Forecasting the Weather Impact on Corn Production with Machine Learning

Corn is a foundational commodity in the modern economy, supporting food systems, fuel production, and numerous industrial applications. For businesses across the agricultural supply chain, accurately forecasting both corn yield and price is essential for managing costs, anticipating revenue, and reducing financial risk. This study evaluates the extent to which machine-learning models can predict corn yield and price in Iowa, Illinois, and Nebraska—three states that collectively produce approximately 40% of U.S. corn (United States Department of Agriculture, National Agricultural Statistics Service, n.d.). Using historical weather and production data from 1980 to 2019, a series of supervised learning models were developed to assess how effectively climate and economic variables forecast annual outcomes.

Among the yield models, linear approaches performed the best. The climate-only models revealed a clear inverse relationship between heat stress and yield, aligning with other known models. Overall, the final models explained roughly 30% of the variation in annual corn yield. In contrast, corn price proved far more predictable, with linear models accounting for approximately 82% of the variation in annual price levels. These results highlight the differing complexity of the two prediction tasks: yield is influenced by many factors not captured in climate summaries, while price responds more directly to broad economic signals.

Several features were created to better represent the ideal growing environment for corn.  Number of extreme heat days in a year, number of good degrees of heat needed to grow corn, and the Max dry spell.  Number of extreme heat days was the best feature.