Machine learning-driven spatio-temporal zonation of groundwater depletion using hydro-climatic trend analysis in Purulia, West Bengal

Sustainable groundwater management in semi-arid regions requires a precise understanding of the complex interactions between climatic variables and subsurface storage. To address this, the primary objective of this study is to develop a robust machine learning framework for the spatio-temporal zonation of groundwater depletion across Purulia, India, by integrating 20 years (2005 to 2024) of multi-parametric hydro-climatic trends. In terms of methodology, the non-parametric Mann-Kendall test and Sen’s Slope estimator were utilized to quantify the velocity of change in Ground Water Levels (GWL), Actual Annual Rainfall (AAR), Evapo-Transpiration (ET), and Annual Soil Moisture (ASM). These dynamic trend signatures were subsequently clustered using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to delineate three distinct hydrological regimes: high depletion, moderate depletion, and potential zones, with model validity confirmed via one-way ANOVA (F = 17,796, p < 0.001) and silhouette analysis. The key results reveal that 14.71% (920.69 km²) of the study area falls within the High Depletion Zone, exhibiting an average decline of − 0.048 m/year. Crucially, a distinct decoupling between groundwater trends and precipitation was detected in these hotspots ( r = -0.15), where depletion persists despite rising rainfall (+ 2.71 mm/year), signalling dominant anthropogenic over-extraction. From an operational perspective, this research provides a statistically validated, high-resolution diagnostic tool enabling regional planners to implement targeted artificial recharge, prioritized water budgeting, and sustainable pumping regulations in critical aquifer stress zones.