Assimilating Multicycle Rupture Simulations into Probabilistic Forecasting Models
Abstract
A general problem in earthquake forecasting is how to assimilate deterministic physical simulations into probabilistic forecasting models. Here we focus on combining long earthquake catalogs (~ 106 yr) from the multi-cycle Rate-State Quake Simulator (RSQSim) of Dieterich & Richards-Dinger (2010) with the time-independent Uniform California Earthquake Rupture Forecast Version 3 (UCERF3), of Field et al. (2014). Rupture statistics are compared by establishing a mapping of RSQSim ruptures into the UCERF3 rupture set, which we optimize to preserve the seismic moment. Our Bayesian approach uses the UCERF3 logic tree to construct a prior distribution, which we update using an RSQSim catalog, modeled as a time-independent Poisson process. We consider two forms of the multivariate prior, a gamma distribution and a log-normal distribution. Because the gamma distribution is a conjugate prior to the Poisson likelihood function, it is computationally simpler to implement, but the log-normal prior is a better choice because it can properly handle correlations in the rupture rates. We assess the efficacy of the updating schemes by logarithmic scoring of the forecasts against independent RSQSim catalogs. We can then employ the calibrated time-independent forecasts as prior distributions in the development of time-dependent forecasts that are consistent with the RSQSim event statistics.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMS040...02V
- Keywords:
-
- 7299 General or miscellaneous;
- SEISMOLOGY