Making Earthquake Early Warning Faster and More Accurate Using Etas Seismicity Models as Bayesian Prior

Abstract eng:
Conventional Earthquake Early Warning (EEW) algorithms focus on time-series analysis of waveform data, which unfortunately requires delay between data collection and alert delivery. In order to provide reliable warning as quickly as possible before the arrival of damaging ground shaking, we propose a new algorithm that uses Bayesian probabilistic inference to provide optimally fast estimates of earthquake location; in many cases, the earthquake location is available as soon as the first P-wave arrival at the station located closest to the epicenter. The algorithm uses Epidemic-Type Aftershock Sequence (ETAS) seismicity forecast model as Bayesian prior to provide a intuitive initial approximation; the analysis of seismic waveform information is incorporated into the solution as data becomes available over time. We have evaluated the algorithm for all 504 M4+ earthquakes in Southern California between 1990 and 2005. For the earliest epicentral location estimation at 0.5 seconds after P-wave detection, the median location error using seismicity forecast with waveform analysis improved by 58% relative to results using waveform analysis only. We also present location estimations of a M5.2 Lone Pine Earthquake and a M5.4 Chino Hills Earthquake in detail which highlight the importance of Bayesian seismic prior and waveform likelihood interaction. Our new strategy has shown promising results and implementation of this methodology should significantly enhance the performance of EEW systems.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket