The use of social information in vulture flight decisions

Abstract

Animals rely on a balance of personal and social information to decide when and where to move next in order to access a desired resource. The benefits from cueing on conspecifics to reduce uncertainty about resource availability can be rapidly overcome by the risks of within-group competition, often exacerbated toward low-ranked individuals. Being obligate soarers, relying on thermal updraughts to search for carcasses around which competition can be fierce, vultures represent ideal models to investigate the balance between personal and social information during foraging movements. Linking dominance hierarchy, social affinities and meteorological conditions to movement decisions of eight captive vultures, Gyps spp., released for free flights in natural soaring conditions, we found that they relied on social information (i.e. other vultures using/having used the thermals) to find the next thermal updraught, especially in unfavourable flight conditions. Low-ranked individuals were more likely to disregard social cues when deciding where to go next, possibly to minimize the competitive risk of social aggregation. These results exemplify the architecture of decision-making during flight in social birds. It suggests that the environmental context, the context of risk and the social system as a whole calibrate the balance between personal and social information use.

1. Introduction

Animals must constantly decide where and when to move next in order to find resources such as food, water, shelter or a mate, necessary for life. To make these decisions, they can rely on two sources of information: personal information and social information. Personal information includes knowledge of the spatio-temporal patterns of resource distribution that individuals may perceive or have memorized from previous encounters [1]. For example, food-storing birds are able to return to locations where they stored or saw food in the past, based on prior expectation of the resource availability [2]. Social information, on the other hand, is obtained by observing the behaviour of others [3–5]. Feeding, fleeing or mating individuals provide discrete information about the availability and locations of food, predators or potential mates.

For resources that are heterogeneously distributed in the environment, ephemeral and unpredictable, using only personal information for movement decisions may be prone to inaccuracies, either because the knowledge is erroneous per se and/or because previous experience has been too limited [6]. In such conditions, social animals may benefit from companions' knowledge and may follow the dominant or oldest individual(s) considered as knowledgeable (e.g. homing pigeons, Columba livia, or elephants, Loxodonta africana, [7,8]), follow the largest group through shared decision-making [9], or stay with preferred affiliates [10–12]. Because using social information can considerably reduce uncertainty in finding resources, individuals should favour this source of information to achieve cost-efficient movement [13–15]. However, relying heavily on social information can also lead individuals to aggregate on resources. While this aggregation could potentially facilitate early detection of predators [16] and access to resources (e.g. in seabirds preying on fish schools [17]), it could also induce competition by exploitation or interference if the resource is monopolizable and depletable [18]. Since both social and personal information are often available to social animals and may differ in their quality and quantity [1], they need to balance their relative importance, depending on the availability and predictability of the resource. When deciding on the next movement step, social animals must trade off the decreased uncertainty of locating a resource through social information, with the potential increase in competition risk. Such a balance may be dictated by the immediate needs of the individual and its risk sensitivity [19] but also by the group social organization. For example, low-ranked individuals are known to suffer more from within-group competition compared with high-ranked individuals [20]. To minimize future interactions for food with their conspecifics (and certainly high-ranked individuals), low-ranked individuals should forage on their own and be reluctant to use social information which may trigger aggregation [21].

Vultures rely on two unpredictable resources: carcasses to feed and thermal updraughts (i.e. masses of hot air rising from heated surfaces) to move. During foraging flights, these large soaring birds gain altitude by circling into thermal updraughts and glide across the landscape to the next updraught while scanning the ground for carcasses [22]. Although some topographic features are clearly favourable to updraughts presence [23], at the individual level, challenging local meteorological conditions (e.g. high wind speed, low temperature, high cloudiness) can make thermal locations and availability hard to predict [24]. If they fail to detect an updraught, vultures may be forced to switch to flapping flight, or worse to land and take off again, significantly increasing their energy expenditure [25,26]. While both thermals and carcasses are relatively unpredictable, the use of thermals by individuals does not impact their availability to other birds contrary to carcasses, which are a depletable resource. When a vulture discovers a carcass, its sharp drop in altitude while circling before landing is used as a signal by conspecifics, dragging tens of individuals to the food source in a few minutes [27,28]. As the number of vultures around the carcass increases (up to 100–120 individuals, [29,30]), individual feeding rates decrease due to reduced access to the resource, resource depletion by competitors, and increased agonistic interactions [29]. Therefore, in these social birds, individuals should have to balance the advantage of conspecific presence to locate thermal updraughts (as demonstrated in [31]) with the ultimate cost of competition around the carcasses that can be fierce [32–35]. As such, vultures are ideal models to investigate the role of conspecifics in shaping their foraging movement decisions.

