Disentangling urban habitat and matrix effects on wild bee species

Study area and study sites

All study sites were located in Marzahn-Hellersdorf (61.8 km², 256,173 inhabitants), a district in the northeast of the German capital Berlin (Fig. 1). Large-scale residential housing areas built during the 1980s, but also featuring areas with detached housing and commercial areas, characterize the district’s landscape matrix. The provision with greenspaces in the study area is high as the public greenspaces in the district (1,262 ha, nearly 5% of the district’s surface), are accompanied by abundant privately owned greenspaces inside and outside the apartment blocks that are generally accessible to the public. These formal green spaces often consist of a mix of wooded areas and grasslands, including lawns and meadows, which are subject to high and low frequency of mowing, respectively. Moreover, extensive wastelands emerged due to infrastructure removal at the beginning of this millennium. These wastelands usually undergo only irregular or low management, and often, harbour ruderal grasslands and related types dominated by forbs. In order to reduce costs, most housing companies and public land owners perform a generally low management regime of these grasslands.

Within this context, we chose 30 study sites, well-defined by private and public greenspace property lines, and both inside and outside the apartment blocks. These sites were rather evenly distributed throughout the district and represented a range of open habitats: (1) wastelands with a mix of grassland and herbaceous vegetation; (2) ruderal grasslands; (3) infrequently mowed lawns; or (4) well-kept meadows and lawns. Additionally, some of the wasteland sites were previously subject to a restoration treatment to enhance the native grassland flora (for details, see Fischer et al., 2013; Fischer, von der Lippe & Kowarik, 2013). In the following, we address these types of greenspaces, dominated by grasses and forbs and managed at least irregularly, as grasslands.

Bee sampling

Bees were sampled during two periods, one during early summer (15th and 16th of June) and one in midsummer (30th and 31st of July) in 2012 on sunny days without precipitation and maximum temperatures of at least 20 °C. Pan traps were used to collect the individuals. This is a common and efficient bee sampling method, well suited for all geographical regions and types of habitats (Leong & Thorp, 1999). Pan traps consisted of 15 cm Ø plastic bowls, which were spray painted in UV-bright yellow, white, and blue (Sparvar Leuchtfarbe; Spray-Color GmbH, Merzenich, Germany) and were filled with ∼300 ml of 4% formaldehyde solution as well as a drop of detergent to reduce surface tension. Three pan traps (one of each colour) were randomly positioned at each study site with at least 5 m distance from each other to avoid interference. In order to minimize strata collection bias, pan traps were attached to wood sticks 30 cm above the ground (Cane, Minckley & Kervin, 2001).

After two days, caught bees and bycatch were transferred into vials. Bee specimens were dried and identified to species level with the aid of a binocular microscope and standard identification keys. The European honeybee Apis mellifera was excluded from the list for all analysis since its abundance follows different seasonal patterns than wild bees (Tommasi et al., 2004) and its management in hives complicates the assessment (Cane, 2005).

Environmental variables

To characterize study sites, we ascertained eight environmental variables (Table 1): (1) Area size in km² for each site was measured using the free online tool Google Planimeter (http://www.acme.com/planimeter). (2) To reflect the degree of urbanization we used the ‘UrbanizationScore’ software developed by Seress et al. (2014). This program takes a 1 km² area around each site and generates semi-automated scores based on the method by Czúni, Lipovits & Seress (2012). (3) Since past studies indicate that at least some bee species are affected not only by artificial barriers like buildings but also by natural barriers like forests (Bhattacharya, Primack & Gerwein, 2003; Farwig et al., 2009), the variable isolation was included. We calculated this variable based on the surrounding area of each cardinal direction (north, east, south, west) of the site’s borders. Each of the borders was sorted into one of four categories (Fig. 2) and was given respective points ranging from 0 to 3. The points of all four directions were added up to get a single isolation score for each site with 0 being the lowest, and 12 being the highest possible score. (4) The variable management reflected the mowing regime of a site and was differentiated between extensive (at least annually) and irregular mowing (less than once a year). Data on this variable was gained from respective authorities and additionally evaluated during field work. (5) As represented by the variable restoration, a third of all investigated sites had previously been part of a 5 year in-situ experiment to restore urban grassland vegetation on wasteland sites and thereby enhance the number of native plant species. The exact details of this experiment are given in Fischer et al. (2013) and Fischer, von der Lippe & Kowarik (2013). (6) The variable site type described the human influence on the layout of a site, that is, if and how a site was designed. This classification was based both on site history and the current status. We used the three categories ‘wastelands’ with no known intentional or directed design influence, and where vegetation was able to develop freely (e.g., after the complete destruction of infrastructure, and where the rubble was levelled out); ‘residual’ sites with only low design influence (e.g., vacant lots between houses where traces of past landscape design and plantings were still visible), and intentionally created, ‘designed’ sites (e.g., sites that included flower beds and lawns). (7) Mean degree of coverage with currently flowering plants was recorded during both bee-sampling periods to reflect the short term situation. Each site was put into one of four categories ranging from no visible flowers to maximum observed flowers. (8) The number of plant species was derived from records of the restoration experiment mentioned above (Fischer et al., 2013; Fischer, von der Lippe & Kowarik, 2013) and associated research activity in the other plots.

