Structure of a metacommunity of urban bees Species diversity and spatio-temporal modularity

mellifera (32 %), were sampled monthly in pan traps. ( i ) Information about species traits was extracted from the literature, and trait values were correlated and used to characterize the fauna. Most were soil-nesters, pollen generalists, and common. ( ii ) Habitat diversity within five concentric circles with trap at the centre and radii from 50 m to 1000 m was related to bee α diversity. The relationship was significant only within 1,000 m for all bees and for bumblebees. Solitary bee diversity was uncorrelated with habitat diversity at all spatial levels. ( iii ) Spatio-temporal structure was analysed as two networks, one for bees linked to sites, and one for bees linked to months. Link patterns were analysed for levels of nestedness, modularity, and spatio-temporal β diversity. The two networks were weakly and non-significantly nested, but strongly modular, being composed of five and four modules of co-occurring bees, respectively. ( iv ) Finally, we studied total β diversity, β TOTAL , being the sum of species turnover, β TURN , and species loss/gain or nestedness, β NEST . For both site and season, β TURN was higher than β NEST , and site β TOTAL was higher than season β TOTAL . One reason for this metacommunity structure may be a high spatio-temporal habitat patchiness, sustaining a rich biodiversity. Thus, a few large areas may not compensate for the loss of several small patches. Consequently, establishment of many green, even small habitats is recommended.


Introduction
Growth of urban areas is certainly one of the most severe encroachments of nature, causing a disappearance of many species and their biotic interactions (Seto et al., 2012).This loss and change in species composition due to urbanization, in general, are now well understood.However, the effects of urbanisation on biodiversity are complex because urban areas are highly heterogenous in ways that still are poorly understood (Fauviau et al., 2022).Recent reviews even suggest urban areas should be regarded as novel ecosystems and thus considered as specific habitats and not just degraded versions of former natural habitats (Hall et al., 2017;Silva et al., 2023).
We focus here on bees (Hymenoptera, Apoidea), a well-studied and diverse group of insects that provide essential pollination services to wild plants and crops (IPBES, 2016;Klein et al., 2007;Ollerton et al., 2011).The literature presents many studies of bee communities in urban areas, but there is no strong consensus about which generalizations to be expected (Hernandez et al., 2009).One reason is that urban areas represent a highly variable patchy mosaic of habitats, ranging from concrete deserts to diverse green spaces (Baldock, 2020).
Here, we investigated how bees, as an ecological guild, respond to urban spatio-temporal patchiness; in particular, habitat diversity and seasonal change, which both affect flower richness and nesting site availability, two of the most important drivers of bee diversity (e.g.Kleijn & van Langevelde, 2006;Steffan-Dewenter et al., 2002).
As pollinators of many plants, bees are a key element of both natural habitats and urban areas (Baldock et al., 2015).However, they constitute a very heterogeneous group, being not just rich in species, but also in their species-specific ecologies, such as body size, diet specialisation, sociality, nesting requirements, and foraging range (Baldock et al., 2015;Gathmann & Tscharntke, 2002;Greenleaf et al., 2007;Westphal et al., 2006;Zurbuchen et al., 2010aZurbuchen et al., , 2010b)).At our study sites in a North European metropolitan region, bees also make up a diverse community, including solitary species, bumblebees and their kleptoparasites, in addition to the ubiquitous honeybee.These groups and their species members are all expected to respond to urban patchiness according to their own specific ecology, e.g.bumblebees are social, long-range flying, polylectic (pollen generalists), and active most of the season.Consequently, they perceive the landscape and season more coarse-grained than solitary bees like Lasioglossum species, which are short-range flying, oligolectic (pollen specialists), and active during short periods (Cane, 2021).For a North European city, we analysed (i) the bee metacommunity in both space and time, focusing upon its α diversity and trait correlations (Leibold et al., 2004), (ii) the relationship between habitat diversity and α diversity of the metacommunity, (iii) the spatio-temporal network structure, i.e. the spatial bee species-site network and the temporal bee species-month network and their link patterns, especially levels of nestedness and modularity, and finally, (iv) the structure of the bee β diversity.

