Stable Alert Decision Method and Machine Learning based P-Wave Detection for On-Site Earthquake Early Warning

Abstract Proactively responding to earthquakes is challenging. Researchers have developed methods to reduce structural damage from earthquakes. The Earthquake Early Warning (EEW) service is more effective than seismic design at reducing human casualties. We developed an on-site EEW that can provide alerts faster than the network-based EEW. However, on-site EEW is not used as a public service due to the low alarm accuracy of the existing on-site alarm. Therefore, we applied both methods to improve the alarming accuracy of on-site EEW. First, the developed system increased the accuracy of the identification rate for the initial P-wave using deep learning. Second, the method utilizing a nearby seismometer was added to improve the accuracy of the alarm. We conducted an on-site EEW trial operation and checked the performance from 2020.10.08 to 2022.03.31. We reviewed the warning cases for small and large events during the trial and confirmed the possibility of automated alert decisions.