Classifying fish schools and estimating their species proportions in fishery-acoustic surveys

Automated or computer-assisted tools are needed for estimating the proportion of species and their biomass in echosounder surveys of marine ecosystems. Operational systems rely mainly on school morphologies or the frequency responses of scatterers to identify target species in echograms. This paper presents two complementary methods for classifying schools and estimating their species proportion in a multispecies, pelagic environment. One method relies on the training of probabilistic school classifiers; the other exploits echogram similarities to infer species proportions directly from the proportions known at trawled sites. The methods are demonstrated with empirical and simulated data. School classifications and species-proportion estimates are compared to demonstrate the effectiveness of the proposed methods.