Delineating the Controlling Factors of Hydraulic Fracturing-Induced Seismicity in the Northern Montney Play, Northeastern British Columbia, Canada, With Machine Learning

Abstract

Abstract Recent studies confirm that the distribution of injection-induced earthquakes (IIE) can be related to both natural (e.g., tectonic, geological, and hydrological) settings and operational details. However, the relative importance of operational factors with respect to the natural ones has not been fully understood for the western Canada sedimentary basin. In this study, we train the eXtreme Gradient Boosting (XGBoost) machine-learning algorithm to comprehensively evaluate six geological and seven industrial operational factors suspected to be correlated with the distribution of IIE in the northern Montney play (NMP), British Columbia. We then derive the Shapley Additive Explanations values to quantitatively interpret the outputs from XGBoost. Our results reveal that operational and geological factors have comparable contributions to the IIE distribution. The top four features that contribute most to the seismicity pattern are horizontal distance to the Cordilleran deformation front, cumulative injected volume, shut-in pressure and vertical distance to the Debolt formation (with respect to the hydraulic fracturing [HF] depth). Features with secondary influence are the thickness of the Montney formation, breakdown pressure, cumulative fault length per unit area, and vertical distance to the basement (with respect to the HF depth). Other remaining features (e.g., the average treating pressure and injection rate) appear the least related. Our results provide critical information to establishing a comprehensive susceptibility model that includes key geological and operational factors affecting the IIE distribution in the NMP area.