Content uploaded by Madeleine Beekman
Author content
All content in this area was uploaded by Madeleine Beekman on Jun 21, 2018
Content may be subject to copyright.
How does an informed minority of scouts guide
a honeybee swarm as it flies to its new home?
MADELEINE BEEKMAN*,ROBERTL.FATHKE†& THOMAS D. SEELEY†
*School of Biological Sciences, University of Sydney
yDepartment of Neurobiology and Behavior, Cornell University
(Received 2 December 2004; initial acceptance 20 January 2005;
final acceptance 4 April 2005; published online 1 December 2005; MS. number: 8369)
When a honeybee swarm lifts off to fly to a new nest site, only the scouts know in what direction the
swarm must fly, and they constitute only about 5% of the bees in a swarm. Nevertheless, a swarm will
fly quickly and directly to its destination. How does the small minority of informed scouts indicate the
swarm’s flight direction to the large majority of uninformed bees? Two hypotheses have been suggested.
The first proposes that the flying scouts streak through the swarm cloud in the direction of the goal, thereby
indicating the travel direction visually (vision hypothesis). The second proposes that flying scouts release
pheromones from their Nasanov glands at the front of the cloud of flying bees, thereby indicating the
travel direction chemically (olfaction hypothesis). We tested both hypotheses by studying the flights of
normal swarms and comparing them to the flights of swarms composed of bees whose Nasanov glands
were sealed shut. Our results support the vision hypothesis and contradict the olfaction hypothesis. We
identified fast-flying bees (‘streakers’) in swarms, as predicted by the vision hypothesis, but we found no
effect of sealing the Nasanov glands of swarming bees. Sealed-bee swarms were perfectly capable of flying
directly to a new nest site.
Ó2005 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.
In many animal species, individuals move about in groups
as they perform seasonal migrations, travel to food sources
and return to safe havens (Boinski & Garber 2000; Krause
& Ruxton 2002). An intriguing question about such group
movements is how they are oriented. In some species, all
individuals in a group share a genetically determined pro-
pensity to travel in a certain direction (Berthold &
Querner 1981; Berthold et al. 1992), or all are involved
in choosing a particular travel direction (Neill 1979; Gru
¨n-
baum 1998). In other species, relatively few of the individ-
uals within a group have pertinent information about the
group’s travel destination, usually because of differences
between individuals in age or experience, and these in-
formed individuals guide those that are not informed.
For example, a few informed individuals within a fish
school can determine the foraging movements of the
group and can steer a group towards a target (Reebs
2000; Swaney et al. 2001). In this study, we investigated
a striking form of group movement that relies on guidance
by a small subset of informed individuals: the flight of
a honeybee swarm.
A swarm of honeybees consists of one queen and several
thousand workers. Swarms are generally produced in the
spring, when large colonies divide themselves for repro-
duction. In this process of colony fissioning, the mother
queen and approximately half the worker bees leave the
parental nest to establish a new colony, while a daughter
queen and the balance of the workers stay behind to
perpetuate the old colony (reviewed in Winston 1987).
The swarm bees leave en masse, forming a cloud of bees
just outside the parental nest, but within about 20 min
they coalesce into a football-sized cluster at an interim
site, usually a nearby tree branch. From here they choose
a nest site.
Several hundred scout bees fly from the swarm cluster to
search out tree cavities and other potential dwelling
places. The dozen or so scouts that find suitable cavities
report their locations by means of waggle dances per-
formed on the surface of the bivouacked swarm, and other
scouts decode the dances, visit the sites themselves, and
may dance in turn. There ensues a process of ‘friendly
Correspondence: M. Beekman, School of Biological Sciences A12, Uni-
versity of Sydney, Sydney, NSW 2006, Australia (email: mbeekman@
bio.usyd.edu.au). R. L. Fathke and T. D. Seeley are in the Department
of Neurobiology and Behavior, Mudd Hall, Cornell University, Ithaca,
NY 14853, U.S.A.
161
0003–3472/05/$30.00/0 Ó2005 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.
AN IM AL BEH A VI OU R, 2006, 71, 161–171
doi:10.1016/j.anbehav.2005.04.009
competition’ among the scouts affiliated with the differ-
ent sites, at the end of which one site comes to dominate
in visits and dancing (Lindauer 1955; Seeley & Visscher
2004). Once this agreement is reached, the swarm takes
flight again and moves to the chosen site, often several kil-
ometres away (Seeley & Morse 1977; Villa 2004).
An intriguing feature of the flight of a honeybee swarm
is that only about 5% of a swarm’s members have visited
the new nest site before swarm liftoff (Seeley et al. 1979).
Furthermore, the other 95% of the swarm’s members have
remained quiescent throughout the decision-making pro-
cess, that is, they have not paid attention to the dances in-
dicating the location of their future home. Therefore,
most of the bees in an airborne swarm do not know in
what direction they must fly until the scouts give them
this information. How do the scouts do so? Two mecha-
nisms of swarm guidance have been proposed. Lindauer
(1955) observed in airborne swarms that some bees fly
through the swarm cloud at high speed and in the correct
travel direction, seemingly ‘pointing’ the direction to the
new nest site. Lindauer suggested that these fast-flying
bees (which we call ‘streakers’) are scouts that have visited
the chosen nest site, and that their behaviour guides the
other, uninformed, bees towards their new home. We refer
to Lindauer’s hypothesis as the vision hypothesis. An al-
ternative to the vision hypothesis is the olfaction hypoth-
esis of Avitabile et al. (1975), who proposed that the scouts
provide guidance by releasing assembly pheromone from
their Nasanov glands (a gland between the last two ter-
gites of the bee’s abdomen) on one side of the swarm
cloud, thereby creating an odour gradient that can guide
the other bees in the swarm.
So far, neither the vision hypothesis nor the olfaction
hypothesis has been tested empirically, although other
investigators have confirmed Lindauer’s report that there
are streakers in flying swarms (Seeley et al. 1979; Dyer
2000). The vision hypothesis has been tested theoretically
in a modelling study in which a small proportion of the
bees in a simulated swarm repeatedly fly through the cen-
tre of the swarm in the direction of the nest site (Janson
et al. 2005). The simulated swarms behave much like
real swarms, which indicates that streakers are, in princi-
ple, able to guide swarms. This modelling study shows,
however, only that the vision hypothesis is a possible
mechanism of swarm guidance; it does not prove that it
is the actual mechanism.
