Bayesian-Maximum Entropy method applied to seabed geoacoustic models using broadband and narrowband acoustic measurements

Abstract

Listen

Translate article

Conditional probability distributions P(H|D,M) are computed using a Bayesian-Maximum Entropy method applied to acoustic measurements collected during the Seabed Characterization Experiment in the spring of 2017. Marginal distributions are computed for both seabed geoacoustic and source parameter values. A prior range-dependent seabed model M is derived from CHIRP survey measurements made in 2015. The prior bounds of parameter values forming the hypothesis vector H are obtained from an extensive set of sediment core measurements made in the region. The acoustic data, D, were produced by small explosive and combustive sources and towed tonal sources. Two acoustic models are employed to sample the N-dimensional H space: Range-dependent parabolic equation RAM and a normal mode method that includes shear wave effects. The dimensionality of M is optimized via a Gaussian Mixture Model (GMM) and compared to the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC). [Work supported by Office of Naval Research.]