While the role of conspecifics in attracting individuals once a carcass is found has been well documented in vultures [36,37], the role they may play on individuals’ movement during carcass search is far less known in this soaring bird. Using a group of captive but freely flying ‘griffon’ vultures, Gyps fulvus and Gyps rueppellii, tagged with high-resolution GPS loggers, we studied how conspecifics' presence shapes individuals’ movement decisions during foraging flights. Despite being trained birds released for public shows, these individuals sometimes detected and fed on carcasses at surrounding farms (five times in the 24 days of data collection). We therefore consider these flights comparable to natural flights [38] whose patterns are mostly driven by food search given the opportunistic behaviour of these birds. Focusing on the movement steps from thermal to thermal, we first assessed when do individuals preferentially discover new thermals (i.e. use of personal information) compared with using thermals already discovered by conspecifics (i.e. use of social information). We expected that vultures would favour the use of social information when unfavourable meteorological conditions increased thermal unpredictability and when flight conditions (e.g. low altitude) increased risks of landing [1]. Furthermore, given the hierarchy in vulture groups, we expected low-ranked individuals to be more prone to use personal information than high-ranked individuals to try to find the food source first, in order to avoid large aggregation [39,40]. As a result, low-ranked individuals would explore the environment and high-ranked individuals would prioritize following low-ranked ones. Second, we investigated the drivers underlying thermal selection when individuals had to choose between simultaneously available thermals. We expected individuals to select thermals providing the maximal positive vertical speed (i.e. climb rate) as it may provide a reliable proxy of the thermal current strength helping them maximize their height gain [31]. To decrease uncertainty about resource finding and risks mentioned above, we expect that individuals should favour thermals hosting the maximum number of individuals to maintain cohesion and secure the possibility to cue on as many conspecifics as possible [31]. Finally, social preferences may also influence decision, with individuals preferentially moving together with preferred affiliates [12,41,42], as it could reduce competition due to familiarity between individuals [43].

2. Material and methods

(a) . Study site, vultures housing conditions and experimental settings

The study was carried out in 2021 and 2022 at the Rocher des Aigles falconry centre, Rocamadour, France and divided between winter and summer periods each year. During winters, vultures were housed within an aviary (6.7 × 6 × 6 m) equipped with four perches: three of them measuring 3.10 m, and one the full width of the aviary (electronic supplementary material, figure S1). This setting was used to estimate vulture social bonds (see Social bond estimation). In addition, besides being fed daily on small pieces of meat to prevent conflicts, five feeding events (one each week during a five-week period) were organized in the aviary on a butchery carcass occurring after a 1-day fasting (to motivate feeding). These feeding events were used to assess dominance hierarchy within the group (see Hierarchy estimation). In summer, these trained vultures were kept perching on individual logs, released several times per day to execute free flight shows for the public within a landscape composed of plateaus interspaced by canyons, similar to ‘Causses’ landscape typically used by French wild vultures [44]. The falconry centre is located near a 120 m deep canyon and offers natural soaring conditions for raptors, making this study site a great place to investigate natural group flight behaviour (see Group flights) [26].

We used GPS data and visual observations to characterize the social and flight behaviour of eight captive vultures (seven Eurasian griffon vultures, Gyps fulvus, and one closely related Rüppell's vulture, Gyps rueppellii), including five females and three males (electronic supplementary material, table S1). Each year, we conducted experiments on a group of six individuals (two griffon vultures were replaced in 2022; electronic supplementary material, table S1). Experiments followed the animal ethical guidelines of France and the Centre National de la Recherche Scientifique. Handling of birds to fit GPS loggers followed the protocol of telemetry study of vultures authorized in the Programme Personnel 961, coordinated by O.D., under the supervision of the French ringing centre, CRBPO, Paris. Furthermore, experiments, observations, handling and flight events were systematically performed under the guidance of the head of animal caretakers, B.N.

(i) . Social bond estimation

During five weeks in both years (December/January 2020–2021 and November/December 2021), we recorded pictures of vultures in the aviary from 8.00 to 19.00 (local time) at 5 min intervals, using three camera traps (Wosport Big Eye D3 and Reconyx HyperFire HC600).

We identified birds using repeated colours on plastic rings and marks on the ruff and back-head feathers, using harmless colour sticks (Raidex GmbH, figure 1a). We removed pictures on which we observed agonistic interactions, and then processed the remaining ones to extract the individuals' ID and location based on their bill/head location to estimate inter-individual distances with a purpose-built image annotation program in Julia software, JuliaHub Inc. [45]. For subsequent analyses, we relied on R software (v. 4.2.2, R software, 2022 [46]).

Figure 1.

