Study site and species
Field work was carried out on the University of Sussex campus in southeast England (50.8671° N; 0.0879° W). We repeated 10 identical three-day exclusion trials from May to September 2017. Data on bee foraging were collected only on days considered suitable for foraging, > 12 °C, with light winds and no rain, when honeybees and bumblebees were seen to be actively foraging on the lavender plants and/or on other flowers in the study area. There were two apiaries belonging to the Laboratory of Apiculture and Social Insects within < 1 km of the study site each (with between 6 and 10 colonies in total during the study period), plus three further apiaries within < 2 km, and a high density of colonies managed by beekeepers in the wider local area. Honeybees mainly forage for nectar and pollen from March to October (Couvillon et al. 2014a), and healthy colonies consist of between 20 and 40,000 adult bees in May–June and some 40,000 in September (Hooper 1991). Therefore, it is certain that foraging honeybees were present and abundant in the area throughout the study period.
We used the same lavender variety, Lavandula x intermedia ‘Grosso’ (Lamiaceae), as the previous research that demonstrated exploitative competition for nectar between Apis and Bombus in summer (Balfour et al. 2013, 2015a).
A total of 700 Grosso plants in 3 L pots were obtained from Downderry Nursery, Sussex (www.downderry-nursery.co.uk), the same supplier as for the previous competition studies (Balfour et al. 2013, 2015a). The plants had been grown in ways to cause bloom at different times. 300 plants were kept in greenhouses and polytunnels by Downderry Nursery to induce early flowering in May and June. 150 plants were grown normally, without treatment, and flowered in late July. A final batch of 250 plants were trimmed during the summer to delay bloom until September, with 150 plants used in the final two trials, 9 and 10. Some of this batch flowered in late August and 96 spare plants were used to replace plants that were near the end of their bloom in Trial 8 (21–24 August), to ensure a similar level of bloom across trials. Different growth regimes did not affect the general appearance of the plants and average nectar secretion rate was similar between batches (Results).
Trial design and experimental exclusions
The May to September study period was categorised into three seasons, spring (May and June), summer (July and August), and autumn (September). July and August were combined as summer since honeybee foraging distances are highest in the study area in these months, which indicates a dearth in overall nectar availability (Couvillon et al. 2014a). Pre-July study months were combined as spring. Autumn was defined according to the National Met Office definition of meteorological autumn as starting on 01 September (National Met Office 2019), and also coincided with the flowering of ivy (Hedera spp.) in the study area from early September, following Couvillon et al. (2014a).
Each trial consisted of three exclusion days. Exact trial dates were dependent on suitable weather conditions. We aimed to carry out an even number of trials per season, but this was not possible due to poor weather conditions in spring and the lack of lavender plants in full bloom in autumn following the final trial. We achieved three trials in spring (1–3: 23–25 May, 31 May–02 June and 13–15 June), five in summer (4–8: 04–06 July, 10–13 July, 31 July–04 August [data not collected on 02–03 August due to bad weather], 14–16 August and 21–23 August) and two in autumn (9–10: 12–14 and 19–22 September) making ten in total. We alternated trials between two sites 600 m apart on the University campus to reduce any potential local-effect bias.
Following Balfour et al. (2015a), each three-day trial was set up using 150 plants in three patches of 50 pots, separated by 100–200 m. Plants were selected at the start of the trial to give approximately equal total bloom per patch. Each patch was randomly assigned to a treatment: honeybees excluded (HBE), bumblebees excluded (BBE) and control (CON, no bees excluded). Following established methods (Balfour et al. 2015a), bees of the “wrong” type were excluded throughout each day using a light tap with a bamboo cane. On all patches we excluded male wool carder bees (Anthidium manicatum), since these are highly territorial and aggressive towards other bee species, and the conopid fly (Sicus ferrugineus) which lays its eggs on foraging bumblebees (Falk 2015), in case these insects were causing honeybees and bumblebees to avoid the lavender; both were rarely present.
We estimated the total number of flowers in each patch once during each trial by counting the number of flowering inflorescences in the patch and multiplying this by the average number of flowers calculated from 40 randomly-selected inflorescences.
Bee count data
Data collection followed established and effective methods for counting bees visiting flowers (Garbuzov and Ratnieks 2014b; Balfour et al. 2015a). We counted bees foraging in each patch from 09:00 to 17:30 on each trial day. To do this we made a near instantaneous count every 30 min in which we scanned the patch by eye for approximately 30 s and recorded any bees and other insects actively foraging at that time (Garbuzov and Ratnieks 2014b). In general, bees spend < 30 min in a patch during a single foraging attempt. Therefore, although individuals will revisit patches, the 30-min interval between counts means that the count data represent different visits (Garbuzov and Ratnieks 2014b). After 17:30 all patches were covered with netting to prevent insect access until targeted exclusions resumed the following morning (Balfour et al. 2015a).
