Integrating genetic algorithms and machine learning for spatiotemporal groundwater potential zoning in fractured aquifers

Abstract

Study region

Groundwater overexploitation in Jharkhand’s fractured hard-rock aquifers threatens sustainability amid rising domestic, agricultural, and urban demands.

Study focus

This study develops an integrated framework that combines Genetic Algorithm (GA)-optimized clustering, Random Forest (RF) regression, and Gradient Boosting (GB) classification to map Groundwater Potential Zones (GWPZs) in the Jharkhand state of India (2013–2023) using 103 monitoring wells and multiple hydrogeological, topographic, and remote-sensing variables. GA was applied to optimize hydrostratigraphic clustering. The Mann-Kendall (MK) test assessed temporal groundwater trends, the RF regression predicted groundwater depth at unmonitored sites, and the GB classification was implemented for spatial mapping. Model interpretability was boosted using Local Interpretable Model-Agnostic Explanations (LIME).

New hydrological insights for the region

The framework identified three GWPZs (high, medium, and low), validated by strong clustering indices (Silhouette = 0.90, Dunn = 0.94). MK analysis revealed significant groundwater depletion across all clusters (Z = -2.66 to −1.47, p < 0.05). RF regression achieved high predictive accuracy (R² ≈ 0.91, WI = 0.89, PBIAS = 0.25), highlighting curvature and lineament proximity as dominant factors. GB classification yielded an F1-score of 95.56 %. Spatially, high-potential zones were concentrated in West Singhbhum, East Singhbhum, and Gumla, while Giridih, Pakur, and Garhwa exhibited low potential with aquifer depletion. These findings provide scientific support for Jharkhand’s 2025 Groundwater Act and demonstrate the transferability of the framework to other hard-rock and data-scarce aquifers like the Brazilian Shield and African cratons.