Vertical continuation of seismic waveforms through the shallow structure with neural operators

Shuye Huang, Yehuda Ben-Zion, & Jamison H. Steidl

Submitted September 7, 2025, SCEC Contribution #14491, 2025 SCEC Annual Meeting Poster #TBD

Seismic waveforms at the surface are strongly affected by linear and nonlinear wave propagation in the geotechnical top layer, topography, various site effects, and local noise sources, which limit the ability to both simulate ground motion and interpret seismograms recorded by surface sensors. To mitigate these effects, downward continuation of surface waveforms is sometimes employed as a denoising strategy. Conversely, upward continuation of borehole‑recorded data can simulate more realistic surface ground motion and enhance seismic‑hazard assessments. In this study, we utilize data recorded in a vertical borehole array and a neural operator approach to perform upward/downward continuation of seismic waveform through the top-most structure of the Earth. Specifically, we use 3-channel quality-controlled waveforms of over 5000 earthquakes recorded at the Garner Valley Downhole Array (GVDA) in southern California, paring surface and 150 m-depth bedrock instruments. We keep a signal content of up to 20 Hz to address both earthquake‑engineering needs for ground motion analysis and seismicity‑monitoring requirements. Two U‑shaped neural operators are trained to map the wavefield upward (150 m depth to surface) and downward (surface to 150 m depth), with the objectives of predicting seismograms of high coherency with observed ones and accurate ground motion parameters (peak ground acceleration, PGA, and peak ground velocity, PGV). Our results show that the trained continuation model significantly improves the waveform simulation compared to baseline algorithms and enhances both PGA and PGV estimation. Although it was not part of the UNO model’s original scope, it predicts waveform spectra matching the accuracy of the empirical transfer function. With future modification towards generalization, this approach can be extended to other sites that have both surface and borehole data for model training, and possibly including transfer approaches for additional locations.

Key Words
Waveform continuation, Ground Motion, Machine Learning

Citation
Huang, S., Ben-Zion, Y., & Steidl, J. H. (2025, 09). Vertical continuation of seismic waveforms through the shallow structure with neural operators. Poster Presentation at 2025 SCEC Annual Meeting.

Related Projects & Working Groups
Ground Motions (GM)