Statistical analysis

One sample unit includes the pooled data from all three pan traps per site and both sampling periods. To detect groupings of sample sites, the bee community’s response to environmental variables (Table 1) was tested using non-metric multidimensional scaling (NMDS) as multivariate analysis (McCune & Grace, 2002). In order to enhance accuracy of statistical analyses and to reduce the statistical noise, bee species detected only once or twice were omitted from the data set beforehand (McCune & Grace, 2002). We also excluded Apis mellifera, which occurred at all study sites. Thus, 37 bee species, with 1,246 individuals totally, were subjected to NMDS. Relative abundances of these species were standardized performing Wisconsin double standardization and square root transformation. Bray–Curtis dissimilarity matrix of bees was used as base for the ordination. A maximum number of 100 random starts were conducted in search for a stable solution. Environmental variables were fitted onto the ordination with 10,000 permutations to assess significance of correlations between species and environmental data.

To find the distinctive species between restored and unrestored sites as well as among the three categories of urbanization (low = −3.4–−1.27, medium = −1.26–1.67, high = 1.68–3.18), a similarity percentage (SIMPER) analysis was conducted. The SIMPER analysis performs pairwise comparisons of groups of sampling units and finds the average contribution of each species to the average overall Bray–Curtis dissimilarity (Clarke, 1993). The function displays the most important species, which contribute to at least 70% of the differences between groups. The level of significance was P < 0.05.

Finally, individual responses of the distinctive bee species (nine species from SIMPER analyses) to all environmental variables were analyzed using Poisson generalized linear models (GLM). Due to overdispersion, standard errors were corrected using a quasi-Poisson distribution. The most appropriate models were selected by means of analysis of deviance. As a goodness-of-fit measure the pseudo R² based on null and residual deviance was calculated (Zuur et al., 2009). All statistical analyses were performed using the R software environment (version 3.0.1, R Core Team, 2013) including package VEGAN (version 2.0-9, Oksanen et al., 2013).

Structure of bee assemblage

The NMDS plot (stress = 0.19) showed no clear groupings of the study sites (Fig. 3). Urbanization significantly affected species composition (P = 0.037) while all other environmental variables had no effect. Matching this underlying environmental gradient, sites with a higher urbanization value were located more to the right of the plot and likewise sites with a low urbanization value were located more to the left—indicating differences in species distribution among the three urbanization levels.

SIMPER analysis revealed that five species (Bombus lapidarius, B. terrestris, Colletes similis, Dasypoda hirtipes and Halictus subauratus) seem to thrive well in highly urbanized sites while four species (Andrena flavipes, Bombus pascuorum, Lasioglossum malachurum and L. villosulum) appeared to be sensitive to urbanization (Table 2).

Species responses to environmental variables

Results from the GLM indicated that species showed multidirectional responses to environmental variables (Fig. 4). A consistent trend was not recognizable, and environmental variables did not exceed three species responses each. Some of the species showed even contrasting responses but restoration and flower density had overall positive effects. In contrast, area and isolation had a negative impact on tested species. All other variables were related to increasing or decreasing abundances, depending on the species, and even varying within the same genera. However, urbanization, management and plant species richness had more positive than negative effects.

