GSA Annual Meeting in Denver, Colorado, USA - 2016

Paper No. 299-13
Presentation Time: 4:35 PM

DEVELOPMENT OF EMPIRICAL MODELS OF NATURAL METHANE OCCURRENCE IN SHALLOW GROUNDWATER OVERLYING THE MARCELLUS SHALE USING MACHINE LEARNING METHODS


LAUTZ, Laura K.1, CHRISTIAN, Kayla1, HOKE, Gregory D.1, SIEGEL, Donald I.1, LU, Zunli1 and KESSLER, J.D.2, (1)Department of Earth Sciences*, Syracuse University, 204 Heroy Geology Laboratory, Syracuse, NY 13244, (2)Department of Earth and Environmental Sciences, University of Rochester, Rochester, NY 14627, lklautz@syr.edu

Shale gas provides >45% of U.S. natural gas production, up from ~5% just a decade ago. There has been significant public and scientific attention given to the risk of groundwater contamination from gases released during unconventional shale gas production. But, assessing such contamination in the Marcellus region has yielded contradictory findings and is complicated by the fact that methane occurs naturally in shallow groundwater in areas overlying the Marcellus. High volume hydraulic fracturing (HVHF) is currently used to produce shale gas in all states overlying the Marcellus shale, with the exception of New York (NY), where HVHF is permanently banned. Given the similar geology, climate, and land use across areas underlain by the Marcellus, studies of methane in domestic wells in NY are representative of methane occurrence prior to HVHF in an area with ongoing conventional gas production.

We measured methane concentrations in 137 domestic wells in southern NY covering an area of 10,230 km2. For each well, we determined the topographic position (valley or upland), the geologic unit of water extraction, the chemical water type, and distances to the nearest fault, lineament, and active or other conventional gas well. Observed data from our study, similar to others, show significant correlations between methane concentrations and potential explanatory variables, which are linked to conceptual models of hydrogeologic processes driving methane occurrence (e.g. water type; landscape position). Our work further suggests empirical models predicting natural methane occurrence are most effective if they consider interaction among explanatory variables using machine learning methods, such as decision tree analysis, rather than simply assessing individual correlations between methane and isolated explanatory variables.

Combining methane and water quality data from this study and other prior studies in NY and Pennsylvania pre-HVHF (n=724), we found that although only 7.7% of domestic wells had >1 mg/L dissolved methane, 52% of valley wells producing Na-rich water had >1 mg/L dissolved methane. Our results suggest high methane concentrations in valley wells producing Na-rich water are likely to be naturally occurring, rather than the result of gas production by HVHF.