Study sites
The bee fauna was sampled at 13 sites within the city of Aarhus, Denmark (Fig. 1, Appendix A: Material 1).In 2016, the year of field work, the city covered an area of 468 km 2 with c. 331.000 inhabitants (Aarhus i tal, 2022).Sites were chosen according to three criteria: 1st-Geography: We chose sites along a gradient line crossing the city from suburban sites in north-east, through the city centre, to suburban sites in south-west (Fig. 1; for geographical coordinates and definitions of 'City' and 'Suburban', see Appendix A: Material 1).2nd-Between-site distance: Since maximum flight range of most bees is < 1 km (Darvill et al., 2004;Gathmann & Tscharntke, 2002;Zurbuchen et al., 2010aZurbuchen et al., , 2010b)), we chose sites separated by ≥ 2 km to reduce spatial autocorrelation.However, many solitary bees forage over shorter distances (Greenleaf et al., 2007).3rd-Diversity of habitat types: Sites were selected to include a highly diverse cityscape of up to ten habitat types (Table 1, Appendix A: Material 5).To some extent, categorization of the city into these habitat types was subjective.However, we assumed bees to respond differently to this level of habitat differentiation with respect to availability of food sources and nesting site opportunities (see also the paragraph Habitat diversity and bee species α diversity below).

Traps
At each site, bees were sampled using pan traps, i.e. one trap per site.Each trap consisted of three semi-spherical bowls, a white, blue, and yellow-coloured one (radius 6.5 cm).This tricoloured palette attracts a wide range of bees (Campbell & Hanula, 2007;Leong & Thorp, 1999), but a few taxa and body size classes may ignore such traps (pers.obs.).Each bowl contained 250 ml of a 1:10 aqueous solution of Rodalon©.The latter was added to conserve trapped insects and break water surface tension and thus reduce any escape of bees.Bowls were placed on a metal plate mounted on a wooden stake in an open place at level with surrounding vegetation.At each site, traps were positioned near to the most abundant type of vegetation, preferably close to flowering plants.However, owners of private gardens decided the exact location of the trap.In public areas, traps were placed at the edge of lawns.At individual sites, attraction to trap colours might be biased by floral colour of local flowers.However, this was not considered here.During the study period of six months, traps at all 13 sites were sampled monthly.Traps were installed for a 5-day continuous period once a month and regularly spaced throughout the season: viz.[16][17][18][19][20][21][15][16][17][18][19][20][17][18][19][20][21][22][16][17][18][19][20][21][16][17][18][19][20][21] August, and 14-18 September; deviations were due to extended periods of rain.Thus, each monthly 5-day period includes one sample.i.e. adding up to 78 trap samples, which were stored in 70 % ethanol.Later, we discarded non-bee insects and subsequently identified all bees to species (Hymenoptera, Apoidea).However, as Bombus cryptarum, B. lucorum, B. magnus, and B. terrestris are difficult to distinguish, they were lumped into a 'Bombus terrestris species complex' (Williams et al., 2012).Bombus magnus and B. cryptarum may be ignored as options because they are unknown to the region (Rasmussen et al., 2016).Number of months a species was observed to be active was termed its phenophase length.If a species was observed, for example, in April and June, it was assumed to be active in May as well.
These traits define the life history of most bee species (Westrich, 1990).We used Cramer's V to get a measure of strength of association between any two nominal variables: '0′, negative association between variables, and '1′, positive association (R function cramersv(x) in R (R Core Team, 2023)).Red list status was not included in this part of the analysis, because all wild bees were of Least Concern (LC) except Andrena fulvida (NT, Near Threatened).

