16:15 〜 16:30
[MIS16-03] 機械学習手法を用いた安野層で見られるタービダイトの逆解析
キーワード：混濁流、数値モデル、房総半島
This study performed an inverse analysis of a turbidite bed using data from 12 outcrops in the Pliocene Anno Formation. A preestablished DNN inverse model alongside a 2D shallow-water forward model was used for the inverse analysis. The goal is to reconstruct the flow conditions of the ancient turbidity current that deposited the turbidite. The Pliocene Anno Formation is distributed in the southern part of Boso Peninsula, Chiba, Japan. It is a forearc basin-fill deposit consisting mainly of alternation of fine-grained turbidites and hemipelagic deposits. A series of tuff marker beds can be observed within the Anno formation, which made it possible to trace one single turbidite bed over a range of more than 25 km. Sampling from the sandstone bed correlated in the formation was conducted at 12 outcrops across a 20 km range.
Inverse analysis using the DNN model involves three steps. First, artificial outcrop datasets are generated using the 2D forward model. Then, training of DNN is performed using the artificial outcrop datasets. Finally, inverse analysis of artificial outcrop datasets and actual sampled outcrop data are conducted using the trained DNN. So far, the viability of this reconstruction was tested with artificial outcrop datasets and the reconstructed flow conditions were very high in accuracy. This proves that the DNN inverse model along with the 2D shallow-water forward model can potentially provide an accurate reconstruction of the flow conditions of turbidity currents from the turbidites deposited. The reconstructed flow conditions can be used to estimate the scale of flow that occurred in the past. Also, by performing calculations using the 2D forward model and the reconstructed flow conditions, the detailed two-dimensional distribution of turbidites can be reconstructed.
Inverse analysis using the DNN model involves three steps. First, artificial outcrop datasets are generated using the 2D forward model. Then, training of DNN is performed using the artificial outcrop datasets. Finally, inverse analysis of artificial outcrop datasets and actual sampled outcrop data are conducted using the trained DNN. So far, the viability of this reconstruction was tested with artificial outcrop datasets and the reconstructed flow conditions were very high in accuracy. This proves that the DNN inverse model along with the 2D shallow-water forward model can potentially provide an accurate reconstruction of the flow conditions of turbidity currents from the turbidites deposited. The reconstructed flow conditions can be used to estimate the scale of flow that occurred in the past. Also, by performing calculations using the 2D forward model and the reconstructed flow conditions, the detailed two-dimensional distribution of turbidites can be reconstructed.