Probabilistic geoacoustic inversion

Abstract

The problem of estimating seabed geoacoustic parameters from ocean acoustic measurements has received considerable attention in recent years. Geoacoustic inversion represents a convenient alternative to direct measurements (e.g., coring) and provides sensitivity relevant to acoustic source localization applications; however, it requires solving a strongly nonlinear inverse problem. A variety of approaches have been developed (by a number of researchers) based on seeking geoacoustic parameters that provide the optimal match to measured acoustic fields using global search techniques. Other approaches include inversion of bottom-loss or seabed-reflectivity data and ambient noise. Topics of current interest include range-dependent inversion, coherent spatial/temporal processing, and uncertainty estimation. This paper reviews the above approaches in terms of a general probabilistic formulation for geoacoustic inversion. The goals of the probabilistic approach are to fit the acoustic data and available prior information to within their uncertainties, and to estimate geoacoustic parameters, their uncertainties, and inter-relationships. This is accomplished using a Bayesian formulation and Markov chain Monte Carlo approach (Gibbs sampling) to extract features of the posterior probability density such as the maximum a posteriori estimate, marginal probability distributions, and correlations. The approach is illustrated for matched-field inversion, inversion of seabed reflectivity, and source localization with environmental uncertainty.