11:30 AM - 11:45 AM
[MIS20-08] Inverse analysis of tsunami deposits: past and future
★Invited Papers
Keywords:natural hazard, morphodynamics, machine learning, depositional process
Why is it necessary to perform the inverse analysis of tsunami deposits? The purpose of studying tsunami deposits is to detect tsunami events that occurred in the past in the geological records. It is generally difficult to determine the recurrence period of tsunamis from historical records alone because it often exceeds several hundred years. In particular, the 2011 Tohoku-oki Earthquake tsunami was a milestone event in that it not only made the public aware of the great risk of tsunamis but also described the detailed characteristics of tsunami deposits. On the other hand, it was also recognized that the mere recognition of past tsunami occurrences is not sufficient for estimating disaster risk. A tsunami of 10 m in depth is a great risk, but a tsunami of 1 m is not so alarming. Therefore, it is vital to use inverse analysis to predict the size of tsunamis (or source fault) based on tsunami deposits.
Tsunami deposit inversion methods have evolved from a technique that can be used only in very limited situations with strong assumptions about inundation flow to a technique that can be applied to more general situations. Jaffe and Gelfenbuam (2007) developed a method to estimate tsunami velocity from the grading pattern of tsunami deposits at a single site. The method was superior in that it could be inverted from the sediment data at only one site; however, they assumed extreme situations where the inundation flow is (a) steady and spatially uniform and (b) such a flow suddenly stops, and all particles settle out. It is doubtful how many situations such an assumption would be valid since tsunami inundation flows are extremely unsteady and non-uniform. In contrast to these models, Naruse and Abe (2017) proposed a one-dimensional shallow-water equation model of tsunami inundation flows. Although this model assumes that tsunamis inundate at a constant velocity and stop abruptly and that the topography is flat, the model is significantly more applicable than the forward models used in previous inverse analysis models in that it is unsteady, uniform, and considers turbulent diffusion of suspended sand. In fact, the results of the model inversion reproduced the hydraulic conditions of the 2011 Tohoku-oki Tsunami in the Sendai Plain very well.
The DNN inverse analysis method solves the problems of conventional tsunami sediment inverse analysis. First, the DNN inversion method can remove the restriction of the forward model. In previous studies, strong assumptions on tsunami inundation flow conditions had to be made to reduce the computational load on the forward model. In the case of the DNN inverse analysis method, it is essential that the computation of the inverse analysis using the trained model is completed instantaneously. The ability to iterate the inverse analysis calculation allows the estimation error of the inverse analysis to be discussed, e.g., using the Jackknife method. It is also important to note that data along a linear transect is no longer necessarily required since the inverse analysis can be performed on a two-dimensional model.
The next target of the tsunami deposit inversion study is the analysis of past tsunami sediments. As the inverse study of the modern tsunami deposits progresses, it becomes clear that the method is effective and that the tsunami deposits contain much more information than previously thought. The velocity and depth of the tsunami inundation flows, the shape of the source fault, and the amount of slip can be obtained from the 2D inverse analysis of the tsunami sediments. The reconstruction of past wave source faults will be a valuable source of information for the recurrence cycle of plate boundary earthquakes. However, unlike the modern deposits, past tsunami deposits cannot necessarily be analyzed at a large number of sites. Thus, it is expected that the quantitative estimation of the uncertainty of the inverse analysis will be a major challenge for this field.
Tsunami deposit inversion methods have evolved from a technique that can be used only in very limited situations with strong assumptions about inundation flow to a technique that can be applied to more general situations. Jaffe and Gelfenbuam (2007) developed a method to estimate tsunami velocity from the grading pattern of tsunami deposits at a single site. The method was superior in that it could be inverted from the sediment data at only one site; however, they assumed extreme situations where the inundation flow is (a) steady and spatially uniform and (b) such a flow suddenly stops, and all particles settle out. It is doubtful how many situations such an assumption would be valid since tsunami inundation flows are extremely unsteady and non-uniform. In contrast to these models, Naruse and Abe (2017) proposed a one-dimensional shallow-water equation model of tsunami inundation flows. Although this model assumes that tsunamis inundate at a constant velocity and stop abruptly and that the topography is flat, the model is significantly more applicable than the forward models used in previous inverse analysis models in that it is unsteady, uniform, and considers turbulent diffusion of suspended sand. In fact, the results of the model inversion reproduced the hydraulic conditions of the 2011 Tohoku-oki Tsunami in the Sendai Plain very well.
The DNN inverse analysis method solves the problems of conventional tsunami sediment inverse analysis. First, the DNN inversion method can remove the restriction of the forward model. In previous studies, strong assumptions on tsunami inundation flow conditions had to be made to reduce the computational load on the forward model. In the case of the DNN inverse analysis method, it is essential that the computation of the inverse analysis using the trained model is completed instantaneously. The ability to iterate the inverse analysis calculation allows the estimation error of the inverse analysis to be discussed, e.g., using the Jackknife method. It is also important to note that data along a linear transect is no longer necessarily required since the inverse analysis can be performed on a two-dimensional model.
The next target of the tsunami deposit inversion study is the analysis of past tsunami sediments. As the inverse study of the modern tsunami deposits progresses, it becomes clear that the method is effective and that the tsunami deposits contain much more information than previously thought. The velocity and depth of the tsunami inundation flows, the shape of the source fault, and the amount of slip can be obtained from the 2D inverse analysis of the tsunami sediments. The reconstruction of past wave source faults will be a valuable source of information for the recurrence cycle of plate boundary earthquakes. However, unlike the modern deposits, past tsunami deposits cannot necessarily be analyzed at a large number of sites. Thus, it is expected that the quantitative estimation of the uncertainty of the inverse analysis will be a major challenge for this field.