Row crop fields provide mid‐summer forage for honey bees

Abstract Honey bees provide invaluable economic and ecological services while simultaneously facing stressors that may compromise their health. For example, agricultural landscapes, such as a row crop system, are necessary for our food production, but they may cause poor nutrition in bees from a lack of available nectar and pollen. Here, we investigated the foraging dynamics of honey bees in a row crop environment. We decoded, mapped, and analyzed 3459 waggle dances, which communicate the location of where bees collected food, for two full foraging seasons (April–October, 2018–2019). We found that bees recruited nestmates mostly locally (<2 km) throughout the season. The shortest communicated median distances (0.474 and 0.310 km), indicating abundant food availability, occurred in July in both years, which was when our row crops were in full bloom. We determined, by plotting and analyzing the communicated locations, that almost half of the mid‐summer recruitment was to row crops, with 37% (2018) and 50% (2019) of honey bee dances indicating these fields. Peanut was the most attractive in July, followed by corn and cotton but not soybean. Overall, row crop fields are indicated by a surprisingly large proportion of recruitment dances, suggesting that similar agricultural landscapes may also provide mid‐summer foraging opportunities for honey bees.

to the landscape increased seed number in high yield conditions (Blettler et al., 2018). Cage studies, which exclude flower-visiting insects from blooms, showed that honey bees improve yield in some cotton cultivars, including total boll and seed mass (Rhodes, 2002).
Despite these yield improvements, and perhaps because of row crop nondependency on pollinators, pollinators in row crop environments remain understudied. For example, it is unclear if these habitats can provide nutrition for pollinators: while previous studies have focused on the impact of insect pollinators to row crops, fewer studies have investigated the reciprocal relationship of the row crop habitat on the insect pollinators.
The honey bee is a valuable, managed insect pollinator, contributing c. $14-20 billion annually to the US economy (Morse & Calderone, 2000). The number of honey bee hives is decreasing in the United States, Britain, and many western European countries (Aizen & Harder, 2009a, 2009b: in the United States alone, there has been a 50% reduction in the number of hives in the last century (Aizen & Harder, 2009a, 2009bEllis et al., 2010;Levy, 2011;Neumann & Carreck, 2010). Honey bees use a recruitment behavior called the waggle dance to communicate the location of a good source of food, which is usually nectar or pollen (Couvillon, 2012;von Frisch, 1967). Nestmate recruits then use the waggle dance to determine the location of a profitable food source, which they can then forage on themselves. The communication therefore allows the hive to exploit the resource in an efficient and timely manner. The dance is also visible to the eye and can be observed and decoded by scientists to determine how, where, and when a forager, the dancer, is collecting food within a landscape (Beekman & Ratnieks, 2000;Carr-Markell et al., 2020;Couvillon, Riddell Pearce, et al., 2014;Couvillon et al., 2014aCouvillon et al., , 2014bSponsler et al., 2017;Waddington et al., 1994).
Here, we monitored honey bee foraging in southeastern Virginian agricultural land, which includes numerous corn, cotton, soybean, and peanut fields. We allowed freely flying honey bee foragers to collect nectar and pollen in a row crop environment for two complete foraging years and analyzed the honey bee waggle dances.
Our objectives were: (1) to determine the communicated foraging distance, as this serves as a proxy for food availability, across the seasons within a row crop environment and (2) to calculate the percent recruitment by foragers to peanuts, soybeans, corn, and cotton. Our unique study design allowed us to observe not only where honey bees highly value forage (i.e., by decoding the waggle dance), but also to calculate how much of that recruitment occurs in fields of interest.

| Study organism and experimental set-up
We studied three predominantly Apis mellifera ligustica honey bee colonies (labeled A-C) that included a queen and approximately 5000 workers plus brood housed in plexiglass-walled observation hives containing three American Standard Deep frames. The observation hives were placed inside a shed (approx. 14 ft × 8 ft), and the colonies were connected to the outside via a 3 cm × 30 cm plastic tube, which afforded the foragers unimpeded access to the landscape. We wired the shed for electricity to provide diffuse lighting and, when needed, air conditioning or heating. Because observation hives are small and the bees cannot maintain large honey stores (i.e., sufficient for a week of bad weather), we provided when needed a sugar solution for supplemental feeding using a canning jar with small holes in the lid inverted over a mesh opening in the observation hive.

