Road verges support pollinators in agricultural landscapes, but are diminished by heavy traffic and summer cutting

1. Supporting pollinators in agricultural landscapes is important for reversing their global decline. Road verges and hedges are used by pollinators for feeding and reproduction, but few studies consider entire pollinator communities, and it remains unclear how they are distributed across adjacent verges, hedges and fields, or how they are affected by traffic and verge cutting. 2. We surveyed flowers and pollinators, using transect counts and pan traps, to explore the role of road verges and their associated hedges in supporting pollinators in an agricultural landscape in southwest England, and the impacts of traffic and verge cutting. At 19 sites, we surveyed the road verge (verge edge and verge cen-tre), the verge hedge (both sides), a field hedge and the field interior. Road verges and hedges much greater flower flower species richness and pollinator than field interiors. Verge hedges far less woody cover than field hedges, greater flower species richness.


| INTRODUC TI ON
Supporting biodiversity in agricultural landscapes is important for global nature conservation (Kremen & Merenlender, 2018), and for human economies due to the ecosystem services upon which agriculture relies (Zhang, Ricketts, Kremen, Carney, & Swinton, 2007).
Although the importance of hedges is well-known, there is increasing interest in road verges and their potential as a conservation resource (Gardiner, Riley, Bommarco, & Öckinger, 2018), especially given the large areas that they cover, for example an estimated 2,400 km 2 in Great Britain, or 1% of land (Plantlife, 2013). In Europe and North America, most rural road verges are cut once or twice per year, or not at all, though cuts are more frequent in some regions, and grazing, burning or herbicides are sometimes used (Bernes et al., 2017). However, there is growing mainstream pressure to change management to benefit wildlife, for example a public campaign by the charity Plantlife in Great Britain has received over 70,000 signatures and is providing management guidance to Councils (Plantlife, 2019). But there is a need for clarity on the impacts of traffic and verge cutting, which will determine how verges can best be utilized and managed.
The value of road verges as habitats for plants and pollinators is the result of low-intensity management, compared to regular grazing, tillage and chemical application in adjacent agricultural fields. Agricultural practices in field interiors can also affect adjacent hedges (Aude, Tybirk, & Bruus Pedersen, 2003), for example herbicide drift and fertilizer run-off may lead to a simplified flora (Staley et al., 2013). It is likely that hedges with fields on both sides, hereafter 'field hedges', are affected by agricultural practices to a greater extent than hedges with a road verge on one side, hereafter 'verge hedges'. This may lead to verge hedges having a greater capacity to support plant and pollinator communities than field hedges. For example Hanley and Wilkins (2015) found that the road-facing side of verge hedges contained a greater flower species richness, flower abundance and bumblebee abundance than the field-facing side. However, no research has compared verge hedges to field hedges.
Although road verges provide an opportunity for nature conservation, there is an array of impacts of roads and traffic that may negatively affect pollinators, including light pollution (Knop et al., 2017), noise pollution (Davis, Schroeder, Yeager, & Pearce, 2018) and traffic collision (Keilsohn, Narango, & Tallamy, 2018). This presents a management challenge to enhance the benefits of road verges to pollinators while reducing the negative impacts of roads and traffic. These negative effects are likely to be most prominent along the edge of the verge by the road, hereafter the 'verge edge', which may have management implications. Furthermore, as most of these negative effects are caused by traffic, they are likely to have a greater impact on busier roads, but no research has explored the impacts of traffic and proximity to roads on pollinator communities.
Road verge management affects their conservation value, though research has mostly focused on plants (Jakobsson, Bernes, Bullock, Verheyen, & Lindborg, 2018). A plot experiment along a single road verge found that cutting twice a year resulted in more flowers and insects than cutting once or no management (Noordijk, Delille, Schaffers, & Sýkora, 2009). It remains unclear, however, how these findings scale up because verge cutting results in the concurrent loss of floral resources across large areas, which probably impacts pollinators much more than suggested. To further understand the impacts of road verge cutting on pollinators, it is necessary to complement plot-scale experiments with larger scale studies.
In this study, we explore the role of road verges and their associated hedges in supporting insect pollinators in an agricultural landscape, and the impacts of traffic and verge cutting. We recorded flower and pollinator communities (orders: Hymenoptera, Diptera, Lepidoptera and Coleoptera) across the flowering season, in fields, hedges and different parts of road verges. Previous studies of road verges have often focused on single taxa (mostly Lepidoptera; e.g. Munguira & Thomas, 1992), few study sites (e.g. Noordijk et al., 2009), or do not explore the factors affecting pollinator communities in road verges (e.g. Cole et al., 2017). Our study is the first to explore the impacts of traffic and management on entire pollinator communities across a large number of sites. We tested the following hypotheses: H1 Road verges and hedges support a greater flower abundance, flower species richness and pollinator abundance than do field interiors.

