Machine learning-based groundwater potential mapping and factor analysis in tropical lateritic terrains using...

Groundwater potential mapping is essential for sustainable water resource management, particularly in tropical lateritic terrains where communities depend heavily on groundwater for domestic and agricultural needs. This study delineates groundwater potential zones (GWPZs) in the Ithikkara River Basin, South Kerala, India, through an integrated geospatial and machine learning framework that combines self-organizing maps (SOM), K-means clustering, and random forest (RF) feature importance analysis. Eight hydro-environmental parameters-land use/land cover (LULC), geomorphology, geology, slope, relative relief, lineament density, drainage density, and mean depth to water table (MDTW)-were normalized and processed in a GIS environment. SOM was trained and optimally clustered into five groundwater potential classes, as supported by a Davies-Bouldin Index (DBI). The clusters were reclassified into very low, low, moderate, high, and very high groundwater potential. RF analysis identified LULC, geomorphology, and geology as the dominant controls on groundwater occurrence. Validation using observed well yield and water table depth confirmed strong agreement, with high-potential zones coinciding with high-yield wells and shallow aquifers in the southwestern and central alluvial-fractured zones. The results provide practical insights for groundwater exploration, artificial recharge planning, and sustainable extraction in hydrogeologically complex terrains of the Western Ghats. Beyond regional applications, the proposed methodology is scalable and interpretable, offering a transferable framework for groundwater potential mapping in other tropical river basins. Future research should incorporate long-term hydroclimatic variability, socio-economic drivers, and climate change projections to further strengthen sustainable groundwater management strategies.