2 columns that are due to meteorological variability. The part of the meteorological variability that cannot be explained by our model is treated as random noise. Because values of emission rates are positive, it is reasonable to assume in a general case that their uncertainties satisfy lognormal distribution. Accordingly, we use the natural logarithms of the estimates defined by Eq. (9) as input data (y) for our analysis. Such time series were composed for 13 largest cities and urban agglomerations (with the total population more than four millions) covered by the domain of our model. Additionally, in case of Cairo, for which the EMEP emission data are not provided, the NOx emission changes were estimated directly from the measured NO2 columns: E(ti)/E(t0) @ Co(ti)/Co(t0), where t0 indicates the year 2001.
Results Decadal Trends in NOx Emissions in the Period from 1996 to 2005 Figure 1 presents time series of NOx emissions averaged over several countries. Rather close agreement between the new EMEP data and our estimates is found,
228
I. Konovalov et al.
Fig. 1 Time series and linear trends in anthropogenic NOx emissions averaged over several countries. Values provided in round brackets are estimates of uncertainties. Values reported in square brackets are the statistical uncertainty (the standard deviation) of a linear fit to the EMEP data (not including systematic uncertainty)
particularly for France, Germany and Great Britain. It is important to note that our estimates of the trends for these countries are in better agreement with new EMEP emission data than with old ones. Statistically significant differences between our results and the expert estimates are found for Italy, where our data suggest that the decrease in NOx emissions in the EMEP inventory is strongly overestimated. The differences between our results and the expert emission data are larger outside of Western Europe. Particularly, the directions of the trends in the measurements and the new EMEP data differ in Russia and Turkey. In Russia, there is also a big difference between the old and new EMEP data, and the new EMEP emissions do not agree with our estimates than the old ones. These observations and other results not shown here indicate that the current knowledge about emissions and their inter-annual changes in former USSR countries, in the Balkans and in the Middle East is still very incomplete and probably inadequate.
Estimates of Nonlinear Trends in Megacities in the Period from 1996 to 2008 Our nonlinear estimates of NOx emission trends for several megacities (with the total population of more than four millions) are shown in Figs. 2 and 3. Nonlinear trends are detected in Bagdad, Barcelona, Madrid, Milan, Moscow and Paris.
Fig. 2 Estimates of NOx emission trends in several large cities. The interannual changes (solid line with dots) are evaluated as a local slope of the trends. The EMEP emissions were averaged over 10 grid cells of our model, surrounding the city center (5 grid cells in the west-east direction and 2 grid cells in the south–north direction).
Fig. 3 The same as in Fig. 2 but for other cities
230
I. Konovalov et al.
These results should be considered as preliminary. Specifically, for Madrid and Moscow we got positive changes in 1990s and negative changes in more recent years. An interesting nonlinearity which indicates the growth of the rate of the negative trend in recent years is found also for Paris. Most probably, the latest tendencies in Madrid, Milan, Moscow, and Paris are related to the increase of the fraction of modern cars equipped with catalytic converters. In Bagdad the change of the direction of the trend is chronologically associated with the well known events of 2003. In the cases of Alexandria, Berlin, London, and Ruhr agglomeration nonlinearities are not detected. Our method gives negative linear trends. In Tehran, the nonlinearities are also not revealed but an interesting result here is a strong positive linear trend (about 5%). This trend can be regarded as an indication of stable development of the Iranian economy. Smaller positive linear trends are also revealed in Istanbul and Cairo. The EMEP data are in reasonable agreement with our estimates in most of the considered cities in Western Europe (Barcelona, Berlin, Madrid, London and Ruhr agglomeration) and also in Istanbul. In Paris, the EMEP data show a considerably stronger downward trend than that in our estimates until 2005, which is followed by an upward “jump” between 2005 and 2006. It seems very unlikely that emissions in such a big city as Paris can “suddenly” increase by about 15% during 1 year. Most probably, this jump and similar irregularities in EMEP data for several other cities are due to changes in methods used by EMEP for spatial allocation of emissions. In Milan, satellite data do not confirm the negative trend predicted by EMEP. The same kind of disagreement between satellite measurement based estimates and the EMEP data was found for the whole of Italy in the case of decadal trends discussed above (see Fig. 1). In Alexandria our analysis yields a negative trend, while the EMEP data show a positive trend. In Moscow, our estimates are in reasonable agreement with the EMEP data in the period from 1996 to 2005, but strongly disagree in the latest years. The fact that the EMEP data for Bagdad and Tehran are constant indicates that EMEP did not have sufficient information about emissions in these cities. The emissions for Cairo are not provided by EMEP. Taking into account that our model (based on the EMEP emission data) could not simulate the NO2 column over Cairo properly, the NOx emission trend in this city was retrieved directly from satellite data for NO2 columns. This approach can be associated with some underestimation of the magnitude of emission trend (up to about 30% according to our analysis).
hecking the Agreement Between the Measurement C Data and Simulations Obtained with the Measurement-Based Emission Estimates Optimization of NOx emissions should lead to improvements in the agreement between the modelled and measured data. A rather critical test of our results can be provided by comparison of independent data such as near surface NOx concentrations
Satellite Monitoring of Nitrogen Oxide Emissions
231
with corresponding simulations performed with the estimated NOx emission trends. Specifically, in order to validate the results of the inverse modelling, the following near-surface measurements have been used: (i) measurements of nitrogen oxides from the United Kingdom Automatic Urban and Rural Network (AURN) between 1996 and 2008, (ii) the measurements of NOx in Madrid (2000–2007) and NO2 in Milan (1999–2007) (the data obtained from the Airbase data base), and (iii) the measurements of NOx concentrations in Paris from Airparif. The raw hourly data have been processed to yield the seasonally averaged (over the summer months) daily mean NOx (or NO2) concentrations. Note that the amount of data available for validation is very limited particularly because the multi-annual measurements have to be performed at the same sites. Because of the low spatial representativeness of surface measurements (compared to satellite measurements), we have to combine data from several sites. The idea of validation of our nonlinear trend estimates is to calculate linear trends of concentrations over the periods where the corresponding nonlinear trends are monotonous. The results are presented in Figs. 4 and 5. Clear improvements (see Fig. 4) both in the trends and RMSE (defined as the RMS difference between respective time series) calculated for NOx concentrations in UK
Fig. 4 Comparison of NOx results of a model run for which changes of NOx emissions were specified using either (1) results of this study or (2) the most recent and (3) older expert data of the EMEP inventory with (4) measurement data for NOx near surface concentrations at AURN in Great Britain. The measurement and modelled data from individual monitors were combined to equalise contributions from sites with different level of air pollution. One sigma uncertainties in trends are indicated in brackets
232
I. Konovalov et al.
Fig. 5 Comparison of measurement data for NOx (or NO2) near surface concentrations in Madrid, Milan, Paris and London with corresponding results of a model run for which changes of NOx emissions were specified using both nonlinear and linear estimate of NOx emission trend (except for London where only a linear trend is estimated). The linear trends in concentrations and emissions are evaluated separately for the periods indicated below the figures
take place when the old EMEP data are replaced either with the new expert data or with our own estimates. The difference between results obtained with our emission estimates and new EMEP data is very small and statistically insignificant: the RMSE calculated with our emission estimates is slightly larger than that obtained with the new EMEP data, the trend in simulated data gets slightly closer to the trend in the measurement data, and improvements in the agreement between the data for
Satellite Monitoring of Nitrogen Oxide Emissions
233
individual years take place only for 6 years. These results indicate that the new EMEP data for Great Britain are already rather good. In the case of megacities (see Fig. 5), we have an almost perfect agreement between the linear trends in measured and simulated NOx concentrations in London. In the other cities considered, the simulations performed with our nonlinear estimates of emissions are more consistent with measurements than the simulations based on the linear estimates of emission trends. Specifically, the changes of “partial” linear trends in measurements are significantly better reproduced in simulations based on the nonlinear than linear emission trends.
Conclusions Our studies presented in this paper demonstrate the feasibility of monitoring NOx emissions by means of satellite. Specifically, we have shown that satellite measurements can be used for estimating multi-annual changes of NOx emissions either considered on a regular grid or representing the sources of air pollution in megacities. Here we investigated the decadal changes (between 1996 and 2005) in NOx emissions in Europe, the Mediterranean and the Middle East and estimated nonlinear trends of NOx emissions in several megacities over the period of 13 years (from 1996 to 2008). We used the data on tropospheric NO2 column amounts derived from the long-term GOME and SCIAMACHY measurements which were combined with calculations performed with the CHIMERE chemistry transport model. Our results indicate that, in agreement with expert estimates, NOx emissions in Western Europe have been mostly decreasing. Much larger differences between the satellite measurement based estimates and the EMEP data was found outside of Western Europe, indicating that emission inventories for those territories are not adequate. The analysis of NOx emission changes in megacities revealed statistically significant nonlinearities in emission trends in Baghdad, Barcelona, Madrid, Milan, Moscow and Paris. In particular, the NOx emissions in Moscow increased in 1990s but show a negative trend in more recent years. In Paris, the decrease of NOx emissions accelerated since about 2002. Available independent ground based measurements of nitrogen oxides in Madrid, Milan and Paris suggest that our nonlinear estimates of NOx emission trends describe actual emission changes more adequately than linear trends. Acknowledgements This research was funded by the Russian Foundation for Basic Research (grant No. 08-05-00969-a) and the European Commission through the GEOMON FP6 project.
References Bovensmann H, Burrows JP, Buchwitz M et al. (1999) SCIAMACHY – mission objectives and measurement modes. J Atmos Sci 56:127–150 Burrows JP, Weber M, Buchwitz M et al. (1999) The Global Ozone Monitoring Experiment (GOME): Mission concept and first scientific results. J Atmos Sci 56:151–175 Enting IG (2002) Inverse problems in atmospheric constituents transport. Cambridge University Press, New York
234
I. Konovalov et al.
Konovalov IB, Beekmann M, Vautard R et al (2005) Comparison and evaluation of modelled and GOME measurement derived tropospheric NO2 columns over Western and Eastern Europe. Atmos Chem Phy 5:169–190 Konovalov IB, Beekmann M, Richter A, Burrows JP (2006) Inverse modelling of the spatial distribution of NOx emissions on a continental scale using satellite data. Atmos Chem Phys 6:1747–1770 Konovalov IB (2007) Regional differences in decadal changes of the atmospheric emissions of nitrogen oxides in the European part of Russia: results of the inverse modeling based on satellite measurements. Doklady Earth Sci 417:685–688 Konovalov IB, Beekmann M, Burrows JP, Richter A (2008) Satellite measurement based estimates of decadal changes in European nitrogen oxides emissions. Atmos Chem Phys 8:2623–2641 Martin RV, Jacob DJ, Chance K, et al (2003) Global inventory of nitrogen oxide emissions constrained by space-based observations of NO2 columns. J Geophys Res 108. doi:10.1029/2003JD003453 Richter A, Burrows JP, Nüß H, Granier C, Niemeier U (2005) Increase in tropospheric nitrogen dioxide over China observed from space. Nature 437. doi:10.1038/nature04092 Vestreng V, Breivik K, Adams M, et al (2005) Inventory review 2005, emission data reported to LRTAP convention and NEC Directive, Initial review of HMs and POPs. Technical report MSC-W 1/2005, ISSN 0804-2446
Detection of Desertification Zones Using Multi-year Remote Sensing Data Lev Spivak, Irina Vitkovskaya, Madina Batyrbayeva, and Alex Terekhov
Abstract Desertification is one of the most important problems for Central-Asia countries. Since 2000 space monitoring of vegetation has been being done in Kazakhstan with NOAA AVHRR data. Integral Normalized Difference Vegetation Index (IVI) was used for identification of desertification area. This index was calculated by summation of decadal NDVI composites for the vegetative season. This paper presents the results. Keywords Desertification • Integrated NDVI • Vegetation degradation
Introduction Kazakhstan Republic occupies 2.7 million km2 area. Most the area is located in arid and semi-arid zones and is used as pastureland. Desertification is one of the most important problems which is connected to the impacts of climate variation/changes and increased anthropogenic activities in the recent 30–40 years. Since 2000, space monitoring of the republic’s area has been implemented in order to register vegetation condition changes and estimate desertification dynamics. In order to identify desertification areas it is necessary to separate seasonal changes caused by weather conditions from the long lasting sustainable vegetation degradation. Special method was developed for analysis and of the long-term component of vegetation dynamics. This paper describes the methodology and its application.
L. Spivak (*), I. Vitkovskaya, M. Batyrbayeva, and A. Terekhov Join-stock Company National Centre of the Space Research and Technology, Almaty, Kazakhstan e-mail:
[email protected];
[email protected]
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_26, © Springer Science+Business Media B.V. 2011
235
236
L. Spivak et al.
Methodology NOAA AVHRR data were used for vegetation and desertification monitoring. Reflectances in the visible (VIS) and near infrared (NIR) parts of solar spectrum observed by the AVHRR instrument were used. The method includes the following procedures (Spivak et al 2006; Spivak et al 2007): 1. Processing of NOAA satellite images and a reference of images in Geographic Lat/Long Projection for WGS-84 spheroid 2. Calculation of daily Normalized Difference Vegetation Index NDVI = (NIR − VIS)/(NIR VIS) 3. Construction of decadal composite values of NDVI for each pixel based on maximal value of NDVI during a 10-day period 4. Calculation of the Integral vegetation index (IVI) following the equation below 27 IVI = ∑ NDVI i , i =10
where i is the number of decades from the beginning of a year. IVI describes seasonal volume of green biomass in each pixel and is effective for long-term analysis.
Results Based on the IVI for the 2000–2008 period Kazakh territory was divided into the following five zones with different vegetative productivity: Zone E – a very low (desert) Zone D – low Zone C – medium Zone B – temperate Zone A – high Figure 1 shows IVI-based 5-zone distribution during 2000–2008. As seen, the zones have latitudinal distribution and clearly show higher vegetation productivity in the south, deserttype of the environment in the central and mid-vegetation productivity in the north. As seen in Fig. 2 the size of the zone depends on seasonal weather conditions. The most favorable for vegetation year was 2002 and the worst year was 2006. Considerable increase in zone A and corresponding contraction of zone E are observed in years with favorable weather conditions. In unfavorable years, desert zone expands very much and the high productivity zone contract. In order to eliminate seasonal weather contribution we used the method of “transit zone”, which is the zone with minimal weather impact. Location of the “transit zone” with the area less than 13% (from the Kazakhstan total) is shown in Fig. 3. Note that such area as Semipalatinsk (eastern Kazakhstan) nuclear site is not taken into account.
Detection of Desertification Zones Using Multi-year Remote Sensing Data
2000
2001
2002
2003
2004
2005
2006
Zone A 20
30
2008
2007
% (IVImax)med 10
237
Zone B
Zone C
Zone D
Zone E
40
Fig. 1 IVI-based zones’ distribution during 2000–2008 in Kazakstan
square/1000. km**2
1200,0 1000,0 800,0 600,0 400,0 200,0 0,0 2000
2001
zone A
2002 zone B
2003 zone C
2004
2005
zone D
2006
zone E
2007
2008 year
Fig. 2 Area under A to E zones during 2000–2008
The “transit zone” is shown in Fig. 3 in the central Kazakhstan. The IVI-based estimates in Fig. 4 indicate that this zone is quite stable in size and equal approximately average long-term integral vegetative index value. Figure 5 shows dynamic of IVI/(IVImax)med for moderate weather conditions years.
238
L. Spivak et al.
%, normaliz. IVI 15
35
Fig. 3 Location of the “transit zone”, Kazakhstan
IVI/(IVI)maxmed, averaged by tranzit zone territory
Fig. 4 Dynamics of integrated vegetative index for the “transit zone” 0,290 0,280 0,270 0,260 0,250 0,240 0,230 0,220 0,210 2000
2001
2003
year
2005
Fig. 5 IVI/(IVImax)med changes dynamic for close on weather condition years
2007
Detection of Desertification Zones Using Multi-year Remote Sensing Data
239
Fig. 6 Areas of “transit zone” with low vegetation level during 2000–2007
Figure 6 shows a part of Kazakhstan map’s “transit zone” only and confirms worsening of vegetation conditions in this zone (dark color). The area with a low level of vegetative productivity in the transit zone gradually increases.
Conclusion Zoning of Kazakhstan territory with the seasonal Integrated NDVI was carried out during 2000–2008. Location of “transit zone” where weather impact on vegetation productivity is minimal was determined. Sites of sustainable vegetative degradation which is typical for desertification were identified.
