The application of passive acoustic monitoring to study a population of white-bearded gibbons (Hylobates albibarbis) in a heterogeneous mosaic forest in Central Kalimantan, Indonesia

Abstract

Passive acoustic monitoring has made it possible to capture datasets so large that they cannot be analysed manually in full in a reasonable timeframe. In this thesis, I propose a method to extract ecological information automatically from long-term acoustic datasets. Chapter 1 describes the development of a convolutional neural network, using an open-source framework for deep learning in bioacoustics, to automatically identify Bornean white-bearded gibbon (Hylobates albibarbis) vocalisations. This thesis details the steps involved in developing this solution such as the collection of recordings, developing a training dataset, training neural–network models, and evaluating the models’ performance. The model performed at a comparable level to similar approaches, drastically reducing analysis times when compared to a human observer and showing no significant difference in detection frequency over time when compared to a manually annotated dataset. In Chapter 2, the trained model was then applied to a 19,660-hour acoustic dataset captured in an area of unprotected forest with a heterogenous mosaic of different habitats, uncovering information on variation in rates of gibbon calling activity. We found that the rate of female great calls varied significantly through time and found no significant effect of habitat on calling rate. This represents an example of how passive acoustic monitoring, combined with an automated detector, can greatly increase the temporal resolution of studies concerning vocal species when compared to manual methods. In the final discussion, I then suggest the next steps necessary to develop this method further, to study population densities and spatial distributions of H. albibarbis and other vocal species on an ever-greater scale, helping to keep pace with a global demand for a greater quantity and scale of wildlife population monitoring programmes as a result of anthropogenic pressures on the environment.