[S-IT19] Coupling of deep Earth and surface processes

[SIT19-P02] Convolutional Neural Networks for Seismic Velocity Model Building and Uncertainty Quantification

*于 凡¹、Jamali Hondori Ehsan²、朴 進午¹ (1.東京大学大気海洋研究所、2.株式会社ジオサイエンス)

キーワード：マルチチャンネル地震探査、機械学習、畳み込みニューラルネットワーク

Velocity model building is a key step in seismic imaging processing, and an accurate interval velocity model is required for reliable depth migration of multichannel seismic reflection data. The conventional Seismic imaging technologies such as traveltime tomography, reverse time migration or full waveform inversion are generally operator dependent, time-consuming, and most of the time require a reasonably accurate initial velocity model to converge to a stable solution. In order to solve this problem, we propose a method to estimate the interval velocity models faster and more accurately by using Convolutional Neural Networks (CNN). In the recently developed velocity model building tools, CNN have proven to be effective for simple geological settings (Araya-Polo et al., 2018; Simon et al., 2023). However, in more complex geologies such as subduction systems, CNN have a gap compared to traditional model building methods. In addition to training the CNN model with the simple layered geological settings, we also implemented more complex synthetic geological settings into the training process to solve this issue and try to find a better solution. On the other hand, uncertainty quantification is another important step for the inversion problem in seismic imaging processing, as it provides a measure of the results that allows us to evaluate the final models. In this study, we implemented Monte-Carlo Dropout to perform uncertainty quantification of the predicted P-wave velocity model. The CNN model is trained by synthetic data and then tested on real seismic data.