Bayesian full-wavefield reflection coefficient inversion and uncertainty estimation

Abstract

This paper presents a Bayesian inversion technique to recover seabed geoacoustic properties of multi-layered media from measured broadband reflection-loss data taking full-wavefield effects into account. Seismo-acoustic traces are time windowed into packets with reflection effects of different sets of layers to increasing depth. Each packet is processed to yield reflection coefficients as a function of angle and frequency. The reflection coefficient data are inverted using a full-wavefield forward model. The posterior probability density is sampled with a Gibbs sampler that steps through the packages, using the results for one package as prior knowledge to constrain subsequent inversions. Prior knowledge about the number of layers and their thicknesses is obtained through a separate Bayesian inversion of picked reflection travel times. The posterior probability density for the final package is considered the full solution to the inverse problem and is interpreted in terms of parameter estimates, marginal probability distributions, credibility intervals, and parameter correlations.