G-larmS: Integrating Real-Time GPS into Earthquake Early Warning

In an effort to improve earthquake parameter estimation in earthquake early warning for large earthquakes (such as moment magnitude and finite fault geometry), the BSL is working to integrate information from real-time GPS and now generates and archives real-time position estimates using data from 62 GPS stations in the greater San Francisco Bay Area. This includes 26 stations that are operated by the BSL as part of the Bay Area Regional Deformation (BARD) network, 8 that are operated by the USGS, and 29 stations operated by the Plate Boundary Observatory. Data from these sites are processed in a fully triangulated network scheme in which neighboring station pairs are processed with the software trackRT. Positioning time series are produced operationally for 172 station pairs; additional station pairs will be added as more real-time stations become available. G-larmS, the geodetic alarm system, sits on top of real-time GPS processors such as trackRT and analyzes real-time positioning time series, and determines and broadcasts static offsets and quality parameters from these. Following this, G-larmS derives fault and magnitude information from the static offsets and broadcasts these results as well. This prototype Python implementation is tightly integrated into seismic alarm systems (CISN ShakeAlert, ElarmS) as it uses their P-wave detection alarms to trigger its processing. Testing the results of real-time GPS for earthquake early warning (EEW) under realistic conditions, and for scenarios that are relevant to the San Francisco Bay Area's tectonic environment, is a major step toward having our work accepted for integration with an operational EEW system. While Northern California has many small earthquakes (i.e., Mw<4) that are used to validate the seismic system, it is only for very large earthquakes (Mw >6.5) that real-time GPS is expected to provide a significant contribution. This is because for larger events seismic systems need additional information to correctly estimate magnitude and finite fault extent and because real-time GPS suffers from a lower signal-to-noise ratio than post-processed data. Here, we follow two strategies to test G-larmS: (1) add simulated static offsets to (archived) real-time time series, and (2) replay archived data that contain static and dynamic motion due to a real event. We test the prototype system for the Bay Area using synthetic data for a Mw 6.9 Hayward Fault Scenario and on data for the 2010 Mw 7.2 El Mayor-Cucapah earthquake. We compare offset estimate time series and the evolution of event characteristics (magnitude, fault geometry, and slip) to model predictions and post-processed results. We find that the dynamic motion of the El-Mayor Cucapah event impacts the evolution of co-seismic offset estimates initially. However, the results converge quickly (about 1 S-wave wavelength) towards co-seismic offsets (within real-time noise) estimated from post-processed data.