THE IMPORTANCE OF WILD BEE COMMUNITIES AS URBAN POLLINATORS AND THE INFLUENCE OF HONEYBEE HIVE DENSITY ON WILD BEE CROP VISITATION RATES

While urban beekeeping is on the rise, data on the role of wild bee communities as crop pollinators in cities is still scarce. We analysed wild bee visitation rates on apple, plum, cherry, pear, blackberry, raspberry, and strawberry in a Bavarian city with a very high honeybee density of c. 19 hives/km. During 137.5 hours of observation time, we observed 52 wild bee species on the studied crop plants. During more than 50 h of observation time on fruit trees in flower, we found that wild bees provided 41% of the total bee visits, honeybees the remaining 59%. Honeybee hive density had a significantly negative effect on wild bee abundance. Bumblebees appeared more tolerant to poor weather conditions than all other bee groups. Wild bee species richness on apple flowers was not significantly impacted by flower diversity in the surroundings of the trees. Together, our results suggest that species-rich wild bee communities in urban areas are important for pollination success in common fruit crops, especially under unstable spring weather conditions. Bee-friendly management of urban spaces should be prioritised to support wild bee communities as well as the increasing number of honeybees in cities. In order to reduce competition with endangered wild bees, the bee keepers should always consider the available floral resources in their surroundings from early spring to late autumn and adapt their number of hives accordingly. Keywords—Biodiversity, urban beekeeping, agroecology, apple pollination, wild bees, bee pollination


INTRODUCTION
Recent studies highlight the importance of urban agriculture for food security (Edmondson et al. 2020), positive effects of urban agriculture on biodiversity in general (Lin et al. 2015), and on bee species diversity in particular (Normandin et al. 2017;Baldock et al. 2019;Lanner et al. 2020). At the same time, a large share of crops -including many of the species popular in urban gardens like pumpkin, raspberry and apple -depend on biotic pollination. Studies quantifying the relevance of animal pollination at the global scale showed that 85% of all flowering plants, c. 76% of the leading global food crops and 35% of global food production depend on animal pollination (Ollerton et al. 2011;Klein et al. 2007).
Despite increasing interest in bee diversity and conservation potential of urban habitats (Hernandez et al. 2009;Baldock et al. 2015;Hall et al. 2017;Baldock 2020;Krahner & Greil 2020) much less attention has been directed towards pollination services in urban environments. This is somewhat worrying since e.g., a survey of community gardens in New York City found that 92% of the crops require bee pollination to some degree to set fruit or seed (Matteson & Langellotto 2009). Studies comparing pollination efficiency in urban, natural and agricultural habitats found that fruit and seed set of herbaceous plants (Cussans et al. 2010;Theodorou et al. 2020) and wild bee flower visitation rates and foraging activity (Leong et al. 2014;Theodorou et al. 2020;Kaluza et al. 2016) were highest in urban environments. Spatial modelling approaches assessing pollinator supply and demand across Iowa City found that pollinator supply meets demand only in 72% of the city (Zhao et al. 2019). In one of the rare studies on urban food crop pollinators, Lowenstein et al. (2015) found cucumber (Cucumis sativus), eggplant (Solanum melongena) and purple coneflower (Echinacea purpurea) in gardens in Chicago to be visited by different groups of pollinators with honeybees constituting less than 7% of visits, and fruit and seed set to increase with pollinator visitation and taxonomic richness.
Extensive research has been investigating potential drivers of species richness and abundance of flower visiting communities in agricultural and natural systems and found that local plant diversity and (semi-)natural habitat support species richness and abundance of crop flower visitors (Ricketts et al. 2008;Kremen et al. 2004;Martins et al. 2015;Nayak et al. 2015;Kammerer et al. 2016;Joshi et al. 2016;Motzke et al. 2016;Fisher et al. 2017;Alomar et al. 2018;Ganser et al. 2018;Saunders & Luck 2018;Fijen et al. 2019). Fewer studies focus on urban drivers of plant-pollinator interactions (Harrison & Winfree 2015) and their results indicate some divergences from non-urban settings: while visitation rates on flowering plants in Toledo (USA) declined with increasing impervious surface and increased with floral resource availability (Burdine & McCluney 2019a), a negative correlation to the amount of green area and a positive correlation to human population density have been identified in studies in Leuven (Verboven et al. 2014) and Chicago (Lowenstein et al. 2014), respectively.
Studies assessing the impact of honeybees on wild bees in agricultural and natural settings found evidence for negative effects of rising honeybee hive densities on wild bee communities and visitation rates through competition for food resources (Torné-Noguera et al. 2016;Dupont et al. 2004;Lindström et al. 2016;Mallinger et al. 2017;Geldmann & González-Varo 2018;Geslin et al. 2017). With urban beekeeping being on the rise, competition from increased numbers of honey bee hives has been identified as one of the main threats for urban pollinator conservation (Baldock 2020). While no significant effects of honeybee hive density on wild bee species richness were detected on study sites in Montréal with densities below 0.5 hives/km 2 (McCune et al. 2020), the only urbanbased studies to our knowledge specifically investigating the impact of honeybee colony density on plant-pollinator interactions identified negative effects on wild pollinator visitation rates at average hive densities of 6.5 hives/km 2 in Paris (Ropars et al. 2019) and 16 to 22 hives/km 2 in Munich (Renner et al. 2021). In their attempt to disentangle the "modern gordian knot of urban beekeeping", Egerer & Kowarik (2020) identify challenges in balancing potential risks of urban beekeeping (e.g. floral resource competition, parasite and pathogen transmission, and changes in wild flora community composition due to changing plant-pollinator mutualisms) with the benefits of its contribution to pro-environmental behaviour, and call for scientific research on the matter oriented towards science-city partnerships for pollinator-friendly cities.
Our study focuses on five common fruit crops of the rose family in meadow orchards and gardens in an urban setting in Southern Bavaria. We analysed the flower visitor community composition of different crop species to determine the importance of wild bee pollinators, especially in comparison to honey bees. We tried to account for confounding factors like abundance of flowering plants in the surroundings and the impact of weather conditions. We hypothesized that 1) a large proportion of the flower visits is performed by wild bees; 2) urban environments have a negative effect on wild bee abundance; 3) honeybee-hive density has a negative effect on wild bee abundance; 4) wild bees with bigger body sizes are less sensitive to low temperatures, higher humidity and wind. J Poll Ecol 29(16)

