Fusion of Multiresolution Seismic Tomography Maps Using Physics-informed Probability Graphical Models

Zheng Zhou, Kim B. Olsen, & Peter Gerstoft

Published September 10, 2023, SCEC Contribution #13244, 2023 SCEC Annual Meeting Poster #004 (PDF)

The resolution of velocity models obtained by tomography varies due to multiple factors and variables, such as the inversion approach, ray coverage, data quality, etc. Fusing such tomography models with different resolutions is desired when updating community models, to enable more accurate ground motion simulations. Toward this goal, we introduce a novel approach called the Physics-Informed Probability Graphical Model (PIPGM) designed to integrate seismic models with varying resolutions and uneven data point distribution. The PIPGM is able to capture relationships between subdomains of multiple resolutions, such as well-defined high-resolution (HR) embedded into low-resolution (LR) regions. By leveraging information from the HR subdomain, the PIPGM enhances details within LR areas through a maximum likelihood problem that incorporates prior knowledge from the HR regions. We assess the efficacy of the proposed methodology using both 2D and 3D velocity models, including synthetic checkerboard models as well as a fault zone model derived from the 2019 Ridgecrest, CA, earthquake sequence. Our findings demonstrate a ~38% reduction in travel time residuals compared to conventional Gaussian kernel smoothing in the 2D experiments, with similar reductions expected in 3D. These improvements stem from the PIPGM's adaptive weight selection, which effectively accommodates the complex structure of the Ridgecrest model. In contrast, traditional techniques struggle to handle nonuniformly distributed data uniformly. Our proposed PIPGM holds significant potential for enhancing our understanding of Earth's structure and offers promising advancements in other seismic research applications, such as earthquake ground motion prediction.

Key Words
Seismic Tomography, Multiresolution model fusion, Physics-informed machine learning

Citation
Zhou, Z., Olsen, K. B., & Gerstoft, P. (2023, 09). Fusion of Multiresolution Seismic Tomography Maps Using Physics-informed Probability Graphical Models. Poster Presentation at 2023 SCEC Annual Meeting.

Related Projects & Working Groups
Seismology