Integrated GIS, Remote Sensing and Machine Learning Approach for Mapping and Predicting Groundwater Pollution Zones

Abstract

In semi-arid regions, the deterioration of groundwater quality due to industrialization and intensified agriculture is a serious problem. Because groundwater pollution is so common in Bathinda, Punjab, thorough mapping and forecasting are necessary for sustainable management. The current study uses GIS, remote sensing and machine learning to map and forecast groundwater contamination zones using water quality index (WQI) models. The overarching goal of the framework is to create an integrated geospatialmachine-learning mapping of the risk of groundwater pollution; the secondary goal is to use regressionbased model selection to determine the most significant hydrochemical markers on WQI. When more variables were included in the model, subset regression clearly demonstrated an increase in prediction accuracy. There was some success with single-variable models (Cl), but their predictive power was just 0.08 (R²). Generally speaking, adding more parameters significantly boosted explanation power. Eleven parameters (NO₃, SO₄, HCO₃, K, pH, TDS, TH, Ca, Mg, Na and Cl) in the best-fitting model produced nearly flawless predictions (R2 = 1, MSE ≈ 0) and reduced information criteria (AIC = 0, SBC = 0). This demonstrates how physicochemical characteristics, anions and cations all work together to predict groundwater quality. A great way to monitor Bathinda's water quality is to combine machine learning techniques with GIS tools.