Conversion of lands to agroecosystems has resulted in a decline in bird biodiversity. Analyzing functional diversity is a central tool for detecting changes in the ecological functions performed by birds in these landscapes. This paper aims to investigate the responses of bird taxonomic and functional diversity to landscape heterogeneity and native forest cover in Neotropical vineyards. We sampled 19 vineyard landscapes in southeastern Brazil. These landscapes covered a gradient of forest cover and heterogeneity resulting from various land uses. To assess bird diversity, we considered both taxonomic diversity and functional diversity (i.e., functional richness, evenness, and divergence). To examine the potential interactions between landscapes and bird assemblages, we employed generalized linear models (GLM). Taxonomic diversity showed no correlation with any landscape metrics. On the other hand, variation in the three metrics of functional diversity was related to landscape heterogeneity. However, in heterogeneous landscapes, these communities can be structured by limiting similarity processes. We highlight the impact of landscape homogenization on the ecological functions performed by birds in vineyards while finding no significant effect on species diversity. These findings can provide valuable support for the formulation of public policies aimed at striking a balance between agricultural production and biodiversity conservation.

A. Landscape metrics

Fieldwork was conducted in a rural area of ~40,000 ha in São Miguel Arcanjo, a significant region for grape production in Brazil. The study areas were carefully chosen based on high-resolution images obtained from ArcGIS 10.3 base map imagery, which was captured by DigitalGlobe satellites in 2016 with a scale of 1:5000 and a resolution of 0.5 m2. Seven distinct land-use classes (remnants of Atlantic Forest, regenerating forests, Eucalyptus plantations, grape plantations, other agricultures, open areas, and urban areas) were manually delineated using ArcGIS (ESRI) within the study area. To ensure accuracy, field validation was conducted in 2016 and 2017, and any errors in interpretation were rectified. For each polygon identified as a grape plantation, a central point within the vineyards was selected, and buffers with a radius of 100, 250, 500, 750, and 1,000 m were established around it. Nineteen sampling sites located at least 1,000 m from each other were selected using FRAGSTATS v4.2.1 with a forest cover percentage ranging from 18% to 55% and environmental heterogeneity assessed by Shannon's diversity index (SHDI) within a range of 0.95 to 1.78.

B. Bird Surveys

Bird surveys were conducted during the grape harvest period, from January to April 2018. Four points were established at each study area (1,000-m radius buffer), ensuring a minimum distance of 200 m between each point and the center. To assess bird communities, we employed 50-m 10-min fixed-radius point counts at each point. These counts were carried out four times per area (40-min), spanning four different days in the mornings (between 6 am and 10 am). This resulted in 160 minutes of bird counts conducted at each area (i.e. 1,000 m radius buffer) and a cumulative duration of 3,040 minutes across all 19 sampled areas. All visually and acoustically detected birds were recorded, except those flying over. The abundance was determined by aggregating the total number of contacts for each species and dividing by 16, representing the average of four daily repetitions across each of the four sampling points within a landscape.

C- Data analysis (all references in the original Manuscript)

We used species richness and abundance as measures of TD. Functional traits that may indicate ecological functional diversity; Luck et al. 2012) were applied to characterize the functional structure (i.e., the distribution of species and their abundance in the space they occupy; Villéger et al. 2008). We considered traits related to seven categories to calculate the components of FD, which include the ability of resource use, diet, foraging stratum, foraging strategy, reproductive strategy (nest location), migratory status, and activity period (Table S1). These trait categories were selected based on previous studies by Barbaro et al. (2017), Luck et al. (2012, 2015), and information from Wilman et al. (2014), Del Hoyo et al. (2019), and BirdLife International (2019).

The Gower distance (Gower 1971) was used to create a distance matrix from the trait's matrix using the ade4 package (Dray & Dufour 2007, Thioulouse et al. 2018). Subsequently, we calculated FD indexes according to the equations proposed by Villeger et al. (2008), using the FD package (Laliberté & Legendre 2010; Laliberté et al. 2015) in R v.3.5.1 (R Core Team 2018). Complementary indexes were employed to better represent the community by capturing its entire functional structure (Barbaro et al. 2017). These indexes include functional richness (FRic), functional evenness (FEve), and functional divergence (FDiv) (Laliberté & Legendre 2010; Mason et al. 2005; Petchey & Gaston 2002; Villéger et al. 2008).

FRic is the convex hull volume of the functional trait space, summarized by a principal coordinate analysis (Laliberté et al. 2015). This concept refers to the diversity of functions performed by various species within an ecosystem, representing the extensive array of roles and interactions these species play in a biological community. Theoretically, the greater the FRic, the more resilient and stable an ecosystem is, as the loss of a species can be compensated by others that perform similar functions. FEve refers to the uniformity in the distribution of the functional contributions of different species, assessing whether they perform similar functions in terms of their abundance or relative importance within an ecosystem. When FEve is high, their functions are well distributed and balanced, increasing stability. However, when FEve is low, some species may dominate and perform more important functions, increasing the vulnerability of the ecosystem. FDiv assesses the trait abundance distribution within this volume and increases with extreme trait values. It’s a more comprehensive measure that considers not only the number of functions, but also the variety and distribution of these functions in the ecosystem, assessing how different functions are distributed among species, and considering whether they overlap or are complementary (Mason et al. 2005; Villéger et al. 2008; Laliberté & Legendre 2010).

Potential influences of species richness on some FD metrics were controlled using standardized effect sizes (SES) (Mason et al. 2013). For this purpose, we employed a simulation approach to create a null model with expected values by chance. This involved maintaining the number of species constant and randomizing the abundances among species, which generated 999 random communities per site (Mason et al. 2013). Subsequently, we calculated the FD metrics and used the means and standard deviations to compute the standard effect size (SES) for each metric, following the formula SES (observed values - mean of expected values) / standard deviation of expected values (Gotelli & McCabe 2002). We tested the significance of SES values using the p-value of a one-tailed z-score (less or greater depending on the SES value). Negative values of SES indicate that the observed metrics are less than expected by chance, suggesting stronger environmental filtering and greater species similarity within the community. Conversely, positive SES values indicate that the observed metrics are higher than expected by chance, suggesting greater niche complementarity and a lower level of similarity within the community (Petchey & Gaston 2007).

Linear models with a Gaussian error distribution were conducted to analyze the impacts of forest cover and landscape heterogeneity on TD metrics, FD metrics, and standardized effect sizes for FD metrics. We used a full model structure with heterogeneity, forest cover, and their interaction, a model structure without the interaction, and simple models in which only heterogeneity or forest cover was used to explain diversity metrics. To determine the significance of each variable, we compared models and assessed the goodness of fit using the likelihood-ratio test (Quinn & Keouh 2002). In addition, we examined the relationships between TD and FD metrics through Pearson correlations. To further understand the process of bird community assembly, we compared observed values of FD and expected values using the Wilcoxon test for paired samples. The comparison involved contrasting the observed values of FD with the mean expected values obtained from 999 randomizations. We also examined the relationship between expected values of FD and species richness using simple linear regression (SLR).

To test the spatial autocorrelation of our results, we calculated Moran’s I for landscape predictors, diversity metrics, and model residuals using the spdep package (Bivand & Wong 2018).