References Spivak L, Vitkovskaya I, Batyrbayeva M (2006) Space monitoring of desertification processes in Kazakhstan with the use of long-term remote sensing data. IGARSS Proceeding, 2006, Anchorage, Alaska, pp. 23–27 Spivak L, Vitkovskaya I, Batyrbayeva M, et al (2007) Analysis of inter seasons vegetation dynamic on Turkmenistan area. Deserts assimilation problems. Ashgabad. 4(65):25–29
Satellite Desertification Monitoring in Sahara Mikhail A. Popov, Sergey A. Stankevich, Alexei I. Sakhatsky, Menny O. El Bah, Daoud Mezzane, and Igor A. Luk’yanchuk
Abstract Combating desertification in Sahara is important due to immense territory and lack of infrastructure. The objective of this article is the application of RS satellite methods for desertification monitoring in Sahara countries Mauritania and Morocco. It was demonstrated that RS methods is a useful tools to monitor desertification in Mauritania and Morocco. They require: dynamic nonlinear model of the ecosystem for long-time desertification forecast; database of the multi spectral satellite images, thematic maps correlated with other geo-meteo-data; and passive microwave and optical satellite data fusion procedures. Keywords Desertification monitoring • Microwave and optical data fusion
Introduction Desertification is land degradation, reduction of the bio-potential, intensification of aridity and wind-induced sand-mass transfer. This problem is very important in the Sahara countries Mauritania and Morocco leading to socio-economic problems. Being incapable to combat with desertification alone, Mauritania emerges as M.A. Popov (*), S.A. Stankevich, and A.I. Sakhatsky Scientific Centre for Aerospace Research of the Earth NAS of Ukraine, Kiev, Ukraine e-mail:
[email protected];
[email protected];
[email protected] M.O. El Bah University of Nouakchott, Nouakchott, Mauritania D. Mezzane Cadi, Ayyad University, Marrakech, Morocco e-mail:
[email protected] I.A. Luk’yanchuk University of Picardy, Amiens, France e-mail:
[email protected] F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_27, © Springer Science+Business Media B.V. 2011
241
242
M.A. Popov et al.
the important geo-political link between Central and North Africa. An explosive growth of Moroccan economy substantially stimulates the cultivation of virgin dry lands (ORMVAT 2007). The modern remote sensing (RS) satellite-image technology is the cheapest and most appropriate tool to deliver the reliable information about decertification, especially in remote regions (Campbell 2007). RS observations are normally used to take decisions about land exploitation and management especially in agriculture, hydrology, water management, geology, mineral resources, cartography, national security and, environmental changes. RS methods are used occasionally in Mauritania and Morocco due to a lack of experience, coordination and administrative difficulties (Abdellaoui 1989). The major problem is the interpretation of RS images together with geo-informatic data. Following our RS application in Emirates, Libya, Nigeria, Algeria, Kazakhstan and Turkmenistan (Lyalko and Popov 2006) we present the distributed network of desertification monitoring to transfer, locally adopt and apply the progressive RS technology to Mauritania and Morocco. The program of counter-desertification with RS methods should include: (a) evaluation of desertification with RS and geo-information data; (b) acquisition of multi-spectral and radar satellite images and creation of distributed database; (c) development of the nonlinear dynamical model of ecosystems; (d) development of image recognition algorithms to classify the elements of desert landscape (dunes, massifs, barchans) and to parameterize their dynamics; (e) collection of in situ measurements and matching them with geo-meteo-cartographical information; (f) development of long-range forecasts of eco-dynamics, availability of water and mineral resources. Implementation of this program will help to evaluate the risks of desertification and address water and natural resources. This article’s goals are to present the application of RS satellite methods for desertification monitoring in Sahara countries Mauritania and Morocco.
Implementation of RS Technology in the Sahara Countries Morocco and Mauritania having a very different level of economic development have common historical, social language and cultural traditions. Moreover they form a common block with European – African commercial, human, transportation and technological links. In the last decade Morocco had explosive economic development which triggered exploration and monitoring of natural resources and environment. Several national and international programs and centres introduced the application of RS methods for desertification monitoring (ORMVAT 2007; Web 1 2010; Web 2 2010; Web 1 2010). The economical and political situation in Mauritania is less stable then in Morocco that creates problems application of RS methods. The Nouakchott University created a laboratory for processing satellite images to address such natural hazards as desertification, flood, landslide, insect invasion etc. Figure 1 demonstrates application of RS methods in Mauritania.
Satellite Desertification Monitoring in Sahara
243
Fig. 1 Desertification monitoring of town Nouakchott: (a) general view, (b) RS-based GIS image
Fig. 2 AMSR-E/MODIS data fusion for desertification mapping in Morocco on May 28, 2009: (a) AMSR-E 26 km soil moisture product; (b) AMSR-E 1 km wrapped soil moisture product; (c) MODIS 1 km band 4, band 6 and land surface temperature product RGB synthesis (13:35); (d) MODIS 1 km band 4 (visible)/band 6 (SWIR) NWI product; (e) AMSRE-E/MODIS microwave/optical fused 1 km soil moisture product
Passive microwave data were applied to obtain soil moisture using inverted microwave radiometer response model (Chauhan et al.1994, Chauhan 2002). These data were enhanced by applying optical satellite data. For this purpose the Normalized water index (NWI) was used (Sakhatsky 2006):
NWI =
Eλ = 0.55 µ m − Eλ =1.65 µ m Eλ = 0.55 µ m + Eλ =1.65 µ m
(1)
where El = 0.55 mm – spectral radiance in visible band and El = 1.65 mm – spectral radiance in SWIR one. In addition land surface temperature is also applied for desertification monitoring (Liu et al. 2002). Figure 2 demonstrates the land cover moisture spatial resolution enhancement using AMSR-E (passive microwave 26 km resolution) and MODIS (multispectral optical 1 km resolution) data products. The algorithm for enhanced resolution land
244
M.A. Popov et al.
cover moisture assessment by microwave/optical fusion has been preliminary validated over north-west Sahara under non-cloudiness condition (correlation coefficient is 0.62). This technique is useful also for land cover classification (Popov et al. 2008).
Conclusion To combat desertification in agricultural and industrial regions of west part of Sahara a reliable and relatively non-expensive method is required to control the components of Sahara ecosystem (vegetation, sand and water). It was demonstrated that RS methods is a useful tools to monitor desertification in Mauritania and Morocco. They require: dynamic nonlinear model of the ecosystem for long-time desertification forecast; electronic database of the multi-spectra satellite images, thematic maps correlated with other geo-meteo-data; and passive microwave and optical satellite data fusion. Acknowledgement This research was supported by the NATO Science for Peace and Security (SPS) Programme, Grant Reference SPS MD SFP 984085.
References ORMVAT Office of agriculture (2007) Monographie agricole de la region Meknès – Rafilalet”. http://www.aui.ma/enhanced/regionmeknes/html/monogr_de_la_rmt_oct_2007.pdf Campbell JB (2007) Introduction to remote sensing. Taylor & Francis, New York, 626 p Abdellaoui A (1989) Développement et télédétection des ressources naturelles au Maghreb Central Télédétection en francophonie. AUPELF-UREF John Libbey Eurotext, Paris. Linsenbarth A (1996) A Lyalko VI, Popov MO (eds) (2006) Multi-spectral remote sensing for nature management. Naukova dumka, Kiev, 360 p Web 1 (2010) http://www.crts.gov.ma/desertification/forma_suivi_global.pdf Web 2 (2010) http://www.ucam.ac.ma/cners/ Web 3 (2010) http://doc.abhatoo.net.ma/doc/img/pdf/desertification-2.pdf Chauhan N, LeVine D, Lang R (1994) Discrete scatter model for radar and radiometer response to corn: comparison of theory and data. IEEE Trans Geosci Remote Sens 32:416–426 Chauhan N (2002) Soil moisture inversion at L-band using dual-polarization technique: A model based sensitive analysis. Int J Remote Sens 23:3209–3227 Sakhatsky AI (2006) Satellite data application for the water exchange modeling in geosystems (Ukrainian). Proc NAS Ukraine 4:118–126 Liu L, Zhang B, Xu G, Zheng L, Tong Q (2002) Vegetation classification and soil moisture calculation using land surface temperature (LST) and vegetation index (VI). Proc SPIE 4730:319–323 Popov MA, Stankevich SA, Sakhatsky AI, Kozlova AA (2008) Land cover contextual classification using space imagery for wetland and forest monitoring. Proceedings of the United Nations/ Austria/European Space Agency Symposium Space Tools and Solutions for Monitoring the Atmosphere and Land Cover, Graz (Austria)
Part V
Satellite & In Situ Long Records For Trend Analysis, Modeling & Monitoring
Global Vegetation Health: Long-Term Data Records Felix Kogan, Wei Guo, and Aleksandar Jelenak
Abstract The new Global Vegetation Health (GVH) data set has been developed for operational and scientific purposes. The GVH has advantages before other long-term global data sets, being the longest (30-year), having the highest spatial resolution (4-km), containing, in addition to NDVI, data and products from infrared channels, originally observed reflectance/emission values, no-noise indices, biophysical climatology and what is the most important, products used for monitoring the environment and socioeconomic activities. The processed data and products are ready to be used without additional processing for monitoring, assessments and predictions in agriculture, forestry, climate change and forcing, health, invasive species, deceases, ecosystem addressing such topics as food security, land cover land change, climate change, environmental security and others. Keywords Vegetation health • 30-year 4-km data records • Vegetation Condition Index (VCI) • Temperature Condition Index (TCI) and Vegetation health indices (VHI) • NDVI and BT
Introduction One of the most important long-term (30 years) satellite-based data records characterizing land surface, air near the ground and climate were created from the Advanced Very High Resolution Radiometer (AVHRR) flown on the National Oceanic and Atmospheric Administration (NOAA) polar-orbiting satellites. Several F. Kogan (*) NESDIS/NOAA, Center for Satellite Application and Research (STAR), Washington DC, USA e-mail:
[email protected] W. Guo IMSG Inc., Washington DC, USA A. Jelenak University Corporation for Atmospheric Research, Washington DC, USA F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_28, © Springer Science+Business Media B.V. 2011
247
248
F. Kogan et al.
global data sets were developed from the AVHRR records since the early 1980s. They were NOAA’s Global Vegetation Index (GVI and GVI-2), NASA’s Pathfinder and GIMMS (Tarpley et al. 1984; James and Kalluri 1994; Kidwell 1997; Tucker et al. 2004). These data were focused only on the Normalized Difference Vegetation Index (NDVI), ignoring infrared measurements, which are very useful for monitoring land, climate and socioeconomics. Therefore, NOAA has developed new data set entitled the Global Vegetation Health (GVH). The GVH has advantages before other long-term global data sets, being the longest (30-year), having the highest spatial resolution (4-km), containing, in addition to NDVI, data and products from infrared channels, originally observed reflectance and emission, many indices with suppressed noise, biophysical climatology and what is the most important, products used for monitoring the environmental and socioeconomic activities (Kogan 1995, 1997). This paper describes the new, considerably improved and currently available to users the NOAA’s global AVHRR-based operational GVH data set at 4-km (0.036°) resolution.
Satellites, Sensor, Data, Noise, Noise Removal, GVH Method The 30-year, 4-km, 7-day composite GVH data records were developed from the measurements made by the AVHRR instrument flying on board NOAA polarorbiting operational satellites. The NOAA/AVHRR is a cross-track scanning system (Kidwell 1995, 1997; Cracknell 1997) sensing the Earth and the atmosphere near the ground continuously through the 30-year history (from the early 1980s to the present) in the following wavelength of the solar spectrum: the visible (VIS, 0.58–0.68 mm, channel 1 (Ch1)), near infrared (NIR, 0.725–1.1 mm, channel 2 (Ch2)) and two infrared (IR, 10.3–11.3 mm, channel 4 (Ch4) and 11.5–12.5 mm, channel 5 (Ch5)). The AVHRR instrument scans the Earth continuously at a 1.1-km ground resolution and the measurements are sampled and recorded for the entire globe at 4-km resolution contributing continuously to the NOAA’s Global Area Coverage (GAC) data set (Cracknell 1997). From the 14 NOAA satellites flying in sun-sinchronous orbit and carrying the AVHRR instruments, the GVH system and data sets were developed from seven afternoon satellites: NOAA-7, 9, 11, 14, 16, 18 and 19 launched on June 23, 1981 (local day time at launch 14:30), December 12, 1984 (14:20), September 24, 1988 (13:30), December 30, 1994 (13:30), September 21, 2000 (13:44), May 20, 2005 (13:50) and June 2, 2009 (13:44), respectively. These satellites operated during 1981–1985, 1986–1989, 1989–1994, 1995–2000, 2001–2005, 2005 to present and 2009 to present, respectively. During September 1994–January 1995 no afternoon operational observations were produced since NOAA-11 satellite was malfunctioned and new NOAA-13 satellite failed soon after launch. Also, during January–June 2005, NOAA-16 was malfunctioning from time to time and its data were replaced with NOAA-17 (morning satellite) preliminary calibrated to the
Global Vegetation Health: Long-Term Data Records
249
NOAA-16 data. From the indicated satellites, NOAA-7 and 9 carried AVHRR-1 instrument, NOAA-11 and 14 – AVHRR-2 and the rest – AVHRR-3. All of them have identical design but with slightly different response functions. The VIS and NIR channels were pre-launch calibrated for converting counts into reflectance. The IR channels were in-flight calibrated and their count values were converted to brightness temperature. Due to chlorophyll and carotenoid pigments, leaf interior (mesophyll structure) and water content in green vegetation, the VIS and NIR measurements provide a means of monitoring vegetated surfaces. Difference between the NIR and VIS increases when vegetation becomes greener, more vigorous (more water) and denser. This is the main principle for the vegetation indices. The (NIR-VIS) difference was normalize representing the Normalized Difference Vegetation Index (NDVI = (NIR − VIS)/(VIS + NIR)). The IR channels were corrected for non-linear behavior of the instrument (Cracknell 1997). The GVH system algorithm starts form data extraction from the AVHRR/ CLAVR-x processing system (Jacobowitz et al. 2003; Heidinger and Pavolonis 2009) and collating the data onto a global GVH grid. This grid is based on the Plate Carre map projection. The global data spans from 75.024° (north edge) to −55.152° (south edge) in the latitudinal and from −180° (west) to 180° (east) in longitude directions. This processing supports nominal grid cell length of 4-km (3,616 * 10,000 grid elements). The GVH input includes the CLAVR-x navigation (NAV), observation (OBS), and geo-location (GEO) files for each Global Area Coverage (GAC) Level 1b orbit. Daily data are aggregated to a 7-day period using compositing method (saving the day which has the highest NDVI during the period). The compositing starts on the first day of a year and a period must have at least 4 days in the same year. The GVH output is a single file for each processing period containing metadata for each output variable, sensor and solar zenith angle, relative azimuth angle and ch1, ch2 counts and ch4, ch5 brightness temperature in the Hierarchical Data Format (HDF) similar to CLAVR-x output. One of the important steps in the primary data processing is radiometric calibration of visible and correction of thermal channels. Visible channels’ calibration consists of generally two steps: pre- and post-launch calibration. Based on Kidwell (1995), the following pre-launch linear formula (A = S * C + I ) is applied, where (A) is albedo, (S) is slope and (I) is intercept. Since the instrument output does not remain the same after launch, post-launch calibration was applied (R = S * (C − cd)) to NOAA-7 to 14 satellites, where C is 10-bit radiance count and cd – dark count) following Rao and Chen (1993, 1999). For NOAA-16 through 19 a dual slop calibration method was applied. Long- and Short-term Noise. Noise in AVHRR data creates fundamental constraints to the remote sensing of the Earth. The noise sources are physical, geometrical, mechanical, mapping, environmental, random etc.; some of them long-term, some short-term and some both (Kogan et al. 1996; Rao and Chen 1993, 1999; Cracknell 1997; Kidwell 1997). Clouds and other atmosphere constituents (aerosol, water vapor etc.) obscure the land surface reducing NDVI considerably. In case of unusual events, such as sharp volcanic aerosol increase, NDVI can be
