Poster #262, Earthquake Forecasting and Predictability (EFP)

Pseudo-prospective testing of five year time-independent California earthquake forecasts with inlabru

Kirsty Bayliss, Mark Naylor, Farnaz Kamranzad, & Ian Main
Poster Image: 

Poster Presentation

2021 SCEC Annual Meeting, Poster #262, SCEC Contribution #11511 VIEW PDF
Probabilistic forecasts estimate the likelihood of future seismicity in some specified time-space-magnitude window, but a forecast can only truly be considered meaningful if it demonstrates a degree of proficiency at describing future seismicity. Log-Gaussian Cox processes with a spatially varying, random intensity field may be used to flexibly model the spatial pattern formed by the locations of earthquakes. Using the Bayesian inlabru approach we fit models that use different combinations of spatial covariates that might help describe observed seismicity, including fault location, slip rate and strain rate. We use these spatial models to develop time-independent earthquake forecasts for Cal...ifornia using both full and declustered earthquake catalogs. We then test these models in a pseudo-prospective way by comparing with observed events over three contiguous 5-year time periods, using forecast tests developed by the Collaboratory for the Study of Earthquake Predictability (CSEP) and implemented in the PyCSEP package (Savran et al, 2021).
We compare the inlabru seismicity forecasts with previous results for the California testing region and explore the differences in forecast performance arising from different input data sets and the use of grid-based or simulated catalog-based tests. We demonstrate that the inlabru models perform well overall in pseudo-prospective testing, especially when using the simulated catalog-based tests that make use of full model posteriors.