| Study location, crops, and bloom times
We located the colonies at the Virginia Tech Tidewater Agricultural Research and Extension Center (TAREC) (36°41′06.4″N 76°46′01.6″W) in Suffolk, Virginia. The 465-acre facility researches southeastern Virginia row crops (e.g., corn, cotton, peanuts, soybeans, and small grains) and swine. The TAREC comprised the immediate area around the honey bee hives, but within a wider foraging range (approx. 8.3 km radius) there was a mixed-use landscape of forests (33.5%), cropland (30.4%), wetlands (18.75%), developed land (4.9%), grassland/pasture (3.55%), and other land types (8.9%) (USDA-NASS 2018USDA-NASS 2019, 2019. Because there were unrealistic fluctuations from year to year in the CropScape land type characterizations, and because some of the land types in our crops of interest were demonstrably false (i.e., we had first-hand knowledge that a particular field in TAREC was, for example, one crop whereas it was mislabeled as another in CropScape), we manually mapped and digitized the crops of interest within the region where most of the foraging occurred. Our crops of interest were peanuts, soybeans, corn, and cotton, which accounted for 34-36% of the landscape around the hives within a 2 km radius in 2018 and 2019 (Table 1; Figure 6).
Row crops vary in their bloom times ( Figure 1). Small grains and cover crops (i.e., crops planted for a variety of beneficial effects that are not harvested and sold as the main cash crop in an operation) bloom from early April to mid-May. Major row crops begin to bloom around mid-June to early July. For this study, we were interested TA B L E 1 Percent of crops of interest within an area that represents the majority of dances (i.e., within a 2.0 km radius form the hives). Land-type percentages are from the (manually corrected) 2018 and 2019 National Cropland Data layers and therefore are subject to slight yearly variations in examining honey bee foraging on peanuts, soybeans, corn, and cotton because they all bloom around the same time, beginning between mid-June to early July with the bloom season lasting until mid-to late-August. Additionally, it is often anecdotally assumed that bees prefer soybeans to cotton, but, to our knowledge, it has never formally been investigated. In our calculations of percent foraging of specific fields of interest (see below), we restricted the date ranges to the time intervals when c. 75% of the crop is in bloom, which we consider full bloom ( Figure 1).

| Data collection-waggle dance filming
Foragers may perform waggle dances to recruit nestmates to high quality resources, generally 100 m to many kilometers from the hive (von Frisch, 1967). Importantly, foragers only dance for the best resources at any given time (Couvillon, 2012;Seeley, 1994), and this allows us to determine the location of where the bees are most preferring to forage, not the location of all the available forage including low quality forage. When we decode waggle dances, we do so because we are interested in where the dancer herself has foraged: by definition, every dancer is (1) a recently returned forager who (2) has found a highly profitable resource, usually nectar or pollen. We,