Data collection. (a) Perched vultures. Distance between vultures during perching events was used to estimate social-bond strength. (b) Feeding event around a butchery carcass. Agonistic interactions during those feeding events were used to estimate dominance hierarchy. (c) Flying vulture. Vultures were released for free flight into a 120 m canyon, equipped with high-resolution GPS loggers.

We considered the social bond between a dyad of individuals i and j based on spatial proximity following the simple ratio association index (SRI, equation (2.1), [47,48])

$${\mathrm{SRI}}_{i,j}={\mathrm{SRI}}_{\hspace{0.17em}j,i}={\displaystyle \frac{n_{\mathrm{tog}}}{n_{\mathrm{tot}}},}$$ 2.1

where n_tog is the number of pictures in which individuals i and j were on the same perch at a Euclidean distance of less than 1.55 m and n_tot is the total number of pictures in which individuals i and j were both detected on the same perch. SRI values varied between 0 and 1, where 0 represented dyads that were never seen associated and 1 represented dyads that were always observed sitting at less than 1.55 m from each other. The distance of 1.55 m was chosen as matching to the mode of the inter-individual distances distribution (electronic supplementary material, figure S2). This was also consistent with the aviary setting, as it corresponded to half the length of most available perches, and was biologically relevant (as it matches with the maximum distance (a step forward, body tilted toward the opponent and neck extended) at which an individual can attack and bite another one). Our analyses were robust to choices of a lower distance threshold (see electronic supplementary material, ESM01).

(ii) . Hierarchy estimation

Each winter, we estimated hierarchy within the vulture group by monitoring feeding interactions during the five carcass-based feeding events in the aviary (10 in total, figure 1b) using a remotely controlled video camera (GoPro Hero 4, GoPro Inc.) fixed at 2 m height on the aviary wall. These feeding events lasted on average 34 min (s.d. ± 4 min).

We computed individuals’ rank relying on the randomized Elo-rating approach [49,50], which accounts for potential temporal instability of the rank using permutations in the agonistic interaction series (‘elo_scores’ function, aniDom package, [50,51]; using 1000 randomizations and fixing the rank adjustment speed along the series, K-factor, to 200). The interaction series consisted of identifying the ‘wins’ and ‘losses’ during agonistic interactions [52] with other individuals recorded ad libitum from video footage of feeding events (annotated with BORIS video analysis software, [53]). We used the ethograms from Bosé & Sarrazin [32] and Valverde [54] to characterize griffon vulture feeding behaviour and between-individual interactions. An individual won the interaction when it interrupted another individual's feeding bout (by pecking it, displacing it or engaging in a fight), and finally accessed the carcass before its opponent. In other cases, the interaction was considered as a ‘loss’ for the initiator. We assessed the reliability of the dominance hierarchy through individual Elo-rating repeatability (‘estimate_uncertainty_by_repeatability’ function, aniDom package, [50]).

(iii) . Group flights

We recorded vulture flights decisions during 42 flight sessions (21 sessions each year, electronic supplementary material, table S2) in the vicinity of the Rocher des Aigles. In general, birds were released for a flight session three times per day (in rare occasions from two to four times), at around 11.00, 14.30 and 16.00 (local time) for a mean duration of 26.03 min (s.d. ± 14.15 min) of flight. Once released, the birds were not forced to fly and no food was placed in the landscape to attract them. They only received a meagre reward (ca 60 g) at the end of the show (as they are fed at the end of the day; see electronic supplementary material, ESM02 for details on captive vulture daily life). These captive vultures were trained to fly freely, searching for thermals, gaining altitude and coming back to their trainers (electronic supplementary material, video S1 [55]). After about 20 to 25 min of free flight, the birds were signalled by their caretakers to return to the Rocher des Aigles. It could take several minutes for them all to return, with a longer delay if the flying conditions were particularly good and when birds were more motivated to remain in flight. Vultures were equipped with a high-resolution GPS logger (4 Hz, TechnoSmart, models Gipsy 1, Gipsy 5 or Axytreck) positioned at their lower back using a Teflon leg-loop harness (figure 1c, [56]). They were released in two groups of three individuals. The second group was released 2 min after the first group. The groups were built according to social preferences, with the three most socially bonded birds together, and the composition of these groups remained stable across all flight sessions of the same year. Release order alternated between consecutive days. For each flight session, we recorded and considered as stable the cloudiness (i.e. the proportion of clouds covering the sky, on a scale from 0—no clouds—to 8—sky fully covered by clouds), horizontal wind speed (four categories estimated locally from the Beaufort scale) and temperature (extracted from meteofrance.com).

To further investigate how vulture thermal choices were shaped by personal and social information, we pre-processed flight tracks in three consecutive steps. We subsampled individuals' tracks from 4 to 1 GPS fix per second by taking the first record, and segmented their flight behaviour into gliding, linear soaring and circular soaring. We then created spatio-temporally dynamic maps of thermal availability based on the spatial clustering of individuals' circular soaring phases. Leaning on these maps, we retraced the history of thermal use/choice by individuals.

