Optimized groundwater quality evaluation using unsupervised machine learning, game theory and Monte-Carlo simulation.

Assessing groundwater quality is essential for achieving sustainable development goals worldwide. However, it is challenging to conduct hydrochemical analysis and water quality evaluation by traditional methods. To fill this gap, this study analyzed the hydrochemical processes, drinking and irrigation water quality, and associated health risks of 93 groundwater samples from the Sichuan Basin in SW China using advanced unsupervised machine learning, the Combined-Weights Water Quality index, and Monte-Carlo simulations. Groundwater samples were categorized into three types using the self-organizing map with the K-means method: Cluster-1 was Ca-HCO type, Cluster-2 was dominated by Ca-HCO, Na-HCO, and mixed Na-Ca-HCO types, Cluster-3 was Ca-Cl and Ca-Mg-Cl types. Ion ratio diagrams revealed that carbonate dissolution and silicate weathering primarily influenced the hydrochemical characteristics. Cluster-1 samples exhibited high NO contents from intensive agricultural activities. Cluster-2 samples with high Na contents were characterized by positive cation exchange, while Cluster-3 samples with elevated Ca and Mg contents were influenced by reverse cation exchange. Combined-Weights Water Quality Index indicated that 62.37% of total samples were suitable for drinking, predominantly located in the central part of the study area. Irrigation Water Quality Index revealed that 33.34% of total samples were suitable for irrigation, mainly in the northeastern region. NO concentration and electrical conductivity (EC) value were the main indicators with the highest sensitivity for drinking and irrigation suitability, respectively. Probabilistic health risk assessments suggested that a significant portion of the groundwater samples posed a health risk greater than 1 to children (63%) and adults (52%) by Monte-Carlo simulation. The high-risk areas (hazard index >4), primarily in the eastern region, are closely associated with nitrate distribution. Sensitivity analysis demonstrated that NO concentration is the primary indicator accounting for health risks. Reducing the application of nitrogen-based fertilizers on cultivated land is the most effective approach to improve drinking quality and mitigate the associated health risks to the population. This study's findings aim to produce a novel groundwater quality evaluation for promoting the sustainable management and utilization of groundwater resources.