A Bayesian inverse approach to identify and quantify organisms from fisheries acoustic data

Abstract Identifying sound-scattering organisms is a perennial challenge in fisheries acoustics. Most practitioners classify backscatter based on direct sampling, frequency-difference thresholds, and expert judgement, then echo-integrate at single frequency. However, this approach struggles with species mixtures, discards multi-frequency information when integrating. Inversion methods do not have these limitations, but are widely used because identifications often ambiguous the algorithms complicated to implement. We address shortcomings using probabilistic, Bayesian inversion method. Like other methods, it handles uses all available frequencies, extends naturally broadband signals. Unlike previous approaches, leverages priors rigorously incorporate from sampling biological knowledge, constraining reducing ambiguity identification. Because well-specified model should produce solutions that both wrong confident. The physical scattering processes, so its output fully interpretable, unlike some machine learning methods. Finally, can be implemented existing libraries easily parallelized for large datasets. present examples simulations field data Gulf of Alaska, discuss possible applications extensions