Neural operator improves subsurface velocity model updates in inversion
Full Waveform Inversion with a Learned Preconditioner
Machine Learning
Summary
Building accurate underground maps can be tricky because the math involved is tough and slow. The authors looked at whether a special kind of AI called a neural operator could help speed up and improve this process. They found that a version of this AI that understands certain math details works better than simpler ones, helping to get more reliable underground pictures. However, just predicting images with AI isn’t enough; the AI also needs to use math correctly to guide the improvements.
What this means in practice
- •For geophysical data analysts: Use learned neural preconditioners informed by Jacobians to improve the speed and accuracy of velocity model updates in seismic inversion.
- •For oil and gas exploration teams: Implement neural operator-based preconditioning to obtain better subsurface velocity reconstructions during seismic imaging workflows.
Authors
Jeongjin Park
Abstract
Full-waveform inversion (FWI) is a powerful approach for reconstructing subsurface velocity models, but its practical performance is limited by ill-conditioning, sensitivity to the update direction, and the high cost of curvature-based methods. In this project, we investigate whether a neural operator trained on reverse-time-migration (RTM) images can provide a useful learned preconditioner for FWI. We compare projected FWI against neural-preconditioned FWI using either a standard Fourier neural operator (FNO) or a Jacobian-informed FNO (JI-FNO). The neural preconditioner applies the normal operator of the learned RTM map to modify the true Devito FWI gradient. Our experiments show that the JI-FNO preconditioner achieves the lowest model error while maintaining data-residual decay comparable to baseline projected FWI. In contrast, the standard FNO preconditioner is unstable and gives worse model recovery, despite using the same neural-preconditioning framework. We also identify a failure mode of a nonlinear neural Gauss Newton-type inner solve: directly matching the neural RTM output to the FWI gradient creates an objective-space mismatch and does not yield a reliable descent direction. These results suggest that forward RTM-image prediction alone is insufficient for learned preconditioning; accurate learned derivatives and objective-consistent use of the neural operator are critical for producing useful inversion updates.