Selection of Tsunami Observation Points Suitable for Database-Driven Prediction

During the Great East Japan Earthquake in 2011, real-time estimate of the earthquake’s magnitude was quite low, and consequently, the first report about the tsunami also understated its severity. To solve this issue, some proposed a massive overhaul of Japan’s offshore tsunami observation networks and methods to predict tsunamis in real time. In this study, we built a database containing 3,967 scenarios of tsunamis caused by earthquakes with hypocenters along the Nankai Trough, and tested a tsunami prediction method that uses this database along with offshore tsunami observation networks. Thus, we found that an uneven distribution of observation points had a negative effect on predictive accuracy. We then used simulated annealing to select the observation points to be used at each observation site and found that the predictive accuracy improved while using a few selected observation points compared to using every point.