Evaluating Possible Drivers of Diurnal and Seasonal Patterns of Avian Singing Behavior Utilizing Autonomous Recording Units (ARUs) and Machine Learning Techniques

Autonomous recording units (ARUs) and machine learning-based species classifiers represent new emerging technologies with the potential to transform avian ecology research. In particular, these methods have the potential to shed light on patterns of diurnal and seasonal singing activity of entire bird communities with much less effort compared to traditional research methods. Using the open source BirdNET species classifier, I characterized the diurnal and seasonal singing behavior of 55 species during the breeding season from 2019-2025. From the over 27 million vocalizations classified by BirdNET, I characterized the diurnal and seasonal vocalization patterns using generalized additive models. The GAMs revealed that patterns in avian singing behavior differed among species. Therefore, I was able to group species into distinctive seasonal and diurnal activity groups. From there, I tested whether body size, residency status, foraging guild, foraging strata, sexual dimorphism, or phylogenetic relatedness explained variation in diurnal and seasonal patterns of vocalizations. Specifically, we explored how these traits impacted correlated behavior amongst species and similarities in their first peak in vocalizations. The trait axis by which diurnal singing activity groups differed the most was foraging strata, with some groups more represented by generalist species, while other groups were more dominated by perching species. We detected no phylogenetic signal for diurnal peak timing and sexual dimorphism, and body size did not predict peak singing timing. However, pairwise t-tests revealed meaningful differences in peak timing across multiple diurnal behavior groups. Our seasonal groups were influenced greatly by environmental noise; therefore our groups were not distinctive enough to yield meaningful differences in traits for this section of our analysis. Overall, we found that passive acoustic monitoring, along with the use of BirdNET, presents an efficient and standardized process for characterizing diurnal avian vocalization patterns. However, challenges with ARU degradation and environmental noise suggest that these tools complement, rather than outright replace, traditional point counts. These methods should be further explored and optimized in order to better inform future researchers on important periods of sampling in various species.