Advanced Groundwater Forecasting: AI-driven Optimization for Accurate Resource Mapping

Groundwater is a critical resource for sustaining life in semi-arid regions with limited surface water availability. Accurate groundwater potential mapping (GPM) is crucial for sustainable water resource management; however, existing machine learning (ML) approaches frequently encounter challenges in predictive accuracy and optimization efficiency. This study introduces an innovative hybrid modeling framework—Random Forest combined with the Whale Fruit-Fly Optimization Algorithm (RF-WFOA)—that, to the best of our knowledge, has not been previously applied to GPM. RF-WFOA integrates the global search capability of the Whale Optimization Algorithm (WOA) with the rapid convergence of the Fruit Fly Optimization Algorithm (FOA), thereby overcoming local optima and enhancing the robustness of hyperparameter tuning. The approach was applied to Andika County, Iran, using an integrated hydrological, topographical, and geological dataset. Comparative experiments with RF-FOA and RF-WOA demonstrated that RF-WFOA achieved the highest predictive accuracy (AUC = 0.982) and superior computational efficiency, outperforming both RF-FOA (AUC = 0.966) and RF-WOA (AUC = 0.972), with all models showing statistically significant differences (p < 0.0001). Feature importance analysis revealed elevation as the most influential factor (importance score = 0.49), followed by rainfall (0.15) and distance to fault (0.057). The enhanced robustness and efficiency of RF-WFOA indicate its suitability for deployment in other water-scarce regions. The resulting groundwater potential maps can guide policymakers and local water authorities in prioritizing groundwater exploration zones, optimizing resource allocation, and implementing targeted conservation strategies to ensure long-term water security. © 2025 Elsevier B.V., All rights reserved.