Quantifying bee assemblages and attractiveness of flowering woody landscape plants for urban pollinator conservation

Urban and suburban landscapes can be refuges for biodiversity of bees and other pollinators. Public awareness of declining pollinator populations has increased interest in growing plants that provide floral resources for bees. Various publications and websites list “bee-friendly” plants, but such lists are rarely based on empirical data, nor do they emphasize flowering trees and shrubs, which are a major component of urban landscapes. We quantified bee visitation to 72 species of flowering woody landscape plants across 373 urban and suburban sites in Kentucky and southern Ohio, USA, sampling and identifying the bee assemblages associated with 45 of the most bee-attractive species. We found strong plant species effects and variation in seasonal activity of particular bee taxa, but no overall differences in extent of bee visitation or bee genus diversity between native and non–native species, trees and shrubs, or early-, mid-, and late-season blooming plants. Horticulturally-modified varieties of Hydrangea, Prunus, and Rosa with double petals or clusters of showy sterile sepals attracted few bees compared to related plants with more accessible floral rewards. Some of the non-native woody plant species bloomed when floral resources from native plants were scarce and were highly bee-attractive, so their use in landscapes could help extend the flowering season for bees. These data will help city foresters, landscape managers, and the public make informed decisions to create bee–friendly urban and suburban landscapes.


Introduction
Many wild bee species, including important crop pollinators such as bumble bees (Bombus spp.), are declining in abundance or range [1][2][3][4][5][6]. Loss of floral resources, associated with agricultural intensification and habitat loss, is one of the major drivers of pollinator decline [5,7]. Protecting natural areas and restoring agricultural lands are important strategies for pollinator conservation, but urban landscapes, which offer a variety of forage and nesting sites, can also be refuges for bees [8][9][10]. Indeed, substantial portions of native bee communities can persist and even thrive in urban and suburban areas with support from gardens [11][12][13][14][15][16], parks [17], low-input lawns [18][19], and other properly designed and managed urban green spaces. PLOS  insect-pollinated trees and shrubs that are attractive to consumers in part because of their showy flowers or fruits. We compared overall attractiveness and bee genus richness and diversity between native and non-native plant species, trees and shrubs, and early-, mid-, and lateseason blooming species. Patterns of preference and seasonal activity of different bee taxa based on their abundance in collections from each plant species were quantified. We identified numerous bee-attractive species of woody landscape plants and documented clear differences in the assemblages of bees attracted to different plant species.

Plant species
In total, 72 species of flowering woody plants were sampled from 2014-2017 (Tables 1 and 2). Sampling took place from February to November each year. Plant species were selected based on recommendations from land care professionals, their suitability for planting within the Ohio River Valley region, and availability and frequency of use in urban landscapes. Both native and non-native plant species were included in order to compare their usage by bees. Plants listed as an invasive or nuisance species by the USDA National Invasive Species Information Center [52] or by the state governments of Kentucky, Tennessee, Missouri, Illinois, Indiana, Ohio, West Virginia, or Virginia were not included. Additionally, we sampled three sets of plant species (Hydrangea spp., Ilex spp., and Rosa spp.) to compare bee-attractiveness and bee genus diversity among cultivars differing in horticultural characteristics, and between closely-related native and non-native plants.

Sample sites
All 373 sample sites were located within the urban landscape and were separated by at least 1 km for same-species sites to ensure minimal overlap of bee populations. Sample sites included street-side and municipal plantings, commercial and residential landscapes, campuses, parking lots, and urban arboreta and cemeteries. Most (93%) of the sample sites were within the Lexington, Kentucky USA metropolitan area; the remainder were in urban or peri-urban cemeteries or arboreta in Louisville or northern Kentucky, or near Cincinnati in southern Ohio. All sample sites were within 145 km of the Lexington city limits. Individual sample sites ranged from single trees or large shrubs, to groupings or hedges of a particular plant species. We sampled five different sites for most (56) of the 72 plant species, four sites for 10 plant species, and three sites for each of the remaining six, harder-to-locate woody plants. An additional 35 sites were used for comparisons between native and hybrid tea roses (Rosa spp.). Permissions to collect samples of bees were granted by grounds managers, staff, or by property owners depending on site type.

