Validating a Discriminant Analysis Model Used to Distinguish Salinity Contamination from Deicers vs Produced Water

Baseline water quality measurements taken prior to unconventional energy development have been advocated as a method to better understand if groundwater quality has changed following resource extraction. However, direct comparisons of pre- and post-development data can produce equivocal results due in part to other potential sources of groundwater contamination. In previous work, we developed a model that uses linear discriminant analysis (LDA) to identify sources of salinity in groundwater samples based on their geochemical fingerprints. Our approach has the advantage of utilizing solute concentrations commonly available in existing databases, rather than novel tracers typically not measured in routine water chemistry studies. While our model appeared successful, there was no clear way to determine the accuracy because the saline groundwater samples we evaluated did not have known sources of contamination. Following up on that work, we have applied a modified version of the salinity-fingerprinting model to a new dataset of groundwater with known sources of contamination compiled from two studies of groundwater quality in Illinois: Panno et al., Illinois State Geol. Survey, Open File Series 2005-1 and Hwang et al. Environ. & Eng. Geosci., 11: 75-90 (2015). By predicting the source of salinity in groundwater samples for which sources of contamination are known, we were able to validate model predictive-accuracy. Results show high classification accuracy for groundwater samples impacted by formation water and road salt, with diminishing success for other contamination sources. Posterior probabilities, a statistic inherent to LDA, provides a proxy for prediction confidence in cases where the source of contamination is unclear. The results indicate that this model could be used as an additional tool in baseline water quality assessments to identify any major changes in the source of groundwater salinity.