STUDY SITES
The study was carried out in the city of Freising in southern Bavaria, Germany, which has a population of about 50,000 inhabitants and a total area of c. 89 km 2 (www.kreis-freising.de). Freising is situated in the lower Isar valley at an altitude of c. 460 m asl. The climate is temperate with annual rainfall of 807.32 mm and temperatures ranging from -15.2 to +33.7 (long-term average based on the values of the years 2010 to 2019, www.wetterby.de). A total of 517 wild bee species have been recorded in Bavaria, and c. 300 species are estimated to exist in the administrative district of Freising (Scheuchl et al. 2018;Bayerisches Staatsministerium für Landesentwicklung und Umweltfragen 2001). Pollinator surveys on fruit trees were performed in five fruit orchards distributed all over the city (Figure 1, sites A-E; A-C).
The surface area of the orchards varies between 0.8 -4.25 ha, the number of fruit trees per orchard ranges between 50 -216 trees of ten different species, the age of the trees ranges from 10 to more than 50 years (Table 1). The minimum distance between two orchards is 350 m, the maximum distance 2,500 m. The orchards have been subject to different management types and intensity ranging from fallow to grazing, mowing, and multiple mulching per season (Table 1, A-C). Additional observations on berry patches (Rubus, Fragaria) were performed in 23 private gardens, community gardens and fallow land sites throughout the city (Figure 1, red and green dots; D-H).

STUDY SPECIES
We selected nine species from the rose family: plum (Prunus domestica L.), cherry plum (Prunus cerasifera Ehrh.), sweet cherry (Prunus avium L.), sour cherry (Prunus cerasus L.), pear (Pyrus communis L.), apple (Malus domestica Borkh.), strawberry (Fragaria x ananassa Duchesne), raspberry (Rubus idaeus L.) and blackberry (Rubus sect. Rubus, mostly Rubus armeniacus Focke). These fruit crops are common and widespread for food production in urban environments and their simple flower morphology makes pollen and nectar available to a broad range of insects. In our study orchards, this selection covered all present fruit trees except for three rare Rosaceae (Cydonia oblonga Mill., Sorbus domestica L. and Mespilus germanica L.), and the wind-pollinated walnut, Juglans regia L. (Table 1). From early spring (cherry and plum) to late summer (blackberry), our study species spanned the entire flowering season. The selected crops vary in different cultivars from selfincompatible to self-compatible, but fruit-set and fruit quality of all these species are increased through insect-pollination (Abrol 2015).

