Research articleSynchronization, coordination and collective sensing during thermalling flight of freely migrating white storks
Abstract
Exploring how flocks of soaring migrants manage to achieve and maintain coordination while exploiting thermal updrafts is important for understanding how collective movements can enhance the sensing of the surrounding environment. Here we examined the structural organization of a group of circling white storks (Ciconia ciconia) throughout their migratory journey from Germany to Spain. We analysed individual high-resolution GPS trajectories of storks during circling events, and evaluated each bird's flight behaviour in relation to its flock members. Within the flock, we identified subgroups that synchronize their movements and coordinate switches in their circling direction within thermals. These switches in direction can be initiated by any individual of the subgroup, irrespective of how advanced its relative vertical position is, and occur at specific horizontal locations within the thermal allowing the storks to remain within the thermal. Using the motion of all flock members, we were able to examine the dynamic variation of airflow within the thermals and to determine the specific environmental conditions surrounding the flock. With an increasing amount of high-resolution GPS tracking, we may soon be able to use these animals as distributed sensors providing us with a new means to obtain a detailed knowledge of our environment.
This article is part of the theme issue ‘Collective movement ecology’.
1. Introduction
Many birds exploit vertical air motion to facilitate travel with minimum energetic costs [1–4]. Soaring migrants gain altitude using rising columns of air created by solar radiation (often referred to as thermals), after which they glide in order to obtain distance. To thermal efficiently, birds need to adjust their flight speed and circling radius to find, and remain close to, the centre of the thermal where updraft is highest. Manoeuvring within updrafts can be challenging because thermal shape and strength are influenced by horizontal winds, and often change rapidly [5–8]. In addition, the potential energy that birds can gain during circling depends also on individual features such as wing morphology [9], age and experience [3,10]. Many avian species travel in cohesive flocks to reach their wintering area because being in a group can help individuals gather reliable information about the surrounding environment [11–14]. For soaring birds, the presence of conspecifics can provide information on the location and strength of updrafts [15,16]. But the response to social cues adds another level of complexity to the already challenging task of thermal soaring, because directional changes within the group need to propagate quickly, and often across the entire flock [17,18]. Thus, to achieve coordination, and maintain flock cohesion while circling in a thermal, group members need to react appropriately not only to their neighbours but also to the surrounding air.
While some bird migrants travel in flocks consisting of subgroups of familiar and/or related individuals [11,19,20], others form large groups without differentiated social relations. The latter form of group organization can lead to aggregations of hundreds to hundreds of thousands of birds at hotspots with favourable flight conditions [21]. Yet, although these flocks may possess a spatio-temporally dynamic group structure, ultimately leading to so-called fission–fusion systems [22], they still require coordination and synchronization to maintain cohesion and maximize the advantages arising from group membership. Local inter-individual interactions can provide substantial benefits to individuals, for example, by facilitating emergent sensing of long-range environmental gradients [23]. By functioning like a distributed sensory array, individuals can increase their effective perceptual range, which may be particularly important for migrating species that move along long-distance gradients [16,24–26]. But also on a smaller scale, individuals could benefit from collective sensing, allowing them, for example, to improve the finding and exploitation of constantly changing structures such as thermal updrafts.
Here our goal was to explore the fine-scale, collective movement decisions of a group of white storks (Ciconia ciconia) during their long-range travel. White storks from the ‘western’ subpopulation travel up to 4000 km through Western Europe to the northwestern parts of Africa, mostly by soaring–gliding flight [27]. During migration, they congregate in large flocks potentially to follow more experienced conspecifics [12]. Here, we analysed the individual high-resolution GPS trajectories of stork flocks flying in thermals (in total 51 thermals with a minimum of 10 circling birds, figure 1a,b). Henceforth, we refer to thermalling flight as circling. First, we assessed each bird's individual flight behaviour in relation to its flock members, with the predictions being that (i) individuals coordinate their movements to move in synchrony while trying to remain inside the complex thermal structure, and that (ii) the spontaneous interactions between neighbouring individuals shape dynamic subgroups. Second, we used the birds' movements within the flock to estimate the dynamic variation of the moving air that forms the thermals to create a detailed picture of the specific environmental conditions through which the flock is migrating. Taken together, we evaluated each individual's flight behaviour in relation to its flock members thereby providing us with detailed insights into the organization, dynamics and functioning of a group of long-distance migrants.
