S33B-2764
The G-FAST Geodetic Earthquake Early Warning System: Operational Performance and Synthetic Testing

Wednesday, 16 December 2015
Poster Hall (Moscone South)
Brendan W Crowell1, David A Schmidt1, Paul Bodin1, John Emilio Vidale1, Timothy Ian Melbourne2 and Victor M Santillan3, (1)University of Washington, Seattle, WA, United States, (2)Central Washington University, Ellensburg, WA, United States, (3)Central Washington Univ, Ellensburg, WA, United States
Abstract:
The G-FAST (Geodetic First Approximation of Size and TIming) earthquake early warning module is part of a joint seismic and geodetic earthquake early warning system currently under development at the Pacific Northwest Seismic Network (PNSN). Our two-stage approach to earthquake early warning includes: (1) initial detection and characterization from PNSN strong-motion and broadband data with the ElarmS package within ShakeAlert, and then (2) modeling of GPS data from the Pacific Northwest Geodetic Array (PANGA). The two geodetic modeling modules are (1) a fast peak-ground-displacement magnitude and depth estimate and (2) a CMT-based finite fault inversion that utilizes coseismic offsets to compute earthquake extent, slip and magnitude. The seismic and geodetic source estimates are then combined in a decision module currently under development. In this presentation, we first report on the operational performance during the first several months that G-FAST has been live with respect to magnitude estimates, timing information, and stability. Secondly, we report on the performance of the G-FAST test system using simulated displacements from plausible Cascadian earthquake scenarios. The test system permits us to: (1) replay segments of actual seismic waveform data recorded from the PNSN and neighboring networks to investigate both earthquakes and noise conditions, and (2) broadcast synthetic data into the system to simulate signals we anticipate from earthquakes for which we have no actual ground motion recordings. The test system lets us also simulate various error conditions (latent and/or out-of-sequence data, telemetry drop-outs, etc.) in order to explore how best to mitigate them. For example, we show for a replay of the 2001 M6.8 Nisqually earthquake that telemetry drop-outs create the largest variability and biases in magnitude and depth estimates whereas latency only causes some variability towards the beginning of the recordings before quickly stabilizing towards a single solution. Each modeling module has different limitations imposed by latency, network architecture, noise, drop-outs, and earthquake size/location, and we finish by discussing how to best improve the robustness and timing of the warnings from G-FAST and challenges associated with integration into the ShakeAlert system.