Estimation of Tsunami Characteristics from Deposits: Inverse Modeling Using a Deep‐Learning Neural Network

Abstract

Tsunami deposits provide information for estimating the magnitude and flow conditions of paleotsunamis, and inverse models have potential for reconstructing hydraulic conditions of tsunamis from their deposits. The majority of the previously proposed models are based on oversimplified assumptions and possess some limitations. We present a new inverse model based on the FITTNUSS model, which incorporates nonuniform and unsteady transport of suspended sediment and turbulent mixing. The present model uses a deep neural network (DNN) for the inversion method. In this method, forward model calculations are repeated for random initial flow conditions (e.g., maximum inundation length, flow velocity, maximum flow depth, and sediment concentration) to produce artificial training data sets of depositional characteristics such as thickness and grain-size distribution. The DNN was then trained to establish a general inverse model based on artificial data sets derived from the forward model. Tests conducted using independent artificial data sets indicated that this trained DNN can reconstruct the original flow conditions from the characteristics of the deposits. Finally, the model was applied to a data set of 2011 Tohoku-oki tsunami deposits. The predicted results of flow conditions were verified by the observational records at Sendai plain. Jackknife resampling was applied to estimate the precision of the result. The estimated results of the flow velocity and maximum flow depth were approximately 5.4 ± 0.1 m/s and 4.1 ± 0.2 m, respectively, after the uncertainty analysis. The DNN shows promise for reconstruction of tsunami characteristics from its deposits, which would help in estimating the hydraulic conditions of paleotsunamis. Plain Language Summary This study presents an inverse model that uses an artificial intelligence technique to estimate the hydraulic conditions of paleotsunamis from deposits. The estimated flow conditions are essential tools for disaster resilience and tsunami hazard mitigation to reduce socioeconomic impact of tsunamis on coastal cities.

Reconstruction of flow conditions from 2004 Indian Ocean tsunami deposits at the Phra Thong island using a deep neural network inverse model

“…Code and data availability. The source codes and all other data of the DNN inverse model are available in Zenodo (https://doi.org/10.5281/zenodo.4744889, Mitra et al, 2021).…”

Recovering 3D Basement Relief Using Gravity Data Through Convolutional Neural Networks

“…The feed-forward calculation results of the inversion network during the training process are evaluated by the loss function. We choose the mean squared error (Mitra et al, 2020) as the loss function of the inversion network:…”

“…The feed‐forward calculation results of the inversion network during the training process are evaluated by the loss function. We choose the mean squared error (Mitra et al., 2020) as the loss function of the inversion network: where is the number of elements in the sediment‐basement‐interface grid, is the depth of the sediment‐basement‐interface model (label), and is the predicted depth calculated by CNN. The values of this loss function, which quantifies how close the mapping from the gravity anomaly to the sediment interface, is averaged over the entire data set (Patterson & Gibson, 2017).…”

Washover deposits related to tsunami and storm surge along the north coast of the Shimokita Peninsula in northern Japan

“…In the 10 years since the 2011 Tohoku-oki earthquake (Mw 9.0), tsunami-deposit research has shown dramatic progress not only in Japan but also across the world (Goto et al 2014(Goto et al , 2021Costa and Andrade 2020;Sawai 2020;Sugawara 2021). In addition to the discovery of new tsunami deposits, research on numerical modeling of tsunami and sediment transport (Sugawara 2021) and on tsunami size estimation (Namegaya and Satake 2014;Ishimura and Yamada 2019;Naruse and Abe 2017;Mitra et al 2020Mitra et al , 2021 has been conducted. In particular, geological surveys were performed at multiple sites along the Pacific Coast of the Tohoku region where the 2011 tsunami inundated provided thousands of years of tsunami history (e.g., Goto et al 2015;Miyauchi 2015, 2017;Takada et al 2016;Inoue et al 2017).…”