Tsunami Wavefield Reconstruction and Forecasting Using the Ensemble Kalman Filter

Offshore sensor networks like DONET and S‐NET, providing real‐time estimates of wave height through measurements of pressure changes along the seafloor, are revolutionizing local tsunami early warning. Data assimilation techniques, in particular, optimal interpolation (OI), provide real‐time wavefield reconstructions and forecasts. Here we explore an alternative assimilation method, the ensemble Kalman filter (EnKF), and compare it to OI. The methods are tested on a scenario tsunami in the Cascadia subduction zone, obtained from a 2‐D coupled dynamic earthquake and tsunami simulation. Data assimilation uses a 1‐D linear long‐wave model. We find that EnKF achieves more accurate and stable forecasts than OI, both at the coast and across the entire domain, especially for large station spacing. Although EnKF is more computationally expensive than OI, with development in high‐performance computing, it is a promising candidate for real‐time local tsunami early warning. Recent years have seen more research on tsunami early warning using data from seafloor pressure sensors. Forecasts of the tsunami wave height can be computed by incorporating the pressure observations into the physical model of wave propagation. The current approach uses a simple, constant linear interpolator to blend the data with the model forecast. To improve the accuracy and stability of the forecast, we propose a more mathematically sophisticated approach that dynamically updates this interpolator, as the physical model evolves and as more data become available. This will incorporate more information into the forecast and optimize it. Using a scenario tsunami from our earthquake‐tsunami coupled simulation in the Cascadia subduction zone, we run our proposed data assimilation approach. The predicted wave heights across the ocean and at the coast are more accurate and consistent over time. More reliable forecasts can therefore be issued to coastal residents earlier in the event of a destructive tsunami. Although our method takes longer to run, with greater computing power and more efficient implementation, it is a promising candidate for real‐time tsunami early warning. We propose an alternative tsunami wavefield assimilation method, the ensemble Kalman filterThe methods are tested on scenario tsunami in Cascadia from fully coupled dynamic earthquake and tsunami simulationEnsemble Kalman filter achieved more accurate and stable forecasts compared to optimal interpolation