Development of a conservative-automated tonal detector with high performance at a large temporal scale

In the context of global warming, monitoring changes in marine fauna distribution is crucial, especially in remote regions. Passive acoustic monitoring (PAM) has become a valuable method to detect vocal species over long periods, particularly in Arctic waters. However, the growing volume of data collected raises new analytical challenges. In this study, we developed an automatic tonal detector combining energy detection with a contour shape fitting algorithm, tested on bearded seal (Erignathus barbatus) vocalisations. This Arctic species, highly vocal during the breeding season and living in remote areas, is well-suited to PAM-based monitoring. We assessed the detector’s performance across areas and seasons. While the detector showed high accuracy in identifying individual vocalisations (100% correct detections), it also missed a substantial number of calls (77.5% false negatives). Despite this conservative detection approach, it effectively captured seasonal presence patterns, with 80% to 96% agreement with manual annotations in Svalbard, showing that this low recall rate does not compromise the reliability of the study in an ecological approach. This tool offers a rapid, first-order estimation method for detecting bearded seals – or other tonal sound-producing species – from large acoustic datasets collected over extended time periods.