Bumble bees exploit known sources but return with partial pollen loads when foraging under low evening light

Diel activity of diurnal pollinating insects declines well before the end of day, but a few species, such as bumble bees, can continue foraging in the period before sunset, when light levels decrease. Foraging during this time could be costly for individuals, and key decisions regarding the daily cessation of foraging behaviour have to be made during this period of rapid abiotic change. However, little is known about the behaviour of diurnal bees under these light conditions. We conducted a ﬁ eld experiment in a mixed agricultural landscape to investigate pollen foraging in the hour before sunset in the buff-tailed bumble bee, Bombus terrestris , comparing it to an hour during the afternoon. We monitored activity at the colony entrance and collected pollen sacs from returning foragers. Overall, there were fewer exits and returns in the evening period than in the afternoon period. Evening foragers returned to the colony with partial pollen loads when compared to daytime foragers, and those returning in lower light levels had smaller loads. Additionally, we found that the same ﬂ ower taxa were being exploited in the evenings as in the daytime. We found no effect of individual size on the likelihood of evening foraging or the size of pollen loads. Our results suggest that B. terrestris are facultative evening foragers that opportunistically exploit pro ﬁ table ﬂ owers they have encountered during the day. Evening foragers decide to return to the colony with partial pollen loads in this period of the foraging day, which results in a largely reduced pollen in ﬂ ux to the colony at the end of the day. Our ﬁ ndings indicate that the gradual decline in light is likely to be an important factor guiding the decisions of bumble bees and other diurnal pollinators of whether to extend or end foraging, and when to return to the colony.

Relatively rapid changes in light conditions occur daily after sunrise and in the approach to sunset, when abiotic light is getting brighter or dimmer. While these conditions only prevail for a limited time each day, diurnal pollinators may utilize this intermediary time between day and night to maximize their active hours. However, it appears that these times are not part of the typical pollinating day, as few diurnal pollinators are active, probably due to constraints on their vision, which guides flight and navigation (Kelber et al., 2006;Warrant & Dacke, 2011). Bumble bees are diurnal pollinating insects which have been observed foraging in the early morning and late evening (Corbet, 1978;Hall et al., 2021;Spaethe & Weidenmüller, 2002); however, surprisingly little is known about the activity of foragers at times when light levels are low. In the present study, we therefore observed bumble bee foragers as they departed and returned to their colony in the late evening under rapidly declining light levels. Predictions for the foraging success of bumble bees in dimming light conditions can be informed by the foraging habits of true nocturnal bees. Somanathan et al. (2020) presented a comprehensive field study of the foraging niche of the Indian carpenter bee, Xylocopa tranquebarica. Through observation and pollen collection from wild individuals, the study revealed that the flower species visited by X. tranquebarica significantly overlapped with the choices of sympatric diurnal bees from the same genus. The pollen loads collected by X. tranquebarica were smaller than those on their diurnal counterparts, and floral constancy was low. Furthermore, there was no observed relationship between the nocturnal bees' foraging choices and floral availability. Together, the results indicated that nocturnal bees retrieve 'leftover' pollen, which has already been depleted during daylight hours, and that they forage opportunistically (Somanathan et al., 2020). Eveningforaging diurnal bumble bees could experience this same effect, and an evident cost of evening foraging could be a lack of abundant resources, leading to lower profitability of foraging bouts. However, for the X. tranquebarica population there is enough pollen available at night to sustain this niche (Somanathan et al., 2020). Diurnal bumble bees going out in the evening must also have high enough levels of resources available to motivate foraging at this time.
Visual constraints are likely to be the main limiting factor reducing pollinator activity in the late evening. The ability of a forager to exploit a floral resource effectively or complete a foraging bout may be compromised in dim light due to the lower sensitivity of their diurnally adapted eyes. For comparison, tropical nocturnal bees such as X. tranquebarica or Megalopta genalis, and other colletid and halictid bees (Kelber et al., 2006;Liporoni et al., 2020;Somanathan et al., 2019), have evolved adaptations in their apposition compound eyes that enhance their sensitivity and ability to forage under extremely dim light. For diurnal bumble bees, laboratory experiments have revealed the behavioural changes that occur when individuals fly in dim light conditions; in particular, they flew slower, and completed more meandering routes down a flight tunnel (Reber et al., 2015). Furthermore, there is some evidence that locating resources takes longer under dim light (Chittka & Spaethe, 2007). Although dim light had very little effect on the landing accuracy of bumble bees in experiments in which they were presented with a simple platform (Reber et al., 2016), this could be different for landings and handling of real flowers. If dim light negatively impacts the flight performance, navigation or resource-handling ability of bees in the field, we may expect a reduction in pollen acquisition efficiency from evening foragers. Additionally, foraging flights are energetically costly and risky for individuals, and conducting a foraging bout in the approach to sunset increases the risk of it being too dark for the bee to navigate home at the end of the bout (Kelber et al., 2006), and therefore increases the risk of staying out of the safety of the colony overnight. These theoretical costs in the late evening could lead to individuals deciding not to leave the colony to forage or deciding to end the current foraging bout before the light levels become too dim. Other abiotic factors that may change at sunset and could affect behaviours alongside light include temperature and humidity, which are well documented as impacting bee activity (e.g. Sanderson et al., 2015;Willmer & Stone, 2004).