Thermal use identification. To segment vulture flight between circular soaring, linear soaring and gliding flight, we first calculated turning angle and vertical speed between consecutive locations using the move R package [57]. We applied a moving window of 30 s to calculate the absolute cumulative sum of the turning angles (hereafter cumulative turning angle) and a moving window of 5 s to calculate the average vertical speed. We then applied a k-means approach (k = 2, ‘kmeans’ function, stats R package) on the smoothed vertical speed (positive speed when flying upwards, negative when flying downwards) to distinguish between soaring (ascending flight) and gliding (descending flight, [58,59]). We further classified soaring locations into circular soaring (indicating use of thermal updraughts) and linear soaring (also called slope soaring, expected to occur outside of thermals), with circular soaring being associated with a cumulative turning angle greater than or equal to 300°. A result of segmentation is illustrated in figure 2b. Finally, we inferred the use of a thermal when the individual engaged in circular soaring for more than 30 s, with no interruption of more than 5 s of gliding (electronic supplementary material, table S1).

Figure 2.

Flight data pre-processing. Pre-processing steps of group flight GPS data, example of one flight session. The altitude ranges from 200 to 600 m. (a) shows a group flight (see electronic supplementary material, video S1), with colours corresponding to each individual. (b) illustrates the segmentation of an individual's flight (blue individual in (a)), with the orange segments corresponding to circular soaring phases. (c) illustrates the three-dimensional density-based spatial clustering of individuals' circular soaring phases, with colours indicating the three thermals identified in this flight session.

Dynamic mapping of available thermals. Within each flight session, we created a dynamic map of thermals (figure 3). First, we spatially clustered vulture circular soaring locations (reflecting the use of the same thermal updraught) independently of time by using a three-dimensional density-based spatial clustering approach (‘dbscan’ function, dbscan R package, [60]). This algorithm relies on a spherical neighbourhood to perform density-based neighbour joining, i.e. clustering (figure 2c). We assumed this neighbourhood to be of a 40 m radius, and a minimum number of five locations within this range for the algorithm to consider the neighbourhood further. This 40 m threshold corresponded to the largest 4-nearest-neighbour distance observed when considering locations attributed to thermal use only (‘kNNdistplot’ function, dbscan R package) and matched with empirical expectations of radius during circular soaring phases [61].

Figure 3.

Illustration of the step selection framework used to investigate thermal selection. We focused on the movement of vultures released from the Rocher des Aigles when flying from thermal to thermal (i.e. a step). To do this, we mapped each thermal used during a flight session based on movement segmentation and clustering (see Material and methods section) to create dynamic maps of thermal availability over the flight session (as represented by the aerial views). The illustrated example focuses on the decision of a vulture (V₁; step 1) when leaving the thermal (T_A) and having to choose between two available thermals (T_B, close but not currently used by another vulture, and T_C, further away but currently used by another individual). T_D was not available until step 2, when it was discovered and used by another individual, and is therefore shown in grey at step 1. At step 2, V₁ joined V₃ in T_C and both thermals T_A and T_B were no longer available. A thermal was available from the moment when the first individual entered it until the last individual left it. Therefore, the number of available thermals could change during the flight session (see differences between maps in step 1 and 2).

We then made those maps dynamic in time by considering the lifetime of each thermal. We considered a thermal as ‘available’ from the moment when the first individual entered it until the last individual left it (figure 3). Note that using vulture tracks for this mapping may induce limits such as the underestimation of the real lifetime of a thermal but also the lack of detection of thermals never used by vultures.

(b) . Statistical analyses

We defined collective flight events as any time of a flight session when at least two individuals were flying. For each of these events, we first analysed the use of social information (the tendency to join thermals already discovered by conspecifics) as a function of external (meteorological) and internal drivers (individual traits). We then used step selection functions to define, at each movement step, which drivers determined the selection of the chosen next location (thermal updraught) relative to other potential locations.

(i) . Drivers of social information use

We investigated the effect of local meteorological context, individual traits and flight mechanics on the use of social information, defined here as the tendency to join thermals already discovered by conspecifics. We considered that an individual discovered a thermal when it was the first, among all individuals, to adopt circular soaring flight into it. For the analysis, we discarded the discovery of the first thermal in each flight session (as this thermal was necessarily discovered).

