Skip to main content

Young frigatebirds learn how to compensate for wind-drift

Cite this dataset

Wynn, Joe et al. (2020). Young frigatebirds learn how to compensate for wind-drift [Dataset]. Dryad.


Compensating for wind drift can improve goalward flight efficiency in animal taxa, especially amongst those that rely on thermal soaring to travel large distances. Little is known, however, about how animals acquire this ability. The great frigatebird (Fregata minor) exemplifies the challenges of wind drift compensation because it lives a highly pelagic lifestyle, traveling very long distances over the open ocean but without the ability to land on water. Using GPS tracks from fledgling frigatebirds, we followed young frigatebirds from the moment of fledging to investigate whether wind drift compensation was learnt and, if so, what sensory inputs underpinned it. We found that the effect of wind drift reduced significantly with both experience and access to visual landmark cues. Further, we found that the effect of experience on wind drift compensation was more pronounced when birds were out-of-sight of land. Our results suggest that improvement in wind drift compensation is not solely the product of either physical maturation or general improvements in flight control. Instead, we believe it is likely that they reflect how frigatebirds learn to process sensory information so as to reduce wind drift and maintain a constant course during goalward movement. 


GPS Tracking Tracking was carried out on Europa Island in the Mozambique Channel (-22.36oN, 40.37oE) on 13 adult female and 10 juvenile frigatebirds (see figure 1). Adult males were not used in this analysis as their role in chick provisioning is limited and, hence, they show little homing motivation. Devices measured 130×30×12mm and weighed 30g (PS-RF, e-obs GmbH, Munich, Germany), representing between 1.88%–3.55% of the frigatebirds' mass. Devices were deployed dorsally using Tesa tape and were set to record location every 2 or 5 minutes. 3-dimensional accelerometer data were also gathered but is not presented here (27). Environmental and landmark cue data Wind data were derived from the NOAA Global Forecast System at a temporal resolution of 3-hours and a spatial resolution of 0.5o longitude and 0.5o latitude. Whether or not birds could see any piece of land was ascertained using their altitude, measured using GPS and smoothed using a rolling median over a window of 4 consecutive fixes; the elevation of local topography, derived from USGS Global Multi-resolution Terrain Elevation Data; and the curvature of Earth. Smoothed altitude was used in analyses as the GPS-derived altitudinal error is substantially higher than that observed in both the longitudinal and latitudinal dimensions (28, 29). Birds were assumed to be able to see land if a line of-sight could be drawn between their position and the maximum elevation of any piece of land without the Earth’s surface intervening. GPS points taken at night were removed from the analysis because it was not known whether reduced visual salience might affect access to landmark cues, and there were insufficient night-time GPS points to statistically test for an effect of daytime/night-time (980 individual fixes, representing only 24 trips from 6 unique individuals). Analysis including GPS fixes taken at night is included in the supplementary materials. Track processing, statistics and analysis All statistics and processing were conducted in R (30). Tracks were interpolated using a cubic spline function (31) so that fixes were positioned at precise 5-minute intervals. Tracks were also divided into trips out from the colony, with a trip defined as a continuous set of points recorded > 500m from the island’s coastline with a maximum distance from the colony of > 3km. Since juveniles were tracked from their very first trips to sea, for a given trip we attempted to quantify the experience of the bird at that point in its development, measuring experience as the number of trips the focal bird had been on prior to the trip in question. In total, 19,732 interpolated GPS fixes were used in the analysis of fledgling frigatebirds, representing 1001 trips from 10 individual birds (with a mean of 100 and a median of 122 trips per individual), whilst 35,430 interpolated GPS fixes were used in the analysis of adult frigatebirds, representing 345 trips from 13 individuals (with a mean of 26 and a median of 12 trips per individual). We chose to analyse only the homing sections of trajectories as we had an a priori expectation of where birds were aiming for. Because we had no expectations about the form of homing behaviour in frigatebirds, we conservatively defined homing as any points that occurred after the maximum distance to the colony was recorded on the trajectory. Due to the mechanisms by which frigatebirds generate lift (principally thermalling) and search for prey items (area restricted search behaviour) we expected individuals of all ages to engage in tortuous, non-navigational behaviours (21) . Track tortuosity was measured using the rolling standard deviation of track bearing over a window of 5 consecutive fixes, and non-navigational behaviours were parsed out using a mixture model to separate GPS fixes into 2 groups based on tortuosity (32, 33). Only points with tortuosity lower than the mixture model-defined cut-off (of 52o) were retained for use in navigational analyses. We repeated all analyses with multiple tortuosity cut-off points so as to ensure the significance of any findings were robust and unbiased by the threshold at which points were removed based on tortuosity (see supplementary materials). For each point along a homing track, a beeline direction to the colony was assigned along the Great Circle route, from which instantaneous deflection was calculated (34). Orientation behaviour was modelled using this instantaneous deflection from the beeline (measured in degrees on a -180 o to 180o scale) as a response variable. From the calculated beeline direction home, the cross-beeline and along-beeline wind components were calculated per interpolated GPS position. Using these wind components, we modelled the effect of wind drift in a 3-way interaction between the cross-beeline wind component, fledgling experience and whether or not the bird could see land as a binary factor (see figure 2). The along-beeline wind component was also included in all models. This was because we expected that an increased headwind component might reduce groundspeed, thus halting a bird’s forward progress and increasing instantaneous deflection per unit crosswind. By including the along-beeline component we, therefore, sought to standardise model output coefficients with respect to the along-beeline wind component so as the results presented were not the result of a confound between any variables of interest and the along-beeline wind component. The effect of wind drift was modelled using linear mixed-effects models with trip ID, nested within bird ID, used as random effects (34, 35). P-values were calculated using likelihood ratio tests between the hypothesised (alternative) model and a null model that did not contain the variable or interaction of interest.


University of Oxford

European Research Council, Award: ERC-2012-ADG_20120314

Agence des Aires Marines Protégées

Biotechnology and Biological Sciences Research Council, Award: BB/M011224/1

University of Oxford

French National Centre for Scientific Research

Templeton World Charity Foundation, Award: TWCF0316

Mary Griffiths Award