| Data collection-waggle dance decoding
Dance decoding was based on an updated version of . Briefly, we converted video files filmed at 30 frames per second to AVI using Ubuntu (v. 2004Ubuntu (v. .2021.0) and imported them into ImageJ (version 1.52i). At the start of each video, we measured our vertical reference plumb line to determine the angle offset (i.e., if the cameras and/or observation hive are not completely vertical, then measuring the angle offset corrects for that, usually only by a few degrees or fractions of a degree). Then we played the video from the beginning until we found the first dancer, which usually occurred within the first few seconds of the video. A returning forager performing a waggle dance will repetitively oscillate her abdomen while moving linearly (waggle run phase) across the comb, will stop, F I G U R E 1 Approximate bloom times of row crop cultivars in Suffolk, Virginia. Bloom intervals are divided based on the percentage of bloom (i.e., >75% bloom refers to greater than 75% of fields of interest in bloom). We investigated percent honey bee foraging in fields of interest when >75% of the crop of a field was actively flowering turn either to the left or right, and then walk back (return phase) to the starting position. She may then do another waggle run, repeating this pattern of waggle run and return phase 1-100+ times (Seeley et al., 2000;von Frisch, 1967).
Dance decoding involves extracting two pieces of information per waggle run to obtain a vector: the duration of the run and the direction of the run relative to vertical (Couvillon, Riddell Pearce, et al., 2012;von Frisch, 1967). We assessed the duration by noting the start and end frame of each waggle run and calculating the difference, which we then convert from number of frames to seconds of duration by dividing by the 30 fps. We assessed the direction of each waggle run by using the straight-line tool in ImageJ to measure the angle from vertical that the dancer makes with her body as she performs a waggle run (Couvillon, Riddell Pearce, et al., 2012;von Frisch, 1967).
This measurement is automatically corrected for the angle offset (see above) in Excel. Lastly, since honey bees orient relative to the sun but dance relative to vertical, we noted the time of day and date during filming and corrected later for the solar azimuth to obtain the bearing.
For each dance, we decoded four mid-dance, consecutive waggle runs per dancer, with an equal number of left and right-handed turns (Seeley et al., 2000;von Frisch, 1967). We averaged the four waggle run durations and azimuth-corrected angles to obtain a single duration and direction for each dance. This number and selection of waggle runs was chosen because the calculated mean direction and mean duration was not significantly different from the means obtained if all the waggle runs per dance are decoded (Couvillon, Riddell Pearce, et al., 2012). We converted duration to distance using the universal calibration reported by Schürch et al. (2019).
Taken together, these averaged components (direction +distance) give the approximate location of the advertised resource.
Once the first dance within that video frame had been decoded, we then decoded the other dances that were occurring simultaneously on the dance floor. In videos with few dancers present, we watched the entire video and decoded every dance. However, in videos with higher activity (20+ per hour), we skipped ahead 6 min (or 10,789 frames) after decoding all the simultaneous dances that occurred at the start of the video. By skipping ahead, we avoided potential pseudo-replication, where we sample from the same dancer twice [i.e., a single dancer can dance 1-100 times (Seeley et al., 2000), depending on the resource quality, which can last many minutes and can involve the dancer moving in and out of the screen].
Our objective was to achieve at least 20 dances per hour of video.
These dances should provide a good representation of colony foraging during the respective hour of filming (Couvillon, Riddell Pearce, et al., 2012;Sponsler et al., 2017). In all, we decoded 3459 dances, 2066 from 2018 and 1393 from 2019. We used R (R Core Team, 2020) to perform all the analyses.

| Data management and validation
For waggle dance decoding, we recorded all raw video data on SD cards, which were subsequently uploaded to a GTD as a permanent, online, cloud-based repository. We immediately entered decoded dance data into Excel, and each decoder would upload his or her Excel file daily to the GTD. To avoid version confusion, the latest raw data as entered the Excel files were obtained from the GTD through R code using the googledrive package. Lastly, we created a GitHub repository to manage R files.
We validated dance data before analysis. Data were scanned and rows with missing values, due to human error, were removed. Next, we examined our data for erroneous values by calculating, for each dance, the standard deviation (SD) per dance for each dance component (i.e., the SD between the four runs for waggle run duration and for waggle run angle). We then identified the highest ten values (i.e., the dances with the most intra-dance variation) within the data set for each respective dance component (i.e., duration, angle).
We located these 20 dances and examined them on the data entry sheet to determine if the SD was due to an obvious entry error. If the error source was unclear, we located the dances on the original videos to verify whether high SD was based on error inherent in the dance (i.e., a "noisy" dancer with high intra-dance variation), which is normal (Couvillon, Phillipps, et al., 2012;Couvillon, Riddell Pearce, et al., 2012;Schürch et al., 2019) and acceptable, or due to human error (i.e., dance decoding error), in which case we re-decoded the dance. We noted any changes to initial values in our dataset. The process of examining the 10 dances with the highest SD values in either duration or angle was repeated until all 10 dances' SDs were solely from error inherent in the dance (i.e., were just noisy dances). Overall, this required us to validate six rounds of highest SD value dances for both duration and angle and involved our correcting (re-decoding) 30 dances out of a dataset of 3459 dances.

