Fish length prediction from acoustic descriptors of Anchovy (Engraulis anchoita) schools

In a fishery assessment context, Artificial Neural Networks (ANNs), among other techniques, have been intensively tested for the identification of fish species and pursuing the automatic classification of the schools, based on descriptors extracted from digital echosounder recordings. Prediction of year-class strength is also a critical challenge for fisheries scientists and managers. In this study we extend the use of automatic classification methods, to the prediction of fish length, based on the acoustic descriptors of the fish schools. Anchovy (Engraulis anchoita) is the most abundant pelagic fish species in the SW Atlantic. We used data from a comprehensive anchovy surveys data base, comprising 9 acoustic surveys carried out between 1995 and 2008, for training and testing ANN s of different architecture. In this experience, by using only acoustic descriptors of anchovy schools ensembles, together with concurrent information on the fish size distribution obtained by trawling, we were able to satisfactory predict schematics of the ensembles age structure. Correct classification rates up to 70% were obtained.