Study Area

Golden Eagles in eastern North America breed in the Canadian provinces of Québec, Labrador, and Ontario and winter in the central Appalachian Mountains and surrounding lowlands (Katzner et al. 2012). Each spring and autumn these birds migrate long distances along Appalachian ridgelines between their summering and wintering grounds. Migratory decisions, straightness of flight, and flight modes used are strongly influenced by weather, topography, and the experience of the bird (Duerr et al. 2014, Katzner et al. 2015, Miller et al. 2016b). The northern portion of the migratory routes of these eagles is over hilly terrain, boreal forest, and in areas characterized by long cold winters and short summers (Duerr et al. 2014). The southern portion of their migratory route takes them over more diverse landscapes of mountain ridges and dendritic valleys, plateaus, and coastal plains. Land cover is generally mixed conifer and deciduous temperate forest. The climate in southern portions of migratory routes is more mild than farther north.

Data Collection

Golden Eagles were captured on wintering grounds during 2008–2015 and fitted with 80−95 g CTT-1100 (Cellular Tracking Technologies, Rio Grande, New Jersey, USA) solar-powered GPS telemetry units. The units weighed <2.5% of body mass and were attached to the eagles as a backpack using a nonabrasive Teflon ribbon harness (Kenward 1985). Telemetry units collected, at 30 s or 15 min intervals, GPS data on location, instantaneous speed, fix quality, course over ground, and altitude (Lanzone et al. 2012). Data at 30 s intervals were collected predominantly in or near the long and linear ridges of Pennsylvania (Lanzone et al. 2012), whereas 15 min interval data were collected across the entire migratory route, spanning dramatically more diverse topography. Data were stored on the units and, at regular intervals when in mobile phone coverage, sent across the GSM (Global System for Mobile Communications) network.

We estimated the age of captured eagles using molt patterns (Jollie 1947, Bloom and Clark 2001). We classified birds as juvenile (first year of migration), sub-adult (2nd–4th autumn or spring migration), or adult (> 4th autumn or spring migration).

Data Processing

We focused statistical analyses exclusively on “migratory” movements (Duerr et al. 2014, Miller et al. 2016b). We identified the start of migration when an eagle left a summer or winter home range and made a directed flight toward its ultimate winter or summer range (Duerr et al 2014). We identified the end of migration when an eagle stopped making continuous northward or southward progress. During these periods, we only considered “migratory flight,” which we defined as occasions when eagles moved >10 km hr⁻¹ (Katzner et al. 2012, Duerr et al. 2014). We filtered data separately for hourly and daily scale analyses. For daily scale analyses we considered only days on which we could clearly identify a roosting site (i.e. repeated points at one spot with zero speed and altitude above ground level generally <50 m) in the morning and evening and in which there were no data gaps >1 hr in duration. For hourly scale analyses, we considered only hours with at least 3 GPS points (for 15 min data) or with at least 30 GPS telemetry points (for 30 s data).

We measured distances eagles traveled using the Tracking Analyst “Points to Lines” tool (ArcGIS 10.1; ESRI, Redlands, California, USA) to connect sequential GPS locations and then to measure the length (in km) of tracks. At a daily scale, we calculated distance traveled for each bird by summing the lengths of all the tracks between the presumptive morning and evening roosts. Time in flight thus could vary considerably among days. At an hourly scale, we calculated the distance traveled for each bird-hour by summing the lengths of all the tracks between the points nearest to the start and end of each nominal hour (e.g., 0800–0900). We then calculated speed measurements (in km hr⁻¹) from these data. We calculated flight speed in 3 manners: (1) once at a daily scale, using only 15 min data (subsampling 30 s data where required); (2) once at an hourly scale, using only 15 min data (again, subsampling 30 s data when required); and (3) once again at an hourly scale, but this time using only 30 s data.

