Self-attention summary networks for subsurface velocity-model building from common-image gathers

2026-10-06

AI Summary

This study addresses the limitation of conventional workflows in which common-image gathers (CIGs) are used solely for diagnostics rather than leveraging their physical information to assist probabilistic velocity inversion. To overcome this, we propose a multi-scale self-attention summarization network that innovatively compresses three-dimensional CIG volumes into compact embeddings while preserving offset-dependent kinematic structures. By integrating these embeddings with a conditional flow matching model, the method enables efficient transport from a source distribution to the posterior of velocity fields. Numerical experiments demonstrate that, compared with directly conditioning on raw CIGs, the proposed approach significantly improves posterior velocity inference accuracy and robustness against background model mismatches, while effectively reducing predictive uncertainty.

Abstract

Common-image gathers (CIGs) contain physically meaningful information about velocity-model errors through reflector focusing and residual moveout, but in conventional imaging workflows they are typically used only as diagnostic tools. In this work, we propose a multiscale self-attention summary network that maps high-dimensional 3D CIG volumes into compact conditioning embeddings for probabilistic subsurface velocity inversion. These learned embeddings preserve offset-dependent kinematic structure and spatial coherence while reducing variability caused by background-velocity mismatch. Conditioned on these summary embeddings, a flow-matching model learns a transport from a Gaussian source distribution to the posterior distribution of plausible velocity fields. Numerical experiments show that, compared with direct conditioning on raw CIGs, the proposed summary network improves posterior velocity inference. In particular, the multiscale attention design provides greater robustness to background-model mismatch, yielding more accurate posterior reconstructions and lower predictive uncertainty.

Problem

• subsurface velocity inversion
• common-image gathers
• probabilistic inversion
• velocity-model building
• posterior uncertainty

Innovation

• self-attention summary network
• common-image gathers
• flow-matching model
• velocity-model building
• probabilistic inversion