POLLINATOR SURVEY
Our systematic observation periods sum up to a total observation time of 137 hours and 30 minutes, with 101 h and 40 min dedicated to the trees, the remaining time spent on the berry patches (Table 2).
Pollinator surveys were performed between 9 AM and 6 PM during sunny weather with temperatures over 12°C on 19 days between April 16 and May 4, 2018, during full bloom of the fruit trees. Flower visitors were photographically recorded within an area of approximatively 4 m 2 during a standardized observation period of 10 minutes with the aim to at least differentiate between the four categories honeybee, bumblebee, large solitary bee (body size > 9 mm, which means larger than honeybee workers), and small solitary bee (body size up to 9 mm) and enable identification to genus or species level whenever possible. Only flower visitors touching stamen or  Pollinator surveys of berries were performed on 32 days between May 3rd and August 6, 2018. Citizen scientists, who had been trained on guided walks and through personal visits, contributed observations for 24 standardised observation periods of 10 minutes. Because of the lower number of flowers per site in the berry patches compared to the orchard trees, it was not possible to stick to a standardized area of 4 m 2 . Therefore, each bee individual of the same species and sex was only counted once per observation period.
Bees were identified as far as possible to species level based on our field identification guide (Weissmann & Schaefer 2020). Species that are not distinguishable in the field (e.g., Andrena minutula, A. minutuloides, A. falsifica, A. strohmella and the very rare A. subopaca, A. alfkenella, A. nana, A. semilaevis, A. rugulosa, A. floricola, A. niveata, A. pusilla, A. nanula, A. saxonica, A. anthrisci) were treated as species groups with reference to the most common species (e.g., Andrena cf. minutula). For each species, information on body size and nesting ecology was compiled from Amiet et al. (1999), Amiet et al. (2001), Amiet et al. (2004), Amiet et al. (2007), Amiet et al. (2010) and Westrich (2018). We differentiated between ground nesting solitary bees (nesting in self-dug cavities in the soil), cavity nesting solitary bees (nesting in selfdug or pre-existing holes in dead wood, plant stems, walls, crevices, snail shells), and bumblebees (nesting predominantly in preexisting holes like old rodent nests in the ground or under moss and turfs in the herb layer, some also in pre-existing holes aboveground e.g. behind walls, in piles of stones, in bird nests, in dead wood). Cuckoo bees were categorised according to the nesting preference of their host.

FLORAL RESOURCES IN THE HERB LAYER
In each of the studied tree patches, we mapped all insect-pollinated flowering plant species in the herb-and shrub layer of each orchard. The surveys took place over the entire spring and early summer season between April 30 and July 29, 2018, when most of the meadows had been mown.

LANDSCAPE METRICS
We quantified the proportion of urban habitat, agricultural habitat, lawns, flower-rich green spaces and woody habitat in 100 m, 300 m and 600 m zones surrounding the orchards based on aerial photographs (QGIS Development Team 2016).
Due to the close proximities of the study sites, there is considerable overlap in these zones at the 600 m distance but much less at the 300 m scale and none at the 100 m scale ( Figure 2).

QUANTIFICATION OF HONEYBEE HIVES
The number of honeybee hives on the study sites and in zones of 300 m and 1,000 m diameter around each orchard was estimated based on data from the local veterinary office and personal observations on site. In total, c. 580 hives are registered at the regional veterinary office for the city of Freising and c. 900 hives for the entire municipal area, which represents a honey bee hive density of c. 19 hives/km 2 in the city (an area of c. 30 km 2 ) and 10 hives/km 2 for the whole municipal area of c. 88.6 km 2 .

METEOROLOGICAL DATA
Meteorological data was obtained from the agrometeorological weather station of the Bavarian State Research Center for Agriculture (LfL) in Freising. For each observation period, we used the reference data of the previous full hour, e.g., we attributed weather data of 2 PM to all observation slots between 2-3 PM (Appendix I).
The following weather parameters were obtained: temperature at 2 m elevation in °C, relative humidity in %, wind speed in m/s, and global radiation in Wh/m 2 .

Species richness
For each crop species, the cumulative wild bee species richness was summarized from all observation periods (137 h 30 min of total observation time). For each observation period on the fruit trees, we added the species recorded during the 10 minutes of systematic observation to those recorded in the 10 additional minutes dedicated to catch flower visitors. Values for Prunus domestica, Prunus cerasifera, Prunus avium and Prunus cerasus were pooled due to the low number of the available data for two of the tree species. We calculated the Sörensen index IA to account for similarity between wild bee communities of each crop according to the following formula IA = (2g / (a+b)) * 100 with g = total number of bee species occurring both on plant  species A and on plant species B; a = total number of bee species occurring on plant species A; b = total number of bee species occurring on plant species B. We analysed the correlation of wild bee species richness to crop species based on all crop species on all sites. Species-time relationships were calculated as the mean number of wild bee taxa per 10 min. As the total number of observation periods per species varied from 37 on plum to 190 on apple, we calculated mean rarefied species richness for 35 sampling units with rarefaction curves (function rarc).

Abundance
Flower visitor abundance on fruit trees was summarized using the maximum values of individuals per species during each of the 305 systematic observation periods. In the comparison of bee abundance on trees and berries, each individual of the same species and sex was counted only once per 10 min observation period. We analysed the correlation of abundance of different functional groups of bees (based on bee size, genus and nesting type) to crop type (orchard trees vs. garden crops) based on all crop species on all sites.
To test whether honeybee and wild bee abundance on orchard trees are significantly different, we used an unpaired two-samples Wilcoxon test (function wilcox.test) for each tree species separately because the data is metric but not normally distributed. An unpaired two-sample t-test (function t.test) and a poisson regression (function glm) gave similar results. To test significant differences in abundance of different insect groups, we performed a Kruskal-Wallis rank sum test (function kruskal.test) for each tree species because the assumptions for an ANOVA were not met. As significant differences were detected by this global test for each of the tree species, we applied post-hoc tests for pairwise multiple comparisons (function dunn.test with bonferroni correction for p-values) to see which groups differ significantly.
To test whether the number of bee species per category is similarly distributed across all crop species, we performed Pearson's Chi-squared test (function chisq.test). Because some of the counts were less than 5, we confirmed that Fisher's Exact Test for Count Data (function fisher.test) gave similar results. As significant differences were detected by these global tests, we performed the same tests on the orchard trees and garden crops separately. As no significant differences were detected within these groups, we summed the number of bee species per category for the orchard trees and garden crops respectively and tested for differences between these two crop groups, where significant differences were detected.

Landscape effects
Using the data from the most common tree species, apple, we tested for a correlation of flower visitor abundance to the proportion of urban habitat at a 100 m, 300 m, and 1,000 m scale around each site using a Poisson regression model (function glm). Each bee type and each landscape scale represented a separate model.
To study the effects of local characteristics on wild bee diversity and abundance, we first pooled data of the pollinator surveys on apple blossoms for each orchard to obtain total wild bee species richness per orchard and then performed a correlation test (function cor.test) and a Poisson regression model (function glm) to investigate whether wild bee species diversity and abundance on apple flowers, respectively, are related to species diversity of flowering plants in the meadows.
To study the impact of honeybee-hive density at different scales on honeybee and wild bee abundance on apple flowers, we fitted a Poisson regression model (function glm). Wild bees and honeybees and each landscape scale (on site, 300 m radius, 1,000 m radius) represented a separate model.

Weather effects
To study the impact of weather conditions on bee abundance on all studied tree species per bee category, we fitted a Poisson regression model (function glm). As response variable, we used honeybee, bumblebee, large solitary bee, and small solitary bee abundance on all fruit tree flowers. As continuous predictor variables, we used temperature at 2 m elevation, relative humidity, wind, and global radiation. Each bee category represented a separate model. We used a stepwise automated model selection program to construct all possible models based on the set of explanatory variables and then rank these models based on their AIC (Package MASS, function stepAIC, mode of stepwise search: "both").

Joint models explaining the abundance of different bee groups on apple flowers
Finally, we included all explanatory variables in a single model to specify which of the abovementioned predictor variables best explain bee abundances on apple flowers. Three of the explanatory variables are site-bound (honeybee hive abundance, proportion of urban habitat, floral diversity on site), but all 190 observations are time dependent regarding date and time of day and have individual values per sampling unit for all weather parameters. We did not include the variable "site" as a random effect as its influence was already included in the model through the three site-dependent explanatory variables. As the changes in the predictor variables honeybee hive abundance, floral diversity and proportion of urban habitat correspond to changes in study site, confounding factors like pollution or pesticide use might have an impact that is not accounted for in the model. For weather related variables, we only kept the parameter "global radiation", because this variable had been selected in all the models analysing the effect of different weather parameters on bee abundances. To test whether the remaining predictor variables are correlated, we performed linear models with small solitary bee, large solitary bee, bumblebee, total wild bee and honeybee abundance as response variables and computed the variance inflation factors (function vif). As all values were between 0 and 6, we left all predictor variables in the models and fitted a Poisson regression model (function glm) with small solitary bee, large solitary bee, bumblebee, total wild bee and honeybee abundance as response variables and proportion of urban habitat, species diversity of flowering plants, honeybee hive abundance and global radiation as predictor variables. Each landscape scale (honeybee hive abundance on site with proportion of urban habitat at the 100 m scale, honeybee hive abundance with proportion of urban habitat at the 300 m scales, honeybee hive abundance at the 1000 m scale with proportion of urban habitat at the 600 m scale) and each bee type constituted a separate model.

