Foraging distance distributions reveal how honeybee waggle dance recruitment varies with landscape

Abstract

Honeybee (Apis mellifera) colonies use a unique collective foraging system, the waggle dance, to communicate and process the location of resources. Here, we present a means to quantify the effect of recruitment on colony forager allocation across the landscape by simply observing the waggle dance on the dancefloor. We show first, through a theoretical model, that recruitment leaves a characteristic imprint on the distance distribution of foraging sites that a colony visits, which varies according to the proportion of trips driven by individual search. Next, we fit this model to the real-world empirical distance distribution of forage sites visited by 20 honeybee colonies in urban and rural landscapes across South East England, obtained via dance decoding. We show that there is considerable variation in the use of dancing information in colony foraging, particularly in agri-rural landscapes. In our dataset, reliance on dancing increases as arable land gives way to built-up areas, suggesting that dancing may have the greatest impact on colony foraging in the complex and heterogeneous landscapes of forage-rich urban areas. Our model provides a tool to assess the relevance of this extraordinary behaviour across modern anthropogenic landscape types.

Fig. 1. The honeybee waggle dance carries information about the location of a resource.

a The honeybee waggle dance communicates the direction of the resource, relative to the direction of the sun through the angle of the dance relative to the vertical. The duration of the waggle run indicates the distance to the resource. b This information, in the form of a bearing and a distance, allows other foragers to locate the resource in the landscape (blue circle).

Fig. 2. Honeybee foraging.

a A hypothetical illustration of a typical foraging honeybee, foraging with scouting only. Foragers leave the hive on a search path with straight segments (white lines) and continue until they encounter a resource (coloured circles). b An illustration of honeybees foraging with recruitment. Foragers that have identified resources in scouting trips (white lines) convey this information on the dance floor (brown disc), where foragers can sample dances reporting on both scouting and recruiting trips and follow these directions (yellow lines). c Complementary cumulative frequencies of foraging distances reported from scouting and recruit trips from a simulation model (see Materials and Methods). Note the difference in the shape of the distributions. The scout distribution is best fit by an exponential (grey straight line), the recruit distribution by a Rayleigh distribution (black curve).

Fig. 3. The rationale underlying the derivation of foraging distributions.

a The distance distribution of resources encountered by scouts is approximated by an exponential distribution. b These distances are communicated on the dance floor. c Dances for resources that are nearer or higher in quality are repeated more often. The most profitable resources (red circles) are danced for more often for a given distance than resources of lesser quality (yellow and green circles). d Scout trip distances are translated into dance duration. e Recruits select a resource location from the dance floor (brown disc) and visit the resource (yellow disc). f The distance distribution of resources encountered on recruit trips is translated into dance duration. g By taking together the dance duration distributions for scouts and recruits the distributions of all dances on the dancefloor can be found, leaving a distinctive imprint in the form of a shoulder, or hump (h), which also leaves a similar imprint in the cumulative distance distribution of resources visited (i). The dots in the panels a, d, f, g, h and i are from simulations and are for illustrative purposes only.

Fig. 4. The honeybee foraging model fitted to data from 20 hives.

a Location of study hives in South East England, shaded area in the main plot indicates Greater London. For 16 hives the collective foraging model provided best explanation (circles), for 4 hives the individual search model provided the best explanation (triangles) as indicated by lowest AIC score, and for one hive (square) neither model gave a good fit. Locations of hives that provided data for subsequent panels indicated by c and d. b Distribution of goodness of fit confidence values for each model fit to waggle run durations from each site. The P-value is derived from a bootstrapped two-sided KS test comparing the fitted model predictions to the empirical data, the red dashed line marks the significance threshold of 0.05. The number of data points used for each site is given in Supplementary Table 1. c Representative example of the dance duration distribution for a hive in which the individual model (blue line) provided a better fit than the collective foraging model (red line). d Dance duration distribution for a hive in which the collective foraging model (red line) provided a better fit than the individual model (blue line). The typical shoulder or “hump” in the distribution is indicative of contribution of recruitment dances. Panels show the complimentary cumulative frequencies with binned frequency distributions as inset.

Fig. 5. Collective foraging correlates with land-use.

a Within both the agri-rural and urban landscapes, we found considerable variation in waggle dance use across sites. b We thus sought to explore whether this could be explained by differences in land-use across sites using PLS analysis. For the agri-rural landscape beta regression shows the relationship (black line) between first principal component and waggle dance use, with 95% confidence interval shown by the grey shaded area. c Correlation scores for the PLS. Boxplots illustrate the range of correlations (median, upper and lower quartiles) between the first principal component and each land-use type following jack-knife resampling. Correlations outside the shaded area significantly contribute to the first principal component. Going from left to right in panel b, the landscape will contain fewer landscape types with a low correlation score and more with a higher correlation score. NAUMGS and NAMGS stands for non-agricultural unmanaged green space and non-agricultural managed green space, respectively.