Predicting Corn Yields Using MODIS EVI, LAI, and Lst Images in County Level.

Accurate in-season yield prediction maps are beneficial to growers for planning and management decisions and other commercial purposes including marketing and crop insurance.  Previous work relied on images collected over the growing season to create crop yield maps with training data of a post-season; however, there are many advantages if an in-season accurate map can be created.  The objectives of this study  were to:  1) develop a model using MODIS images to predict in-season yield at a county level and 2) determine the relationship between sampling date of crop stage and accuracy prediction. The study area for corn yield mapping was the state of Illinois in 2010. The 2009 CDL was used to provide training for 250-m MODIS data.  The collection of 5 gridded and georeferenced 8-day composite L3 MODIS data [250-m EVI (MOD09Q1G_EVI), 1-km LAI (MOD15A2), and 1-km LST (MOD11A2)] were downloaded from NASA website.  In 2010, MODIS data from day-97 (April 7) to day-297 (October 24) were used.  For the dates of planting, emerging, and growth stages, USDA NASS reports were used. Regression trees were generated with MODIS images and used for classification.  Stepwise regression was used to find the best multiple linear regression model using MODIS data to predict corn yield.  The selected regression model (R²=0.8113) generated from 2010 images used data up to day-201 (July 20, silking). The estimated 2010 corn yields from the model was significantly correlated (R²=0.8135^**) with the NASS 2010 data. For validation, the predicted 2011 corn yields using the 2010 model was significantly correlated (R²=0.6345^**) with the NASS 2011 data.  These results show that:  1) MODIS data can be used to model corn yields and 2) the silking stage timing could be used to predict corn yields.