Passive acoustic monitoring has underutilized potential for studying animal behavior at a landscape scale. We used passive acoustic recordings collected across the entire western slope of the Sierra Nevada, USA, and the artificial intelligence (AI) algorithm BirdNET to investigate avian vocal responses to a predator (American goshawks) throughout this ecosystem. We found a significant decrease in total avian vocal output following the first goshawk call of the day, and this effect diminished with latitude. By retraining BirdNET on a subset of labeled mountain chickadee vocalizations, a common species with distinct predator-associated versus non-predator-associated calls, we were able to classify chickadee vocalizations by call type with high accuracy. Chickadees produced more "fee-bee" songs relative to "chick-a-dee" calls at sites with more open vegetation and at lower latitudes, and they showed a greater drop in the probability of fee-bee songs after a goshawk call at more open sites. This may reflect a tradeoff between territory quality and predation risk. The habitat-behavior interactions we observed likely could not have been uncovered without surveying a large geographical area. This highlights the value of passive bioacoustics for studying animal behavior, as well as the utility of BirdNET for classifying bioacoustic signals into categories beyond species identity.
Pardo et al. (Fri,) studied this question.