250
F. Kogan et al.
depressed for a long time (Kogan et al. 1994). Changes in viewing and illumination geometry can lead to both an increase and decrease in NDVI depending on location, type of vegetation, position of sun and sensor. Satellite orbital drift, sensor degradation, and satellite change create long-term noise in NDVI data, especially after a satellite has been in service for more than 3 years. A few techniques have been designed to detect and reduce some noise in AVHRR data such as cloud and aerosol-screening, bi-directional effects, pre- and post-launch calibration, geometry effects, sensor degradation satellite orbital drift. Regardless of all these achievements, a complete physically based correction for all effects over various land surfaces, able to eliminate high, medium and low frequency noise, is not available. Unfortunately, NDVI and IR annual time series values experience large fluctuations (Kogan 1995) introducing some errors when this data is used for monitoring purposes. Many of these fluctuations are associated with non-physical causes, such as method of data sampling and processing, satellite navigation and orientation, observation and communication errors, and other random noise. It is unlikely to develop corrections for this type of noise. Besides, if clouds are detected the data are discarded, leaving a hole on a map. This put additional constraint on AVHRR data utility. In the development of GVH method and algorithm, major long- and short-term noise creating problem for satellite data interpretation and applications was removed from the data. They are (a) satellite orbit and sensor degradation; (b) jumps between the satellites; (c) excessive stratospheric aerosols; (d) difference in Equator crossing time; (e) difference between AVHRR sensors; (f) high frequency (short-term) noise; (g) random noise. Some of the noise sources are interrelated, some have additive impact, some has either short- or long-term contribution and some provides a combine input. Visible channels are affected by all noise sources, while infrared mostly by short-term because they are calibrated on board. A large satellite data distortion in the visible channels and NDVI occurs due to such long-term noise as satellite orbital drift, AVHRR sensor degradation and excessive stratospheric aerosols from volcanoes. The orbital drift and sensor degradation affected visible measurements on all space platforms; they normally start 2–3 years after a satellite was in space. NDVI reduction was observed on the data collected from NOAA-9 during 1987 and 1988, NOAA-11 (1993 and 1994) and NOAA-14 (1999 and 2000). In addition, reflectance/emission measured from NOAA-11 were distorted due to elevated stratospheric aerosols during 1991–1993 resulted in NDVI reduction. The aerosol was built after eruption of Mt Pinatubo volcanoes in the Philippines in mid-June 1991. In a few weeks after eruption, the aerosols encircled the Earth with the major air flow and stayed in the area between 30–20°N and 20–25°S for nearly 1.5–2 years. As the result the NDVI in that area dropped almost in half during second part of 1991 through 1992 (Kogan et al. 1996; Vargas et al. 2009). Similarly, NDVI reduced in tropical latitudes in April 1982 for 1–2 years after eruption of El Chichon volcano in Mexico (Stowe et al. 1992). Some difference between NDVI calculated from NOAA-16 and 18 satellites reflectance and from NOAA-14 and 11 existed during 2000–2010. This occurred because the AVHRR-3 instrument has slightly different characteristics than
Global Vegetation Health: Long-Term Data Records
251
AVHRR-2. Specifically, the area under response function curve for the first two was 8–12% narrower than for the last two. As the result the NDVI from NOAA-16 and 18 turned out to be higher than for the previous satellites. Moreover, in addition to NDVI reduction for NOAA-7 and 9 due to stratospheric aerosol and sever orbit degradation, general level of NDVI for the 1981–1988 period was 10–15% smaller than for the next two satellites because the time of the equator crossing for the first two satellites was almost one hour later. As the result of these problems and also orbit-degradation errors, large jumps in NDVI were observed between the end of previous and the beginning of the next satellites. The IR channels are also affected by stratospheric aerosol and orbit degradation (although less than the VIS) and must be corrected. Finally, high frequency noise inside each year created by clouds, variable transparency of the atmosphere (water vapor, dust, chemicals etc.), surface anisotropy, geometry of the sun and sensor, position of satellite, methods of data processing, random noise (including human errors) and long-term sources of noise distort considerably reflectance/emission of both NDVI and BT creating difficulties for satellite data application. Noise removal. As it has been mentioned, quite often different sources of noise affect NDVI and BT: either reducing their values (sensor degradation, orbital drift, equator crossing time, atmospheric attenuation, volcanic eruptions etc.) or increasing (spectral response function, off-nadir view etc.) or both depending on parameters (sensor type, surface anisotropy, forward/backscattering, sampling, random errors etc.). Moreover, quite often, several sources of noise affect the measured parameters the same time and with different intensity. It is hard to develop individual procedure for each source of noise because it is unknown how to separate them (for example elevated stratospheric aerosol and sensor/orbit degradation). Therefore, the noise in GVH data was removed empirically by comparing satellite and in situ observations applying the methods of mathematical statistics plus validation. Bias related to sensor degradation, satellite orbital drift, jumps in the indices while transitioning from one satellite to the next/previous and an elevated stratospheric aerosol were removed by applying the Empirical Distribution Function (EDF) method. (Crosby et al. 1996). An EDF is based on the assumption that for large areas, the NDVI reduction due to technical and external forces (orbital drift, volcanic eruptions, etc.) is larger than the weather-related NDVI changes from year to year. This can be expressed through probability that a random variable X is less than a given value x [F(x) = Pr{X < x}]. Following this assumption, large and stable changes in NDVI and BT can signal unexpected disturbances due to non-weather related causes discussed here. As a rule, sensor and satellite orbital degradation, increase and aerosol reduce NDVI and BT. This negative effect can be reduced by adjusting the EDF of parameters for the affected years with a benchmark EDF for the non-affected. The benchmark EDF for NDVI and BT were statistically composited from data of the 5 non-affected years: 1989, 1990, 1995, 1996, and 1997 (Vargas et al. 2009). The benchmark EDF’s for NDVI and BT were developed for each latitude lines and week of the year. The normalization of
252
F. Kogan et al.
the distorted data was performed for each pixel inside every latitude line and for each week by adding the difference between distorted and benchmark EDFs to each distorted pixel values of NDVI and BT for each latitude line and week. The thresholds (0.01 for NDVI and 2°C for BT) for the correction was selected based on data analysis. A stable long-term bias between NDVI from NOAA-16, 18 (2000–2010) and previous satellites due to response function differences was removed by calibrating the distorted NDVI against the NDVI measured by an on-ground radiometer during the growing season of 2002 over soybeans and corn at the experimental station of the University of Nebraska-Lincoln. During that period the fields’ vegetation fraction (VF) changed from 10 at the beginning of the season to 90% at the end. The in situ NDVI were compared with the top-of-canopy NDVI (obtained by running radiative transfer model) from AVHRR-3 sensor on NOAA-16. The comparison showed that NOAA-16 and later NOAA-18 NDVI must be reduced by 10% for all VF in order to remove response function curve differences with previous satellites. Another stable long-term bias, lower NDVI level for NOAA-7 and 9 (1981– 1989) compared to the later satellites, was investigated. A half an hour data collected from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) on the Meteosat Second Generation (MSG-2) European satellites were used to investigate NDVI dynamics in relation to the local time of observations. The results indicated that the bias developed because for the earlier satellites the equator crossing time was half to one and a half hours later than for the following satellites. SEVIRI’s NDVI analysis for the 2009 indicated that NDVI drops between 6% and 11% if observation local time moves from 13:00 to 14:30. In order to correct this distortion the EDF method was used. High frequency noise (clouds, aerosol, bi-directional reflectance, sun and sensor angles, human errors, other random noise) were removed from NDVI and BT by applying statistical methods. The vegetation-oriented method for a comprehensive noise reduction stems from a statistical approximation of the vegetation and temperature annual time series. The idea was to (a) single out the seasonal cycle; (b) suppress high frequency noise, and (c) enhance medium and low frequency variations related to large-scale and persistent weather fluctuations. This technique considers smoothing the weekly time series with a combination of a compound median filter and the least squares technique (Kogan et al. 1997). Numerous tests showed that this smoothing eliminated completely the high frequency outliers, including random, approximated accurately the annual NDVI and BT cycles, and, more importantly, singled out medium-to-low frequency weather-related fluctuations (valleys and hills in the NDVI and BT time series) during the annual cycle (Kogan 1995). Figure 1 shows time series of completely processed no noise NDVI averaged over three 1.0° latitudinal circles. Each diagram covers several ecosystems: broadleaf forest (54.5–55°N), mostly desert with some contribution from Southeast Asia forest (24.5–25°N) and mostly tropical forest (5.0–5.5°S). As seen, the time series (a) after do not show neither long-term trend nor short-period (5–10 years) trends;
Global Vegetation Health: Long-Term Data Records
253
Fig. 1 Processed mean NDVI (left) and BT (right) for three global latitude bands
(b) there are no any jumps between the satellites (1985–1986, 1988–1989, 1994–1995); (c) no NDVI reduction due to satellite orbit, sensor degradation and excessive stratospheric aerosols (1982–1984 and 1991–1994); (d) no difference due to Equator crossing time; (e) no difference due to AVHRR sensors changes; and (f) no high frequency and random noise are observed. GVH Method. After noise removal, weather-driven differences in NDVI and BT between the years become apparent: lower NDVI and higher BT in dry years and opposite in normal and wet years. This principle of comparing NDVI and BT for a particular year with their dry–wet range calculated from 30-year observations was laid down in the GVH algorithm development. The absolute maximum and minimum of NDVI and BT during 1981–2005 were calculated for each of the 52 weeks and for each pixel. They were then used as the criteria to estimate the upper (favorable weather) and the lower (unfavorable weather) limits of the ecosystem resources. Further, for estimation of weather impacts on vegetation condition, NDVI and BT values for a particular time (1 week or several weeks) were normalized relative to the absolute max/min interval. Following this procedure, NDVI and BT were rescaled based on Eqs. (1–3). They were named the Vegetation Condition Index (VCI), Temperature Condition Indices (TCI) and Vegetation Health Index (VHI) designed to characterize moisture (VCI), thermal (TCI) and total vegetation health (VHI) conditions in response to weather impacts
VCI = 100 * (NDVI − NDVI min ) / (NDVI max − NDVI min ) TCI = 100 * (BTmax − BT )/ (BTmax − BTmin ) VHI = a * VCI + (1 − a )* TCI
(1) (2) (3)
254
F. Kogan et al.
Fig. 2 Dynamics of vegetation health indices during the 2009 spring drought in southern Taxes (for a box with longitude 98.0–97.7°W, latitude 28.0–28.3°N) from NOAA-18 satellite
where NDVI, NDVImax, and NDVImin (BT, BTmax, and BTmin) are the smoothed weekly NDVI (BT), their multi-year absolute maximum, and minimum, respectively. The VCI, TCI and VHI approximate the weather component in NDVI, BT and their combination values. They fluctuate from 0 to 100, reflecting changes in vegetation conditions from extremely bad to optimal. An example of VH assessment and analysis of the 2009 spring drought in the southern Taxes is shown in Fig. 2. As seen, all indices show drought (below 40) from January 2009. The worse conditions developed due to extremely high temperatures (TCI close to 0), which continued through the entire spring (almost 4 months). As the result of the extreme heat, VCI (moisture index), which indicated mild (30–40) drought in January start deteriorated quickly reaching extreme drought level of 5–10 during March and April. Since both VCI and TCI showed extreme drought conditions the total vegetation health conditions were extremely unfavorable. In May, drought recovery has started, because the temperature cooled off.
Conclusions The new Global Vegetation Health (GVH) data set has been developed for operational and scientific purposes. The GVH has advantages before other long-term global data sets, being the longest (30-year), having the highest spatial resolution (4-km), containing, in addition to NDVI, data and products from infrared channels, originally observed reflectance/emission values, no-noise indices, biophysical climatology and what is the most important, products used for monitoring the environment and socioeconomic activities. The processed data and products are ready to use without additional processing, since the noise was completely removed and the products were developed based on vegetation requirements and the laws governing the vegetation. In addition to monitoring, the GVH data can be used for predictions since some parameters such as NDVI, VCI and partially VHI have “memory” reflecting cumulative impacts of the environment on vegetation canopy (Kogan 1995). GVH time series can be also used for climate-related trend analysis
Global Vegetation Health: Long-Term Data Records
255
such as vegetation green up, start/end of the growing season, gross primary production, losses in vegetation productivity, land cover change etc. It is important also to mention a few data correction procedures ignored in other global data. A few of them are removal of high frequency noise, also difference in response functions between AVHRR-3 and previous versions of the instruments, inconsistencies between the satellites in equator crossing time parameter. Moreover, some corrections were verified versus in situ data to see if the data quality is improving. The next steps in this research is comparison of GVH with other global datasets and analysis of climate induced vegetation green up.
References Cracknell AP (1997) The advanced very high resolution radiometer. Taylor & Francis, USA, 534 p Crosby DS, Goldberg MD, and Chung W (1996) Inter-satellite calibration using empirical distribution functions. 8th Conference on Satellite Meteorology and Ocean, Atlanta GA. American Meteorological Society, pp 188–190 Heidinger AK, Pavolonis MJ (2009) Global daytime distribution of overlapping cirrus cloud from NOAA’s Advanced Very High Resolution Radiometer. J Climate 18(22):4772–4784 Jacobowitz H, Stow LL, Ohring G, Heidinger A, Knapp K, Nalli N (2003) The advanced very high resolution radiometer PATHFINDER Atmosphere (PATMOS) climate data set: a resource for climate research. Bull Am Meteorol Soc June:785–793 James ME, Kalluri SN (1994) The Pathfinder AVHRR land data set: an improved coarse resolution data set for terrestrial monitoring. Int J Remote Sensing 15:3347–3363 Kidwell KB (ed) (1997) Global vegetation index user’s guide. National Oceanic and Atmospheric Administration, National Environmental Satellite Data and Information Services, National Climatic Data Center, Camp Springs MD, USA Kidwell KB (ed) (1995) NOAA polar orbiter data users guide. National Oceanic and Atmospheric Administration, National Environmental Satellite Data and Information Services, National Climatic Data Center, Camp Springs MD, USA Kogan FN (1997) Global drought watch from space. Bull Am Meteorol Soc 78:621–636 Kogan FN, Sullivan JT, Ciren PB (1996) Testing post-launch calibration for the AVHRR sensor on world desert targets during 1985–1993. Adv Space Res 17(1):47–50 Kogan FN (1995) Droughts of the late 1980s in the United States as derived from NOAA Polar Orbiting Satellite Data. Bull Am Meteorol Soc 76:655–668 Kogan FN, Sullivan J, Carey R, Tarpley D (1994) Post-pinatubo vegetation index in Central Africa. Geocarto Int 3:51–58 Rao CRN, Chen J (1993) Calibration of the visible and near-infrared channels of the Advanced Very High Resolution Radiometer (AVHRR) after launch. Proceedings the International Society of Optical Engineering, Orlando, FL, pp 56–66 Rao CRN, Chen J (1999) Revised post-launch calibration of the visible and near-infrared channels of the Advanced Very High Resolution Radiometer on the NOAA-14 spacecraft. Int J Remote Sensing 20:3485 Stowe LL, Carey RM, and Pellegrino PP (1992), Monitoring the Mt.Pinatubo aerosol layer with NOAA/11 AVHRR data, Geophys. Res. Lett., 19:159–162. Tarpley JP, Schneider SR, Money RL (1984) Global vegetation index from NOAA-7 meteorological satellite. J Climate Appl Meteorol 23:491–494 Vargas M, Kogan F, Guo W (2009) Empirical normalization for the effect of volcanic stratospheric aerosols on AVHRR NDVI. Geophys Res Lett 36:L07701 Tucker CJ, Pinzon JE, Brown MB, Slayback DA, Pak EW, Mahoney R, VermoteEF, El Salcous N (2004) An extanded AVHRR 8-km NDVI data set compatible with MODIS and SPOT vegetation NDVI data. Int J Remote Sensing 15(7):340–354
Aero-Space Radar Online Monitoring of Disasters in Ukraine Mariya Belobrova, Dmitry Bychkov, Anatoly Boev, Alexandre Gavrilenko, Valentin Efimov, Alexandre Kabanov, Ivan Kalmykov, Alexandre Matveev, and Valery Tsymbal
Abstract This paper provides examples of utilization of space radar (SLRARS) data for natural and anthropogenic disasters prevention. Ukrainian scientists have a good expertise in using low-cost, high-performance spaceborne and airborne sidelooking radars (SLRARS). These instruments help to keep tracking different natural and technogenic catastrophes and evaluate the most essential parameters characterizing these phenomena and events.
Environmental Phenomenon In October 1983, the heavy masses of multiyear near-polar ice started moving southward, and the convoy of 22 vessels got nipped in the Longa strait to the south of the Wrangel Island. Soon the most powerful atomic ice-breaker headed to the distressed vessels. But the multiyear ice whose thickness was >5 m was really “hard nut to crack”. Finally, one vessel was mercilessly crushed by the ice and sank and the other was badly damaged. The lost ships with cargoes were estimated at over $8 billion plus thousand people had to be airlifted from those far-flung areas of Chukotka. The first radar images of this area showed that100 km to the north of the Wrangel Island there was a vast zone covered with thin ice, which could open the clear way to the long-suffering convoy. The radar images allowed one to discern the fissures and patches of ice-free water in heavy multiyear ice shown
M. Belobrova () Usikov Institute of Radiophysics and Electronics, NAS of Ukraine, Kharkov, Ukraine e-mail:
[email protected] D. Bychkov, A. Gavrilenko, V. Efimov, A. Kabanov, I. Kalmykov, A. Matveev, and V. Tsymbal Kalmykov Center for Radiophysical Sensing of the Earth, NAS of Ukraine, Kharkov, Ukraine A. Boev Institute of Radioastronomy, NAS of Ukraine, Kharkov, Ukraine
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_29, © Springer Science+Business Media B.V. 2011
257
258
M. Belobrova et al.
Fig. 1 (a) Initial SLRAR image of the Longa Strait (acquired from “Cosmos-1500”, October 20, 1983). A large “polynia” (ice-free path) is clearly seen near Wrangel island. It is located 100 km northward of the area where the convoy had been ise-trapped; (b) an interpretation map of the same area; different types of sea ice and convoy route are indicated
in (Fig. 1), right up to the Wrangel Island. The patches of ice-free water stretching out as far as that zone were found in the hummocked ice fields. The convoy promptly altered its course and headed north ward (Kalmykov et al. 1993). As the fleet of cargo ships followed the ice-breakers, it soon succeeded in getting closer to the young ice zone and sailing further safely arriving to the port of Pevek in a few days. Another example helped to prevent disaster in the Dnieper basin. In spring 1988, fast and vigorous snow melting was predicted in Ukraine based on synoptic situation. In order to forestall the possible failures of the hydrotechnical facilities on the Ukrainian rivers an initial desire was to drain water from the man-made storages of the Dnieper cascade. However, this option was rejected because the water from one of the man-made reservoir was saturated with radioactive nuclides from the Chernobyl disaster. The data from the SLRAR of the “Cosmos-1766” satellite showed that snow melting rate is going to be far below the predictable the overflowing of those artificial water storages was not expected. Similar snow melt monitoring was successfully performed in the spring 1996.
