Applying a Deep Learning Algorithm to Tsunami Inundation Database of Megathrust Earthquakes

Abstract

We explore recent developments in computer science on deep learning to estimate high-resolution tsunami inundation from a quick low-resolution computation result. Deep network architecture is capable of storing large information acquired via a training/learning process by pairing low-and high-resolution deterministic simulation results from precalculated hypothetical scenarios. In a real case, with a real-time source estimate and linear simulation computed on a relatively low-grid resolution, optimized network parameters can be used to rapidly and accurately transform low-resolution simulation outputs to higher-resolution grids. We generate 532 source scenarios of interplate earthquakes (Mw 8-9) along the Japan Trench subduction zone to simulate tsunamis and utilize a precalculated library for deep learning. To test the proposed method, we consider a realistic source model inferred from static ground displacements and offshore tsunami data presumably available in real-time and compare forecasted inundation heights against observations in Rikuzentakata and Otsuchi cities associated with the 2011 Tohoku-oki tsunami. Our results show that the proposed method exhibits comparable accuracy to the conventional physics-based simulation but achieves approximately 90% reduction of real-time computational efforts. Thus, it has good potential as a future tsunami forecasting algorithm.

Discriminating the occurrence of inundation in tsunami early warning with one-dimensional convolutional neural networks

“…Fauzi and Mizutami 52 used a CNN to classify low resolution tsunami inundation maps, and MLP to model and map these low resolution series to the inundation map. Mulia et al 53 expanded on this, by incorporating a larger number of scenarios to calibrate a Feed Forward model with several hidden layers, aimed to characterize more complex attributes. Both works focus on inundation maps, whereas Makinoshima et al 56 use a deep 1D CNN to estimate tsunami inundation time series based on the input series obtained from a dense network of tsunameters deployed in Japan, as well as geodetic information.…”

“…The relative performance of the models varied significantly, which was associated to both the use of uniform slip for the initial condition which may not be representative enough of the tsunami characteristics and the limited number of scenarios used. Mulia et al 53 used a similar approach, considering 532 source scenarios with uniform slip distributions, using also 30 arcsec and 1.11 arcsec resolutions for the modeling. However, a Deep Feed Forward Neural Network using tsunami inundation from a low resolution LSWE model was used as input (instead of coastal tsunami amplitudes), and a high resolution inundation from a NLSWE as model output.…”

“…The present work aims to addressing these questions while providing criteria for defining the training and testing data sets. To this end, a similar configuration to that of Makinoshima et al 56 is evaluated, but considering three main differences: (i) similar to prior research, a low resolution forecast obtained from the numerical modeling of tsunami propagation is considered to be the input data 52 , 53 . The reason for this is to assess how capable is the network model to forecast in cases where offshore sea surface time series are not readily available or suitable to be used by a neural network, as occurs in many countries; (ii) A wider range of scenarios are tested.…”

Geophysical model generation with generative adversarial networks

“…Over the past few years, methods based on deep neural networks (DNN) received significant attention in the geoscientific community. They have been widely applied to various problems such as seismic data processing and interpretation (Zhu et al 2019), earthquake and tsunami prediction Mulia et al 2020;) and many others. Of particular interest is the application of modern deep learning (DL) methods to inverse problems in geophysics, i.e.…”

The 2021 Loyalty Islands Earthquake (Mw 7.7): Tsunami Waveform Inversion and Implications for Tsunami Forecasting for New Zealand

“…Facing the enduring threat of tsunamis affecting their coastal populations and infrastructures, many countries have built tsunami pre‐computed scenarios databases to support tsunami preparation and response, for example, Japan (Hoshiba & Ozaki, 2014; Tatehata, 1997), French Polynesia (Reymond et al., 2012), Turkey (Onat & Yalciner, 2013), Australia (Greenslade et al., 2011), Indonesia (Harig et al., 2019), Portugal (Matias et al., 2012), and New Caledonia (Duphil et al., 2021). High resolution tsunami inundation forecasting through scenario selection of pre‐computed scenarios, or deep learning using pre‐computed scenarios, have also been considered (Gusman et al., 2014; Mulia et al., 2018, 2020). Ways to improve the use of those databases, and the accuracy of impact forecasting, especially for scenarios whose magnitude or location lie outside the ranges of the existing ones, are of major interest.…”