Real‐Time Earthquake Early Warning With Deep Learning: Application to the 2016 M 6.0 Central Apennines, Italy Earthquake

• Published in: Geophysical research letters Vol. 48; no. 5
• Main Authors: Zhang, Xiong, Zhang, Miao, Tian, Xiao
• Format: Journal Article
• Language: English
• Published: 16.03.2021
• Subjects: earthquake early warning; earthquake location; earthquake magnitude; machine learning; neural network; the 2016 Central Italy earthquake
• ISSN: 0094-8276; 1944-8007
• DOI: 10.1029/2020GL089394

Earthquake early warning (EEW) systems are required to report earthquake locations and magnitudes as quickly as possible before the damaging S wave arrival to mitigate seismic hazards. Deep learning techniques provide potential for extracting earthquake source information from full seismic waveforms instead of seismic phase picks. We developed a novel deep learning EEW system that utilizes fully convolutional networks to simultaneously detect earthquakes and estimate their source parameters from continuous seismic waveform streams. The system determines earthquake location and magnitude as soon as very few stations receive earthquake signals and evolutionarily improves the solutions by receiving continuous data. We apply the system to the 2016 M 6.0 Central Apennines, Italy Earthquake and its first‐week aftershocks. Earthquake locations and magnitudes can be reliably determined as early as 4 s after the earliest P phase, with mean error ranges of 8.5–4.7 km and 0.33–0.27, respectively. Plain Language Summary Earthquake early warning (EEW) systems detect hazardous earthquakes, estimate their source parameters, and transmit warnings to the public. Conventional EEW algorithms depend on picking and analyzing the first seismic compressional wave (i.e., P wave). Seismic waveforms contain more information and can potentially be used to estimate earthquake source parameters with the fewest possible number of stations and to promptly transmit warning information. Deep learning techniques provide opportunities for extracting and exploiting the features behind seismic waveforms. In this study, we develop a fully automatic real‐time EEW system by directly mapping seismic waveform data to earthquake source parameters using deep learning techniques. We apply this system to the 2016 M 6.0 Central Apennines, Italy Earthquake and its first‐week aftershocks. Our results show EEW can be reliably issued as early as 4 s after the earliest P arrival. Key Points A fully convolutional network is designed for real‐time earthquake detection, location, and magnitude estimation Earthquake locations and magnitudes can be determined as early as a few seconds of earthquake signals received at very few stations The system evolutionarily improves and updates earthquake source parameters by receiving continuous data