Geoacoustic inversion using Bayesian optimization based on Riemannian geometry

Geoacoustic inversion is a pivotal technique in ocean acoustics for estimating seabed acoustic properties, which have an important influence on underwater acoustic detection and communication in shallow water. Traditional methods, such as Bartlett matched-field inversion, encounter challenges with high-dimensional parameter spaces, leading to trade-offs with respect to computational costs and inversion uncertainty. This study presents a novel approach to geoacoustic inversion, which employs a Gaussian process-based Bayesian optimization algorithm to optimize the Riemannian processor. The method estimates geoacoustic parameters through geodesic distance computations of acoustic cross-spectral density matrices. Simulation and experimental validation using data collected during the SWellEx-96 experiment, which took place near San Diego, California, have demonstrated that the information geometry-based method achieves accuracy comparable to that of the Covariance Matrix Adaptation Evolution Strategy, while significantly reducing the computational time by 3.9 to 5.8 times. Furthermore, the Riemannian matching approach is also shown to be more robust than the traditional Bartlett method.