| 研究課題/領域番号 |
19J20293
|
|---|
| 研究機関 | 東京大学 |
|---|
研究代表者 |
WANG YUCHEN 東京大学, 理学系研究科, 特別研究員(DC1) (80943290)
|
|---|
| 研究期間 (年度) |
2019-04-25 – 2022-03-31
|
| キーワード | Tsunami / Tsunami Early Warning / Data Assimilation / Real-Time Detection / Resonance Analysis |
|---|
| 研究実績の概要 |
In the past year, I have been working on tsunami data assimilation. To put tsunami data assimilation into practice, the offshore tsunami signals should be provided in real time as the input for data assimilation. To achieve this, I proposed a real-time tsunami detection algorithm using Ensemble Empirical Mode Decomposition (EEMD). EEMD decomposed the time series into a set of Intrinsic Mode Functions (IMFs). The tsunami signals of Offshore Bottom Pressure Gauge (OBPG) were automatically separated from the tidal components, seismic waves, as well as background noise. Moreover, I combined the tsunami data assimilation approach with the real-time tsunami detection algorithm. I used the 2016 Fukushima earthquake as an example. The combination facilitates a satisfactory tsunami forecast.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
1: 当初の計画以上に進展している
理由
I successfully proposed an algorithm for real-time tsunami detection. It detects the tsunami arrival with a short detection delay, and characterizes the tsunami amplitude accurately (Wang et al., 2020 SRL). By combining it with tsunami data assimilation, it facilitates a satisfactory tsunami forecast and enables the establishment of a real-time tsunami early warning system. The application to the 2016 Fukushima earthquake shows data assimilation works effectively on the S-net observation in Tohoku region (Wang & Satake, 2021 SRL). Moreover, I am working on the tsunami resonance analysis in Japan due to far-field earthquake sources. This part is in collaboration with researchers from Centro de Investigacion para la Gestion del Riesgo de Desastres (CIGIDEN), Chile.
|
| 今後の研究の推進方策 |
My next step is to further improve the tsunami early warning by using deep learning. The EEMD algorithm could detect the tsunami signals by comparing the IMF value with a pre-defined threshold. However, the threshold is set artificially according to the past experience. It is challenging to select an appropriate threshold. Therefore, we could use the technique of deep learning to improve the real-time tsunami detection algorithm and to avoid setting a threshold artificially. After being trained, the deep learning model will be able to detect the tsunami arrival, and to characterize the amplitude automatically. With the help of deep learning, the input for tsunami data assimilation can be more accurate, which leads to a better forecasting of the tsunami early warning system.
|
