Leveraging Machine Learning for the Identification of Artificial Groundwater Recharge Potential Zones in Shivamogga District.

Groundwater depletion has become a major challenge in India due to extensive extraction and limited natural recharge. Identifying suitable locations for artificial groundwater recharge (AGR) is essential for ensuring long-term water sustainability. This study presents a hybrid machine learning framework that integrates K-Means clustering, Convolutional Neural Networks (CNN), and XGBoost to delineate AGR potential zones. The method combines unsupervised spatial zoning, deep feature extraction, and ensemble classification to capture non-linear interactions among hydrogeological, topographic, and climatic factors. Using nine thematic layers, including rainfall, geology, geomorphology, slope, and drainage density, the framework was applied to the Shivamogga district of Karnataka, India. The proposed model achieved an overall accuracy of 99.67%, outperforming conventional and two-stage hybrid approaches. High recharge potential was observed in lateritic and valley regions with favorable infiltration characteristics, while low-potential zones were found in steep and less permeable terrains. The results demonstrate the robustness of multi-stage integration for reliable groundwater recharge mapping and provide a scalable approach for sustainable water resource management in data-rich regions.