Habitat diversity and bee species α diversity
At each site, we estimated habitat type diversity within a set of concentric circles with trap at centre and radii of 50 m, 100 m, 200 m, 500 m, and 1000 m, i.e. the 'landscape dimension' or buffer zones around each trap (Levé et al., 2019).This was done in ArcGIS Pro 2.5.1.Habitat type classification followed Basemap02 (Levin et al., 2017), i.e. a raster-based GIS-layer, where each cell was covered by habitat types: (1) urban/ residential area, (2) forest/ park, (3) industrial area, (4) roads, (5) railway, (6) recreational area, (7) farmland, (8) natural vegetation, (9) water bodies, and (10) others.For each circle and site, the area of each habitat type was calculated by summing up the area of the cells of a particular type within the circle.Input data were from May-September 2016, except farmland area data (habitat type 7), which were from 2011.
Using the Shannon-Wiener diversity function (Good, 1953), -Σp i ln p i , we analysed, if habitat type diversity and bee species diversity per site and circle size, i.e. bee α diversity, were correlated.p i was the proportion of area of each habitat type i at a site and circle size, or the relative abundance of a bee species i in the sample at a site.In empirical diversity studies, this function ranges from 1.5 to 3.5 (Gaines et al., 1999).

Nestedness and modularity of bee-site and bee-month networks
Based on species occurrence data, bee-site and bee-month networks were created and visualized as matrices.A link was given as either present ('1′) or absent ('0′) (qualitative data) or as its strength (quantitative data), which here was absolute species abundance at a site or month (Appendix A: Materials 1, 2).Connectance, defined as the percent observed bee species presences out of the total potential presences, i.e. c = 100 x number of observed links/ (number of bee species x number of sites/months), was calculated for the two matrices.Connectance c indicates, how well connected the networks are, i.e. how generalized species are and how species-rich sites/months are.
In ecological network studies, typically two link patterns are analysed: nestedness (Almeida-Neto et al., 2008;Bascompte et al., 2003;Dupont et al., 2003) and modularity (Newman, 2003;Olesen, 2022;Olesen et al., 2007).Most networks show both patterns, but to a varying degree.Thus, the relative dominance of these two link patterns gives a description of the spatio-temporal heterogeneity of the urban bee metacommunity.
We calculated levels of nestedness and modularity for both the beesite and the bee-month network.Several indices of nestedness level are available.We chose NODF2, because it corrects for variation in matrix size and connectance and is recommended for comparisons across different networks (Almeida-Neto et al., 2008).If NODF2 → 0 or 1, the network becomes increasingly non-nested or nested, respectively.NODF2 was calculated using qualitative link data (R library bipartite).Level of modularity Q was calculated using quantitative/weighted link data with algorithm DIRTLPAwb+ (Beckett, 2016).To test the significance of the observed link patterns, 100 random networks were generated, keeping the observed interaction probabilities between a bee and a site/month fixed.Finally, we evaluated if metrics fell within the 0.95 confidence interval estimated from the randomizations.If significant, standardized NODF2' and Q were also calculated as z = (observed metric mean of random runs)/ SD of random runs.

Spatial and temporal β diversity of the bee metacommunity
We calculated the spatial and temporal β diversities of the bee metacommunity, i.e. a landscape or regional β diversity of bees across all 13 study sites and a temporal β diversity across all six months.As measure, we used total β diversity between pairs of sites or months, where a is number of shared species between two sites or months, b is number of unique species at one site or month, and c is number of unique species at the other site or month.In Appendix A: Material 6, we give the formula for multiple-sites and multiple months dissimilarity according to Baselga (2010).
β TOTAL can be broken down into two components, describing in detail the metacommunity's structure and dynamics, which include information with strong implications for management and planning of urban biodiversity (Baselga & Orme, 2012;Carstensen et al., 2014).Relative strength of the two components can be disentangled and compared using the analysis developed by Balsega and Orme (2012).The two components of β TOTAL are (1) turnover of species β TURN between one site or month and another site or month, and (2) nestedness β NEST , i.e. species loss and gain at a site or month compared to the other sites or months.
Thus, β TOTAL = β TURN + β NEST (see later Fig. 8).β TURN measures the extent to which sites or months have a unique bee diversity, and β NEST measures the extent to which poorer species assemblages are subsets of richer ones.Species turnover (β TURN ) favours modularity, whereas species loss/gain favours nestedness (β NEST ).We obtained estimates of all three indices, using R library betapart (Baselga & Orme, 2012).If β TOTAL → 0, sites/months become more similar in species composition; if β TOTAL → 1, sites/months become more dissimilar.For both bee-sites and bee-months, 100 randomizations of each network were used to generate frequency distributions of multi-site and multi-month dissimilarity measures.To test if actual indices deviated from random expectations, they were compared to mean indices and standard deviations of random runs.

