Uso de assobios na identificação de delfinídeos (Odontoceti: Delphinidae) para o oceano Atlântico Sul ocidental

Abstract

Cetaceans are vocally active animals which rely almost exclusively on sound to survive in the marine environment. Delphinids (Odontoceti: Delphinidae) present a complex acoustic repertoire, comprised of echolocation clicks, burst pulses and whistles. Among these, whistles may be used in the process of acoustic identification, due to their species-specific parameter properties. In the present study, we used these vocalizations to classify delphinid species which were not visually identified in South and Southeast Brazil. To record the sounds, three Auset® matrix models were used, all composed by three omnidirectional elements (hydrophones), varying among each other in total length (250 to 300 m) and distance between hydrophones (3 to 10 m). The adopted sampling rates were of 96 kHz and 48 kHz / 24 bits, a difference which did not interfere with the extraction of acoustic parameters of whistles, as the signals were recorded in all of their contour and duration, not being limited by the equipment’s frequency. The classification analysis was carried out through the Random Forest algorithm, which demonstrated the qualify of a classification via percentage attribution. These percentages of correct and incorrect classifications, together with global accuracy, balanced accuracy and p- value, were used to evaluate the models. In general, ROC curves and their respective areas under the curve demonstrated well-adjusted models for the definition of species, while Gini and decrease in accuracy values indicated that duration, central frequency, frequency variation and highest frequency were the acoustic parameters that better contributed to the classification process. The exclusively acoustic encounters were classified as with species usually sighted along the southern and southeastern continental slope of Brazil, namely Delphinus delphis, Stenella frontalis and Stenella longirostris, which confirms the occurrence and distribution of these species in the study area, as well as the efficiency of the classification systembased on the acoustic parameters of whistles for the western South Atlantic Ocean.