Quantifying and transferring environmental uncertainties in underwater acoustic modeling

Abstract

Uncertainties in environmental inputs represent a major source of uncertainty in underwater acoustic model outputs and applications thereof (e.g., transmission-loss estimation, source localization). Seabed geoacoustic parameters are often estimated by the inversion of ocean acoustic data. Hence, rigorous quantification of geoacoustic inversion uncertainties and the transfer of these uncertainties to modeling applications are of key importance. Uncertainty estimation in geoacoustic inversion is naturally accommodated in a Bayesian formulation, which combines data and prior information to form the posterior probability density (PPD) of seabed parameters. Important components of this approach include quantitative model selection for seabed parameterizations consistent with the information content of the data; an appropriate model for residual data errors that specifies the likelihood function; and nonlinear estimation of the PPD, which is normally carried out using Markov-chain Monte Carlo (MCMC) methods. MCMC characterizes the PPD using a large ensemble of dependent random samples of the geoacoustic parameters, which can be computationally demanding. However, these uncertainties can be transferred efficiently to subsequent propagation-modeling applications using a much-smaller, randomly chosen (independent) subset from the ensemble. The approach is illustrated using simulations and inversion of ship noise recorded on a horizontal array of hydrophones at the New England Mud Patch.