Integration of geospatial techniques and artificial neural networks for groundwater potential zonation in India

This study delineates Groundwater Potential Zones (GWPZs) in the Mahi River Basin (MRB), where increasing groundwater stress is driven by urbanization, intensive agriculture, and limited recharge. Ten predictive factors: geology, geomorphology, slope, lineament density, drainage density, land use/land cover (LULC), Normalized Difference Vegetation Index (NDVI), Soil Adjusted Vegetation Index (SAVI), Normalized Difference Water Index (NDWI), and Toposoil Grain Size Index (TGSI), were derived from multisource satellite imagery (Landsat 8 OLI/TIRS, Sentinel-2, SRTM DEM) and field observations. An Artificial Neural Network (ANN) model assigned nonlinear contribution weights to these inputs and generated a GWPZ map with five classes: Very Good (5.24%), Good (16.35%), Moderate (29.89%), Poor (30.88%), and Very Poor (17.64%). Model performance was validated quantitatively using the Area Under the ROC Curve (AUC = 0.850). Additional validation using 350 well locations showed strong spatial correspondence between predicted zones and groundwater depth. The Kruskal-Wallis H test (H = 42.87, p < 0.001) confirmed significant hydrogeological differences among ANN-predicted classes, while Moran's I (0.037, p = 0.46) demonstrated spatial independence of prediction residuals. More than 62% of the basin falls within Moderate to Very Poor zones, emphasizing the need for strategic groundwater recharge planning. The integration of ANN with remote sensing and GIS offers a robust, data-driven model for sustainable groundwater resource management in semi-arid and data-poor regions.