Modeling of Cotton Yields in the Amu Darya River Floodplains of Uzbekistan IntegratingMultitemporal Remote Sensing and Minimum Field DataZhou Shi, Gerd R. Ruecker,* Marc Mueller, Christopher Conrad, Nazar Ibragimov, John P. A. Lamers,Christopher Martius, Guenter Strunz, Stefan Dech, and Paul L. G. VlekABSTRACTIncreased knowledge about the spatial distribution of cotton(Gossypium hirsutum L.) yield in the Khorezm region in Uzbekistansupports the optimal allocation of resources. This research estimatedthe spatial distribution of cotton yields in Khorezm by integratingremote sensing, field data, and modeling. The agro-meteorologicalmodel used was based on Monteith’s biomass production model withmultitemporal MODIS (Moderate Resolution Imaging Spectroradiom-eter)-derived parameters from 2002 as primary inputs. The photosyn-thetically active radiation (PAR) and environmental stress scalars oncrop development were estimated with meteorological information.Using high-spatial-resolution Landsat 7 ETM1 images, the cottonarea was extracted and the cotton fraction determined within thecoarse spatial resolution MODIS pixels. The spatial resolution of theMODISFPARdatawasimprovedbyusinganempiricalrelationshiptothe higher-resolution MODIS NDVI (Normalized Difference Vegeta-tion Index) data. The estimated raw cotton yield ranged from 1.09 to3.76 Mg ha 21 . The modeling revealed a spatial trend of higher yieldsin upstream areas and in locations closer to the irrigation channelsand lower yields in downstream areas and at sites more distant to thechannels. The validatedyield estimations showed a 10%deviationfromofficial governmental statistics. The established agro-meteorologicalmodel based on freely available MODIS data and a minimum of fielddata input is a promising technique for economic and operational late-season estimationof spatially distributedcotton yield overlargeregionson which management adjustments could be made.ACHIEVING ECOLOGICALLY SUSTAINABLE AGRICULTURALDEVELOPMENT in the Aral Sea basin is not an easytask. Cotton production plays a dominant role, covering46% of all irrigated land in Uzbekistan. However, thevolume of water from the Amu Darya River, which isone of the major sources of water for irrigation, grad-ually becomes limited due to increasing water demandin upstream irrigation regions and in neighboringcountries (Ressl et al., 1998; Vlek et al., 2003). The rela-tionship between cotton yield and water balance is im-portant for targeting water allocation and planning landuse restructuring.Remote sensing is the premier technology for givingan unbiased view of large areas by providing spatiallyexplicit information repeatedly collected, and it has thusbeen widely used to estimate crop yields at a regionalscale (Quarmby et al., 1993; Baez-Gonzalez et al., 2002;Bastiaanssen and Ali, 2003; Doraiswamy et al., 2003).One commonly applied methodology is to develop em-pirical relationships between the normalized differencevegetation index (NDVI) and crop yield (Groten, 1993;Dalezios et al., 2001; Conrad et al., 2004a). However,a simple empirical relationship is only of local and in-stantaneous significance. Furthermore, an empirical as-sessment requires excessive measurement programs tocollect yield data in the field, which on a regional scale istime consuming and costly (Moulin et al., 1998).The agro-meteorological model based on solar radia-tion and leaf development, and therefore also referredto as the light-use-efficiency model (Monteith, 1972),has great potential for estimating crop yield using satel-lite and agro-metereological data as inputs (Moulinet al., 1998; Lobell et al., 2003). Satellite data such asthose from NOAA-AVHRR (National Oceanic and At-mospheric Administration-Advanced Very High Reso-lution Radiometer) and MODIS (Moderate ResolutionImaging Spectroradiometer) with high temporal fre-quency have mainly been used for model runs at a dailytime step. Bastiaanssen and Ali (2003), for example,applied the linear relationship between the fraction ofphotosynthetically active radiation (f) and NDVI de-rived from AVHRR to estimate the fraction of absorbedphotosynthetically active radiation (FPAR) for the Mon-teith model. In a study in Burkina Faso, regression mod-els based on AVHRR NDVI time-series data were usedby Groten (1993) for modeling crop yield.The newly available satellite images from the MODISsensor incorporate enhanced atmospheric correction,cloud detection, improved geo-referencing, comprehen-sive data quality control information, and an enhancedability to monitor vegetation development (Runningetal.,1999;Hueteetal.,2002).Moreover,aseriesofstan-dard MODIS products, such as FPAR, vegetation index(VI), leaf area index (LAI), and net primary productiv-ity, is provided by the MODIS land science team. TheMODIS land data have mainly been applied in global-scale change research, but the regional application andZ. Shi, Institute of Agricultural Remote Sensing and InformationSystem, Zhejiang Univ., Hangzhou 310029, China; G.R. Ruecker, G.Strunz, and S. Dech, German Aerospace Center, German RemoteSensing Data Center, Oberpfaffenhofen, 82230 Wessling, Germany;M. Mueller, Institute for Prospective and Technological StudiesEuropean Commission–Joint Research Center Edificio Expo C/IncaGarcilaso, 41092 Sevilla, Spain; C. Conrad, Dep. of Geography, RemoteSensing Unit, Bayerische Julius-Maximilians-University Wuerzburg,Am Hubland, 97074 Wuerzburg, Germany; N. Ibragimov, UzbekistanNational Cotton Growing Research Institute, P.O. Box Akkavak,702133 Tashkent, Uzbekistan; and J.P.A. Lamers, C. Martius, and P.L.G.Vlek, Center for Development Research, Univ. of Bonn, Walter-Flex-Str. 3, 53113 Bonn, Germany. Received 12 Sept. 2006. *Correspondingauthor (gerd.ruecker@dlr.de).Published in Agron. J. 99:1317–1326 (2007).Modelingdoi:10.2134/agronj2006.0260ª American Society of Agronomy677 S. Segoe Rd., Madison, WI 53711 USAAbbreviations: CDF, cumulative distribution function; DOY, day ofyear; ETM+, Enhanced Thematic Mapper Plus; FPAR, fraction ofabsorbed photosynthetically active radiation; LAI, leaf area index;MODIS, Moderate Resolution Imaging Spectroradiometer; NDVI,normalized difference vegetation index; PAR, photosynthetically ac-tive radiation; VPD, vapor pressure deficit.Reproduced from Agronomy Journal. Published by American Society of Agronomy. All copyrights reserved.1317