Results
The urban bee metacommunity in space and time A total of 313 individuals of bees belonging to 40 species were collected, viz.124 solitary bees belonging to 29 species, 89 bumblebees (Bombus spp.) belonging to ten species, and 100 honeybees (Apis mellifera) (Fig. 2A, Appendix A: Material 1).
Pooling data from the entire season, frequency distributions of number of bee species and individuals per site turned out to be leftskewed (Fig. 3A, B, Appendix A: Material 1), whereas the frequency distributions for monthly counts were approximately normal (Fig. 3C, D; Appendix A: Material 2).Number of species per site ranged from three at site L (urban/ residential) to 17 at site M (railway, industrial area), and number of individuals ranged from only three at site L to 57 at site H (roads, urban/ residential) (Fig. 3A, B, Appendix A: Material 1).Monthly species number ranged from two in September to 23 in July, and number of individuals ranged from 13 in September to 97 in July (Fig. 3C, D; Appendix A: Material 2).
On average, a bee species was present on 2.9 ± 2.78 sites (mean ± standard deviation) with Bombus and Halictus species being most widespread and Andrena species being the most local (Appendix A: Material 1).Fourteen or about half of all solitary bees were only sampled at one site.Mean phenophase length was 2.2 ± 1.57 months (Appendix A: Material 2).Andrena species richness peaked in April-May with a mean species phenophase of 1.2 ± 0.40 months, and Lasioglossum species richness peaked in July with 2.6 ± 1.40 months, whereas bumblebees were almost omnipresent with a phenophase of 3.2 ± 1.87 months.
Species richness and total species abundance were significantly correlated (Fig. 4).Site H, however, was an outlier, because of relatively high abundances of Apis mellifera and Andrena haemorrhoa.
The α or Shannon-Wiener diversity index at a site ranged from 1.10 at site L to 2.42 at site B (Table 1).

Ecological traits of the urban bees
Data on the distribution of ecological traits of the collected species are found in Fig. 5  were present at fewer sites than solitary polylectics (1.8 vs. 2.5 sites; t = 31.4,P < 0.001).Here, 'lectic' only includes pollen consumption.C. Activity period-Most species were either active in mid-late season (18 species) or earlier (13 species) (Fig. 5C).D. Number of generations-Almost all (37) species were univoltine, except Andrena bicolor, A. flavipes, and A. minutula.Social species were regarded as univoltine (Fig. 5D).E. Red list status IUCN-Only Andrena fulvida was red-listed in the nearthreatened (NT) category, whereas all other species were of least concern (LC).Apis mellifera was not evaluated in the list (NA) (Fig. 5E).
Most traits (including number of study sites but excluding Red list status) were not significantly associated, except (i) nesting site-activity period, (ii) number of study sites (proxy of regional species distribution)-activity period, and (iii) number of study sites-number of generations (Appendix B: Supplemental Fig. 1).