To investigate the drivers underlying the use of social information we modelled the probability to join a thermal already discovered by others (binary response: 0 if the selected thermal has not been previously discovered by a conspecific, 1 otherwise) using generalized linear mixed models (GLMMs) with binomial error structure and a logit link function [62]. Our full model contained the following 10 fixed effects: meteorological variables with the (i) wind speed (categorical predictor, four categories with null (Beaufort 0–1), low (Beaufort 2), medium (Beaufort 3) and high (Beaufort greater than or equal to 4)), (ii) cloudiness and (iii) temperature (both continuous predictors); social variables with (iv) the age (continuous predictor) and (v) rank in the dominance hierarchy of the individual (ordinal categorical predictor, with the dominant individual ranked first); and variables related to the mechanics of flight with (vi) the glide-ratio (horizontal distance travelled during a 1 m altitude loss, only measured on glides with straightness greater than 0.95 in each flight), (vii) the altitude of and (viii) the three-dimensional distance to the exit location from the previous thermal used (all continuous predictors). We also added (ix) the group in which individuals have been released (first or second group released for the flight) and (x) the time elapsed since the first individual take-off (continuous predictor) as control variables. Individual ID was considered as a random factor. We predict that the use of social information should increase with thermals unpredictability (i.e. higher wind speed, lower temperature), increase with inexperience and competitive abilities (i.e. younger and higher-ranked individuals), and increase with flight challenge and landing risk (i.e. lower glide ratio, lower previous thermal exit altitude and larger distance from previous thermal).

To compare the relative importance of the fixed effects we scaled all non-categorical variables to use their estimate as dimensionless effect size [63]. We examined the significance of each variable by comparing the goodness of fit of models with and without the variable of interest using a likelihood ratio test (‘drop1’ function, stats R package). Assumptions required for these statistical approaches (homoscedasticity, Gaussian distribution of residuals) were checked with plot diagnosis (histogram of residuals, residual Q-Q plot, distribution of residuals versus fitted values, DHARMa R package, [64]). We also tested for the presence of outliers, and calculated the variance inflation factor (VIF) to test for collinearity (VIF values greater than or equal to 3 suggesting a strong collinearity [65]). We did not detect collinearity in our predictors (VIF_max = 1.74) (electronic supplementary material, figure S3). Furthermore, we extracted the marginal coefficient of determination $(R_m^2)$ and the conditional coefficient of determination $(R_c^2)$, which describe, respectively, the proportion of variance explained by fixed effects and by the fixed and random effects combined [66]. Finally, as the flight time period and the tested individuals differed, we cross-compared models fitting on the two years separately (see electronic supplementary material, ESM01).

(ii) . Drivers of thermal updraught selection

To study the drivers underlying thermal selection, we embedded our work in the Step Selection framework [67] in which we investigated the determinants of vulture movement decisions to fly from a thermal to another specific one among all simultaneously available at this time (i.e. during a ‘step’). In practice, we considered the series of thermals used by each individual. In that series, we focused on movement steps involving a flight to a thermal previously (or currently) used by a conspecific when other thermals were available. Using a conditional logistic regression, we compared the ‘chosen’ thermal characteristics with those ‘available’ but not chosen. The conditional logistic regression included seven predictors, respectively characterizing the thermal profitability with (i) the distance to it and (ii) maximum vertical speed reached in the thermals by any individual since the focal individual has been released in the flight session (continuous predictors), individual personal experience considering whether (iii) the thermal was previously used by the focal individual (binary predictor), and social information with (iv) the presence of the focal individual's preferred affiliates in the thermal or not (binary predictor), (v) the number of individuals present in the thermal, (vi) the weighted mean (by the number of previous visits to the thermal) of the social bond with individuals that used the thermals, and (vii) the negative cubed difference of ranks between the focal individual and those in the thermals (all continuous predictors, set to 0 for the two latter if no individuals used it/were present). We used the negative cubed difference to consider an attraction–repulsion effect. In cases where the difference in rank is large, high-ranked individuals should be attracted to conspecifics (i.e. higher probability to join the thermal in which the difference of ranks is large and positive), while low-ranked individuals should be repulsed (i.e. lower probability to join the thermal in which the difference of ranks is large and negative). When the difference of rank is weak, this should have a close to null effect on the probability of selecting a thermal. To model this effect, and because the dominant individual is rank 1, the negative cubed difference of ranks was used. For example, following the curve of the negative cube function, if the difference of rank was five (e.g. the focal individual is ranked sixth—a low rank—a conspecific in another thermal is ranked first—a high rank) the probability that the focal individual joined the conspecific should be drastically decreased, mimicking a repulsion effect. We predict that individuals should select the thermal with the highest profitability (i.e. closest in distance and the one with largest positive vertical speed), the most familiar (i.e. if previously used), with the most valuable social information (i.e. hosting the most and preferred affiliates), and minimizing competition risk (i.e. when the cube rank difference is the largest).

