Using the BirdNET algorithm to identify wolves, coyotes, and potentially their interactions in a large audio dataset

Abstract

Passive acoustic monitoring has emerged as a scalable, noninvasive tool for monitoring many acoustically active animals. Bioacoustics has long been employed to study wolves and coyotes, but the process of extracting relevant signals (e.g., territorial vocalizations) from large audio datasets remains a substantial limitation. The BirdNET algorithm is a machine learning tool originally designed to identify birds by sound, but it was recently expanded to include gray wolves (Canis lupus) and coyotes (C. latrans). We used BirdNET to analyze 10,500 h of passively recorded audio from the northern Sierra Nevada, USA, in which both species are known to occur. For wolves, real-world precision was low, but recall was high; careful post-processing of results may be necessary for an efficient workflow. For coyotes, recall and precision were high. BirdNET enabled us to identify wolves, coyotes, and apparent intra- and interspecific acoustic interactions. Because BirdNET is freely available and requires no computer science expertise to use, it may facilitate the application of passive acoustic surveys to the research and management of wolves and coyotes, two species with continental distributions that are frequently involved in high-profile and sometimes contention management decisions.