Sea-crossing along migratory flyways is limited more strongly by wind than by lack of uplift

The open sea is considered an ecological barrier to terrestrial bird movement. However, over-water journeys of many terrestrial birds, sometimes hundreds of kilometers long, are being uncovered by bio-logging technology. To understand how these birds afford their flights over the open sea, we investigated the role of atmospheric conditions in subsidizing sea-crossing behavior at the global scale. By analyzing forty years of temperature data, we show that the spatio-temporal patterns of sea-crossing in terrestrial migratory birds correspond to favorable uplift conditions. We then analyzed route selection over the open sea for four bird species with varying levels of dependence on soaring flight, representing all major migratory flyways worldwide. Our results showed that favorable uplift conditions, albeit not as common and as powerful as over land, are not rare over the open seas and oceans. Moreover, wind, which is more variable than uplift in its spatio-temporal distribution, is the determining factor in the birds’ route selection over the open sea. Our findings suggest a need for revisiting how ecological barriers are defined, to reflect what we know of animal movement in the era of bio-logging.


Introduction
Dynamic atmospheric conditions largely define the energetic costs of flight for birds. Tail winds for example, permit birds to reduce air speed while maintaining the speed of travel, helping them to save energy [1,2]. Likewise, rising air as a consequence of the interplay between wind and topography causing orographic uplift or rising warm air creating thermal uplift, can push flying animals upwards and reduce the energetic costs of remaining airborne [3][4][5]. The energy landscape is however interspersed with ecological barriers where energetic subsidies in the atmosphere are weak or absent, impeding efficient movement. Flight over ecological barriers becomes thus energetically costly, yet is unavoidable for some animals, particularly during migration [6,7]. How birds afford their flights across such seemingly hostile environments remains an open and important question in movement ecology.
The open sea is considered such a major ecological barrier for terrestrial bird migration, particularly for soaring birds [7]. This generally accepted assumption roots in observational studies of birds gathering in large numbers at bottlenecks prior to setting out over even relatively short over-water passages [7,8]. Early studies of bird migration, particularly for soaring birds, interpreted these large aggregations as a sign of a general unwillingness of the birds to fly over the open sea [9]. Yet, with the advancement of bio-logging technology, sea-crossing journeys of many terrestrial bird species, in all major migratory flyways, are being uncovered.
The atmospheric conditions, in particular wind, are suggested to subsidize terrestrial birds' flight over the open sea, facilitating sea crossings [10][11][12][13][14]. Yet, some studies also show that wind support (i.e. the length of the wind vector in a bird's flight direction) is not imperative for sea-crossing [15][16][17]. In fact, high wind speeds can hinder movement, by preventing migratory birds from initiating sea-crossing [18], or drifting birds from their paths and resulting in death at sea [19]. Wind, therefore, seems to have a variable impact on the propensity and the energetics of sea-crossing, as it can both reduce and add to the costs of flight, depending on its speed and direction.
A recent and emerging hypothesis is that uplift also plays a role in the energy seascape for terrestrial bird migration. This is particularly important for soaring birds, because the energetic costs of flapping flight for them is exceptionally high [4,5]. Recent bio-logging studies measured flight altitude in soaring birds to provide indirect proof of thermal soaring behavior at sea [20,21]. More recently, high resolution GPS tracking documented the circling flight pattern and vertical aerial climb of migrating ospreys over the Mediterranean Sea [22]. These studies also confirmed the earlier suggestions that ∆T, defined as the difference in temperature between the sea surface and the air can be used as a proxy for uplift potential [23] and predicts the occurrence of soaring flight at sea. Positive ∆T values correspond to upward moving air, while negative values can be interpreted as sinking air (i.e. subsidence). This proxy was consequently adopted to quantify the energy seascapes that enable juvenile European honey buzzards to cross the Mediterranean Sea successfully [16].
We have now come to the realisation that sea-crossing might not be a rare behavior and that it can be subsidized by atmospheric conditions. To reveal general patterns in the role of uplift and wind in the energy seascape for terrestrial bird migration, we investigate sea-crossing behavior at a global scale. We hypothesize that at the global scale, the spatio-temporal patterns of sea-crossing in soaring birds corresponds to positive values of ∆T. Consequently, we hypothesize that wind, which is more variable than ∆T in its spatio-temporal distribution, as well as its influence on sea-crossing, is the limiting factor in terrestrial birds' route selection over the open sea. We tested the first hypothesis by modeling the spatio-temporal distribution of ∆T in the main sea-crossing regions of the world. To test for the second hypothesis, we used biologging data to investigate route selection over the open sea.