Figure 1. Collective flight trajectories and examples of synchronization in circling direction. (a) Flock trajectories (1 Hz GPS data) of migrating storks showing different flight types (circling and gliding). Grey arrows indicate the flight direction. Tracks are colour-coded based on the horizontal path curvature, κ. |κ| is close to 0 during gliding and typically larger than 0.01 during circling. Positive (yellow-red) and negative (cyan-blue) values of κ indicate counterclockwise and clockwise circling, respectively. The thermal is drifting with the wind resulting in distorted trajectories even if the bird flies in a perfect circle relative to the moving air. This can be seen in peak values of κ. (b) Enlarged view of the tracks marked in (a). (c) Curvature against time for each individual. Each function is shifted upwards by the ID of the bird. Black circles show a switch in curvature during circling (IDs of the birds are shown on the right). Arrows at the top depict the time delays between the switches relative to the first individual. (d) Horizontal trajectories of the birds that switched their circling direction (same time interval as in (c)). Tracks (except bird 2) are shifted according to a grid of 150 m for better visibility. Numbers indicate bird ID.
2. Material and methods
(a) Dataset
To study collectively migrating white storks (C. ciconia), we equipped 61 juvenile storks one week prior to fledging with high-resolution, solar GSM-GPS-ACC loggers (e-obs GmbH, Munich, Germany). With birds originating from 22 different nests, we tracked the entire offspring of a small stork colony in the south of Germany (47°45′10.8″ N, 8°56′2.4″ E). Tagging occurred from June to July 2014; the tagging permit number G-13/28 was issued by the Regierungspräsidium Freiburg (Federal State of Baden-Württemberg, Germany).
The transmitters (weight 54 g) were attached using a Teflon-nylon harness (weight approx. 12 g). The total weight of transmitter and harness was 66 g, corresponding to approximately 2% of the mean body mass of white storks. To capture group movements during migration at high temporal resolution, we recorded GPS locations and three-dimensional body acceleration for 18 h a day (between 04.00 and 22.00 local time at the natal ground). The GPS provided high frequency (1 Hz) observations every 15 min for 2 or 5 min; positions have a positional accuracy of ±3.6 m (i.e. when stationary, 50% of fixes remain within a radius of 3.6 m within 24 h). Each GPS fix consisted of geographical position and elevation in WGS84 coordinates, speed and heading. Owing to limited tag memory and high data transmission costs, these settings required a data download via an ultra-high frequency radio link every 3–5 days (from a distance of approx. 300 m). This is why we only used this resolution for the birds' autumn migration (approx. till September 2014). Accelerometers collected data every 10 min for a short period (3.8 s). These bursts did not overlap with the GPS recordings, and we did not use it for the analyses of this study. The data that were used for this study are part of the MPIO white stork lifetime tracking data [27] and are available through the Movebank Data Repository (https://www.movebank.org/node/15294).
(b) Flock flights
Of the 61 tagged storks, 27 individuals departed together as a flock from their natal ground and migrated as a group to the north of Spain (41°55′49″ N, 2°15′17″ E) providing us with 1 414 226 locations of approximately 1000 km of collective migration. During the first 5 days of migration, some individuals stopped migration completely (n = 1), joined different flocks (in total n = 8) or died (n = 1), resulting in a group of 17 birds arriving together in Spain.
(c) Data analyses
(i) Trajectories and circling
We converted the geodetic coordinates provided by the 1 Hz GPS into x, y and z coordinates using the Flat Earth model with an (x, y) = (0, 0) origin at the beginning of each burst. We smoothed these coordinates using a Gaussian filter (σ = 1 s). Birds were assigned to a flock if they had at least 10 flock mates within 200 m. Because we were mainly interested in individual and group behaviour while exploiting long stretches of thermal upwinds, we removed non-circling parts using a minimum vertical speed of vz ≤ 1 m s−1. This way we were able to analyse continuous circling segments that gave reliable information about group synchronization. Because the bird and the thermal are drifting with the wind, the measured GPS trajectories represent the sum of the animal movement vector (heading and speed) and the wind vector resulting in a ‘trochoidal’ path in the horizontal plane (figure 1d).