The decision to forage during the evening could also be influenced by individual size. Bumble bee workers vary considerably in size (Frasnelli et al., 2021;Goulson et al., 2002;Jandt & Dornhaus, 2014;Kapustjanskij et al., 2007;Yerushalmi et al., 2006), which may lead to alloethism (task differentiation by size). Larger workers, which have larger eyes (Kapustjanskij et al., 2007;Spaethe & Chittka, 2003;Taylor et al., 2019), may be able to forage in dimmer light than their smaller counterparts. In colonies of the buff-tailed bumble bee, Bombus terrestris, larger individuals were more likely to initiate foraging at earlier times and under dimmer light conditions after dawn (Hall et al., 2021). It may be more costly for smaller foragers to perform visually guided behaviours in dim light, and they may be more at risk of dying from cold if they are unable to return to the colony overnight. We predicted that smaller bees would be less motivated to forage before sunset, and more likely to return in brighter conditions than larger bees, mirroring earlier findings by Hall et al. (2021). If so, fewer workers will be motivated to forage in dim light compared to in bright conditions, and although larger workers are thought to carry larger loads (Goulson et al., 2002), an overall reduction in resource influx would be expected.
To summarize, evening foragers will hypothetically gain less resource and risk higher costs than when foraging in the daytime. However, since bumble bees have been observed on flowers during the evening, there must be some benefits of foraging in dimmer light, such as competition avoidance (Smith et al., 2017;Wcislo et al., 2004). They could also maximize the potential foraging window, but importantly also access profitable flowers, which would motivate bees to utilize this period even if there are higher costs.
We conducted a field experiment in a mixed agricultural landscape to investigate pollen foraging in B. terrestris, in the hour before sunset, comparing it to an hour during the afternoon. We predicted that foragers would continue to exploit known resources in the evening that were accessed during the day. Overall, we expected individuals to bring back less pollen per foraging trip in the evening, as a result of dimming light levels. We hypothesized that in the approach to sunset, light levels would impact the decision to return to the colony, either by synchronizing returns across all bees or by triggering returns relative to the size of the individual. Alternatively, we hypothesized that individuals would return once they carried a full pollen load, similar in size to those collected during the day.

Bumble Bee Colonies
We used eight commercially reared colonies of the U.K. native buff-tailed bumble bee, B. t. audax (Koppert Biological Systems, Rotterdam, Netherlands) during the experiment, in two groups of four colonies, during July and August, 2020. During delivery and transfer to the field site, bees were not permitted to leave the colony and were supplemented with sugar syrup (Attracker, Koppert Biological Systems) and pollen until the first day in the field, after which no supplements were provided.
The field site was in a mixed agricultural landscape (South Hams, Devon, U.K.) which is typical for the southwest of England. We kept the bee colonies in custom-made outdoor boxes containing corridors and separating sections for capturing and marking returning bees. We placed the boxes approximately 5 m apart in a mixed wildflower and wild grass border of a barley field, adjacent to fields containing turnips, winter wheat, pasture and red clover. Each field was bordered with healthy hedgerows. As the experiment took place during July and early August, no mass flowering field crops such as oilseed rape, Brassica napus napus, were in flower in the local area, although floral availability is still high around this time of the year in similar U.K. landscapes (Timberlake et al., 2019).

Procedures
Bees were allowed to exit and re-enter the colony throughout the experiment. We undertook individual tagging of workers on the first 2 days after transfer to the field and opportunistically afterwards during the pollen collection process, using coloured, numbered queen-marking tags. Newly emerging queens, which were easily distinguished by their large size, were not tagged. This period of tagging allowed time for foragers to explore the site and locate exploitable resources.
