Seismic Simulations for Structure and Source Characterization in the Bay Area: Foundations for ML Acceleration of Waveform Modelling

Claire Doody, Jiun-Ting Lin, Qingkai Kong, Luis Vazquez, Caifeng Zou, Youngsoo Choi, Arthur J. Rodgers, Kamyar Azizzadenesheli, Zachary E. Ross, & Robert W. Clayton

Submitted September 7, 2025, SCEC Contribution #14968, 2025 SCEC Annual Meeting Poster #TBD

Fourier Neural Operators (FNOs) are a machine learning technique that quickly and accurately determines solution operators of partial differential equations. This method shows great promise for improving the speed of forward calculations in full waveform seismic imaging and ambient noise tomography (e.g., Kong et al., 2025; Yang et al., 2021, 2023; Zou et al., 2024). We present the foundational work for an FNO model trained for a 160kmx160kmx80km region in the greater Bay Area. We computed training data for the FNO model using Salvus (Afanasiev et al., 2019) and present our current method for computing training data. We compare the results of our model trained on a subset of our training dataset with observed data from moderate-magnitude (Mw 3.5-6) Bay Area earthquakes to show good fit to data in our target period band of 5-30 seconds. We also show that the FNO architecture can be used for faster focal mechanism inversions, converging to similar solutions to published catalogues even if the model’s first guess is significantly different from the final solution.

Key Words
Machine Learning, Full Waveform Inversion

Citation
Doody, C., Lin, J., Kong, Q., Vazquez, L., Zou, C., Choi, Y., Rodgers, A. J., Azizzadenesheli, K., Ross, Z. E., & Clayton, R. W. (2025, 09). Seismic Simulations for Structure and Source Characterization in the Bay Area: Foundations for ML Acceleration of Waveform Modelling. Poster Presentation at 2025 SCEC Annual Meeting.

Related Projects & Working Groups
Seismology