Bee-attractiveness ratings
Given the wide variation in plant height and form, and in floral density, size, and morphology across such a wide range of trees and shrubs planted at hundreds of sites, it was not possible to standardize sampling on the basis of floral area such as has been done in studies [e.g., 15,40] quantifying bee visitation to same-sized replicated plots of herbaceous flowering plants in a common-garden setting. Instead, each plant species' relative bee-attractiveness was rated based on two 30-second "snapshot" counts [16] per site for in most cases 10 (minimum of six) snapshot counts per plant species. The snapshot counts were also used to justify the exclusion of relatively non-attractive plants from more extensive bee sampling. Snapshot counts were taken at or near peak bloom of a given plant. During each 30-second period, bees actively foraging on the flowers of the target plant(s) were counted, taking care to avoid counting the same insects more than once. Snapshot counts were taken while walking slowly around the tree or shrub, or along hedges or other sites with long, continuous plantings (> 2.5 m), whereas for smaller shrubs they were taken while stationary. For relatively tall trees, snapshot counts were taken only as high up in the canopy as the observer was able to distinguish bees from flies or other insects. Because of the large number of sample sites and distances between them, variable weather conditions, and the relatively brief (1-2 week) and overlapping bloom periods for many of the plants, in most cases it was not possible to visit and sample a given site more than once. Sampling conditions were as consistent as possible within a given species; e.g., same-species sites were sampled on the same day or within a few days of each other in a given year, snapshot counts were taken between 10:00 to 18:00 EST during non-inclement weather (e.g., sunny to partly cloudy, winds < 16 kph), and sampling of early spring-blooming species was done only on days with temperatures >10˚C and bees were active. Snapshot counts at a given site were taken immediately before collecting each 50-bee sample (see below) to minimize disturbance of the bees.

Sample collection
We sampled the bee assemblages associated with 45 of the 72 aforementioned plant species, excluding relatively non-attractive ones with average snapshot counts of < 5 bees. Samples were collected from 213 total sites including five sites for 35 of the plant species, four sites for eight species, and three sites for two of the rarer plants (213 total sites). Bees were collected immediately after taking snapshot counts and represented the first 50 bees observed on the flowers after the counts were finished (250 total bees collected for most plant species). Most samples were collected using aerial insect nets that could be extended to collect from heights up to about 5 m above ground level, when necessary. Some shrubs with fragile flowers were sampled by knocking individual bees into plastic containers filled with 75% EtOH. Sampling time ranged from < 15 min to more than 2 h per site. Bee samples were washed with water and dish soap, rinsed, then dried using a fan-powered dryer for 30-60 min, and pinned. All bees were identified to genus [53,54]. Bumble bees (Bombus spp.) and honey bees (Apis mellifera L.) were identified to species. Reference specimens are deposited in the University of Kentucky Department of Entomology Insect Collection.

Statistical analysis
Snapshot counts, bee genus diversity, and abundances of each of the five predominant families of bees (Andrenidae, Apidae, Colletidae, Halictidae, Megachilidae) and of Bombus spp. and A. mellifera, were compiled across sampling years and analyzed for main effects of plant species, plant family (as a proxy for plant species due to limited degrees of freedom), provenance (native or non-native), plant type (tree or shrub), and Julian date number for peak bloom using General linear models procedure (SAS, Version 9.4; SAS Institute, Cary, NC, USA). Species diversity was based on the inverse of Simpson's D (hereafter 1/D), which calculates a number between 0 and 1, with higher numbers indicating more species-rich and even samples (Margarun 2004). For analysis of bee taxa abundance, we counted the number of individuals in each sample belonging to one of five North American bee families (Apidae, Andrenidae, Colletidae, Halictidae, Megachilidae) and two additional taxa, Apis mellifera and Bombus spp. and analyzed abundance of each for main effects. Sampling date was standardized by converting to a Julian date number, which assigns each calendar date a unique integer starting from 0 on January 1. We also attempted to analyze main effects of flower color, flower type, and inflorescence type on bee snapshot counts, taxa abundance, and diversity but were unable to do so due to the uneven distribution of the data among class variable levels.  effects on snapshot counts (Table 4). There were small but statistically significant differences in snapshot counts between trees and shrubs, with trees having higher snapshot counts than shrubs (Fig 1). Snapshot counts increased slightly as the growing season progressed. There were no significant differences in snapshot counts between native or non-native species (Fig  1).