We collected pollen from foragers returning to each colony in the afternoon and in the evening, by gently capturing individuals as they entered the tunnels of the external box, but before they entered the colony, and removing the pollen with a toothpick. We collected 372 pollen samples from 277 individuals over 15 nonconsecutive days (Group 1: 8 days; Group 2: 7 days).
Specifically, the collection periods were (1) an hour between approximately 1400 and 1500 hours (daytime), and (2) an hour before sunset (evening), with daily sunset times taken from the Met Office mobile weather app (The Met Office). Sunset was at 2120 hours on the first collection day, and advanced to 2048 hours on the last collection day. On 3 days it was not possible to complete both recordings due to rain. Light level (lx) and temperature ( C) were recorded every 30 s throughout the experiment using an automated data logger (HOBO Pendant #UA-002-64, Onset Computer Corporation, Bourne. MA, U.S.A.), which was attached to the exterior of one of the four colony boxes and remained in the same location throughout the experiment.

Data Acquisition and Analysis
During the observation times, we videorecorded bees entering and leaving each nest entrance. Exit and entry events were extracted automatically using a custom-written code in MatLab (R2018b, MathWorks, Inc., Natick, MA, U.S.A.). Extracted frames allowed us to categorize events as departures or returns and to determine the corresponding light level by matching the time stamp against the light recordings. As some videos were shorter than the expected 1 h due to technical issues, exit and return counts for a full hour were extrapolated and rounded to the nearest integer for videos longer than 50 min (N ¼ 15 of 105 total). We did not use videos shorter than 50 min (N ¼ 10) in the statistical analysis of exit and return counts.
We froze pollen samples at e20 C until the end of the field work and then transported them to the laboratory for analysis. We defrosted each pollen sample before weighing it on 0.0001 g precision Pioneer PA114C scales (OHAUS Corporation). Most samples consisted of two sacs, one from each corbicula (N ¼ 276); however, some contained only one sac (due to loss in the field; N ¼ 51) or were fragmented (N ¼ 45). We weighed all samples in full. The weighing boat was cleaned with ethanol between samples to prevent cross-contamination.
Following weighing, we mounted each sample on a microscope slide, following a procedure adapted from Sawyer (1981); 371 samples were mounted as one was lost following weighing. For each sample, we dissected a thin cross-section of the centre of one pollen sac onto a microscope slide and added a drop of distilled water. For fragmented samples, we mounted an equivalent amount of fragment since dissection was not possible. We mixed the pollen into the water using a toothpick and left the slide on a hotplate at 50 C until the water evaporated. We then added a drop of melted glycerine jelly with fuchsin dye (Brunel Microscopes Ltd., Chippenham, U.K.) as a mountant, and topped it with a cover slip, which was gently pressed to spread the mountant across the sample. The slide was left on the hotplate for 10 min before being left at room temperature for 24 h to set, after which we sealed the cover slip with clear nail varnish.
To create a reference pollen collection, we collected flowers from 33 plant species at the field site. These flowers were found in the experimental and adjacent fields, and along an approximately 200 m stretch of road-facing hedgerow leading to the field. We visually identified the flowers to the lowest taxonomic ranking possible using a combination of Blamey et al. (2013) and the Seek mobile app (v2.9.0, iNaturalist, California Academy of Sciences and National Geographic, San Francisco, CA, U.S.A.). We then collected pollen from the anthers and mounted these samples in the same manner as the bee-collected samples.
We photographed the bee-collected samples and the floral reference collection at 100Â, 400Â and 630Â magnification using a Leica DM 1000 microscope and integrated Leica DC295 camera (Leica Microsystems, Wetzlar, Germany). We used these photographs to identify pollen to the lowest possible taxonomic level by its morphological features, including size, by: comparing photographs to the study reference collection, using the identification key in Sawyer (1981) and accompanying digitized reference photographs (Chandler & Rennison, 2005), and using the online repositories Pollen Wiki (pollen.tstebler.ch) and Global Pollen Project (globalpollenproject.org). Each identified taxon contained only one morphotype (i.e. morphotypes were not grouped under higher taxonomic ranks). We also recorded the number of different taxa in each of the 371 samples, and calculated the percentages of samples (or subsets of samples where relevant, e.g. afternoon samples only) containing each taxon. We categorized taxa in each mixed-taxa sample into abundance categories ('majority' and 'minority') by visual inspection of the slide; if two taxa were assessed as equally abundant, we allocated both as 'majority', and where fewer than five grains were present in the sample, we allocated the taxon as 'trace'. To determine whether a taxon was frequently found as a single taxon, a majority taxon within a mix, or a minority taxon within a mix, we calculated the percentage occurrences of the taxon as that category (majority or minority) within single-or mixed-taxa samples, respectively, for both time periods. Those taxa that achieved greater than 10% representation in a specific category (e.g. classified as a minority taxon in >10% of afternoon mixed samples) are identified in Fig. 4 in the Results. We did not investigate trace taxa in this way, although they were included in the count of different taxa per sample.
