Methods of acoustic data processing affect species detectability in passive acoustic monitoring of multi-species playback

Abstract

Passive acoustic monitoring (PAM) efforts have recently been accelerated by the development of automated detection tools, enabling quick and reliable analysis of recordings. However, automated methods are still susceptible to errors, and human processors achieve more accurate results. Our study evaluates the efficacy of three detection methods (auditory, visual and automated using BirdNET) for 43 European bird species (31 diurnal, 12 nocturnal), analysing the impact of various factors on detection probability over different distances. We conducted transmission experiments in two forest types from March to June, examining the effect of call characteristics, weather conditions and habitat features, to assess their impact on detection probability at different distances. Our findings reveal that species detection distance varies with each detection method, with listening to recordings obtaining the highest detectability, followed by the visual method. Although BirdNET is less accurate, it still proves useful for detection, especially for loud species. Large diurnal and small nocturnal species were most detected. Our study emphasizes the importance of considering detection methods to maximize species detectability for effective PAM research.