GROUNDWATER QUALITY AND SUSTAINABILITY ANALYSIS USING MODELING TECHNIQUES: CASE STUDY OF MAHARASHTRA STATE, INDIA

Abstract

This research presents a novel machine learning framework to assess groundwater quality across Maharashtra State using data from 680 monitoring stations. Each station's groundwater quality index (GWQI) was evaluated and categorised into five classes: unfit for drinking, poor, moderate, good, and Excellent. These classifications were based on established thresholds. The study identified several key factors influencing groundwater quality, including maximum and minimum temperatures, precipitation, population density, soil moisture, groundwater table, and agricultural Inputs such as manure production and fertilisers (N, P and K). These input variables were integrated and analysed within a Geographic Information System (GIS) environment. To analyse the relationships between these input variables and the GWQI, we used five different machine learning classifiers: Random Forest (RF), Support Vector Machine (SVM), XG-Boost, Artificial Neural Networks (ANN), and K-Nearest Neighbor (KNN). The effectiveness of these algorithms was assessed using the Receiver Operating Characteristic (ROC) curve along with other statistical metrics such as overall accuracy, precision, recall, and F-1 score. The Random Forest model proved the most effective, demonstrating high overall accuracy (0.92) and an ROC score of 0.95. Consequently, this model was used for current and future predictive groundwater quality analyses. Our results from 2016 indicated a predominance of the poor GWQI class covering 21.5% of the area, followed closely by the moderate class at 21.6%. The Good, Unfit for Drinking, and Excellent classes comprised 19.4%, 17.6%, and 9.9%, respectively. Looking forward to the year 2040, the predictive analysis suggests a significant increase in areas classified as poor (35.7%) and unfit for drinking (30.4%), with decreases observed in the moderate, Good, and Excellent classes (24.7%), (6.2%) and (2.9%), respectively. The analysis of input variables showed that maximum temperature, precipitation, soil moisture, and groundwater table levels are the key factors of groundwater quality in Maharashtra. This study introduces a cost-effective, machine learning-based method for modelling groundwater quality, providing essential insights for future groundwater management and planning strategies.