At the end of the experiment, we froze and subsequently dissected the bee colonies, and measured the individuals that had retained their identification tags and had an associated pollen sample (N ¼ 128), as bumble bee foragers vary greatly in body size (e.g. Goulson et al., 2002). We took two measurements, using digital callipers (Louisware): the thorax width (distance between the wing attachment points, i.e. intertegular distance, in mm), and the length of the wing (distance from the wing attachment to the wingtip of the right wing, in mm). We made each measurement three times per individual and recorded the mean values. The two measurement types were highly correlated (Pearson correlation coefficient ¼ 0.81); therefore, only thorax width was subsequently used as this is an established measure of bumble bee body size (Cane, 1987;Hagen & Dupont, 2013).
We used generalized linear mixed models (GLMMs), using the lme4 package (Bates et al., 2015), to assess the difference in colony activity (exit count and return count, separately; N ¼ 95 counts for each) between the afternoon and the evening; both models used a negative binomial residual distribution and log link function. The random factors that were included in each model were colony identification (1e8) nested within Group (1 or 2) and calendar date nested within Group to account for daily variation such as in the weather or resource availability and other seasonal changes.
We analysed the masses of the full pollen samples (i.e. samples with two sacs) to explore the effects of time period and light condition on how much pollen the bees returned with. The following three GLMM models used a gamma residual structure and inverse, inverse and square-root link functions, respectively. The random factors were the same as described above.
To determine whether foragers acquired the same amount of pollen during the evening as in the day, we modelled pollen load mass by the fixed factor period (afternoon or evening). One sample was included per bee (N ¼ 210). For bees with repeat measures, the first chronological evening sample was preferentially selected due to the larger number of afternoon samples; if no evening sample was present, we selected the first chronological afternoon sample from that bee for inclusion.
To investigate the effects of declining light levels on pollen load mass, we modelled the pollen load mass of evening samples by (1) time until sunset upon the return of the forager to the colony (N ¼ 68) and, separately, (2) light level at time of return (N ¼ 63). These two models also included only one sample per bee: for bees with more than one evening sample, we included the first chronological sample. Light level was scaled by its mean to account for the large difference in scales of the variables.
To assess the masses of pollen from repeat samples, we conducted a paired Wilcoxon signed-rank test on samples collected from bees with at least one measure from each of an afternoon and an evening period. We included the first chronological measure of each period for each bee (i.e. two samples per bee, N ¼ 19 bees).
Using a one-way ANOVA test, we compared the thorax width of bees in three categories: bees only sampled in the afternoon, bees only sampled in the evening, and bees that were sampled in both time periods.
We modelled the thorax width of bees returning with samples in the evening as a response variable by the fixed factor 'time until sunset at return' (N ¼ 61), and separately modelled against scaled light level at time of return (N ¼ 60). For bees with repeat evening samples, we only included the first chronological evening sample in this analysis. We used GLMMs with a Gaussian residual structure and inverse link function. For both models the random factors were the same as described above.
To determine whether pollen load mass was related to bee size, we modelled load mass by thorax width in a GLMM, using a gamma residual structure with inverse link function and the same random factors as described above. We included one sample per bee (N ¼ 105), selecting the first chronological evening sample, or if absent the first afternoon sample, for bees with repeat samples.
GLMM output values are given with residual degrees of freedom, and mean values are presented with standard errors.

Ethical Note
This research adheres to the ASAB/ABS Guidelines for the Use of Animals in Research, the legal requirements of the U.K. and the institutional guidelines of the University of Exeter.

Colony Activity
The number of exits and the number of returns were both significantly lower in the evening than in the afternoon period (exits: GLMM: z 88 ¼ e8.32, P < 0.001; returns: z 88 ¼ e6.19, P < 0.001; Fig. 1). Exit counts decreased from a mean of 34.1 ± 4.4 in the afternoon hour to 9.7 ± 1.5 in the evening hour, and return counts decreased from a mean of 54.1 ± 4.5 to 29.4 ± 3.2 in the afternoon and evening hours, respectively.
