S33B-2756
Testing and Development of the Onsite Earthquake Early Warning Algorithm to Reduce Event Uncertainties

Wednesday, 16 December 2015
Poster Hall (Moscone South)
Jennifer Rebecca Andrews, California Institute of Technology, GPS, Pasadena, CA, United States
Abstract:
Primary metrics for measuring earthquake early warning (EEW) system and algorithm performance are the rate of false alarms and the uncertainty in earthquake parameters. The Onsite algorithm, currently one of three EEW algorithms implemented in ShakeAlert, uses the ground-motion period parameter (τc) and peak initial displacement parameter (Pd) to estimate the magnitude and expected ground shaking of an ongoing earthquake. It is the only algorithm originally designed to issue single station alerts, necessitating that results from individual stations be as reliable and accurate as possible.

The ShakeAlert system has been undergoing testing on continuous real-time data in California for several years, and the latest version of the Onsite algorithm for several months. This permits analysis of the response to a range of signals, from environmental noise to hardware testing and maintenance procedures to moderate or large earthquake signals at varying distances from the networks. We find that our existing discriminator, relying only on τc and Pd, while performing well to exclude large teleseismic events, is less effective for moderate regional events and can also incorrectly exclude data from local events. Motivated by these experiences, we use a collection of waveforms from potentially problematic 'noise' events and real earthquakes to explore methods to discriminate real and false events, using the ground motion and period parameters available in Onsite's processing methodology.

Once an event is correctly identified, a magnitude and location estimate is critical to determining the expected ground shaking. Scatter in the measured parameters translates to higher than desired uncertainty in Onsite's current calculations We present an overview of alternative methods, including incorporation of polarization information, to improve parameter determination for a test suite including both large (M4 to M7) events and three years of small to moderate events across California.