| Data analysis-distance
Honey bees have evolved to be economic in their foraging decisions and, considering the energetic impact of a flight, a bee will only fly and then recruit her nestmates to a resource as necessary, especially given the distance from the hive (Couvillon et al., 2014b;Schmid-Hempel et al., 1985;Seeley, 1986). Therefore, communicated foraging distance can serve as a proxy for the availability of nectar and pollen (i.e., when bees recruit for a resource relatively further away, it is likely because food cannot be found closer to the hive).
We interpreted the communicated foraging distance as a function of time to determine seasonal fluctuations in forage availability. We began by plotting the raw data as distance across time, which included a LOESS regression. We then aggregated time by month, which allowed us to compare median foraging distance per month to previous studies (Couvillon et al., 2014b). Lastly, because our study was in an environment with semi-predictable bloom times highly relevant to our study, we aggregated median foraging distance across the growing season (e.g., pre, during, and post bloom) by crop. We expected to see communicated foraging distance decrease as we go from pre-to during bloom and increase as we go from during bloom to postbloom if the crop is impacting food availability.

| Data analysis-percent recruitment by foragers and mapping in fields of interest
We were primarily interested in honey bees' use of soybean, cotton, corn, and peanut fields ( Figure 1, Table 1) because of our regional land cover, the timing of the crops, and the gaps in the previous research. Honey bees are imprecise and inaccurate in their dance communications and there is an inherent error in both components of the dance (Okada et al., 2014;von Frisch, 1967

| Honey bees in row crops recruit mostly locally with some long-range events
The median distance foragers recruited to in the landscape during

| Foraging distance varies by month and year
Foraging distance, as communicated by waggle dancers, varied by years, which suggests that forage is most available at this time. July is the month when all target row crops are blooming (Figure 1).

| Communicated median foraging distance varies during crops' full bloom
The median foraging distance travelled during pre, during, and post bloom followed our prediction for corn ( (2019), especially if one considers the representation of these crops (%) in the landscape (Table 2, Figure 5).
This increase was most evident in peanut and in corn in 2019. For example, peanuts accounted for approx. 6% of the landscape within their immediate foraging range for both years, and yet the bees foraged upon and recruited to peanuts 7% and 19% in 2018 and 2019, respectively, during bloom. This means that bees used peanuts during bloom time 3-5× more than one would expect given its representation in the landscape in 2019. The bees' indication through the waggle dance of the peanut fields also followed the predicted downup-down pattern that one would expect if the foraging were driven by that crop's bloom (Figures 5 and 6).
In contrast, bees decreased their recruitment to soybeans from pre-bloom to bloom in both years and to cotton in 2018 (  Figure 5 and 6). The bees' indication through the waggle dance of the soybean fields was opposite of peanut and instead followed an up-down-up pattern for both years.
Corn also was an attractive foraging option for honey bees during the mid-summer. Corn accounted for 8% and 11% of the landscape, and approximately 9% and 16% of recruitment indicated the fields during bloom (Table 2, Figures 5 and 6).  (Table 2, Figures 5 and 6). In 2019, while the fields accounted for 11% of the landscape, the bees indicated the fields the most during the bloom (14%), which was an increase from pre bloom (11%).
As these row crops bloom over roughly the same period, taken together, the row crops account for a large proportion of foraging during full bloom (c. 37-50% in total, Table 2, % foraging during bloom), representing a large effort by honey bee hives in exploiting row crops for forage.

| Honey bees recruit in July especially to peanuts and also to corn and cotton across entire study
After we correct for distance, honey bee dances indicated the highly

