Incorporation of real-time earthquake magnitudes estimated via peak ground displacement scaling in the ShakeAlert Earthquake Early Warning system

The United States earthquake early warning (EEW) system, ShakeAlert®, currently employs two algorithms based on seismic data alone to characterize the earthquake source, reporting the weighted average of their magnitude estimates. Nonsaturating magnitude estimates derived in real time from Global Navigation Satellite System (GNSS) data using peak ground displacement (PGD) scaling relationships offer complementary information with the potential to improve EEW reliability for large earthquakes. We have adapted a method that estimates magnitude from PGD (Crowell et al., 2016) for possible production use by ShakeAlert. To evaluate the potential contribution of the modified algorithm, we installed it on the ShakeAlert development system for real‐time operation and for retrospective analyses using a suite of GNSS data that we compiled. Because of the colored noise structure of typical real‐time GNSS positions, observed PGD values drift over time periods relevant to EEW. To mitigate this effect, we implemented logic within the modified algorithm to control when it issues initial and updated PGD‐derived magnitude estimates (⁠MPGD">MPGD">MPGDPGD⁠), and to quantify MPGD">MPGDPGD uncertainty for use in combining it with estimates from other ShakeAlert algorithms running in parallel. Our analysis suggests that, with these strategies, spuriously large MPGD">MPGDPGD will seldom be incorporated in ShakeAlert’s magnitude estimate. Retrospective analysis of data from moderate‐to‐great earthquakes demonstrates that the modified algorithm can contribute to better magnitude estimates for Mw>7.0">Mw>7.0w>7.0 events. GNSS station distribution throughout the ShakeAlert region limits how soon the modified algorithm can begin estimating magnitude in some locations. Furthermore, both the station density and the GNSS noise levels limit the minimum magnitude for which the modified algorithm is likely to contribute to the weighted average. This might be addressed by alternative GNSS processing strategies that reduce noise.