Foragers continued to exit and return to the colony throughout the evening period, although activity declined towards sunset (Fig. 2). The latest recorded exit time was 1.3 min before sunset, at 376.7 lx, and the latest recorded returns were at sunset (i.e. 0 min before sunset), under slightly brighter conditions at around 440e600 lx. It is also possible that either exits or returns could have occurred after sunset, although our trend suggests that counts would be very low. Since return times in the final hour of the day were not clustered, it is unlikely that a set light threshold triggers the return, even if bumble bees were returning from different distances from the colony (Osborne et al., 2008;Walther-Hellwig & Frankl, 2000).

Pollen Mass
Comparing the pollen loads between the two time periods, foragers brought back significantly smaller pollen loads in the evening, when mean mass was 0.0153 ± 0.0012 g, compared to the afternoon, when mean mass was 0.0268 ± 0.0012 g (GLMM: t 203 ¼ 5.12, P < 0.001; Fig. 3). A similar result was found for paired repeat data from foragers that had samples in the afternoon and evening periods; these foragers also returned with less pollen in the evening than the afternoon (Wilcoxon signed-rank test: For samples collected during the evening, foragers returning in lower light had smaller loads (GLMM: t 56 ¼ 2.258, P ¼ 0.024). However, the time until sunset at the return of the forager did not have a significant effect on pollen load mass (GLMM: t 61 ¼ e1.193, P ¼ 0.233).
All taxa with more than one occurrence were found in both afternoon and evening samples. Blackberry was found in 68% of afternoon samples and 65% of evening samples, while red clover was found in 26% of afternoon and 25% of evening samples. The greatest difference in the percentage occurrences between afternoon and evening was found in thistle, which was found in 17% of afternoon samples and 11% of evening samples. All other taxa showed small differences in the percentage representation between afternoon and evening samples (Fig. 4a). Two taxa (lime, Tilia Â vulgaris, and sunflower, Helianthus sp.) were only found in one evening sample each, and two taxa (mallow, Malva sp., and scabious, Scabiosa sp.) were only found in one afternoon sample each.
The majority of samples, 62%, contained only one pollen taxon, while 25% contained two taxa, and the remaining 13% comprised three, four or five taxa. Of single-taxon samples, 91% were either blackberry (68%) or red clover (23%). Common toadflax, birdsfoot trefoil, umbellifer and the unidentified taxon comprised the remaining 8%. Of the afternoon samples, 61% were single taxon, compared to 64% of the evening samples. Samples with two taxa comprised 24% of afternoon samples and 27% of evening samples. For samples with three or more taxa, these percentages were 15% and 8%, respectively (Fig. 4b).
Blackberry and red clover were the only taxa comprising more than 10% of single-taxon samples, both in the afternoon (blackberry 70%, red clover 23%) and in the evening (blackberry 65%, red clover 25%). For samples that contained two or more taxa, blackberry and red clover were again the predominant species in both periods, with the highest percentage of majority taxa occurrences in mixed samples (blackberry: afternoon 60%, evening 66%; red clover: afternoon 31%, evening 21%). Birdsfoot trefoil was also represented well in these samples, appearing as a majority taxon in 17% of afternoon and 15% of evening mixed-taxa samples. Common toadflax was also a majority taxon in 17% of afternoon samples; however, it was only a majority taxon in 6% of mixedtaxa evening samples. Thistle and umbellifer were the most common minority taxa in mixed-taxa samples in both periods (thistle: afternoon 38%, evening 32%; umbellifer: afternoon 18%, evening 28%), but neither met the 10% threshold as major taxa. Grass was also present as a minor taxon in 13% of afternoon and 11% of evening mixed-taxa samples. Honeysuckle (Lonicera sp.) did not meet the 10% threshold in afternoon samples (7%) but was present in 13% of mixed-taxa evening samples. These taxa are labelled in Fig. 4a. No other taxa met the 10% threshold in any category.

Forager Size
The thorax size of measured foragers ranged from 4.03 to 6.89 mm, with a mean value of 4.92 ± 0.07 mm. Forager size did not vary significantly between bees that were sampled in the afternoon only, in the evening only, or in both periods (ANOVA: F 2 ¼ 0.008, P ¼ 0.992; Fig. 5).
Within evening samples, the body size of the individual did not relate to the timing of its return to the colony, when measured either by time from return until sunset (GLMM: t 54 ¼ 0.149, P ¼ 0.882) or by light level at return (t 53 ¼ 0.383, P ¼ 0.702).
