Machine Learning‐Based Analysis of Geological Susceptibility to Induced Seismicity in the Montney Formation, Canada. Issue 22 (23rd November 2020)

• Record Type: Journal Article
• Authors: Wozniakowska, Paulina; Eaton, David W.
• Is Part Of: Geophysical research letters. Volume 47:Issue 22(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-11-23
• Subjects: induced seismicity -- seismogenic potential -- supervised machine learning; Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1029/2020GL089651
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: We analyze data from 6, 466 multistage horizontal hydraulic fracturing wells drilled into the Montney Formation over a large region in western Canada to evaluate the impact of geological, geomechanical, and tectonic characteristics on the distribution of hydraulic fracturing‐induced seismicity. Logistic regression was used to obtain a machine learning estimate of the seismogenic activation potential of each well. Our results fit the observed spatial variability, including an enigmatic change in seismicity at 120°W that does not correlate with any change in industrial activity. Feature importance analysis provides insight into data types that have the greatest impact on the results. Based on current data, seismogenic activation potential is most strongly influenced by depth of injection and distance of the well to the Cordilleran thrust belt. Plain Language Summary: A supervised machine learning model is applied within a zone of oil and gas development in western Canada to determine areas where hydraulic fracturing operations are more likely to induce earthquake activity. Geological characteristics are investigated to determine factors that exert the greatest influence on earthquake triggering during or within 3 months after hydraulic fracturing. Our results indicate that an enigmatic change in seismicity patterns at 120°W arises from a combination of parameters, including distance to the northern Canadian Rockies. These inferences may be useful for a general understanding of localization of induced seismicity. Key Points: A supervised learning method reveals seismogenic activation potential for a Canadian shale play in response to hydraulic fracturing The results match an enigmatic, abrupt change in the distribution of seismicity across a provincial boundary The most influential parameters are distance to the Cordilleran deformation front and injection depth