Explainable artificial intelligence for hydroclimatic controls and mid-century cotton yield projections across Texas

Climate change leads to greater hydroclimatic stress due to higher temperatures and increased atmospheric evaporation demand. This creates major challenges for sustainable cotton production in Texas by increasing crop water stress, indicating the need for tools that identify key hydroclimatic drivers and support region-specific climate adaptation. In this study, a hydroclimatic environmental intelligence framework was developed by combining LOCA2-GFDL-CM4 CMIP6 climate projections (SSP245 and SSP585), Multiple Linear Regression (MLR), Random Forest (RF), and Explainable Artificial Intelligence (SHAP). This framework helps to assess the historical hydroclimatic factors that influence cotton yield and to project future cotton yield (Overall, Irrigated and Non-irrigated) for the middle of the century (2030–2050) across six selected climate divisions in Texas. RF performed better than MLR in 11 out of the 18 combinations of climate division and production system. The correlation coefficients (r) for RF varied from 0.24 to 0.72, as opposed to the range of 0.06 to 0.63 for MLR. The future RF projections showed that the Lower Valley division had the highest median projected yield (0.61–0.62 t ha⁻¹) and was then followed by the High Plains division (0.56–0.57 t ha⁻¹). The North Central division had the lowest median yield (0.37–0.38 t ha⁻¹). Cotton grown with irrigation always had higher projected yields (0.67–1.04 t ha⁻¹) than cotton grown without irrigation (0.35–0.57 t ha⁻¹) under both climate scenarios. The framework proposed offers an interpretable method for determining the main hydroclimatic controls, for predicting future cotton productivity, and for aiding agricultural decision-making that is resilient to climate change.