Groundwater recharge in semi-arid mountainous environments is controlled by complex climatic, geological, topographical and anthropogenic interactions. In the Moroccan Atlas Mountains, snowmelt plays a crucial role in sustaining surface and groundwater resources, which are increasingly threatened by agricultural expansion and climate variability. In this context, the Midelt Aquifer within the Ansegmir River Watershed (ARW) represents a strategic water resource for irrigated agriculture, particularly apple orchards that heavily depend on groundwater extraction. This study aims to map the groundwater recharge potential of the Midelt aquifer (241 km²), in response to the expansion of orchards within the Oued Ansegmir watershed. The study integrates piezometric data, remote sensing (RS), GIS-based algorithms and machine learning (ML) approaches. Three modeling techniques were compared: Frequency Ratio (FR), Random Forest (RF) and Extreme Gradient Boosting (XGBoost). The models were calibrated using 271 field measured piezometric observations used directly as the target variable (70:30), and 16 conditioning factors retained after a Pearson correlation and variance inflation factor (VIF) analyses. Performance was evaluated on an independent test set (n = 82) using ten complementary metrics, ROC curves, confusion matrices, radar plots, an inter-model agreement matrix, and pixel by pixel spatial consensus analysis. XGBoost outperformed the other two models (AUC = 0.769, F1 = 0.590, Kappa = 0.356) and demonstrated remarkable consistency in the spatial validation, particularly in identifying areas with low recharge potential. Therefore, it was selected to produce the final groundwater recharge potential map. These results provide a scientific basis for sustainable groundwater management and the prioritization of future drilling activities within the ARW.