Dimensions of biodiversity loss: Spatial mismatch in land‐use impacts on species, functional and phylogenetic diversity of European bees

Abstract Aim Agricultural intensification and urbanization are important drivers of biodiversity change in Europe. Different aspects of bee community diversity vary in their sensitivity to these pressures, as well as independently influencing ecosystem service provision (pollination). To obtain a more comprehensive understanding of human impacts on bee diversity across Europe, we assess multiple, complementary indices of diversity. Location One Thousand four hundred and forty six sites across Europe. Methods We collated data on bee occurrence and abundance from the published literature and supplemented them with the PREDICTS database. Using Rao's Quadratic Entropy, we assessed how species, functional and phylogenetic diversity of 1,446 bee communities respond to land‐use characteristics including land‐use class, cropland intensity, human population density and distance to roads. We combined these models with statistically downscaled estimates of land use in 2005 to estimate and map—at a scale of approximately 1 km2—the losses in diversity relative to semi‐natural/natural baseline (the predicted diversity of an uninhabited grid square, consisting only of semi‐natural/natural vegetation). Results We show that—relative to the predicted local diversity in uninhabited semi‐natural/natural habitat—half of all EU27 countries have lost over 10% of their average local species diversity and two‐thirds of countries have lost over 5% of their average local functional and phylogenetic diversity. All diversity measures were generally lower in pasture and higher‐intensity cropland than in semi‐natural/natural vegetation, but facets of diversity showed less consistent responses to human population density. These differences have led to marked spatial mismatches in losses: losses in phylogenetic diversity were in some areas almost 20 percentage points (pp.) more severe than losses in species diversity, but in other areas losses were almost 40 pp. less severe. Main conclusions These results highlight the importance of exploring multiple measures of diversity when prioritizing and evaluating conservation actions, as species‐diverse assemblages may be phylogenetically and functionally impoverished, potentially threatening pollination service provision.

.1: Map of sites in Europe from which bee abundance data were available. The size of the points correspond to the number of bee species sampled at the site (log-transformed). Note that all points are semi-transparent; points that appear opaque are therefore indicative of areas where multiple sites have been sampled. Table S1.2: Land-use class and intensity definitions as used in Hudson et al. One or more disturbances of moderate intensity (e.g., selective logging) or breadth of impact (e.g., bushmeat extraction), which are not severe enough to markedly change the nature of the ecosystem. Primary sites in suburban settings are at least Light use.
One or more disturbances that is severe enough to markedly change the nature of the ecosystem; this includes clear-felling of part of the site too recently for much recovery to have occurred. Primary sites in fully urban settings should be classed as Intense use.
Primary Non-Forest Pasture with significant input of fertiliser or pesticide, and with high stock density (high enough to cause significant disturbance or to stop regeneration of vegetation).

Human use (urban)
Urban Extensive managed green spaces; villages.
Suburban (e.g. gardens), or small managed or unmanaged green spaces in cities.
Fully urban with no significant green spaces.

Appendix S2 Supplementary Methods, Phylogeny
We used a birth-death polytomy resolver (?) to estimate the placement of missing species given their taxonomic affinities: congeners were constrained to be a sister in a monophyletic clade, unless the phylogenetic tree provided evidence against genus monophyly. Where species did not have congeners in the phylogenetic tree, we used higher-level taxonomic constraints for species placement; we only used such constraints where nodes had greater than 95% bootstrap support.
Caenaugochlora, Chlerogella, Pereirapis, Pseudaugochlora, Chalepogenus and Agapostemonoides were restricted to their respective tribes, where monophyly was strongly supported (100% bootstrap support, Hedtke et al., 2013). Note that Agapostemonoides was constrained within the tribe Caenohalictini, which is sometimes considered only a subtribe within the Halictini tribe (Danforth et al., 2008). Pachyprosopis (Euryglossinae: Colletidae) was constrained to be sister to the tribes Euryglossinae, Scrapterinae, and Xeromelissinae, but species were not permitted to enter the clades formed by the Xeromelissinae or Hylaeinae (Almeida & Danforth, 2009;Hedtke et al., 2013). The genus Ceylalictus was constrained to be placed within its subfamily, Nomiodinae. Where synonyms were identified using the ITIS database (taxize package), these were merged (e.g., Homalictus punctatus was synonymised with Lasioglossum punctatus). Where the published phylogeny had species placements that appear very discrepant (i.e., the placement of Ceratina japonica and Anthophora pillipes outside of their otherwise monophyletic groups and placed with fairly distantly related species) and were noted as such by the authors of the tree (Hedtke et al., 2013), these were considered missing species and their placement was estimated using pastis.
For twenty-eight bee clades with missing species (these were usually subfamilies or tribes which had greater than 85% bootstrap support; Table S2.1), a phylogenetic tree was developed using a birth-death model with zero extinction rate and exponential speciation rate using MrBayes, for at least 100,000,000 generations and four runs, with samples taken every 10000 generations. Tracer v1.4.1 was used to track effective sample sizes to assess convergence of parameter estimates. The standard deviation of split frequencies of the four independent runs were also assessed (with a value of less than 0.01 taken as evidence that the models were reaching convergence). From each of the converged runs, we then sub-sampled from the post-burn-in S13 posterior distribution of each bee clade to produce 1000 within-clade trees. A large clade of apid bees did not reach parameter convergence in any of the four independent runs, but measures of phylogenetic signal within the clade were not significantly different between runs (analysis of variance, F df =3 = 1.27, n.s.), so a random sample from all runs were taken. Random samples (without replacement) were then taken from the set of within-clade trees for each bee clade and were grafted onto the original phylogenetic tree. The original tree had first been rate smoothed, using PATHd8 (with the root age constrained to one) (Britton et al., 2007), which is a computationally efficient methods for large phylogenetic trees, and incomplete clades were pruned. To graft the clades onto the original tree, the clade was first scaled to have a depth of 1; the edge lengths were then scaled by the age of the crown node of the clade.  (Thomas et al., 2013) and MrBayes (Ronquist et al., 2012) Figure S3.1: Percentage of each 1km 2 grid cell that is low-intensity cropland.  Figure S3.2: Percentage of each 1km 2 grid cell that is medium-intensity cropland.

Legend
High-Intensity Cropland  Figure S3.3: Percentage of each 1km 2 grid cell that is high-intensity cropland.

Appendix S4.1 Spatial Autocorrelation Results
Moran's I was used to assess spatial autocorrelation in model residuals (spdep package : Bivand & Piras, 2015;Bivand et al., 2013). There was no significant evidence of spatial autocorrelation in the model residuals for species (Moran's I = -1.06, p = 0.86), functional (Moran's I = -0.94, p = 0.83) or phylogenetic diversity (Moran's I = -0.98, p = 0.83). Spatial autocorrelation was also assessed for the residuals of each study in turn. At most, two studies showed spatial autocorrelation in their residuals, but this was not more than expected by chance (one-sided χ 2 test: χ 2 = 5.0294e − 32, p = 0.5). The first level of the land-use and intensity factor (Natural/semi-natural vegetation) forms part of the intercept term and so does not explicitly appear in the coefficients