Bee abundance by taxon and genus diversity
Overall, 11,275 bees were collected from 45 species of flowering woody plants that attracted, on average, � 5 bees in the snapshot counts. Apid bees comprised 44.0% of all bees sampled and were present on all 45 plant species sampled (Tables 3 and 5). Halictid bees were similarly ubiquitous on all plant species and accounted for 23.6% of total bees. Andrenid bees accounted for 21.4% of total bees and often dominated the bee assemblages of early blooming plants. Colletid and megachilid bees were the least abundant bees overall, comprising only 5.0 and 5.9%, respectively, of the total bees in our samples. Apis mellifera and Bombus spp. were collected from 44 and 39 of the sampled plant species, respectively, and accounted for 21.4 and 11.9% of the total bees. Plant species (Table 4), and by extension plant family, played a key role in abundance of all bee taxa analyzed (Table 6) and both were the only significant factors for Andrenidae, Apidae, and A. mellifera. Most woody plants attracted bees from at least four families; one exception was mock orange (Philadelphus) from which >95% of the bees collected were Chelostoma philadelphi (Robertson), a small megachilid. Colletidae, Halictidae, and Bombus all showed strong seasonal patterns in abundance, with the proportion of Colletidae in our samples declining sharply with increasing Julian date, while proportionate abundance of Halictidae and Bombus increased. Colletidae were proportionately more abundant on trees than on shrubs, and on native as opposed to non-native plant species (Table 6). All other bee taxa, including nonnative A. mellifera and native Bombus, were equally proportionately abundant on native and non-native plants. Twenty-three bee genera were represented in our samples (Table 5), the most abundant being Apis (22.1% of total bees), Andrena (21.4%), Lasioglossum (19.6%), and Bombus (12.2%).  Table 4.

Cultivar comparisons
Snapshot counts were compared among four Hydrangea species, H. arborescens 'Annabelle' (native, shrub), H. macrophylla (non-native, shrub), H. paniculata (non-native, shrub), and H. quercifolia (native, shrub) which differ in their floral characteristics (Table 2). Most notably, H. paniculata has exposed fertile flowers while the other three species lack fertile flowers or have them hidden beneath showy sterile outer sepals (Dirr 2011). Non-native hydrangeas had higher average snapshot counts than native hydrangeas (14.0 and 2.8, respectively, F 1,34 = 6.19, P = 0.02), but this was entirely because H. paniculata, a non-native, was the only species that was highly attractive to bees. Bee genus diversity was not analyzed because H. arborescens, H. macrophylla, and H. quercifolia had extremely low bee visitation rates, and were not sampled for bees. Snapshot counts and bee assemblages were compared between four Ilex species: I. × attenuata (native, shrub), I. × meserveae (non-native, tree), I. opaca (native, tree), and I. verticillata (native, shrub). All four had similar floral characteristics (Tables 1 and 2) and differed mainly by height and spread of the plant. There were no significant differences between the average snapshot counts of native and non-native Ilex (18.8 and 15.5, respectively, F 1,34 = 0.77, P = 0.39), nor were there significant differences between the average genus diversity of native and non-native Ilex species (0.56 and 0.60, respectively, F 1,18 = 0.28, P = 0.60).
