To minimise the risk of colliding with the ground or other obstacles, flying animals need to control both their ground speed and ground height. This task is particularly challenging in wind, where head winds require an animal to increase its airspeed to maintain a constant ground speed and tail winds may generate negative airspeeds, rendering flight more difficult to control. In this study, we investigate how head and tail winds affect flight control in the honeybee Apis mellifera, which is known to rely on the pattern of visual motion generated across the eye – known as optic flow – to maintain constant ground speeds and heights. We find that, when provided with optic flow cues in both the longitudinal and transverse directions of flight, honeybees maintain a constant ground speed but fly lower in head winds and higher in tail winds, a response that is also observed when longitudinal optic flow cues are minimised. This change in height with wind does not appear to result in a constant rate of optic flow in the ventral visual field, suggesting that honeybees may rely on a combination of mechanosensory and visual information when controlling flight in wind. We also find that, when the transverse component of optic flow is minimised, or when all optic flow cues are minimised, the effect of wind on ground height is abolished. We propose that the regular sidewards oscillations that the bees make as they fly may be used to extract information about the distance to the ground, independently of the longitudinal optic flow that they use for ground speed control. This computationally simple strategy could have potential uses in the development of lightweight and robust systems for guiding autonomous flying vehicles in natural environments.

Training

The experiments were carried out in an All Weather Bee Flight Facility at the Australian National University’s Research School of Biological Sciences. The temperature inside the facility was maintained at 24 ± 5 °C during the day. A beehive mounted on the wall supplied the honeybees (Apis mellifera L.) used in the experiments. For each experiment, up to 16 honeybees were individually marked and trained to fly to a feeder containing sugar solution placed at the end of the test section nearest the fan.

In each experiment, the pattern was displayed in the experimental tunnel two days prior to testing and the bees were allowed to forage freely during this time. The bees were trained in still air and were only presented with the wind conditions during the testing periods, which lasted for 30 min. Between test conditions, bees were allowed to visit the feeder in still air for at least 30 min to minimise learning effects across trials. Each experiment was conducted over three days with each test (wind) condition being presented in a randomised order on each day.

Experimental apparatus 

The experiments were conducted in a square cross-section wind tunnel that was constructed of clear Perspex (figure S1). The tunnel consisted of an entrance section, a test section, and a settling chamber. The entrance section was 1.1 m in length with a cross-section that contracted from 400 mm x 400 mm at the open end to 200 mm x 200 mm where it attached to the test section. The test section of the tunnel was 3.5 m in length and had a uniform cross-section of 200 mm x 200 mm. An opening at the fan-end of the test section allowed access to the tunnel, which was required for the placement of the sugar-water feeder. The end of the test section was attached to a settling chamber which consisted of three component sections: 1) a 1.1 m long chamber with a cross section that expanded from 200 mm x 200 mm to 400 mm x 400 mm, 2) a 400 mm x 400 mm honeycomb section that consisted of a 150 mm wide array of plastic PVC tubes (each 15 mm in diameter), 3) a 1.1 m long section that contracted from 400 mm x 400 mm to 200 mm x 200 mm, where it connected to the fan. The role of the settling chamber was to reduce the turbulence created by the rotation of the fan blades and to ensure that the air flow in the test section was following a smooth path. The fan was driven by a variable voltage, bi-directional DC motor to generate head winds or tail winds at speeds of up to 10 m.s^-1 in the test section of the tunnel. Wind speeds in the test section of the tunnel were measured at using both a hot-wire anemometer and a fan anemometer at 50 mm, 10 mm and 150 mm from the walls and floor at the fan-end and entry of the test section, and then calibrated with the voltage settings of the fan. The air speeds did not change across the test section of the tunnel (and were equal to those provided in the analysis) as it had been specially designed to produce laminar air flow.

The tunnel and the camera that was positioned above the tunnel were covered with a white cloth to minimise reflections from the Perspex roof and enable a clear view of the honeybees’ body orientation and position. This also had the effect of minimising optic flow cues in the bee’s dorsal visual field, although the lens of the camera filming from above would have been visible as a small black circle in the white covering.

Image analysis

Flights in the test section en route to the feeder were filmed at a rate of 100 Hz using two synchronised CMOS cameras (MotionPro 10k, Redlake Inc.). The optical axes of the cameras were positioned orthogonal to each other such that one camera provided a top-view of the honeybees’ flight trajectories along the tunnel, whilst the second camera provided an end-on view along the length of the tunnel. Flights were recorded over a distance of 400 mm in the tunnel’s mid-section and later tracked in image sequences recorded from each camera view using an automated tracking program developed in-house.

Data analysis

Ground speed was calculated from the camera that was positioned above the mid-section of the tunnel; ground height was calculated from the camera that was positioned with an end-on view of the tunnel. The pixel positions of the honeybee in each camera view were converted to mm values using the known size and position of objects in the tunnel. From this, it was possible to calculate the coefficients required for the pixel to mm conversion for each position in the tunnel from each camera view. The error associated with this method of converting pixel coordinates to three-dimensional world coordinates was determined to be below 10 mm from calculations of objects of known size

The mean ground speed and ground height values were calculated for individual flights. Unless otherwise stated, all values are given as the mean ± standard deviation.