The size of the forager did not significantly affect the mass of the pollen load carried (GLMM: t 98 ¼ e0.744, P ¼ 0.457).

DISCUSSION
This study is a novel exploration into the behavioural consequences when a diurnal pollinator decides to exploit flowers under naturally dimming light, at the interface between differing light niches. We directly compared foraging under daytime and sunset light conditions, and our results suggest that the visual environment of the late evening contributes towards changes in colony level activity and individual foraging behaviour in bumble bees, affecting pollen acquisition. Investigating the variation in behaviour of diurnal pollinators during crepuscular periods of the day broadens our understanding of behavioural plasticity in foraging insects in relation to the sensory environment. It will also help us to predict the potential effects of the exposure of both diurnal and   nocturnal pollinators to artificial light, not only at night but also when in the transition between night and day.
In the present study, we found that activity levels declined at the end of the day. Our observation that fewer bees exited the colony to initiate a foraging trip in the late evening compared to the daytime indicates that only a proportion of potential foragers are motivated to leave the colony during this time, with others choosing to stay in the colony during the approach to sunset. Similarly, the return rate was lower in the evening than in the afternoon, demonstrating that fewer bees were out foraging during this late evening period. To our knowledge, these are the first quantitative measurements of bumble bee colony activity in the late evening under naturally dimming light, confirming previous common anecdotal observations. We conclude that bumble bees are facultative evening foragers, and that the decision to forage in the late evening is probably made in relation to the lower profitability and higher predicted costs of foraging at this time.
The finding that individuals may choose to return with a partial pollen load has implications for our understanding of the flexibility of foraging behaviours in pollinators, in that typical choices on when to move on from a foraging patch, or indeed cease foraging for that bout, may be different for animals foraging in crepuscular periods as compared to the choices made by the same animal in similar resource patches during the daytime. The potential alteration to foraging tactics in response to large, regular changes in the sensory environment is an intriguing route for future study.
The use of the crepuscular period by foraging diurnal pollinators could lead to these species being exposed to light pollution due to temporal overlap with the period when artificial lights are turned on and become dominant as the natural light levels fall. The prevalence of artificial lighting is increasing on a global scale and is of growing concern to ethologists (reviewed by Gaston et al., 2014). Recent work has shown that artificial light exposure can negatively affect foraging invertebrates indirectly, via light-induced changes in their food plants (Bennie et al., 2015(Bennie et al., , 2018Grenis & Murphy, 2019), and via behavioural changes in predators and parasites (Bennie et al., 2018;Sanders et al., 2018). Additionally, artificial light can decrease pollinator visits to flowers and have knock-on effects across ecosystem networks (Knop et al., 2017;Macgregor et al., 2017). However, research has largely focused on nocturnal species, and this pollutant may be overlooked as a potential threat to foraging behaviours and pollination by their diurnal counterparts. We suggest bumble bees could be a good model for future studies into the reach of artificial light beyond the night-time.
However, it is still unclear how facultative foraging during the evening is regulated. The departures and late returns strongly suggest that individuals were motivated to forage in the late evening, which in turn means that there must have been enough resource available to support foraging trips. Additionally, it is evident that evening foragers did not require full pollen baskets to trigger their return to the nest, as many had smaller pollen loads than they had collected during the day or during brighter parts of the evening. Two main factors probably influence a bee's decision to stop pollen foraging and return to the nest with a partial pollen load. First, the decreasing light in the approach to sunset may put visual constraints on its ability to reach and exploit flowers, or to navigate home (for a review, see Warrant & Dacke, 2011). Second, resource availability may be lower in the evenings with resources being depleted during the day (Somanathan et al., 2020) or flowers closing. We also cannot exclude the possibility that some bees may have switched from pollen foraging to nectar foraging part way through their foraging bout, as we were unable to accurately weigh the bees in the field. We note that the circadian rhythms of the foragers, which are entrained by external light cues (e.g. Chittka et al., 2013;Merling et al., 2020;Spangler, 1973; and drive periodicity in activity Yerushalmi et al., 2006), will also be a factor in the cessation of daytime activity, although we did not manipulate or measure circadian rhythms here (for a review, see Bloch et al., 2017). Furthermore, other abiotic factors may covary with light, for example temperature; our data showed temperature declined before sunset (see Appendix Figs. A1, A2), although temperatures remained high enough for bumble bee thermal requirements during our study period (Couvillon et al., 2010;Heinrich, 1974;Lundberg, 1980;Peat & Goulson, 2005). Thermoregulatory costs were unlikely to play a major role in our study, whereas the changing light environment is likely to be an important influence on activity at this time. While it is not possible to tease apart all factors in our study, the decision by bumble bees to end foraging probably involves integration of information from the sensory environment encountered by the individual together with internal circadian cues, and some assessment of resource profitability.
