G-larmS: An Infrastructure for Geodetic Earthquake Early Warning, applied to Northern California

Wednesday, 17 December 2014: 11:35 AM
Ronni Grapenthin1,2, Ingrid A Johanson1 and Richard M Allen1, (1)University of California Berkeley, Berkeley, CA, United States, (2)New Mexico Institute of Mining and Technology, Socorro, NM, United States
Integrating geodetic data into seismic earthquake early warning (EEW) is critical for accurately resolving magnitude and finite fault dimensions in the very largest earthquakes (M>7). We have developed G-larmS, the Geodetic alarm System, as part of our efforts to incorporate geodetic data into EEW for Northern California. G-larmS is an extensible geodetic EEW infrastructure that analyzes positioning time series from real-time GPS processors, such as TrackRT or RTNET. It is currently running in an operational mode at the Berkeley Seismological Laboratory (BSL) where we use TrackRT to produce high sample rate displacement time series for 62 GPS stations in the greater San Francisco Bay Area with 3-4 second latency. We employ a fully triangulated network scheme, which provides resiliency against an outage or telemetry loss at any individual station, for a total of 165 basestation-rover pairs.

G-larmS is tightly integrated into seismic alarm systems (CISN ShakeAlert, ElarmS) as it uses their P-wave detection alarms to trigger its own processing and sends warning messages back to the ShakeAlert decision module. Once triggered, G-larmS estimates the static offset at each station pair and inputs these into an inversion for fault slip, which is updated once per second. The software architecture and clear interface definitions of this Python implementation enable straightforward extensibility and exchange of specific algorithms that operate in the individual modules. For example, multiple modeling instances can be called in parallel, each of which applying a different strategy to infer fault and magnitude information (e.g., pre-defined fault planes, full grid search, least squares inversion, etc.). This design enables, for example, quick tests, expansion and algorithm comparisons.

Here, we present the setup and report results of the first months of operation in Northern California. This includes analysis of system latencies, noise, and G-larmS’ response to actual events. We also test how differential positions over relatively short baselines (like those produced at the BSL) compare to absolute positions in the case of a very large earthquake. We perform this analysis using data from the 2011 Mw 9.0 Tohoku earthquake, add randomly selected real-time noise, and invert for slip along the subduction zone interface.