Identification of acoustic targets off Angola using General Discriminant Analysis
AbstractPelagic fish off Angola constitute about 80% of the total fish landed and some 75% of the animal protein in the diet of the coastal population. The primary means of estimating stock abundance is through acoustic surveys, a method that is sensitive to identification of echo traces. This paper describes commonly encountered pelagic fish aggregations in terms of acoustic properties, and morphological and spatio-temporal descriptors. A knowledge base of reference observations validated by an experienced team of regional operators was used in a stepwise General Discriminant Analysis (GDA) that sought to identify traits characteristic to common species groups. A range of descriptors provided significant discriminant power (p < 0.01), and these were used to construct a classification algorithm. The resulting overall classification success emanating from the GDA was 46% when only considering the greatest posterior probability of group membership. This success rate increased to 62% and 71% respectively when including the second and third highest group membership probabilities. For Cape horse mackerel Trachurus trachurus capensis and Sardinella spp., success rates of 80% and 75% were obtained when including classifications that fell into groups with probabilities ranging from greatest to third greatest respectively.
Keywords: discriminant analysis; hydroacoustics; pelagic; species identification
African Journal of Marine Science 2006, 28(3&4): 525–533