Discriminant analysis as a decision-making tool for geochemically fingerprinting sources of groundwater salinity

Concern over contamination of groundwater resources in areas impacted by anthropogenic activities has led to an increasing number of baseline groundwater quality surveys intended to provide context for interpreting water quality data. Flexible screening tools that can parse through these large, regional datasets to identify spatial or temporal changes in water quality are becoming more important to groundwater scientists. One such tool, developed from previous work by the authors, makes use of linear discriminant analysis (LDA) to identify the most probable source of chloride salinity in groundwater samples based on their geochemical fingerprints. Here, we applied the model to a dataset of shallow groundwater with known sources of contamination compiled from two studies of groundwater quality in Illinois: Panno et al. (2005) and Hwang et al. (2015). By predicting the source of salinity in groundwater samples for which the sources of contamination are known, we validated model prediction-accuracy. Results show high classification accuracy for groundwater samples impacted by basin brines (e.g. deep saline groundwater) and road salt (> 80%), with diminishing success for those impacted by organic sources of chloride, such as septic effluent and animal waste. Posterior probabilities, a statistic inherent to LDA, provide a proxy for prediction confidence that enables the model to be used for assessment and accountability measures, such as identifying parties responsible for contamination. LDA is complementary to fingerprinting using halogen ratios (e.g. Cl/Br) because it implicitly relies on halogen ratios for classification decisions while providing a clearer, more quantitative classification of contamination sources. Our model is ideal for regional assessment or initial screening of salinity sources in groundwater because it makes use of commonly measured solute concentrations in publicly available water quality databases.