Probabilities of Earthquakes in the San Andreas Fault System: Estimations from RSQSim Simulations

Jacquelyn J. Gilchrist, Thomas H. Jordan, & Kevin R. Milner

Submitted July 31, 2018, SCEC Contribution #8237, 2018 SCEC Annual Meeting Poster #031

The RSQSim model of Dieterich & Richards-Dinger (2010) is now capable of simulating very long (> 10^6 yrs) catalogs of M ≥ 5 earthquakes using the same fault geometry and slip rates as UCERF3 (Field et al., 2014), the latest seismic hazard model for California. Time-independent hazard maps derived from RSQSim and UCERF3 using NGA ground motion prediction equations are surprisingly similar (Shaw et al., 2018). The potential for incorporating information from RSQSim simulations into probabilistic forecasting models motivates us to address how it might be done in practice. We consider the time-independent forecasting of events composed of M binary subevents, where the mth subevent is failure/no-failure of the mth fault section. The M marginal distributions can be well estimated by simulation, but a complete forecast must specify the frequencies of N = 2^M composite events, many of which are rare and poorly sampled, even at long simulation times. We consider a Bayesian scheme in which the prior on the forecast probabilities is a Dirichlet distribution of order N. The mean values are specified either by a prior forecast, such as UCERF3, or as the product of the M marginals (i.e., assuming subevents independence). The relative uncertainty in the prior is specified by a pseudo-count. We take the likelihood function for the event counts to be the multinomial distribution, which is conjugate to the Dirichlet prior; therefore, the posterior can be calculated analytically, allowing M to be fairly large. We apply this forecasting method to the southern San Andreas fault system, focusing on the probabilities of composite events involving faults that interact in the El Cajon Pass and San Gorgonio Pass regions. We are particularly interested in how these regions of fault complexity act as earthquake gates, governing the propagation of ruptures from one fault to another. The subevents in our models include ruptures on sections of the San Andreas, San Jacinto, and major thrust faults in the eastern Transverse Ranges. In the restricted world of the current RSQSim model, rupture propagation through the San Gorgonio Knot is extremely rare. Propagation between the San Andreas and San Jacinto is also rare and tends to be mediated by ruptures on nearby thrust faults.

Key Words
Earthquake Forecasting, Earthquake Simulators

Citation
Gilchrist, J. J., Jordan, T. H., & Milner, K. R. (2018, 07). Probabilities of Earthquakes in the San Andreas Fault System: Estimations from RSQSim Simulations. Poster Presentation at 2018 SCEC Annual Meeting.


Related Projects & Working Groups
Collaboratory for Interseismic Simulation and Modeling (CISM)