Bio-logging data set
We compiled a bio-logging data set containing migratory trajectories of birds that regularly perform sea-crossings. Our data was comprised of five species: the Oriental honey buzzard Pernis ptilorhynchus and the grey-faced buzzard Butastur indicus in the East Asian flyways, the osprey Pandion haliaetus and the peregrine falcon Falco peregrinus, in both the African-Eurasian and the American flyways, and the Eleonora's falcon Falco eleonorae in the African-Eurasian flyway. These birds are all capable of soaring flight, although their dependence on uplift varies, with the falcons and the osprey being less dependent on uplift than the buzzards [9].
We restricted the geographic extent of our study to the 0 • -60 • latitudinal zone in the northern hemisphere. The majority of sea-crossing instances happen within this zone (but see for example [24]). Moreover, we only investigated autumn migrations, as spring migration data were scarce.
We focused only on sea-crossing behavior during migration to ensure a common flight purpose among all the species and individuals in the analyses. We only included adults as they actively select their route based on experience, unlike juveniles that follow an innate direction of migration, likely without any route selection criteria (e.g. [25]). We limited our analysis to sea-crossing trips longer than 30 km. Shorter sea-crossings do not always represent individuals' decision-making over the sea. They can be accomplished, for example, by gaining height over land and gliding over the sea to the opposite shore [17]. In addition, due to the spatio-temporal coarseness of the atmospheric data used, we did not consider it informative to model sea-crossings at shorter distances.
The bio-logging data were collected using a variety of devices and recording schedules (Supplementary Table  1). To ensure a uniform temporal resolution and to reduce spatio-temporal auto-correlation, we re-sampled all data to one-hourly intervals (with a tolerance of 30 minutes; see thinTrackTime in package Move [26]).

Spatio-temporal modeling of ∆T
To show the spatio-temporal variation in ∆T at the global scale, we used 40 years of global temperature data. We downloaded sea-surface temperature and temperature at 2 meters above the sea for the entire globe and for all the years that data was available (1979-2019) from the European Center for Medium-Range Weather Forecast (ECMWF; https://www.ecmwf.int) Era-Interim reanalysis database (temporal and spatial resolution of 6 hours and 0.7 • , respectively). After downloading the data, we randomly sampled 50,000 data points for each year, in order to reduce computing requirements in further processing and analysis. We then spatially filtered the data to exclude lakes, as we were only interested in the open seas and oceans. To include a proxy for time of the day, we calculated the solar elevation angle for each data point. We then created a categorical variable with three levels, night, low sun elevation, and high sun elevation, corresponding to sun elevation angles below -6, between -6 and 40 degrees, and over 40 degrees, respectively.
We loosely followed Nourani et al. [16] to construct energy seascapes. In brief, we modeled ∆T as a function of latitude, longitude, day of the year, and time of the day using the generalized additive mixed modeling (GAMM) approach. We constructed four models in total, one for each region, where regular long-distance sea-crossing is performed by terrestrial birds, namely South-East Asia, the Indian Ocean, Europe, and the Americas. We extracted the timing and location of sea-crossings from our bio-logging data set. We did not have empirical data for bird migration over the Indian Ocean and therefore consulted the relevant literature to extract the spatiotemporal pattern of the Amur falcon's Falco amurensis sea-crossing over the region [27]. Each model included two smoothers, one cyclic cubic regression splines smoother for the day of the year and a spline on the sphere for latitude and longitude. For both of these parameters, one smoothing curve was estimated for each level of time of day. Year was added as a random intercept in the models to control for annual variations in ∆T. We also included a variance structure in the models to account for the heteroscedasticity caused by higher ∆T variance in higher latitudes. Models were fitted using the mgcv package [28] in R ver. 4.0.2 [29]. We used each model to predict energy seascape maps for the autumn migration season (August-November). We spatially interpolated the prediction rasters to a 1 km resolution, for visualization purposes.