Aero-Space Radar Online Monitoring of Disasters in Ukraine
259
Monitoring of Extratropical and Tropical Cyclones, Fronts, Squalls and Other Dangerous Phenomena in Sea–Air System The SLRAR X-band data enable one to evaluate the intensity at which the energy from atmosphere to ocean is transferred. In other words, the intensity of a scattered radio signal is directly related to the spectral density of short ocean waves (ripples). The simultaneous application of both radar and optical data allows the energy exchange between ocean and atmosphere to be adequately monitored (Mitnik and Viktorov 1980). Figure 2 illustrates two radar images of swiftly developing young polar cyclone. The images were acquired by the spaceborne “Cosmos-1500” SLRAR several days apart. No optical images are given, because the radar surveying operations were performed during the polar night and the cloud cover was not visible. As seen from Fig. 2a, the radar image shows that in the Norwegian Sea, close to the ice sheet, a young polar cyclone (size 80–100 km) is in its initial formation phase and the near-sea surface wind inside it does not exceed 12–15 m/s. In this early phase the typical vortex pattern of this cyclone is pronounced much in the same way as its structural elements: specifically, a small-size, calm, windless area in the centre and the atmospheric fronts. The same polar cyclone occurs (Fig. 2b) south-west of the Spitsbergen. Its size is 300 km, with wind speed 17–20 km/s.
Fig. 2 Radar images of the young polar cyclone: (a) the initial incipience phase in the Norwegian Sea close to the ice sheet, the near-sea surface wind inside it (a according to the SLR data) does not exceed 12 to 15 m/sec. Its size varies between 80–100 km. A spin-like structure of this cyclone and its constituent elements are pronounced: a small-size calm area in the centre and the atmospheric fronts; (b) the same cyclone is observed south-west of the Spitsbergen. Its size has grown up to 300 km and the wind speed is between 17 and 20 m/sec
260
M. Belobrova et al.
Fig. 3 Satellite images of occluded cyclone above the Okhotsk Sea. from spaceborne “Cosmos-1500” SLRAR: (a) radar image; (b) optical image of the visible range
Figure 3 shows both the SLRAR and optical images of the occluded cyclone above the Okhotsk Sea. In the optical image a cloud spiral is clearly discernible, which is originated from the small-size cloudless space at centre 1. The radar image illustrates the central zone 1, which 20 × 25 km area is shifted nearly 25 km northward (clearly seen by the south ward-inclined axis). According to estimates, zone 1 looks dark, the wind speed inside is less than 5 m/s. Dark patch 1 is surrounded by a brighter horseshoe-shaped cloud where the wind speed V ≅ 7–10 m/s. In the optical image this area is featured by the white cumulus clouds. Still further away from the centre the wind speed tends to decrease. In the radar image V ≅ 4 ÷ 5 m/s is seen in region 3. Part of this area as shown in the optical image is marked by stratus clouds 3 with a slightly elevated upper boundary interspersed with separate thicker cumulus-shaped elements. No disturbances occur in the sea surface wind field; even at a boundary of cloud-free area 4 and cloud cover 3 the values V remain unchanged. Tropical Cyclone “Diana” was formed on September 8, 1984 near Bahamas Islands, and soon was upgraded to a highly intense tropical storm (Mitnik and
Aero-Space Radar Online Monitoring of Disasters in Ukraine
261
Fig. 4 Radar (right) and optical (left) images of the tropical cyclone “Diana” in its initial development phase from “Cosmos-1500” satellite SLRAR
Viktorov 1980). Figure 4 presents radar and optical images of tropical cyclone “Diana” in its initial stage. Among the features that are quite discernible in the SLRAR image is a vortex-like near-sea surface wind structure of the large-scale cyclone. In this development phase the cyclone-driven near-sea surface wind is 17–20 m/s. The cloud shape was not typical for tropical cyclones. The second radar surveying of the tropical cyclone “Diana” was made just at a time when it had reached its peak developed. The radar image shown in Fig. 4b was acquired on September 11, 1984 when the atmospheric pressure within the cyclone centre dropped to 952 gPa, whereas the maximum wind speed increased to 56 m/s. Radar images features of the “Diana” hurricane, dated September 11, showed a dark area of 13–14 km in diameter, typical hurricane “eye”. The wind speed decreased, gravitational-capillary waves reduced resulted in a reduced level of a sea surface-scattered radar signal. The outer diameter of the “eye” wall is about 30 km, similar to visible image. Analysis of Fig. 5c indicates that the “calm” zone in the hurricane “eye” is diminished. Two spiral-shaped chains of rain cells inside the cloud appear to be brighter than the surrounding background. The chains are located at 120 and 270 km from the tropical cyclone centre, with the width 10–12 km. The SLRAR images acquired 12 h later indicated that “Diana” hurricane has shifted to north-northeast by about 85 km with the speed 7 km/h (Utkin et al. 1986).
262
M. Belobrova et al.
Fig. 5 Satellite images of the tropical cyclone “Diana”: (a) visible image at 18 h 30 min September 11, 1984; (b) and (c) SLRAR images at 18 h 30 min September 11, 1984 and 8 h 30 min September 12, 1984 from “Cosmos-1500”
To determine the “Diana” hurricane parameters two near-sea surface wind field cross-sections were produced (to the east and to south of the “eye” centre). Data processing was performed according to the algorithms presented in (Kalmykov et al. 1989). Figure 6 shows the module of the “Diana” hurricane wind speed as a
Aero-Space Radar Online Monitoring of Disasters in Ukraine
263
|V|, m/s
50 1 40 30 20
3 2 4
10 100
200
r,km
Fig. 6 Dependences of wind speed module of the “Diana” hurricane from distance to its centre (SLRAR (1 – latitudinal and 2 – meridian crosscuts) and NOAA reconnaissance aircraft (3 – latitudinal and 4 – meridian crosscuts))
Fig. 7 Images of the eastern part of the Black Sea acquired by EOS “Sich-1” on March 17 (a) 20 (b), 1996 (07 h 33 min and 19 h 19 min GMT). March 17 (a) and (c) on March 20 (07 h 11 min GMT). Illustration of jet-like eastern wind from the Transcaucasian region, caused by the terrain
function of the distance to its centre based on the SLRAR and data provided by NOAA reconnaissance aircraft (Lawrence and Clark 1984). As seen the wind speed is reduced considerably moving from the hurricane centre.
Widespread Effects of Hazardous Atmospheric Processes in Coastal Areas and Locked-Land Seas Figure 7 shows SLRAR images of the eastern part of Black Sea acquired by EOS “Sich-1” on March 17 (07.33 GMT, (a) and (b) with 12 h interval) and 20 (07.11 GMT), 1996. All three show jet-like wind (“wind tunnel”-type) resulting from the impact of terrain. The wind jet extended up to 150 km and was around ~40–50 km wide. Figure 8 presents SLRAR image (acquired by the EOS “Sich-1” October 26, 1996) atmospheric cyclone in the eastern part of the Black Sea. The spatial scale of
264
M. Belobrova et al.
Fig. 8 An image of the cyclone in the eastern part of the Black Sea (acquired by the SLR EOS “SICH1”, orbit N6226). The near-sea surface wind speed shows a sharp increase from 2 to 3 m/s and may run up to more than 20 m/s at the boundary of the atmospheric front in its southern part, close to the coast of Turkey. White arrows – marked of the atmospheric fronts; black arrows – run-down winds
the developed vortex was well in excess of 250 km. The vortex structure is clearly visible, especially the most hazardous front in the southern part, close to the coast of Turkey. The near-sea surface wind speed at the boundary of this front can reach more than 20 m/s in a zone 5–10 km long. This particular phenomenon is thought of as highly dangerous to sea-going vessels.
Conclusion The 20-years experience of the Ukrainian spaceborne SLRARs application convincingly testifies that such systems (characterized by a wide swath can be effectively used to create the real satellite system for on-line warning and monitoring
Aero-Space Radar Online Monitoring of Disasters in Ukraine
265
natural catastrophes. The group of 3–4 satellites equipped with SLRAR systems will be capable of providing real-time information about hazardous processes. This data will be updated between 2 and 3 h apart.
References Kalmykov A, Pichugin A, Tsymbal V (1989) Determination of the driving wind using the side looking radar system of the Cosmos-1500 satellite. Sov J Remote Sens 5:668–690 Kalmykov A, Velichko S, Tsymbal V (1993) Observations of the marine environment from spaceborne side-looking real aperture radars. Remote Sens Environ 45:193–208 Lawrence M, Clark G (1984) Atlantic hurricane season of 1984. Mon Weather Rev 113:1228–1237 Mitnik L, Viktorov S (1980) Radiolocation of the Earth Surface from the Space. Gidrometeoizdat, Leningrad (in Rusian) Utkin V, Shestopalov V, Kalmykov A et al (1986) Determination of the characteristics of tropical cyclones from spaceborne radar images. Dokl Akad Sci USSR 286:331–333 (in Russian)
Comparison of AVHRR-Based Global Data Records Felix Kogan, Marco Vargas, and Wei Guo
Abstract Several global data sets have been developed from the AVHRR instrument measuring reflectance/emission of the Earth since the early 1980s. The longest datasets currently available for users are NOAA’s Global Vegetation Health (GVH), NASA’s Global Inventory Modeling and Mapping Studies (GIMMS) and Land Long Term data Records (LTDR). The GVH has 30-year records (1981–2010), GIMMS – 26 (1981–2006) and LTDR – 19 (1981–1999). These datasets have different spatial and temporal resolutions, processing methods (sampling, calibration, noise removal, mapping, gap treatment etc.), applicability, availability, distribution etc. They have been used frequently for monitoring earth surface, atmosphere near the ground and analysis of climate related land surface trends. Since one of the common features of these datasets is the Normalized Difference Vegetation Index (NDVI) this paper is focusing on comparison of NDVI time series, specifically comparing time series dynamics and trends. It is shown that GIMMS NDVI is two to three times higher and has steeper long-term trend compared to GVH and LTDR. Keywords Vegetation health • 30-year 4-km data records • Vegetation condition index (VCI) • Temperature (TCI) and Vegetation health (VHI) indices • NDVI and BT
Introduction The Advanced Very High Resolution Radiometer (AVHRR) has been in space on NOAA operational polar-orbiting satellites for the past 30 years observing earth surface and atmosphere and will continue this endeavor in the future. The AVHRR F. Kogan (*) and M. Vargas NESDIS/NOAA, Center for Satellite Application and Research (STAR), Washington DC, USA e-mail:
[email protected] W. Guo IMSG Inc., Washington DC, USA
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_30, © Springer Science+Business Media B.V. 2011
267
268
F. Kogan et al.
data are extremely useful for monitoring weather and weather-related disasters, land ecosystems, agriculture, forestry, human health, invasive species, climate forcing and other human activities. In the recent 20 years, following such a wide spectrum of applications, AVHRR-based Normalized difference Vegetation Index (NDVI) was used for the analysis of climate trend and land surface changes (Zhou et al. 2001; Myneni et al. 1997; Nemani et al. 2003). That stimulated development of AVHRR-based historical records. Several global data sets were developed from AVHRR measurementss since the early 1980s. The most popular were NOAA’s Global Vegetation Index (GVI, since 1985 and GVI-2 since the early 1990s) and Global Vegetation Health (GVH since 2000), NASA’s Pathfinder (since the early 1990s), Global Inventory Modeling and Mapping Studies (GIMMS, from the late 1990s) and currently under development the Land Long Term data Records, LTDR (Tarpley et al. 1984; James and Kalluri 1994; Kidwell 1995; Tucker et al. 2005; LTDR 2010; GIMMS 2010; GVH 2010; Kogan et al. 2010). These data were focused (except GVI-2 and LTDR) on NDVI, ignoring infrared measurements, which are very useful for monitoring land, climate and socioeconomics (Kogan 1995). The available data have different spatial and temporal resolutions and processing methods (sampling, calibration, noise removal, mapping, gap treatment etc.), applicability, availability, distribution and others. Although the data were produced from the same original source they are often mismatch characterizing the same phenomena. However, NDVI from each dataset was available for the global community (GVH 2010; LTDR 2010; GIMMS 2010) and was used frequently, especially for analysis of climate related land surface trends (Zhou et al. 2001; Nemani et al. 2003; Myneni et al. 1997). Therefore, this paper is focused on comparison of NDVI time series and specifically comparing time series dynamics and trends.
Data Sets Description Global AVHRR data of the Global Area Coverage (GAC) format at 4 km special and daily temporal resolution have been archived from 1981 through present for all NOAA series polar-orbiting operational satellites (Kidwell 1995; Cracknell 1997). The observations for the entire period are available for four channels (Ch): visible (VIS, 0.58–0.68 mm, Ch1), near infrared (NIR, 0.725–1.1 mm, Ch2) and two infrared (IR, 10.3–11.3 mm, Ch4 and 11.5–12.5 mm, Ch5). The data are presented in 10-bit digital count values for each 4-km pixel between latitudes 75.024° N and 55.152° S and longitudes 180°W and 180° E. The GAC data were used in the development of all datasets. Three data sets GIMMS, LTDR and GVH were selected for further analysis. Below is a brief description of the data. GIMMS dataset provides NDVI only calculated from the VIS and NIR; satellites – NOAA-7, 9, 11, 14, 16 and 17; period – 1981–2006; resolution: 8 km (0.072°) special and 15-day maximum value composites (MVC) temporal; calibration – vicarious from Vermote and Kaufman (1995); corrections – volcanic stratospheric aerosol during 1982–1984 and 1991–1994, and satellite orbital drift using empirical
Comparison of AVHRR-Based Global Data Records
269
mode decomposition/reconstruction (EMD) method; projection – Albers; web http://glcf.umiacs.umd.edu/data/gimms/. LTDR dataset: provides daily NDVI, daily surface reflectance (five channels), three angles and quality flag; satellites – NOAA-7, 9, 11 and 14; period – 1981– 1999; resolution – 5.5 km (0.05°), daily temporal; calibration – Vermote and Kaufman (1995); satellites NDVI product available from 1981 to 1999. web – http://ltdr.nascom.nasa.gov/ltdr/products2007.html. GVH dataset: provides – original measurements (VIS, NIR, IR4, IR5, NDVI, Brightness temperature (BT), three angles); no noise NDVI and BT, climatology of NDVI and BT, products (Vegetation (VCI), Temperature (TCI) condition indices and Vegetation health index (VHI), Fire risk index, Drought index; satellites – NOAA-7, 9, 11, 14, 16, 17, 18 and 19; period – 1981–2010; resolution – 4-km (0.036º) and 7-day MVC; calibration – vicarious from Rao and Chen (1999); corrections – volcanic stratospheric aerosol during 1982–1984 and 1991–1994, satellite orbital drift, difference between AVHRR-2 and AVHRR-3 instruments; difference in equator crossing time, high frequency noise; projection – Plate Carree (latitude–longitude); validation – in 27 countries; focus: – globe; web – http://www .star.nesdis.noaa.gov/smcd/emb/vci/VH/index.php.
