| Summary: | Many bioacoustic studies have been able to identify individual mammals from variations in the fundamental frequency (F0) of their vocalizations. Other characteristics of vocalization which encode individuality, such as amplitude, are less frequently used because of problems with background noise and recording fidelity over distance. In this thesis, I investigate whether the inclusion of amplitude variables improves the accuracy of individual howl identification in captive Eastern grey wolves (Canis lupus lycaon). I also explore whether the use of a bespoke code to extract the howl features, combined with histogram-derived principal component analysis (PCA) values, can improve current individual wolf howl identification accuracies. From a total of 89 solo howls from six captive individuals, where distances between wolf and observer were short, I achieved 95.5% (+9.0% improvement) individual identification accuracy of captive wolves using discriminant function analysis (DFA) to classify simple scalar variables of F0 and normalized amplitudes. Moreover, this accuracy was increased to 100% when using histogram-derived PCA values of F0 and amplitudes of the first harmonic. When this method was extended to wild Eastern wolf howls, a similar result was achieved of 100% for solo howls and 97.4% for chorus howls from 119 wolves using histogram derived PCA values. This was a new result for wild Eastern grey wolves. Individuality in howls was then tested in 10 other subspecies. The results showed that all wolf subspecies tested showed individuality in the F0 and amplitude changes of their howls and could be identified with 74.0% to 100% accuracy. Finally, the use of artificial neural networks (ANNs) to survey howls using novel data was assessed. The ANNs achieved higher accuracy than DFA, where DFA did not achieve 100%, and were capable of attributing novel howls to known wolves. Therefore howls could be used as a survey method in situ.
|
|---|