Also for this model, we scaled all non-categorical variables to better compare their relative importance. We fitted the conditional regression considering all individuals together, yet considering data stratified at the individual-step level. We finally reported the relative selection strength (RSS) of significant variables which provides the magnitude of estimated selection coefficients, holding all other covariates fixed at their mean value [68,69].

3. Results

Vulture dominance hierarchy was steep (electronic supplementary material, figure S4) and reliably inferred (individual Elo-rating repeatability = 0.82 and 0.83 in 2021 and 2022, respectively). The rank orders among individuals present in both years were relatively consistent and uncorrelated to sex or age (Wilcoxon test: w = 6, p = 0.70 and w = 3, p = 0.80, Pearson's correlation coefficient [95% confidence interval]: ρ = −0.19 [−0.87, 0.73], p = 0.71 and ρ = 0.72 [−0.21, 0.97], p = 0.11, respectively, for sex and age in both years; electronic supplementary material, table S1). During the 21 flight sessions performed each year, we identified a total of 520 and 578 thermalling events in 2021 and 2022 respectively. On average, 63% (s.d. ± 7%, electronic supplementary material, table S1) of these circular soaring behaviours took place in thermals discovered by a conspecific.

(a) . Flight risks and hierarchy shape the use of social information

Our model was significantly better than the null model (considering only control effects; ${\chi }_{10}^2=195.3,$ p < 0.001, AIC = 1237.4 and 1412.7 respectively) and explained 30% of the variance (electronic supplementary material, table S3). The probability for an individual to use a thermal previously discovered by a conspecific decreased with temperature (from 0.74 at 17°C to 0.43 at 31°C; figures 4a, 5a, electronic supplementary material, table S3), but tended to increase with cloudiness and wind speed (figure 4a, electronic supplementary material, table S3). This probability dropped also with the distance from the previous thermal and the altitude at which the bird left it (from 0.63 when being at a distance of 12 m from the last thermal used to 0.16 at a distance of 6776 m and from 0.76 when exiting the last thermal at an altitude of 195 m to 0.039 at 1574 m of altitude; figures 4a, 5b,c, electronic supplementary material, table S3). Individuals lower in the dominance hierarchy were approximately twice as likely to discover new thermals than high-ranked individuals (figures 4a, 5d). We did not detect significant effects of age and glide-ratio on the probability to use thermal previously discovered by conspecifics (figure 4a, electronic supplementary material, table S3). Fitting the same model structure on 2021 and 2022 data separately yielded the same overall results, suggesting that the observed pattern was robust to changes in hierarchy and between-year conditions (electronic supplementary material, figure S5, and table S3).

Figure 4.

Estimates of models investigating the drivers of social information use (a) and thermal selection (b). Rows correspond to each predictor, with positive values indicating increased probability to discover a thermal (a) or to move to the thermal (b). For their biological meaning in (a) and (b), see Material and methods sections 2(b)(i) and 2(b)(ii), respectively. Each point represents the standardized estimate value. Segments give the associated 95% confidence intervals.

Figure 5.

The probability to use a thermal already discovered by conspecifics decreases with temperature, distance to the thermal, flight altitude and hierarchy rank. This is the visual representation of significant variable effects presented in figure 4a. Points represent the probability of using a thermal already discovered by a conspecific, estimated on the raw data. Their size is relative to the number of thermals used. Because vultures were encouraged to leave the Rocher des Aigles to perform free flights, but not forced to, and can fly for different lengths of time, some differences in the number of thermals used by individuals can appear (e.g. small sample size for individuals of rank 4 in panel d). To estimate the probability in (a), (b) and (c), predictors were binned in six bins. Black lines with grey shades show the GLMM estimated probability with its 95% confidence interval (N = 1098 thermals).

(b) . Vultures select thermal updraughts hosting the most conspecifics

We identified 178 movement steps where an individual entered a thermal while at least one other thermal was available simultaneously. In these movement steps, individuals were approximately 28 times more likely (RSS [95% confidence interval] = 27.94 [5.99, 131.63]; figure 4b, electronic supplementary material, table S4) to select a thermal hosting the largest number of conspecifics compared with a thermal hosting only one individual. On the contrary, the probability to choose a thermal tended to decrease when the preferred affiliate was using it. The distance to the previous thermal, the maximal vertical speed reached in the thermal, and whether individuals used this thermal in the past did not significantly affect thermal selection (figure 4b, electronic supplementary material, table S4). At time of decision (i.e. when individuals decided to move from one thermal to another), the difference in dominance ranks as well as the presence of its preferred affiliate did not drive the individual's probability of selecting the thermal. This pattern was consistent when considering only movement steps where individuals had to choose between thermals currently used by other vultures at time of decision (N = 61, electronic supplementary material, figure S6 and table S4). Furthermore, considering all decision events, the sensitivity analysis on the inter-individual distance threshold for the social bond strength estimation yielded the same results (i.e. 1.55, 1.30 and 1 m; see electronic supplementary material, ESM01, figure S7 and table S5).

