10:45 〜 12:15
[HCG20-P04] Reconstruction of Sea-Level Change From 2D Stratigraphic Sections: An Inverse Model Using Convolutional Neural Network (Preliminary Studies)
キーワード：Stratigraphy, Sea-Level Change, CNN, Forward Model
This study examines how 2D stratigraphic profiles of sedimentary stacks can be connected to sea level fluctuations (eustasy change) via an effective Convolutional Neural Network (CNN) inverse model using DionisosFlow software as its forward model. DionisosFlow parameters are linked to stratigraphy by a complex 3D numerical simulation, and the CNN inverse model is trained using the images of stratigraphic sections produced by this forward model. Thus, it seems essential to first understand the necessary tasks in dealing with the forward model (DionisosFlow) prior to creating CNN inverse model. This presentation examines the sensitivity of the stratigraphic sections output by the forward model to input parameters and its response to sea-level changes. It has been found out that DionisosFlow model-making process can be automated by the application of macro codes. Data demonstration (2D+3D) and manipulating subsidence and bathymetry maps are also possible outside the DionisosFlow framework via unconventional manipulation of DionisosFlow input and output files. A study also took place regarding the relationship between DionisosFlow parameters, which proved that eustasy change has one of the highest effects on changing the configuration of sedimentary layers. Studying boundary effects (the influence of the calculation domain size) also showed that reducing basin volume considerably increases sedimentary depositions within the system, which can be normalized by reducing the initial sedimentary supply accordingly. Finally, equations for creating random eustasy change, needed in the forward model, have been created, and different types of forward model outputs for initiating the first CNN inverse model have been investigated.