Benchmarking Range-Dependent Propagation Modeling in Matched-Field Inversion

Abstract

This paper considers how the accuracy of range-dependent propagation modeling affects the results of matched-field inversion (MFI) for seabed geoacoustic parameters. In MFI, the forward problem of computing the acoustic fields associated with candidate geoacoustic models is solved a large number of times. Given significant mismatch due to measurement and theory errors, together with the computationally intensive nature of MFI, the appropriate tradeoff between modeling accuracy and computational speed is not obvious. This tradeoff is considered here in terms of the degradation in information content for the geoacoustic parameters that results from inaccurate propagation modeling. The information content is quantified using the marginal posterior probability distributions of the geoacoustic parameters, as computed from a fast Gibbs sampling approach to Bayesian inversion. A synthetic example of this analysis is presented in which the parabolic equation is used to model acoustic fields for a shallow-water, upslope environment, with different levels of modeling accuracy/speed controlled by the range and depth step sizes of the computational grid.