A simple and effective approach to investigate the dominant contaminant sources and accuracy in water quality estimation through Monte Carlo simulation, Gaussian Mixture Models (GMMs), and GIS machine learning methods

Abstract

Water quality variables, linked with natural and human activities, significantly impact groundwater quality. Assessing these links and impacts is a complex method, but it can be simplified using innovative approaches like Monte Carlo simulation and Gaussian Mixture Models (GMMs). This approach was applied by examining data on water quality parameters for the years 2009 and 2021 in varied hydrogeology in West Bengal, India. Geospatial distribution of water quality was analyzed using 479 samples in 2009 and 734 samples in 2021 through a recently suggested Water Pollution Index (WPI). The WPI uncovered a deterioration in the quality of water as the percentage of samples categorized as “excellent” decreased from 30.5 % to 28 %, while “polluted” samples increased from 44 % to 45 %. Statistical analyses, including analysis of variance (ANOVA) for yearly comparisons and Friedman tests for variable group comparisons, showed significant differences (p < 0.0001) in both analyses. Sensitivity analysis identified total iron as the predominant factor influencing groundwater quality in both years, along with significant contributions from fluoride, electrical conductivity (EC), total dissolved solids (TDS), magnesium, and nitrate. Uncertainty analysis revealed a shift in the WPI distribution centre from 0.45 to 0.9 in 2009 to 0.5–1.1 in 2021, with shorter probability tails suggesting more consistent water quality measurements in 2021 compared to 2009. Cluster-wise source apportionment for groundwater contaminants using machine learning-based Gaussian Mixture Models (GMMs) suggested that various factors influenced the groundwater quality in the study region. This study highlights the potentiality of applying the Monte Carlo simulation to simply find out the drivers of groundwater pollution as well as linking the GMMs with geospatial distribution, which could be a simple, innovative and best solution to take targeted actions to address the issue and manage water quality effectively.