Route selection analysis
We investigated route selection by fitting a step selection function [30] to relate the probability of presence over the sea with atmospheric conditions. Every two consecutive points along a track were considered a step. Atmospheric conditions were compared between the observed step and a set of alternative steps that were available to the birds in space and time. The grey-faced buzzard was excluded from this analysis because of the insufficiently low resolution of the satellite-tracking data.

Data preparation
We filtered our data set to include only points over the open sea. Trajectories that intersected land, e.g. islands, were broken into sea-crossing segments and were analysed separately. Sea-crossing segments shorter than 30 km, and those that included fewer than 3 tracking points, were removed.
We prepared a stratified data set for analysis: along each sea-crossing segment, for each step, we generated 69 spatially alternative steps based on the step lengths and turning angles of the empirical data, so that each stratum had a total of 70 steps (1 used and 69 alternative). All data were then annotated using the ENVdata track annotation service [31] provided by Movebank (www.movebank.org ). Each point was annotated with u (eastward) and v (northward) components of the wind, sea-surface temperature, and temperature at 2 m above the sea, all provided by ECMWF. We selected the bilinear and the nearest-neighbour methods of interpolation for the wind and temperature data, respectively. We then calculated wind support [32], wind speed, and ∆T (sea surface temperature minus overlaying air temperature) using the annotated data. Additionally, to investigate whether predictability of atmospheric conditions affected the sea-crossing route choice, we annotated each point with long-term variances (over the 40 years that data was available from ECMWF Era-Interim database) for wind support and ∆T.

Data analysis
We checked the annotated data set for multicollinearity and only considered variables that were not highly correlated (r < 0.6). Prior to analysis, we converted all values of explanatory variables to z-scores (i.e. the distance of a raw value from the mean, measured in units of standard deviation) to ensure comparability among predictors.
Step selection functions were then estimated using the Integrated Nested Laplace Approximation (INLA) method, as suggested by [33] using the INLA package [34] in R ver. 4.0.2 [29]. We constructed a multilevel model with fixed effects for ∆T, wind support, wind speed, longterm variances of ∆T and wind support, as well as an interaction term for wind speed and ∆T. Species and individual IDs (nested within species) were included as random effects on the slopes. To construct alternative models, in a step-wise manner, we removed explanatory variables with insignificant coefficients (i.e. with credible intervals including zero). We then compared the alternative models using the Widely Applicable Information Criterion (WAIC) and the log marginal likelihood (MLik) scores ( [35]). The model with the lowest WIAC and MLik scores was considered the best.

Results
The spatio-temporal pattern of sea-crossing in the six terrestrial bird species corresponded with positive uplift potential over the open sea ( Fig. 1; see sub-plots for withinyear and within-day variations in each region). The osprey was the only species flying over the open sea when the sea surface was colder than the air (i.e. negative ∆T ). This pattern occurs over both the Mediterranean and the Caribbean Seas (Fig. 1). However, this behavior did not set the osprey apart from the other species in route selection criteria, as we did not find species-specific differences in route selection in our step selection function analysis ( Supplementary Fig. 1).
We analyzed route selection in 210 sea-crossing segments of 46 individuals (Supplementary Table 1). The most important variable determining over-water route se-  lection was wind support, with a positive effect. The interaction between ∆T and wind speed also showed a positive, yet smaller, effect (Fig. 2). ∆T had a small and non-significant impact on route selection (Fig. 2). Neither of the long-term variances were retained in the best model (i.e. model 3; Table 1; see Supplementary  Table 2 for detailed GAMM outputs).  Table 1.

