Data comprise flower abundance and diversity data and bee abundance, diversity and activity data collected during extensive surveys carried out on farms in Hampshire and West Sussex, southern England between 2013 and 2015. The pollen diets of wild solitary bees were quantified using direct observations and pollen load analysis. The purpose of the study was to provide valuable information to scientists, governments and land managers in designing more effective measures to conserve the broader wild bee community on agricultural land. The work was funded by the Natural Environment Research Council grant NE/J016802/1 and the Game and Wildlife Conservation Trust.
Publication date: 2016-06-06
Nine HLS and ten ELS farms were selected in Hampshire and West Sussex, UK. The selected HLS farms had been implementing an average of 5.56±0.13 ha of pollinator-focused flower-rich schemes representing 2.17±0.05% of the farm area by ownership for a minimum of three years. As 70% of farms in England were at the time in some form of environmental stewardship (Elliot et al. 2010), ELS farms were chosen as the control group for this study. Flower-rich schemes were available under ELS, but these schemes had a low uptake so only basic ELS farms without such management were selected for this comparison. Pollinator-focused flower-rich schemes were typically established with a seed mix containing c.15-30 flowering forb species (Carvell et al. 2007; Pywell et al. 2011). Additional plant species such as Hypochaeris radicata and Trifolium repens are sometimes included in experimental mixes (i.e. Scheper et al. 2015), but this did not represent the situation in our study area and so these species were not characterised as sown. Whilst there were no such flower-rich areas on ELS farms, most of the species included in these seed mixes can be found growing in a wild state on these farms. Consequently, in order to allow a comparison of pollen choice preferences and relative rates of utilisation across farm types, plant species included in pollinator-friendly agri-environment schemes were characterised as 'sown' even when found growing wild as part of the wider plant community. For a full list of the plant species characterised as being sown as part of pollinator-focused management see Appendix S1 in Supporting Information. Farms were predominantly arable, or mixed arable/dairy with wheat, barley, oilseed rape and permanent/silage grassland as the major crops.
Bee and floristic surveys
In 2013 and 2014, a standardised 3 km transect was designed for each farm, passing through all major habitat types present. For HLS farms this included pollinator-focused flower-rich schemes (HE10 floristically enhanced grass margins, HK7 species-rich grassland restoration, HF4 pollen and nectar mixes), non-agricultural grass margins and hedgerow and woodland edge habitats. For ELS farms only non-agricultural grass margins and hedgerow and woodland edge habitats were surveyed, as no pollinator-focused schemes were present. Crops and areas of agricultural grassland were not surveyed. Each transect was subdivided into discrete sections, with each section covering a distinct habitat type. Transects on HLS farms were designed to survey as many pollinator-focused schemes as possible whilst remaining contiguous and passed through an average of 1496±148m of flower-rich habitat in an average of 3.77±0.24 discrete habitat patches per farm.
Bee activity was recorded along the transect following standard bee walk methodology (Carvell et al. 2007), with all bees within 2 m of the recorder identified to species level. Individuals that could not be identified in the field were netted for later identification. The first flowering plant species visited and the purpose of the visit, for either pollen or nectar, was recorded. Hylaeus species, which lack scopal hairs on their body, instead ingesting pollen and regurgitating it in the nest, cannot reliably be determined to be foraging for pollen and so all plant visits were recorded simply as visits. On each transect, the number of species of flowering plants and the number of flowering units of each plant species within 2 m of the recorder was estimated within each discrete transect section. Grasses, sedges and rushes were not recorded as these plant species are not attractive to bees in the study region. This assessment followed Carvell et al. (2007) with one flower cluster (e.g. an umbel, a head, a capitulum) counted as a single unit. Sixteen farms (eight HLS, eight ELS) were surveyed in 2013. Transects were walked three times through the season, between 25th May to 5th June, 26th June to 15th July and 3rd to 11th August. Seventeen farms (eight HLS, nine ELS) were surveyed in in 2014. Transects were walked three times through the season, between 17th and 27th May, 21st June to 9th July and 3rd to 15th August. These discrete sampling blocks are henceforth referred to as 'sampling rounds'.