Results and Discussion The currently available three datasets have the same start year (1981) for the time series but different end year (for GVH 2010; GIMMS 2006; and LTDR 1999). The comparison was done for the period 1982–1999, considering the earliest year (1999) of LTDR data end and that the 1981 data covered only the last 3 months of the year. Since the description of GIMMS and LTDR data processing is quite sketchy the comparison analysis does not explanation causes of the differences. Three regions in Asia, Africa and South America were selected (Table 1) to characterize all major ecosystems. Figure 1 shows 19-year average NDVI time series for the major areas indicated in Table 1 for the three continents Asia, South America and Africa. As seen, the GVH and LTDR time series are quite similar while the GIMMS data are very different having two to three times higher NDVI. Moreover, the amplitude of NDVI variation inside a year is also two times larger for GIMMS compared to two other data sets. These differences are consistent for the three continents. Although LTDR data are daily and GVH – weekly they match quite well though LTDR has slightly smaller NDVI and larger variation over time, which can be explained by GVH’s Table 1 Regions selected to average NDVI
Region Asia Africa South America
Coordinates (º) Latitude 40°N–60°N 15°N–30°S 15°N–25°S
Longitude 15°W–30°E 15°W–40°E 80°W–35°W
270
F. Kogan et al.
0.8
GIMMS
Asia
GVI-x
NDVI
0.6 0.4 0.2 82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99 LTDR
0.8
GIMMS
Soutn America
GVI-x
NDVI
0.6 0.4 0.2 LTDR GIMMS GVI-x
NDVI
0.6 0.5
Africa
0.4 0.3 0.2 0.1 82
83
84
85
86
y1 = 0.00088*x - 1.5
87
88
89
90
91
92
93
94
95
96
y2 = 0.0001*x + 0.24
97
98
99
y3 = 0.00014*x + 0.025
Fig. 1 1982–1999 NDVI time series for GIMMS, LTDR and GVH
Fig. 2 1982–1999 slope trend (*10−3) for GIMMS, LTDR and GVH datasets
temporal data sampling with MVC procedure which gives preference to a larger NDVI compared to no sampling (daily values) for LTDR. Also, more pronounced seasonal cycle is observed in LTDR and GVH compared to GIMMS in South America. Evaluation of the three datasets for the trend existence during 1982–1999 indicates that all three have small mostly upward trend since the slope is positive, Fig. 2 shows slope change for the NDVI averaged for each degree latitude. As seen, the
Comparison of AVHRR-Based Global Data Records
271
trend is changing for both latitude and for the area. The northern regions have slightly larger trend. Moreover, in sub Sahara Africa the trend for GIMMS and GVH is changing from positive above equator to slightly negative below the equator. From the three datasets, GVH shows the smallest trend since it is closer to zero line compared to the other two datasets.
Conclusion The three datasets produced from the same GAC data showed different NDVI values since they have different sampling, calibration and processing. GVH dataset is the longest, has the highest special resolution, also has all original measurements (VIS, NIR, IR4, IR5, NDVI, Brightness temperature (BT), three angles); no noise, climatology and what is the most important products (Vegetation (VCI), Temperature (TCI) condition indices and Vegetation health index (VHI)). All three datasets have small mostly upward trend; the GVH has the smallest slop. In Africa, the GVH and GIMMS data change slope from positive to negative while crossing the equator.
References Cracknell AP (1997) The advanced very high resolution radiometer. Taylor & Francis, USA, 534 p GIMMS (2010) http://glcf.umd.edu/data/gimms/ GVH (2010) http://www.star.nesdis.noaa.gov/smcd/emb/vci/VH/vh_browse.php James ME, Kalluri SN (1994) The Pathfinder AVHRR land data set: an improved coarse resolution data set for terrestrial monitoring. Int J Remote Sensing 15:3347–3363 Kidwell KB (ed) (1995) NOAA polar orbiter data users guide. National Oceanic and Atmospheric Administration, National Environmental Satellite Data and Information Services, National Climatic Data Center, Camp Springs MD Kogan FN (1995) Droughts of the late 1980s in the United States as derived from NOAA polar orbiting satellite data. Bull Am Meteorol Soc 76:655–668 Kogan F, Guo W, Jelenak A (2010) Global vegetation health: long-term data records. In: Kogan F, Powell A, Fedorov O (eds) Use satellite and in situ data to improve sustainability. Springer, New York (in this book) LTDR (2010) http://ltdr.nascom.nasa.gov/cgi-bin/ltdr/ltdrPage.cgi Myneni RB et al (1997) Increased plant growth in the northern high latitudes from 1981–1991. Nature 386:698–702 Nemani RR, Keeling CD, Hashimoto H, Jolly WM, Piper SC, Tucker CJ, Myneni RB, Running SW (2003) Climate-driven increases in global terrestrial net primary production from 1982 to 1999. Science 300(5625):1560–1563 Rao CRN, Chen J (1999) Revised post-launch calibration of the visible and near-infrared channels of the advanced very high resolution radiometer on the NOAA-14 spacecraft. Int J Remote Sensing 20:3485 Tarpley JP, Schneider SR, Money RL (1984) Global vegetation index from NOAA-7 meteorological satellite. J Climate Appl Meteorol 23:491–494
272
F. Kogan et al.
Tucker CJ, Pinzon JE, Brown ME, Slayback D, Pak EW, Mahoney R, Vermote E, El Saleous N (2005) An extended AVHRR 8-km NDVI data set compatible with MODIS and SPOT vegetation NDVI data. Int J Remote Sensing 26(20):4485–5598 Vermote EF, Kaufman YJ (1995) Absolute calibration of AVHRR visible and near-infrared channels using ocean and cloud views. Int J Remote Sensing 16(13):2317–2340 Zhou L, Tucker CJ, Kaufmann RK, Slayback D, Shabanov NV, Myneni RB (2001) Variation in northern vegetation activity inferred from satellite data of vegetation index during 1981 to 1999. J Geophys Res D17:20069–20083
Merging Remote Sensing and In Situ Data for Estimation of Energy Balance Components Under Climate Change Conditions: Ukrainian Steppe Zone Tatiana Ilienko and Elena Vlasova
Abstract Climate change is recognized as one of the most important natural events in the past decades. Energy balance components (surface temperature, evapotranspiration, etc.) are closely related with climate change. Therefore adaptation of the existent models is needed for future climate. A shortage of surface weather observations due to the reduction of permanent weather networks requires combining remote sensing and in situ observations. This work is devoted to merging of these two data types in estimation of energy balance components of the Ukrainian steppe zone under climate warming condition. Keywords Climate change • Energy balance • Evapotranspiration • Remote sensing data
Introduction Climate change is recognized as one of the most important natural events in the past decades. The Intergovernmental Panel on Climate Change (IPCC) concluded that the global surface temperature increased 0.74 ± 0.18°C during the last century. A warmer climate will affect both environmental and sustainable development of the world. The climate system reacts by adjusting the earth’s energy balance to a new equilibrium. These processes include a release of latent heat through increase of evaporation, plant transpiration and precipitation accelerating hydrologic cycle. Regarding water resources, the difference between precipitation and evapotranspiration determines the amount of water available for runoff and groundwater recharge. T. Ilienko () Agroecological Institute, Kyiv, Ukraine e-mail:
[email protected] E. Vlasova Institute for Hydraulic Engineering and Land Reclamation Ukrainian Academy of Sciences, Kyiv, Ukraine
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_31, © Springer Science+Business Media B.V. 2011
273
274
T. Ilienko and E. Vlasova
Precipitation changes will be critical, but evapotranspiration which is controlled by changes in other climate variables such as temperature, humidity, radiation and wind speed will also play a major role. Evapotranspiration being a key component of the energy and water balance, plays also important role in the water cycle of irrigated lands of Ukraine. Scientific community has formulated the demands for data frequency, update and accuracy (Sellers 1993). One of the important requirements is to validate energy balance components’ calculation. Shortage of surface weather observation due to a reduction of permanent weather network requires combining remote sensing and in situ observations. Unlike point weather station observations, satellite sensors provide spatial information on energy balance components. Satellite remote sensing is able to provide the required frequency of data update with high accuracy (Li and Garand 1994). Therefore it is expedient to merge both in situ and remote sensing data. Over the last two decades a considerable number of research has been undertaken to determinate actual evapotranspiration, regional distribution of energy balance components over heterogeneous land surfaces using a combination of satellite and in situ data (Kustas et al. 2003; Roerink et al. 2000). The aim of this research is to estimate energy balance components under warming climate conditions by merging in situ and satellite data. The study of evapotranspiration as an element of water balance will give basic knowledge of water consumption of vegetation, water supply and demand as well as water shortage.
Data and Methods Meteorological data were collected for the steppe zone of Ukraine in the Zaporozhsky region (oblast), Kamenko-Dneprovsky district during 1939–2004. Analysis of data in Fig. 1 shows a decrease in total annual precipitation (Fig. 1a) and an increase in average temperature (Fig. 1b). Satellite data of the cloud-free Landsat images of Zaporozhsky region (August 21, 2000 and July 17, 2001) have been used for the study. Mathematical and statistical analysis was used for data processing. FAO techniques of evapotranspiration calculation (Allen et al. 1998) and Surface Energy Balance Algorithm (SEBAL) with modification (Bastiaanssen et al. 1998, Roerink et al. 2000) were also used in this research. Satellite data processing included image acquisition and preprocessing (mapping, radiometric and geometrical correction, data conversion to specified format, assessment of cloud conditions), thematic image processing: hybrid classification for crop identification; specification of field boundaries, evapotranspiration modeling, energy balance components calculation. Multispectral satellite images and in situ data were used for evapotranspiration calculation (Ilienko 2009). Energy balance components were calculated by radiation balance method using space images. The relationship between radiation and energy balance was approximated by the Eqs. (1) and (2) and is shown in Fig. 2.
Merging Remote Sensing and In Situ Data for Estimation of Energy Balance
a
275
Annual precipitation amount (1939-1969) 650
precipitation amount, mm
600 550 500 450 400 350 300 250 200 150 1932 1937 1946
1951 1956 1961 1966 Years
b
Percipitation amount (1973, 2004 years)
Average monthly temperature (1973, 2004 years)
160
25 20 1973 year
15 10
2004 year
5 0 march april may june july august
percipitation amount, mm
Temperature (C)
30
120 80 40 0
1973 year 2004 year
march april may june july august
−40
Fig. 1 Meteorological data for Zaporozhsky oblast (1939–1969): (a) annual total precipitation and trend; (b) average monthly temperature and precipitation amount
R n = K ↓ −K ↑ + L ↓ −L ↑,
(1)
R n = G 0 + H + λ E,
(2)
where Rn – net radiation, K↓, K↑ – short wave radiation, L↓, L↑ – long wave radiation, G0 – soil heat flux, Н – sensible heat flux, lЕ – latent heat flux. Net radiation and radiation balance components (1) is derived from satellite image surface reflectance (G0), surface temperature (Т0), incoming solar radiation (К↓) and long wave incoming radiation (L↓) by the Eq. (3) written as
R n = (1 − ρ0 ) τ K ↓ −σε T0 4 + L ↓,
(3)
where t is atmosphere transmissivity coefficient, s – Stephan–Boltzmann constant,e0 – surface emissivity coefficient. Energy balance components (2) are derived from radiation balance components because G0 = f1(Rn,r0, T0NDVI), H = f2(Rn, G0, L), lЕ = f3(Rn, G0, L), L = (TH – T0)/
276
T. Ilienko and E. Vlasova Space images Visible
Near IR
Thermal IR
Input data Albedo – ρ0
Surface temperature – T0
Vegetation index – NDVI
Radiation balance short wave radiation
KØ
K≠ net
long wave radiation
L≠
LØ
radiation
Rn Energy balance soil heat flux
G0
sensible heat flux
H
evaporative latent heat fraction flux
lE
Actual evapotranspiration
ETa
Fig. 2 Relation between radiation, energy and evapotranspiration
(TH – TlE), where TH, TlE are boundary temperatures which can be distinguished in the reflectance–temperature relationship: for wet and dry surface conditions (Roerink et al. 2000). The flow chart of the energy balance components calculation is shown in Fig. 3. Daily net radiation Rn,24 and daily potential evapotranspiration ЕТр,24 = 0.408(Rn – G0) are calculated (G0 = 0 daily) using the Penman–Monteith method (Allen et al. 1998). Daily evapotranspiration ЕТа,24 = LЕТр,24 was derived from the energy balance components.
Study Area The investigated Kamenko-Dneprovsky district of Zaporozhsky is located in Ukrainian steppe zone and is characterized by a shortage precipitation, their irregular distribution through the year, high temperature and low air humidity. The highest temperature and the lowest precipitation amount are in August when the main crops are maturing. The main activity in the region is agriculture and some crops
Merging Remote Sensing and In Situ Data for Estimation of Energy Balance
277
Fig. 3 Flow-chart of energy balance components calculation
are irrigated. There are five irrigation systems which irrigate about 27,460 ha of land. Stepovy farm was selected as pilot territory because its environment is typical for the entire region.
Results and Discussion Landsat ETM 7 images (August 21, 2000 and July 16, 2001) of the entire region and farm Stepovy are shown in Fig. 4. Hybrid classification for crop identification; specification of field boundaries is shown in Fig. 5. Using regression analysis the expression for surface albedo (r0) through planetary albedo (rp) was obtained:
( = (ρ
( ) − 0.048) ⁄ 0.516 (R
)
ρ 0 = ρp + 0.0746 ⁄ 0.432 R 2 = 0.905, August 2000 ,
ρ0
2
p
)
= 0.9425, July 2001 .
Albedo for different surface types is presented in Table 1.
278
T. Ilienko and E. Vlasova
Fig. 4 Landsat ETM 7 images of the entire region (left) and farm Stepovoy (right) for August 21, 2000 and July 16, 2001 Land use maps (Stepovy farm)
a
N W
E S
corn_silo fallow pears soybean sunflower corn_beans spring barley winter barley winter wheat alfalfa water 0
3
6
12
Kilometers 24
18
N
b
E
W S
water pease corn spring barley winter wheat fallow sainfoin sunflower silo corn soybean perennial herbs
0
3
6
12
18
Kilometers 24
Fig. 5 Hybrid crop classification, farm Stepovy: (a) August 2000, (b) July 2001
Merging Remote Sensing and In Situ Data for Estimation of Energy Balance
279
Table 1 Albedo and evapotranspiration for different surfaces August 21, 2000 July 16, 2001 Actual daily Actual daily evapotranspi evapotranspi Relative daily evapotranspi Surface ration Surface ration ration albedo (mm/day) Surface type albedo (mm/day) Water 0.071 10.35 0.92 0.099 10.5 Fallow 0.185 3.06 0.51 0.189 1.18 Winter wheat 0.194 1.03 0.11 0.258 3.73 Corn 0.22 4.3 0.43 0.198 7.08 Sunflower 0.23 2.9 0.2 0.21 7.32 Alfalfa 0.209 2.6 0.19 0.163 2.07
a
Relative daily evapotranspi ration 0.91 0.12 0.46 0.73 0.79 0.24
Daily net radiation Rn
−2 −1
MJm d
36.7- 47.78 47.79- 49.99 50- 51.57 51.58- 52.94 52.95- 54.42 54.43- 57.9 57.91- 61.59 61.6- 63.59
b −2 −1
MJm d
42.63-50.65 50.66- 53.01 53.02-54.66 54.67- 56.07 56.08- 57.25 57.26-62.44 62.45-68.57 68.58-102.8
Fig. 6 Daily net radiation: (a) August 21, 2000, (b) July 16, 2001
The relationship between boundary temperatures TH and TlE and surface albedo was obtained using satellite data: TH = 325 − 50 ρ0 , Tλ E = 291.47 + 41.17 ρ0 (August 2000 ) and TH = 321.77 − 26.06 ρ0 , Tλ E = 294.21 + 17.87ρ0 (July 2001). Energy balance components were derived using the above mentioned method and input parameters calculated from satellite data. Figures 6–9 present energy
280
Instantaneous soil heat flux G0
a
-2 -1
Jm s
3.25 - 4.72 4.73 - 5.4 5.41 - 5.96 5.97 - 6.55 6.56 - 7.09 7.1 - 7.63 7.64 - 8.2 8.21 - 9.01
b -2 -1
Jm s 4.41 - 5
5.01 - 5.59 5.6 - 6.19 6.2 - 6.99 7 - 7.58 7.59 - 8.18 8.19 - 8.78 8.79 - 9.78
Fig. 7 Instantaneous soil heat flux: (a) August 21, 2000, (b) July 16, 2001 Instantaneous sensible heat flux H
a -2 -1
Jm s
2.928-6.209 6.21-12.77 12.78-19.34 19.35-22.62 22.63-25.9 25.91-29.19 29.2-32.47 32.48-42.31
b -2 -1
Jm s
3.645-7.728 7.729-12.15 12.16-16.23 16.24-19.98 19.99-23.38 23.39-27.12 27.13-30.87 30.88-34.27 34.28-81.57
Fig. 8 Instantaneous sensible heat flux: (a) August 21, 2000, (b) July 16, 2001
Merging Remote Sensing and In Situ Data for Estimation of Energy Balance
a
281
Instantaneous latent heat flux lE -2 -1
Jm s
8.64 - 16.5 16.6 - 21.8 21.9 - 27 27.1 - 32.3 32.4 - 37.5 37.6 - 48 48.1 - 58.6 58.7 - 69.1
b -2 -1
Jm s
6.85 - 11.4 11.5 - 16.4 16.5 - 21.5 21.6 - 26.1 26.2 - 30.7 30.8 - 36.1 36.2 - 42.4 42.5 - 49.5 49.6 - 58.2
Fig. 9 Instantaneous latent heat flux: (a) August 21, 2000, (b) July 16, 2001
balance components: net radiation Rn, instantaneous soil heat flux G0, sensible heat flux H and latent heat flux lE, respectively. Daily net radiation Rn,24, daily potential evapotranspiration ЕТр,24 and daily actual evapotranspiration ЕТа,24 were also calculated. Relative evapotranspiration ЕТrel was determined as ЕTa,24/ЕТр,24. The results are presented in Table 1. The resulting and maps of daily actual (ЕТа,24) and relative (ETrel) evapotranspiraration for August 21, 2000 and July 16, 2001 are shown in Figs. 10 and 11.
