Using unsupervised machine learning and positive matrix factorization models to drive groundwater chemistry and associated health risks in a coal − mining rural region

Abstract

Identifying and quantifying the geogenic sources and anthropogenic sources of heavy metals and nitrate from groundwater is essential for securing the groundwater environment in mining rural areas. However, the integrated approaches for clarifying groundwater chemistry, specific pollutant sources, and associated probabilistic health risks in the mining rural area have yet to be raised. In this study, unsupervised machine learning, compositional data analysis based on principal component analysis, positive matrix factorization, and Monte-Carlo simulating health risks were used to quantify pollutant sources and groundwater drinking suitability in a coal − mining region of northeastern Chongqing, SW China. Three groups of groundwater samples were recognized by unsupervised machine learning. Group A was Ca − HCO₃ type. Group B was dominated by Ca − SO₄ and mixed Ca − Na − HCO₃ types. Group C consisted of Ca − HCO₃ type and Ca − SO₄ types. Group A and Group C were controlled by carbonate rocks and silicate dissolution, while Group B was dominated by the dissolution of silicate rocks, pyrite and oxides of heavy metals. Positive cation exchange was identified in all types of groundwater samples. Agricultural activity and mining sewage discharge was the primary sources of nitrate contamination. Compositional data analysis (CoDa) based on principal component analysis (PCA) and positive matrix factorization (PMF) model identified three primary hydrochemical processes and five factors for all hydrochemical components, respectively. CoDa-PCA corroborated the former analysis of hydrochemical diagram, mineral saturation index. According to the PMF analysis for all hydrochemical components, natural background levels (NBLs) and the PMF (five factors) for heavy metals indicated the concentrations of Fe and Mn originated from the dissolution of Fe and Mn oxides in red beds (25.91 %). The concentrations of Co, Ni, Ba, Zn, Cu, and Hg were derived from the dissolution of oxides (22.35 %), barite (17.87 %), sphalerite (17.85 %), chalcopyrite and cinnabar (16.02 %). The combined weighted water quality index (CWQI) and heavy metal pollution index (HPI) values of all groundwater samples satisfied the drinking permission limit, revealing the groundwater was suitable for drinking purposes in the study area. The hazard index (HI) values depicted that there was approximately a 6.01 % probability of groundwater posed health risks above the acceptable limit (>1) to children. The most sensitive factors to human health risks were exposure frequency to contaminated water and NO₃⁻ concentration. Our study is expected to provide a reliable and robust basis for groundwater sustainable management in mining rural regions.