WILD BEE COMMUNITIES
The wild bee flower visitor community of the study crops comprised a total of 52 taxa observed in 137 h 30 min (Table 4, Figure 3).
The total bee species diversity ranged from 14 (plum) to 34 species (rasp-and blackberries). The bee communities of the different plant species were most similar across the fruit trees and differed most between strawberries and the fruit trees and rasp-& blackberries and the fruit trees, respectively (Table 5).
The species-time relationship (the mean number of wild bee taxa per observation period), ranged from 0.08 for apple flowers to 0.31 on raspberry and blackberry flowers. Mean rarefied species richness for 35 observation periods ranged from 9 on strawberry to 23 on raspberry and blackberry flowers (Table 6).  The number of small solitary bees was highest on berry flowers, the number of bumblebees and large solitary bee species was highest on apple flowers (Figure 4). The number of bee species per bee category was not similarly distributed across the study plants (Pearson's Chi-squared test: Xsquared = 25.636, df = 10, P-value = 0.004262; Fisher's Exact Test for Count Data: P-value = 0.006635). It was, however, similarly distributed within the tree species (Pearson's Chi-squared tests: X-squared = 0.80446, df = 6, P-value = 0.992) and the berries (X-squared = 3.3976, df = 2, P-value = 0.1829) and not similarly distributed between the orchard trees and garden crops (X-squared = 21.223, df = 2, P-value = 2.463e-05 Overall, the wild bee community on the berry crops (straw-, rasp-and blackberries) included 37 species of 14 genera ( Figure 5). Among the group of non-bee flower visitors on orchard trees, 75% of all visits were by flies (Diptera), mainly bee flies (Bombylius spec.) and hoverflies (Syrphidae), 13% by Coleoptera (9% of all visits by the rose chafer, Cetonia aurata), 8% by non-bee Hymenoptera (7% of all visits by Vespidae, 1% by Symphyta) and 4% by Lepidoptera (c. 50% Pieridae, and 50% Nymphalidae).
Ground nesting solitary bees performed 43% of flower visits on all crops and were the most abundant flower visitors both on orchard trees and berry patches. All ground nesters combined (ground nesting solitary bees, ground nesting bumblebees and bumblebees which are flexible in their nesting behaviour) comprised 76% of all bee visits on orchard trees and 74% on berries. Cavity nesting solitary bees constituted 17% of flower visits on orchard trees and 22% of visits on berries ( Figure 6).

TREES
In total, we recorded 822 bee visits on the fruit trees in 50 h 50 min observation time. More than half of the visits were by honeybees, 41% by wild  bees. On plum and pear, wild bee and honeybee abundance were similar with about one bee individual of each category per 10 min per 4 m 2 (Figure 7). On cherry and apple, honeybee abundance was significantly higher than wild bee abundance: for cherry, on average 1.1 honeybees and 0.5 wild bees per 10 min and for apple 2 honeybees and 1.3 wild bees per 10 min.
Among the wild bee categories, large solitary bees were the most abundant flower visitors on all fruit trees, followed by bumblebees on cherry and apple, and by small solitary bees on plum and pear ( Figure 8). All other flying insects performed 18% of all recorded flower visits.

FLORAL RESOURCES SURROUNDING THE TREES
In the herb-and shrub-layer of the meadow orchards, we recorded a total of 144 insectpollinated flowering plant species of 34 families (Appendix II). Flowering plant species diversity ranged from 22 species on the mulched orchard to 79 species on the fallow orchard. The number of flowering plant species on site had a slightly positive impact on wild bee abundance on apple flowers (poisson model: estimate = 0.006459, stderror = 0.003008, Z-value = 2.147, P-value = 0.0318*), but the effect was not significant in the joint models. Species richness of wild bees on apple flowers is not significantly positively affected, probably due to the low number of study sites (correlation test: T-value = 1.8535, P-value = 0.1638, N = 5).

LANDSCAPE METRICS
When considered as sole variable, the proportion of urban habitat had a positive effect on total wild bee abundance at all scales. The positive effect was higher for small solitary bee abundance than for large solitary bee abundance. Bumblebee abundance was negatively affected by the proportion of urban habitat (Table 7).
In the joint models, the proportion of urban habitat positively affected small and large solitary bee abundance and negatively affected bumblebee and honeybee abundance: when the proportion of urban habitat increases by 10%, the mean number of small solitary bees visits on apple flowers per observation period more than doubles at the 100 m scale, the mean number of large solitary bees increases by 20% at all scales, the mean number of bumblebees is reduced by 20% at the 100 m and 600 m scales, and the mean number of honeybees is reduced by 10% at all scales (Table 11).