Conclusions This research was based on mathematical and statistical analysis, satellite data processing and energy balance component calculations. For estimation of energy balance satellite and in situ data were merged with GIS techniques. The method was adapted to the steppe zone of Ukraine (Kamenko-Dneprovsky district, Zaporozhsky oblast). For pilot territory components of energy balance, actual and relative evapotranspiration for different crops were calculated and mapped.
282
Actual daily evapotranspiration ETa,24
a
N E
W S
mm/d 0.7562 - 2.377 2.378 - 3.673 3.674 - 4.808 4.809 - 5.942 5.943 - 7.401 7.402 - 9.508 9.509 - 12.75 12.76 - 15.99 16 - 18.42 18.43 - 23.93
0
3
6
12
18
Kilometers 24
b
N E
W S
mm/d 0.1047 - 1.352 1.353 - 2.599 2.6 - 3.846 3.847 - 4.47 4.471- 5.094 5.095 - 6.341 6.342 - 7.588 7.589 - 9.459 9.46 - 10.71 10.72 - 11.95
0
3
6
12
18
Kilometers 24
Fig. 10 The resulting maps of actual daily evapotranspiration: (a) August 21, 2000, (b) July 16, 2001 Relative daily evapotranspiration ETrel
a
N E
W S
0
3
6
12
18
0.08 - 0.2 0.21 - 0.29 0.3 - 0.4 0.41 - 0.49 0.5 - 0.58 0.59 - 0.67 0.68 - 0.76 0.77 - 0.88 0.89 - 1.12 1.13 - 1.85
Kilometers
24
b
N E
W S
0.01 - 0.14 0.15 - 0.22 0.23 - 0.3 0.31 - 0.4 0.41 - 0.47 0.48 - 0.55 0.56 - 0.63 0.64 - 0.7 0.71 - 0.83 0.84 - 1.01
0
3
6
12
18
Kilometers
24
Fig. 11 The resulting maps of relative daily evapotranspiration: (a) August 21, 2000, (b) July 16, 2001
Merging Remote Sensing and In Situ Data for Estimation of Energy Balance
283
References Allen RG (1998) Crop evapotranspiration: guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper No. 56, Rome, Italy Bastiaanssen WGM, Menenti A et al. (1998) A remote sensing surface energy balance algorithm for land (SEBAL), part 1: Formulation. J Hydrol 212–213:198–213 Ilienko TV (2009) Using satellite and in situ data for evapotranspiration assessment. Ukrainian steepe zone case. J Agroecol June:122–126, Special edition. Kyiv Kustas WP, Diak GR, Moran MS (2003) Evapotranspiration. Remote sensing of earth. Encyclopedia of water science. Marcel Dekker, New York, pp 267–274 Li ZQ, Garand L (1994) Estimation of surface albedo from space – a parameterization for global application. J Geophys Res 99:8335–8350 Roerink GJ, Su Z et al. (2000) S-SEBI: a simple remote sensing algorithm to estimate the surface energy balance. Phys Chem Earth B 25(2), 147–157 Sellers PJ (1993) Remote sensing of the land surface for studies of global change, NASA/GSFC International Satellite Land Surface Climatology Project Report, Columbia, MD
Atmosphere Aerosol Properties Measured with AERONET/PHOTONS Sun-Photometer over Kyiv During 2008–2009 Vassyl Danylevsky, Vassyl Ivchenko, Gennadi Milinevsky, Michail Sosonkin, Philippe Goloub, Zhengqiang Li, and Oleg Dubovik
Abstract The PHOTONS network, as a part of the AERONET ground network for aerosol remote sensing of Earth’s atmosphere, covers more than 40 sites in Europe, Africa and Asia, providing sun-photometer measurements, calibration and data processing. Within the framework of scientific cooperation between the Lille 1 and the National Taras Shevchenko Kyiv Universities, the CIMEL CE 318-2 sun-photometer has been operated at Kyiv from the end of March, 2008. This article describes the AERONET/PHOTONS measuring equipment, procedure, data processing and the preliminary analysis of columnar aerosol properties retrieved during April 2008–March 2009. Spectral aerosol optical thickness (AOT), Angström parameter and precipitable water vapor thickness were measured and analysed. Keywords Aerosol remote sensing • AERONET/PHOTONS network • Aerosol optical thickness • Angström parameter • Precipitable water vapor
Introduction In recent years, scientific community, governments and non-government organizations are giving much attention to research of the atmospheric aerosols content, dynamic and physical properties since it is one of the air pollutants that can be potentially hazard for biosphere and also contributor to global climate change (Penner et al. 2001; Forster et al. 2007). Present increase in the amount of aerosols V. Danylevsky (*), V. Ivchenko, and G. Milinevsky National Taras Shevchenko University of Kyiv, Kyiv, Ukraine e-mail:
[email protected] M. Sosonkin Main Astronomical Observatory of National Academy of Science of Ukraine, Kyiv, Ukraine P. Goloub, Z. Li and O. Dubovik Université de Lille, France F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_32, © Springer Science+Business Media B.V. 2011
285
286
V. Danylevsky et al.
in the atmosphere creates negative radiative forcing counteracting to global warming (Forster et al. 2007). Key parameters for determining both direct and indirect radiative forcing are: (1) the aerosol optical properties, which vary as a function of a wavelength and relative humidity, (2) the atmospheric loading and geographical distribution of the aerosols, which vary as a function of time, and (3) the aerosol particles sizes, shapes and chemical compositions. Lack of aerosol temporal and spatial data and insufficient accuracy of aerosol properties determination create some problem for accurate estimation of aerosol radiative forcing (Penner et al. 2001; Forster et al. 2007; Kokhanovsky 2008; Dubovik et al. 2002). It is important to separate radiative forcing created by anthropogenic aerosol contribution from radiative forcing created by the natural aerosol. The atmosphere aerosol particle properties are usually estimated by an inverse problem solution (King et al. 1999; Dubovik et al. 2002; Kokhanovsky 2008). The Earth atmosphere-surface system is characterized by great number of parameters which have to be retrieved simultaneously. The best results are obtained by joint analysis data of space-borne and ground-based remote sensing. In order to monitor aerosol properties and dynamics at regional and global scales, a network of ground-based sites, equipped with standardized measuring devises was set up. Ground-based network for passive aerosol measurements is the AERONET (AERosol Optical NETwork, http://aeronet.gsfc.nasa.gov/)–established in early 1990 by NASA and Laboratoire d’Optique Atmosphérique (LOA) University Lille 1, the Centre National d’Etudes Spatiales (CNES) and Centre National de la Recherche Scientifique (CNRS) of France (Holben et al. 1998). The AERONET consists of hundreds of automatic sun-photometers. The PHOTONS (PHOtométrie pour le Traitement Opérationnel de Normalisation Satellitaire, http:// loaphotons.univ-lille1.fr) is French subdivision of the AERONET, operates about 45 observational sites: about 30 in Europe (France), 10 in Africa and 5 in Asia. They provide sun-photometer measurements, calibration and data processing. But AERONET/PHOTONS sites distributed unevenly, especially in East Europe (in Ukraine particularly). At the end of 2007, following scientific cooperation between LOA, Lille 1 (France) and National Taras Shevchenko (Kyiv, Ukraine) universities, the AERONET/PHOTONS site was set up. This article describes equipment, data reduction procedures and the preliminary analysis of columnar aerosol properties retrieved from Kyiv site.
Instrument and Data Currently, automatic sun photometers (spectral radiometers) CIMEL CE-318 (http://www.cimel.fr/photo/sunph_us.htm) are used by the AERONET/PHOTONS as the main instrument (Holben et al. 1998). The CE-318 sun tracking photometers have been designed and realized to be a very accurate motorized, portable, autonomous (powered by solar battery) and automatic instruments. The most
Atmosphere Aerosol Properties Measured with AERONET/PHOTONS
287
currently widespread models of CIMEL sun-photometers over AERONET/ PHOTONS sites are standard СЕ 318-1 and polarized model СЕ 318-2. In order to derive total column aerosols properties, water vapor and ozone, these photometers measure solar irradiance, sky radiance (aureole brightness), polarization (if polarized model is used), along the almucantar and vertical of the Sun with certain angle intervals. The photometers have two optical channels with two collimators and independent detectors of different sensitivities to measure bright direct sun irradiance and dark sky radiance in the standard wavebands. The photometers equipped with sets of optical filters, which wavelengths were selected to avoid strong influence of gaseous constituents of atmospheric extinction and to take into consideration water vapor absorption. The CE 318-1 model is equipped with filters for wavelengths of 340, 380, 440, 675, 870, 940, 1020 nm, and the CE 318-2 model for wavelengths 440, 675, 870 (2 polarization analyzers added), 940, 1020 nm. Spectral bandwidth of each channel is equal to 10 nm at half maximum. The 940 nm channel is used to determine the water vapor amount in atmosphere column because water vapor has a maximum of absorption at this wavelength. The data are transferred from sun-photometer to the AERONET/PHOTONS data base in two ways: via the special data collection systems on a geostationary satellite, or via Internet. Description of the sunphotometers characteristics is provided by http://www.cimel.fr, http://aeronet. gsfc.nasa.gov/new_web. The polarized CE 318-2 sun-photometer model has been installed at Kyiv site. The pre-programmed microprocessor control measurement procedure provides several scenarios depending on Sun position on celestial sphere, season and time of the day (Holben et al. 1998). The data obtained from observations are used both for aerosol optical thickness (AOT), water vapor content measurements and for the instrument self-calibration. Following Holben et al. (1998) and Li et al. (2008), calibration techniques is used to convert the instrument outputs to AOT and radiance (W/m2 sr mm). Two types of calibration procedures are used: direct-Sun irradiance and diffuse-sky radiance. Also different techniques are used to calibrate reference (master) and field instruments. Sun-channels of reference sun-photometers are usually calibrated at special high-altitude sites with clear stable atmosphere conditions by the Langley plot method, which uses the Sun as a reference light source. Field instruments are generally calibrated by comparison with the master instrument at low-altitude calibration sites (e.g., Goddard Space Flight Center, USA, and Carpentras, France). Sky-radiance channels are calibrated in the laboratory by using an integrating sphere or a “vicarious” calibration method (Li et al. 2008). The errors is less than 2% for solar channels and less than 5% for sky-radiance channels. These values correspond to the total uncertainty in AOT from a newly calibrated field instruments under cloud-free conditions typically not more than 0.01 for l ³ 440 nm. The aerosol optical depth is computed for three data quality levels: level 1.0 – unscreened data, level 1.5 – screened for cloud contamination, and level 2.0 – cloud-screened and quality-assured data. Level 2.0 data are also corrected after photometer’s yearly recalibration.
288
V. Danylevsky et al.
Method For the most reliable Level 2.0 data the following parameters are determined: spectral AOT, Angström exponent and water vapor content (thickness of precipitation water layer) in atmosphere column over observational site (Holben et al. 1998, 2001). AOT is proportional to the number of aerosol particles in sun-photometr’s field of view and is a proper measure of aerosol content in atmosphere over the observational site. Spectral AOT approximation is based on the Beer–Lambert–Bouguer law: E (λ ) = E0 (λ ) · exp (−τ (λ )) where E(l) is spectral solar irradiance measured with the calibrated sun-photometer at the time of observations, E0(l) is solar irradiance at the top of the atmosphere, computed using the sun-photometer calibration coefficients, and t(l) is atmosphere optical thickness in the direction of the Sun, it is computed from the equation above. To obtain AOT the optical thickness due to water vapor, Rayleigh scattering and trace gases (O3, CO2, NO2 etc.) must be subtracted from t(l): AOT (λ ) = τ (λ ) − τ (λ )water − τ (λ )Rayleigh − τ (λ )CO − 2
Water vapor content is a very important factor for deriving AOT. The total column water vapor is derived from three spectral channels: 675, 870 and 940 nm. Firstly atmosphere optical thickness is computed for 675 and 870 nm using Rayleigh optical thickness and AOT only. Then the atmosphere optical thickness for 940 nm is computed extrapolating the data obtained above. Hence, the water vapor optical thickness tW for 940 nm is found using measured and extrapolated data:
(
)
ln (τ W ) = ln (τ 940 measured )− ln τ 940 extrapolated . The total thickness TW of the precipitable water layer in atmosphere column is determined using equation: 1
− ln (τ W ) b a TW = mW where a and b are filter-dependent constants, and mW is water vapor optical air mass. Angström parameter a is power exponent in equation that is used for calculation of AOT dependence on light wavelength: AOT (λ ) = B · λ − α where B is AOT at l = 1 mm. The parameter a is calculated from data measured at two or more wavelengths, using a least squares fit, as a=
d ln (AOT (λ )) d ln (λ )
.
Atmosphere Aerosol Properties Measured with AERONET/PHOTONS
289
AOT obtained for 440 and 870 nm is used for a calculations, as a rule. Angström parameter determined in this way is the simplest qualitative indicator of aerosol particle size averaged on atmosphere column over observational site because aerosol particles optical properties and, as a consequence, spectral extinction coefficient of aerosol depend on the ratio 2p · a/l, where a is the characteristic size of the particle. The coefficient a increases when the particles’ size decreases. Studies of optic atmosphere properties show the Angström parameter change range from −0.1 (coarse particles with a ~ 1–10 mm) to 2.5 (fine aerosol fraction with a ~ 0.01–0.1 mm). Representative value of a for inland aerosol of various sources is about 1.3, but its peak value for practically molecular atmosphere (with a << l) can reach 4 (Dubovik et al. 2002; Kokhanovsky 2008). But detailed research shows that Angström parameter determined from two spectral channels mentioned above is more sensitive to the ratio Vfine/Vtotal, where Vfine is the aerosol fine particles volume and Vtotal is the total particles volume, than to effective radius of particles observed in atmosphere column. The dependence of the ratio Vfine/Vtotal on a for the sun-photometer spectral range can be more precisely determined from AOT obtained at more than two spectral channels using a second-order polinomial fit of the logarithm of equation AOT(l) = B · l−a (Schuster et al. 2006). Special inversion algorithm and software of Version 2 have been developed by AERONET team for aerosol optical and physical properties retrieval (Dubovik and King 2000; Dubovik et al. 2002, and Version 2 Inversion Products/Inversion Product Description at http://aeronet.gsfc.nasa.gov/new_web/publications.html). The software inverts sky radiances simultaneously at all available wavelengths for the complete solar almucantar scenario or principal plane scenario together with AOT measured at the same wavelength. The retrieval accounts for different levels of accuracy in the measurements: the standard deviation for error in AOT is assumed ±0.01, the standard deviation of error in the sky radiance measurements is assumed ±5%, and the standard deviation for error in scattered angles are ±0.1°. The inverse solution is based on a set of assumptions, principal of them are: (1) atmosphere is plane-parallel; (2) aerosol particles are partitioned into two components: spherical, which is modeled by an ensemble of polydisperse homogeneous spheres, and non-spherical, which is a mixture of polidisperse randomly-oriented homogeneous spheroids; (3) vertical distribution of aerosol is homogeneous in the almucantar inversion and bi-layered in the principal plane inversion; (4) errors of measurements are uncorrelated and log-normally distributed. The output includes both retrieved aerosol parameters (particles size distribution, volume concentration, volume median and effective radii, complex refractive index, partition of spherical/ non-spherical particles etc.) and calculated on the basis of retrieved aerosol properties (single scattering albedo, phase function and its asymmetry). The output also provides estimates for both random and possible systematic errors for most of the retrieved characteristics. According to those estimates, 68% confidence intervals are presented for most of retrieved characteristics. The detailed retrieved aerosol properties are used for calculating downward and upward radiant fluxes in broad spectral range and aerosol radiative forcing.
290
V. Danylevsky et al.
Study Area The Ukraine’s AERONET/PHOTONS site is located at the Golosiiv forest, 10 km from the center of the Kyiv. The major sources of the anthropogenic atmosphere pollution over Kyiv are fumes from fuel-burning boilers, combustion plants, cars and airplanes. The surrounding landscape enables sun-photometer to scan entire celestial hemisphere. The site has been equipped with the polarized sun-photometer model CIMEL CE-318-2. Level 2.0 data were produced during April 2008–March 2009 and Level 1.5 thereafter. There are other facilities to study aerosol remotely: Sevastopol AERONET site, which is at very southern coast of Ukraine, approximately 700 km from Kyiv; Chişinău (Moldova), about 450 km and Minsk (Belarus), about 550 km.
