Forecasting corn yield in Iowa using remotely sensed data and vegetation phenology information

摘要

Iowa is the number one com-producing state in the U.S. annually producing 1.77 billion bushels of corn, generating around 1.5 billion dollar annually. Consequently, fluctuations in total corn production have great economic impacts on U.S. agriculture. Pre-harvest estimates of final corn yields are important to identify areas of surplus or deficit early in the year. Time integrated (TI) Normalized Difference Vegetation Index (NDVI) and Vegetation Phenological Metrics (VPM) data sets were derived from a 10-year (1989-1998) multi-temporal Advanced Very High Resolution Radiometer (AVHRR) NDVI imagery data set were derived to forecast crop reporting district (CRD) yields. To ensure model validity and robustness, mean absolute percent errors (MAPEs), were calculated using jackknife regression techniques. Results showed that models built using the VPMs (average MAPE = 13.7 percent) performed significantly better than those built using the TI NDVI (average MAPE = 14.2 percent). Overall, yield forecast models derived from the TI NDVI and VPM provided acceptable pre-harvest estimates of final com yields in Iowa,