Habitat diversity and bee species α diversity
All sites and their surroundings formed a complex mosaic of habitat types (Appendix A: Material 5).We analysed, if habitat diversity at each of the 13 sites and landscape dimension (50 m to 1000 m) was correlated with bee α diversity at sites (Fig. 6).Total bee, wild bee (total bee, excluding honeybee), and Bombus α diversities were all positively correlated with habitat type diversity within the largest radius of 1000 m, but never within any smaller circle.Solitary bee α diversity was not correlated with habitat diversity at all.These findings are summarized in Appendix C (Supplemental Fig. 2), where the F-statistics of the correlation analyses are plotted against landscape dimension.Thus, only at the largest landscape dimension (1000 m) did we see any effect of habitat diversity on bee α diversity.

Nestedness and modularity of bee-site and bee-month networks
The link pattern of the 2-mode bee-site/month networks is shown in Fig. 7.The 40 urban bee species and their 13 sites with 115 links gave the spatial network a connectance c sites of 22 %, and the 40 bee species and the six months with 83 links gave the temporal network a higher c months of 35 %.
Both networks were significantly modular, but not nested Table 2.In fact, they were even anti-nested, i.e. their NODF2-values were significantly lower than the mean NODF2 of the null model runs.The bee-site and bee-month networks had five and four modules, respectively Fig. 7.The strongest hub or most generalized bee species in each of the five sitemodules were Apis mellifera, Lasioglossum morio, L. leucopus, Halictus tumulorum, and Bombus terrestris complex.The strongest in each of the five month-modules were Apis mellifera, Bombus terrestris complex, B. lapidarius, and B. hortorum.Modularity level Q was slightly higher in the bee-month network than in the bee-site network, which also was reflected in the distribution of links within and between modules.For the bee-site network, number of links between and within modules was L b = 55 and L w = 60, respectively, i.e.L b / L w = 0.92.For the bee-month network, L b = 38 and L w = 45, respectively, i.e.L b / L w = 0.84.Thus, season made the bee metacommunity denser (higher c) and more modular or heterogeneous (lower L b / L w ) than space.
Early (June) and late summer (August and September) constituted one module, whereas each of the other modules encompassed only a single month.Andrena spp.were only present in the April and May modules (except the later A. fulvida), and most Lasioglossum spp.were found in the July module.Every site-module had representatives from three to four month-modules, i.e. each site-module had a monthly sequence of species, and each month-module had its characteristic bee fauna.

Regional and seasonal bee β diversity
Site and month bee β diversities are given in Table 3. Site species turnover (β TURN ) contributed much more to total β diversity than species loss/gain (β NEST ) (β TURN: β NEST = 11.6;Fig. 8).Similarly, seasonal species turnover (β TURN ) contributed more to total β diversity than species loss/gain (β NEST ) (β TURN : β NEST = 8.8).Temporal turnover was slightly less, perhaps because turnover here took place among solitary bee genera and not species (Appendix A: Material 2).Hence, total β diversity of the urban bee metacommunity in the city of Aarhus was characterized by a high turnover of species among both sites and months.
Honeybee and bumblebees had a wide range (mean 4.5 sites/species) and long phenophase (mean 3.3 months/species), connecting sites and months.In this way, they increased nestedness (β NEST ) and reduced modularity (β TURN ).Solitary bees had the opposite effects upon network structure.They had a short range (mean 2.2 sites/species) and a short phenophase (mean 1.7 months/species), disconnecting sites and months (Appendix A: Materials 1 and 2).In this way, they decreased nestedness (β NEST ) and increased modularity (β TURN ).