Results and Discussion The first year sun-photometer data were used to estimate climatology and microphysics of aerosol particles over Kyiv. Figure 1 shows variations of the aerosol properties obtained from Level 2.0 data during April 2008–March 2009. As seen, AOT, the Angström parameter and precipitable water layer thickness are changing considerably in the course of the year. The largest AOT and water vapor (WV) were in August–September. The range of Angström parameter indicates that the fine mode aerosol dominates in the Kyive’s atmosphere (Fig. 1b). This is mainly anthropogenic aerosol, which is confirmed by other studies (Penner et al. 2001; Forster et al. 2007). But coarse-mode aerosol particles dominate in some days, especially during spring and summer of 2008 (see Fig. 1b). Table 1 shows the number of days in each month with the measurements of Levels 1.5 and 2.0. As seen, the number of observations in spring and summer is significantly larger than during November–February. Comparison of the climatology of aerosol optical properties obtained at Kyiv with other AERONET sites is shown in Table 2 (Holben et al. 2001; Dubovik et al. 2002). The AOT over Kyiv during April 2008–March 2009 was rather low compared to other continental sites except for Dalanzadgad presented by desert dust aerosol. The lower limit of the Angström parameter over Kyiv is close to lower limit of a at maritime and desert continental sites. The upper limit of the Angström parameter over Kyiv is closer to sites with urban-industrial and biomass burning aerosols. Table 2 also shows that AOT over Kyiv was lower than over boreal forests, tropical forests and savannas where biomass burning aerosol predominates. Besides, AOT was lower than at urban-industrial regions. The Angström parameter range at Kyiv indicates larger variation probably due to the sizes and, may be, microphysics of aerosol particles. The AOT and Angström parameter over Kyiv for two days one in summer and one in spring are shown in Fig. 2.
Atmosphere Aerosol Properties Measured with AERONET/PHOTONS
291
Fig. 1 Dynamics of aerosol columnar climatology parameters (sun-photometer СЕ-318): (а) monthly average АОТ, standard deviations »±0.01; (b) the Angström parameter a (daily average) in the range 440–870 нм; (c) precipitable water layer thickness (daily average)
Table 1 Numbers of days with the measurements, Kyiv
Month, year April 2008 May 2008 June 2008 July 2008 August 2008 September 2008 October 2008 November 2008 December 2008 January 2009 February 2009 March 2009
Level 1.5 9 25 19 22 29 14 18 6 2 3 5 10
Level 2.0 9 25 19 22 29 14 18 6 2 3 4 9
292
V. Danylevsky et al.
Table 2 Aerosol climatology properties at Kyiv in comparison with other AERONET sites. The AOT daily average for Kyiv site (column 2), whereas the AOT ranges for other sites include all data obtained at 440 nm (Dubovik et al. 2002) and monthly average at 550 nm (Holben et al. 2001) Range of the Mean value Angström of АОТ for Range of АОТ l = 440 nm or parameter for l = 440 nm Site and time of l = 500 nm or l = 500 nm observations ll = 440–870 нм Aerosol type – Kyiv 2.1 ³ a ³ 0.4 0.76 ³ t440 ³ 0.06 t440 = 0.23 2008–2009 1.0 ³ t440 ³ 0.1 t440= 0.24 2.5 ³ a ³ 1.2 UrbanGSFC, Greenbelt industrial (MD, США) and mixed 1993–2000 Crete-Paris, 0.9 ³ t440 ³ 0.1 t440 = 0.26 2.3 ³ a ³ 1.2 UrbanFrance 1999 industrial and mixed Mexico City 1.8 ³ t440 ³ 0.1 t440 = 0.43 2.3 ³ a ³ 1.0 Urban1999–2000 industrial and mixed 3.0 ³ t440 ³ 0.1 Amazonian forest, t440 = 0.74 2.1 ³ a ³ 1.2 Biomass Brazil 1993– burning 1994; Bolivia 1998–1999 1.5 ³ t440 ³ 0.1 African savanna, t440 = 0.38 2.2 ³ a ³ 1.4 Biomass Zambia burning 1995–2000 2.0 ³ t440 ³ 0.1 Boreal forest, USA t440 = 0.40 2.3 ³ a ³ 1.0 Biomass and Canada burning 1994–1998 0.25 ³ t500 ³ 0.05 t500 = 0.13 Dalanzadgad, 1.94 ³ a ³ 0.61 Desert dust Mongolia 1997–2000 Lanai, Hawaii 0.12 ³ t500 ³ 0.06 t500 = 0.08 0.96 ³ a ³ 0.56 Oceanic 1995–1999 San Nicolas Island, 0.13 ³ t500 ³ 0.04 t500 = 0.08 Oceanic 1.10 ³ a ³ 0.78 California 1998–2000
On May 20 (Fig. 2a and b) coarse aerosol particles predominated relative volume more than 50% over a day (a £ 1). AOT increase was accompanied by a decrease in a, because relative content of coarse aerosol particles was increasing. The Angström parameter for August 19 values indicate that aerosol fine mode predominated (a » 2), which relative volume was more than 50% over a day, and AOT diminution in the morning with the corresponding a increase (Fig. 2c and d). It indicates that coarse particles number is decreasing.
Atmosphere Aerosol Properties Measured with AERONET/PHOTONS
293
Fig. 2 Spectral AOT and the Angström parameter variations at Kyiv site during one day in the spring (a, b) and one in the summer (c, d)
Conclusions For the first time measurements of aerosol properties are performed in the atmosphere column over Kyiv with the sun-photometer CIMEL CE-318-2 (polarized model) starting from the end of March, 2008. The PHOTONS network, as a division of AERONET, is in charge of the Kyiv site providing the sunphotometers calibration and data processing. Version 2 of the AERONET inversion retrievals techniques have been applied to derive atmosphere column of aerosol optical and microphysical properties and aerosol climatology over Kyiv using data of Level 2.0 quality obtained from the Sun direct irradiation and the Sky radiation measurements for a period from April 2008 to March 2009. Daily means AOT at 440 nm were changed in the range from 0.06 to 0.76 over Kyiv for the period mentioned, and daily means AOT at 675, 870 and 1020 nm have lower values. Maximum of AOT was observed in August–September, and minimum in November. Yearly average AOT at 440 nm over Kyiv is equal to 0.23 and is rather low as compared with some other continental sites except the site of desert dust aerosol. The precipitable water vapor thickness on atmosphere column over Kyiv had maximum during summer months and was not more than 3 cm, but in winter it could be less than 0.2 cm. Range of the Angström parameter values obtained during of a year for spectral range of 440–870 nm (2.1 ³ a ³ 0.4) is rather wide as compared with some other AERONET sites, but the parameter a vary between approximately 1.5
294
V. Danylevsky et al.
and 2.0 during the most part of a year showing that aerosol fine mode predominated in atmosphere over Kyiv. Spectral AOT and the Angström parameter can vary appreciably during a day pointing on variations of relative content of the fine or coarse aerosol particles in atmosphere column over Kyiv. This work was supported by the Ministry of Education and Science of Ukraine, and EGIDE, France.
References Dubovik O, King M (2000) A flexible inversion algorithm for retrieval of aerosol optical properties from Sun and sky radiance measurements. J Geophys Res 105:20,673–20,696 Dubovik O, Holben B, Eck T et al (2002) Variability of Absorption and Optical Properties of Key Aerosol Types Observed in Worldwide Locations. J Atmos Sci 59:590–608 Forster P, Ramasvamy V, Artaxo P et al (2007) Changes in atmospheric constituents and in radiative forcing. In: Solomon S, Qin D, Manning M (eds) Climate Change 2007: the physical science basis. Contribution of working group I to the fourth assessment report of the intergovernmental panel on climate change. Cambridge University Press, Cambridge, UK, New York, USA Holben B, Eck T, Slutsker I et al (1998) AERONET – a federated instrument network and data archive for aerosol characterization. Remote Sens Environ 66:1–16 Holben B, Tanré D, Smirnov A et al (2001) An emerging ground-based aerosol climatology: Aerosol Optical Depth from AERONET. J Geophys Res 106:12067–12097 King M, Kaufman Y, Tanre D et al (1999) Remote sensing of tropospheric aerosols from space: past, present, and future. Bull Am Meteorol Soc 80:2229–2259 Kokhanovsky A (2008) Aerosol optics. Light absorption and scattering by particles in the atmosphere. Springer and Praxis Publishing, New York, London Li Z, Blarel L, Podvin T et al (2008) Transferring the calibration of direct solar irradiance to diffusesky radiance measurements for CIMEL Sun-sky radiometers. Appl Opt 47:1368–1377 Penner J, Andreae M, Annegarn H et al (2001) Aerosols, their direct and indirect effects. In: Houghton J, Ding Y, Griggs D et al (eds) Climate Change 2001: the scientific basis. Contribution of working groupe I third assessment report of the intergovernmental panel on climate change. Cambridge Univers Press, Cambridge, UK, New York, USA Schuster G, Dubovik O, Holben B (2006) Angstrom exponent and bimodal aerosol size distributions. J Geophys Res 111:D07207,1 – D07207,14
Global Distribution of Magnetic Storm Fields and Relativistic Particles Fluxes Olga Maksimenko and Galyna Melnyk
Abstract The results of a total magnetic field and ring current (RC) field calculations in the inner magnetosphere (±10 RE) are presented, using Tsyganenko’s T01 empirical model for disturbed magnetospheric magnetic field. The maps of the spatial distribution of model magnetic fields for an intensive magnetic storm on April 6–7th, 2000 were analysed. Moving the large-scale non-uniform structural boundaries change is picked out in the geomagnetic model field distribution both for the main phase of the magnetic storm and for quiet day. The difference of the contributions of RC and cross-section tail current magnetic fields in the total magnetic field during the main and the recovery phases of the magnetic storm is determined. Different dynamics of the relativistic proton and electron fluxes at a geostationary orbit near the magnetic storms has been shown. Keywords Magnetic field • Model • Magnetic storm
Introduction Magnetic storms have always been a central element in space weather studies. They have a global character with the complicated distribution of the magnetospheric magnetic field variations. Three phases in the development of magnetic storms – initial and main (several hours) and longer (several days) recovery phases are allocated. The storm intensity is defined by the maximum value of the horizontal component field depression relative to the quiet level in low and middle latitudes (the disturbance storm-time index – Dst). Geomagnetic field space-time variations during storms are caused by current system structures in inner magnetosphere and an ionosphere which condition is supervised by parameters of an interplanetary
O. Maksimenko (*) and G. Melnyk Institute of Geophysics, Ukrainian National Academy of Sciences, Kiev 252601, Ukraine e-mail:
[email protected]
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0_33, © Springer Science+Business Media B.V. 2011
295
296
O. Maksimenko and G. Melnyk
magnetic field (IMF) and a solar wind (SW). They depend on the type of the interplanetary activity: magnetic clouds (МC), shock waves, CIR – areas of the plasma compression in the solar wind high-speed stream. Basic sources in magnetosphere are: the ring current RC (symmetric SRC and partial PRC), the cross-section tail current (TC), field-aligned Birkeland’s currents, magnetopause currents, etc. It is impossible to determine geomagnetic field variations each of these sources on observation data. Modern empirical magnetospheric field models (Tsyganenko 2002; Tsyganenko et al. 2003) allow us to solve this problem. The difference between the time of development and life of each current source depends on the storm intensity (Tsyganenko 2002; Tsyganenko et al. 2003; Kozyra and Liemohn 2003). They cause dynamics of their relative contribution to the total magnetic disturbance field defines the character of field intensity changes depending on a storm’s development phase. In turn, the storm magnetic field intensity depends also on the geophysical storminess level and previous background energetic particles population in the magnetosphere. Mainly the RC contains protons (60 keV), oxygen ions (80 keV) and electrons (30–60 keV). Its life time and densities depend on nonlinear processes of scattering and acceleration of energetic particles inside the magnetosphere. Different changes of relativistic electrons (>0.6 MeV, Ee > 2 MeV) and protons (Ep > 1 MeV) fluxes are noted in an external radiation belt and in most cases on L > 4RE during intensive magnetic storms. Their spectra and fluxes are measured on low-altitude and geostationary satellites LANL, GOES-8, 10 (L = 6.6 RE) (Reeves et al. 2003; Kanekal et al. 1999). Magnetosphere magnetic field model calculations are extremely necessary at identification of precipitating energetic particles areas borders in the ionosphere, auroral electrojets locations and establishment of criteria of their correct identification with of sources borders in corresponding magnetosphere plasma domains (Feldstein et al. 2006; Newell et al. 2004). In this paper the global distribution of model (Tsyganenko’s Т01) magnetic fields on distance <15 RE is discussed. The dynamic of relativistic particles fluxes variations, measured onboard satellites GOES 8, 10 on 135°W and 225°W longitude at the geostationary orbit, near a storm is considered on an example of magnetic storms with different sources in SW on April 6–7, 2000 and on May 15, 1997. The magnetic fields calculations also have been used for an estimation of the difference of the RC and TC fields relative contribution in the total magnetosphere disturbances field for the main and recovery phases of the strong magnetic storm on April 6–7, 2000.
Data WIND satellites data of solar wind parameters and IMF, INTERMAGNET observatories (Argentine Islands AIA, Alibag ABG) 1-min data of geomagnetic field are used for the analysis. Changes of the magnetic field during storms were determined as difference between disturbed and quiet levels. Quiet level was calculated as average value between last quiet day before a storm and first quiet day after it. As the characteristic
Global Distribution of Magnetic Storm Fields and Relativistic Particles Fluxes
297
of magnetic storminess level the indexes SYM and Dst from World Data Centre for Geomagnetism, Kyoto (http://wdc.kugi.kyoto-u.ac.jp) are used. For studying relativistic particles fluxes dynamic (electrons Ee > 0.6 MeV and Ee > 2 MeV and protons Ep > 1 MeV and Ep > 10 MeV) during magnetic storms their daily data of geostationary satellites GOES 08, 10 (www.swpc.noaa.gov/Data/goes.html) were used. Model calculations were carried out using empirical data-based Tsyganenko’s model Т01 for the inner magnetosphere magnetic field. Input parameters of storm magnetic field model include the geodipole inclination angle, Ву, Bz IMF components and solar wind dynamic pressure, Dst-index. Also in this model changes IMF and solar wind velocity for the previous period of the calculations moment are considered by introduction of two functions g1 and g2, depending on Bz and By IMF and solar wind velocity. As a result the total magnetosphere disturbances field is represented as the sum of the basic current sources fields: B = BCF + BRC + BTC + BFAC1 + BFAC 2 + Binf , where BCF – magnetopause currents field, BRC – ring current field including separately the symmetric BSRC and the partial BPRC ring currents fields, BTC – cross-tail currents field, field-aligned currents of areas 1 BFAC1 and 2 BFAC2 fields. Last member Binf represents an interaction field between the geomagnetic field and IMF.
eomagnetic Variations During April 6–7, 2000 G Magnetic Storm The April 6–7, 2000 magnetic storm was the second strongest in the year 2000 if quantified by the peak of the Dst-index. It has arisen after coronal mass ejection (CME) on April 4, 2000 observed near to the western Sun limb. The front of shock wave CME has reached the Earth’s magnetosphere on April 6. The magnetic cloud has touched magnetosphere only by the flank in recovery phase. After analyzes this sequence of events using observations of several spacecraft in the solar wind and at geostationary orbit as well as recordings from more than 80 magnetometer stations at latitudes higher than 40°N, it was determined that such intensive storm (Dst = −320 nT, Kp = 9) was caused by the very large solar wind magnetic pressure, which compressed the dayside magnetopause inside geostationary orbit for a period of more than 6 h (Huttunen et al. 2002). Changes IMF and SW parameters and also the geomagnetic field on low-latitude (ABG) and mid-latitude (AIA) observatories are presented on Fig. 1. The storm sudden commencement SC has been registered by ground observatories at 16:40 on April 6. Before SC, about 16:20, the solar wind velocity V has increased by 200 km/s, the magnitude B IMF has grown from 5 to 28 nT, By from −7 nT to −27 nT, component Bz from positive became negative (−10 nT). Such changes also have served as the reason of the ring current development on the Earth almost right after SC. The solar wind dynamical pressure Psw finds out two accurate maxima: at 23:00–24:00 on April 6 (I) and at 02:00–04:00 on
298
O. Maksimenko and G. Melnyk
Fig. 1 Variation of interplanetary magnetic field components (Bz, By), dynamic pressure (Psw) and solar wind velocity (V) according to satellites WIND measurements; variations geomagnetic field horizontal components (dH) based on INTERMAGNET observatories and the index of intensity magnetic field H-components of symmetric ring current SYM based on World Data Centre for Geomagnetism, Kyoto
April 7 (II). The first of them was observed at negative Bz and accurately showed in the form of the substorm in the American sector (in AIA +700 nT). In the maximum II at 02:00–04:00 such effect is not shown, because Bz is positive. The substorm is observed at the entire Australian-Asian sector (in ABG +60 nT). Difference
Global Distribution of Magnetic Storm Fields and Relativistic Particles Fluxes
299
between substorms at 23:00–24:00 and at 02:00–04:00 is caused by change Bz IMF which from negative became positive at 24:00UT. The first substorm proceeds in a peak of the storm main phase; the second is appeared on the recovery phase. This storm has not the initial phase and the main phase begins right after SC and then, after 6 h, the recovery phase begins. During this storm several cases short-term (less than 8 mines) huge induced currents (to 34А) in high-latitude regions were fixed (Huttunen et al. 2002).
