Stochastic optimization and ensemble learning towards robust irrigation advisory to enhance water productivity under a changing climate

This study investigates irrigation suitability patterns across Morocco to align with the nation’s irrigation policy reforms promoting climate-smart, water-saving, and resilience-based agricultural practices. A novel hybrid modeling framework integrating physics-inspired metaheuristic algorithms with ensemble machine learning (ML) was developed to improve predictive accuracy and guide efficient irrigation planning. A novel framework integrating physics-inspired metaheuristic optimization with ensemble ML was developed to enhance predictive performance and guide precision irrigation planning. eXtreme Gradient Boosting (XGB) was optimized using four metaheuristic algorithms Quantum-Behaved Avian Navigation Optimizer (QANO), Photon Search Algorithm (PSA), Nuclear Reaction Optimization (NRO), and Kepler Optimization Algorithm (KOA), demonstrating superior accuracy over conventional models. The irrigation suitability models for drip, flood, and sprinkler systems achieved AUC values above 90%. Spatial variability and reliability were analyzed using the Coefficient of Variation (CV), identifying uncertainty zones under both historical (1960–2000) and projected climate scenarios (SSP2-4.5 and SSP5-8.5, 2015–2030) from CMIP6 ensemble data. Explainable AI (XAI) through SHAP analysis revealed actual evapotranspiration (AET), temperature extremes, vapor pressure deficit (VPD), soil moisture, and elevation as dominant factors. The findings demonstrate the potential of climate-smart irrigation planning, precision water management, and adaptive modeling approaches to minimize water use, enhance agricultural resilience, and strengthen Morocco’s transition toward sustainable and resource-efficient agriculture. Despite high model performance, future work should integrate finer-scale soil and socio-economic data to enhance local applicability and policy relevance.