Full text loading...
-
An Uncertainty Quantification Method Using Normalizing Flow for Wavefield Reconstruction Inversion with Non-Gaussian Convex Priors
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 86th EAGE Annual Conference & Exhibition, Jun 2025, Volume 2025, p.1 - 5
Abstract
Summary
Quantifying uncertainties arising during seismic processing is crucial for exploration and production decisions. Typically, uncertainty quantification (UQ) for waveform inversions is complex and costly, particularly if the prior distribution isn’t Gaussian. This study introduces a new UQ approach for waveform reconstruction inversion (WRI) with non-Gaussian priors using the normalizing flow method. Unlike traditional full-waveform inversion (FWI), WRI has a more convex objective function, allowing the likelihood distribution to be closely approximated by a Gaussian. Moreover, the normalizing flow technique facilitates generating approximate probability density functions (PDFs) for intricate target distributions by minimizing the Kullback-Leibler (KL) divergence between the approximated and actual PDFs. This capability allows analysis of PDFs with non-Gaussian convex priors. Numerical examples demonstrate that the proposed method provides reasonable uncertainty analysis within acceptable computational time.
© EAGE Publications BV