Bioacoustics and machine learning for automated avian species monitoring in global biodiversity hotspots

Abstract

The complex realities of changing climate and biodiversity are often imperfectly understood. As a conservation tool, bioacoustic monitoring with machine learning (ML) models can provide valuable insights for informed decision making in conservation efforts. In this study, we built deep convolutional neural networks to classify calls of two rare bird species detected in ambient field recordings from the mountains of Nepal. With limited amount of training data, we used data augmentation techniques to effectively increase the size of training set and thus boost the model performance. The model output provides insights of species activity and abundance over time across multiple ecosystems, which can be used as a biodiversity change indicator, and also helps scientists and conservation experts to better understand species behavior, diversity, and habitat preference. This modeling methodology and its framework can be easily adopted by other acoustic classification problems.