(ii) Curvature
Curvature was calculated as κ = (vxay − vyax)/(vx2 + vy2)1.5 from the horizontal components of the velocity (vx, vy) and the acceleration (ax, ay) after smoothing all components with a running average over three neighbouring time points (t − 1 s, t, t + 1 s). To avoid values with high curvature, we removed data points with |κ| > 1/m (i.e. 0.3% of the total data points were removed).
(iii) Synchronized path segments
We determined the group's level of synchronization by calculating the directional correlation delay [28] between pairs of birds in a 30 s time window. The directional correlation delay method identifies similar path segments and quantifies the temporal relationship of the birds' switches in flight direction. Individuals circling in the same direction obtain high correlation values; these increase even more when birds circle at the same phase following a similar path (i.e. the same part of the circle, but at different altitudes).
At each time step t, Cij(τ, t) correlation values were calculated for an interval [t – 15 s; t + 15 s], where Cij(τ, t) = 〈vi(t′)·vj(t′ + τ)〉t′∈[t–15s;t + 15s]. A path segment was synchronized when of all possible time delays (τ) the maximum correlation value, Cij(t) = maxτ∈[−10s,10s]Cij(τ, t), was higher than a threshold, Cij(t) ≥ Cmin = 0.9 (for details see [28]; for details on how GPS accuracy affects directional correlation functions see the supplementary information of [28,29]).
(iv) Size of subgroups
At each time point, we calculated for each individual the number of flock members with which it was synchronized. For each burst, we chose the largest number of individuals as the size of the biggest group circling in synchrony.
(v) Proportion of birds that circled in the same direction
We defined the ratio of individuals circling in the same direction as the number of birds circling in the same direction divided by circling birds (circling either in the same or in opposite directions).
(vi) Synchronization ratio
Synchronization ratio (rsynch) was defined as the time spent moving on a synchronized path segment (see above) divided by the total time of circling together for each pair of birds. rsynch was calculated for every burst, and pooled separately for each migration day.
(vii) Switch in circling direction
An individual performed a switch in circling direction at time t, if (i) the sign of curvature, κ, did not change throughout at least 80% of the preceding interval [t − 11 s, t − 1 s]; (ii) did not change for at least 80% of the following interval [t + 1 s, t + 11 s]; (iii) |κ| ≥ 0.01 m−1 within both intervals before and after the switch; and (iv) the mean value of κ in the interval before and after the switch had opposite signs.
(viii) Synchronized switches in circling direction
We defined a synchronized switch in circling direction (SSCD) for a pair, if (i) both individuals switched their circling direction and circled in the same direction before and after the switch, and (ii) the time between the switches was smaller than 20 s. Although this is an arbitrarily chosen threshold, we decided upon a conservative measure. In 91% of all SSCD, the time between the switches was smaller than 10 s.
(ix) Estimation of the centre of the thermal for each altitude
We defined the centre of the thermal at altitude z as the weighted sum of the horizontal locations of all individuals in the range of [z − 5 m, z + 5 m]. The weights were defined by the relative altitude difference using a Gaussian distribution with σ = 3 m. Typically, 102 ± 33 (mean ± s.d.) locations were used for each altitude resulting in a smoothly changing ‘spine’ for the thermal.
(x) Estimation of air velocity at different locations of the thermal
The vertical component of the air velocity was estimated for a 5 × 5 × 4 m3 (x, y, z) sized grid using the average vertical speed of the individuals. While rising in thermals, storks sink in the air that rises around them at a faster speed. We defined the minimum sinking speed as 1.2 m s−1 which corresponded to the peak value of the sink rate during gliding. We estimated the air's vertical speed to be 1.2 m s−1 larger than the vertical speed of the storks. The horizontal speed of the air is estimated from the path of the centre of the thermal (see above). We used the average vertical speed of the individuals for the analysis; the estimated air speed was used for the visualization of the thermal only.
(xi) Estimation of circling period, speed and radius
We identified the period of the circling, Tcircle, as the average time between neighbouring local minima calculated separately for the two vertical components of the velocity, vx and vy. We calculated the average circling speed, vcircle, as half of the difference between neighbouring local maxima and minima points separately for the two vertical components of the velocity, vx and vy. We derived the radius of the circling, rcircle, from the estimated period and speed of circling as rcircle = vcircle · Tcircle/(2π).
