Quantifying information content of modal dispersion data in geoacoustic inversion

Abstract

This paper illustrates the information content of ocean acoustic modal-dispersion data to constrain parameters and uncertainties of seabed geoacoustic models. In particular, time-frequency warping analysis is applied to extract dispersion data consisting of arrival times as a function of frequency for 18 of the first 21 propagating modes from recordings of an impulsive sound source on a vertical hydrophone array. To quantify the information content of these dispersion data to resolve seabed structure, a Bayesian inversion formulation is applied that includes rigorous approaches to model selection and data error modeling. Model selection considers both layered and gradient representations of seabed profiles using trans-dimensional inversion and Bernstein-polynomial basis functions, respectively. In both cases, model parameterizations are determined probabilistically from the data as part of the inversion. The error model assumes a multi-variate Gaussian distribution with unknown variance and covariance for each mode; covariance estimation is formulated in terms of trans-dimensional sampling of zeroth- and first-order autoregressive processes. The applicability of these assumptions/approaches is validated with qualitative (graphical) and quantitative residual analyses. Results are considered as marginal probability profiles for geoacoustic properties, which quantify the resolution of seabed structure versus sub-bottom depth. [Work supported by the Office of Naval Research.]