Discussion
Our global-scale analysis of sea-crossing behavior showed that soaring birds were more selective for wind than uplift while flying over water. We found that uplift conditions are mostly favorable, particularly in autumn (Fig.  1). Wind conditions however were more variable at small spatio-temporal scales than uplift ( Supplementary Fig.  2). Thus, birds were more responsive to instantaneous wind conditions, particularly wind support, when selecting their sea-crossing routes. We used 40 years of temperature data to model ∆T as a function of space and time. This allowed us to make energy seascape maps that corresponded to when and where major sea-crossing events regularly take place. The mostly positive values of ∆T on these maps indicate that soaring birds are not facing strong subsidence when flying over water. By looking at the within-year trends in ∆T in these regions, we can further see that, with the exception of the temperate zone, ∆T has low seasonal variation (Fig. 1). This indicates that uplift could be available to migratory birds crossing the open sea during spring migration, though further studies need to confirm this with empirical data. We further observed that, during autumn migration, the range and mean of ∆T values were similar between the observed and alternative steps in our step selection function estimation (Supplementary Fig. 2). The birds thus face more variability in wind conditions than in uplift, which explains why wind support was the most important criteria for route selection (Table 1).
We also found an influence of the interaction between ∆T and wind speed in sea-crossing route selection. On the one hand, at moderate wind speeds, even small values of positive ∆T are enough for supporting over-water soaring flight [36]. On the other hand, strong winds can destroy thermals and impede soaring flight. The energy seascape therefore is formed by a complex interplay of uplift and wind conditions. To understand the energetics of terrestrial birds' flight over the open sea, future studies can employ theoretical modeling as well as highresolution bio-logging, to quantify the birds' response to different combinations of wind and uplift conditions during over-water flight.
In the literature, sea-crossing behavior is commonly associated with morphological capability of sustaining long bouts of flapping flight, aided with favorable wind. Although evidence for soaring flight over the sea has been around for at least a decade now, many studies still overlook uplift potential when analyzing sea-crossing behavior. Studies that try to include uplift do so by using a variety of proxies, including air temperature [15], air temperature gradient [12], vertical air velocity [12], boundary layer height [37], and solar irradiance [17]. These studies rarely find the uplift proxy as the most important determinant of sea-crossing. The reason for this could be that, similar to our step-selection analysis, the naturally low variance in uplift conditions make it an insignificant criterion for route selection, despite its positive effects. Another explanation could relate to the meaningfulness of the proxies themselves. These proxies are all related to upward movement of air in the atmosphere, but have not been shown to correlate with soaring flight in birds. ∆T however, has been proven, both by direct observations [36] and bio-logging technology [22], to be related to soaring flight. We encourage future studies to take advantage of this proxy, not least because widespread use of ∆T makes comparisons among studies possible. Yet, there remains a strong need for quantifying the amount of energy, or the realized uplift, that a bird can gain from ∆T. Theoretical studies could investigate this for different morphological characteristics, as well as wind and other environmental conditions.
Uplifts reduce the energetic costs of remaining airborne for flying animals. While obligate soaring birds can be dependent on this resource for crossing the sea, flapping flyers also benefit from a lack of subsidence over water. We showed that different species of raptors, with different morphological characteristics and soaring flight dependencies, all benefit from the positive uplift conditions when sea-crossing, while maximizing wind support.
Our results can further shed light on the energy seascapes for other animals flying over the open sea. The relatively high uplift during the night compared to daytime (Fig.  1) could reduce drag and lead to energetically cheaper flight in nocturnal migrants, for example over the Mediterranean [38,39] and the Caribbean [40,41] Seas. Moreover, over the Indian Ocean, dragonflies [42] and cuckoos (Cuculus spp.; https://birdingbeijing.com/the-mongoliacuckoo-project/ ) migrate within the time window indicated in Fig. 1 for the Amur falcon and can take advantage of the energetic subsidy that the atmosphere provides. The magnitude of benefit that uplift provides for these animals might be small in relation to the amount of energy they spend on flapping, but it might be what makes the difference between success or failure over the open sea.

Conclusion
The common narrative of many papers reporting seacrossing behavior in terrestrial birds is based on the premise that flying over the open sea, particularly by soaring, is an exceptional behavior. We show that this is not the case, because favorable uplift conditions, albeit not as common and as powerful as over land, are not rare over the open seas and oceans. The spatio-temporal pattern of sea-crossing in many terrestrial animals coincides with favorable uplift conditions. We further show that wind, which can increase or reduce the costs of flight depending on its strength and direction, is the main limiting factor for the propensity of sea-crossing at a global scale. Our findings suggest a need for revisiting how ecological barriers are defined, to reflect what we know of animal movement in the era of bio-logging.

Authors' contributions
EN conceived the study and designed the analyses with input from KS and PB. EN carried out the analyses and wrote the paper. ROB, OD, SG, HH, CK, OK, NL, FM, IP, AS, JFT, NT, NMY, and MW contributed data. All authors commented on and edited the manuscript drafts.