PASSIVE ACOUSTIC MONITORING DETECTS DIFFERENT FREQUENCIES AND DIVERSITY OF SONGBIRDS THAN A POINT COUNT METHOD: A CASE STUDY OF EPHEMERAL WETLANDS
No Thumbnail Available
Williamson, Ansley Vaughan
There is a growing interest in using artificial intelligence (AI) for the passive acoustic monitoring (PAM) of avian biodiversity due to its potential to provide cost-effective and non-invasive methods for processing large amounts of data. Traditional point count survey methods provide essential data regarding the diversity of many songbird communities; however, they are often costly and cumbersome compared to PAM. The purpose of our study was to compare species occurrences and community diversity detected by PAM and point count detection of bird biodiversity in ephemeral wetlands while varying the PAM confidence threshold used to score a species detection and the spatial scale. Our PAM system used Solo audio recorders and the BirdNET AI algorithm to identify species. We sampled 17 ephemeral wetlands and 5 uplands for 3 days each between May 15, 2022 - June 15, 2022. The degree of difference between PAM and point counts depended strongly on the confidence threshold and spatial scale. Specifically, the median species occupancy detected between methods was significantly different for several species. Lower confidence thresholds (e.g., 0.1) in BirdNET resulted in greater species occupancy estimates than point counts, while higher thresholds (e.g., 0.9) yielded lower estimates. BirdNET estimated lower species occupancy than point counts. The confidence threshold also had a strong effect on community-level estimates of diversity, but this effect changed across the sample, site, and study spatial scales. In general, when the confidence threshold was low PAM detected higher diversity than point counts and this effect was stronger at larger spatial scales. Our results demonstrate that the confidence threshold and spatial scale will strongly influence comparisons between PAM and the point count method. PAM and point counts can be used complementarily, but the differences in monitoring shown in this analysis should be considered when designing a study that uses PAM.