HONEYBEE-HIVE DENSITY
The number of honeybee-hives on site ranged between zero and ten. In the 300 m radius, values ranged from nine to 37 and in a 1 km radius from 90 to 147. When considered as sole variable, hive number had a slightly positive impact on honeybee abundance and a slightly negative impact on total wild bee abundance on apple flowers. The effect decreased with increasing distance of the hives from the site. The negative effect was higher for small solitary bees than for large solitary bees. For bumblebees, it was positive Table 7. Results of poisson regression models assessing the effect of the proportion of urban habitat around each site on wild bee and honeybee abundance on apple flowers. Each landscape scale and bee type represent a separate model. Estimated parameters, standard errors, z-values and p-values are given. Significance is marked with * at P  0.05, ** at P  0.01 and *** at P  0.001. at the 300 m scale (p-values are significant at p  0.05 except for the impact of the number of honeybees on site on small solitary bee and bumblebee abundance, of the number of honeybee hives at the 300 m radius on honeybee abundance, and of the number of honeybee hives at the 1000 m radius on bumblebee abundance) (Table 8).
In the joint models, honeybee hive density had a negative effect on small solitary bee abundance at the 300 m and 1000 m scales (the mean number of small solitary bee visits per observation period is reduced by 72% and 50% when the number of bee hives increases by ten hives at the 300 m and the 1000 m scale, respectively) and a positive effect on honeybee abundance at the local scale (the mean number of honeybee visits per observation period increases by 56% when the number of bee hives on site is raised by ten hives) (Table 11).

WEATHER CONDITIONS
In 2018, when the fieldwork was performed, the flowering period of fruit trees in the orchards was earlier and shorter than usual. The year 2018 was the warmest year in Germany since the beginning of records in 1881 (deviation of 2+3 Kelvin from the period of reference . After low temperatures in February and March (Germany: -2.3 K and -1.1 K, Freising: -1.8 and -0.1), April and May 2018 were particularly warm (Germany: + 4.9 K and + 3.9 K, Freising: + 5.8 and + 4.4). While the beginning of the flowering period of sweet cherry in mid-April was average, the start of apple flowering on April 20th was c. 5 days earlier than the average of the period of reference (Imbery et al. 2018), (Agrarmeteorologie Bayern 2020). These particular weather conditions led to a shortened overall flowering period with a strong overlap between the studied fruit tree species.
When all parameters represented separate models, increases in temperature and radiation and decrease in humidity had a positive effect on honeybee, large solitary bee and small solitary bee abundance (p < 0.05 for all models except the one correlating temperature and large solitary bee abundance). The same weather conditions had the opposite effect on bumblebee abundance. Wind only had a significant (positive) effect on small solitary bee abundance (Table 9).
In the models selected by stepAIC, abundance of honeybees, large solitary bees and small solitary bees were best explained by humidity (negative impact) and radiation (positive impact). For large solitary bee abundance, temperature was included in the best model as a third parameter (with a positive impact). Abundance of bumblebees was   Table 10. Results of poisson regression models assessing the effect of temperature at 2 m elevation, relative humidity, and global radiation on the abundance of different bee groups on fruit trees. Models were selected using the stepAIC function (selection in both directions). Only best models are reported and model estimates, p-value and model selection statistics are given. Significance is marked with * at P  0.05, ** at P  0.01 and *** at P  0.001. best explained by temperature and radiation (both with a negative impact) (Table 10).
In the joint models only considering flower visitors on apple flowers, global radiation had a significant and positive effect on the abundance of all bee groups except bumblebees (Table 11).

HOW DIVERSE IS THE FLOWER VISITOR COMMUNITY?
Overall, the observed wild bee community in Freising comprised a large number of species and genera characterized by a variety of functional traits. With 32 species on orchard trees, 17 on strawberries and 34 on raspberries, our results are Table 11. Results of poisson regression models describing the effect of the number of honeybee hives at different landscape scales, of floral diversity on site, of the proportion of urban habitat at different landscape scales, and of global radiation on bee abundances on apple flowers. Each landscape scale and bee type represent a separate model. Estimated parameters, standard errors, z-values, p-values and model selection statistics are given. Significance is marked with * at P  0.1, * at P  0.05, ** at P  0.01 and *** at P  0.001.  Poll Ecol 29(16) comparable to findings in agricultural study sites. For example, 47 bee species have been identified in mixed meadow orchards in Baden-Württemberg (Schwenninger & Wolf-Schwenninger 2012), 22 bee species on sweet cherry orchards in Flanders , 27 bee species on apple orchards in Hungary (Földesi et al. 2016), 18 bee species on strawberry fields in Lower Saxony (Ahrenfeldt et al. 2015), and 47 bee species on raspberry in forests in Maine (Hansen & Osgood 1983). Since our species identification approach based on high resolution photographs instead of collected bees leads most likely to an underestimation of total species numbers in difficult genera like Lasioglossum and Nomada, the real diversity in our study sites can be expected to be even higher.
The high proportion of bumblebee visitation in our study plants is in line with previous studies on these crops (Pardo & Borges 2020;Quinet et al. 2016;Ellis et al. 2017;Lye et al. 2011). Large wild bees were predominant on fruit trees in meadow orchards and small wild bees were more abundant on straw-, rasp-and blackberries in gardens.
Considering that many small bodied wild bee species are more restricted in their flight range (Zurbuchen et al. 2010;Gathmann & Tscharntke 2002;Greenleaf et al. 2007;Hofmann et al. 2020), their higher share in garden crop visitation might be related to the fact that these sites provide a larger variety of microstructures at a small scale. Similarly, Bennett & Lovell (2019) found abundances of small-bodied and larger-bodied bee species in urban agricultural sites to respond most strongly to local site and landscape variables, respectively. The late flowering period of straw-, rasp-and blackberries might have an additional impact, as many small bees (e.g. Hylaeus spp., Hoplitis spp.) are active later in the season than larger-bodied species (e.g. Andrena spp., Osmia spp., and Anthophora spp.) (Westrich 2018).
The overall large share of ground-nesters reflects their predominance in the German wild bee fauna since c. 74% of all non-parasitic bee species known from Germany nest in self-dug or pre-existing cavities in the ground (Westrich 2018). The slightly higher share of hypogeic cavitynesters on garden crops compared to orchard trees might be related to higher heterogeneity and abundance of aboveground-microstructures within these sites (Fortel et al. 2014;Matteson et al. 2008). In particular, species of three of the genera forming the group of small bodied bees -Hylaeus, Ceratina, Osmia (Hoplitis) -nest in stems of raspand blackberries (Westrich 2018).