Bee visitation to two species of roses (Rosa) was compared. Rosa setigera, a single-flowered native rose with pollen prominently displayed during most of its bloom, was sampled at five sites. Hybrid tea roses are non-native, and they are typically double-or triple-flowered and either lack stamens and pollen, or have pollen that is concealed by multiple layers of petals during bloom. We sampled a variety of hybrid tea roses which we divided into seven categories based on color and flower form: light pink with single petals, dark pink with single petals, red with single petals, white with double or triple-petals, light pink with double-or triple petals, dark pink with double or triple petals, or red with double or triple-petals. Rosa setigera had a significantly higher average snapshot count than all hybrid tea roses sampled (16.1 and 0.1, respectively, F 1,78 = 146.8, P < 0.001). Bee genus diversity was not analyzed because the hybrid tea roses, which had very low visitation rates, were not sampled for bees.

Discussion
To our knowledge, this is the first scientific study to quantify variation in bee-attractiveness and bee assemblages across a wide range of flowering woody landscape plants. We identified 45 species of trees and shrubs that could be useful for augmenting floral resources for bees in urban and suburban settings. Although all of our sampling took place in Kentucky and southern Ohio, most of the bee-attractive plants on our list should grow satisfactorily throughout USDA Plant Hardiness Zone 6, which covers extensive regions of the United States [55].
As with all studies assessing diversity of bees [56], our sampling methodology has limits and biases. Counting bees on the wing, as in our snapshot counts, leaves room for misidentification (e.g., counting bee mimics as bees) and miscounting, but we attempted to reduce this by replicating counts and using skilled observers with training in bee identification. Snapshot counts and 50-bee samples were based on one visit to each site because of the large number of sample sites, the distances between sites (up to 145 km), and the relatively short bloom periods of some plants. While it is unlikely that a sample of 250 bees collected from five sites would capture the full bee species richness and diversity of a given plant species during the entirety of its bloom, our data do provide a measure of which tree and shrub species attract and support robust bee assemblages. Although some studies have used replicated plots with similar-aged plants to compare bee visitation rates [16,42], establishing 72 species of trees and shrubs in a replicated common garden plot for eventual pollinator sampling would have been impractical because of the cost, space, and time required for establishment. Moreover, results from common garden experiments can be location-specific, reflecting the relative abundance of different pollinator taxa at that particular site. Our sampling from multiple (in most cases five) plantings of each species across hundreds of existing urban landscape sites doubtless encompassed more of the variation in soil conditions, potential nesting sites, and other landscape-level factors that would affect bee diversity than if all sampling had been done at a single location.
The premise that augmenting floral resources benefits bees is based on the assumption that local bee populations are often food-limited. Floral resource availability is thought to be a major driver of population abundance and diversity of wild bees [7]. Long-term abundance of bumble bees and other wild bees has declined in parallel with widespread declines in floral abundance and diversity in Europe [1,4], and populations of solitary bees are enhanced by mass-flowering crops, suggesting that floral resources are indeed limiting [57,58]. There is some debate [59] that the dense, high-resource displays of wildflower mixes or other urban plantings intended to augment resources for bees might have unintended ecological consequences for remnant native plant biodiversity (e.g., by competing for generalist pollinators, functioning as hubs for pollinator-transmitted plant pathogens, or decreasing he likelihood of conspecific pollen transfer). However, such plantings might also increase pollination of remnant native plants through a spillover effect [59] similar to that observed in agricultural crops bordered by wildflower strips [60]. Although those types of potential ecological interactions warrant future research, they are beyond the scope of this study. Together with studies documenting that four common city tree species attracted a fifth of all native bee species occurring in Berlin, Germany [22], and that nine of the main tree species planted along streets of European cities, including some non-native species and hybrids, provide nectar and pollen of high nutritional suitability for pollinators [45] our results suggest that urban landscapes can be made even more valuable as refuges for pollinators by incorporating additional bee-attractive woody plants.
