2.4. Taxonomic, functional and phylogenetic diversity metrics
For abundance estimates of each bird species, we pooled the data using the maximum abundance recorded between the two survey periods, per point, per year. We calculated several diversity metrics to characterize each community and community type, based on taxonomic diversity (TD), functional diversity (FD) and phylogenetic diversity (PD). As a measure of TD, we used the species richness observed. To have a comprehensive understanding of the functionality of bird communities, we computed two metrics of FD based on the 10 functional traits: functional divergence (FDiv), and functional evenness (FEve). FEve was used as proxy for the resilience of communities (Lee & Martin, 2017; Morelli et al., 2020). We assessed the amount of phylogenetic variation of each community through one PD metric: Mean Pairwise Distance (MPD). To calculate phylogenetic diversity, 1,000 phylogenetic trees based on the Hackett backbone (Hackett et al., 2008) were downloaded from http://birdtree.org/ (Jetz et al., 2012). Since PD metrics are dependent on species richness, we calculated standardised effect size (ses) for the MPD metric (sesMPD). To do so, we compared the observed MPD of a community with the expected MPD of 999 null communities generated with null models dependent on the “independent swap” algorithm (Swenson, 2014). This procedure removed the effect of species richness (Swenson, 2014). Both MPD and sesMPD metrics were calculated as a mean value from the 1,000 values for each community (based on the 1,000 trees; Cosset and Edwards, 2017). Then, we used a two-tailed Wilcoxon signed rank test to test the difference between observed and expected MPD (Erdős et al., 2022). We calculated all the metrics using the packages ‘FD’ (Laliberte & Legendre, 2010), ‘adiv’ (Pavoine, 2020) and ‘picante’ (Kembel et al., 2010) in R version 4.1.1 (R Development Core Team, 2021).
2.5. Data analysis
Comparison of diversity metrics between community types
We compared TD, FD and PD metrics between community types, following the three tests used in Loiola et al., 2018: 1) comparing non-invaded and invaded communities (non-invaded vs invaded; Test 1); 2) comparing native species of non-invaded communities and native species of invaded communities (non-invaded vs invaded no alien; Test 2); and 3) within invaded plots including and excluding alien species (invaded vs invaded no alien; Test 3). We performed the above tests across and within the main three land-uses, trying to assess the effect of environmental filtering on the specific species pool within a given land-use (Loiola et al., 2018). To evaluate whether differences of the various metrics between community types were significant, we performed univariate analysis of variance (ANOVA) followed by Tukey’s test.
Effect of landscape configuration and composition on diversity metrics
To assess the effect of landscape variables on diversity metrics, we built linear mixed-effect models (GLMM) with normal errors using the maximum log-likelihood method, considering all landscape composition and configuration variables as fixed effects. Moreover, to account for the dependence of observations from the same location in different years, we used point ID as a random effect. Models were run separately for each community type. We calculated Akaike’s Information Criterion (AIC) to rank each candidate model. Then, we carried out model averaging (Richards, 2008) on all models with < 2 ∆AICc using the R package MuMIn (Bartoń, 2022). To account for spatial autocorrelation, we used spline correlograms (Bjørnstad & Falck, 2001) with 1,000 bootstrap resamples, both on raw data and on model residuals (Santana et al., 2017). We inspected spline correlogram plots of full model residuals (Zuur et al., 2009), and assumed absence of spatial autocorrelation when 95% Confidence Intervals included (see Supporting Information 2).