Recent progress in tsunami early warning using deep learning

Donsub Rim, Assistant Professor of Mathematics, Washington University Department of Mathematics & Statistics

In this talk, we will discuss recent progress in the development of tsunami prediction models based on standard deep learning techniques. Real-time tsunami forecasting can be useful for emergency management, but constructing an accurate model that is computationally efficient enough for real-time deployment remains a challenge. There are two aspects that make it challenging. First, during the early stages of a tsunamigenic event, there is large uncertainty regarding the earthquake source itself that produces the tsunami. Second, the tsunamis are nonlinear waves and this makes the construction of simple surrogate models difficult. Despite these difficulties, there have recently been promising results using deep learning models, showing that they can produce real-time and accurate forecasts of tsunamis within a few minutes of an earthquake. In our study, a convolutional neural network model was trained to use as input Global Navigation Satellite System (GNSS) data that measures seismic displacement and outputs a prediction of the tsunami waveform at a location interest. The deep learning models were trained on simulated data, building on decades of progress in numerical methods that is now able to create tsunami waveforms based on random synthetic earthquakes. We will provide an overview of both the numerical and the deep learning models and also discuss future directions. 

This talk is based on a joint work with R. Baraldi, C. M. Liu, R. J. LeVeque, and K. Terada. 


Host: Slava Solomatov

Register to attend via Zoom