Soybean Area and Yield Estimation Using MODIS and Landsat Data in the Conterminous United States

Thursday, 17 December 2015
Poster Hall (Moscone South)
Xiao-Peng Song1, Matthew Hansen2, Peter Potapov2, Stephen V. Stehman3, Alexander Krylov2, LeeAnn King1 and Bernard Adusei1, (1)University of Maryland College Park, College Park, MD, United States, (2)University of Maryland, College Park, MD, United States, (3)SUNY College of Environmental Science and Forestry, Syracuse, NY, United States
The world’s population is projected to grow to 9 billion by 2050. The increasing population, amplified by people’s increasing consumption of animal products will create a massive demand for food and feed from grain production. As such, global food security will remain a worldwide concern for the next half century. Addressing the food security issue requires data and information support, including research and operational programs for crop monitoring, modeling and yield forecasting. Satellite observations, owing to their synoptic and repetitive nature, have the unique advantage of providing timely information on crop growth at regional to global scales. However, it remains a challenge to accurately identify crop type, estimate areal extent and forecast crop yield with satellite data. Here we employ a stratified random sampling framework for estimating soybean area and yield in the conterminous United States using satellite data collected by the MODIS and Landsat sensors. Complementing each other, the temporally-rich MODIS data are used to capture rapid phenological transitions of soybean crops, whereas the moderate-resolution Landsat data are used to delineate more spatial details for accurate area estimation. For every sample, we derive generic phenological metrics from MODIS and Landsat data and employ machine learning algorithms to identify soybean pixels with reference data generated from RapidEye images and verified by extensive field visits. We also characterize empirical relationships between satellite metrics and soybean yield compiled by the USDA National Agricultural Statistics Service (NASS). Preliminary results suggest that MODIS data alone underestimate soybean area considerably, whereas Landsat data can provide accurate estimate on soybean area. However, soybean yield can be predicted using MODIS-based reflectance metrics. Our sample depict well the spatial variation of soybean yield over the conterminous United States. In addition, the area-weighted national soybean yield derived from the sample can also capture the year-to-year variation in national yield as compared with NASS’s yearly, survey-based yield estimates. These results indicate that the stratified random sampling approach is efficient and accurate for soybean area and yield estimation at national scales.