This web application uses cookies and other tracking technologies to ensure you get the best experience.
SC/69A/SH/05
Offline resourceN/A |
Request |
| Additional files | |
SC_69A_SH_05.pdf1.8 MB |
Download |
| View in browser | |
Resource ID
20051
Access
Open
Document Number
SC/69A/SH/05
Full Title
WhatsUpp? Proof of concept automated detection of ‘upcalls’ of southern right whales on Antarctic feeding grounds
Author
Brian S Miller, Meghan G Aulich, Susannah V Calderan, Emma L. Carroll, Russell Leaper, Jamie Macauley
Authors Summary
A proof-of-concept study was conducted to explore the viability of automated acoustic detection of upcall vocalisations produced by southern right whales (Eubalaena australis) on their high-latitude Antarctic feeding grounds. The automated acoustic detector used was that of Shiu et al (2020) which was based on a convolutional neural network and deep learning, and was implemented via the open source Pamguard Deep Learning Classifier software module. Results from the detector were presented after application to a long-term acoustic dataset recorded in the Eastern Indian sector of the Southern Ocean that contained upcalls discovered during manual inspection of recordings for calls from other species. Performance assessments were largely qualitative, but revealed encouraging performance, despite the detector being trained on data recorded decades earlier and in a different hemisphere (with North Atlantic right whale calls). However, uncertainty remains about the capacity to discriminate between southern right whale and humpback whales as the source of at least some of the upcalls detected. Recommendations for future follow-up work were presented, with a key message being that a future, broad-scale, passive acoustic study of the spatial and temporal patterns in detections of SRW upcalls would likely be viable, tractable, and may require only modest additional resources by focusing on the large extant passive acoustic datasets that have already been collected throughout the Southern Ocean.
Publisher
IWC
Publication Year
2023
Abstract
A proof-of-concept study was conducted to explore the viability of automated acoustic detection of upcall vocalisations produced by southern right whales (Eubalaena australis) on their high-latitude Antarctic feeding grounds. The automated acoustic detector used was that of Shiu et al (2020) which was based on a convolutional neural network and deep learning, and was implemented via the open source Pamguard Deep Learning Classifier software module. Results from the detector are presented after application to a long-term acoustic dataset recorded in the Eastern Indian sector of the Southern Ocean that contained upcalls discovered during manual inspection of recordings for calls from other species. Performance assessments were largely qualitative, but revealed encouraging performance, despite the detector being trained on data recorded decades earlier and in a different hemisphere (with North Atlantic right whale calls). However, uncertainty remains about the capacity to discriminate between southern right whale and humpback whales as the source of at least some of the upcalls detected. Recommendations for future follow-up work are presented, with a key message being that a future, broad-scale, passive acoustic study of the spatial and temporal patterns in detections of SRW
upcalls would likely be viable, tractable, and may require only modest additional resources by focusing on the large extant passive acoustic datasets that have already been collected throughout the Southern Ocean.