BERRIES FROM THE ROSE FAMILY?
More than half of the flower visits on apple in the urban meadow orchards were performed by honeybees. The large proportion of honeybees can be explained by the high density of honeybee hives within the city with a minimum of 90 honeybee hives in a 1 km radius around each site (containing in late spring up to 4.5 Mio worker bees), and by the large flight range of honeybees as they routinely forage in distances of 1.5 km, and can travel more than 10 km in search of floral rewards (Beekman & Ratnieks 2000;Waddington & Visscher 1994;Visscher & Seeley 1982;Zurbuchen et al. 2010). At the same time, our study shows that wild bees provide an important share of total bee flower visits. The wild bee proportions of 33% of total bee visits on cherry, 39% on apple, 46% on plum, and 52% on pear is comparable or higher than rates recorded in orchards outside urban areas. In a study on mixed orchards in Baden-Württemberg, 16% of bee visits on apple, 38% on Prunus spp. and 40% on pear were performed by wild bees (Schwenninger & Wolf-Schwenninger 2012), in a study on orchards in northern Hesse, about one third of the bee visits on sweet cherry were from wild bees (Holzschuh et al. 2012).
The relevance of wild bees for urban food crop production might exceed their mere share in flower visitation rate because of pollination effectiveness, interspecific facilitation and complementary pollination behaviour. In a study on 41 crop systems worldwide, Garibaldi et al. (2013) found fruit set to increase twice as strongly with increased visitation by wild insects as with an equivalent increase in visitation by honeybees. Studies investigating the pollination behaviour of different insect groups have shown that many hairy large solitary bees like Andrena and Osmia ssp. outperform honeybees in pollination effectiveness because the dry-collected pollen in dense brushes on the hind legs or beneath the abdomen is more likely to be transferred from flower to flower and because of their flower visitation rate and behaviour (Stavert et al. 2016;Eeraerts, Vanderhaegen, et al. 2019;Pardo & Borges 2020;Mallinger & Gratton 2015;Russo et al. 2017). Pollination observations on raspberries and apple blossoms showed that bumblebees outperform honeybees in pollen deposition on stigmata and forage over longer periods of the day and during poorer weather conditions (Thomson & Goodell 2002;Willmer et al. 1994). Research on interspecies facilitation in almond and sweet cherry orchards provided evidence that interspecies interactions between wild bees and honeybees increase the pollination effectiveness of honeybees (Brittain et al. 2013;Eeraerts et al. 2020). Studies on strawberry fields identified the relevance of complementary pollination behaviour of wild and honeybees for fruit quality (Chagnon et al. 1993). All this points to an important role of wild bees as urban pollinators.