hanges in Relativistic Particles Fluxes C at Geosynchronous Orbit The analysis of the relativistic protons changes and electron fluxes on L = 6.6 RE in a storm vicinity on an example of magnetic storms with the sudden commencement, caused by a shock wave from the rapid propagation of the СМЕ is made. The intensification of protons fluxes 1 day before the storm, (April 4, 2000), is well visible on schedules of variations of relativistic particles daily fluxes (Fig. 2). The proton fluxes with Ep > 10 MeV was more in 6.6 times in comparison with its value (3.7 × 105 pr/sm2-day-sr) in day of the storm main phase 6.04.2000. For protons concerning low energy (Ep > MeV) the flux reduction in 1.2 times to 5.9 × 107 pr/sm2-day-sr 6.04.2000 is noticed also. We will especially pay attention to particles
4
1
3
0.5
2 30.04.97
Main phase 06.07.04.00
�108
1 0.5 30.03.00
�107
�1010
3 2.5 2 1.5 1 0.5 0 01.03.00
0 29.04.00
10 8 6 4
>2MeV
30.04.97
6 5 4 3 2 1 0 30.05.97
2.5 1.5
4
0 01.03.00
1 30.05.97
�106
2
6
2
>2MeV
>0.6MeV
8
Electron/sm2-day-sr
�1010
6 5 4 3 2 1 0 01.04.97
�107
>10MeV
1.5
0 01.04.97
b
5
>10MeV
>1MeV
2
�104
>1MeV
Main phase 15.05.97
�106
>0.6MeB
a
Proton/sm2-day-sr
2 30.03.00
0 29.04.00
Fig. 2 Changes of the relativistic particles daily fluxes registered on geostationary satellites GOES-8, GOES-10 (http://www.ngdc.noaaa.gov), during 2 months near to magnetic storms: – for a moderate storm on May 15, 1997 (Fig. 2a) and for the strong storm on April 6–7, 2000 (Fig. 2b). On the top panels variations of daily proton flux with energy Ep > 1 MeV (an empty circle, an axis at the left) and Ep > 10 MeV, and on bottom – the flux of relativistic electron with energy Ee > 0.6 MeV (an empty circle, an axis at the left) and Ee > 2 MeV are shown
300
O. Maksimenko and G. Melnyk
flux recession for next days after the storm main phase, in a recovery phase 7.04.2000: almost 10 times for protons with Ep > 1 MeV and in 25 times for electron with energy Ee > 2 MeV. It is interesting that after storm flux of a electron (Ee > 2 MeV) increases with peak 7.4 · 107 el/sm2-day-sr in 4 days after the storm main phase and flux of electron with low energy (Ee > 0.6 MeV) is appeared for the third day April 9, 2000 to 2.7 × 1010 el/sm2-day-sr. It is in 13 times larger for the day of April 6, 2000 of the main phase. The total flux of protons with Ep > 1 MeV has decreased by 50 times and with Ep > 10 MeV by 44.8 times while similar electron flux with Ee > 0.6 MeV and Ee > 2 MeV have grown by three and seven times, accordingly. The more energetic relativistic protons fluxes did not change almost after a storm compare to its level before the storm. On the other hand, on a day of the storm main phase (April 6, 2000) the increase in daily fluxes of protons with Ep > 1 MeV and Ep > 10 MeV are 40 and 6.6 times, accordingly (mainly in low energy range) in comparison with their values after a storm was observed. Daily electron fluxes with Ee > 0.6 MeV and Ee > 2 MeV have on the contrary decreased by 50 and 100 times, accordingly. Thus, it has been revealed by geostationary orbit that the strong magnetic storm was accompanied by loss of electron and increase in the protons fluxes in the storm main phase and the general recession of relativistic particles fluxes in the storm recovery phase when vertical component Bz IMF has changed the direction for the northern. However, through 3–4 days after the storm main phase the electrons fluxes not only has exceeded reference values, but also there were additional energetic particles, is possible owing to non-adiabatic acceleration mechanism (Reeves et al. 2003). For comparison, the dynamics of relativistic particles on May 1997 has been analysed (Fig. 2а), when the moderate magnetic storm on May 15, 1997 (SC at 02:00UT; Dst = −115 nT at 12:40 UT) was observed. The beginning of its main phase has coincided with the arrival of long (9:06–01:00 UT) magnetic cloud (http://lepmfi. gsfc.nasa.gov/mfi/mag_cloud_pub1.html). The electron fluxes increase in the storm recovery phase with a maximum (1.1 × 108el/sm2-day-sr) for 5 days after the magnetic cloud, 20.05.1997, is well visible as well as the enough long interval (11–15.05.1997) its minimum level before a storm. The protons fluxes with Ep > 10 MeV strongly increases 3 days prior to the storm during solar flare. We pay attention that growth of protons fluxes (Ep > 1 MeV) was observed in day of the storm main phase, 5.15.1997 (to 9.6 × 105 pr/sm2-day-sr) and for 7 days after a storm, 22.05.1997 (1.3 × 105 pr/sm2-day-sr).
patial Distribution of Modelled Magnetic Fields in Inner S Magnetosphere (<10 RE) During the Magnetic Storm on April 6–7, 2000 For understanding global distribution of geomagnetic storm fields in the inner magnetosphere, the maps of modelled magnetic field of Bz-component in the GSM system co-ordinates for total storm magnetic field BzG0 (Fig. 3а) and ring current magnetic field BzG4 (Fig. 3b) were constructed. Analysis show that there is a shell
Global Distribution of Magnetic Storm Fields and Relativistic Particles Fluxes
301
Fig. 3 Distribution of (a) total magnetic field BzG0 and (b) ring current magnetic field BzG4 in magnetosphere on distance 1RЕ < L < 10RЕ for planes ZGSM = + 2RE (panel І), ZGSM = 0RE (panel ІІ) and ZGSM = −2RЕ (panel ІІІ) in the storm main phase on April 6, 2000 and in magnetic quietest day on April 11, 2000, ZGSM = 0RE (panel ІV)
302
O. Maksimenko and G. Melnyk
like area of large negative values BzG4 which has sharper external edge and the minimum on the distance of 3.8 RE. The minimum BzG4 is observed more close to a midday meridian, than minimum BzG0 displaced by midnight (21:00). The subsun boundary of magnetopause was visible on distance ~8 RE. The zero values line of BzG0 from an outer side of the negative field area caused BzG4, crosses equator on XGSM = 5.4 RE at midday and XGSM = − 6.2RE at night and in the morning when it is extended on 0.8–1 RE during a magnetic storm. The topology of a magnetic field in magnetosphere can be investigated, analyzing similar maps on a set of planes, it is consecutive displaced on different distances from an equatorial plane (ZGSM = ±1 RE, ZGSM = ±2 RE, ZGSM = ±4 RE) and including azimuth and radial magnetic field component (By, Bx). Figure 3 (panels I) shows the maps for BzG0 and BzG4 magnetic fields on the plane parallel equatorial and displaced on ZGSM = 2 RE. Owing to an inclination of a dipole axis a magnetosphere boundary position is further on 0.2 RE on ZGSM = 2 RE, than on ZGSM = 0 RE (panels II). Minimum BzG4 is displaced by midnight (20.5 h) on plane ZGSM = 2 RE while on ZGSM = −2 RE it is displaced in an opposite side (panels III). Contribution of BzG4 in BzG0 is decreased (BzG4/BzG0 = −90 nT/−135 nT = 0.63) with an expansion of the evening minimum BzG4 area on 0.5 RE in comparison with plane ZGSM = 0 RE. Table 1 presents, values of modelled magnetic field contributions to BzG4 and cross-tail current BzG2 in total magnetosphere magnetic field BzG0 along a radial profiles in different daily sectors. Simulations were fulfilled based on Tsyganenko’s Т01 empirical model for main (19:10 UT 6.0 4.2000), recovery (15:10 UT 7.04.2000) phases of the magnetic storm 6/7.04.2000, and for the quiet day, 11.04.2000. The data is received for indexes Dst close values (−125 nT and −139 nT) located at displaced on −4 h and +15 h from the moment of Dstmin in storm main and recovery phases, accordingly. Contribution of the BzG4 into BzG0 of the main phase can reach 100% during the day and slightly less than 90% in evening. In a quiet day, when the ring current was symmetric, its field accounts less than 70% from the full field. At the same time, in the main phase, we should underline strengthening of BzG4 and occurrence of a maximum of its size with the evening asymmetry, obviously appreciable intensity owing to increase in the PRC. Thus, the model T01 shows that at the beginning of the magnetic storm main phase (19.10 on April 6, 2000 (Dst = −125 nT), in 2.5 h after the sudden commencement of the storm (16:41)), the boundary of dayside magnetopause was at Table 1 Relative contributions of the maximum values of ring current (BzG4/BzG0) and cross-tail current (BzG2/BzG0) magnetic fields in the total magnetic storm full field for different local time in the main (19:10 UT on April 6, Dst = −125 nT) and recovery (15:10 UT, April 7 Dst = −139 nT) phases of the magnetic storm on April, 6–7 and on April, 11, (quiet day) 2000 Date Morning Evening Night Day On April, 6th main phase 0.76 0.90 0.60 1.00 BzG4/BzG0 BzG2/BzG0 0.64 0.41 0.49 0.80 On April, 7th recovery phase BzG4/BzG0 0.47 0.60 0.40 0.61 BzG2/BzG0 0.29 0.19 0.70 – On April, 11th quiet day 0.71 0.71 0.60 0.70 BzG4/BzG0
Global Distribution of Magnetic Storm Fields and Relativistic Particles Fluxes
303
8 RE that is closer to the Earth than its position during quiet conditions (11.4 RE) (Fig. 3, panels IV). The crossing section size of the daily magnetosphere on distance XGSM = 3.5 RE was 22 RE. At displacement from equator it is observed the magnetosphere expansion on meridian with north-southern asymmetry. According to the estimations, the area of negative magnetic fields is also extended asymmetrically along a line day-night with boundaries XGSM = 5.4 RE in the afternoon and XGSM = −6.5 RE at night. The allocated spatial minimum of field intensity BzG4 = −110 nT in the area of the developed ring current was observed on a distance XGSM = 3.8 RE, in evening. It was amplified during the great magnetic storm by 2.2 times relative to quiet field level BzG4 = −50 nT on April 11, 2000. During the main phase, the area of negative field BzG4 was expanded on 1.8 RE and the line of a field change was located at −8.2 RE to the night part and 7.9 RE in a subsun point. The last reflects day magnetosphere compression by the solar wind plasma with increase in magnetic field gradient which values is defined by SW parameter and depends on a magnetic storm phase and its intensity. The azimuth displacement of minimum BzG0 from midnight (XGSM = −3.6 RE, YGSM = 2.0 RE) in an afternoon direction to XGSM = −2.2 RE, YGSM = 3.2 RE at strengthening of magnetic field value BzG0 by 1.4 times (to BzG0 = −145 nT). At the beginning of the main phase the relation BzG4min/ BzG0min was 0.76. The relative contribution of the basic current sources fields in the total magnetic field of magnetosphere disturbances does not remain to constants and changes during a magnetic storm depends on a source of a storm and its intensity that does not contradict the data of other researchers (Kozyra and Liemohn 2003; Kalegaev et al. 2005).
Conclusion Based on the Tsyganenko’s Т01 empiric model, large-scale structures inside the global distribution of a storm’s magnetic field and the distance <15 RE in the magnetosphere were defined. Areas with the large negative magnetic fields in the form of a belt on an equatorial plane are caused by the ring current (RC) with evening maximum on L = 3.5 RE which is displaced in a direction of afternoon sector relative to the maximum of the total magnetic field with an increase in intensity of a storm, Dst. Boundary of day magnetopause in quiet magnetic conditions are found with a subsun point on distance 11.5 RE which appears closer to the Earth (8 RE) at the beginning of the main phase of a storm on April 6, 2000 at Dst = −125 nT. Zero lines of magnetic fields RC have asymmetry in a direction day-night with marks on 6.2 RE on the night party of equator and further on 0.8–1 RE from a surface of the Earth. The relative contribution of magnetic field RC’s BzG4 to total magnetic field BzG0 during the magnetic storm on April 6, 2000 continue on April 7 contributing to the cross-tail magnetosphere of the current magnetic field BzG2. Along YGSM-axis the values were 0.9 and 0.43 at the beginning of the main phase decreasing to 0.6 and 0.2 for a cross-tail current magnetic field at a storm recovery phase (Dst = −139 nT).
304
O. Maksimenko and G. Melnyk
At the beginning of the main phase the primary contribution of RC magnetic field reached 100% unlike its rather constant level 60% in all sectors during quiet conditions on April 11 in the afternoon.
References Feldstein Y, Popov V, Cumnock J et al (2006) Auroral alectrojets and boundaries of plasma domains in the magnetosphere during magnetically disturbed intervals. Ann Geopys V 24(8):22–2276 Huttunen K, Koskinen H, Pulkkinen T, et al (2002) April 2000 magnetic storm: solar wind driver and magnetospheric response. J Geophys Res A12 107:1440. doi: 10.1029/2001JA009154 Kalegaev VV, Ganushkina NY, Pulkkinen TI et al (2005) Relation between the ring current and the tail current during magnetic storms. Ann Geopys V 23(2):523–533 Kanekal S, Baker D, Blake J et al (1999) Magnetospheric response to magnetic cloud (coronal mass ejection) events: relativistic electron observations from SAMPEX and Polar. J Geophys Res 104:24885–24894 Kozyra J, Liemohn M (2003) Ring current energy input and decay. Space Sci Rev 109:105–131 Newell P, Ruohoniemi J, Meng C-I (2004) Maps of precipitation by source region, binned by IMF, with inertial convection streamlines. J Geophys Res 109 A10206. doi: 10.1029/2004JA010499 Reeves G, McAdams K, Friedel R et al (2003) Acceleration and loss of relativistic electrons during geomagnetic storms. Geophys Res Lett 30(10) 1529. doi: 10.1029/2002GL016513 Tsyganenko N (2002) A new magnetospheric magnetic field model. 1. Mathematical structure 2. Parameterization and fitting to observations. J Geophys Res V107 A8. doi: 10.1029/2001JA000220 Tsyganenko N, Singer H, Kasper J (2003) Storm-time distortion of the inner magnetosphere: How severe can it get? J Geophys Res V108 A5. doi: 10.1029/2002JA009808
Appendix 1
−45
45
135
−45
45
135
−15
−15
2007
− 45
− 45
15
15
45
45
−135
45
45
−135
− 45
− 45
−15
−15
15
15
2008 −135 STRESSED 6
− 45
45
135 FAVORABLE
FAIR 12
24
36
48
60
72
84
Vegetation Health in mid-July Fig. 1 Vegetation health index in mid-July 2007 and 2008 (For the paper “Early Detection and Monitoring Droughts form NOAA Environmental Satellite” by Felix Kogan and Wei Guo in Part I)
F. Kogan et al. (eds.), Use of Satellite and In-Situ Data to Improve Sustainability, NATO Science for Peace and Security Series C: Environmental Security, DOI 10.1007/978-90-481-9618-0, © Springer Science+Business Media B.V. 2011
305
51
49
Cereal Yield 27.5
45
45
47
47
51
49
Appendix 2
2002 34
38
51
30
51
49
26
49 45
45
47
47
2003 30
34
38
28.4
45
45
47
47
49
49
51
51
26
18.4
2004 26 6
12
30 24
36
34 48
60
38 72
84
Fig. 1 Mid-May VH and Ukraine’s average cereal yield, 2002–2004 (For the paper “SatelliteBased Crop Production Monitoring in Ukraine and Regional Food Security” by Felix Kogan, Tatiana Adamenko and Mikola Kulbida in Part II)
307
Appendix 3
Fig. 1 Correlation of VH indices (VCI, TCI, VHI) with SST anomaly in 3.4 tropical Pacific during the ENSO years in 1982–1997 (For the paper “ENSO Impact on Vegetation” by Felix Kogan in Part III)
309
Appendix 4
Fig. 2 Lag correlation of VHI for December–February with SST anomaly for December– February (lag0), September–November (lag3), June–August (lag6) and March–May (lag9). Red box indicates 3.4 area where SST data were collected (For Fig. 2 “ENSO Impact on Vegetation” by Felix Kogan in Part III)
311
Appendix 5
0.8
GIMMS
Asia
GVI-x
NDVI
0.6 0.4 0.2 82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99 LTDR
0.8
GIMMS
South America
GVI-x
NDVI
0.6 0.4 0.2 LTDR GIMMS
Africa
GVI-x
NDVI
0.6 0.5 0.4 0.3 0.2 0.1 82
83
84
85
86
y1 = 0.00088*x − 1.5
87
88
89
90
91
92
93
y2 = 0.0001*x + 0.24
94
95
96
97
98
99
y3 = 0.00014*x − 0.025
Fig. 1 Processed mean NDVI (left) and BT (right) for three global latitude bands (For Fig. 1 “Comparison of AVHRR-Based Global Data Records” by Felix Kogan, Marco Vargas and Wei Guo)
313
