USING BIOACOUSTICS AND OCCUPANCY MODELS TO UNDERSTAND SHIFTS IN BIRD VOCAL TIMING

Abstract

Northern ecosystems are being altered by rapid climate change as a result of changes in temperature, vegetation structure, and seasonal environmental conditions, which greatly influence avian communities. Understanding how these environmental changes cause birds to respond is vital for predicting future ecological dynamics in subarctic ecosystems. Current technological advances in passive acoustic monitoring (PAM) in conjunction with automated species identification provide a robust tool for examining long-term patterns of species occurrence and seasonal responses across a wide range of spatiotemporal scales. In this study, approximately 4.5 TB of continuous autonomous recording unit (ARU) data from Denali National Park and Preserve’s soundscape monitoring project in Alaska was used to investigate (1) environmental drivers of avian site use using dynamic occupancy models and (2) long-term shifts in vocal onset timing using linear regression species specific models. Continuous audio recordings collected between 2009-2023 were segmented into 15-minute recordings and analyzed using BirdNET, a deep learning-based convolutional neural network (CNN) automated bird identification program that detects and identifies birds from audio recordings using vocal characteristics and assigning confidence scores to each detection. Dynamic occupancy models were developed for White-crowned Sparrow (Zonotrichia leucophrys), Dark-eyed Junco (Junco hyemalis), and Swainson's Thrush (Catharus ustulatus) to estimate occupancy parameters (occupancy, colonization, extinction, and detection probability) while taking imperfect detection into consideration. Vocal onset timing was estimated for 17 focal species using a rolling-window approach using sustained seasonal daily detections of avian vocalizations, and linear models using Akaike’s information Criterion to examine relationships with climatic, elevational, and habitat category covariates. Occupancy relationships were modest in strength and mostly species-specific, whereas detection probability had a better supported and more consistent environmental structure. Across the study’s focal species, Air temperature consistently increased detection probability and suggests that environmental conditions strongly influence acoustic detectability. Dark-eyed Junco showed positive associations with herbaceous habitat and negative associations with precipitation, exhibiting the strongest occupancy structure. Alternatively, White-crowned sparrow demonstrated relatively diffuse habitat relationships. There was a positive association with shrub and herbaceous habitats for Swainson’s Thrush, but occupancy environment relationships were less definitive. Similarly, Vocal phenology revealed species-specific responses rather than a broader community wide trend. Two species demonstrated strong vocal onset responses through time; Wilson’s warbler showed earlier singing, whereas Gray-cheeked Thrush sang later across time. However, the remaining species mostly demonstrated weak or uncertain temporal trends after accounting for environmental and habitat variables. Our findings underscore that climate and habitat influence both the probability of acoustically detecting birds, and the phenology of seasonal vocal activity, however, species’ responses varied tremendously. This research integrates a reproducible workflow for passive acoustic monitoring, automated species detection and identification, dynamic occupancy modeling, and vocal phenology analysis to investigate long term, large scale environmental and habitat changes across northern landscapes and their ecological responses. These findings provide a baseline for robust avian monitoring in Denali National Park and Preserve and highlight the value of combining passive acoustic monitoring with analytical modeling to better understand how subarctic bird communities are responding to ongoing climate shifts.