A geoacoustic Bayesian inversion of seafloor sediment properties using a large calibrated single-beam echosounder dataset

Seabed characterization is of interest to those studying benthic environments, pursuing offshore development, and improving national defense capability, however, direct sampling is costly even for a limited geographic scope. The National Oceanic and Atmospheric Administration (NOAA) maintains a database containing more than 200 000 km of calibrated, multifrequency, single beam echosounder (SBES) data taken over the last 15 years. Bayesian geoacoustic inversion using these data provides a promising and impartial framework by which the posterior probability distributions (PPDs) of model parameters are estimated given the information content in the data. Here, an established time domain scattering model is used, containing scattering contributions from the bulk properties, interface roughness, and heterogeneity in the sediment. The forward model is augmented, coupling the bulk geoacoustic model parameters using empirical relationships based on index of impedance, and allowing for a depth-dependent bulk density. Efficient sampling of model parameter space is employed using parallel tempering methods and principal component analysis. To provide proof-of-concept, sandy, silty, and muddy sites are chosen that contain both NOAA SBES returns and ground truth. PPDs presented agree well with ground truth, demonstrating promise for seabed characterization and associated uncertainty at unprecedented spatial scales. [Work sponsored by ONR, N000142312819.]