Discussion
Studies of urban pollinators are carried out around the world, particularly in Western Europe and North America, but also in the Tropics, and several of these focus upon bees (Silva et al., 2023), e.g. in UK and France (Baldock et al., 2015;Fauviau et al., 2022;Sirohi et al., 2022;Tassin de Montaigu & Goulson, 2024).temperate, urban areas, Apis and Lasioglossum are among the most frequently recorded bee genera (Silva et al., 2023).In addition to these, Andrena was the most species-rich genus and the most abundant genus of solitary bees in the current study.
The high abundance of solitary bee species observed by us (40 % of all individuals) is remarkable, being seven times higher than in a comparable study of bees in British cities (6 %) (Baldock et al., 2015), although bee faunas of the two countries are similar in size (296 for Denmark and more than 270 for Britain (Falk, 2019;Rasmussen et al., 2016).These national counts may be methodologically flawed, because the pan trap method used in the current study is reported to have a better coverage and generally detects higher bee species richness than transect walks (e.g.Westphal et al., 2008), which was the sampling method used in the British study.However, another study of bee-plant networks in a larger British town and nearby nature reserves (Sirohi et al., 2022) observed 29 non-parasitic solitary bee species along transects compared to 27 in Aarhus; 13 species overlapped between the two studies.Therefore, spatial variation in solitary bee fauna may either reflect different sampling methods or if real, be dependent upon factors ignored in our studies, e.g.density of small green areas with a unique flora intermingled with open spaces for nesting.
In our study, most urban bee species were polylectic (pollen generalists), ground-nesting, univoltine and relatively common, being active either early or mid-late in the season.This is in accordance with the ecology of the national bee fauna (Rasmussen et al., 2016).Baldock et al. (2015Baldock et al. ( , 2019) also observed a predominance of pollen generalist species.Some studies find more cavity-nesting bees, which they suggest is due to the high tarmacked surface cover, being an obstacle to soil-nesting species (Cane et al., 2006;Fauviau et al., 2022;Hernandez et al., 2009).In conclusion, the main ecological and perhaps taxonomical composition of the bee fauna in our study seems to be a random sample of the national fauna, although bumblebee species are relative more common in the city.
Urban expansion is a major driver of biodiversity loss (e.g.-Cibicka, 2012;Bates et al., 2011;Deguines et al., 2012;Hernandez et al., 2009).However, moderate levels of urbanisation, particularly of certain urban habitat types, such as small green areas, may support higher bee diversity compared to agricultural and even natural areas (Baldock, 2020;Baldock et al. 2015;Fauviau et al., 2022;Hall et al., 2017;Theodorou et al., 2017).The explanation for this may be, that urban areas can be highly heterogeneous, composed of a very fine-grained mosaic of habitat types within short distance of each other, including high quality bee habitats (Baldock, 2020).The latter might be private gardens, public parks, allotments, and cemeteries, which all are considered important habitats for bees, because of their rich flora and nesting sites (Baldock, 2020;Baldock et al., 2019;Goddard et al., 2010;Silva et al., 2023).In the current study, we showed that within the urban area of the medium-sized city of Aarhus in Denmark, total bee α diversity was not correlated with habitat diversity, at least at smaller spatial scales (<1000 m), and the included habitat types, except at the largest spatial scale (1000 m radius) for bumblebee α diversity and thus also for total bee α diversity, because bumblebee species constituted a large proportion of the total diversity.Solitary bee α diversity did not correlate with habitat diversity at any of the five spatial scale levels.This was surprising, because most solitary bees forage within a few hundred metres of their nest (Gathmann & Tscharntke, 2002;Hofmann et al., Fig. 7. Visualization of 2-mode networks between bee species and sites and bee species and months.Size of each species or site/month box is proportional to its number of links.Nodes of similar colour belong to the same module.

Table 2
Nestedness and modularity of bee-site and bee-month networks and comparisons with null model.100 runs per null model.c.i., confidence interval.z = (Q -Q 0 )/ standard deviation of null model runs.

