Bayesian inversion of seabed reverberation and scattering data via parallel tempering

Abstract

This paper describes Bayesian inversion of reverberation data for seabed scattering and geoacoustic parameters using the method of parallel tempering. The seabed is modeled as a sediment layer over a semi-infinite basement, with interface scattering occurring at the rough upper and lower boundaries of the sediment and volume scattering within the layer. The scattering mechanisms are considered to be independent and are modeled using perturbation theory and the Born approximation. Unknown parameters include seabed geoacoustic properties (sediment thickness and sound speeds, densities, and attenuations for the sediment and basement) and scattering properties (roughnesses and scattering strengths for upper and lower layer boundaries and volume scattering strength for the sediments). The reverberation inversion problem is found to be strongly nonlinear with a highly multi-modal posterior probability density (PPD). Standard Markov-chain Monte Carlo (MCMC) methods, such as Metropolis–Hastings sampling, are inef...