| DISCUSS ION
Here, we have shown that honey bees in an agricultural landscape forage extensively in nearby row crops, especially during the midsummer full bloom period. Recruitment was generally local, <2 km away from the hive throughout the foraging year (April-October) ( Figure 2). We found some variation per month, with communicated foraging distance increasing (i.e., forage availability decreasing) in April, May, June, and August ( Figure 3). The month where distance was shortest, indicating that food was most abundant, was observed for both years in July, the only month when all target crops (peanuts, soybeans, corn, cotton) were in full bloom ( Figure 1) we demonstrated that peanuts, followed by corn and cotton (but not soybean) is most indicated in July compared to other fields (Figure 7).
Taken together, all these data suggest that row crops provide abundant forage for honey bees during the mid-summer, a period that has previously presented as a time of forage dearth in other landscapes (Couvillon et al., 2014b;. Other studies have investigated honey bee foraging dynamics in row crops, with the majority focusing on the pollination services bees provide to crops (Blettler et al., 2018;Erickson, 1975;Girardeau & Leuck, 1967;McGregor et al., 1955;Pires et al., 2014). Row crops are wind or self-pollinated and do not depend upon insects for successful pollination; however, they have flowers that produce nectar and pollen and, in the case of cotton, extrafloral nectaries that attract pollinators and provide forage (Martin, 1980;Röse et al., 2006;Willmer, 2011). Our study demonstrates that, when placed in a row crop environment, honey bees will forage and recruit to the row crops, primarily during the periods of full and, sometimes, postbloom ( Figure 6, Table 2). Percent recruitment during full bloom was especially high in 2019 to peanut (19%), corn (16%), and cotton (14%).
Because we possessed first-hand knowledge of field locations, we also anecdotally could see how crop rotation might affect bee recruitment choices: in 2019, some peanut fields were located closer to the hive, which could explain the large increase in the use of those fields. Of course, though, our distance correction still demonstrated peanuts as the most preferred crop (Figure 7).
During bloom in July, while peanut, corn, and cotton accounted for c. 28% of the landscape within the foraging range, the honey bees indicated these fields c. 49%. Therefore, dancing bees are recruiting to the row crops nearly twice as much as one would predict based on how prevalent the fields are in the landscape and their distance from the hive. Additionally, communicated foraging distance serves as a proxy for forage availability, where high availability of forage results in a lower communicated foraging distance and lower availability of forage results in higher communicated foraging distance (Couvillon et al., 2014b). The communicated foraging distance for peanut and corn in 2019 is low (Figure 4), supporting that the foraging bees are finding abundant resources close to the hive. In contrast, soybean was not appreciably indicated by our recruiting bees, with the bees use of these fields decreasing in the bloom compared to the pre-bloom, which is also reflected in the noneffect of bloom on the foraging distance (heavily overlapping CIs, Figure 4).  TA B L E 2 Mean percent foraging (95% CI) in fields of interest by bloom in 2018 and 2019, as determined by waggle dance decoding, mapping, and analysis. During bloom is the period when >75% of fields are in bloom. Pre-and post-bloom periods comprise percent foraging in fields of interest before and after the period of full bloom. Percent land cover displays the proportional representation of that land type within the landscape reached physiological maturity, although the percentage of blooming plants is usually <75% (Quisenberry & Roark, 1976;Ritchie et al., 2007). Cotton was relatively more attractive again in September (Figure 7), which could suggest the bees return to the remaining cotton blooms when the other crops have ceased blooming but before autumn forage, like goldenrod, becomes available (Richardson et al., 2015). One limitation of using the dance language alone is that we cannot immediately test between these alternative explanations.
Our multinomial logistic regression analysis, which corrects for distance, demonstrates that honey bees possess a strong preference for peanut, then corn and cotton, fields. In other words, dancing honey bees are 2-3 times more likely to recruit to peanut, corn, or cotton fields in July compared to other fields and compared to that preference in April (Figure 7). In contrast, peak soybean recruitment happened in August, which does still fall within their bloom time ( Figure 1). Soybeans do continue to bloom into mid-August ( Figure 1). Therefore, like the September cotton peak, this small increase might also represent the bees demonstrating a renewed interest in soybean once the more attractive peanut and corn are past their peak blooms.

