Climate change impacts on crop yields: A review of empirical findings, statistical crop models, and machine learning methods

Understanding crop responses to climate change is crucial for ensuring food security. Here, we reviewed similar to 230 statistical crop modeling studies for major crops and summarized recent progress in estimating climate change impacts on crop yields. Evidence was strong that increasing temperatures reduce crop yields. A 1 degrees C warming decreased the yields by 7.5 +/- 5.3% (maize), 6.0 +/- 3.3% (wheat), 6.8 +/- 5.9% (soybean), and 1.2 +/- 5.2% (rice) across the world, but spatial heterogeneity was noticeable, due partly to asymmetric nonlinear crop responses to temperature (e.g., warming-induced gains in cold regions). Yield responses to precipitation were not consistent across the studies or geographical areas. On average, climate explained 37% of yield variability. We also observed a methodological shift from linear regression to machine learning (e.g., explainable AI and interpretable machine learning), which on average reduced predictve errors by 44%. Furthermore, we discussed the opportunities and challenges facing statistical crop modeling, such as ensemble modeling, physics-informed machine learning, spatiotemporal heterogeneity in crop responses, climate extremes, extrapolation under novel climates, and the confounding from technology, management, CO2, and O-3.