Acoustic approach as an alternative to human-based survey in bird biodiversity monitoring in agricultural meadows

types and visibility (i.e., percent of the area within 100 m radius which can be visually inspected by observer). Meadows presented the main type of habitat within a 100 m radius around the recording sites (86%, SD = 18.6), while forests and buffer strips covered 8% (SD = 11.6%), arable fields 3% (SD = 12.0%), water reservoirs 3% (SD = 7.4%) of radius area (mean coverage of each habitat type around recording site and standard deviation (SD) are given). An average visibility within 100 m radius around a recording site was 85% (SD = 14.7%) and ranged from 40% to 100%. At each recording site highly skilled fieldworker (MJ or PS) surveyed birds using the point- count method [45]. During a 10-minute survey, all birds seen or heard were assigned to one of three distance categories: � 50 m, � 100 m, and � 100 m. Observers were trained to estimate distance before they start field surveys. At each recording site we conducted two 10-minute surveys during the 2019 breeding season: early survey (from April 11 to May 20; hours from 05:08 to 09:53 AM, local time) and late survey (from May 27 to June 30; hours from 05:01 to 08:59 AM, local time). The observations were noted in a previously prepared paper field forms. Such field methodology is commonly applied to evaluate effectiveness of agricultural measures dedicated to birds’ protection in Poland. Additionally, during each survey the soundscape (10 minutes of recording done while the human observer surveyed birds) was recorded using a portable digital recorder (Zoom H1n; two different recorders were used dur-

ing the study) with a built-in microphone (unidirectional condenser, 90˚ XY stereo format) with windscreen. The quality of the recorder used is comparable with other models of autono- mous sound recorders commonly used for soundscape recording [24]. The recorder was put horizontally on a tripod at 2 m above ground level and kept at an average of 5 m from the field- worker. We recorded WAV audio files with sampling rate 48 kHz/16 bit. The same settings of recording were used throughout the study (input level: -10 dB; low-cut filter: off; limiter: off). Acoustic analyses Soundscape recordings were analysed by the same observer who surveyed birds in the field (MJ or PS). The observer analysed recordings only from recording points at which they sur- veyed birds in the field, two months after the end of fieldwork. Observers did not have oppor- tunity to check the list of bird species detected during point-count field survey before scanning the recording from that survey. This way we kept the observer skills consistent at a recording site and eliminated potential effect of previous experience from field survey on results of a recording scanning. The observers listened to recordings and manually scanned spectrograms using Audacity 2.3.2 software, without limit of time spend on analysing a single recording. Each time they analysed a 10-minute recording sample collected exactly at the same time as a point-count done by human-observer in the field. For each recording, a list of detected bird species was prepared. Observers focused on both songs and calls produced by individuals to

recognise a species. Special attention was given to recordings where many species sang at the same time, which made species detection and identification difficult. When the fieldworker had difficulties with recognition of some birds vocalisations, he compared them with the sam- ples of vocalisations available on Xeno-Canto (Xeno-canto foundation). However, we found some sound samples which we were not able to assign to a particular species. These unrecog- nised sounds were excluded from the analysis. Statistical analyses For each survey at recording site, we calculated the number of species detected by recorder and observer over distance � 50 m, � 100 m, and unlimited range. We applied Generalised Estimating Equations (GEE), which extend the generalized linear model to allow for analysis PLOS ONE Acoustic bird biodiversity monitoring in agricultural meadows PLOS ONE | https://doi.org/10.1371/journal.pone.0266557 April 8, 2022 4 / 16