(c) (d)
Carr -Markell et al., 2020;Garbuzov et al., 2015;Sponsler et al., 2017), demonstrating its wide suitability for determining how honey bees use particular landscapes, fields, or flora for food. Overall, how honey bees use a landscape is a complex interaction of colony requirements, season, distance to field/crop of interest, and availability of alternative forage. In the springtime, when abundant alternative forage will be present, honey bees visit oilseed rape fields 0-26%, depending on whether the hives were in an urban or rural location respectively (Garbuzov et al., 2015). In our study location, at nearly 50% of all foraging in mid-summer was in row crops, especially peanut, corn, and cotton. Cotton, soybean, and peanut produce both nectar and pollen, and cotton secretes nectar through extrafloral nectaries; however, these are not usually viewed as high quality food sources (Martin, 1980). Indeed, our data suggests that the use of cotton and soybean is dependent on a relative lack of other options in the landscape.
Our bees largely foraged locally (median communicated foraging distance = 0.702 km), with some long-range recruitment indicating as far as ~8.3 km. While honey bees can fly several kilometers from the hive for resources, they will not typically do so unless resources near the hive are extremely scarce (Couvillon, Riddell Pearce, et al., 2014;Seeley, 1994) or if a highly rewarding F I G U R E 6 Honey bee dancers indicate the row crops during bloom compared to pre-and post-bloom. Transparency denotes level of recruitment, where increased color indicates higher recruitment from foragers. In particular, the large black patches that are not row crops are visible in pre-and post-bloom. During bloom, the transparency moves inwards to focus on the colorful row crop fields F I G U R E 7 Honey bees indicate peanut (yellow) the most in July compared to other fields and compared to April. Corn (orange) and cotton (blue), but not soybean (green), are also highly attractive. Odds ratio of crops displaying relative attractiveness by month with distance correction. Soybean did demonstrate peak recruitment in August, and cotton showed a second peak in attractiveness in September. Year included as a random factor Month Odds ratio (relative to the preference of focal crop vs. other in April) alternative suddenly becomes available (Beekman & Ratnieks, 2000). A study that investigated distance over time, as indicated by the waggle dance, reported that the maximum foraging distance for bees in Southern England was ~6 km (Couvillon et al., 2014a(Couvillon et al., , 2014b et al., 2013). Perhaps because of the challenging conditions, we had two hives die in early July 2019, and although they were quickly replaced with additional observation hives, there was a small gap in our data that overlapped with the bloom of corn (Figure 1), which already possesses a relatively short full bloom period. These colony deaths and the few days it took to replace them makes the bees' use of corn even more impressive: corn accounted for 8% and 11% of the landscape, and approximately 9% and 16% of recruitment indicated the fields during bloom (Table 2, Figure 5). After a distance correction, we see that honey bees have a definite preference for corn relative to other fields during July (Figure 7).

May
About 35% of major food crops are insect pollinated (Klein et al., 2007). Largely because of this reliance in pollinator-dependent crops, flower visiting insects are estimated to provide c. $200-250 billion worldwide in critical benefits to agricultural ecosystems (Gallai et al., 2009). Currently the demand for food crop pollination is outpacing the availability of pollinators (Aizen & Harder, 2009b;Garibaldi et al., 2011), as approximately 41% of insect species are declining globally (Hallmann et al., 2017;Kosior et al., 2007;Sánchez-Bayo & Wyckhuys, 2019). Of these insect species, major classes of pollinators, such as Hymenoptera and Lepidoptera, are some of the most affected (Berenbaum et al., 2007).
Overall, our honey bees foraged in rows crops of peanut, corn, and cotton much more than initially predicted, especially given that the select row crops are not pollinator-dependent, are generally poor nectar sources, were not overly represented in the landscape (Table 1), and can present a cumbersome challenge in pollen transport in the case of cotton (Jones & McCurry, 2012;Vaissière & Vinson, 1994;Vansell, 1944). Although our study is limited to one study site, and its location on a mixed crop research farm concentrated multiple crops within the immediate vicinity, it nevertheless presents the tracking of honey bee foraging through waggle dance decoding that spanned two full foraging years. As such, it represents an advancement in our understanding of where honey bees forage in a row crop environment and suggests that, in a similar agricultural landscape, hives might also be sustained by such crops during the summer. However, several unknowns remain: just because honey bees can feed themselves on row crops does not mean that it is the ideal option for their well-being. Future studies could determine how honey bees might exploit row crops in landscapes with greater or lesser heterogeneity, especially given that periods of forage dearth have been observed in less heterogeneous landscapes (Dolezal et al., 2019). It would also be interesting to investigate whether a diet of row crop forage might also represent an over-exposure to agricultural chemicals, which could have important implications for pest management. Couvillon: Conceptualization (lead); data curation (supporting); formal analysis (equal); funding acquisition (lead); investigation (lead); methodology (lead); project administration (lead); resources (supporting); software (supporting); supervision (lead); validation (supporting); visualization (supporting); writing -original draft (equal);

AUTH O R CO NTR I B UTI
writing -review and editing (lead).

DATA AVA I L A B I L I T Y S TAT E M E N T
The data corresponding to this manuscript has been curated as a static dataset for use by other researchers and the public and is stored at the Virginia Tech Data Repository (https://doi. org/10.7294/19755016).