5:15 PM - 6:45 PM
[MIS20-P06] Estimation of hydraulic conditions using a DNN inverse model: Application to paleotsunami deposit at a coastal marsh facing Beppu Bay
Keywords:Tsunami deposit, Submarine active fault, Beppu Bay, Inverse analysis
In Beppu Bay, surveys of submarine active faults and onshore tsunami deposits have revealed the chronology of fault movements and the depositional ages of paleo-tsunami deposits. According to these surveys, a total of five tsunamigenic earthquakes are thought to have occurred in the last 7300 years (Yamada et al., 2021). On the other hand, there are no historical records of tsunamis before the 1596 CE Keicho-Bungo earthquake, and their magnitude has yet to be fully investigated. In recent studies, inverse analysis methods have been proposed to quantitatively reconstruct behaviors of paleotsunamis that have not been recorded in written documents. The inverse analysis method can reconstruct the hydraulic conditions of a tsunami based on information such as thickness and grain size distribution of tsunami deposits. In this study, we aimed to quantitatively reconstruct the hydrological conditions of the paleotsunami that occurred in Beppu Bay by applying the inverse analysis method FITTNUSS-DNN (Mitra et al., 2020) to the paleotsunami deposit formed approximately 6500 years ago in a coastal marsh located on the north coast of Beppu Bay. FITTNUSS-DNN has successfully reconstructed the hydraulic conditions of medium to small paleo-tsunamis and proved to be effective for relatively small tsunamis and paleo-tsunami deposits that have been degraded over time (Batubo et al., 2024).
We conducted an additional tsunami deposit survey to gain detailed data on tsunami deposit thickness and grain size distribution for FITTNUSS-DNN. The surveyed region was a relatively small marsh extending approximately 30 m from a shoreline, about 85 m wide, 150 m deep. In this survey, cores were excavated using a gouge auger at a total of 43 sites along a shore-perpendicular transect used in the previous study (Yamada et al., 2022). In all sediment cores, sand layer was identified in an organic-rich mud sequence. This sand layer is separated from the above and below organic mud layers by a clear stratigraphic boundary, suggesting that it was formed by a sudden event. The thickness of the sand layer shows a landward thinning trend, and the mean grain size of the sand layer shows a landward fining trend.
In this study, we first developed a FITTNUSS-DNN inverse analysis model suitable for the depositional environment of the tsunami deposits investigated in this study. Its performance was tested using the artificial datasets, proving that it is close to the model performance of previous studies (e.g., Mitra et al., 2020; Batubo et al., 2024). The test results suggested that the model has sufficient capability to estimate the hydraulic conditions of a tsunami. The hydraulic conditions reconstructed by the inverse analysis were a maximum inundation distance of 74.41 m, a flow velocity of 5.440 m/s, and a maximum flow depth of 1.261 m. The thickness and grain size distribution were calculated from the forward model using the reconstructed hydraulic conditions, and it was confirmed that the estimated results of this model are well similar to the measured distributions. Uncertainty was evaluated by using the jackknife method, and the ranges of error were ±8.596×10−15 m for inundation distance, ±0.097 m/s for flow velocity, and ±0.046 for flow depth, confirming that the error ranges of the estimated values were relatively small. The reconstructed tsunami hydraulic conditions in the marsh are consistent with the characteristics of tsunamis observed to date. In the future, we aim to improve the accuracy and precision of the estimation of the paleo-tsunami behaviors by improving the inverse model and applying the model to paleo-topographic reconstructions and sediment data at multiple sites. Furthermore, we expect to clarify the scale of the prehistoric tsunami that occurred in Beppu Bay and to identify the fault rupture zone by combining the model with a tsunami simulation model that considers the behavior of the source fault.
We conducted an additional tsunami deposit survey to gain detailed data on tsunami deposit thickness and grain size distribution for FITTNUSS-DNN. The surveyed region was a relatively small marsh extending approximately 30 m from a shoreline, about 85 m wide, 150 m deep. In this survey, cores were excavated using a gouge auger at a total of 43 sites along a shore-perpendicular transect used in the previous study (Yamada et al., 2022). In all sediment cores, sand layer was identified in an organic-rich mud sequence. This sand layer is separated from the above and below organic mud layers by a clear stratigraphic boundary, suggesting that it was formed by a sudden event. The thickness of the sand layer shows a landward thinning trend, and the mean grain size of the sand layer shows a landward fining trend.
In this study, we first developed a FITTNUSS-DNN inverse analysis model suitable for the depositional environment of the tsunami deposits investigated in this study. Its performance was tested using the artificial datasets, proving that it is close to the model performance of previous studies (e.g., Mitra et al., 2020; Batubo et al., 2024). The test results suggested that the model has sufficient capability to estimate the hydraulic conditions of a tsunami. The hydraulic conditions reconstructed by the inverse analysis were a maximum inundation distance of 74.41 m, a flow velocity of 5.440 m/s, and a maximum flow depth of 1.261 m. The thickness and grain size distribution were calculated from the forward model using the reconstructed hydraulic conditions, and it was confirmed that the estimated results of this model are well similar to the measured distributions. Uncertainty was evaluated by using the jackknife method, and the ranges of error were ±8.596×10−15 m for inundation distance, ±0.097 m/s for flow velocity, and ±0.046 for flow depth, confirming that the error ranges of the estimated values were relatively small. The reconstructed tsunami hydraulic conditions in the marsh are consistent with the characteristics of tsunamis observed to date. In the future, we aim to improve the accuracy and precision of the estimation of the paleo-tsunami behaviors by improving the inverse model and applying the model to paleo-topographic reconstructions and sediment data at multiple sites. Furthermore, we expect to clarify the scale of the prehistoric tsunami that occurred in Beppu Bay and to identify the fault rupture zone by combining the model with a tsunami simulation model that considers the behavior of the source fault.