H2
Verge hedges support a greater flower abundance flower species richness and pollinator abundance than do field hedges, because they are less exposed to agricultural practices.

H3 Verge edges have a lower flower abundance, flower species richness
and pollinator abundance than do verge centres, because they are more exposed to traffic pollution and disturbance.
H4 Road verges next to busier roads have a lower flower abundance, flower species richness and pollinator abundance, because they are more exposed to traffic pollution and disturbance.
H5 Road verges that have been cut have a lower flower abundance, flower species richness and pollinator abundance.

| Study sites
The study was carried out in 2018 at 19 study sites in Cornwall, United Kingdom (Appendix A). Study sites were selected based on: (a) Road type-not along a trunk road due to access restrictions and not in a residential area due to distinctive verge management; (b) Physical characteristics-road verge primarily consisting of grassland habitat, at least 5 m wide and 50 m long to allow the desired sampling regime, with adjacent farmland separated by a hedge rather than a fence; and (c) Safety-parking nearby and visibility for at least 75 m in both directions. A database of possible sites was created and, from this, permission was sought to access fields. Fields were those directly adjacent to the selected road verges, and were either arable or pasture, with a range of crops and management reflecting local agriculture. From this, we found 19 study sites that were at least 1 km apart. Road verges were 5-24 m wide. In the study area, most hedges are a characteristic regional style, called 'Cornish hedges'.
They consist of a stone-faced earth bank (generally 1.5 m high and 1.5 m wide at the base) with herbaceous vegetation growing from the top and sides, with or without significant woody vegetation.

| Transect locations
At each study site, we set up 50 m transects at six locations

| Site characteristics
We measured the width of each road verge at distances of 5, 15, 25, 35 and 45 m along each verge centre, and calculated a mean value.
We estimated the percentage cover of woody vegetation on the top of each verge hedge and field hedge transect to the nearest 10%. We measured traffic density by counting the number of vehicles passing by the road verge in either direction for 10 min, and repeated this on three separate days, between 09:00 and 16:30. During each site visit, road verge management was recorded, providing an estimated time of cutting that was accurate to within a two week period.

| Transects
We used transect surveys to compare the abundance and species The sampling regime at each study site. The six transect locations refer to: verge edge (VE), verge centre (VC), verge hedge verge-facing side (VHV), verge hedge field-facing side (VHF), field centre (FC), and field hedge (FH). Surveys of flowers and pollinators were carried out at each transect in three survey rounds: spring, early summer and late summer. Pan traps were placed out once in summer for 48 hr defined as one or multiple flowers that can be visited by an insect without having to fly between them, following Baldock et al. (2015).
Pollinator surveys were conducted when the wind speed was below Beaufort scale 5 and the temperature was above 17°C, or between 13-17°C if there was no more than 40% cloud cover (Pollard & Yates, 1993), except in the spring round when temperatures rarely met the minimum requirements. Temperature (°C) and wind speed (Beaufort scale) were recorded at the start of each survey. We walked each transect in both directions at a steady pace over 10 min and recorded all pollinators within 1 m either side of the transect and 2 m ahead. When a pollinator was observed visiting a flower, we also recorded the species of flower. Generally, bees, butterflies and hoverflies were identified to genus or species (if necessary taking voucher specimens), whereas beetles, non-syrphid flies and moths were identified to order (except for common and distinctive species) (full details in Appendix D). Given this variability in the level of taxonomic identification across and within groups, even for key groups such as bees and hoverflies, we did not calculate pollinator species richness from transect data, but instead used pan traps for this purpose (see below).
Sixteen plant species were excluded from further analyses because they had small flowers (< 5mm) that produce little nectar and pollen (Baude et al., 2016) and were rarely observed being visited by pollinators, or were horticultural varieties that are not visited by pollinators (Appendix E).

