Coastal Tsunami Prediction in Tohoku Region, Japan, Using S-net Observations based on Artificial Neural Network
| Autor: | Wang, Y., Imai, K., Ariyoshi, K., Miyashita, T., Takahashi, N. |
|---|---|
| Rok vydání: | 2023 |
| Zdroj: | XXVIII General Assembly of the International Union of Geodesy and Geophysics (IUGG) |
| DOI: | 10.57757/iugg23-2222 |
| Popis: | Seafloor observation network for earthquakes and tsunamis along the Japan Trench (S-net) was installed offshore Japan. It provides observations for tsunami early warning. Coastal tsunami prediction using offshore tsunami observations contains two main types: tsunami inversion (Tsushima et al., 2009) and tsunami data assimilation (Maeda et al., 2015). The first type must consider the source information, whereas the second type is affected by coseismic deformation. Recently, artificial neural network was introduced to tsunami prediction (Mulia et al., 2020).We adopted a denoising autoencoder (DAE) model for coastal tsunami prediction. It is a neural network with the encoder-decoder structure trained to denoise or correct corrupted data. We used 1,000 stochastic earthquake models (M7.0–8.8) in Tohoku region, Japan, and calculated synthetic tsunami waveforms at 50 S-net stations and two coastal stations. The DAE model was trained using 800 synthetic scenarios. Then, we tested the model against 200 unseen synthetic scenarios and two real tsunami events: the 2016 Fukushima earthquake and the 2022 Tonga volcanic eruption. DAE model accurately predicted coastal tsunami waveforms for synthetic events. The maximum amplitude was accurately predicted for the 2016 Fukushima tsunami with a forecast accuracy of over 90%. The entire waveforms were also fairly predicted. However, coastal waveforms were not satisfactorily predicted for the 2022 Tonga volcanic tsunami, likely due to its different generating mechanism (i.e., meteorological tsunami). Our research is the first study to apply artificial neutral network to tsunami prediction using real observations. In the future, we will adopt more tsunami scenarios for model training. The 28th IUGG General Assembly (IUGG2023) (Berlin 2023) |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|