A Bayesian Method for Real‐Time Earthquake Location Using Multiparameter Data

Abstract

A primary task of a network‐based, earthquake early warning system is the prompt event detection and location, needed to assess the magnitude of the event and its potential damage through the predicted peak ground shaking amplitude using empirical attenuation relationships. Most of real‐time, automatic earthquake location methods ground on the progressive measurement of the first P‐wave arrival time at stations located at increasing distances from the source but recent approaches showed the feasibility to improve the accuracy and rapidity of the earthquake location by using the additional information carried by the P‐wave polarization or amplitude, especially with unfavorable seismic network lay‐outs. Here, we propose an evolutionary, Bayesian method for the real‐time earthquake location which combines the information derived from the differential P‐wave arrival times, amplitude ratios and back‐azimuths measured at a minimum of two stations. As more distant stations record the P‐wave, the posterior pdf is updated and new earthquake location parameters are determined along with their uncertainty. To validate the location method, we performed a retrospective analysis of mainshocks (M > 4.5) occurred during the 2016–2017 Central Italy earthquake sequence by simulating the typical acquisition layouts of in‐land, coastal and linear array of stations. Results show that with the combined use of the three parameters, 2–4 s after the first P‐wave detection, the method converges to stable and accurate determinations of epicentral coordinates and depth even with a nonoptimal coverage of stations. The proposed methodology can be generalized and adapted to the off‐line analysis of seismic records collected by standard local networks.

Comparing and integrating artificial intelligence and similarity search detection techniques: application to seismic sequences in Southern Italy

“…Large uncertainties arise from the complexity of the structure across which waves propagate, often showing converted phases preceding the S-wave or emergent P signals (De Landro et al 2015). Future direction for picking improvement could be grounded on transfer learning to refine picking criteria based on local data and analyst measurements (e.g., Chai et al 2020), on arrival time consistency across multiple stations or including more observables, such as wave polarization (Zollo et al 2021). Template matching provides accurate relative arrival times that can be further improved by narrowing the time window around the main phases (Schaff & Waldhauser 2005).…”

Comparing and integrating artificial intelligence and similarity search detection techniques for seismic sequences in Southern Italy

“…Future direction for picking improvement could be grounded on transfer learning to refine picking criteria based on local data and analyst measurements (e.g., Chai et al 2020), on arrival time consistency across multiple stations or including more observables, such as wave polarization (Zollo et al 2021). Template matching provides accurate relative arrival times that can be further improved by narrowing the time window around the main phases (Schaff and Waldhauser, 2005).…”

Design, Implementation and Testing of a Network-Based Earthquake Early Warning System in Greece

“…Additional observables could be included to better constrain the hypocentre position in real-time. Among them, for example, the joint use of time, amplitude ratio and backazimuth estimates, as proposed by Zollo et al, (2021), could represent a valid strategy to avoid wrong location estimates for off-network events.…”

“…In this case, the initial P-wave signals are analysed and used as proxies of the late arriving, strongest shaking waves at the same site, with no or poor available information on the earthquake source parameters (such as location and magnitude) (Wu et al, 2005;Caruso et al, 2017). Finally, hybrid approaches are based on the joint use of both regional and on-site configurations, as proposed by Zollo et al (2021) and Colombelli et al (2012).…”