(a) Minor effect of urban matrix variables

Urbanization significantly affected species composition. This can be explained by a species turnover as the SIMPER analysis (Table 2) showed that some species seemed to thrive very well in urban sites (Bombus lapidarius, B. terrestris, Colletes similis, Dasypoda hirtipes and Halictus subauratus) while others negatively responded to urbanization (Andrena flavipes, Bombus pascuorum, Lasioglossum malachurum and L. villosulum). Thus, urbanization did not seem to have an overall negative impact, but it affected urbanization winners and losers in different ways as suggested by previous studies (e.g., Banaszak-Cibicka & Żmihorski, 2012; Hülsmann et al., 2015). Correspondingly, Turrini & Knop (2015) found that urbanization also favors specific species. Community composition may be determined by stochasticity (Sattler et al., 2010), and urbanization impacts depend on geographical, historical and economic factors of cities (McKinney, 2008). Our results and those of recent studies indicate that urbanization can also have positive effects for certain species.

In our study, most of the detected bee species were not affected by any of the environmental variables while few showed multidirectional responses (e.g., Colletes similis, Lasioglossum pauxillum). Matrix variables were found to be important for structuring pollinator communities and explaining patterns of species richness, abundance and biotic interactions in rural landscapes (Steckel et al., 2014; Clough et al., 2014). Matrix variables (urbanization, isolation, area) affected only a few species, and in different ways:

• -

  Urbanization had a negative effect only for one species, Lasioglossum leucozonium, although this species is known to occur in a broad spectrum of rural as well as urban habitat types (Saure, 2005).

• -

  Similarly, isolation negatively affected only Bombus terrestris, which is surprising because this species is known to be one of the most abundant bumblebee species in many rural and urban habitats of Europe (Parmentier et al., 2014; Hülsmann et al., 2015). However, since the goodness-of-fit measure for this model was quite low (R² = 17, Fig. 4), results should not be overestimated. In fact, it appears that isolation plays a minor role for bees that might easily move between their foraging habitats due to their high mobility (Zurbuchen et al., 2010; Krewenka et al., 2011).

• -

  Only two species were affected by area size, being more abundant in smaller areas. Size of green area can have positive effects on bee assemblages (Henning & Ghazoul, 2011) but other studies did not find size effects in cities at least for bumblebees (Gunnarsson & Federsel, 2014). In our study, responses of only two species, Colletes similis and Lasioglossum pauxillum, may therefore be explained by stochasticity or undetected side effects.

(b) Positive effect of restoration

Site variables (site type, management, restoration) affected several species, and restoration had an overall positive effect on species abundances. This is substantiated by the results from the SIMPER analysis, which revealed that 11 species were strongly associated to sites with restoration (Appendix S1). From these results we conclude that restoration can significantly shape bee assemblages in urban environments (see below).

(c) Positive effect of vegetation variables

Flower density and number of plant species had an overall positive effect on bee species as shown in other studies in churchyards and cemeteries (Bates et al., 2011), parks (Matteson, Grace & Minor, 2013) and urban habitats in general (Henning & Ghazoul, 2012; Hülsmann et al., 2015).

Restoration was previously determined to have positive effects on bee species in rural areas, e.g., due to increased plant diversity (Albrecht et al., 2007). Our study demonstrates that restoration of grassland on urban wasteland sites increased the abundance of wild bees in urban settings as well. Regarding the vegetation of the sites, the restoration treatments increased plant species richness by adding a considerable pool of native dry-grassland species (Fischer, von der Lippe & Kowarik, 2013) and resulted in an increased floral diversity (Senße, 2013). This is in line with a study that revealed the positive effect of increased floral diversity for bumblebee and hoverfly abundances in amenity grassland (Blackmore & Goulson, 2014) and findings of Wastian, Unterweger & Betz (2016). Whereas Hanley, Awbi & Franco (2014) showed for urban gardens that bumblebees have no dietary preference for native or non-native plants, there is also evidence from highly disturbed sites (roadverges), where restoration of the native vegetation especially resulted in increased bee richness (Hopwood, 2008).