Towards large‐scale abundance assessments through automated birdsong detection and distance estimation

Abstract

Passive acoustic monitoring (PAM) is emerging as a powerful tool for wildlife research. In this study we evaluate the effectiveness of PAM for automatic bird detection and distance estimation, focusing on Golden-crowned Warbler Basileuterus culicivorus and Southern Antpipit Corythopis delalandi. We found how the birds´ spatial position influences sound pressure levels (SPLs) by building suitable attenuation models. Models adjust well to the data, even though B. culicivorus exhibited a spherical spreading, whereas C. delalandi exhibited a somewhat more cylindrical model, probably reflecting a larger influence of ground and canopy. By automatically detecting the songs using template matching we were able to automatically measure received level (RL) from the recordings and use the values to estimate the distance of 271 songs of B. culicivorus. Our distance estimates ranged from 2 to 8 m, occasionally reaching 16 m, much smaller than the distances found in the literature for manual inspection of PAM recordings. We failed to automatically detect the songs of C. delalandi, as template matching efficiencies were quite low. Overall, our results suggest a promising automatic workflow to estimate the distance of the singer. Nevertheless, we found no significant correlation between the overall number of detections and the species abundance on point counts (2011–21), underscoring PAM´s limitation to provide somewhat more traditional ecological data on species populations. Current models require refinement to improve their accuracy for a better understanding of species´ spatial distribution and habitat use. Enhanced attenuation models and improved detection techniques could provide more reliable distance estimates in the future.