Model-based clustering of hydrochemical data to demarcate natural versus human impacts on bedrock groundwater quality in rural areas, South Korea

Abstract

• This study aims to demarcate natural versus anthropogenic impacts on groundwater. • Clustering with a Gaussian mixture model was conducted for hydrochemical data. • Better selection of optimal variables and cluster numbers is crucial. • Bivariate normal mixture model using nitrate and fluoride was very robust. • This method can be used to evaluate contamination susceptibility and background levels. Improved evaluation of anthropogenic contamination is required to sustainably manage groundwater resources. In this study, we investigated the hydrochemical measurements of 18 parameters from a total of 102 bedrock groundwater samples from two representative rural areas in South Korea. We used model-based clustering with a normal (Gaussian) mixture model to differentiate the contributions of natural versus anthropogenic processes to the observed groundwater quality. Water samples varied in hydrochemistry from a Ca–Na–HCO 3 type to a Ca–HCO 3 –Cl type. The former type reflected derivation of major ions largely from water–rock interactions, while the latter type recorded varying degrees of anthropogenic contamination. Among the major dissolved ions, fluoride and nitrate were shown to be good indicators of the two types, respectively. The results of model-based clustering showed that the bivariate normal mixture model, which was based on the covariance of nitrate and fluoride, was more robust than multivariate analysis, and provided better discrimination between the anthropogenic and natural groundwater groups. Model-based clustering to measure the degree of cluster membership for each sample also showed a gradual change in groundwater chemistry due to mixing between the two water groups. This study provided an example of the successful application of model-based clustering to evaluate regional groundwater quality and demonstrated that better selection of the dimensional structure (i.e., selection of optimal variables and number of clusters) based on hydrochemistry was crucial in obtaining reasonable clustering results.