Detecting anomalous methane in groundwater in shale gas production areas using big data

The pubic is concerned about the environmental impact of the rapid development of shale gas in the U.S. The leakage of methane out of unconventional wells has drawn most attention because methane is a potential explosion hazard and contributes to global warming.

Many environmental forensic studies using geochemical tracers (e.g., organic compounds, water chemistry, and carbon and noble gas isotopes) have been conducted to investigate the source and transport pathway of dissolved methane in groundwater. These studies have reached conflicting conclusions partly because of the variable dataset size (small vs. large). In addition, the natural variability of these geochemical tracers can make assessments of the provenance of dissolved methane uncertain. As more and more pre-drill water chemistry data are released to the public, some studies (e.g., [1]) that use large datasets of water chemistry have shed light on the correlation between dissolved methane and natural/geologic/anthropogenic features, e.g., the proximity to anticlines, faults or shale gas wells. However, the discrimination between methane leaking from oil and gas wells and naturally-occurring methane remains complicated.

We have developed a geochemical fingerprinting tool from published data ( 11,000 water samples) on groundwater quality in northeastern Pennsylvania to distinguish between methane from natural and anthropogenic sources. We apply machine learning models to predict methane concentrations based on other water chemistry parameters and to subsequently study the importance of features in predicting methane concentration. In turn, this has helped us find helpful chemical analyte(s) to distinguish between natural and anomalous methane dissolved in groundwater. The learned models have also been tested on compiled data from many previously published studies in the U.S.

[1] Wen et al. (2018) Environmental Science & Technology 52 (12), 7149-7159.