Estimating bird detection distances in sound recordings for standardizing detection ranges and distance sampling

Abstract

Autonomous sound recorders are increasingly used to survey birds, and other wildlife taxa. Species richness estimates from sound recordings are usually compared with estimates obtained from established methods like point counts, but so far the comparisons were biased: Detection ranges usually differ between the survey methods, and bird detection distance data are needed for standardizing data from sound recordings. We devised and tested a method for estimating bird detection distances from sound recordings, using a reference recording of test sounds at different frequencies, emitted from known distances. We used our method to estimate bird detection distances in sound recordings from tropical forest sites where point counts were also used. We derived bird abundance and richness measures and compared them between point counts and sound recordings using unlimited radius and fixed radius counts, as well as distance sampling modelling. First we show that it is possible to accurately estimate bird detection distances in sound recordings. We then demonstrate that these data can be used to standardize the detection ranges between point counts and sound recordings with a fixed‐radius approach, leading to higher abundance and richness estimates for sound recordings. Our distance‐sampling approach also revealed that sound recorders sampled significantly higher bird densities than human point counts. We show for the first time that it is possible to standardize detection ranges in sound recordings and that distance sampling can successfully be used too. We revealed that birds were flushed by human observers and that this possibly leads to lower density estimates in point counts, although sound recorders could also have sampled more birds because of their earlier deployment times. Sound recordings are more amenable to distance‐sampling modelling than point counts as they do not exhibit an observer‐induced avoidance effect, and they can easily collect more replicates for obtaining more accurate bird density estimates. Quantifying bird detection distances was so far one important shortcoming that hindered the adoption of modern autonomous sound recording methods for ecological surveys.

Comparing the sampling performance of sound recorders versus point counts in bird surveys: A meta‐analysis

“…Our analysis of alpha species richness at identical ranges revealed a strong impact of the simultaneity of survey methods, with two possible explanations: either earlier or more random sampling times of sound recorders are favourable, or a possible avoidance effect causes a drop in detected species. Such an avoidance effect in point counts has been shown using distance sampling (Darras, Furnas et al., ), and it has also previously been demonstrated using alert distances (Fernández‐Juricic, Jimenez, & Lucas, ). Birds also presumably avoided observers in other studies (Sin et al.…”

“…With a similar set‐up using one SMX‐II microphone and an ultrasound microphone (SMX‐US) that had a lower signal‐to‐noise ratio, Darras, Furnas et al. () estimated a maximum detection range of 40 m in closed forest, as only 5% of birds were detected above that distance. Hutto and Stutzman () used identical microphone capsules in an open habitat; we chose a range of 100 m as farther bird vocalizations were mainly undetected in sound recordings.…”

“…Trees are often used as installation poles; however, microphones should not be close to their trunk to avoid impeding sound transmission. Separating the microphones by approximately 25 cm would yield results most comparable to human point counts, and also ensure the best sound source localization (Maxon & Mazor, ), useful for getting a spatial image of the soundscape and for estimating bird detection distances in sound recordings (Darras, Furnas et al., ). Defining more exact measures would be impractical and have only little effect on the audio quality.…”

“…Sound recorders did not sample larger detection ranges than point counts, so when bird detection distances were available, the point count data were filtered to include only detections below a distance that corresponded to the sound recorders' detection range as estimated by the authors. In some cases, bird detection distances were available for sound recordings as well, which helped to truncate detection ranges of both survey methods to a more sensible threshold (Darras, Furnas, Fitriawan, Mulyani, & Tscharntke, ; Sedláček et al., ; Vold, Handel, & McNew, ; Wa Maina et al., ). This resulted in species richness values for identical or at least similar ranges (hereafter termed “identical”).…”

Autonomous sound recording outperforms human observation for sampling birds: a systematic map and user guide

“…In more heterogeneous montane habitats, McGrann and Furnas () detected only 1% of birds just visually and in forest, Darras et al. detected only 4% of birds just visually. Moreover, visual detections mostly concern birds flying over the sampling point, which have large ranges and are relatively unrelated to the sampled location (Kułaga and Budka ).…”