Statistical analysis

Linear mixed models using the ‘lme’ function from the ‘nlme’ package in R (R Core Team 2018) were used to analyse the effect of wind speed on the tested flight parameters. Students t-tests were used to test for significance at the 5% level (see electronic supplementary material for the data set, details of the models used and the output of the analyses). Bee identity was included as a random effect in the statistical models to account for the variation between and within flights from individual bees. The number of individual bees and flights are provided in Table 1.

Experimental conditions

In all experiments, flights were recorded in head and tail winds of 1 m.s^-1 or 2 m.s^-1, or in still air when the tunnel walls and floor displayed one of four different visual textures. The wind speeds were limited by the maximum speed that bees would fly at in a tail wind. At speeds superior to 2 m.s^-1, the bees would either land on the floor or turn around and fly out of the tunnel.  The patterns used in each experiment were created by attaching strips of red electrical tape (18 mm in width) on to sheets of white laminated paper, 3.5 m long  and 200 mm wide, and attaching these to the outside of the walls or the floor of the transparent Perspex tunnel – attaching them to the inside may have led to unwanted turbulence – leading to a 210 mm distance between the visual textures on the walls (calculated by adding the width of the Perspex on the walls, 5 mm to the internal width of 200 mm). The strips of tape were distributed evenly along either the length or width of the paper (depending on the pattern) with a 40 mm or 25 mm edge-to-edge separation for the longitudinal and transverse strips, respectively.

The visual textures used in these experiments were designed to test the effect of different optic flow cues on the flight control of bees flying in wind and still air. In Experiment 1, the tunnel was lined with a pattern consisting of transverse and longitudinal stripes that created a cross hatch pattern (figure 1a). This pattern provided strong longitudinal and transverse optic flow cues for bees flying to the feeder. In Experiment 2, the tunnel was lined with stripes oriented along the longitudinal axis of the tunnel to generate strong transverse optic flow cues, whilst minimising longitudinal optic flow cues (figure 1b). In Experiment 3, the tunnel was lined with stripes oriented along the transverse axis of the tunnel to generate strong longitudinal optic flow cues, whilst minimising transverse optic flow cues (figure 1c). In Experiment 4, the tunnel was lined with just the white sheets, minimising both transverse and longitudinal optic flow cues (figure 1d).

To investigate whether the bees were adjusting their ground speed and ground height in wind to maintain a constant rate of longitudinal optic flow in the lateral and ventral visual fields, we calculated the average maximum rate of longitudinal (front-to-back) optic flow that the they would experience in each flight trajectory in both their ventral and lateral visual fields. This analysis was performed for Experiment 1, where the bees would have experienced strong longitudinal and transverse optic flow cues. The maximum rate of longitudinal optic flow in the ventral visual field was calculated in deg.s^-1 as (V/h)*(180/π), where V is the mean ground speed in mm.s^-1 and h is the height above the ground in mm. The maximum rate of longitudinal optic flow in deg.s^-1 that would have been experienced in each flight in the lateral visual field was calculated as (V/d)* (180/π), where V is the mean ground speed in mm.s^-1 and d is the distance to the wall pattern, 105 mm.

Effect of visual texture on the transverse component of flight

An initial observation of the flights suggested that they contained regular transverse oscillations (figure 1e-h). To investigate whether these oscillations were indeed regular and whether they were affected by the properties of the visual texture, three additional analyses were conducted on the flight data for the 0 wind speed condition and compared across Experiments 1, 3 and 4. Data from Experiment 2 (longitudinal stripe pattern) was excluded from this analysis because there were too few data points per flight to perform an accurate frequency analysis due to the high ground speeds in this experiment.

Fourier transform analyses were performed on the transverse velocity values obtained from each individual flight using Matlab (Mathworks, USA). Observations of the individual power spectra for each flight confirmed the initial observation that, in almost all flights, the transverse oscillations contained a single, dominant frequency (that is, the frequency with the highest power). The dominant frequency value for each flight was then averaged and compared across treatments.

The maximum transverse velocity for each oscillation within individual flights was identified by locating the point where the preceding and succeeding values of transverse velocity were smaller, indicating a maximum. A Gaussian filter of 10 frames (standard deviation = 2.5 ms) was applied to the velocity data to remove the noise (caused by tracking inaccuracies) before performing this analysis in order to be able to accurately identify the maximum velocity for each transverse oscillation. The mean and standard deviation (std) of all the maximum transverse velocity values were then calculated for each flight.

The statistical models used in the paper were run in R (R Core Team, 2018) and can be found in the 'Statistical analysis' supplementary material along with the output of the analyses presented in the paper. The variables have the same name as the headings in each of the columns in the Data.xlsx file. Treatment is wind speed but reordered so that the still air condition is used as the contrast. Each sheet in the Data file is data from each experiment, with DataExpt1 being the data imported from the data sheet entitled ‘Experiment 1’ etc. There are two extra columns in the data sheet Experiment 1 and these are the calculations for the average maximum optic flow experienced in the lateral visual field (LOF) and ventral visual field (VOF). For details of how this was calculated, please see the Methods section of the paper.

The x, y, z position data for each flight given in the Data.xlsx file can be found as individual '.mat' files in the 'Raw flight data.zip' file. The names of each flight correspond to the data in the first column of each sheet for each experiment in the 'Data.xlsx' file.