We tested both the vision and olfaction hypotheses of
swarm guidance using real honeybee swarms. We studied
the flights of both normal honeybee swarms and swarms
in which each bee’s Nasanov gland was sealed shut.
METHODS
Study Sites
Our main study site was the Liddell Field Station of
Cornell University at Ithaca, New York, U.S.A. We set up
swarms, one at a time, on the edge of a large, unmowed
field (330 800 m) bounded by woods. Near the centre of
this field stands a large white ash tree, Fraxinus americana,
on which we mounted, 4 m off the ground, a small, unoc-
cupied hive (a six-frame nucleus box). This bait hive pro-
vided the swarms with an attractive nest site in a
controlled location. To enhance the hive’s attractiveness
to the bees, we put inside it a frame of old, empty comb
and a pheromone lure containing an artificial blend of
the main compounds in the secretion of the Nasanov
gland of worker bees (Brushy Mountain Bee Farm, Moravian
Falls, North Carolina, U.S.A.). On the eastern edge of the
field, 270 m from the tree-mounted hive, we set up our
swarms. Each swarm was placed on a swarm mount
(Weidenmu
¨ller & Seeley 1999), positioned in the centre
of a 20 20-m mowed area. Within this mowed area, we
erected 36 1.3-m stakes, 4 m apart, to create a grid for mea-
suring the length and width of each swarm cloud immedi-
ately after liftoff. We also erected a 6-m pole, with 1-m
markings, to measure the heights of the top and bottom
of each swarm cloud. We mowed several paths between
the swarm mount and bait hive so that we could easily fol-
low our swarms once they were in flight. Every 30 m along
the path that ran directly from the mount to the hive, we
erected a distance marker consisting of a 2-m stake topped
with a bright flag.
We used a second study site to record the flight of one
swarm over a greater distance than was possible at our
main study site. This second site, 7 km from the main one,
was a much larger (1200 2000 m) field that was com-
pletely mowed. We established one swarm in the middle
of this field using the same mount as for our main study
site. We also placed two bait hives in trees on the edge
of the field, as described above, but the swarm flew instead
to a hollow tree more than 3 km outside the field.
Swarm Preparation
All our swarms were artificial swarms prepared from
colonies headed by ‘New World Carniolan’ queens (mainly
Apis mellifera carnica) purchased from Strachan Apiaries,
Yuba City, California, U.S.A. We worked with swarms of
two sizes: large ones (ca. 11 500 bees) for making general
observations on swarms in flight, and small ones (ca.
4000 bees) for the experimental test of the olfaction hy-
pothesis. The size of the large swarms corresponds to the
median size of natural swarms, and the size of the small
swarms is within the size range of natural swarms (Fell
et al. 1977).
We prepared each large swarm by removing the queen
from an established colony in a normal hive, caging her in
a small queen cage (3.2 10 1.6 cm), suspending the
queen cage in a larger ‘package’ cage (15 25 25 cm)
of wood with wire screen sides, and then shaking 1.5 kg
of the colony’s worker bees (ca. 11 500 bees) into the
cage using a large funnel. The worker bees quickly clus-
tered around their caged queen. We kept the caged bees
indoors at room temperature and fed them a granulated
sucrose:water solution (1:1 by volume) ad libitum until
the bees started to produce wax scales 3–4 days later. We
then placed the swarm on the swarm mount; we opened
the swarm cage, fastened the caged queen to the mount,
and shook the worker bees on to the base of the mount.
ANIMAL BEHAVIOUR, 71, 1
162
The workers then clustered around their caged queen, and
some of them, the nest site scouts, flew off to search for
possible nesting cavities.
The small swarms were prepared differently from the
large swarms, with each bee receiving either a paint seal
over the Nasanov gland (treatment swarms) or a paint dot
on the thorax (control swarms). For each swarm, we first
collected batches of worker bees in wire cages
(10 10 25 cm) using the large funnel, putting about
1000 bees in each cage. We then shook bees from these ca-
ges into plastic zip-lock freezer bags (ca. 40 bees in each
bag). We placed 10–20 bee-filled bags into a refrigerator
(4C) until the bees were motionless. We then emptied
one bag at a time, spreading the chilled bees in a single
layer on a container of reusable ice (‘blue ice’) to keep
them immobile. To prepare a swarm of bees with sealed
Nasanov glands, we applied a thick layer of paint (Testors
Gloss Enamel, The Testor Corporation, Rockford, U.S.A.)
over the entire dorsal surface of the distal end of each bee’s
abdomen. To prepare a control swarm of unsealed bees, we
applied paint to each bee’s thorax instead of her abdomen.
Each batch of painted bees was then transferred from the
blue ice to a paper towel, to help the bees to warm up and
the paint to dry. When the bees began to crawl about, we
checked their paint marks before placing them in a ‘pack-
age’ cage, as above, that contained their queen, confined
in a queen cage. The bees eventually recovered fully from
being chilled and clustered around the queen. A sugar
solution feeder provided the caged bees with food. Once
we had prepared all 4000 bees in a swarm, we treated
the swarm as described above for the large swarms (i.e.
we fed them with sugar solution until wax scales were
produced, then released them on the swarm mount). We
prepared three treatment (sealed-bee) swarms and three
control (unsealed-bee) swarms. When placing the treatment
swarms on the swarm mount, we assisted the bees in
clustering around their queen by tying an artificial Nasanov
lure to the stand until the bees had formed a cluster.
Flight Data Recording: Large Swarms
We studied the flights of five large swarms. Three made
270-m flights to our bait hive and two made considerably
longer flights to hollow trees. The three swarms that made
relatively short flights flew along a route known in
advance (straight from swarm mount to bait hive), which
enabled us to record their flights with precision. We
encouraged these three swarms to fly to our hive in the
field, rather than to some tree in the forest, by steadily
monitoring the waggle dances performed on each swarm
and removing any dancers that advertised sites other than
our hive. We did not censor the dances in the other two
swarms which flew to hollow trees. One of the two long-
flight swarms was observed at the primary study site and
the other was observed at the secondary study site. We
hoped that by letting these two swarms make long flights,
they would show us the maximum flight speed of a swarm.
