Details of the Abstract
| Title of paper | Attempts to detect tsunami-induced electromagnetic fields using machine learning methods: Towards tsunami early warnings |
| List of authors | "Author, Chiaki Mita, co-author, Takuto Minami, co-author, Hiroko Sugioka, co-author Hiroaki Toh" |
| Affiliation(s) |
"Graduate School of Science, Kobe University, 234s420s@stu.kobe-u.ac.jp, Graduate School of Science, Kobe University, tminami@port.kobe-u.ac.jp, Graduate School of Science, Kobe University, hikari@perl.kobe-u.ac.jp, Graduate School of Science, Kyoto University, toh@kugi.kyoto-u.ac.jp" |
| Summary | Tsunamis possibly cause significant damage to our lives. With significant tsunamis, it is known that observable tsunami-induced electromagnetic (TEM) variations arise. The electromagnetic fields are generated when conductive sea water moves in the Earth’s main magnetic fields as the tsunami propagates. Recent research findings indicate that the initial rise of TEM fields occurs prior to the variation in the tsunami wave height in deep sea cases. This phenomenon could contribute to tsunami early warnings. However, there are some obstacles to detecting TEM fields immediately. TEM phenomena are rarely reported because TEM detection is limited to the cases where the signal-to-noise ratio is large enough. So far, previous studies have been attempted to detect TEM visually in the time or frequency domain. In this study, we propose a new approach using a machine learning method. We developed 1-D CNN models composed of three convolutional layers. We set up a classification problem to categorize the input data as either containing TEM data or not containing TEM data. As input data, we prepared three components of magnetic fields, three components of the Earth’s main magnetic fields and ocean depth at the observation site. We employed several seafloor observation sites in the Philippine Sea and in the northwest Pacific Ocean. Our model was trained well, as evidenced by a significant decrease in the loss value. Currently, our experiments with real data imply difficulties in application of machine learning for detection of TEM variation in real data. |
| Session Keyword | 5.0 Monitoring: of GICs, environmental, tectonic and geomorphological hazards |
| File upload |
5.0_attempts_to_detect_tsunam_mita.pdf
|