Japan Geoscience Union Meeting 2022

Presentation information

Print

[E] Oral

S (Solid Earth Sciences ) » S-CG Complex & General

[S-CG44] Science of slow-to-fast earthquakes

Fri. May 27, 2022 9:00 AM - 10:30 AM 103 (International Conference Hall, Makuhari Messe)

convener:Aitaro Kato(Earthquake Research Institute, the University of Tokyo), convener:Yoshiyuki Tanaka(Earth and Planetary Science, The University of Tokyo), Asuka Yamaguchi(Atomosphere and Ocean Research Institute, The University of Tokyo), convener:Takahiro Hatano(Department of Earth and Space Science, Osaka University), Chairperson:Yoshihiro Ito(Disaster Prevention Research Institute, Kyoto University), Hiroyuki Tanaka(Earthquake Research Institute, the University of Tokyo)

9:00 AM - 9:30 AM

[SCG44-13] Improved Earthquake Monitoring with AI

★Invited Papers

*Gregory C Beroza1 (1.Stanford University)

Keywords:Earthquakes , Seismicity, Artificial Intelligence, Machine Learning

Artificial Intelligence, and particularly machine learning, is having an impact in nearly all aspects of seismology. To date it is most well-established, and is having the greatest impact, in earthquake monitoring. The well-established workflow for earthquake monitoring consists of a sequence of tasks that includes: phase detection, phase association, location, and event characterization. This workflow is used to develop seismicity catalogs around the world and across scales. Because the number of earthquakes increases rapidly as magnitude decreases, cataloging smaller earthquakes dramatically increases the number of earthquakes, and hence the amount of information available, to study earthquake processes. In the past decade, the methods of AI – initially data mining, and more recently machine learning – have been applied to seismic monitoring to great benefit. Appropriate architectures, accurate data labels, and data augmentation all play important roles in developing effective models that generalize well. The simplest approach for improvement is modular, in which individual earthquake monitoring tasks are replaced one-by-one with neural network models that are applied in serial. There should be advantages in multi-task models, with a complete end-to-end model as an extreme end member, that take advantage of contextual information within a waveform and across a seismic network. AI-based earthquake monitoring is now being deployed for real-time monitoring, and there is no reason for it not to be applied comprehensively to available archived data. Seismologists are now able to develop seismicity catalogs that often include an order of magnitude more small earthquakes. Most effort to date has been directed towards fast earthquakes, but the same approaches should be useful for slow earthquakes as well. The next challenge will be to use this more complete view of seismicity to improve understanding of earthquake mechanics. The methods of AI should also be useful for this effort as well.