Four large swarms lifted off at the primary study site. In
each case, the cloud of flying bees remained essentially
stationary for about a minute after all the bees had taken
flight, and during this time we estimated the dimensions
of each airborne swarm. To do so, we counted the number
of metres in the 20 20-m grid over which the swarm ex-
tended when viewed from the side (i.e. looking along
a line perpendicular to the flight direction) and when
viewed from the back (i.e. looking along the line of flight).
We also recorded the heights of the top and the bottom of
each swarm with the aid of the metre marks on the 6-m
pole. We recorded observations with continuously run-
ning tape recorders that we started when each swarm
began to take-off.
We measured flight speed for each swarm in one of two
ways depending on whether the swarm made a short-
distance flight to our hive or a long-distance flight to a tree.
For each of the three swarms that flew to our bait hive, and
hence along a route with flagged stakes at 30-m intervals,
we recorded when the centre of the swarm passed over
each stake. For each of the two swarms that flew to a far-off
tree, we ran beneath the swarm and planted a flagged stake
every 30 s for as long as we could stay with the swarm. We
later retraced the flight paths of the two swarms and mea-
sured the distances between the stakes.
For the three swarms that flew to the hive, we also
recorded the time that each swarm needed to enter the
hive (‘entry time’). We measured this as the interval
between when the front of a swarm reached the hive
and when the air became quiet again. A swarm produces
a loud buzzing sound while the bees are flying about
outside their new home, but this drops off as the bees land
at its entrance and move inside.
Flight Data Recording: Small Swarms
All but one of the six small swarms that were used to test
the olfaction hypothesis performed a 270-m flight from
swarm mount to bait hive. The one (control) swarm that
did not fly to the bait hive also flew towards the bait hive,
but continued flying past it a few hundred metres to take
up residence in a hollow tree in the woods. For these
swarms, as for the large swarms, we baited the hive with
a pheromone lure to make our hive attractive to the bees.
Using the small swarms to test the olfaction hypothesis
required that we eliminate the artificial pheromone from
the hive before each swarm flew to it. We did this at least
30 min before each swarm took flight, by replacing the
bait hive that contained a swarm lure with one that had
never contained a swarm lure. The scout bees showed no
signs of being disturbed by this swapping of hives. In
one case (treatment swarm 3) the artificial pheromone
lure was removed from the nestbox only moments before
the swarm moved into the bait hive and this bait hive was
not replaced by a clean one, because the events overtook
the observer.
For each of the six small swarms (three treatment and
three control), we recorded its dimensions when starting
flight, speed throughout flight, and entry time at the end
of flight (except in the case of the control swarm that flew
past the bait hive), using the methods described above for
the large swarms. At the end of each swarm’s flight, we
counted the bees on the front of the bait hive that were
BEEKMAN ET AL.: GUIDANCE OF HONEYBEE SWARMS 163
standing in the scenting posture: abdomen raised and
abdomen tip tilted to expose the Nasanov gland. This
behaviour is normally performed to emit Nasanov pher-
omone, but in the case of our treatment swarms little or
no such pheromone was emitted. We made these counts
every minute for 10 min (or less when the counts became
unreliable because of the large number of bees landing
near the hive entrance), starting when the front of a swarm
reached the tree with the bait hive. Finally, within an hour
of when each treatment (sealed-bee) swarm had moved
into the bait hive, we checked the seals of these bees. To
do so, we sampled and inspected 250 bees using the fol-
lowing methods: after removing the bait hive’s lid to ex-
pose the swarm bees, an experimenter randomly plucked
one bee at a time from inside the hive, pinched her be-
tween thumb and index finger, and observed her as she
tried to sting. Each bee curled her abdomen strongly while
trying to sting, so it was easy to see whether the paint seal
over her Nasanov gland was intact or broken.
Photographic Analysis
To investigate the variation in flight speed and flight
direction among the individual bees within a flying
swarm, we took photographs of the second large swarm
during its flight to the bait hive. Photographs were made
with a 35–100-mm zoom lens mounted on a camera body
for 35-mm film. We used colour transparency film with
a slow film speed (DIN 64) and took the photographs with
a moderately long exposure time (1/30 s). The camera was
positioned 20 m from the centre of the swarm and the axis
of photography was perpendicular to the axis of swarm
flight (Fig. 1). The camera was positioned as low as possi-
ble (about 1 m off the ground) so that the background
for most of the swarm was clear sky. This set-up yielded
photographs that captured a 10-m-wide image at the
distance of the swarm centre (20 m) and in which individ-
ual bees appeared as small, dark streaks on a light back-
ground. In each photograph there was a 45-cm size
reference, either the swarm mount or a portable marker,
positioned near the centre of the swarm. We analysed
two of the photographs of large swarm 2: one taken 33 s
after the swarm had completed liftoff, and was starting
to move away from the mount, and one taken 57 s later
(90 s after liftoff), when the swarm was moving steadily
(although still slowly) and the centre of its cloud had
reached the 30-m mark. Each photograph was analysed
by projecting it on to a white surface to create an enlarged
(600 900 mm) image. We measured the length (mm)
and the angle (degrees, relative to horizontal) of each
dark streak that was in focus in the enlarged image
(N¼1333 and 1553 streaks, representing 12 and 13% of
the bees in the swarms).
Because each photograph had a size reference and
recorded the bees’ movements during a known interval
(1/30 s), we were able to calculate for each photograph the
conversion factor between streak length and flight speed
(for both photographs: 1 mm of streak length ¼0.3 m/s
of velocity in the plane of the photograph). This conver-
sion factor is accurate only for bees whose distance from
the camera matched that of the size reference, i.e. 20 m.
Although some bees were closer and some were more dis-
tant, those whose distance from the camera was not with-
in 1.5 m of 20 m were not in focus in the photograph and
so were not included in this analysis. The maximum errors
of our flight speed estimates were, therefore, 8% of 20 m.
