Abstract Earthquake early warning (EEW) programs, such as the U.S. ShakeAlert system, provide users with critical and time-sensitive earthquake information with seconds to tens of seconds of lead time prior to experiencing strong ground motions. ShakeAlert has been active on the U.S. West Coast since 2018, amassing a large dataset of detected earthquakes. Using information about these earthquakes and how they are observed at regional seismic stations, we can improve on established EEW algorithms. Here, we propose an update to the scaling relationship used in the Earthquake Point-Source Integrated Code (EPIC) algorithm to calculate real-time magnitudes. This update is based on an analysis of over 15,000 earthquakes detected through EPIC. Our reconfiguration calculates station magnitudes using one scaling relationship for stations located within 30 km of the epicenter and another scaling relationship for mid- to far-field stations. When tested against a large dataset of past events, our new scaling methodology reduces EPIC’s magnitude error from 0.41 to 0.27 magnitude units with additional improvements possible if improved epicentral location information is available. Applying the new scaling to the past event catalog reduces cases in which EPIC’s magnitude erroneously triggers an alert, without introducing many missed alerts due to underestimates. The time it takes to reach an alertable magnitude under the new scaling is also unchanged. In addition, because the new scaling accommodates the inclusion of S-wave data, rather than filters for their removal, no new delays related to processing are incurred in EPIC. This work is motivated by a goal to lower the current magnitude overestimate as a means to increase confidence in future alerts.
Williamson et al. (Thu,) studied this question.