← Back to Sessions

Full Waveform Inversion of the Los Angeles Basin Using Neural Operators

Description: 

Full waveform inversion using conventional numerical methods is computationally expensive for large 3D problems. Neural operators offer a data-driven alternative that can approximate solutions to wave equations orders of magnitude faster while remaining reasonably accurate. Building on recent advances in neural operators for elastic wave modeling and inversion, we apply the machine learning approach to the LAB2022 data set, a nodal seismic array spanning the Los Angeles Basin. We use a transformer-based Helmholtz neural operator as the forward modeling engine and automatic differentiation for full waveform inversion, enabling extensive parameter search that is prohibitive with conventional methods. Our results demonstrate the feasibility of large-scale neural-operator–based full waveform inversion using real seismic data and highlight its potential to refine velocity structures in regions with comparable spatial footprints, which is important for earthquake hazard assessment and tectonic studies.

Session: New Frontiers in Seismic Observations and Modeling with Innovative Methods and Emerging Data on Earth and Other Planets - III

Type: Oral

Date: 4/17/2026

Presentation Time: 05:00 PM (local time)

Presenting Author: Caifeng Zou

Student Presenter: Yes

Invited Presentation: 

Poster Number:

Authors