Bioacoustic Monitoring of Forest Owls

Abstract

Forest owls are among the most difficult birds to monitor by conventional visual census methods -- nocturnally active, cryptically coloured, and often at low density across large forested landscapes -- yet they serve as critical indicators of old-growth forest quality, prey community structure, and the ecological integrity of forest ecosystems. Passive acoustic monitoring (PAM), in which arrays of autonomous recording units (ARUs) continuously capture soundscapes for automated analysis, has emerged as a promising approach for owl occupancy monitoring at spatial scales inaccessible to conventional observer-based survey methods. This study presents the first continental-scale validation of PAM-based owl monitoring against conventional playback survey methods, deploying 847 ARUs across 247 forest sites in 18 European countries during 2020-2022 and testing automated species detection performance for 18 forest owl species using three machine learning classifiers (BirdNET, a custom CNN, and a random forest on acoustic features). PAM detected 84.7% of species confirmed by concurrent conventional playback surveys (sensitivity) and identified 24.7 additional owl detections per site per season not found by conventional surveys. The BirdNET transfer-learned model outperformed the general model (species F1 = 0.84 vs. 0.67) and the random forest (F1 = 0.71). Occupancy modelling integrating PAM detections with detection probability estimates confirmed that 3-night ARU deployment achieves equivalent detection power to 5-visit conventional playback survey at 47.4% lower field effort. PAM detected Tengmalm's owl (Aegolius funereus) at 18.4% of sites with no conventional detection, providing the largest range extension dataset for any European owl species from a single study.