Which way rupture will propagate at the junction of the San Andreas and Garlock fault? – insights from the intersection of earthquake simulation and machine learning

Abhijit Ghosh, Shankho Niyogi, Evan O. Marschall, Roby Douilly, & David D. Oglesby

Published September 10, 2023, SCEC Contribution #13270, 2023 SCEC Annual Meeting Poster #091

It is challenging to anticipate the direction of rupture propagation in a fault junction, as it depends on multiple factors that are often difficult to estimate accurately. A machine learning approach may provide important new insights as it can handle many parameters and outcome scenarios efficiently. The challenge is to come up with a set of reasonable parameters and corresponding outcomes that is large enough for a ‘machine’ to ‘learn’. Here, we have selected the junction of the San Andreas and Garlock fault system for this study. We use Rate-State earthquake simulator (RSQSim) for simulating a reasonably large earthquake catalog, and generating corresponding critical parameters such as normal and shear stress, strength excess, distance and depth of rupture etc. Input parameters for the simulation is taken from the Uniform California Earthquake Rupture Forecast, Version 3 (UCERF3), to obtain realistic results. The simulation results are then used for the next step involving a machine learning approach. We apply a gradient boosting and random forest algorithm for classification of rupture at the junction of the San Andreas and Garlock fault. We show that this approach can produce reasonable testing accuracy (>65%) for certain set of parameters and scenarios. Moreover, it highlights relatively important parameters, or combination of them, that may determine if the rupture will break through the junction, and if so, the direction of propagation. This study demonstrates that a machine learning approach driven by physics-based earthquake simulation may serve as a promising tool to investigate rupture propagation at a junction of major fault systems with complex geometry.

Key Words
San Andreas fault, Garlock fault, machine learning, earthquake simulation, rupture

Citation
Ghosh, A., Niyogi, S., Marschall, E. O., Douilly, R., & Oglesby, D. D. (2023, 09). Which way rupture will propagate at the junction of the San Andreas and Garlock fault? – insights from the intersection of earthquake simulation and machine learning. Poster Presentation at 2023 SCEC Annual Meeting.

Related Projects & Working Groups
San Andreas Fault System (SAFS)