Japan Geoscience Union Meeting 2019

Presentation information

[J] Oral

S (Solid Earth Sciences ) » S-TT Technology & Techniques

[S-TT46] Seismic Big Data Analysis Based on the State-of-the-Art of Bayesian Statistics

Mon. May 27, 2019 1:45 PM - 3:15 PM A08 (TOKYO BAY MAKUHARI HALL)

convener:Hiromichi Nagao(Earthquake Research Institute, The University of Tokyo), Aitaro Kato(Earthquake Research Institute, the University of Tokyo), Takuto Maeda(Graduate School of Science and Technology, Hirosaki University), Keisuke Yano(The university of Tokyo), Chairperson:Nana Yoshimitsu(Earthquake Research Institute, The University of Tokyo), Takeru Matsuda(Graduate School of Information Science and Technology, The University of Tokyo)

3:00 PM - 3:15 PM

[STT46-06] Deep-learning-based Earthquake Detection for Continuous Seismic Network Records

*Keisuke Yano1, Takahiro Shiina2, Sumito Kurata1, Aitaro Kato2, Fumiyasu Komaki1, Shin'ichi Sakai2, Naoshi Hirata2 (1.Graduate School of Information Science and Technology, The University of Tokyo, , 2.Earthquake Research Institute, The University of Tokyo)

Keywords:Seismology, Deep learning

Over the last decade, continuous seismic data have been enormously acquired on seismic networks consisting of multiple sensors at distributed locations. Analyzing these data efficiently and thoroughly offers substantial benefits to seismology. The first important step in the analysis is earthquake detection, that is, detecting earthquakes in continuous massive datasets.

In this talk, we present a deep-learning-based scheme for earthquake detection from continuous records in a seismic network. We work with a convolutional neural network (CNN), which is one of the most powerful supervised learning techniques, to capture features discriminating between earthquakes and noises. Our scheme has an advantage of utilizing multiple stations in a seismic network to discriminate between earthquakes and noises.

We apply our scheme to continuous data on Metropolitan Seismic Observation network (MeSO-net) from September 4, 2011 to September 16, 2011. We show our scheme improves on CNNs based on few stations especially in preventing mis-detection. In addition, the trained network in the last fully connected layer has quasi-sparsity, by which we identify features important for CNN to recognize earthquakes.