Gradient-based Bayesian geoacoustic inversion for sediment properties at the New England mud patch

Abstract

This paper presents nonlinear Bayesian inversion of wide-angle seabed reflection-coefficient data for fine-grained/cohesive sediments recorded in the ONR Seabed Characterization Experiment at the New England mud patch. The inversion is applied to high-resolution broadband reflectivity data from a site characterized by smooth bathymetry and a thick mud layer. Since smooth, continuous gradients in seabed properties are common for low-speed mud layers, this work develops a Bayesian inversion based on a Bernstein-polynomial representation of sediment geoacoustic profiles. The gradient-based Bernstein-polynomial inversion is compared to a trans-dimensional inversion which samples over profiles defined by an unknown number of discrete seabed layers. [The research was funded by the Office of Naval Research, Ocean Acoustics Program, and the Canadian Department of National Defence.]