Acoustic identification of nine delphinid species in the eastern tropical Pacific Ocean

Abstract

Acoustic techniques have the potential to increase the reliability of cetacean species identification during shipboard surveys. The whistles of nine odontocete species were compared using data collected from a towed array and sonobuoys deployed during dolphin abundance surveys in the eastern tropical Pacific. Twelve variables were measured manually from spectrographic displays of each whistle (n=912). Multivariate discriminant function analysis (DFA) resulted in 49.9% of whistles being classified to the correct species. It was hypothesized that some whistles carry less species-specific information than others, therefore, groups of five whistles were averaged to reduce the effect of these ambiguous whistles. Correct classification increased to 65.4% when DFA was run on the averaged data set. A species identification decision tree that used 7 of the 12 whistle variables was constructed using nonparametric techniques (classification and regression trees) and resulted in 53.1% correct classification when applied to the original data set. Prior probabilities were added to the decision tree based on sighting rates for each species in the study area, resulting in 56.7% correct classification. The species identification decision tree provides a relatively simple acoustic method that can be used to augment conventional visual techniques.