Thus, for example, a bee whose streak was 10 mm long
in the enlarged photograph was a bee whose flight speed,
in the plane of the photograph, was 3 m/s. For the bees in
the photograph, we used streak lengths to evaluate the
distribution of flight speeds and streak angles to get a dis-
tribution of flight angles relative to the horizontal, i.e. the
swarm’s travel direction. We could not tell the polarity
(movement left versus movement right) of each streak,
so we assumed that each bee was flying in the general di-
rection of the swarm’s movement (i.e. from right to left in
the photograph). In principle, the bees’ flight angles could
range from 90(streak straight down) through 0 (streak
horizontal) to þ90(streak straight up), but most of the
flight angles were 45 . Finally, in analysing the second
photograph, which recorded the swarm when it was well
under way, we measured streak length and streak angle
separately for the upper and lower halves of the photo-
graph. We did so to compare the distributions of flight
speed and flight angle between the upper portion of the
swarm and the swarm as a whole.
Flight Speed Measurements: Individual Bees
To estimate the maximum flight speed that a worker bee
can achieve in straight-line flight, we measured the flight
durations of 20 bees as each bee flew to and from sugar
water feeders 500, 700 and 900 m along a line extending
from their hive. The slope of the regression line that relates
flight duration (to or from the feeder) and flight distance
indicates the speed of bees in straight-line flight; it is not
10 m
20 m
Swarm
Flight direction
Camera
Figure 1. Bird’s-eye view of the layout used to photograph flying
swarms. The camera (bottom) was 20 m from the swarm (top)
and photographs were taken at an angle of 90 relative to the
swarm’s travel direction. The recorded image was 10 m wide at
20 m, so included most of the swarm.
ANIMAL BEHAVIOUR, 71, 1
164
affected by time spent in circling flight at the end of each
trip, as the bee prepares to settle at the hive entrance or bee
feeder. To stimulate the bees to fly at top speed, we worked
with bees from a hungry colony as they exploited a rich
feeder (Seeley 1995). To make the study colony (which in-
habited a two-frame observation hive) hungry, we re-
moved its honey stores. To make the feeder highly
desirable, we loaded it with a highly concentrated
(2.5 M/litre) sucrose solution that the bees could collect
ad libitum. Twenty bees from this colony were trained to
the feeder and marked with paint for individual identifica-
tion. Once these 20 bees were marked, all additional bees
arriving at the feeder were captured to prevent crowding
at the feeder and to prevent the colony’s ‘nectar’ influx
from becoming high; under these conditions, foragers re-
main highly motivated to forage. For each bee, we re-
corded the time spent making one flight from hive to
feeder and one flight from feeder to hive, for each of the
three distances between hive and feeder. Flight durations
were measured by two observers, one at the hive and
one at the feeder. Both were equipped with a stopwatch
and a two-way radio. To measure the duration of a bee’s
outward flight, the observer at the hive started a stopwatch
when he saw a marked bee leave the hive, alerted the ob-
server at the feeder to look for the arrival of this bee, and
then stopped his stopwatch when the observer reported
that the bee of interest had landed at the feeder. To mea-
sure the duration of a bee’s homeward flight, the roles of
the two observers were reversed. The feeder was positioned
in three successive locations south of the observation hive,
500, 700 and 900 m. We used the same 20 bees for all fe-
eder locations. Wind speed and direction were recorded
at 30-s intervals with an anemometer and wind vane
mounted on a 5-m mast at the observation hive.
RESULTS
Flights of Large Swarms
The large swarms, when airborne, formed clouds of
flying bees 8–12 m long (horizontal axis in travel direc-
tion), 6–8 m wide (horizontal axis perpendicular to travel
direction) and 3–4 m high (Table 1). We estimated the vol-
ume of each swarm’s cloud by multiplying the measure-
ments of length, width and height, and we used this
estimate to calculate the density of bees in each swarm’s
cloud, knowing that each swarm contained approximately
11 500 bees. The range of density estimates was 30–
80 bees/m
3
(mean 50 bees/m
3
). At this density, adjacent
bees are separated, on average, by 27 cm. The three
swarms that performed a 270-m flight to the bait hive
needed 275–322 s to do so (i.e. time elapsed from when
a swarm completed liftoff to when the front of the swarm
reached the tree holding the hive). The range of entry
times for these three swarms was 10 min 22 s to 24 min.
The pattern of change in speed during the flight was
similar for these three swarms (Fig. 2). Each one accelerated
steadily for the first 90 m, reached a peak flight speed of
6–7 km/h (1.7–1.9 m/s) after flying 90–120 m, reduced
its speed after 210–240 m, moved slowly during the final
30 m, and finally stopped at the tree holding the bait hive.
For the three large swarms that flew only 270 m to the
bait hive we suspected that the peak recorded flight speeds
of 6–7 km/h (1.7–1.9 m/s) did not represent the maximum
speeds that flying swarms can achieve. We therefore al-
lowed two swarms to make long-distance flights to hollow
trees. Based on the dances performed by these swarms’
scouts shortly before liftoff, we estimated that these two
swarms made flights of over 1000 and 4000 m to their
new homes. Tall grass and trees limited our ability to
keep in contact with these swarms to only 347 and
1323 m, respectively, that is, for only about the first third
of each swarm’s flight. Nevertheless, for these two swarms
we recorded markedly higher maximum flight speeds
(9.6 km/h or 2.7 m/s and 11.3 km/h or 3.2 m/s) than for
the first three swarms (Fig. 3). The maximum speed of
the second swarm remained stable for the last 5 min
that we were able to follow the swarm.
Flights of Small Swarms
Both treatment and control swarms, when airborne,
formed similar-sized clouds of flying bees (treatment
swarms: 8 m long, 8–10 m wide, 2–3 m high; control
swarms: 8–10 m long, 6–8 m wide, 3 m high; Table 2).
The density estimates were also similar for the two types
of swarms (treatment swarms: 17–31 bees/m
3
, mean
22 bees/m
3
; control swarms: 20–29 bees/m
3
, mean
22 bees/m
3
). At a density of 22 bees/m
3
, adjacent bees
are separated, on average, by 36 cm.
Both treatment and control swarms flew quickly and
directly to the bait hive (control swarm 1 flew to the bait
hive but then continued past it to occupy a hollow tree).
