← Back to Sessions

Optimizing Real-Time Gnss-Based Magnitude Estimation for Shakealert

Description: 

To improve alerting for M7+ earthquakes, we adapted a magnitude estimation algorithm (“GFAST”, Crowell et al., SRL, 2016) based on real-time, GNSS-measured peak ground displacement (PGD) for use in the ShakeAlert EEW system for the U.S. West Coast. As implemented for ShakeAlert, GFAST runs in parallel with two seismic algorithms. The magnitude estimates from all three are combined in a weighted average (M_AV). What is the optimal logic for determining which magnitude estimates are included in M_AV each time source parameters are updated during an earthquake? Murray et al. (BSSA, in review) presented a framework and preliminary parameters to suppress inclusion of (or “throttle”) GFAST magnitude estimates (M_PGD) that may be unreliable due to noise in the GNSS position estimates. Ahead of formal evaluation for inclusion in the production system, we are monitoring GFAST behavior on ShakeAlert servers that simulate the production environment under typical real-time conditions. This has highlighted aspects of the initial throttling criteria that warrant optimization. The initial criteria required 1) ≥3 GNSS stations to report PGD exceeding time-dependent threshold values determined empirically from analysis of GNSS position noise, and 2) M_PGD uncertainty ≤0.5 unit. We have since added two criteria: excluding position data with large uncertainties and only including M_PGD when M_AV already exceeds 6.0 using seismic data alone. We also re-estimated the PGD thresholds to obtain more conservative values that better eliminate suspect M_PGD and a finer temporal discretization to reduce abrupt changes in M_PGD throttling over time. In parallel we are exploring ways to reduce noise and steps in the GNSS position time series themselves and to better characterize their true uncertainties. Finally, we are considering adopting spatially variable PGD thresholds to account for geographic variation in station density, as well as refinements to the method for deriving M_PGD uncertainty.

Session: Earthquake Early Warning Optimization and Efficacy

Type: Oral

Date: 4/20/2023

Presentation Time: 04:45 PM (local time)

Presenting Author: Jessica R. Murray

Student Presenter: No

Invited Presentation: 

Authors