Compensation for wind drift prevails for a shorebird on a long-distance, transoceanic flight

Background Conditions encountered en route can dramatically impact the energy that migratory species spend on movement. Migratory birds often manage energetic costs by adjusting their behavior in relation to wind conditions as they fly. Wind-influenced behaviors can offer insight into the relative importance of risk and resistance during migration, but to date, they have only been studied in a limited subset of avian species and flight types. We add to this understanding by examining in-flight behaviors over a days-long, barrier-crossing flight in a migratory shorebird. Methods Using satellite tracking devices, we followed 25 Hudsonian godwits (Limosa haemastica) from 2019–2021 as they migrated northward across a largely transoceanic landscape extending > 7000 km from Chiloé Island, Chile to the northern coast of the Gulf of Mexico. We identified in-flight behaviors during this crossing by comparing directions of critical movement vectors and used mixed models to test whether the resulting patterns supported three classical predictions about wind and migration. Results Contrary to our predictions, compensation did not increase linearly with distance traveled, was not constrained during flight over open ocean, and did not influence where an individual ultimately crossed over the northern coast of the Gulf of Mexico at the end of this flight. Instead, we found a strong preference for full compensation throughout godwit flight paths. Conclusions Our results indicate that compensation is crucial to godwits, emphasizing the role of risk in shaping migratory behavior and raising questions about the consequences of changing wind regimes for other barrier-crossing aerial migrants. Supplementary Information The online version contains supplementary material available at 10.1186/s40462-022-00310-z.


Defining the Aerial Migratory Corridor
To establish the preferred migratory direction for our tracked godwits, we adapted methodology from Pearse et al. (2018) [1] to define their migratory corridor. We began with all godwit locations associated with northward migration. Overland locations where ground speeds fell below 3 ms -1 [2] and consecutive movements traversed < 15 km were grouped into discrete stopover events; we retained only the first and last locations for each stopover and retained all other locations. When directional movement ceased or stabilized for longer than 14 days-a pattern indicating arrival at the breeding grounds, mortality, or transmitter failure-we considered this to be the end of northward migration and removed all associated locations.
Next, we created a 95% core area polygon to serve as the migratory corridor. As we were primarily interested in the first stage of migration (i.e., from departure through the main stopover region), we focused on the area south of 50°N. We divided this area into 'windows' of equal height (see methodology below) and calculated window vertices as the medians of the transmitter-reported locations' y-dimensions (i.e., north/south) and percentiles of the locations' x-dimension (i.e., east/west) corresponding to the area between the 2.5 th and 97.5 th percentiles (n = 1,811 locations). We then connected the vertices to form a polygon outlining the migratory corridor.
To determine the optimal spacing for polygon vertices-in essence, one that sufficiently captured both within and among individual variation while avoiding overfitting-we tested multiple window heights from 2° to 6° latitude in 1° increments. We used 1,000 bootstrap estimates to compare how well the resulting polygons fit our transmitter-reported locations, seeking to maximize this value. We then used the fractal dimension index via the landscapemetrics package in R [3] to compare the complexity of the resulting polygons; as lower values were indicative of less-complex shapes, we sought to minimize this value. These comparisons informed our selection of the best-fitting, optimally smoothed polygon to serve as the overall migratory corridor for our tracked godwits.

Tolerance Value Calculation
Behaviors such as full compensation, full drift, and supported flight occur when two or more critical movement vectors are aligned. However, alignment need not be perfect, particularly given the error inherent in vector measurements. To capture cases in which these vectors are nearly aligned, we used tolerance values derived from the known imprecision of our vector estimates. When evaluating if a godwit's realized travel direction is aligned with its preferred direction, we allowed vectors to be separated by up to 0.1 rad or ~5.73°. We estimated an appropriate value using the arctangent of the maximum location error (13.0 km, corresponding with the 95 th error percentile of the least-precise location classes included in our analysis; [4]) divided by the median distance traveled between locations in our filtered, thinned flight tracks (129 km). When evaluating if a godwit's realized direction is aligned with the wind direction, we allowed vectors to be separated by up to 0.2 rad or ~11.46°, a larger value reflecting imprecision both in the godwit locations and in altitude estimation. We estimated an appropriate value using the sum of the smaller tolerance value (0.1 rad) and median difference between wind directions at the optimal and second-optimal altitude (~0.16 rad). Tolerance values can be easily adjusted for more or less precise data sources.

R Code for Behavioral Assignments
We classified behaviors by evaluating the angles between three normalized movement vectors: the angle between subsequent realized travel direction and wind direction (θrw), the angle between realized travel direction and the midpoint of the preferred range (θrd), and the midpoint of the preferred range and the wind direction (θwd), and the midpoint and one edge of the preferred range (θdd) at each location. The following hierarchical decision list assigned one behavior to each location: If θrw < τ1 & θrd > (θdd + τ2), then fully drifting.