The Realization of an Earthquake Early Warning System for Schools and Its Performance during the 2019 ML 6.3 Hualien (Taiwan) Earthquake

Abstract

This article discusses the earthquake early warning system (EEWS) for schools of the National Center for Research on Earthquake Engineering, Taiwan (NCREE’s EEWS, earthquake early warning system [NEEWS]) that was recently completed. The system consists of 98 seismic stations with a complete set of system capabilities and 3514 broadcast stations with only the associated alert broadcast system capabilities. The broadcast stations receive both any on-site alerts issued by the seismic stations and any regional alerts issued by the Central Weather Bureau and then broadcast whatever alert is received earliest. Shortly after the establishment of the NEEWS, the ML 6.3 Hualien earthquake, which had a maximum measured peak ground acceleration (PGA) of 515.17 Gal, struck Taiwan on 18 April 2019. During the earthquake, the performance of both the seismic stations and the broadcast stations of the system was documented. The current study analyzes and discusses the accuracy of the PGA predictions, lead times, and classification performance at both the seismic stations and the broadcast stations of the NEEWS. The results show that the NEEWS is a cost-effective and promising system of EEW.

Onsite Early Prediction of PGA Using CNN With Multi-Scale and Multi-Domain P-Waves as Input

“…On the other hand, the GMPE-based approach forecast shaking based on estimates of the source parameters. Based on the experience of two real earthquake events, the accuracy of the predicted PGAs of SVR and GMPE-based approaches was quite similar, but the SVR approach could provide a longer lead time for near-epicentral sites (Hsu et al, 2018(Hsu et al, , 2021. We believe the CNN approach would show similar performance to the SVR approach because their computation time is similar.…”

“…The alerts will be issued when faster and smaller P-waves are detected after an earthquake has occurred. Public earthquake early alerts during several recent large earthquakes have been provided successfully (Fujinawa and Noda, 2013;Cuéllar et al, 2014;Yamada et al, 2014;Hsu et al, 2016Hsu et al, , 2018Hsu et al, , 2021Kodera et al, 2016;Allen and Melgar, 2019;. The algorithms of EEW techniques can be classified into regional and onsite warning ones based on their concept required to estimate an earthquake's parameters.…”

“…The regression model to predict the PGA according to these P-wave features was established based on the SVR approach. The algorithm they developed has been implemented successfully to issue alerts during several large earthquakes (Hsu et al, 2016(Hsu et al, , 2018(Hsu et al, , 2021. Furthermore, site effects on the PGA have been accommodated by including the horizontal-to-vertical spectral ratio into the input of an artificial neural network prediction model (Hsu et al, 2020).…”

Earthquake Early Warning Systems in Taiwan: Current Status

“…When the 2019 Hualien earthquake occurred, both the seismic stations of NEEWS and the alert broadcast system at all the 3,514 schools were almost ready, hence the performance of the "satellite-based approach" of NEEWS could be discussed. The lead time before the arrival of PGA was approximately 4.2-14.6 seconds for the stations having recorded PGA ≥ 25 gals with an epicenter distance of 11.1-119.7 km, and the IPAR of NEEWS was 31/36=86.1% (Hsu et al 2021). The reported IPAR of NEEWS during all above three earthquakes is supposed on a higher side as the reported best annual IPAR of the EEWS of Japan Meteorological Agency is 86% (JMA, 2016).…”

“…The implemented SVM model in the NEEWS predicts PGA based on six P-wave features extracted from the first 3 seconds of the vertical acceleration. Recently, the prediction algorithm at some of the NEEWS stations is upgraded to predict PGA based on the features extracted every second, hence the lead time could be improved (Hsu et al, 2021). The general performance of PGA prediction of the NEEWS using the SVM approach during March 2015 -July 2021 is shown in Figure 9.…”