Why is it so difficult for physics-based aftershock forecasting models to outperform statistical models?
Video not found
Please check back later. Videos are usually posted within 24 hours.
USGS Earthquake Science Center
- Date & Time
- Yosemite Conference Room, Building 19, Moffett Field
- Grace Parker
I explore why physics-based models of aftershock triggering do not generally outperform statistical models in prospective testing. Pseudo-prospective tests on suites of synthetic aftershock sequences suggest that a significant factor is the level of spatial clustering of the direct aftershocks. The statistical ETAS model performs better the more clustered the aftershocks, while smooth spatial kernels that are typical of physical models do more poorly. Real aftershocks tend to be quite clustered, likely due to heterogeneity of the background physical conditions. This implies that to improve the performance of physical models, we must focus on understanding the interaction between mainshock effects and background physical conditions such as stress state, fault strength, and fluid pressure.
I investigate how the background stress in particular may affect aftershock locations. I propose the stress-similarity triggering model, which hypothesizes that more aftershocks occur in areas where the mainshock stress change is similar in orientation to the background stress. An example is the normal faulting regions of onshore Japan that were activated by extensional stress changes from the MW9.0 Tohoku-Oki earthquake. I apply this model to the 2019 Ridgecrest aftershock sequence, and find that the off-fault aftershocks cluster primarily in locations with high stress similarity. The aftershock density varies substantially inside the high-stress-similarity lobes, however, implying that other variable background conditions also play a role.