2025 Volume 81 Issue 17 Article ID: 25-17003
In conventional tsunami prediction using a sequential data assimilation ensemble Kalman filter, the initial wave source of past earthquakes has been used as an ensemble. In this study, multiple ensembles were set up by perturbing the initial wave source obtained from the magnitude and epicenter of the earthquake, which are real-time earthquake information available immediately after the earthquake occurs, using the earthquake similarity law. Data was assimilated for 5 minutes from the earthquake occurrence at the DONET observation point in the Nankai Trough using three ensemble settings: empirical information only, real-time information only, and considering both information, and predictions were made for 18 coastal points. Verification was performed using tsunamis from the 1946 Nankai earthquake, the 1944 Tonankai earthquake, and Cabinet Office Case 3, and the results showed that the prediction accuracy of the ensemble setting of the initial wave source using both information was the best in all three cases, and the RMSE of the 60-minute time waveform averaged 18 points was 0.11m, 0.25m, and 0.40m, which was an improvement over the conventional method.
