Virtual array inversion in the Seabed Characterization Experiment 2022

Among the main goals of the Seabed Characterization Experiment 2022 were the investigation of fine-grained sediments off the coast of New England and the assessment of various methods for estimating sediment geoacoustic properties. Tonal data collected by a vertical line array during the experiment are used in this work for source localization and geoacoustic inversion, employing both traditional matched-field inversion (MFI) and a Gaussian process (GP)-based MFI approach. The GP-based method utilizes virtual arrays generated through functions that model the coherence of the acoustic field at different ocean depths. Instead of using raw measurements, the inversion relies on predicted data at densely spaced virtual sensors, derived from interpolating the original array data. A Gaussian kernel is incorporated into the prediction process, and various virtual sensor spacings are explored for array interpolation. Genetic algorithms are applied to optimize the inversion process for both methods. The performance of each is compared through their estimates and associated uncertainty. An analysis of the results shows that the GP-based technique outperforms the traditional method, producing estimates that closely align with ground truth data and exhibit reduced uncertainty. [Work supported by ONR.]