A Stress-Similarity Aftershock Forecast Model

Date: 4/24/2019

Time: 03:00 PM

Room: Pike

Statistical models of earthquake clustering have performed well in prospective forecast tests, despite containing no physical mechanisms for the observed clustering. Including physics-based model components may improve the accuracy of the statistical forecasts. In particular, static stress changes have been shown to correlate with aftershock occurrence, and in some cases the addition of static Coulomb stress changes can improve upon purely statistical forecasts (e.g. Cattania et al., SRL 2018; Segou et al, JGR 2013; Bach and Hainzl, JGR 2012) but this has not been uniformly successful (e.g. Woessner et al., JGR 2011). I test a different approach to the inclusion of static stress changes, based on the observation that aftershocks tend to occur in locations where the static stress change is similar to the background stress. An example of this is the onshore normal faulting regions that were activated by the M9 Tohoku earthquake because of the extensional stress changes (Yoshida et al., PAGEOPH 2018). For four M≥6.7 Southern California mainshocks, I show that aftershocks are significantly more likely than preshocks to occur in regions where the static stress change tensor and the background stress tensor are more similar, as measured by the tensor dot product. I use this observation to develop a stress-similarity spatial kernel for the ETAS model, which depends on both distance from the mainshock fault plane and the similarity of the static stress change and the background stress. For the Southern California M≥6.7 mainshocks, the improvement in the likelihood of the best-fitting ETAS parameters usually justifies the additional model parameters. The largest improvement over an ETAS model with a radially-uniform spatial kernel is early in the aftershock sequence, and diminishes as more secondary triggering occurs. Pseudo-prospective tests will be performed to determine if the stress-similarity model can outperform a standard ETAS model in CSEP-style tests.

Presenting Author: Jeanne Hardebeck

Authors