An experimental benchmark for geoacoustic inversion methods

Abstract

Over the past 25 years, there has been significant research activity in development and application of methods for inverting acoustical field data to estimate parameters of geoacoustic models of the ocean bottom. Although the performance of various geoacoustic inversion methods has been benchmarked on simulated data, their performance with experimental data remains an open question. This article constitutes the first attempt of an experimental benchmark of geoacoustic inversion methods. To do so, the article focuses on data from experiments carried out at a common site during the Shallow Water 2006 (SW06) experiment. The contribution of the article is twofold. First, the article provides an overview of experimental inversion methods and results obtained with SW06 data. Second, the article proposes and uses quantitative metrics to assess the experimental performance of inversion methods. From a sonar performance point of view, the benchmark shows that no particular geoacoustic inversion method is definitely better than any other of the ones that were tested. All the inversion methods generated adequate sound-speed profiles, but only a few methods estimated attenuation and density. Also, acoustical field prediction performance drastically reduces with range for all geoacoustic models, and this performance loss dominates over intermodel variability. Overall, the benchmark covers the two main objectives of geoacoustic inversion: obtaining geophysical information about the seabed, and/or predicting acoustic propagation in a given area.