Urban and suburban landscapes typically consist of a diverse mix of native and non-native plant species [16,44,[61][62][63][64][65]. Recently, the long-standing debate [66,67] about whether or not there is any role for non-invasive exotic plants in conservation biology has spurred a fervent movement in gardening circles advocating that urban landscapes be constructed predominantly or exclusively with native plants [68]. One of the main arguments against landscaping with non-native plants; i.e., ones that do not occur naturally in a particular region, ecosystem, or habitat, is their potential to become invasive. Although ornamental horticulture has been a major pathway for plant invasions [69,70], many non-native ornamentals are either sterile hybrids or are considered non-invasive with a low risk of escaping cultivation [71,72]. None of the woody plants included in our study is listed as invasive in Kentucky or surrounding states [52].
Another argument for landscaping with native as opposed to non-native plant species is that natives tend to support higher diversity and numbers of endemic caterpillars and other coevolved plant-feeding insects that help to sustain insectivorous birds and other desirable urban wildlife [68,[73][74][75]. However, ornamental plants and shade trees are also valued for aesthetic appearance, so ones with abundant insect herbivores and associated feeding damage are more likely to be treated with insecticides that could be hazardous to bees. Moreover, some native North American woody plants are far more susceptible to invasive pests than their exotic congeners originating from the pest's natal region [76,77] and thus more likely to receive pesticide applications. We identified a number of non-invasive, non-native woody plants (e.g., Abelia, Aralia, Cornus mas, Heptaconium miconioides, Hydrangia paniculata, Maackia amurensis, Tetradium daniellii, Vitex agnus-castus, and others), that are both highly bee-attractive and relatively pest free, making them good candidates for use in bee-friendly urban landscapes. The present study adds to a growing body of evidence that both native and non-native plants can be valuable in helping to support bees and other pollinators in urbanized habitats [12,16,18,22,45,46,63,64,78]. Because most urban bees are polylectic [13,14] and will forage on a wide variety of plant species, they will readily incorporate non-native plants into their diets so long as they provide sufficient quantity and quality of pollen and nectar [79].
Phenology of bloom is important when considering the value of plants for bees. Bloom time tends to be conserved by geographic origin, with cultivated non-native plants generally retaining the phenology of their source region [63,80]. Our study identified 15 species of bee-attractive woody landscape plants that typically bloom before April or after mid-July, twelve of which are non-native (Table 3). Early or late blooming plants can be especially valuable to seasonal specialists by providing floral resources during critical times of nest establishment in the spring and winter provisioning in the fall [64,81]. Bumble bees, which do not store substantial amounts of pollen and nectar, require a consistent supply of floral resources throughout the growing season including in early spring when post-wintering queens are foraging alone to establish their colony, and late in the growing season to provision the developing queen brood, and as food for new queens that feed heavily in preparation for hibernation in overwintering sites [76][77][78]. In our study, bumble bees constituted a large proportion of the samples from Aesculus × carnea in early spring, and from Abelia × grandiflora, Clethra alnifolia, Heptacodium miconioides, and Vitex agnus-castus late in the growing season. Honey bees, which will forage year-round if weather permits, also benefit from season-long floral resources. We identified a number of trees and shrubs that are highly attractive to honey bees including some that bloom early (e.g., Cornus mas and Prunus subhirtella 'Autumnalis') or late (e.g., Rhus copallinum and Tetradium daniellii) in the growing season. Wind-pollinated (anemophilous) plants, particularly trees, tend to flower earlier than entomophilous species [82], and they, too, can provide critical nutrients for bees, especially in early spring before the springsummer floral resource peaks [49][50][51]. Inclusion of wind-pollinated trees and shrubs, which were not sampled in our study, could change the proportion of bee-attractive natives versus non-natives amongst early-blooming landscape plants. Urban landscapes can be enhanced as habitat for bees by incorporating a variety of entomophilous, anemophilous, and ambophilous flowering plant species, biased toward natives and near-natives but including some nonnatives, to ensure succession of overlapping bloom periods and provide food during periods of poor nutrient availability before and after the spring to mid-summer floral resource peaks [64].
