@

초록

Groundwater is a vital source of freshwater in semi-arid regions, where water shortages are common. Accurately identifying areas with high groundwater potential is crucial for water management and agricultural sustainability. This study introduces an advanced approach using geospatial artificial intelligence (GeoAI) to predict groundwater-prone areas. We optimized the CatBoost algorithm with the Fruit Fly Optimization Algorithm (FOA) to improve prediction accuracy. Additionally, we used Shapley Additive exPlanations (SHAP) to make the model more interpretable and identify key factors influencing groundwater distribution. Our approach was applied in Kazerun and Kooh-Chenar counties, Iran, using 16 environmental factors related to groundwater potential. The results showed that the CatBoost-FOA model performed significantly better than the standalone CatBoost model, achieving higher accuracy (RMSE = 0.086, MAE = 0.068, AUC-ROC = 0.838 vs. RMSE = 0.219, MAE = 0.17, AUC-ROC = 0.741). SHAP analysis revealed that factors such as low elevation, high vegetation index (NDVI), high rainfall, and proximity to rivers significantly affect groundwater availability. Lower elevation areas facilitate surface water accumulation and infiltration, higher NDVI values indicate regions with sufficient soil moisture, more significant rainfall enhances groundwater recharge, and proximity to rivers improves hydraulic connectivity, supporting groundwater presence. This study provides a more accurate and interpretable method for groundwater potential mapping, helping improve water resource planning and management.

키워드