Bayesian model selection applied to self-noise geoacoustic inversion

Self-noise geoacoustic inversion involves the estimation of bottom parameters such as sound speeds and densities by analyzing towed-array signals whose origin is the tow platform itself. As well as forming inputs to more detailed assessments of seabed geology, these parameters enable performance predictions for sonar systems operating in shallow-water environments. In this paper, Gibbs sampling is used to obtain joint and marginal posterior probability distributions for seabed parameters. The advantages of viewing parameter estimation problems from such a probabilistic perspective include better quantified uncertainties for inverted parameters as well as the ability to compute Bayesian evidence for a range of competing geoacoustic models in order to judge which model explains the data most efficiently.