2020
).However, our results are consistent with Tassin de Montaigu and Goulson (2024), who found no effect of habitat quality in the immediate surroundings (<100 m), and Ahrne et al. (2009), who found the strongest correlation between bumblebee diversity and proportion of suitable area within 1000 m in an urban landscape in the UK and Sweden, respectively.However, bumblebee and butterfly α diversity is known to be affected by garden quality in UK (Tassin de Montaigu & Goulson, 2024).
Our findings suggest that bee diversity might be controlled by other factors, e.g.variation in ambient temperature, impervious surface area (Fauviau et al., 2022), and last, but not least, the flora of individual sites (Sirohi et al., 2022).The latter may be an important driver of spatial turnover because the highest bee diversity was observed in industrial areas, which were rich in weeds and other ruderal plants.Unfortunately, we do not have any quantitative data to support this in the current study.However, the distribution of solitary bees among sites suggests that each site supported specific bees, because almost half of all solitary bee species (14 of 29 spp.) were unique to single sites.Several studies find that specialist bees thrive in cities, e.g.Casanelles-Abella et al. (2021), da Rocha-Filho et al. (2021), and Fauviau et al. (2022).Sirohi et al. (2022) suggests that a high diversity of urban solitary bees is due to the many wildflowers, supporting oligolectic species (pollen specialists).In our study, diet width might restrict the dispersion of oligolectic bees, because these were present at fewer sites than polylectic species.
Although the current study did not find a simple relationship between habitat diversity of the immediate surroundings and bee α diversity, the analysis of β diversity revealed a strong spatial and temporal structuring of bee diversity, being far from random.The metacommunity of urban bees was organized into distinct sub-groups or modules in both space and time.Spatial modules consisted of groups of sites with more similar bee faunas compared to other sites in the city, while temporal modules consisted of bees with similar phenologies.
Thus, each site and month contributed to the overall city α diversity, i.e.
spatial and temporal modules were complementary.The decomposition of bee β diversity among sites and months showed in detail that the diversity in our study city was driven by spatiotemporal species replacement (the turnover or modular component), especially of solitary bees, and that, generally, species gain and loss (the nestedness component) only played a minor role.A high turnover of species diversity was also observed by Tew et al. (2022), but not in Banaszak-Cibicka and Żmihorski (2020).Our metacommunity had a weak nestedness, mainly driven by the few, very abundant, widespread generalist species Apis mellifera and Bombus terrestris complex, both being present at all sites and in six and five months, respectively.These two species (groups) have long foraging ranges (Apis mellifera, e.g.Couvillon et al., 2014;Bombus terrestris, e.g. Osborne et al., 2008), potentially linking different sites within the urban area.
Thus, an increasing number of green patches in a city adds to bee α diversity, as each patch and time within the season are expected to support at least some unique species; this being supported by our finding of high turnover values.Thus, the species ratio between widespread social and range-limited solitary bees has a strong impact upon the dynamic interplay between modularity and nestedness and can be managed and manipulated in the planning of density, size, and bee habitat quality of green urban areas.
Finally, as an overall message to conservation managers, we conclude that the high spatial turnover among sites suggest, that in terms of regional or β biodiversity, particularly the contribution from the solitary bee fauna, a few large areas may not compensate for the loss of several small patches (however, see Stewart et al., 2018).Thus, the establishment and protection of many green habitats with a long bee-friendly season and within flight distance from each other is recommended.Y.L. Dupont et al.

Fig. 1 .
Fig. 1.Study sites in the city of Aarhus, Denmark-labelled A to M. Urban areas are greyish in colour, open land (mostly farmland) is white, forests are light-green, and water bodies are blue (map modified from 'Aarhus i tal 2022').Latitude and longitude of sites are included in Appendix A: Material 1.

Fig.
Fig.Correlation between bee species richness per site and total bee species abundance per site.

Fig. 8 .
Fig. 8. Modular (A, C) and nested (B, D) versions of the 2-mode bee-site (A, B)/ months (C, D) networks.In (A) and (C), cell colours are like node colours in Fig. 7. Grey cells are links between modules.

Table 1
Species richness and total absolute abundance of bees at each site.Sites are sorted according to Shannon-Wiener diversity index.
Y.L.Dupont et al.

Table 3 β
diversity of bee-site and bee-month networks and comparisons with null model.100 runs per null model.c.i., confidence interval.