In 2015 farms were surveyed for a fixed period of time rather than using distance based transects. ELS farms were surveyed for 3 hrs with 1.5 hrs spent on non-agricultural grass habitats and 1.5 hrs on woody hedgerow/woodland edge habitats. HLS farms were surveyed for 3 hrs with 1 hr on pollinator-focused flower-rich schemes, 1 hr on non-agricultural grass habitats and 1 hr on woody hedgerow/woodland edge habitats. The survey followed standard bee walk methodology as described above, but at a reduced pace to ensure thorough sampling. All bees within 2 m of the recorder were identified to species level. The first flowering plant species visited and the purpose of the visit, for either pollen or nectar, was recorded. Solitary bees with clearly visible pollen on their body were collected, placed in individual Eppendorf tubes and frozen. The collection of pollen loads from foraging bees may overestimate pollen use of more easily observable flowers. Ideally, pollen would be sampled from bees as they return to their nest, but this method was not chosen for this study as is often time consuming and may lead to low sample sizes for species with difficult to locate nests. All flowering plant species present on the transects were recorded, but their abundance was not quantified. Pollen samples from insect visited flowering plant species present were collected to form a pollen reference library. Pollen reference slides were prepared by transferring pollen-laden anthers to a drop of water on a microscope slide. The slide was gently heated to allow grains to absorb water and achieve their maximum size and to evaporate excess water. The remains of the anthers were removed, molten glycerine jelly stained with fuchsin was added and the slide was sealed with a coverslip. For a full list of sampled flowering plant species see Appendix S2. Fourteen farms (7 HLS, 7 ELS) were surveyed in 2015. Transects were walked four times throughout the season, between 22 April adn 13th May, 26th May to 17th June, 25th June to 4th July and 29th July to 10th August. All bee surveys were conducted between 0930 and 1700 hrs when the temperature was above 13°C with at least 60% clear sky, or above 17°C with any level of cloud. No surveys were conducted when it was raining. All bee and floristic surveys were conducted by the same individual (TJW) to minimise recorder bias.
The scopal pollen load of foraging solitary bees collected in 2015 was analysed by light microscopy using the method outlined by Westrich and Schmidt (1986). Before removing pollen from the scopae, the total load was estimated relative to a full load for that species, ranging from 8/8 (full load) to 1/8 (one eighth load). The pollen grains were removed from the scopae using an entomological pin and transferred to a drop of water on a microscope slide. Pollen that was not clearly held in the scopae was not sampled as this may have become attached to other parts of the body during nectar visits to non-host plant flowers. The slide was gently heated to allow grains to absorb water and achieve their maximum size and to evaporate excess water. Molten glycerine jelly stained with fuchsin was then added and the slide was sealed with a coverslip. The proportion of the load comprised of different plant species was estimated along three randomly selected lines across the cover slip at a magnification of x400. The proportion of the load by volume was estimated by the relative area of the slide occupied by each plant species, rather than the absolute number of grains, in order to better reflect the total volume of pollen collected, an important correction in mixed loads where pollen grains of different plant species often differ widely in size (Cane and Sipes, 2006). Species representing less than 1% of the load were excluded from further analysis as their presence may have arisen from contamination (Westrich and Schmidt 1986).
The proportions of pollen collected were corrected according to the overall size of each load to give a final weight, e.g. a full load (8/8) comprised of 50% Centaurea nigra and 50% Leucanthemum vulgare would receive a final C. nigra weight of 50 and a final L. vulgare weight of 50, whereas a quarter load (2/8) comprised of 100% Hypochaeris radicata would receive a final H. radicata weight of 25. The pollen grains were identified to species using Sawyer (1981) and the reference collection assembled during the project. The majority of samples were identified to species level, but where this was not possible pollen was identified to genus, for example in Brassica, Plantago and Geranium. For a full list of taxa and the level of identification, to either species or genus, see Appendix S3.