The mean SD flight time for the three treatment swarms
(346.7 27.6 s) was not significantly different from that
for the control swarms (267.0 0.0 s; Student’s ttest:
t
3
¼2.183, P¼0.12). The mean entry time for the three
treatment swarms (17 min 22 s) was, however, significantly
higher than that of the two control swarms for which we
had entry time measurements (8 min 53 s; Student’s ttest:
t
3
¼5.612, P¼0.01). The entry time for treatment swarm
3 was much shorter than that of the other two treatment
Table 1. Measurements for three large swarms that flew 270 m to
a bait hive and one large swarm that flew more than 1000 m to a
hollow tree
Length
(m)
Width
(m)
Height
(m)
Density
(bees/m
3
)
Flight
time (s)
Entry time
(min:s)
Swarm
1
12 8 4 30 297 24:00
Swarm
2
8 6 3 80 322 12:30
Swarm
3
12 8 3 40 275 10:22
Swarm
4
88 3 60 ND ND
Each swarm contained approximately 11 500 bees. The dimensions
of each swarm were measured shortly after liftoff. ND ¼no data.
BEEKMAN ET AL.: GUIDANCE OF HONEYBEE SWARMS 165
swarms, probably because in this swarm, but not the other
two, we did not replace the bait hive that contained
a pheromone lure with an unscented bait hive.
The pattern of change in speed during flight was similar
for treatment and control swarms (Fig. 4). As occurred
with the large swarms, the small swarms accelerated
steadily for the first 90 m, reached peak flight speeds after
flying 90–120 m, slowed after flying 210–240 m, moved
slowly during the final 30 m, and finally stopped at the
tree supporting the bait hive. The only deviation from
this pattern came with control swarm 1, which flew past
the bait hive and did not slow down when nearing it.
The maximum speeds of the treatment swarms (6.8, 3.6
and 6.8 km/h) were similar to those of the control swarms
(6.7, 6.4 and 7.2 km/h), with the obvious exception of
treatment swarm 2. This swarm flew much more slowly
than the other swarms (maximum speed only 3.6 km/h),
probably because it flew against a fierce headwind; the
other swarms encountered at most a slight breeze.
The counts of bees performing the scenting behaviour
(Fig. 5) on the front of the bait hive showed no obvious
difference between treatment (Fig. S1 in the Supplementary
Material) and control swarms.
Of the 250 bees sampled from each treatment swarm
after it had moved into the bait hive only 0.8% (swarms 1
and 2) and 1.6% (swarm 3) of the bees had broken seals.
Flight Patterns of Individuals
While watching the flying swarms, we observed con-
spicuous differences between individuals in flight speed
and pattern. Most bees within a swarm flew slowly and
with curved flight paths, but some flew rapidly and
linearly in the direction of the swarm’s destination. The
latter bees (streakers) appeared to be mainly in the upper
region of a swarm.
0
2
4
6
8
10
12
0–30
30–60
60–90
90–120
120–150
150–180
180–210
210–240
240–270
270–300
300–330
330–360
360–390
390–420
420–450
450–480
480–510
510–540
540–570
Interval (s)
Speed (km/h)
Long-distance flight 1
Long-distance flight 2
Figure 3. Flight speed patterns of the two large swarms that flew more than 1000 m (swarm 4) and more than 4000 m (swarm 5) to hollow
trees. We measured each swarm’s speed throughout the first third of its flight (i.e. as long as we could run along with the swarm; 347 m for
swarm 1 and 1323 m for swarm 2) from distance covered per 30 s of flight.
0
1
2
3
4
5
6
7
8
0–30 30–60 60–90 90–120 120–150 150–180 180–210 210–240 240–270
Interval (m)
Speed (km/h)
Large swarm 1
Large swarm 2
Large swarm 3
Figure 2. Flight speed patterns of the three large swarms that flew 270 m to a bait hive. We measured each swarm’s speed throughout its flight
as time elapsed per 30 m of flight.
ANIMAL BEHAVIOUR, 71, 1
166
Shortly after liftoff, the streaks representing individual
bees in the enlarged (600 900 mm) photographic image
(Fig. S2a in the Supplementary Material) were nearly all
shorter than 10 mm, representing bees whose flight speeds
in the plane of the photograph were less than 3 m/s
(Fig. 6a). There were a few longer streaks (10–35 mm) repre-
senting bees that were flying more rapidly (3–10.5 m/s).
When the swarm had increased its speed and had travelled
30 m (Fig. S2b in the Supplementary Material) fewer bees
produced extremely short (1–3 mm) streaks, indicating
that fewer bees were still moving slowly in the plane of
the photograph (Fig. 6b). The streak length distribution
for the top portion of the swarm was shifted noticeably
to higher values relative to the streak length distribution
for the swarm as a whole (Student’s ttest: t
81
¼6.97,
P<0.001; Fig. 6b). This result implies that the bees in
the top portion of the swarm were flying either more rap-
idly or more in alignment with the swarm’s travel direc-
tion, or both, than were bees in the swarm as a whole.
The distributions of streak angles shortly after liftoff and
57 s later when the swarm was en route, for the whole
swarm, were significantly different, with the latter
Speed (km/h)
0
1
2
3
4
5
6
7
8
0–30 30–60 60–90 90–120 120–150 150–180 180–210 210–240 240–270
Treatment
swarm 1
Treatment
swarm 2
Treatment
swarm 3
(a)
0
1
2
3
4
5
6
7
8
0–30 30–60 60–90 90–120 120–150 150–180 180–210 210–240 240–270
Interval (m)
Control swarm 1
Control swarm 2
Control swarm 3
(b)
Figure 4. Flight speed patterns of the six small swarms. (a) Three treatment swarms (Nasanov glands sealed) and (b) three control swarms
(Nasanov glands not sealed). All flew to the bait hive at 270 m, except control swarm 1, which flew past it en route to a hollow tree. We mea-
sured each swarm’s speed as time elapsed per 30 m of flight between swarm mount and bait hive.