(xii) Calculations
Data were analysed using custom-written dedicated scripts in Perl (v. 5.22.1) and Cuda (v. 7.5.17).
3. Results
We determined the group's level of synchronization using two different approaches. First, we identified segments of synchronized circling by examining only the birds' circling direction. Using each bird's trajectory data, we obtained its circling direction by measuring the path curvature, κ, with κ > 0 representing counterclockwise and κ < 0 clockwise circling (figure 1c,d). Second, we explored path segments in more detail by calculating the directional correlation delay [28] between pairs of birds in a 30 s time window. The directional correlation delay (DCD) method identifies similar path segments and quantifies the temporal relationship of the birds' changes in flight direction. Individuals circling in the same direction obtain high correlation values; these increase even more when birds circle at the same phase following a similar path (i.e. the same part of the circle, but at different altitudes, figure 2a). Here we observed that the time delay of 72% of all highly similar path segments (C ≥ Cmin = 0.9) is less than 3 s (table 1). Given that it takes approximately 18 s to complete a full circle [7], these delays indicate that coordinated individuals move in the same direction while being at the same part of the circle (but at different altitudes). Similarly, synchronization can also be detected by looking at the birds’ flight direction/orientation angle (ϕ = atan2(vy, vx)). Coordinated individuals had similar orientation angles indicating that both birds move in synchrony (i.e. at same ‘phase’; figure 2b). Uncoordinated birds (e.g. birds that circle either in opposite directions or not in synchrony in the same direction) had dissimilar orientation angles (figure 2c; see Material and methods for details).
Figure 2. Example of synchronized circling characterized by different measures. (a) Maximum correlation values (Cmax) between bird 26 (bird that switched first in figure 1c; marked as grey area) and all other birds that followed the directional switch. Cmax was calculated for each pair of birds using the directional correlation delay analysis with a 30 s time window. Bars on the right correspond to the average correlation value of the same period. High correlations indicate that birds circle in synchrony which applies to all birds marked on the right of figure 1c. As a counter example, we show bird 25 that is initially circling in synch with bird 26, but does not switch its circling direction. The correlation value drops accordingly. (b) Orientation angle (ϕ = atan2(vy, vx)) of the horizontal velocity against time of all birds synchronized with bird 26 (top line corresponds to bird 26). The orientation angle ϕ is presented on the y-axis and as a colour code. To be able to present many individuals, ϕ/(2π) is shown in the range of [−0.5 : 0.5]; the dashed lines correspond to periodic boundary transition that refers to π = −π (see inset on the right). Switches (around 105 s) from clockwise to counterclockwise circling are depicted as a change from decreasing to increasing values. Similarities in the slopes of the curves indicate that birds circle with similar angular velocity. (c) Orientation angle for birds not synchronized with bird 26 (for simplicity only five birds are shown). For example, bird 27 (top line of panel c) circled in the opposite direction to bird 26 (top line of panel b). Or, bird 25 circled in the beginning of the presented period in synchrony with bird 26, but did not switch its circling direction as bird 26 did.
| τ (s) | no. occurrences | ratio | cumulative ratio |
|---|---|---|---|
| 1 | 650 | 0.399 | 0.399 |
| 2 | 302 | 0.186 | 0.585 |
| 3 | 213 | 0.131 | 0.716 |
| 4 | 154 | 0.095 | 0.810 |
| 5 | 99 | 0.061 | 0.871 |
| 6 | 76 | 0.047 | 0.918 |
| 7 | 53 | 0.033 | 0.950 |
| 8 | 34 | 0.021 | 0.971 |
| 9 | 21 | 0.013 | 0.984 |
| 10 | 13 | 0.008 | 0.992 |
| 11 | 4 | 0.002 | 0.994 |
| 12 | 5 | 0.003 | 0.998 |
| 13 | 3 | 0.002 | 0.999 |
| 14 | 1 | 0.001 | 1.000 |
To detect the size of these synchronized subgroups, we examined only highly correlated segments (Cmin = 0.9) of each circling bout and determined for each individual the number of flock members that circled in synchrony (see Material and methods for details). The flock typically showed highly synchronized subgroups of 10.6 ± 5.9 (mean ± s.d., median = 9, IQR = 10; electronic supplementary material, figure S1). We also explored the proportion of birds that circled in the same direction in relation to the vertical distance between them, finding that storks circled significantly more often in the same direction when their height difference was smaller than 40 m (figure 3; e.g. observed ratio at Δz = 0 m: 0.704 ± 0.107, n = 51, p = 3.4 × 10−41, two-tailed one sample t-test) demonstrating that synchrony is associated with proximity within the flock. Such proximity may be associated with individuals exhibiting non-random associations, or ‘social ties’, to others. However, when we calculated the synchronization ratio, rsynch, defined as the time the pair flew in high synchronization relative to the total time it thermalled together (figure 4a), for each pair of storks, we found a low repeatability of rsynch between different thermals and migration days (figure 4b). This suggests that there are no long-term social links between individuals. Rather subgroups are shaped by ad hoc interactions between individuals that happen to be close to each other within the dynamic flock.