Bumblebees, including parasitic cuckoo species (subgenus Psithyrus), were mostly identified according to species. The two-banded white-tailed bumblebees Bombus terrestris and B. lucorum are difficult to distinguish in the field and were grouped as B. terrestris/lucorum (Fussell and Corbet 1992). Solitary bees were identified according to species where possible, or to genus. Any bees that could not be recognised by eye were caught and identified using a hand lens or microscope. The vast majority of foragers were collecting nectar only and were only ever observed carrying trace amounts of pollen, supporting previous observations in which less than 5% of the foragers on Grosso were observed with pollen in their corbiculae (Balfour et al. 2013).
During each trial we measured secretion rate, standing crop and sugar concentration using microcapillary pipette tubes (Drummond Microcaps 1 µL, 64 mm, 1-000-0010-64 or 0.25 µL, 32 mm, 1-000-00025) inserted into an open flower to extract the nectar from the base of the corolla. The length of nectar drawn up into the tube was measured using a ruler and used to calculate the per-flower volume of nectar as a proportion of the overall tube volume (Corbet 2003; Balfour et al. 2013). Each microcap was used a single time only (Corbet 2003). Nectar measurements were made once per trial, between 12:00 and 14:00 to minimise day-to-day variation.
To measure the per-flower volume of nectar available to insects (standing crop) we extracted nectar from 10 flowers in each patch. Nectar sugar concentration (% Brix) was measured for each sample with sufficient volume using a hand-held refractometer (Bellingham and StanleyTM, 0–50% Brix). To measure hourly nectar secretion rate per flower we used microcaps to empty as fully as possible several flowers in the CON patch, taking care not to damage the nectaries (Corbet 2003). We marked these flowers and bagged the entire inflorescence using fine gauze bags to prevent insect access. After 60 min, we extracted nectar from the marked flowers individually and recorded the volume of liquid contained in the microcap.
We analysed seasonal changes in honeybee visits to lavender flowers when bumblebees were manually excluded (BBE patch) relative to the control (CON) patch over ten trials. The following statistical analysis uses the second and third exclusion days of each trial, when bee numbers and foraging behaviour had stabilised following one full day of exclusions. This is because we observed that honeybee numbers on the BBE patch often varied considerably over the course of the first trial day, which is consistent with previous research in which honeybee numbers took approximately 1.5 days to plateau following the start of bumblebee exclusion from lavender patches (Balfour et al. 2015a, b). To remove this noise in the data, we removed the first trial days from analysis.
As a proxy measure of competition we calculated the absolute difference in per-day mean honeybee counts from 09:00 to 17:30 (n = 18 counts per day) between the two patches [(mean HB(BBE))–(mean HB(CON))], hereafter HB(BBE-CON), since this metric gives a clear indication of the increase in honeybee visits to the BBE patch compared to the control. Using daily average counts removed pseudo-replication from the raw data, and normalised the positively skewed distribution, thereby also correcting for overdispersion. HB(BBE-CON) also accounts for any between-trial variation in the number of flowers.
To analyse between-season variation in HB(BBE-CON) we used a linear mixed effects model [lmer, package lme4 (Bates et al. 2015)] with per-day HB(BBE-CON) as the response (n = 20) and season (spring, summer, autumn) as a fixed effect. Trial was included in the model as a random effect since we expected between-trial variation in HB(BBE-CON), but were not directly testing differences in the response between specific trials in this model (Bolker et al. 2009). Trial day (2 or 3) and site were added as interaction terms to assess any confounding effect on HB(BBE-CON) with the effect of season, but neither were significant and so were not included in the final model. Residuals were visually checked for normality and homoscedasticity, and approved. Differences between seasons were calculated using post hoc pairwise comparisons across groups, using lsm [package lsmeans (Lenth 2016)] within glht [package multcomp (Hothorn et al. 2008)], with P values adjusted for multiple comparisons by the single-step method.
We did not expect honeybee exclusion (HBE) to impact bumblebee visitation, given previous results (Balfour et al. 2015a), although a seasonal effect was possible and worth investigating since the previous study was conducted only in summer (July–August). In fact, honeybee abundance on control patches was consistently low, and there was no increase in bumblebee numbers on the HBE patch relative to the control (Fig. 1; Online Resource 1). Therefore, seasonal changes in bumblebee abundance were not explored statistically. We also did not analyse the effects of honey- and bumblebee exclusion on other insect groups since frequencies of these were too low for statistical analysis.
Nectar standing crop volumes were compared between patch treatments in each trial using per-trial Kruskal–Wallis H tests and post hoc Dunn’s tests for pairwise comparisons between treatments with Bonferroni adjustment of P values (results in Online Resource 3). One-hour nectar secretion rates were compared between batches of lavender plants (n = 4 batches) and between trials (data available for eight of 10 trials) using Kruskal–Wallis H tests for non-parametric data.
Significance was defined at P < 0.05. All analyses were performed using R Studio Version 1.1.419.