Our finding that decreasing light levels in the evening correlated with decreasing size of pollen masses lends support to the hypothesis that diurnal pollinators are less able to access resources due to visual constraints in these conditions. The time until sunset did not affect pollen load mass, which indicates this result is related to the sensory environment and not to circadian rhythms, and we found no relationship between temperature and pollen mass (see Appendix). For any diurnal pollinator active during crepuscular periods, visual guidance involved in the location of resources, flower handling and navigation between resources could be detrimentally impacted by decreasing light, leading to reduced foraging ability and pollination rates. Individual bumble bees could also be assessing the risk of being unable to navigate back to the colony in time before it is too dark to fly, which may motivate them to return before a full load is collected. However, the lowest recorded light level at the return of a bee was around 230 lx, and the lowest light level at an exit was around 290 lx, which are both much higher than their physiological flight limit of 3.4 lx (Reber et al., 2015) or than levels at the initiation of low-light foraging observed at dawn (Hall et al., 2021). Flying and foraging under dim light are clearly functionally distinctive behaviours for which the context of the light condition is important. Furthermore, this highlights the potentially large discrepancy between testing the behaviour of animals under dim light conditions in the laboratory versus under natural field conditions, and future studies must be careful when drawing inferences from one to the other. Regardless, our results provide evidence that declining light is likely to be a key factor influencing the acquisition of resources and the decision to return to the colony in the late evening in a diurnal pollinator.
Bumble bee behaviour can also be influenced by alloethism, the allocation of tasks by body size rather than age, the latter of which is typical for honey bees. We therefore investigated the size distribution of our sampled foragers, which is similar to that previously published for this species (Goulson et al., 2002). Bee size has previously been related to the visual capacity of the individual; therefore, we expected that smaller bees would be less likely to stay out than larger bees in the approach to sunset. However, we found no such relationship among the pollen foragers that returned in the evening. Instead, individuals of the same range of sizes foraged in the evening period as during the day, and smaller bees did not return in higher light levels than larger bees. This suggests that the decision to forage in dimmer light is not exclusively driven by the visual sensitivity of the bee, and indeed is likely to involve assessment of multiple factors, of which light is only one. Recent work by Hall et al. (2021) found that more experienced bees emerge earlier after dawn, some of which were smaller. When comparing with our results, this indicates that both during the early hours of the day and in the late evening, multiple factors, or different weightings of those factors, are being integrated into the decision to forage by smaller bees. Additionally, although there exists some evidence relating bee size to resource-carrying capacity (Allen et al., 1978;Goulson et al., 2002), we did not find a relationship between bee size and pollen load mass, as both smaller and larger bees brought back variable load sizes, both in the afternoon and in the evening. We therefore conclude that the smaller pollen loads found in the evening were not due to the body size of active foragers, in relation to either visual ability or carrying capacity.
Decision making regarding nutrition and the exploitation of flowers may also vary for diurnal pollinators choosing to forage at the end of the day. The last foraging trips of the evening might well be focused on visiting and exploiting well-known sites and profitable plants. Indeed, evening foragers collected pollen from the same plant taxa as those foraging in the afternoon and were slightly more flower-constant in the evenings. Since bumble bees are known to have high patch fidelity (Ogilvie & Thomson, 2016), these results indicate that foragers going out in the evening could have been accessing the same patches of resources that they had already learned the location of during the day. As there is some evidence for lower patch fidelity in different landscapes (Martínez-Bauer et al., 2021), this finding may vary in more resource-diverse habitats than found at our field site. Evening foragers exploiting the same flowers as those active during the day may have access to a depleted amount of resource, due to diminishment by daytime foragers (Somanathan et al., 2020). This potential reduction in resource availability may explain why, in our study, evening bees returned with less pollen; however, as individuals continued to return with partial loads, a high enough level of pollen must have been available to motivate bees to forage at this time. We theorize that at the colony level, foragers switch from a mix of exploration and exploitation flights during the day (Woodgate et al., 2016) to solely exploitation flights in the evening, when resource availability may be lower and risks, such as being unable to navigate back to the colony before night, are higher. The preferred plant taxa were commonplace at the site in large or connected patches (e.g. red clover fields and blackberry in hedgerows), which facilitates navigation between flowers (Cranmer et al., 2012), minimizing the need for exploration. This decision to exploit known resources appears to come at an acquisition cost to the individual, which may be further compounded by the impacts of dim light. However, the alternative low-risk strategy is not to forage at this time; indeed, many individuals chose not to forage in the late evening, as discussed above.
