Tsunami Data Assimilation of the 2015 Torishima Earthquake

The 2015 Torishima earthquake (M5.9) occurred at the Smith caldera on May 2, 2015. It had a CLVD-type focal mechanism and generated larger tsunami waves compared to its seismic magnitude (Sandanbata et al., 2018). Therefore, it was regarded as a 'volcanic tsunami earthquake' - a tsunami earthquake with volcanic origin. The tsunami reached Hachijo Island, Boso Peninsula and Shikoku Island, and were recorded by tide gauges and ocean bottom pressure gauges (Kubota, 2018). Fukao et al. (2018) proposed an opening horizontal sill model to explain its abnormal origin, which makes it difficult for us to forecast tsunamis.

Tsunami data assimilation forecasts the tsunami by assimilating offshore observed data into a numerical simulation, without the need of calculating the initial sea surface height at the source (Maeda et al., 2015). In the Nankai region, the stations of Dense Oceanfloor Network System for Earthquakes and Tsunamis (DONET) and Deep Sea Floor Observatory (DSFO) off Muroto Cape record the water pressure and have real-time data transmission. Using the data of 25 stations, we performed data assimilation approach and retroactively forecasted the tsunami of the 2015 Torishima earthquake. The tsunami waveforms at the tide gauges Tosashimizu and Kushimoto were forecasted, and compared with the actual observations.

The comparison between forecasted and observed waveforms at two tide gauges confirmed that our method could accurately forecast the tsunami arrival time and first-peak amplitude. The forecasted waveforms also matched well with real observations in the frequency domain. We further discuss the relationship between the number of station and the forecasting accuracy. It was found that at least nine stations were required to achieve a fair accuracy. To sum up, our method forecasted the tsunamis merely based on offshore observations, and successfully avoided estimating complex tsunami source.