Exploring the effects of habitat management on grassland biodiversity: A case study from northern Serbia

activity (from 6 to 10 AM). Birds were mapped during a 5- to 15-minute walk around the site perimeters (100 × 100 m). All recorded (observed or heard) individuals, pairs, nests, or families with fledglings were precisely mapped using an orthophoto map of the landscape where indi- vidual trees, bushes, or other structural features were clearly visible. Only records within a 100 m buffer from site boundaries were included in the analysis. Atlas codes that describe breeding probability [47] were attributed to all records. Migrating species and species observed only overflying were excluded from the analysis. Data analysis Species diversity were calculated as Shannon-Weaver index and was analysed for all study sites, while differences across sites and between management measures were tested through one-way ANOVA using Duncan’s test (P � 0.05). To assess the species similarity among sites grouped by management measures, Jaccard similarity index was adopted, whereby a value in the 0−0.25 range indicates small similarity of species composition between sets, 0.25−0.5 is interpreted as medium dissimilarity, 0.5−0.75 indicates medium similarity, and 0.75−1 is indicative of high similarity. In other words, values close to zero are associated with heterogeneous structures, while those close to unity imply homogenisation. To investigate the relationship between species (pollinator and bird) diversity and habitat variables (including plant diversity), we employed generalised linear models (GLMs), based on

the Gaussian distribution family. Separate models were constructed for the responses of polli- nators and birds, and the best model for each response variable was selected based on Akaike’s information criterion corrected for small sample size (AICc) [48]. The AICc considers both the goodness of fit of the model and the number of parameters used, with lower values indicat- ing a better trade-off between model complexity and fit to the data. Linear discriminant analysis (LDA) was performed to quantify and describe sites under dif- ferent management measures, with maximum vegetation height, total vegetation cover, and Shannon-Weaver index treated as predictor variables. Before performing the LDA analysis, data were pre-processed by scaling all the predictor variables. The dataset was then partitioned into a training set (consisting of 75% of all available observations) and a testing set (comprising the remaining 25%). The training set was used to fit the LDA model, while the testing set was used to evaluate the model performance. A ten-fold cross-validation method was adopted dur- ing the training process to tune the model hyperparameters and prevent overfitting. Finally, the accuracy of the LDA model was measured by computing the proportion of correctly classi- fied observations in the testing set. Statistical analyses (ANOVA, GLM, and LDA) were performed in R version 4.2.2 [49] using dedicated packages “vegan” [50], “agricolae” [51], “dplyr” [52], “ggplot2” [53] and “MASS” [54]. A decision tree for plants (as habitat builders) was developed in order to analyse species fre-

quency and their causal relationship at three levels: individual plots, site level, a habitat man- agement measure (G, M, or C). This method aimed to pinpoint vital plant species linked with various habitat management approaches. Decision trees are popular in ecological modeling, providing a visual representation of decision-making processes, particularly valuable for researchers and conservationists. These hierarchical structures involve attribute tests, branch- ing for outcomes, and leaf nodes for predictions [55]. In ecology, they aid in tasks like species classification, habitat assessments, and biodiversity analyses, known for their interpretability even for non-experts. This interpretability is vital for informed conservation and management decisions, allowing stakeholders to comprehend strategies. By identifying key species, habitat PLOS ONE Effects of habitat management on biodiversity PLOS ONE | https://doi.org/10.1371/journal.pone.0301391 March 28, 2024 6 / 19