| Pan traps
Pan trap surveys were carried out in summer (17/07/2018-01/08/2018) to estimate the species richness of solitary bees and hoverflies in verge hedges and field hedges. At each study site, we placed pan traps along the verge hedge (verge-facing side) and field hedge ( Figure 1), with three pan traps per transect, at distances of 10, 25 and 40 m. Pan traps were plastic dishes (radius 7.5 cm) that had been sprayed with fluorescent yellow paint. At each location, vegetation was flattened in a 1 m 2 area where necessary, to ensure pan traps were visible. Each pan trap was raised 6 cm from the ground. Pan traps were filled with water to a depth of 3 cm, and a drop of nonscented detergent was added to break surface tension. After 48 hr, the contents of the pan traps at 10 m and 40 m were collected and stored in 70% ethanol. If a pan trap had been disturbed, we collected the contents of the pan trap at 25 m instead (< 10% of cases), otherwise it was discarded. We identified all bees and hoverflies to species.

| Statistical Analyses
All statistical analyses were carried out in R 3.5.1 (R Core Team, 2018), using generalized linear mixed effects models (GLMM) ('lme4' package; Bates et al.,6). Models were initially fitted using Poisson error structure. Fixed effects were scaled (divided by 10 or 1,000) where necessary to allow model convergence and models were checked visually to meet assumptions. In one case, residuals showed heteroscedasticity so a negative binomial error structure was used.
In each case, the link function was that which provided the lowest AIC. Models were tested for multicollinearity using variance inflation factors, which were < 5 in all cases.
For the transect data, we used two versions of the dataset to test different sets of hypotheses. In the first instance, we used the full dataset to test whether flower abundance, flower species richness and pollinator abundance were affected by transect locations (H1-H3, see Introduction). Flower abundance was modelled using a GLMM with negative binomial error structure. Flower species richness was modelled using a GLMM with Poisson error structure, log link function. Pollinator abundance was modelled using a GLMM with Poisson error structure and square root link function. In all cases, fixed effects included transect location and survey round, and the random effect was transect ID nested within site. In addition, the model for pollinator abundance included flower abundance divided by 1,000, wind speed (Beaufort scale) and temperature (°C) as fixed effects, and an observation-level random factor to address overdispersion. We did not include field type (arable or pasture) as a fixed effect because there were no apparent differences in the plotted data. The significance of the main effects and of pairwise contrasts between survey rounds and transect locations were assessed using likelihood ratio tests (LRT).
In the second instance, we used data for the verge edge and verge centre transects only, to test whether the abundance and species richness of flowers and pollinators in road verges were affected by traffic density, proximity to the road and verge cutting (H3-H5, see Introduction). All models were GLMM with Poisson error structure and square root link function, and fixed effects were survey round, distance from road (m) divided by 10, whether or not the verge had been cut during the study (Y/N), traffic density (vehicles 30 min −1 ) divided by 10, and an interaction between traffic density and distance from road. The random effect was transect ID nested within site, and the models for flower abundance and pollinator abundance also had an observation-level random factor to address overdispersion.
As above, the model for pollinator abundance had additional fixed effects of flower abundance divided by 1,000, wind speed and temperature. We used AICc to identify the best model ('MuMIn' package; Bartoń, 2016), whereby the best-fitting model was that with the lowest AICc. We considered models with ΔAICc < 2 and carried out model averaging on this top model set.
For the pan trap data, we explored whether there were differences in the species richness of hoverflies and solitary bees between verge hedges and field hedges (H2). We used a GLMM with Poisson error structure, square root link function and transect location as a fixed effect.

| RE SULTS
We recorded 143 plant, 45 hoverfly, 28 solitary bee, 8 bumblebee, and 17 butterfly species, as well as many beetles and non-Syrphid flies (summarized in Appendices B-D, and species lists in Appendices E and F).
We present the results of the overall models, and then describe the relevant results for each hypothesis. Transect location and survey round significantly improved models for flower abundance, flower species richness and pollinator abundance (Table 1). An interaction between location and survey round did not significantly improve these models (Table 1)  There was partial support for H2. Verge hedges generally contained little woody cover compared to field hedges, with only 3 of 19 verge hedges containing more than 50% woody cover, compared to 12 of 19 field hedges. Flower species richness was significantly greater along verge hedges (verge-facing side) than along field hedges, but not flower abundance or insect abundance (Figure 2; Appendix H). There was some evidence that pollinator abundance was greater along the verge-facing side of verge hedges than along the field-facing side (p = .0628; Figure 2; Appendix H), but otherwise there were no significant differences between hedge transects (Figure 2; Appendix H). The models examining species richness in pan traps were not significantly improved by including transect location for either hoverflies (LRT X 2 = .10, df = 4, p = .7524) or solitary bees (LRT X 2 = .85, df = 4, p = .3553), indicating no significant differences between verge hedges (verge-facing side) and field hedges (Appendix J).