Table 2. Measurements for six small swarms that flew 270 m to
a bait hive (one continued flying past the hive)
Length
(m)
Width
(m)
Height
(m)
Density
(bees/m
3
)
Flight
time
(s)
Entry
time
(min:s)
Treatment
swarm 1
8 8 2 31 288 21:45
Treatment
swarm 2
8 10 3 17 405 19:36
Treatment
swarm 3
8 10 3 19 347 10:45
Control
swarm 1
863 28NDND
Control
swarm 2
10 8 3 17 267 09:00
Control
swarm 3
8 8 3 20 267 08:45
Each swarm contained approximately 4000 bees. Treatment
swarms: bees whose Nasanov glands were painted shut; control
swarms: bees that received paint but did not have their Nasanov
glands painted shut. The dimensions of each swarm were measured
shortly after liftoff. ND ¼no data.
BEEKMAN ET AL.: GUIDANCE OF HONEYBEE SWARMS 167
distribution clustered more tightly around the travel direc-
tion (Watson–Williams test for differences in two distribu-
tions, Zar 1996:F
1,698
¼4.90, P¼0.027; Fig. 7a). This
result implies that as the swarm started its flight, its bees
increasingly adopted flight directions that were aligned
with the direction of swarm movement. For the distribu-
tions of streak angles for bees in the upper and lower
halves of the image of the swarm en route we found no
difference between the streak angles for bees in the upper
and lower halves of the swarm (F
1,397
¼2.67, P¼0.103;
Fig. 7b). For bees shortly after liftoff, the distributions of
streak angles for bees with streak lengths of over 15 mm
were more aligned with the swarm’s movement than
were those of bees with streaks shorter than 15 mm
(F
1,413
¼4.01, P¼0.04; Fig. 7c). This difference disap-
peared when we compared the distributions of streak an-
gles between slow and fast bees, when the swarm was en
route (F
1,503
¼2.81, P¼0.09; Fig. 7d) when all the bees
showed relatively high alignment with the swarm’s move-
ment direction.
Maximum Speed of Flying Bees
Figure 8 shows the two regression lines relating flight
distance and flight time for bees flying either to or from
a highly desirable food source. The slope of the line for
outward flights, made by essentially empty foragers, is
9.45 m/s and that for homeward flights, made by foragers
with loads of sugar solution, is 6.57 m/s. The flight speed
measurements were all made on one day, when the aver-
age wind speed was just 0.8 m/s towards 103 . The light
wind, moving almost perpendicular to the travel direc-
tions of the bees (180 to the feeder and 0 to the hive),
only slightly influenced our measurements of the bees’
flight speeds.
DISCUSSION
At the heart of this study is the question, how does an
informed minority of scouts guide a honeybee swarm as it
flies to its new home? We addressed two possible answers
to this question, the olfaction hypothesis and the vision
hypothesis. It seems doubtful that the olfaction hypoth-
esis is correct, because if it were, then the small treatment
swarms should have been unable to fly directly and
quickly to the bait hive. The treatment swarms, like the
control swarms, flew directly to the bait hive and did so
with flight speed patterns and flight times nearly identical
to those of the control swarms (Fig. 4,Table 2). Further-
more, the sizes and densities of the two types of swarm
showed no differences.
Given the absence of flight differences between the two
types of swarm, one might wonder whether the Nasanov
glands of bees in the treatment swarms were thoroughly
0
20
40
60
80
100
120
012345678910
Treatment swarm 1
Treatment swarm 2
Treatment swarm 3
(a)
Number of bees attempting to scent
0
20
40
60
80
100
120
012345678910
Time after arrival of swarm at bait hive (min)
Control swarm 2
Control swarm 3
(b)
Number of scenting bees
Figure 5. Number of bees trying to secrete Nasanov pheromone (‘scenting’) for (a) treatment swarms (Nasanov glands sealed) and (b) control
swarms (Nasanov glands not sealed) in the first 10 min after the arrival of the swarm at the bait hive. Data for control swarm 1 are missing
because this swarm did not move into the bait hive.
ANIMAL BEHAVIOUR, 71, 1
168
sealed. Three pieces of evidence indicate that they were.
(1) We needed to assist the bees from the treatment
swarms to form a cluster around the queen on the swarm
mount by using a Nasanov lure. (2) When we sampled
bees from the treatment swarms shortly after they occu-
pied the bait hive, only a few (0.8–1.6%) of the bees had
broken seals. (3) The mean entry time of the treatment
swarms was much longer than that of the control swarms
(Table 2) apart from treatment swarm 3, when the bait
hive was not swapped for a clean one, suggesting that
the odour from the artificial Nasanov lure assisted the
swarm to move in quickly. We would expect slow entry
times if the Nasanov glands were sealed in the treatment
swarms, because scout bees normally use the Nasanov
gland pheromone to help locate the new entrance open-
ing (e.g. specific knothole or crack), once the swarm has
reached its destination (Seeley et al. 1979). Treatment
swarms were able to move into the bait hive without ex-
traordinary delay, even though only about 1% of their
bees could release Nasanov pheromone. This finding sug-
gests that the visual cue of scout bees massed at the en-
trance (see Fig. S1 in the Supplementary Material)is
important for indicating entrance location when a swarm
reaches its new home.
Unlike the olfaction hypothesis, the vision hypothesis
remains viable. Ideally, we would have tested the vision
hypothesis with an experiment analogous to what we did
for the olfaction hypothesis, by comparing swarms in
which the scouts could and could not produce visual
signals by streaking. Nevertheless, two pieces of evidence
strongly support the vision hypothesis. The first is that the
peak flight speeds of swarms (2–3 m/s) were well below the
peak flight speeds of individual bees (9–10 m/s; Figs 2, 3,
8). This result shows that it is possible for scout bees to
streak through a flying swarm. The second piece of evi-
dence shows that some bees do streak through swarms.