Figure 3. Ratio of individuals circling in the same direction as a function of height difference. For each GPS burst that contained at least 30 s of circling, we characterized for every pair whether they circle in the same or in opposite directions. Thin (coloured) lines correspond to the average ratio of birds circling in the same directions. Dashed line depicts a random ratio of 0.5. Thick black line shows the mean of all bursts. Figure 4. Repeatability of synchronization between individuals. (a) Synchronization ratio (rsynch) defined as the time spent moving on a correlated path (derived from DCD analysis) divided by the total time of circling together for each pair of birds. Black circles show values averaged over the first and second day. There is a weak correlation between the days (Pearson's r = 0.31). Coloured markers correspond to five randomly chosen pairs. (b) Synchronization ratio against time for the five randomly chosen pairs. The bottom plot with grey markers shows the mean of the entire flock. Error bars represent standard deviation.
Within these subgroups individuals performed SSCD (synchronized switches in circling direction; figure 1c,d). We identified those birds that successfully initiated a directional switch and were followed by at least one other birds by measuring, for each SSCD, the time delay between the directional switches of the different subgroup members. If a follower copies the directional switch of an initiator, time delays are positive (and are also typically much smaller than the time to complete a full circle in the thermal). To avoid including indirect synchronization, we used only the time delay between the initiator(s) and the average across the rest of the group for the statistical analysis (because with a changing order of A → B → C we cannot distinguish whether C reacted to A or B). We examined whether the altitude of the bird, with respect to others, influences its propensity to initiate a switch in circling direction (i.e. to see if direction changes propagate downwards or upwards), we determined the height differences between each initiator-follower pair. We found no correlation between the difference in height and the time delay between shifts (electronic supplementary material, figure S2) meaning that directional switches can be initiated from any position within the subgroup. Next, we asked whether there exist consistent inter-individual differences among group members in their propensity to initiate switches in circling direction. For each SSCD, we defined the initiator as the first individual to perform the directional switch (e.g. A initiates an SSCD of 3 in a scenario in which birds switch in the order A, B and C). Because of the temporal resolution of the GPS tracks (1 s sampling rate), we observed incidents when several individuals switched their circling direction at the same time point. In such cases, we consider multiple individuals as initiator, but weight their initiation by dividing it by the number of initiators (e.g. in a scenario in which A and B switched at the same time but before C, A and B initiated an SSCD of 1.5). We found a high correlation between the total number of initiations of SSCDs and the total number of times a bird adopts a SSCD as a follower (figure 5, Pearson's r = −0.627, n = 27, p < 0.001). The birds with most SSCD initiations form a strongly connected core cluster in the network of initiations (figure 5b).
Figure 5. Initiation and adoption of synchronized switches in circling direction. (a) The total number of followers that adopted the initiation against the total number of initiations. We defined the initiator as the individual that switched its circling direction first. In cases with multiple initiators, ni, we weighted the initiation with 1/ni. (b) Network visualization of individuals that synchronized their directional switch. Edges are non-directed; their widths represent the total number of times that a pair switched in synchrony (the sum of initiations and adoptions). Only values greater than or equal to 1 are shown. Colour coding of the nodes is identical to (a).
