Digital Twin for tsunami disaster resilience, incorporating data Assimilation of ocean bottom pressure data

Digital twin is recognized as a digital copy of a physical world stored in a digital space and used to simulate the sequences and consequences of a target phenomenon. By incorporating observed data into the digital twin, a full view of the target is obtained through real-time feedback. In our tsunami disaster digital twin platform, the tsunami inundation is first forecasted in real-time using high-performance computing.

 Our target area is the Nankai Trough subduction zone in Japan, where a great earthquake is expected to occur soon and cause a significant tsunami disaster along the coast. In this subduction zone, the dense observation systems, including pressure sensors connected by cables (DONET and N-net) were recently installed at the ocean bottom. Also, the dense GNSS observation network is available on land.

 When a great Nankai earthquake occurs, the source model of the earthquake is quickly estimated from the GNSS data using the REGARD method (Kawamoto et al., 2017). Our digital twin platform can compute the tsunami inundation along the coast of Shikoku in Japan using a high-performance computer within 5 minutes after the earthquake. However, because the GNSS network is on land, the resolution of the slip amount along the plate interface near the Nankai trough is low. 

 Therefore, we developed a novel data assimilation method using dense ocean bottom pressure data to improve the forecasted tsunami wavefield originally estimated from GNSS data using the REDARD method. Then that tsunami wavefield was used to compute the tsunami inundation along the coast by a high-performance computer as an accurate tsunami forecast.

 We tested our data assimilation method for one of the slip distributions of the great Nankai earthquake expected to occur. The reference tsunami wavefield and tsunami inundation were computed from that slip distribution. The pressure data at the actual sensors were calculated from the reference tsunami as inputs for our data assimilation. We also assumed a few preliminary fault models, which are supposed to be estimated from the GNSS data. Results show that the data assimilation method significantly improves the tsunami wavefield; therefore, the forecasted tsunami inundation along the coast is also significantly improved. Especially, the underestimation of the forecast inundation from the preliminary fault model was resolved by using our data assimilation method.

 We conclude that our novel data assimilation method with a preliminary estimated fault model is effective for real-time tsunami inundation forecasting as a tsunami Digital-Twin.