The photographic analysis indicates that in large swarm
2, both when it was starting its flight and when it was
well under way, a minority of the bees were flying at max-
imum speed (>9 m/s ¼streak length >30 mm) in the di-
rection of the swarm’s destination, and a majority were
flying either more slowly or not in the direction of the
swarm’s flight, or both (Figs 6, 8). Our photographic anal-
ysis does not yield an undistorted distribution of the flight
speeds of bees in a swarm, because it measures for each bee
the component of her velocity vector that is in the plane
of the photograph. Thus, the flight speed of any bee that
was flying in or out of this plane was underestimated. The
most extreme underestimation would arise for a bee flying
perpendicular to the plane of the photograph (directly to-
wards or away from the camera); such a bee would have
a streak length of 0 and we would have measured her
flight speed as 0. Therefore, the flight speed distributions
shown in Fig. 6 are skewed to the left. Nevertheless, the
existence of the long (>30 mm) streaks in these photo-
graphs indicates unambiguously that bees were streaking
through the swarm. Furthermore, these long streaks repre-
sent bees flying near the maximum flight speed, so we are
confident that the bees producing these long streaks were
flying in the plane of the photograph, and hence were
aligned with the direction of the swarm’s flight (to pro-
duce such long streaks without flying in the plane of the
photograph, these bees would need to have flown faster
than the maximum flight speed of bees). We conclude,
therefore, that our photographic analysis documents the
existence of bees streaking through the swarm in the di-
rection of the swarm’s movement.
Besides enabling us to evaluate the two key hypotheses
about the mechanism of swarm guidance by scout bees,
our results provide insights about swarms that can guide
future investigations. One is the finding that the fast-
flying bees, the streakers, appear to be most common in
the upper region of a swarm (Fig. 6b). For humans, and
probably also for bees, streakers are much more easily
seen against bright sky than dark ground or vegetation,
so by flying above most of the bees in a swarm, streakers
may facilitate the transfer of their direction information
to the other bees.
A second general finding about the guidance of swarms
is a clearer picture of how a swarm slows itself at the end
of its flight. One possibility is that the scout bees drop out
of the swarm cloud, land at the nest site, and release
Nasanov pheromone to attract the other bees to the
endpoint of their trip (Janson et al. 2005). Observations
on real swarms (Seeley et al. 1979) are consistent with
0
Frequency
0.05
0.1
0.15
0.2
0.25
0.3
0.35
147101316192225283134374043464952
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
147
10 13 16 19 22 25 28 31 34 37 40 43 46 49 52
Streak length (mm)
Total
Top only
(a)
(b)
Figure 6. Frequency distributions of lengths of the streaks in the
photographs shown in the Supplementary Material when these pho-
tographs were enlarged to 600 900 mm based on (a) photo S2a in
the Supplementary Material (N¼1333) and (b) photo S2b in the
Supplementary Material. Two distributions are shown in (b), one
for all the streaks in the photograph (N¼1553) and one for only
the streaks in the top half of the photograph (N¼78). For both
(a) and (b) the conversion factor between streak length and flight
speed is 1 mm ¼0.3 m/s. Streaks longer than 35 mm probably arose
through the overlap of two shorter streaks, because they imply im-
possibly high flight speeds.
BEEKMAN ET AL.: GUIDANCE OF HONEYBEE SWARMS 169
this possibility, but we found a gap of 2–4 min between
when the front of a swarm stopped at the tree holding
the bait hive and when many bees were landing at the
entrance and scenting (Fig. 5). Thus the braking process
seems to involve something other than the scout bees re-
moving themselves from the swarm and flying ahead to-
wards the new nest site. Supporting evidence is that
swarms consistently began to slow themselves 30 m or
more before reaching the bait hive (Figs 2, 4). Scouts are
evidently able to cause the swarm to slow down well
before they leave the swarm cloud to pinpoint the en-
trance opening of the new nest site. A possible mechanism
by which this can be achieved is by simply slowing down
without leaving the swarm (Janson et al. 2005).
Another general finding is that the density of bees in
a flying swarm is usually in the range of 20–60 bees/m
3
,
which implies that the average distance between bees is
about 25–35 cm. This result relates to the theoretical study
of swarm guidance by Janson et al. (2005). To test the vi-
sion hypothesis, Janson et al. assumed that scouts guide
a swarm by flying repeatedly, slightly faster than unin-
formed bees, through the swarm’s centre in the direction
of their new home. Janson et al. then incorporated this as-
sumption into a computer model of a flying swarm and
ran simulated flights of their in silico swarms. In this model,
the bees react to their nearest neighbours so that they
avoid those that are too close while being attracted to other
bees when the distance between neighbouring bees
increases (these behavioural rules are general rules used
in models of group movement). Swarm guidance is
achieved by the bees aligning their movements with those
of the bees within their neighbourhood. The scouts, by
initially flying faster than uninformed bees, have more
influence on the directional movement of the uninformed
bees when the swarm has not yet started to move, thereby
steering the swarm into the direction of the new home (as
soon as the swarm has reached its maximum speed, the
influence of the scouts is greatly reduced). In real swarms,
however, the density of bees appears to be low, suggesting
that an individual bee has relatively few neighbours to
which it reacts. Furthermore, scouts in real swarms appear
to be located mostly in the upper part of the swarm (and
(a)
(b)
Lower half Top half
After liftoff En route
En route
Whole swarm
En route
Slow bees
Slow bees Fast bees
Fast bees
After liftoff
(d)
(c)
Figure 7. Circular distributions of the angles of the streaks in photographs S2a, b (Supplementary Material). A horizontal streak, produced by
a bee flying directly towards the bait hive, is denoted by a dot in the direction of the dashed line (0 ). Solid line: mean vector bearing (MVB).
When the number of data points exceeded 350, a random sample of 350 data points was chosen to draw the plots. (a) Comparison between
just after liftoff (MVB ¼358.8, MVB length ¼0.91) and while en route (MVB ¼2.2, MVB length ¼0.94), for the whole swarm. (b) Comparison
between the top (MVB ¼359.9, MVB length ¼0.97) and bottom (MVB ¼3.6, MVB length ¼0.93) halves of the photograph of the swarm en
route. (c, d) Comparisons between streaks shorter than 15 mm (Slow bees) and longer than 15 mm (Fast bees). (c) Slow bees: MVB ¼359.4,
MVB length ¼0.90; Fast bees: MVB ¼353.5, MVB length ¼0.97. (d) Slow bees: MVB ¼2.8, MVB length ¼0.93; Fast bees: MVB ¼0.0, MVB
length ¼0.97.
0
200
400
600
800
1000
050
100 150 200
Flight time (s)
Flight distance (m)
Outward flights
Inward flights
Figure 8. Regression lines showing the maximum flight speed for in-
dividual bees when empty (Outward flights) and after having for-
aged at a feeder (Inward flights) at 500, 700 and 900 m from the
hive. The data points are the mean SD of 20 bees.