4. Discussion

Using a combination of high-resolution tracking and social structure monitoring, we identified contextual drivers for the differential weighting of personal and social information in movement decisions. We showed that vultures' movement decisions predominantly relied on social information, especially in unfavourable flight conditions that increased thermal unpredictability or put individuals at risk of undesired landing. Overall, individuals preferentially joined thermals with the largest number of conspecifics. However, the use of social information depended on the individual social status: low-ranking individuals were more inclined to use personal information and discovered more thermals on their own than high-ranking individuals.

We found that low-ranked individuals, probably the ones suffering the most from interference competition, had higher probabilities of discovering new thermals, thus probably exploring their environment more intensively than the high-ranked individuals. Such flight strategy would enable subdominant individuals to reach carcasses first, or at least to arrive at the beginning of the feeding event when the rate of interference is lower [29] hence avoiding lost opportunities due to conformity with conspecific behaviour [70]. From this may emerge a producer–scrounger dynamic [71,72] wherein the use of personal information from low-ranked individuals to arrive at food sources with lower competition levels would be exploited by dominant individuals to reduce their own searching effort [18,72,73]. This is coherent with previous observations of low-ranked vultures being ‘pioneers’: the very first individuals to land and feed on the carcasses before being displaced by high-ranked individuals arriving afterwards [29]. This influence of dominance on foraging tactics where low-ranked individuals explore and find food while dominant profit has also been observed in other social bird species such as common cranes, Grus grus, oystercatcher, Haematopus ostralegus, house sparrows, Passer domesticus, and barnacle goose, Branta leucopsis [20,73–75]. Eviction of subordinates from food patches has even recently been identified as a trigger for collective movements in vulturine guineafowl, Acryllium vulturinum [76]. By contrast, in activities where individuals do not experience competition, such as tool-use learning in chimpanzees, naive individuals will generally copy dominant (and knowledgeable) individuals [77]. Because individuals likely to suffer a cost (low-ranked) were reluctant to follow other individuals, while the reverse was not true, our study hence stands as a clear-cut illustration of the ‘copy when asocial learning is costly’ rule [78]. The vulture position in the dominance hierarchy, through the costs it imposes on access to food, seems to calibrate the balance between the use of personal and social information in foraging movements. In some cases, however, trading personal information in favour of social information is inevitable.

When the environment is largely unpredictable or whenever using error-prone personal knowledge can be energetically costly, individuals should tend to eavesdrop and rely more on information provided by conspecifics to reduce uncertainty about resources availability [15,79]. Here, we evidenced both cases. First, vultures prioritized the use of social information when the temperature was low and tended to do so also when cloudiness and wind speed increased (electronic supplementary material, table S3). These weather conditions may translate into fewer and weaker thermals, drifting into the wind, making them less predictable [80–84]. Second, they also favoured social information when the altitude at which they left their previous thermal was low. When exiting a thermal at low altitude, individuals have limited time to glide to the next thermal before having to shift to flapping flight to stay aloft, or else landing in an undesired place, which both would add high energetic cost associated with flapping and take-off [25,26,85,86]. Reaching high altitudes quickly to avoid this risk may also explain why vultures used more thermals previously discovered by conspecifics if those were close to the last thermal they used. While vultures are able to cope with difficult flight conditions (e.g. turbulence and strong wind) by adjusting their banking angles [61], anticipating such risky events may remain the most efficient way to maximize the trade-off between time, energy and risk, which largely dictates their flight strategy [38].

Adult individuals, through experience, are generally better at coping with difficult flight conditions [87], yet we did not find evidence of an effect of age relative to the use of social information, as observed in other group living species (e.g. [88]). More than age per se, the familiarity of individuals with a given situation might shape their tendency to rely or not on social knowledge (e.g. in spider monkeys, Ateles geoffroyi, during collective foraging [89]). The captive individuals tested in this experiment are all adults and fly in the same landscape every day since their birth, thus they are probably very familiar with the areas favourable to thermal emergence. This could explain why we did not detect any effect of age on the use of social information, but also indicates that the relative importance of this source of information is probably underestimated due to the birds' familiarity with the surroundings.