Data Associations

We linearly interpolated weather data at each eagle location using the RNCEP package (Kemp et al. 2011) in R (R Core Team 2012). RNCEP draws on meteorological data modeled by the National Centers for Environmental Prediction (NCEP; Kanamitsu et al. 2002). We focused on weather variables that we expected would most strongly influence migration of these Golden Eagles, per Duerr et al. (2014). These included surface temperature (°C), surface atmospheric pressure (Pa; mmHg), downward solar radiation flux (DSR; W per m²), west–east surface wind vector (U-wind; m s⁻¹), south–north surface wind vector (V-wind; m s⁻¹), Omega, the vertical velocity of air (Pascals s⁻¹), thermal energy at flight altitude (°C per 100 m), and flow assistance, the support or resistance the bird experiences from the air flow (FA; calculated using the Tailwind function based on wind vectors at flight altitude and reported in m s⁻¹; Kemp et al. 2012). Data for each flight segment were then averaged among eagle locations across the segment.

Data Analysis

Because weather variables can be highly correlated, we calculated bivariate Pearson correlation coefficients among all possible pairs of weather variables for each migration season. For variables with |r| > 0.5, we kept the single variable of the pair that we thought provided the greatest ecological insight.

We used one-way repeated measures ANOVA (package ‘nlme' in R; Pinheiro et al. 2015) and Tukey's test for post-hoc multiple comparisons (package ‘multcomp' in R; Hothorn et al. 2014) to compare weather conditions that birds of different age classes experienced during measured migration tracks. Subsequently, for each of the 2 migration seasons (spring and autumn), we created 3 different sets of candidate models to examine the potential effects of weather conditions on migratory performance. These 3 sets of candidate models evaluated performance at a daily scale (one model using 15 min data) and at an hourly scale (2 separate models, one each for 15 min and 30 s data) using migration speed as the metric of migratory performance (i.e. our response variable; see Data Processing). The 2 model sets that considered the 15 min data each had 23 models (described in Table 1), chosen based on our understanding of eagle biology and published information on eagle migration response to weather (Duerr et al. 2014). We used the same candidate models for the 30 s data except, because of the smaller sample size of birds measured at this temporal scale, we were unable to include bird age in models. This model set thus had 11 candidate models (Table 1).

TABLE 1.

List of models in model sets used to evaluate migratory performance of Golden Eagles in eastern North America, 2008–2015, as measured at daily and hourly scales interpreted from 15 min GPS data and at hourly scales interpreted from 30 s GPS telemetry data. Variables used were south–north wind vector (V-wind), east–west wind vector (U-wind), flow assistance (FA), age of the bird (Age: juvenile, sub-adult, and adult), downward solar radiation (DSR), Omega (vertical wind velocity), and thermal energy.

In each case we used linear mixed effects models (lmer, program R) to account for individuals with repeated measurements. At the daily scale, we included eagle ID and day of the year as categorical random effects and bird age (as a categorical variable) and continuous meteorological variables as fixed effects. At the hourly scale, models for the 15 min data included eagle ID, day of the year, and hour of the day as categorical random effects, and bird age and meteorological variables as categorical and continuous fixed effects, respectively. Day of the year and hour of the day were included as random effects in these models because flight behavior of different eagles can be correlated within a season or within a single day (Duerr et al. 2014, Katzner et al. 2015).

Prior to running mixed effects models, we scaled all explanatory and response variables to the range [0,1]. We also examined the residuals of each model and found that the response variable at both daily and hourly scales showed a high degree of heteroscedasticity. As a consequence we used a double square-root transformation of the response variable in models. We confirmed that this transformation removed the heteroscedasticity with a Breusch-Pagen test performed with the R package ‘lmtest' (Hothorn et al. 2015).

We used an information–theoretic approach for model selection (Burnham and Anderson 2004). Within each model set we ranked each model based on its Akaike information criterion value, corrected for small sample sizes (AIC_c), and we considered the top model as best if its AIC_c weight (w_i) was >0.90 (Burnham and Anderson 2002, Grueber et al. 2011). If rankings showed several competing models (w_i < 0.90), we averaged models with w_i > 0.01 using ‘MuMIn' package in R (Barton 2015). Means ± SE are provided unless indicated otherwise and sample sizes are provided in Appendix Table 6.

Meteorological Associations of Migration

In both spring and autumn, cloud cover and downward solar radiation (DSR) were negatively correlated with each other (Appendix Table 7A and 7B). Likewise, flow assistance was correlated with V-wind during spring migration. We therefore removed cloud cover and, in the case of spring models, V-wind, from models. We retained for analysis all other variables.

