Dynamic climate risk modelling for agriculture: A machine learning approach integrating crop phenology into weather index insurance

The Midwest United States experiences significant climate variability that affects the optimal soybean planting date. Inaccurate estimation of planting dates in actuarial models for Weather Index Insurance (WII) can increase basis risk and adverse selection. This study develops a dynamic, transparent WII framework that adapts to temporal climate variability and integrates crop phenology into risk assessment and pricing. A two-step methodology is proposed: first, planting dates are predicted at a 10 km resolution using a machine learning model trained on pre-growing season variables. Second, remote-sensing indices during the growing season are synchronised with predicted phenological stages at weekly intervals to improve county-level crop yield prediction. The framework links these dynamic yield predictions to insurance design, enabling more accurate premium ratings and capital management. The planting date model achieves a mean absolute error (MAE) of 3.33 days and a root mean square error (RMSE) of 4.4 days, outperforming comparable studies. The dynamic yield model improves prediction accuracy by 7.8% (RMSE) and 6.7% (MAE) over a static benchmark based on fixed contract dates. Results show that Gross Primary Production (GPP) during early growth stages drives the dynamic model, whereas the benchmark relies on mid-to-late season climate variables. The stronger early-season signal highlights the dynamic model’s ability to capture critical predictors during germination and early vegetative phases, which are highly indicative of final yields. This framework advances WII design by promoting adaptive, data-driven, and climate-responsive agricultural risk management. • Integrates climate change and land-use dynamics in landslide forecasting • Employs XGBoost with SHAP for explainable AI-based susceptibility mapping • Projects landslide risk in Western Ghats under four future SSP scenarios • Identifies regional hotspots with rising susceptibility by 2100 • Provides data-driven insights for adaptive, climate-resilient planning