Fig. 1: Schematic view of tsunami forecasting by the convolutional neural network (CNN).

The CNN learns the relation between observation data and the resulting tsunami waveform from thousands of numerical simulation results. The tsunami and geodetic observation points are uniformly selected to cover the forecasting areas and are illustrated as red points in the map. After the training process is completed, the trained CNN can predict a time series of inundation tsunami waveform at the onshore forecasting site, which is represented with a green star, solely from observation data for unknown tsunami scenarios.