A Computational Framework for Basin-Scale Pond Recharge Estimation Using Multi-Source Geospatial Data

Water scarcity and groundwater depletion are increasing due to anthropogenic and climatic pressures. This study presents a computational framework for pond recharge estimation in the Ramganga Basin (RGB), India. At present, most ponds in the area are not utilized as Managed Aquifer Recharge (MAR) systems; however, increasing interest necessitates data-driven approaches, and the proposed framework enables systematic evaluation of MAR potential at the basin scale. A large-scale geospatial dataset of 7443 ponds was compiled by integrating field surveys, government records, and remote sensing data. Model validation was conducted using recharge observations from 23 monitoring stations in the Moradabad zone. The framework incorporates ten hydro-environmental predictors within a spatial modeling pipeline to identify high-potential recharge zones. Among the three machine learning models evaluated, Gradient Boosting showed the best predictive performance, with an Area Under the Curve (AUC) of 0.92. It also achieved consistent performance in pond recharge rate prediction, with R of 0.83 and lower RMSE and MAE compared to other models. Overall, machine learning approaches performed better than conventional statistical methods across the selected evaluation metrics. Spatially explicit recharge potential zones were delineated for the entire RGB, revealing clear variability across different geomorphological and hydrological settings. Because validation sites are concentrated within alluvial plain regions, the framework is considered most reliable under similar hydrogeological conditions, while additional validation is required for hilly regions. Overall, the proposed framework provides a reproducible and scalable approach for identifying recharge zones and assessing pond-based groundwater replenishment, supporting data-driven MAR planning in water-stressed basins.