Generalised Linear Mixed-Effect Models (GLMMs) were used to test for the impact of management type on bee and plant species abundance and diversity and the impact of plant species richness on bee species diversity and diet breadth. Models were fit using the maximum likelihood (Laplace Approximation) method. All data analyses were conducted in R version 3.1.1 (R Development Core Team) using the lme4 package for the GLMMs (Bates et al. 2014). All models were fitted with Poisson and negative binomial error distributions and were tested for overdispersion. In all cases negative binomial error structures were the most appropriate and final models were not overdispersed. Final models were compared by ANOVA with a null model containing the same random factor to test for significance.
Differences in the total number of bee and plant species and total floral abundance recorded between different farm types were analysed using GLMMs with management type as a fixed factor. Sampling year was included as a random factor to take account of the temporal data structure and differences in sampling methods. The abundance analysis used the 2013-2014 data and the species richness analysis used the 2013-2015 data. The impact of plant species richness on bee species richness (including Apis, Bombus and kleptoparastic bee species) and oligolectic solitary bee species richness was analysed using GLMMs with plant species richness as a fixed factor and sampling year as a random factor. This analysis used the 2013-2015 species richness data.
The impact of plant species richness on the number of pollen species detected in bee pollen loads was analysed using a GLMM with plant species richness as a fixed factor and sampling round (April/May, May/June, June/July and July/August) as a random factor. The number of pollen species detected in bee pollen loads was also calculated for the seven most common polylectic bee species for which a total of 30 pollen loads had been collected from each species, representing the majority of the pollen load data (759 of 1054 samples, Andrena chrysosceles, A. flavipes, A. haemorrhoa, A. semilaevis, Lasioglossum calceatum, L. malachurum and L. pauxillum). The number of species detected in pollen loads was summed over the year for each species to reduce temporal variation. Farms where no samples of a species were taken were excluded from that species' analysis, as the species may have been absent from the sample for reasons other than floristic composition, e.g. nesting site availability, low detection rate etc. The relationship between plant species richness and the number of pollen species collected by polylectic bee species was analysed using a GLMM with plant species richness as a fixed factor and bee species as a random factor. Both these analyses used the 2015 microscopic pollen load analysis data.
The proportion of sown flowers relative to total flowers was calculated for each farm over the 2013-2014 period. The proportion of observed solitary bee pollen visits to sown flowers and the proportion of solitary bee species visiting sown flowers for pollen was also calculated over the 2013-2014 period. The impact of the proportion of sown flowers on the proportion of observed solitary bee pollen visits and the proportion of solitary bee species visiting sown flowers was analysed using Spearman's rank correlation tests, as in each case the response variable could not be transformed to normality.
Differences in the proportion of pollen collected from different plant types were analysed using binomial tests. For the observational data, the proportion of pollen visits to sown and wild plants was calculated for each sampling round across all years for both farm types. For the pollen load data, a third category of crop plant data was included. A number of pollen loads contained Brassica type pollen, most of which is highly likely to have come from the crop plant oilseed rape Brassica napus. No wild Brassica species such as B. nigra were recorded during floristic surveys with the only other source being small areas of B. rapa that is sometimes sown as part of conservation management for birds. As a result, we are confident that the majority of the Brassica type pollen originated from crop plants and so this was excluded from the comparison between sown and wild plant pollen use. As the pollen load data is non-integer (with variably full pollen loads with mixed species composition), the proportion of each pollen type was used to calculate an appropriate value from the number of collected samples, i.e. where 173 bees were collected with pollen loads in total comprised of 9.7% pollen from sown plants and 90.3% pollen from wild plants by volume this was calculated as 17 samples from sown plants and 156 samples from wild plants. These calculated values were used in the binomial tests.
References avaialable from Supporting Documentation.