A Markov Chain Monte Carlo Approach for Geoacoustic Inversion and Source Localization

Gibbs Sampling, a Markov Chain Monte Carlo technique, has been shown to be a powerful tool for geoacoustic inversion and source localization. By providing estimates of posterior joint distributions, it offers a global optimization route for multi-dimensional matched-field processing that reports uncertainty and covariance in addition to point estimates. In this work, Gibbs Sampling is applied to the extraction of source location and geoacoustic parameters with matching methods that do not rely on full field calculations (unlike matched field processing). In non-dispersive environments, a Gibbs Sampler can provide estimates of path delays and amplitudes efficiently and accurately. Time delay estimates can be subsequently used for source localization, water column depth estimation, and sediment thickness estimation. Amplitudes are typically linked to geoacoustic properties of the sediments. The proposed method is validated through applications to data collected during the Haro Strait experiment.