Passive Acoustic Data as Phenological Distributions: Uncovering Signals of Temporal Ecology

Passive Acoustic Monitoring (PAM) is an increasingly common method for monitoring birds and other sound-producing organisms at scale, but methods that digest these data streams into ecological insight remain underdeveloped. Specifically, using PAM and classification algorithms powered by artificial intelligence (AI) to uncover the phenology of vocal animals is an emerging use of these data but currently lacks standardized, repeatable methods with verified connections to biological phenomena. Here, we articulate specific hypotheses regarding the relationship between avian vocal activity and phenological events, and present a flexible, reproducible methodological pipeline for quantifying avian vocal phenology from PAM data. We applied our pipeline to 18,568 h of audio from 185 recording sites across Olympic National Park, USA. We processed acoustic data through an AI species classifier (BirdNET), then filtered the output using species-specific precision thresholds established through expert review to minimize false positives. For 25 species representing diverse migratory strategies across two elevational strata, we used hierarchical generalized additive models (HGAMs) to estimate daily probabilities of vocal activity from which we extracted standardized "phenometrics" describing the timing, duration, and shape of vocal activity curves. PAM-derived patterns of phenometrics broadly supported expectations, showing promise for future expansion of these methods. Resident species generally exhibited earlier and longer vocal periods than migratory species, and birds at mid-elevations showed delayed and shortened vocal phenology relative to lower elevations. Many species displayed bimodal vocal patterns, with secondary peaks 30-50 days after initial peaks. These generalizable patterns of vocal phenology likely cue transitions in various stages of the avian breeding cycle. Late-season vocal activity, especially in irruptive and resident species, highlighted the method's capacity to capture ecological transitions beyond the breeding season, but robust inferences require further ground-truthing. This study advances the use of PAM for phenological research, offering outputs that can inform long-term monitoring and detect phenological shifts in response to environmental change. We make recommendations for methodological and technological advancement, and highlight the need for studies that integrate PAM data and field-based observation to further strengthen the links between observed phenometrics and confirmed biological states of vocal organisms.