Mimetic birdsong affects automated bioacoustic species classification: A case study using the Northern Mockingbird, a ubiquitous North American mimic

While automated bioacoustic species classifiers play a growing role in avian population monitoring, they can produce false positive misclassifications by conflating acoustically similar sounds, including mimetic birdsong. We quantify the impact of the mimetic song of Northern Mockingbird ( Mimus polyglottos ) on the performance of BirdNET, a popular automated avian species classifier, and find that false positive detections caused by mimetic vocalizations persist in BirdNET results following common filtering methods, are indicated by detections of the mimicking species, and have an overall impact that varies with the focal species and the purpose of the bioacoustic survey. Of the 506 BirdNET false positives caused by Northern Mockingbird song in 22,483 BirdNET detections across 42 h of soundscape audio, 18 passed through species-based and confidence-based output filters. While mockingbird-associated false positives did not materially affect species richness patterns in our BirdNET results, they could impact detection rates or occupancy estimates, especially for some species. Though a case study with limited generalizability, our investigation shows that mimetic birdsong can affect the performance of this globally important avian passive acoustic monitoring tool. This impact may be explored by further study of BirdNET’s response to mimetic song and continued development of output filtering mechanisms.