Benefits of Seasonal Climate Prediction and Satellite Data for Forecasting US Maize Yield

Seasonal agricultural production forecasting is essential for agricultural supply chain and economic prediction. However, to what extent seasonal climate prediction and remote sensing observations can improve crop yield forecasting at regional scale remains unknown. Using a statistical seasonal forecasting framework for U.S. county-level maize yield, we demonstrated that (1) incorporating satellite-based enhanced vegetation index (EVI) significantly improved the yield forecasting performance, compared with other climate-only models using monthly air temperature (T), precipitation (P), and vapor pressure deficit (VPD). (2) The bias-corrected climate prediction from the Coupled Forecast System model version 2 (CFSv2) showed better yield forecasting performance than the historical climate ensemble. (3) Using the T + P + VPD + EVI model with climate prediction from bias-corrected climate prediction from CFSv2 outperformed the yield forecast in the World Agricultural Supply and Demand Estimates reports released by the United States Department of Agriculture, with root-mean-square error of 4.37 bushels per acre (2.79% of multiyear averaged yield) by early August. Plain Language Summary Given the significant advances in both seasonal climate prediction and satellite remote sensing, these data have not been fully used in crop yield forecasting at regional scale, and their benefits are to be quantified compared to survey-based approaches. Here we evaluated the benefits of using seasonal climate prediction and satellite remote sensing data in forecasting U.S. maize yield at both national and county levels. To achieve this goal, we built a seasonal forecasting system for U.S. maize yield by bridging the most advanced seasonal climate prediction products from National Centers for Environmental Prediction (NCEP) with a statistical crop modeling framework. We found we could not achieve a better forecasting performance than the official survey-based forecast from United States Department of Agriculture until we used both climate and remote sensing observations in our model. Compared with using historical climate information for the unknown future in each growing season, using climate prediction from NCEP gave better forecasting performance once we corrected the bias in the seasonal climate prediction products. Using our climate-remote sensing combined model and bias-corrected climate prediction from NCEP, we achieved a better forecasting performance than the United States Department of Agriculture forecast. Our system will be useful for the stakeholders in the agriculture industry and commodity market.