Harvesting insights: interpretable machine learning to understand environmental drivers of U.S. maize and soybean yield

Accurate crop yield prediction is crucial for enhancing food security and agricultural sustainability; however, existing models frequently struggle to capture the intricate relationships between environmental drivers and crop performance. Here we leveraged a large, spatially explicit yield monitor dataset of U.S. commercial maize (Zea mays) and soybean (Glycine max) fields (134 unique crop-site-years). Machine learning models were trained to predict yield with high accuracy (R2 > 0.87, RMSE < 1.13 Mg ha−1), and Shapley Additive Explanations were used to quantify how weather, soil, and terrain properties predict yield variability. Our results highlight the potential of machine learning to disentangle environmental constraints on crop production, thereby providing actionable insights for more resilient U.S. food systems. The results presented here represent a novel approach to identifying maize and soybean yield constraints that can inform the next generation of crop breeding and precision management strategies.