H3 Verge edges have a lower flower abundance, flower species richness and pollinator abundance than do verge centres.
There was partial support for H3. Verge edges and verge centres often had different flower communities, with only four of the top 10 most common flower species at each being the same (Appendix C).
Pollinator abundance was significantly lower along verge edges than along verge centres, but flower abundance and flower species richness were not significantly different (Figure 2; Appendix H).

H4 Road verges next to busier roads have a lower flower abundance, flower species richness and pollinator abundance.
There was strong support for H4. For the subset of the data that included only verge edge and verge centre transects, the best-fitting models for flower abundance, flower species richness and pollinator abundance included survey round, whether or not the verge had been cut and traffic density (Tables 2 and 3). The best-fitting models for pollinator abundance also included flower abundance and distance from the road (Tables 2 and 3). Traffic density ranged from 4 vehicles 30 min −1 to 708 vehicles 30 min −1 and had a significant positive, but small, effect on flower abundance and flower species richness (Appendix K). Pollinator abundance significantly decreased with increasing traffic and significantly increased with distance from the road (Figure 3).

H5 Road verges that have been cut have a lower flower abundance, flower species richness and pollinator abundance.
There was strong support for H5, with flower abundance, flower species richness and pollinator abundance being much lower in verges that had been cut (Figure 4), despite the fact that transect surveys generally took place weeks or months after cutting had taken place. Road verges at 17 of the 19 study sites were cut, and 15 of these were cut just once during the study. Cutting mostly took place within a 2-month period between late-May and mid-July. At 12 study sites, the entire width of the road verge was cut. At the other five study sites, only the verge edge was cut, though in two of these cases the rest of the verge was cut at a later date. Cuttings were never removed from verges.

| D ISCUSS I ON
Our study has demonstrated that road verges and their associated hedges are important for supporting insect pollinators in agricultural landscapes, but that they are negatively affected by high TA B L E 1 The results of the likelihood ratio tests (LRT) testing the significance of the main effects for models assessing the impact of transect location and survey round on flower abundance, flower species richness and pollinator abundance  TA B L E 2 The best-fitting models (ΔAICc < 2.00) resulting from model selection of all possible combinations of terms from global models that assessed the impact of traffic density, proximity to road and verge cutting on flower abundance, flower species richness and pollinator abundance The models used the subset of the data that only included verge edge and verge centre transects. Conditional R 2 is not provided for the models of flower abundance and pollinator abundance because they contained an observation-level random factor to address overdispersion. hedges provided a much greater abundance and species richness of flowers, and were associated with a much greater abundance of pollinators than field interiors. We recorded reasonable numbers of flowers in some fields at some times of year, for example in between rotational grazing in pastures and in arable fields when weeds had flowered, but these occurrences were rare, of short duration, and inconsistent across survey rounds. This suggests that TA B L E 3 The coefficient estimates ± standard error (SE), confidence intervals (CI), test statistic values (z) and significance values (p) for the averaged model for: flower abundance, flower species richness and pollinator abundance. For each, the averaged model was taken from the top models with ΔAICc < 2.00 ( Table 2). The models used the subset of the data that only included verge edge and verge centre transects F I G U R E 3 The impact of traffic density and distance from the road on pollinator abundance at each 50 × 2 m transect during each survey. Data are for verge edge and verge centre transects only, which was the subset of the data upon which analysis was carried out with these explanatory variables. Plotted lines are model predictions (mean ± SE). Full model details are provided in Tables 2 and 3 (Ghazoul, 2006;Vaudo et al., 2015).  (Hanley & Wilkins, 2015). Hanley and Wilkins (2015) found that the road-facing side of verge hedges contained greater flower species richness, flower abundance and bumblebee abundance than the field-facing side. We did not find such clear differences, which may be because farming practices were less intensive in our study fields. Overall, these findings suggest that plant and flower communities are different in verge hedges and field hedges, but that they support similar numbers of flowers and pollinators.

