PREDICTING HIGH BASELINE METHANE CONCENTRATIONS IN GROUNDWATER USING DATA-DRIVEN MODELS

Groundwater methane concentrations change spatially as a function of the hydrogeologic setting. High dissolved methane concentrations in groundwater in shale gas areas can be indicative of either baseline conditions or introduction of stray gas from nearby development of natural gas wells. Data-driven models that predict locations with naturally high methane concentrations could inform assessment of attribution in cases where the source of the elevated methane is not known. We trained decision tree models using well characteristics and water quality data for 360 domestic wells over the Marcellus Shale, in an area where high volume hydraulic fracturing is banned, to predict which wells have naturally high baseline methane concentrations in groundwater. We assessed the performance of decision tree models that use different thresholds for defining high methane concentrations, finding similar results amongst all but the highest threshold. We then evaluated model performance when differing types of information were used to train the decision tree. Our results show that hydrochemical parameters are good predictors of high methane in groundwater, but geospatial parameters are not. Sulfate concentration and specific conductance measurements can be used to effectively predict whether wells have ≥2 mg/L methane. Further, sulfate observations alone can be used to predict groundwater wells with ≥10 mg/L methane. Data-driven decision tree models trained on observational data can be used as a simple screening tool to identify domestic groundwater wells that are likely to contain high baseline methane concentrations in shale gas areas.