Application of machine learning, ambient noise and 2-D seismic interpretation to investigate induced seismicity in western Canada
Date
2024-04-30
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In western Canada, the development of unconventional oil and gas resources in low-permeability reservoirs has increased dramatically in the last two decades. Production of these resources typically requires hydraulicfracturing stimulation to increase permeability, as well as disposal of excess brines into permeable rock units. Both of these industrial processes can lead to induced seismicity (human-caused earthquakes). This thesis applies several approaches in two study areas in western Canada where induced earthquakes have occurred, mainly using existing open-source software. Machine learning (ML), a rapidly growing artificial intelligence approach, is combined with a probabilistic non-linear global search algorithm to construct a new seismicity catalog for the northern Montney play in eastern BC using data from both public and private seismograph stations. PhaseNet, a popular ML method that has been pre-trained using historical earthquake catalogs from around the world, is used for detection of P- and S-wave arrivals (phases). A Gaussian mixture model associator, GaMMA, is used to associate phases into events. Hypocentre locations are determined from the phase picks using the NonLinLoc algorithm. Geological context for the seismicity event distribution is supplied by existing published faults as well as new fault interpretations using 2-D seismic profiles. A growing cloud of seismicity, consistent with the Kaiser effect, is observed along a mapped fault near a disposal zone. This phenomenon is modelled using pore-pressure diffusion, numerically simulated here using code that I developed. In the Musreau Lake region ofwestern Alberta, a different type of analysis is applied, in part because a comprehensive seismicity catalog has been published using a similar ML approach to that described above. Ambient-noise interferometry, which uses cross-correlation of background noise between pairs of seismograph stations, was used to investigate temporal changes in subsurface velocity associated with pore-pressure diffusion and induced seismicity. The results using the NoisePy software show a good recovery of relative velocity changes between pair stations. Further, statistical analysis is required to construct robust observations containing all retrieved signals. Finally, areas for possible future work are proposed, including the use of satellite interferometric synthetic aperture radar (InSAR) as a complementary method for seismological studies of induced earthquakes and slow-slip events, as well as fully coupled physics-based simulation and inversion of induced seismicity processes.
Description
Keywords
Induced seismicity, Machine learning, Western canadian sedimentary basin, Ambient noise interferometry, Pore-pressure modeling
Citation
Rojas-Parra, J. (2024). Application of machine learning, ambient noise and 2-D seismic interpretation to investigate induced seismicity in western Canada (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.