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

of repeated measurements or other correlated observations [46]. GEE were used to examine the effect of the method of detection, observer and time in a season on the number of detected bird species. In the model we specified the recording point as the subject, survey (early or late) and method of detection (recorder, observer � 50 m, � 100 m, and within unlimited radius) as within-subject variables, and the number of detected bird species as the dependent variable. Data were fitted by a negative binomial distribution with a log-link function. We conducted main-effect models which contained three factors: method of detection, survey (early or late) and observer (MJ or PS). Four independent models considering (1) all bird biodiversity, (2) songbird biodiversity, (3) meadow bird biodiversity and (4) farmland bird biodiversity were conducted (see S2 Table for list of species included in each analysis). To test whether detection rates of different species determined by acoustic methods differ from the rates of observers surveying birds within a fixed and unlimited distance categories we chose the most common bird species observed during our study, i.e., those which were recorded during more than 10% of surveys (15 or more surveys from 148 conducted together during early and late surveys). For each of those species we ran Wilcoxon two-related samples test. In these tests we compared detection rate by recorder with detection rate by observer sur- veying birds within 50 m, 100 m, and unlimited distance. We defined the detection rate as a

chance to detect the species present at recording site during a single survey by each of the methods. A species was considered present if it was detected during a single, 10-minute survey by at least one of the methods. In this analysis we considered early and late survey at recording site as independent measurements. We applied simple Bonferroni correction for multiple comparisons (�a ¼ a n, where n is the number of tests). All statistical analyses were conducted in IBM SPSS Statistics 26. P-values are two-tailed. Ethics statement Our study did not involve experiments with animals, therefore we did not need any special permissions. Access to our study area was not restricted in any way. According with Polish law, access to public and private-owners lands is not restricted, excluding areas strictly desig- nated as no entry. We did not encounter areas to which accessibility was limited or illicit. No specific permissions were required for conducting our study. Results Bird biodiversity estimation by recorder and human-observer During the study we recorded 117 bird species in total, including 64 songbird species, 17 farm- land species, and 15 meadow species (S3 Table). 106 species were recorded during the early survey and 94 species during the late survey. 17 species were observed in more than 50% recording points. The most widespread species, the Eurasian skylark Alauda arvensis, was recorded in 92% of recording points. 17 species were recorded at only a single recording point. Seven species were recorded during more than 50% surveys, while 17 species during a single

survey. For more details about species distribution and frequency of occurrence see Fig 1 and S3 Table. We found a significant effect of detection method (χ2 = 1043.124; p<0 .001; df = 3), observer (χ2 = 18.248; p<0.001; df = 1), and no effect of time in a season (χ2 = 0.420; p = 0.517; df = 1) on the number of detected bird species. The recorder method detected (1) significantly more species (on average 9.84 species) than the observer up to 50 m (on average 2.22 species) and (2) observer up to 100 m (on average 6.60 species) but (3) significantly fewer species than the observer within an unlimited radius (on average 14.42 species) (Fig 2, Table 1). When we PLOS ONE Acoustic bird biodiversity monitoring in agricultural meadows PLOS ONE | https://doi.org/10.1371/journal.pone.0266557 April 8, 2022 5 / 16