a) Experimental setup. In a large-scale motion capture system (SMART-BARN), the feeding area, the monitors and the predator running wire were installed (a 3D view on the left, and a top view on the right). b) Still frames captured from the periods before (top) and during (bottom) the predator presentation, respectively. c) Sequence of an event presenting a model predator. Each predator presentation event started with the presentation of a looming stimulus (predator silhouette), which lasted for approximately 10 s, on either of the two monitors, followed by the model predator running on the wire across the room (lasting for approximately 3 s). The “before-cue” period is defined as the 2 minutes directly preceding the onset of the looming stimulus. d) Histograms of the pigeons’ responses. Top: latency to foveate (green) and latency to escape –either running or flying (purple); bottom: latency to run (blue) and latency to fly (red) separately. The vertical line shows the looming stimulus offset.

A set of hypotheses testing the assumptions about vigilance

Pigeons’ foveation during the looming stimulus presentation. a) A pigeon equipped with motion capture markers on its head and back. b) Reconstruction of estimated foveal projections in the local coordinate system of the head (left) and in the global (motion-capture) coordinate system (right). c) Reprojection of each object of interest—the monitor presenting the looming stimulus, the monitor without the cue acting as a control, and the nearest conspecific—within the visual field of all pigeons across all trials. The red squares denote the designated region of estimated foveal projections in the visual field, defined as 75±10° in azimuth and 0±10° in elevation. The color scale illustrates the normalized frequency (normalized to a unit sum) at which each object of interest was observed within each 5×5° cell of the heatmap. d) Normalized frequency (normalized to unit sum then logit-transformed) at which each object of interest was observed within the defined foveal regions on the heatmap. The dots represent single observations (jittered horizontally for visualization), the grey shade represents the violin plot of the distributions, and the boxplot’s boxes represent the 0.25 and 0.75 quartiles (with the median represented as a line inside the box) and the whiskers the minimum and maximum values within the lower/upper quartile ± 1.5 times the interquartile range.

Results of the models from Test 2 and 3. a) Latency to foveate on the predator cue (log-transformed) as a function of behavioral state (feeding or head-up) at the onset of the looming stimulus. The dots represent single observations (jittered horizontally for visualization), the grey shade represents the violin plot of the distributions, and the boxplot’s boxes represent the 0.25 and 0.75 quartiles (with the median represented as a line inside the box) and the whiskers the minimum and maximum values within the lower/upper quartile ± 1.5 times the interquartile range. b-c) Latency to foveate on the predator cue as a function of the proportion of time spent being head-up during the “before-cue” period (b) and as a function of the pecking rate during the same period (c). d-e) Probability of escaping before the looming stimulus offset (d), and latency to fly (e) as a function of the latency to foveate. (f) Histogram of the distribution of the response delay (time between the latency to foveate and the latency to fly). For all depicted results, regression lines were determined with other variables held constant, set to their mean values. A comprehensive table detailing the outcomes of the models can be found in Table S7.

Results of test 4. a-b) Latency to foveate of the first individual foveating on the monitor as a function of the other individuals’ probability of escape before the cue offset (a) and the other individuals’ latency to fly (b). All the regression lines were calculated with other variables constant and equal to their mean value. A detailed table of the models’ results is available in Table S7. c) An overview of the permutation test procedure. d) Results of the permutation test. Histograms denote the distributions of the average time gap in permutated data. The vertical dashed line indicates the mean time gap from the observed data. The proportion of permutated data less than the observed data (on the left of the dashed line) gives the p-value; escaping (top) and flying (bottom); without (left) or with (right) distance control.

definition of the automated classification of the behaviors. See Table S3

Total number of individuals that exhibited foveation, escape, and flight behavior over the course of the trials and events. Each event was composed of a flock of ten pigeons.

Response variables as a function of the trials and events. a) latency to foveate; b) latency to escape; c) latency to fly. The distribution of the data is displayed as violin plots and the color represents the side of the predator event.

Predictor variables as a function of the trials and events. The distributions of the different variables are displayed as violin plots and the color represents the side of the predator event. a-e) behavior during the “before cue” period (2 minutes period straight before the onset of the looming stimulus) as a function of the trial and event: proportion of time head-up (a), proportion of time foveating on either of the monitors (b), proportion of time foveating on one of the conspecifics (c), saccade rate (d), pecking rate (e). f-h) behavioral and spatial state measures at the onset of the looming stimulus as a function of the trial and event: distance to the monitor (f), distance to the nearest neighbor pigeon (g), the number of individuals within a flock exhibiting respective behaviors (feeding, head-up or other behavior) exhibited at the onset on the looming stimulus (h).

Illustration of the defined spherical representations for the pigeon’s head and body.

Explanation of the different responses and variables used in our models.

Description of the behavior classification definitions. Terms definitions: state = behavior that lasts in time; event = punctual behavior. Body-head vector = the vector originating from the body center and projecting to the head centroid. Head direction vector = the vector projecting to the front of the head and towards the horizon when the head is still (corresponds to the y axis of the local coordinate system, projecting 30° above the beak-head center axis). Pitch = head rotation movement corresponding to “nodding” (pitch = 0° corresponds to the head direction vector pointing towards horizon, pitch < 0° corresponds to the head direction vector pointing down, pitch > 0° corresponds to the head direction vector pointing up). Roll = head rotation movement where the head is “tilted” (roll = 0° corresponds to a straight head and roll ≠ 0° corresponds to a head tilted to the right or to the left).

Inter-rater reliability for behavior classification. Pecking count was rated with ICC and other behaviors with Cohen’s kappa. HR1 and HR2 represent the human raters and ACC the automated computer classification.

Histogram of the pigeons’ speed for the whole dataset. Running unusually fast is defined as the body center moving with a speed larger than 0.6 (vertical blue line)

Behavioral changes between the 1 min period before the onset of the looming stimulus and the 1 min period after the disappearance of the predator: proportion of time spent being head-up (a), proportion of time spent foveating on one of the monitors (b), proportion of time spent foveating on one of the conspecifics (c), saccade rate (d), pecking rate (e), distance to the closest monitor (f), distance to the nearest neighbor (g). The dots represent single observations (jittered horizontally for visualization), the grey shade represents the violin plot of the distributions, and the boxplot’s boxes represent the 0.25 and 0.75 quartiles (with the median represented as a line inside the box) and the whiskers the minimum and maximum values within the lower/upper quartile ± 1.5 times the interquartile range.

Model results for behavioral comparison between the 1 min period before the onset of the looming stimulus and the 1 min period after the disappearance of the predator. For all models, the table details the beta estimates, as well as the Chi square value, degrees of freedom and p-values from the likelihood ratio test. Event, predator side, sex, pigeon ID (random effect) and trial (random effect) has been added as control variables.

R formulas used for the different models. Variables starting with “Logit” have been logit-transformed, with “Log” have been log-transformed. As control variables, we have included in all models the event number, the predator side, the sex of the individual, the pigeon ID (random effect) and the trial ID (random effect).

Results of all models related to our main hypotheses. This table presents detailed information for each predictor in every model used in this study, including the beta estimates from the model, as well as the Chi-square value, degrees of freedom, and p-values derived from the likelihood ratio test. Note: (a) was incorporated as a random effect in the model and, as such, does not have a beta estimate.

Results of all repeatability analyses including the observed variable, the repeatability estimate and the p-value.