Adult eagles encountered lower levels of solar radiation and flow assistance during spring migration than did either juveniles or sub-adults (DSR: F = 12.65, df = 2 and 1,302, P < 0.001; FA: F = 6.76, df = 2 and 1,302, P < 0.01; Figure 3A and 3B) and faster west–east wind speeds compared to sub-adults (F = 7.75, df = 2 and 1,302, P < 0.001; Figure 3C). In spite of this, all age classes of eagles traveled at similar daily rates (F = 0.52, df = 2 and 1,302, P = 0.59; Figure 3E). However, hourly speed differed among the age classes (F = 13.36, df = 2 and 6,405, P < 0.001; Figure 3F). A post-hoc comparison suggested that adult eagles traveled at slower hourly speeds than did juveniles or sub-adults (difference of means: juvenile vs. adult: 1.80 km hr⁻¹, 95% CI: 0.50−3.09 95% CI; sub-adult vs. adult: 1.37 km hr⁻¹, 95% CI: 0.51−2.23).

FIGURE 3.

Mean (± 95% CI) for meteorological variables associated with daily migration of and for distances traveled by Golden Eagles in eastern North America, 2008–2015. Plots show (A) downward solar radiation, (B) flow assistance, (C) west–east surface wind vector, (D) south–north surface wind vector, (E) daily distance traveled, (F) hourly distance traveled. Autumn migration is represented by open circles and spring migration by filled circles. Sample sizes are in Appendix Table 5. Age classes of eagles are Juvenile (Juv.), sub-adult (SubAd.), and Adult; for details see Methods.

During autumn, adult eagles encountered lower levels of flow assistance and greater levels of south–north wind than did sub-adults (FA: F = 8.49, df = 2 and 1,302, P < 0.01; V-wind: F = 5.38, df = 2 and 1,302, P = 0.02; Figure 3B and 3D). There were no statistical differences between the 2 age classes in any of the other meteorological variables associated with migration (P > 0.05 in all other cases; Figure 3). As was the case for spring migration, during autumn the daily rate of travel was similar for both age groups (F = 1.44, df = 2 and 1,302, P = 0.23; Figure 3E) but adult eagles traveled at slower hourly speeds compared to sub-adults (t_1,637 = 3.89; P = 0.049; Figure 3F).

Determinants of Migratory Performance in Spring

During spring migration, daily and hourly migratory performance of eagles, as measured with 15 min GPS data, were most strongly influenced by downward solar radiation, flow assistance, and the interaction between flow assistance and age (Table 2). There was no support for a role of any of the other environmental parameters we considered. Daily migratory performance improved with increasing age, but the response to flow assistance was strongest for juveniles and weakest for adults (Table 3). Hourly migratory performance showed relationships to weather nearly identical to those at a daily scale (Table 2), again with response to flow assistance inversely correlated with age (Table 3).

TABLE 2.

Summary of model selection results for models describing migratory performance as a function of meteorological factors during spring migration by Golden Eagles in eastern North America, 2008–2015. We used linear mixed effects models ranked by ascending ΔAIC (the top 5 models are shown). Random effects were bird ID, day of the year, and hour of the day; fixed effects included the east–west wind vector (U-wind), flow assistance (FA), age of the bird (Age: juvenile, sub-adult, and adult), downward solar radiation (DSR), Omega (vertical wind velocity), and Thermal Energy. See Methods: Data Associations for units and additional details on meteorological factors.

TABLE 3.

Coefficient estimates of top model (daily and hourly, 15 min data) and from model averaging (hourly distance 30 s) for models in Table 1 describing migratory performance as a function of meteorological factors during spring migration by Golden Eagles in eastern North America, 2008–2015. Adult eagles were used as a base comparison between the different age groups. Parameters were downward solar radiation (DSR), flow assistance (FA), east–west wind vector (U-wind), and age (AgeS = sub-adult eagles). See Methods: Data Associations for units and additional details on meteorological factors. Prior to analysis, variables were rescaled and double square root transformed.

Speed based on data collected at 30 s intervals was not well described by a single model (Table 2). Instead, there were 5 models with support in the data, each with a flow assistance term and, in 4 cases, one other term. The 4 other terms (one in each model) were for DSR, U-wind, thermal energy, and Omega. The fifth model only had a flow assistance term and very little support in the data. Speed was most strongly influenced by flow assistance (model-averaged estimates; Table 3). DSR was also influential but east–west winds, thermal energy, and Omega were all relatively less important to speed (Table 3).

