Acoustic detection rate can outperform traditional survey approaches in estimating relative densities of breeding waders

Jarrett, David; Willis, Stephen G.

Abstract

Passive acoustic devices are increasingly being used to monitor biodiversity. However, few studies have compared the accuracy of acoustic surveys and traditional surveys against ground‐truthed data. Here, we assess whether acoustic recorders used in conjunction with an artificial intelligence (AI) classifier can predict the relative breeding density of four wader species better than traditional fieldworker transect surveys. In a 27‐km2 upland study site, acoustic data were collected at 83 sampling points and analysed using the BirdNet bird‐sound classifier to estimate vocal detection rate at each location; we also carried out concurrent transect bird surveys. To ground‐truth these approaches, intensive field surveys were undertaken to identify each breeding territory of our focal species. With both the acoustic dataset and the transect dataset, we used similar analytical approaches (random forest regression trees) to predict relative territory density across the study site, and then compared these predictions with the territory density obtained from the intensive field surveys. The classifier performed well at identifying the presence of target species' vocalizations within 3‐s periods for Lapwing (accuracy = 0.911), Curlew (0.826) and Oystercatcher (0.841), but less well for Golden Plover (0.699). For Curlew and Oystercatcher, the predictions obtained from the acoustic approach were a better fit to actual territory density than the transect approach. In contrast, for Lapwing and Golden Plover, the transect predictions outperformed the acoustic predictions, with the acoustic model particularly poor for Golden Plover. We attributed these differences to the performance of the classifier, species' ecology and vocal behaviour. Data gathering for the acoustic approach was more time‐efficient than the transect surveys, requiring less than a quarter of the fieldworker days. We conclude that there is high potential for acoustic approaches to augment traditional methods, although species' ecological characteristics should be considered: species that vocalize more frequently, at higher amplitudes and hold larger territories will be better‐suited to sampling‐based acoustic methods.

Citation

Jarrett, D., & Willis, S. G. (online). Acoustic detection rate can outperform traditional survey approaches in estimating relative densities of breeding waders. Ibis: International Journal of Avian Science, https://doi.org/10.1111/ibi.13375

• Journal Article Type: Article
• Acceptance Date: Oct 16, 2024
• Online Publication Date: Nov 5, 2024
• Deposit Date: Nov 13, 2024
• Publicly Available Date: Nov 13, 2024
• Journal: Ibis
• Print ISSN: 0019-1019
• Electronic ISSN: 1474-919X
• Publisher: Wiley
• Peer Reviewed: Peer Reviewed
• DOI: https://doi.org/10.1111/ibi.13375
• Keywords: long‐term monitoring, AI, bioacoustics, machine learning, breeding waders
• Public URL: https://durham-repository.worktribe.com/output/3090284