Incorporating soil moisture data into a machine learning framework improved the predictive accuracy of corn yields in the U.S

Corn is one of the most important global agricultural commodities, contributing to food, feed, and fuel production. However, in recent decades, it has shown signs of decreasing yield gain and vulnerability to extreme weather events. Understanding environmental factors that influence corn yield is crucial for improving crop management and designing more resilient cropping systems. Leveraging machine learning (ML) techniques capable of handling large-scale datasets offers a promising alternative for uncovering hidden patterns and generating actionable insights to improve crop yield. We hypothesize that incorporating soil moisture and temperature data from land surface models into a ML framework will enhance accuracy of corn yield predictions. To test this, we employed XGBoost algorithm using long-term (2010–2022) yield data collected from multi-environment corn trials conducted by the Variety Testing Program at the University of Missouri, along with weather data from Daymet database, soil data from North American Land Data Assimilation System (NLDAS-2), and POLARIS soil database. Our results highlighted that corn yield was highly influenced by environmental variability, and the effect of planting date on yield was region-specific. The ML(selected) model showed that incorporating soil moisture data, along with carefully selected weather variables, can achieve accuracy (r2 and NSE of 0.44) comparable to models using a full suite of agronomic, soil, and weather inputs (r2 and NSE of 0.47). Furthermore, mid-season evapotranspiration and air temperatures, volumetric water content, saturated conductivity, and early-season precipitation emerged as the most important predictors, underscoring crop’s sensitivity to water-related stress. This finding offers a promising pathway for developing data-driven decision-making tools to optimize corn management and yield under regionally variable weather and soil conditions.