To understand the functional role of these directional switches, we had to explore whether initiators and followers shifted the centre of rotation at specific, non-random locations. Because climb rates are higher at higher altitudes (electronic supplementary material, figure S3), we cannot simply compare the climb rates before and after a directional change. Updrafts vary strongly between, and even within, thermals (electronic supplementary material, figure S4), but birds may still shift their rotation centre at the same locations to avoid losing the thermal (or even move into sinking air). Examining the phases of both birds, by looking at their angle of orientation, we found a strong correlation between the orientations of the birds at the time when the initiator shifts first (figure 6a; Pearson's correlation on periodic values: r = 0.794, n = 823, p = 10−179). By examining the orientation of the birds at the two time points when each performed the switch (i.e. correlating ϕ1(t) with ϕ2(t + τ)), we detected a stronger relationship, indicating that both birds switched at the same location, and in the same direction (Pearson's correlation on periodic values, r = 0.849, n = 823, p = 10−228; figure 6b). As a final test, we repeated the analysis after removing the effect of horizontal winds (using the method of Weinzierl et al. [7]), because the orientation angles of the birds may be biased by wind drift. After removing the effect of the wind from the trajectories, our findings become even stronger (Pearson's correlation on periodic values, r = 0.863, n = 740, p = 10−220; figure 6c) demonstrating that the resulting correlation is not an artefact caused by wind drift.
Figure 6. Bird orientation at the time of the synchronized switch in circling direction. Plots represent the orientation angle of the follower (y-axis) against the orientation angle of the initiator (x-axis) during a synchronized switch. Colour coding corresponds to the time delay, τ, between the switch. (a) Orientation angles of both birds at time t, i.e. when the initiator performed the switch. (b) Orientation of both birds at the time they performed the directional switch, i.e. initiator at time t and follower at time t + τ. (c) Orientation of both birds at the time they performed the directional switch after removing the effects of horizontal winds. We estimated the horizontal wind for each location, and subtracted the wind velocities from the GPS tracks of the birds (see Results, Material and methods and [7] for details).
We also found that our analysis of the motion of birds within thermals allows us to map the dynamic structure of the airflow within the thermals themselves. Each member of the flock explores a different thermal region with different vertical velocities. Although we need to assume the bank angle and corresponding sink rates of the circling birds to estimate the vertical speed of the air flowing inside the thermal, we can still map the quadratic profile of the entire thermal at different altitudes (figure 7; electronic supplementary material, video S1 and S2; see also [30]) by presenting the vertical speed of each stork in a relative coordinate system around the centre of the flock. Using this type of detailed model allows us to explore potential differences in flight performance among birds. Here we give an example by examining the birds' behaviour within that one mapped thermal to identify differences in their abilities to exploit the full potential of updrafts. For example, mapping the complex thermal structure enabled us to estimate its centre so as to compare how far birds circled from the region of strongest uplift. And although all birds had a similar circling radius (electronic supplementary material, figure S5), those farther away from the centre rose with smaller vertical velocities (r = −0.698, n = 25, p = 10−4; figure 8a). Also, birds higher up in the thermal had lower vertical speeds despite thermal strength increasing with altitude (r = −0.500, n = 25, p = 0.011; figure 8b). This finding confirms and complements our additional (forthcoming) study [31] in which we identify leading and following birds. Birds higher up in the thermal exhibited more irregular circling needing to explore the uplifting regions compared to followers who circled on a regular path. Here we show that followers can circle closer to the centre than those birds exploring the thermal further above (r = 0.674, n = 25, p = 2 × 10−4; figure 8c).
Figure 7. Collective sensing to map thermal structure and strength. (a) Bird trajectories colour coded by vertical speed in a 5 min circling bout. (b) The mean path of the flock defines at each altitude the centre of the thermal. Using relative coordinates Δx, Δy from the thermal centre in the horizontal plane the vertical speed produces the velocity profile of the thermal (see also electronic supplementary material, video S1–S3). Vectors correspond to the vertical speed of the birds rising in the moving air. To estimate the vertical speed of the airflow within the thermal, it is necessary to know the bank angle and corresponding sink rate of the circling birds. Figure 8. Individuals' performances in the mapped thermal structure. (a,b) Each circle represents the mean value of one individual. Plots show vertical speed against (a) distance from the estimated horizontal centre of the thermal at the altitude of the circling bird and (b) the bird's height relative to the centre of mass of the flock. We excluded two individuals too far away from the flock (39 and 68 m) from the analysis. (c) Individual's average distance from the horizontal centre of the thermal against the relative height compared to the centre of mass of the flock.
