Spatially explicit artificial intelligence framework for climate risk assessment and adaptation policy of cotton production in Texas

Cotton production in Texas, which supplies approximately 40% of the United States upland cotton, faces escalating climate risk, yet most assessments assume spatial stationarity and employ models rarely validated for spatial transferability. This study develops a spatially explicit multi-model framework that integrates probabilistic inference with machine learning to quantify climate risk to cotton production across 171 Texas counties from 1981 to 2023. Agglomerative hierarchical clustering partitioned the cotton belt into two climate-production zones, and a cluster-based hierarchical Bayesian regression estimated via Markov chain Monte Carlo (MCMC) sampling with the No-U-Turn Sampler (NUTS) revealed a directional reversal of the temperature risk: each 1 °C increase corresponded to a 9.2% production decline (94% HDI excluding zero) in the semi-arid western cluster but a 16.7% increase in the humid eastern cluster, while water stress exerted a consistent positive effect of approximately 16.5% per unit across both zones. Global Moran’s I confirmed strong spatial autocorrelation for all variables, and Multiscale Geographically Weighted Regression identified predictor specific bandwidths ranging from 43 for water stress to 170 for the intercept, revealing scale dependent spatial structure in climate–production relationships. Machine learning benchmarks exposed a critical gap between in-sample fit (XGBoost R 2 = 0.955) and spatially blocked cross validation performance (R 2 = −0.256), under which no model retained meaningful transferable skill, with OLS achieving the highest but near zero spatial CV R2 of 0.012. SHAP analysis confirmed the dominant role of spatial position and water stress. These findings demonstrate that frameworks bridging probabilistic statistical modeling and data driven machine learning yield more robust and transferable climate risk assessments than either approach achieves independently.