WHAT IS THE IMPACT OF LAND USE TYPES SURROUNDING THE CROP SPECIES?
While honeybee and bumblebee abundance on apple flowers were negatively affected by the proportion of urban habitat, small and large solitary bee and total wild bee abundance were positively affected. It is likely that the high proportion of small private gardens and green balconies in these areas made them more attractive to wild bee species than would have been expected. In previous studies, some urban environments have been shown to support wild bee visitation of some bee clades in adjacent crops (Carré et al. 2009;Langellotto et al. 2018;Martins et al. 2018). In particular, urbanised habitats have been found to filter for small solitary bee species, possibly because their lower quantitative requirements in pollen and nectar and smaller flight ranges make them well adapted to urban environments which often provide structurally diverse habitats at a small scale (Banaszak-Cibicka & Żmihorski 2012; Wilson & Jamieson 2019; Buchholz & Egerer 2020). The negative effect on bumblebee abundance in turn might be related to the scarcity of potential nesting sites in largely sealed urban areas. Overall, the partially positive effect of urban habitats on wild bee abundance in J Poll Ecol 29(16) Freising indicates that some urban environments like private gardens can support crop visiting wild bee communities.
We find no significant effect of flowering plant cover in the herb layer on abundance and species richness of wild bees visiting apple flowers. Other studies suggested that additional floral resources in the herb layer of orchards are necessary to support wild bees because the flowering period of fruit trees is shorter than the life cycle of wild bees (Bertrand et al. 2019;Alomar et al. 2018;Saunders & Luck 2018). In our mixed orchards, however, the combined flowering period of plum, cherry, apple, and pear spanned several weeks and this might be the reason why even the orchards with intense mowing regime had a diverse wild bee community.

LANDSCAPE SCALE ON WILD BEE ABUNDANCE?
Honey bee density in Freising is incredibly high with an absolute number of hives almost as high as in the entire city of Paris and a hive density 2-3 times as high as in Berlin, Hamburg (Beckedorf 2015) and Paris, where negative effects on wild bee communities have been demonstrated (Ropars et al. 2019). We also show that wild bee visitation rate on orchard crops was negatively correlated to honey bee hive densities, which is in line with previous studies in natural habitats, agroecosystems and urban environments (Torné-Noguera et al. 2016;Alomar et al. 2018;Ropars et al. 2019;Renner et al. 2021). In Paris, Ropars et al. (2019) found a significant negative effect of honeybee hive densities on large solitary bees and bumblebees but not on small solitary bees, and hypothesised that the latter might be less sensitive to increases of honeybee hive densities because of resource complementarity, smaller species often preferring shallow flowers while larger pollinators prefer deep flowers. Interestingly, in our study focusing exclusively on plants with shallow flowers, the negative effect of honey bee hive density was significant and strongest for small solitary bee species, which indicates that they might be particularly sensitive to direct competition.
These results support the theory that honey bees outcompete other pollinators especially when resources are limited (Cane & Tepedino 2017;Mallinger et al. 2017). In light of social and conservation benefits of beekeeping, calls for inclusive solutions to preserve both biodiversity and beekeeping activities have been articulated (Aebi et al. 2012;Kleijn et al. 2018;Egerer & Kowarik 2020) to promote beekeeping for vs. beekeeping of the city (Sponsler & Bratman 2020). Specific measures should combine adapting hive densities to local floral supply (taking into account floral abundance, species composition and phenology) e.g. through the definition of precautionary zones (Stange et al. 2017;Cane & Tepedino 2017;Henry & Rodet 2020) with the coupling of beekeeping and floral resource enhancement adapted to the specific requirements of diverse wild bee communities e.g. regarding plant species composition and provenience (Baldock 2020;Nichols et al. 2019;Bucharowa et al. 2021).

WHAT IS THE IMPACT OF WEATHER CONDITIONS ON DIFFERENT BEE GROUPS?
The observation that the abundance of honeybees, large and small solitary bees increased with lower humidity and higher temperature and radiation levels while bumblebee abundance decreased under the same weather conditions corresponds to findings that bee species differ in sensitivity to climatic conditions (Burdine & McCluney 2019b). Bumblebees in particular are more tolerant to poor weather (Nielsen et al. 2017;Tuell & Isaacs 2010) and this suggests that they are especially valuable pollinators in early spring and during bad weather periods. Unfortunately, this makes them also more vulnerable to climate change, where the increasing number of very hot days seems to affect the long-term survival of bumblebee populations around the world (Soroye et al. 2020). In cities, where the urban heat effect contributes often to an even higher number of hot days, bumblebees will suffer even faster declines (Hamblin et al. 2017) unless sufficient cooler green space is provided as a refugium.

CONCLUSIONS
Our results demonstrate the importance of diverse wild bee communities for fruit crop pollination in the urban environment. The broad variety of wild bee species differing in size, flight range, nesting behaviour, seasonality and tolerance to weather conditions can guarantee efficient pollination. Honeybees alone are unlikely to reach such a level. Bee-friendly management of green spaces in cities, including private gardens and orchards is a cheap and easy way to support wild bees as well as urban honeybees. Appropriate measures to improve resource availability and connectivity should consider the specific needs of diverse wild bee communities regarding nesting habitat, nesting materials and flowering plant species composition and provenience. This could also reduce potential conflict due to increasing density of honey bee hives in cities and make urban insect communities and fruit crop production more resilient to changing climates.