Acoustic species identification of schooling fish

The development of methods for the acoustic identification of fish is a long-term objective aimed at reducing uncertainty in acoustic-survey estimates. The relative frequency response r(f) measured simultaneously at several frequencies is one of the main acoustic features that characterize the targets, but the relationship between nearest neighbours, school morphology, and environmental and geographical data are also important characteristics in this context. The number of acoustic categories that can be separated with a high spatial resolution is limited by the stochastic nature of the measurements. Because the acoustic categorization of larger ensembles is more reliable than for single targets, spatial smoothing of the backscattering within the school boundaries before that process allows the separation of more categories than is possible with the raw, highly resolved data. Using the mean r(f) of an entire school gives even more reliable categorization, but determining whether or not the school is monospecific sets a new challenge. This problem is evaluated here. The methods are tested and verified. Identification of acoustic categories with similar acoustic properties is done for schooling fish, although the results have limited spatial resolution. The reliability of the categorization is further improved when knowledge of school morphology and geographical distribution of the species are taken into account.