Evaluating GFAST-PGD’s Contributions to ShakeAlert® System Performance for Simulated Cascadia Subduction Zone Earthquakes
Sydney Dybing
University of Washington
- Date & Time
- Location
- Hybrid In Person and Online-only seminar via Microsoft Teams
- Host
- Tara Nye
- Summary
Since 2024, Global Navigation Satellite Systems (GNSS) data have been able to contribute to alerts issued by the ShakeAlert® earthquake early warning system through the GFAST-PGD algorithm, which uses the peak ground displacement (PGD) from GNSS stations and their hypocentral distances to invert for earthquake magnitude (MPGD). GFAST-PGD improves the system’s ability to handle very large magnitude (>M8) earthquakes, such as Cascadia Subduction Zone (CSZ) events. However, the production ShakeAlert® system (with or without GFAST-PGD) has not yet been tested on real North American events of this size due to their rarity. Simulations of CSZ earthquakes offer an opportunity to evaluate how ShakeAlert® may perform in these scenarios, as synthetic waveforms can be run through the system as if they were real-time streaming data. We utilize 3D broadband ground motion simulations for 36 CSZ full-margin rupture scenarios (~M8.7-9.2) from Dunham et al. (2025), an expansion of the M9 Project (Frankel et al., 2018; Wirth et al., 2018), and focus our analysis on the performance and contributions of the GFAST-PGD algorithm, as well as how it interplays with the seismic algorithms (FinDer and EPIC). We identify scenarios for which GFAST-PGD has more accurate MPGD estimates and those for which it performs poorly, and examine the causes of these variations. For example, GFAST-PGD incorporates epicentral distance weighting in the MPGD inversion scheme, in which stations closer to the epicenter are weighted more strongly in the inversion. This can sometimes result in erroneously low MPGD estimates for scenarios in which the amount of slip on the megathrust is lower near the hypocenter than at farther distances, as GNSS stations at those distances would only weakly contribute to the MPGD estimate. We explore potential options for improving GFAST-PGD performance in scenarios like this, such as removal of the epicentral distance weighting scheme. We also investigate how the differences in the performance of GFAST-PGD for the varying scenarios affect the accuracy of the alerts issued by the full ShakeAlert® system.