AUTHORS: Nicole Michel, National Audubon Society; Jennifer Fuller, National Audubon Society; Dale Gentry, National Audubon Society; Michael Worland, Minnesota Department of Natural Resources
ABSTRACT: The use of autonomous recording units (ARUs) is rapidly growing in the field of avian monitoring. While in-person surveys are often limited by the availability of trained observers and survey duration, ARUs enable researchers to control survey timing and drastically increase sampling frequency and duration. This improves the probability of detecting species and accurately assessing characteristics such as species richness or individual species occupancy. However, until recently, ARU recordings were manually processed by human listeners, which is inefficient and highly time-consuming. Today, numerous publicly available classifiers exist to expedite this process, but these classifiers still require human-assisted validation to confirm whether they can accurately detect species of interest. We performed a pilot study examining the effectiveness of a popular classifier, BirdNET Analyzer, for detecting 24 different migratory landbird species at the Riverbend Nature Center in Fairbault, MN. Three ARUs were deployed by the Minnesota Department of Natural Resources May 11th through May 31st, 2023, for four hours after sunrise. ARU recordings were analyzed using BirdNET Analyzer, with up to 100 random samples per species extracted and validated manually in Raven Lite software as either true or false detections. We then identified species-specific confidence thresholds where we were 90-95% confident detections were true positives using a logistic regression. We detected 21 of 24 migratory landbird species, and identified 14 species-specific thresholds for migratory landbirds in Minnesota. Overall, we found that BirdNET Analyzer with human validation was highly effective for remotely monitoring avian species in this region. These findings provide valuable insight for monitoring migratory landbirds in the Upper Mississippi River, as well as for designing effective ARU study designs for a variety of seasons and habitats.
