Toward earthquake early warning based on a very dense seismic network: Automatic classification of seismic waves with a machine learning technique

2020 Fiscal Year Final Research Report

• Project/Area Number: 17K13001
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Natural disaster / Disaster prevention science
• Research Institution: Japan, Meteorological Research Institute
• Principal Investigator: Kodera Yuki 気象庁気象研究所, 地震津波研究部, 研究官 (80780741)
• Project Period (FY): 2017-04-01 – 2021-03-31
• Keywords: 地震防災 / 地震動即時予測 / 緊急地震速報 / 機械学習 / 自動処理

Outline of Final Research Achievements

For the improvement of the prediction timeliness and robustness of earthquake early warning, it is important to use as many observation stations as possible, by incorporating various seismometers including ones under noisy environments. This study develops an unsupervised machine learning algorithm to classify earthquakes and noises recorded in continuous waveforms, which would lead to an automatic data quality check applicable to various seismometers. We showed that earthquakes and characteristic noises recorded in continuous waveforms can be classified with an unsupervised machine learning technique focusing on the similarity in the frequency domain and the adjacency in the time domain.

Free Research Field

地震学

Academic Significance and Societal Importance of the Research Achievements

本研究では，周波数特性の類似性と時間的な近接性という特徴をうまく利用することで，連続波形記録上の地震や特徴的ノイズは教師なし学習で分類可能であることを示すことができた．開発した手法は，地震動即時予測で用いる地震計の品質管理の自動化につながるものであり，地震動即時予測の迅速化・信頼性向上に寄与するだろう．また，本研究で得られた知見は，品質管理の自動化のみならず，様々な物理現象の自動検出にも適用可能なものであり，地震学における新たな機械学習の応用可能性を提示できた．