Aggregated time‐series features boost species‐specific differentiation of true and false positives in passive acoustic monitoring of bird assemblages

Abstract

Passive acoustic monitoring (PAM) has gained increasing popularity to study behaviour, habitat preferences, distribution and community assembly of birds and other animals. Automated species classification algorithms like ‘BirdNET’ are capable of detecting and classifying avian vocalizations within extensive audio data, covering entire species assemblages. PAM reveals substantial potential for biodiversity monitoring that informs evidence‐based conservation. Nevertheless, fully realizing this potential remains challenging, especially due to the issue of false‐positive species detections. Here, we introduce an optimized thresholding framework, which incorporates contextual information extracted from the time‐series of automated species detections (i.e. covariates on quality and quantity of species' detections measured at varying time intervals) to improve the differentiation of true and false positives. We verified a sample of BirdNET detections per species and modelled species‐specific thresholds using conditional inference trees. These thresholds were designed to minimize false‐positive detections while maximizing the preservation of true positives in the dataset. We tested this framework for a large dataset of BirdNET detections (5760 h of audio data, 60 sites) recorded over an entire breeding season. Our results revealed considerable interspecific variability of precision (percentage of true positives) within raw BirdNET data. Our optimized thresholding approach achieved high precision (≥0.9) for 70% of the 61 detected species, while species‐specific thresholds solely relying on the BirdNET confidence scores achieved high precision for only 31% of the species. Conservative universal thresholds (not species‐specific) reached high precision for 48% of the species. Our thresholding approach outperformed previous thresholding approaches and enhanced interspecific comparability for bird community analyses. By incorporating contextual information from the time‐series of species detections, the differentiation of true and false positives was substantially improved. Our approach may enhance a straightforward application of PAM in biodiversity research, landscape planning and evidence‐based conservation.

A ready-to use workflow for automated, sound-based bird identification and localization

“…Our results and those from several recent studies illustrate that confidence levels have to be chosen with care and possibly need to be adjusted (and validated) for different species (e.g. Barré et al, 2019, Cole et al, 2022, Metcalf et al, 2022, Singer et al, 2024. Setting different thresholds for different species could be easily incorporated in the analysis (exemplary code is included).…”

“…In recent years other studies have presented solutions for different aspects of our workflow. While BirdNET currently seems to be the AI of choice for multi-species systems (compare Toenies & Rich, 2021, Cole et al, 2022, Singer et al, 2024, for a comparison to other methods see Xie et al, 2023), the development of AI systems for species identification can be expected to be fast in the next years. Any novel output from possibly more advanced AI models in the future can be easily incorporated into our workflow without much adjustment.…”