One caveat to our data suggesting that native and non-native woody landscape plants may have equivalent usefulness for urban bee conservation is the genus-level taxonomic resolution of our bee data. Without identifying all of the >11,000 bees to species, it is impossible to know whether the non-native trees and shrubs attract and support disproportionate numbers of non-native bee species. Other than the sometimes negative ecological impacts of Apis, Bombus and Megachile spp. that were deliberately introduced to new regions of the world for agricultural pollination [83,84], there is little or no empirical evidence that non-native bees degrade pollination networks or negatively affect the pollination of native plants [84]. However, some non-native bee species exhibit marked preferences for visiting plants from their own natal region [83] which could have consequences should those bees become invasive. In our study, the giant resin bee Megachile sculpturalis Smith, a native of eastern Asia, was collected from flowers of six of the 45 woody plant species from which bee assemblages were sampled, including 12 specimens collected from Aesculus parviflora and Oxydendron arboreum, and 85 specimens from Koelreuteria paniculata, Tetradium daniellii, Vitex agnus-castus, and Maackia amurensis. The first two of the aforementioned plants are native, but the latter four, which accounted for 88% of the collections, are of Asian origin. Megachile sculpturalis has been observed to forcibly evict and occupy the nests of native Osmia sp. and Xylocopa sp. in the United States and Europe [85], so more widespread planting of its favored Asian ornamental trees could facilitate its range expansion, with possible deleterious consequences for native bees in its introduced range. On the other hand, all of the seven Bombus species we collected from woody landscape plants are native, and jointly they foraged equally on native and nonnative woody plants. Indeed, two of the top bumble bee "magnets" were Abelia × grandiflora and Heptacodium miconioides, both late-blooming, non-native plants that were heavily visited by Bombus workers and young queens into late September when most other plants in those landscapes were done blooming.
Another caveat is that, besides being used to inform decisions about woody landscape plants that may promote bee conservation, our data will be used by stakeholders wishing to identify and avoid planting trees and shrubs that attract bees, either to reduce hazard to persons having anaphylactic allergies to bee stings or general hazard and liability around residences or in public settings, or because of general fear of bees. We acknowledge that highly bee-attractive trees and shrubs that are suitable for most landscape settings might be poor choices for sites such as primary school playgrounds, yards frequented by small children, or outdoor public outdoor eating spaces.
Although some plant varieties with double flowers or showy sterile outer sepals that inhibit access to central, fertile flowers may not provide sufficient floral rewards to attract bees [46,86], many horticulturally-modified plants, including hybrids, can be as attractive, or more attractive, than their wild-type counterparts [16,44]. In our study, neither of the native Hydrangea species, having been bred for large clusters of showy sterile sepals, was bee-attractive whereas the open-flowered, non-native H. paniculata had the highest average snapshot count of the 36 shrub species we sampled. Similarly, R. setigera, a native single-flowered rose, was highly attractive to bees, whereas none of the double-and triple-flowered hybrid tea rose cultivars attracted more than a single bee. All four of the Ilex species we compared, representing a mix of native, non-native, and hybrid species, offer easily accessible floral rewards, and all four were attractive to bees. This further illustrates that cultivars and non-native species can be equally attractive to bees as long as floral rewards have not been bred out or obscured. Similar patterns were seen within other plant genera; e.g., Prunus subhirtella and P. virginiana that have single, open flowers, were highly bee-attractive, whereas P. kanzan, a double-flowered species, attracted almost no bees.
In conclusion, this study identified many species of flowering trees and shrubs that are highly attractive to bees and documented the types of bees that visit them. Even so, we did not come close to capturing the enormous diversity of flowering woody landscape plants available in the marketplace [62], so there is great potential for identifying additional plants that could have value for urban bee conservation. Recommendations for bee-attractive plants that are based on empirical data are preferable to the large number of plant lists available to the public that are based only on informal observations or anecdotes [35]. Our data should help to inform and augment existing lists of bee-attractive plants in addition to encouraging the use of sustainable, bee-attractive woody landscape plants to conserve and restore resources for urban pollinators.