FINGERPRINTING SOURCES OF SALINITY TO AQUIFERS OVERLYING SHALE PLAYS USING PUBLICALLY-AVAILABLE BACKGROUND WATER QUALITY DATA AND MULTIVARIATE STATISTICAL METHODS

There are many sources of salinity in shale basins that complicate fingerprinting of shallow groundwater impacted by basin brines. Distinct tracers of basin brines have been suggested (such as strontium isotopes) and these tracers have the advantage of quantifying brine mixing at very low levels and fingerprinting brines from distinct formations, such as the Marcellus. But, there is additional need for complimentary geochemical fingerprinting tools that make use of more commonly measured solutes, such as those commonly reported in publically-available background water quality surveys. Given that mixing of pristine shallow groundwater and different sources of salinity (e.g. road salt, basin brines, septic effluent) can generate similar mixing relationships for individual solutes or solute ratios, multivariate statistical methods may provide an effective way to combine information available from multiple solutes simultaneously to reduce error in fingerprinting sources. Here, we use linear discriminant analysis (LDA) to classify unknown high salinity (>20 mg/L Cl) shallow groundwater samples as most likely impacted by Appalachian Basin brines, road salt runoff, septic effluent, or animal waste, based on the major and minor ion chemistry. The LDA model is developed from the chemistry of synthetic training data, which is created by theoretical two-component mixing of low salinity (<20 mg/L Cl) groundwater with various saline end-members. We explore the effectiveness of the LDA model when developed from different combinations of solutes commonly reported in publically-available water quality databases (e.g. Ca, Mg, Na, K, Cl, Sr, Br). We test the model on samples reported in the literature to be impacted by different sources of salinity. The most effective LDA classification models include halogens (Cl, Br, I), but models developed without Br or I still have correct classification rates of over 75% for samples impacted by brines. Unknown samples classified as impacted by brines and septic effluent have greater methane and nitrate concentrations, respectively, suggesting classifications are correct.