Model
Contrary to H3, the abundance and species richness of flowers were similar between the verge edge and verge centre transects.
However, plant and flower communities changed greatly within the first 1-2 metres from the road, and flower species richness was affected by the amount of traffic on the adjacent road. This is probably an impact of physical disturbance, nutrient enrichment from vehicle emissions, and other forms of pollution. Previous research has shown that vehicle emissions result in nitrogen deposition in road verge soils: busier roads have higher soil nitrogen levels, which decrease with distance from the road but still have a measurable effect at 10 m away (Truscott, Palmer, McGowan, Cape, & Smart, 2005). In support of H4, pollinator abundance in verges decreased with increasing traffic and increased with distance from the road.
On the basis of our observations, we propose that this is primarily The impact of road verge cutting on flower abundance and pollinator abundance at each 50 × 2 m transect during each survey. Data are verge edge and verge centre transects only. Asterisks (***) indicate a significant difference between boxplots with p < .001. Full model details are provided in Tables 2 and 3 due to turbulence from passing road traffic, which makes it difficult for pollinators to forage at the verge edge, especially on busy roads. However, it may also have resulted directly from other forms of pollution, or indirectly due to the described differences in plant and flower communities. A study on honeybees showed that diesel exhaust pollution affected the odours of flowers that are used for foraging (Girling, Lusebrink, Farthing, Newman, & Poppy, 2013), and other roadside pollutants such as heavy metals can impact pollinators (Meindl & Ashman, 2013;Moroń et al., 2012), but have not yet been studied in the roadside environment. The finding suggests that: (a) Road verges that are only a few metres wide are of lower value to pollinators, unless along quiet roads, and (b) The first few metres of wider road verges are of lower value to pollinators.
Road verges that had been cut had many fewer flowers, flower species and pollinators (H5). It is obvious that cutting will initially result in a near-complete removal of flowers, but the numbers of flowers and pollinators were very low even weeks and months later. Grass cuttings were never collected, so road verge vegetation was covered in a thick thatch after cutting that was still visible months later and probably contributed to low flower abundances.
Existing studies suggest that removing verge cuttings benefits plant and insect diversity (Jakobsson et al., 2018), and flower abundance (Noordijk et al., 2009 Another limitation of our findings of the impacts of verge cutting is that the weather in the study year was particularly hot and dry, which may have accentuated the slow recovery of plant and flower communities. Our findings suggest that multiple cuts would be detrimental to flower production in drought years, at least if cuttings were not removed. This is an important consideration because climate change is predicted to increase the frequency of droughts, which reduce the overall availability of floral resources for pollinators (Phillips & Shaw et al., 2018). It is important that recommended management practices benefit pollinators in extreme weather years, as well as in typical years.

| CON CLUS IONS
In this study, we have shown that road verges and their associated hedges are important for supporting insect pollinators in agricultural landscapes, but their capacity to do so is reduced by heavy traffic and summer verge cutting. Generally entire road verges were cut in summer, regardless of whether it was necessary to do so to provide sight lines for road users, which suggests that there is significant opportunity in our study area to improve management for pollinators without conflict with safety requirements.
Our study is the first to show that busier roads have fewer pollinators in adjacent road verges, and that the verge edge next to the road contains fewer pollinators than the verge centre, probably due to turbulence and traffic pollution. In practice, given that pollinators do not use the verge edge as much, it may be beneficial to cut the first couple of metres from the road regularly for safety purposes if this means that areas further away from the road can be managed more favourably (with less cutting or better timing of cutting).
Additionally, regularly cutting the verge edge will remove floral resources, which may reduce the proximity of pollinators to vehicles and subsequent road mortality resulting from collisions (Keilsohn et al., 2018).
On the basis of our findings, we recommend that management of road verges that aims to benefit pollinators should prioritize wider road verges (at least 2 m wide), lower traffic roads, and areas away from the immediate vicinity of the road. Where possible, verge cutting should not be carried out during peak flowering times, or otherwise road verges should not all be cut at the same time, as it results in a period of few flowers.
There is growing mainstream pressure to utilize road verges for nature conservation. Protecting florally diverse road verges that represent species-rich grasslands is important, but these are often the exception rather than the rule. Even modest improvements to the management of the overall network of road verges and hedges will provide major benefits for pollinator populations, given the large areas that they collectively cover.

ACK N OWLED G EM ENTS
We would like to thank Alice Lord for arranging contact with land-