We measured pollen collection in July and August, when in this region, floral resource availability for bumble bees in agricultural landscapes is high (Timberlake et al., 2019(Timberlake et al., , 2021. This is also a critical time for the reproductive success of bumble bee colonies (Goulson, 2003). Our findings indicate that while a few taxa will dominate bumble bee pollen loads, foragers will access a variety of other floral resources in small quantities. In the present study, foragers showed a strong preference for blackberry pollen, which was abundant in the hedgerows throughout the field site. Other taxa commonly utilized included umbellifers and thistles, which were located at the edges of crop fields. Flower resources in hedgerows (Garratt et al., 2017;Hanley & Wilkins, 2015;Hannon & Sisk, 2009;Morandin & Kremen, 2013) and field margins (Lagerl€ of et al., 1992) are known to be a vital source of nutrition for bumble bees in agricultural landscapes, where resources are otherwise sparse. Red clover was also highly favoured by these foragers; clover species are widely known for frequent use by bumble bees and honey bees. Together with birdsfoot trefoil and common toadflax, and various grasses, it forms an effective component for enhancing wild flower resources for pollinators and therefore is frequently found in flower seed mixes for agricultural enhancement schemes, which helps to counteract long-term trends in the reduction of nutritionally important food plants in agricultural landscapes (Carvell, Roy, et al., 2006;Carvell, Westrich, et al., 2006;Kleijn & Raemakers, 2008;Pywell et al., 2011).
Bumble bees gain a nutritional benefit from accessing a broad range of pollen sources. Social bee foragers are known to form individual pollen preferences that can change in different contexts, based on the assessment of species-specific preingestive taste cues, but importantly also reward profitability, which includes how available and easily detectable flowers are, or the time required to collect the pollen on the flower and to improve their handling skills (reviewed in Nicholls & Hempel de Ibarra, 2017). The common stores of the colony will harbour diverse nutrients brought in by the large numbers of foragers, which in itself is sufficient to achieve dietary resilience and secure a balanced diet for the healthy development of the colony (Corby-Harris et al., 2018;Hendriksma & Shafir, 2016;Kriesell et al., 2017;Somme et al., 2015;Vaudo et al., 2016; for a review see Wright et al., 2018). Increasing the diversity of plants in agricultural landscapes has a range of beneficial effects not only for the colonies of social bees and the offspring of other pollinators, but also on crop yield as foragers will access a variety of floral resources in varying quantities in such landscapes (Garibaldi et al., 2014;Vaudo et al., 2015).
In our study, we give novel insight into an as yet little studied period of the day, when declining light levels affect foraging by diurnal pollinators in the approach to sunset, and show that foraging differs from the daytime norm during this period. We conclude that bumble bees are facultative foragers in the late evening, but that those individuals that are motivated to forage at this time incur a resource acquisition cost compared to the daytime. Declining light seems to influence behavioural decisions under these conditions, and therefore should be considered more closely, alongside other cofactors that could change at sunset, when investigating the sensory ecology of pollen-foraging bees and other diurnal foragers that are active in the crepuscular periods.

Author Contributions
Katherine E. Chapman: conceptualization, methodology, project administration, investigation, formal analysis, visualization, data curation, writing e original draft, writing e review and editing. Nathalie E. Cozma: investigation, writing e review and editing. Arran B. J. Hodgkinson: software, writing e review and editing. Roger English: conceptualization, writing e review and editing. Kevin J. Gaston: conceptualization, supervision, writing e review and editing. Natalie Hempel de Ibarra: conceptualization, methodology, formal analysis, visualization, supervision, writing e original draft, writing e review and editing.

Data Availability
Data will be made available on request.

Declaration of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.  Figure A2. Temperature measurements ( C) recorded in the 60 min before sunset (N ¼ 9 days). Each box plot shows the median, interquartile range (IQR), maximum and minimum nonoutlier values (whiskers) and outlier values (outside 1.5 Â IQR, points). Temperature was recorded every 30 s and rounded to the nearest integer, and data were pooled over 5 min.