Assessing the impact of climate indices on corn yield in the continental USA using machine learning approach

Authors: SABUT, A - Texas A&M University; TRIPATHY, K - Texas A&M University; MISHRA, A - Texas A&M University; Anderson, Martha; Cosh, Michael; Kraatz, Simon; Gao, Feng; Cirone, Richard

Submitted to: Agricultural and Forest Meteorology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/12/2025
Publication Date: 5/16/2025
Citation: Sabut, A., Tripathy, K., Mishra, A., Anderson, M.C., Cosh, M.H., Kraatz, S.G., Gao, F.N., Cirone, R.J. 2025. Assessing the impact of climate indices on corn yield in the continental USA using machine learning approach. Agricultural and Forest Meteorology. 371. https://doi.org/10.1016/j.agrformet.2025.110632.
DOI: https://doi.org/10.1016/j.agrformet.2025.110632

Interpretive Summary: Understanding the relationships between climate variations on crop yields is important, both in terms of estimating current year yields and in predicting impacts of longer-term climate change trends on future productivity. In this study, we use machine learning techniques to determine the relationships between county-level corn yield data for 1979-2018 from the USDA National Agricultural Statistics Service (NASS) and a number of climatic indicators relating to precipitation, temperature, and humidity. These indicators include metrics such as maximum and minimum growing season temperature, rainfall totals, and number of extreme heat days, with maximum temperatures exceeding 30 C. Five distinct agroclimatic regions were identified, based on dominant climate and yield characteristics. Within each region, different climatic factors were found to be key to determining yield response. For example, in the West-Central Semi-arid Prairie region, low yields were related to limited water availability and high vapor pressure deficits, whereas in the Southeastern and Mississippi Alluvial and Coastal Plain regions, heat stress and soil fertility appear to play a more important role. Growers can leverage these insights to select crop varieties better suited to regional climatic conditions and optimize management practices. Policymakers, insurers, and agribusinesses can use these findings to develop targeted support programs, refine risk assessments, and make strategic investments that promote resilience in agricultural systems.

Technical Abstract: Climate profoundly impacts crop productivity due to its variability and uncertainty in prediction. This study explores a wide range of climatic indices that represent the various conditions affecting corn growth across the United States. By employing clustering techniques, we categorized rainfed corn-growing regions into distinct zones based on similar climatic characteristics to evaluate how each index influences crop yields. We identified the most effective combinations of these indices and used machine learning models at the county level to map the relationships between climatic factors and crop yields. Our analysis reveals that temperature-related indices, such as the number of days with temperatures exceeding 30°C (HD30), Temperature Variance (Tvar), and Extreme Temperature Range (ETR), are the top three factors that negatively impact yields, while SU (number of summer days) has a positive effect. Precipitation-related indices also contribute positively, highlighting the critical role of balanced water availability during key growth stages. Notably, temperature-related indices emerged as the most effective predictors of yield in most regions, demonstrating stronger influence and higher predictive accuracy compared to precipitation indices. At the county level, machine learning models were used to map these relationships, with XGBoost emerging as the most reliable model. It consistently outperformed alternatives like Random Forest, Support Vector Machine, and LASSO, demonstrating superior accuracy and robustness. This was particularly evident during the extreme climatic conditions of 2012, marked by severe drought and heatwaves, where XGBoost accurately captured yield losses without overestimation. Among all factors, HD30 was identified as the most influential climatic driver of yield reductions under heat stress. This study not only enhances our understanding of climatic influences on crop production but also empowers stakeholders like farmers, policymakers, insurers, and agribusinesses to adopt optimized agricultural practices and develop strategic initiatives that enhance agricultural resilience and food security.