Determinants of Migratory Performance in Autumn

During autumn migration, daily and hourly migratory performance, as measured with 15 min GPS data, were most strongly influenced by south–north wind, flow assistance, eagle age, and an interaction between flow assistance and age (Table 4). At a daily scale, eagle migratory performance was negatively affected by age and flow assistance. V-wind also influenced migration speed; winds from the south (positive V-wind) slowed the speed of migration and winds from the north (negative V-winds) increased the speed of migration. Once again, younger birds responded more strongly to flow assistance than did adults (Table 5). At an hourly scale, V-winds had a similar effect on migratory performance as in the daily-scale model (positive V-winds reduced migratory performance), flow assistance improved migratory performance, and younger birds performed worse than adults but again responded more strongly to flow assistance.

TABLE 4.

Summary of model selection results for models describing migratory performance as a function of meteorological factors during autumn migration by Golden Eagles in eastern North America, 2008–2015. We used linear mixed effects models ranked by ascending ΔAIC and we only show the top 5 models. Random effects were bird ID and day of the year; fixed effects included south–north wind vector (V-wind), flow assistance (FA), age of the bird (Age: sub-adult and adult), downward solar radiation (DSR), Omega (vertical wind velocity), and Thermal Energy. See Methods: Data Associations for units and additional details on meteorological factors.

TABLE 5.

Coefficient estimates of top model (daily and hourly, 15 min data) and model averaging (hourly distance 30 s) for models in Table 1 describing migratory performance as a function of meteorological factors during autumn migration by Golden Eagles in eastern North America, 2008–2015. Adult eagles were used as a base comparison between the different age groups. Parameters were downward solar radiation (DSR), flow assistance (FA), south–north wind vector (V-wind), Thermal Energy, and age (AgeS = sub-adult eagles). See Methods: Data Associations for units and additional details on meteorological factors. Prior to analysis, variables were rescaled and double square root transformed.

No single model described hourly speed of travel calculated using data collected at 30 s intervals (Table 4). Instead, a suite of models was supported. Of the 5 models with the most support in the data, all included a flow assistance term. Four of these models also included one other term, either DSR, Omega, thermal energy, or V-wind. The fifth model had the most support in the data and included only the flow assistance term. Hourly speed of travel was most strongly influenced by flow assistance. Other environmental parameters had relatively little influence on speed of migration. Thus, eagles flew faster especially in response to changes in tailwinds (Table 5).

Seasonal Variation in Migratory Performance

The strong inter-seasonal differences in responses of eagles to meteorological variables are consistent with these birds taking advantage of prevailing conditions to maximize migratory performance. The weather these eagles experience varies between seasons (Duerr et al. 2014), the birds switch between different flight modes in response to weather (Katzner et al. 2015), and their use of those different flight modes has consequences for flight speed (Duerr et al. 2012). Seasonal, daily, and hourly responses of flight behavior are well understood for birds migrating through Europe and Africa, where variation in migratory performance has been linked to spatial variation in tailwinds and thermal convection, especially over arid deserts (Liechti et al. 1996, Shamoun-Baranes et al. 2003, Klaassen et al. 2008, Chevallier et al. 2010, Vansteelant et al. 2015). In eastern North America, in contrast, where humidity is higher and there are no deserts, spatial and seasonal variation in thermal availability is dramatically different. Our analyses show that behavior and performance of eagles in this region responded to this inter-seasonal variation. In particular, in spring, solar radiation, flow assistance, and age helped determine hourly and daily-scale migratory performance, whereas in autumn performance was shaped by wind availability and age.

Influence of Scale of Measurement on Interpretation of Migratory Performance

Although flight data have been collected at intervals as short as 10 s (Vansteelant et al. 2015), we know of no other study that explicitly compares migration performance as recorded with shorter- vs. longer-interval telemetry data. In this case, because our starting models differed slightly (i.e. we were unable to include an age term in the 30 s data) and because the data collected at 30 s intervals were collected only in one region, we were unable to make direct statistical comparisons between measurements at these 2 scales. However, qualitative comparisons that recognize the differences in model starting points can still be informative. For example, it is clear that the 30 s tracks that we collected are a more precise measure than the 15 min data of the distance traveled by the bird. Because of this, the 30 s tracks averaged ∼12% longer than the tracks derived from 15 min data.