4. Discussion
Analysis of the circling behaviour of a flock of white storks during their migration revealed that there exist subgroups that move in synchronization and even coordinate when changing their circling direction. These subgroups exhibited dynamically changing membership with synchrony being strongest among birds that happen to be in close proximity. We found that shifts in circling direction could be initiated by any individual of the subgroup, irrespective of its vertical position within the groups. The adjustments in circling direction exhibited by the storks occurred at specific locations within the thermal allowing the storks to remain within the thermal. Thus, the spontaneous interactions between neighbouring individuals shape dynamic subgroups that may provide advantages while exploiting thermal structures.
We observed that any bird within the subgroups can initiate or follow directional switches, irrespective of its spatial position. It may be possible that leading birds higher in the thermal slowly shift their trajectories during multiple circles, without a directional switch, thereby providing information to the other flock members stimulating them to adjust their rotation centre. Or, alternatively, leaders may decide themselves to circle in another direction while relying on information from the surrounding air. Individuals that are part of a larger synchronized subgroup have a higher number of potential followers, increasing the probability that others will adopt a switch in circling direction.
High variability in the size and composition of the synchronized group may also influence collective sensing [23]. Synchronized and unsynchronized individuals map different regions of the thermal, thus potentially enabling individuals across the group to achieve a better exploration of the complex structure. Flight adjustments while mapping the thermal may not necessarily involve fast, complete switches in circling direction but also slow shifts of the rotation centre taking place over several (irregular) loops without a directional switch. Here we show that birds are responding to each other over very fine time scales (typically < 3 s). Thus, birds could either be responding to the sight of another bird turning (i.e. the cue), or by seeing a bird that turned earlier and already seems to have achieved a higher climb rate. It is also possible that birds adjust their flight path to the shape of the thermal independently of each other, by directly reacting to the thermal [32]. Each bird can sample the uplift during the full circle separately and, when a shift is necessary, they can decide to perform the change at the same location. Individual variation in how fast a bird perceives or responds to the external factors may potentially lead to time delays between initiators and followers. In such cases, the synchronization could be an indirect result of all birds reacting to the same environmental cues.
We also demonstrate that using the motion of the birds within the flock it is possible to estimate the dynamic variation of airflow within the thermals, and thus to estimate the specific environmental conditions in which each individual is flying. This might lead to the possibility to examine the storks’ movements relative to their surrounding environment (electronic supplementary material, video S3), and separate between effects coming from (i) the air of the thermal (e.g. the air drifting and rising because of wind) and (ii) the relative adjustments of the birds in this moving air. Our results highlight that integrating within-group social interactions into migration research will be critical for achieving a complete understanding of the movement decisions of social migrants, and, additionally, provide us with valuable, previously unattainable knowledge on their environment.
Data accessibility
The data used in this study are available in the Movebank Data Repository with doi:10.5441/001/1.bj96m274.
Authors' contributions
A.F. and M.W. conceived the idea and designed the project; A.F., W.F. and M.W. conducted fieldwork and collected the data; M.N. designed the data analyses and visualizations, and analysed the data; A.F., M.N. and I.D.C. wrote the manuscript.
Competing interests
The authors declare that they have no competing interests.
Funding
We acknowledge funding from the Max Planck Institute for Ornithology. A.F. was supported by the German Aerospace Center (DLR) and the Christiane Nüsslein-Volhard Stiftung; M.N. by the Royal Society Newton Alumni scheme; I.D.C. by National Science Foundation (PHY-0848755, IOS-1355061, EAGER-IOS-1251585), ONR (N00014-09-1-1074, N00014-14-1-0635), ARO (W911NG-11-1-0385, W911NF-14-1-0431) and Human Frontier Science Programme (RGP0065/2012).
Acknowledgements
We thank all the people who helped during fieldwork, especially Wolfgang Schäfle, Riek van Noordwijk, Bart Kranstauber, Daniel Piechowski, Babette Eid and Yvonne Flack. We thank Wolfgang Heidrich and Franz Kümmeth (e-obs, Munich, Germany) for their suggestions on logger programming, and Sarah Davidson for setting up the movebank.org data repository. A.F. thanks Andrew Berdahl, Colin Torney and the Santa Fe Institute for the organization of and invitation to the NSF- and SFI-funded workshop ‘Collective Animal Motion in the Wild’ that resulted in this theme issue for, Philosophical Transactions of the Royal Society B.
