Contributions of a Dynamic Model Approach to Yield Forecasts of Corn in the US Corn Belt.

Using simulation models to forecast crop yield during the current growing season can help inform management, marketing, logistics, and policy decisions. At issue is whether a crop simulation model can accurately estimate actual yield at different spatial scales (location, region, country) without need of site-year specific calibration. We have established a Yield Forecasting Center (YFC) to forecast maize yield across the US Corn Belt. The approach consists on a well-validated maize simulation model (Hybrid-Maize), site-specific real-time measured weather, and soil and management data provided by collaborators and industry. Same set of model parameters (besides weather, management, and hybrid maturity) was used across all location-years. Simulations are performed in real-time mode, using measured weather to simulate crop growth and development until the date of the forecast and historical weather data to project the rest of the season, which results in a range of possible end-of-season yield. Upscaling of simulated yield from location to larger spatial domains (district, state, and country) is performed following a bottom-up approach based on an agro-climatic zone scheme and distribution of crop harvested area. Using average yield data from 45 U.S. counties, we evaluated model accuracy at reproducing end-of-season maize yield in four years (2011-2014) with contrasting weather conditions. There was a significant relationship between simulated and county yield average, with simulated yields clearly above actual yields at actual yield levels >10 Mg/ha.  Agreement between simulated and actual yields increased when both were expressed relative to the long-term average. Results from this study indicate that models can contribute to forecast regional yield without need of site-specific calibration.