When faced with a choice between simultaneously available thermals, the previous experience of individuals (i.e. whether the thermal was used previously or not by the focal) or current expertise of the group (i.e. relative age/hierarchy difference) impacted very little vulture movement decisions compared with other social cues (i.e. number of conspecifics present in the thermals, and affiliation status). This result contrasts with previous findings from insects to mammals, including birds [90–94]. In the current system, ascending currents can be very ephemeral phenomena, sometimes only lasting a few minutes [95,96]. Certainly, a ‘live report’ is therefore better provided by the accumulation of convergent information sources (i.e. numerous conspecifics [97]) rather than relying on a unique individual source (i.e. the individual itself or one reference individual). In that line, and surprisingly, the presence of one preferred affiliate in a thermal tended to reduce the probability to join it. There is evidence that social bonds assessed ‘on the ground’ are often unrelated to association in flight [98]. It therefore questions whether collective flights might be used by vultures to strengthen initially weak social bonds. Maintaining association in flight can indeed be important, as evidenced in the migratory behaviour of other soaring bird species to enable accurate collective mapping of the distribution of uplifts [99,100]. Furthermore, for soaring birds, the presence of conspecifics should provide not only information on the location and strength of updraughts [22,100] but could also indicate flight speed and circling radius needed to optimize climb rate, by remaining close to the centre of the thermal where uplift is highest [61]. Yet, the maximum speed reached by individuals using the thermal little affected vulture choices. Possibly, climb rate or individual speed are not as easy to assess at a distance, compared with the number of conspecifics. In other words, vultures tended to favour quantity signals (with the number of conspecifics) over quality signals (maximal vertical speed) [101]. The ‘power of the group’ may indeed in turn drive cohesion, which could itself make social information even more profitable [101,102].

The aforementioned observations relied on an experimental setting involving captive birds. While moving and foraging stand as engrained behaviours underpinning animals' life (see [79] for definition), and are thus likely to be naturally expressed, especially in a long-lived species only recently brought to captivity (two–three generations). Yet, natural foraging conditions can still be very different from those occurring in captivity. This may affect the described dynamic, amplifying or reducing the challenge and necessity of finding food. For example, the studied birds are fed every day (although with limited amounts to keep them lean and responsive to caretakers), thus certainly less motivated in finding food than their wild counterparts which regularly face food-deprived periods and need to adjust their foraging strategy as a consequence [103]. The shift in movement pattern as a result of hunger level may as well affect the balance between personal and social information. Hungry individuals facing energetic emergency may specifically prioritize the use of social information (e.g. in house sparrows [104,105]). In addition, natural feeding events can aggregate up to 100–120 vultures [29] creating conditions in which both the competition and the social information load are much higher than the ones in our experiments. Natural conditions may therefore probably exacerbate the competition and social effect highlighted in this study.

Altogether, our results provide insights into the architecture of decision-making during movement in a social bird. It highlighted the trade-offs between personal and social information these birds have to consider in order to optimize both their flying efficiency and their foraging success. As a first approximation, we considered social cues as coming from ‘conspecifics’. Strictly speaking, however, our study included two species, griffon vulture and Rüppell's vulture, albeit phylogenetically close and with similar biology. The one Rüppell's vulture, in fact, used social information provided by surrounding vultures and did not stand out as an outlier in its behaviour. Despite being from another species, this individual had a stable dominance rank between years and was not the lowest ranked, it also developed a range of affinities similar to other individuals. In this line, in West Africa, both species are commonly seen together in foraging groups. It is known that even phylogenetically distant individuals could be an important source of social information, not only about the presence of carcasses [106], but also about the availability of thermals when sharing the same airspace (e.g. from black kites, Milvus migrans, or common swifts, Apus apus, [107,108]). Interactions with heterospecifics can indeed drastically affect animals' daily life [109], up to shaping the cognitive machinery underpinning their foraging decisions [110]. How heterospecific cues are used when foraging remains clearly overlooked. Future studies in this direction could provide valuable insights into understanding the fundamental rules dictating how animals decide where to go.

Ethics

Mentioned in the paper: experiments followed the animal ethic guidelines of France and the Centre National de la Recherche Scientifique. Handling of birds to fit GPS loggers followed the protocol of telemetry study of vultures authorized in the Programme Personnel 961, coordinated by OD, under the supervision of the French ringing centre, CRBPO, Paris. Furthermore, experiments, observations, handling and flight events were systematically performed under the guidance of the head of animal caretakers, BN.

Data accessibility

Data, scripts and supplementary video S1 are available here doi:10.48579/PRO/WYKS5A [55].

Electronic supplementary material is available here [111].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors' contributions

Y.S.: conceptualization, formal analysis, investigation, methodology, software, visualization, writing—original draft, writing—review and editing; B.N.: investigation, resources; M.S.: software, writing—review and editing; Y.T.: conceptualization, writing—review and editing; O.D.: conceptualization, investigation, supervision, writing—review and editing; B.R.: conceptualization, methodology, software, supervision, validation, visualization, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by the GAIA doctoral school grant, University of Montpellier (Y.S.).