Data Mining Techniques Applied to Marcellus Shale Gas Development in Northeastern Pennsylvania

Abstract

The rapid development in the use of high-volume hydraulic fracturing in Pennsylvania, U.S.A., since 2004 has led to public scrutiny of water quality. To investigate controversies about whether methane concentration in groundwater increases near gas wells, we studied an area with reported problems in Bradford County, Pennsylvania where a large publically available groundwater dataset has been published. Statistical investigation was facilitated by a new data mining technique for analysis of spatial patterns, which is especially useful in this region where drilling-related incidents are rare and sites of natural methane emission are common. Results show that over the whole study area with 1690 analyses, methane concentrations correlate with the distance to conventional, but not unconventional, wells. A smaller subarea (730 analyses) showed a very weak, but statistically significant correlation with unconventional wells that was masked in the larger dataset. Methane concentration also correlates with distance to faults and are most prominent in hotspots near faults. This is consistent with reports of natural emissions of gas in the region, where many east-west striking faults are characterized as low angle, northwest- or southeast-dipping thrust faults that generally outcrop at the surface along valleys. Much of the gas derives from the underlying shales, sandstones and siltstones of the Upper Devonian Catskill and Lock Haven Formations which have been historically explored for gas reservoirs and tends to migrate along the fault lines. After inspection of well completion reports, we have discovered 3 unconventional and 4 conventional wells may be uncemented or uncased at intermediate depths of intersection with faults. Three of these conventional wells are abandoned. These wells constitute a small percentage of the total wells in the study area, approximate 2% and 27% for unconventional and conventional wells, respectively. While we have shown that data mining techniques have implications for best practices, a similar technique may be applied to the potential identification of correlations between production data and subsurface geologic characteristics to draw inferences about variations that occur locally.

AAPG Datapages/Search and Discovery Article #90259 ©2016 AAPG Annual Convention and Exhibition, Calgary, Alberta, Canada, June 19-22, 2016