Geoacoustic inversion of sub-bottom profile data using an autonomous underwater vehicle equipped with a sound source...

Numerical and experimental studies were conducted to investigate bottom geoacoustic inversions using arrival time measurements of wide-angle seabed reflections from an autonomous underwater vehicle equipped with a sound source and a towed hydrophone array. To deal with random returns from inhomogeneous seabed sediment, multi-task Gaussian process (GP) regression is utilized to quantify the return variability, which is then input into a Bayesian inversion scheme to inform the data covariance and ultimately update the prior distributions of geoacoustic parameters. Experimental data were collected during the Seabed Characterization Experiment at the New England Mud Patch. This method provides range-dependent geoacoustic parameter estimates in the experiment area with a resolution on the order of ten meters. Numerical studies indicate that, for timing data with low variance, arrival times can be used to accurately estimate seabed properties. However, the performance of the inversion model deteriorates as the variance of the seabed reflection travel time data increases. The experimental data exhibit a high level of variance in the sub-bottom timing returns, likely due to the presence of inhomogeneities in the sediment layer and roughness between sediment layers. The mean and variance of the direct path, bottom, and sub-bottom arrival time measurements were calculated using multi-task GP regression. Furthermore, the results show that layer thickness and sound speeds are highly coupled. Additional prior information is required to decouple the ambiguity and uniquely determine seabed properties.