ANIMAL BEHAVIOUR, 71, 1
170
hence do not fly through the centre of the swarm as in the
simulation model), which probably increases their visibil-
ity. By increasing their visibility the scouts increase the
number of bees whose behaviour is influenced (i.e.
increasing the number of bees in their neighbourhood),
and doing this may have a similar effect as when the
scouts fly through the centre of a swarm cloud.
Finally, a second recent theoretical paper has shown
that the movement of an animal group can be guided by
a few informed individuals without these individuals
providing explicit guidance signals and even without
anyone in the group knowing which individuals possess
information about travel direction (Couzin et al. 2005).
Couzin et al. presented a model of a moving group in
which each individual attempts to maintain a personal
space by turning away from neighbours within a certain
range, and at the same time each individual attempts to
move in a preferred direction, if it has one. In this model,
only the informed members of the group have a preferred
direction, and their tendency to go in this direction steers
the group. Couzin et al. found that the proportion of in-
formed individuals needed for accurate directional guid-
ance of the group is potentially small, less than 10%.
This theoretical study offers a variant of the vision hy-
pothesis for the mechanism of swarm guidance in honey-
bees, in which the scouts do not produce a visual signal
(i.e. streaking) to steer a swarm, but instead simply tend
to fly in the direction of the chosen nest site. We call
this the ‘subtle guide’ version of the vision hypothesis. A
logical next step in the study of honeybee swarm guidance
would be to test experimentally the streaker bee and sub-
tle guide forms of the vision hypothesis. It will be interest-
ing to see if natural selection has or has not favoured the
use of signals by the informed minority of scouts as they
guide a honeybee swarm to its new home.
Acknowledgments
We are grateful to Adrian Reich, who so patiently analysed
the photographs and helped in many other ways. We also
thank Ben Oldroyd for constructive comments and sup-
port throughout the project and Stefan Janson for com-
menting on the manuscript. M.B. was supported by the
Australian Research Council and received additional fi-
nancial support from the Australian Academy of Science.
The Hughes Summer Scholars Program at Cornell pro-
vided a fellowship to R.L.K. T.D.S. received financial
support from the U.S. National Science Foundation
(research grant IBN02-10541).
SUPPLEMENTARY MATERIAL
Supplementary material associated with this article can be
found, in the online version, at doi:10.1016/j.anbehav.
2005.04.009
References
Avitabile, A., Morse, R. A. & Boch, R. 1975. Swarming honey bees
guided by pheromones. Annals of the Entomological Society of
America,68, 1079–1082.
Berthold, P. & Querner, U. 1981. Genetic basis of migratory behav-
iour in European warblers. Science,212, 77–79.
Berthold, P., Helbig, A. J., Mohr, G. & Querner, U. 1992. Rapid mi-
croevolution of migratory behaviour in a wild bird species. Nature,
360, 668–670.
Boinski, S. & Garber, P. A. (Eds) 2000. On the Move: How and Why
Animals Travel in Groups. Chicago: University of Chicago Press.
Couzin, I. D., Krause, J., Franks, N. R. & Levin, S. A. 2005. Effective
leadership and decision making in animal groups on the move.
Nature,455, 513–516.
Dyer, F. C. 2000. Group movement and individual cognition: les-
sons from social insects. In: On the Move: How and Why Animals
Travel in Groups (Ed. by S. Boinski & P. A. Garber), pp. 127–
164. Chicago: University of Chicago Press.
Fell, R. D., Ambrose, J. T., Burgett, D. M., De Jong, D., Morse,
R. A. & Seeley, T. D. 1977. The seasonal cycle of swarming in
honeybees. Journal of Apicultural Research,16, 170–173.
Gru
¨nbaum, D. 1998. Schooling as a strategy for taxis in a noisy en-
vironment. Evolutionary Ecology,12, 503–522.
Janson, S., Middendorf, M. & Beekman, M. 2005. Honeybee
swarms: how do scouts guide a swarm of uninformed bees?
Animal Behaviour,70, 349–358.
Krause, J. & Ruxton, G. D. 2002. Living in Groups. Oxford: Oxford
University Press.
Lindauer, M. 1955. Schwarmbienen auf Wohnungssuche. Zeitschrift
fu
¨r Vergleichende Physiologie,37, 263–324.
Neill, W. H. 1979. Mechanisms of fish distribution in heterothermal
environments. American Zoologist,19, 305–317.
Reebs, S. G. 2000. Can a minority of informed leaders determine the
foraging movements of a fish shoal? Animal Behaviour,59, 403–
409.
Seeley, T. D. 1995. The Wisdom of the Hive. Cambridge, Massachu-
setts: Harvard University Press.
Seeley, T. D. & Morse, R. A. 1977. Dispersal behavior of honey bee
swarms. Psyche,84, 199–209.
Seeley, T. D. & Visscher, P. K. 2004. Group decision making in nest-
site selection by honey bees. Apidologie,35, 101–116.
Seeley, T. D., Morse, R. A. & Visscher, P. K. 1979. The natural
history of the flight of honey bee swarms. Psyche,86, 103–
113.
Swaney, W., Kendal, J., Capon, H., Brown, C. & Laland, K. N.
2001. Familiarity facilitates social learning of foraging behaviour
in the guppy. Animal Behaviour,62, 591–598.
Villa, J. D. 2004. Swarming behavior of honey bees (Hymenoptera:
Apidae) in southeastern Louisiana. Annals of the Entomological So-
ciety of America,97, 111–116.
Weidenmu
¨ller, A. & Seeley, T. D. 1999. Imprecision in waggle
dances of the honeybee (Apis mellifera) for nearby food sources:
error or adaptation? Behavioral Ecology and Sociobiology,46,
190–199.
Winston, M. L. 1987. The Biology of the Honey Bee. Cambridge,
Massachusetts: Harvard University Press.
Zar, J. H. 1996. Biostatistical Analysis. Upper Saddle River, New
Jersey: Prentice Hall.
BEEKMAN ET AL.: GUIDANCE OF HONEYBEE SWARMS 171
