Flight Response of Slope&#X2010;Soaring Birds To

Home , Orographic lift

Functional Ecology 2014 doi: 10.1111/1365-2435.12381 Flight response of slope-soaring birds to seasonal variation in thermal generation

Adam E. Duerr*,1, Tricia A. Miller1, Michael Lanzone2, David Brandes3, Jeff Cooper4, Kieran O’Malley5, