Salton Trough crustal model optimization: Wasserstein kernel

Rasheed Ajala, & Folarin Kolawole

Published September 8, 2024, SCEC Contribution #13942, 2024 SCEC Annual Meeting Poster #226 (PDF)

Earth models in high seismic risk areas must be optimized using high-frequency data to become relevant for deterministic earthquake hazard assessment and detailed subsurface geologic studies. The computational cost of such pursuit is high but has become more tractable due to advances in multiscale model merging, allowing for inversion focused on local spatial scales. Here, we present ongoing developments in inverting Imperial Valley's Salton Seismic Imaging Project (SSIP) data. Although the hybrid starting model already produces forward models with features comparable to the observed records, the long-wavelength nature of the model induces significant kinematic errors. In addition, the land-based SSIP dataset is high-frequency and noisy. Data misfit sensitivity analysis for various objective measures suggests convergence in local minima. Thus, we implement the optimal transport (Wp) metric, which is known to have better convexity and convergence properties. We specifically adopt the trace-by-trace approach using the linear integral wavefield method to satisfy the data nonnegativity condition while also benefiting from enhancements in convexity resulting from the low-frequency boost. We show compressional wave speed sensitivity kernels in Imperial Valley.

Key Words
Inversion, Optimization, Optimal Transport, Active Source

Citation
Ajala, R., & Kolawole, F. (2024, 09). Salton Trough crustal model optimization: Wasserstein kernel. Poster Presentation at 2024 SCEC Annual Meeting.

Related Projects & Working Groups
Community Earth Models (CEM)