Those highly detailed data that we collected were from eagles flying through a predefined geographic area in the central Appalachian Mountains in Pennsylvania (Lanzone et al. 2012). In this region, the landscape is composed predominantly of long, linear ridges and hilly terrain (Bailey 1993). Such a topographic backdrop creates spatial variation in updrafts, and the 30 s data appear to capture a suite of micro-scale responses eagles used to migrate through such a variable landscape. In contrast, the 15 min data were collected across a much greater spatial extent (the entire geographic range of this subpopulation) and eagle responses are almost certainly more variable across this region than within the area where 30 s data were collected. Thus, the topographic consistency in the area where 30 s data were collected likely explains a large part of the differences in model results. Nevertheless, comparison of these 2 datasets is useful because it highlights the importance of the spatial and temporal scales of measurement when evaluating migratory performance. Similar data sets from other species, especially where age is accounted for more completely than we could do here, would provide insight into the comparative details of soaring birds' responses to variable updraft environments.

Inverse Age Effects on Migratory Performance

Migratory performance of birds typically improves with age (Thorup et al. 2003, Mueller et al. 2013, Sergio et al. 2014). Therefore, we expected experienced adult eagles to migrate in optimal weather conditions, to orient more effectively, and consequently to travel more quickly than younger, more inexperienced, eagles. This was not the case. In fact, although daily distances traveled did not differ among age classes, hourly migratory performance of young eagles was better than that of older eagles (younger eagles flew more quickly than did older eagles; Figure 3F). These patterns run contrary to those in much of the previously published literature (Thorup et al. 2003, Mueller et al. 2013, Sergio et al. 2014). We believe that the explanation for this counterintuitive observation is tied to age-specific timing of migration. In particular, younger eagles migrated later in spring and earlier in autumn than did older eagles (Miller et al. 2016b). As such, they migrated when solar incidence angles are higher, solar radiation was more intense, and thermals were relatively more available. Such conditions are positively associated with higher cross-country soaring speeds (Reichmann 1978). In contrast, adult eagles flew earlier in spring and later in autumn, when thermals were relatively less available and flight speeds consequently slower. These patterns indicate that factors other than experience likely help determine migratory performance in this system.

There may be season-specific selective pressures that cause factors other than experience to determine migratory performance. In this case, adult eagles may see fitness benefits by migrating earlier in the spring season, allowing them to return early to establish breeding territories. Early in the spring, thermals are limited in this region, birds must rely more on orographic updraft or powered flight, and hourly travel speeds are consequently lower (Duerr et al. 2012, 2014). These and other analyses suggest that adults not only are traveling less quickly, but they are also following a less direct migratory pathway (Miller et al. 2016b). However, because daily migratory performance is similar in each age class, we expect that on a daily basis, adult eagles spend more time actively migrating than do younger birds (i.e. they have longer travel days and take fewer breaks en route); previous analyses also support this notion (Miller et al. 2016b).

In contrast, younger birds (juveniles and subadults) do not need to find and defend nesting territories during spring. As they do not breed, they do not face the same fitness pressures as adults to migrate early. Indeed, arriving later in the breeding season may provide adaptive benefit because territory defense by adults grows weaker as the nesting season progresses (Watson 2010). In autumn, non-territorial younger birds also are usually able to leave summering grounds earlier than territorial adults, especially those with dependent young. In both seasons, these strategies allow younger eagles to have more flexibility in choosing the timing of their migration, such that they can choose only to migrate when weather conditions are most suitable for long-distance travel. These timing-based strategies are also likely the mechanism by which a younger, inexperienced bird, whose orientation and weather-compensation abilities are weaker, may still fly farther than an older, more experienced bird that better orients and compensates for poor weather conditions.

Although age is often used as the single proxy for migratory experience and as the only predictor of migratory performance, our data suggest that such a single-faceted approach may cloud interpretation of migratory performance. In fact, our results suggest complex linkages among timing of migration, weather, experience, and stage- or age-specific selective pressures that may drive migratory performance. Key to understanding these patterns is comparing across seasons, when selection may operate differently on age classes, comparing within seasons, when weather may operate differently on age classes, and understanding the effect of the spatial and temporal scales at which migratory performance is measured.