Understanding the Role of Geospatial Factors and Fluid Injection on Induced Seismicity in Oklahoma using Random Forests

Abstract:

The recent surge in induced seismicity in the central part of the United States since 2009 has led to a growing concern among researchers, environmentalists and policymakers to improve seismic hazard assessment. There is a growing need to analyze the contributing factors for increased seismicity and the hazards associated with such an increase. Big data on induced seismicity which have recently become openly available have enabled the possibility to use data-driven machine learning algorithms in order to identify associations between the activity of injection wells and the occurrence of induced seismicity. However, there is a knowledge gap in terms of understanding the most significant factors contributing to increased seismic activities. Our objective is to extract underlying geospatial and injection factors influencing induced earthquakes using earthquake catalog and fluid injection data in Oklahoma, USA. We use a Random Forest algorithm to study the influence of fluid injection wells between 2011 and 2016 on induced seismicity in Oklahoma and found 4 factors – relative basement depth, proximity of earthquake epicenters to injection wells, proximity to existing fault lines and injection volume to be the most significant variables in explaining seismicity within a 20km radius of an injection well. We then use the factors selected by Random Forest to forecast the average annual seismic moment release of injection wells at the county level for the year 2017 using a leave-one-year-out cross-validation technique. We visualized the spatial variation of the predicted seismic moment release rates and correlated it with the underlying individual factors to see how the seismicity varied spatially owing to relative basement depth, proximity to injection wells and cumulative fluid injection rate. (LA-UR-20-25236)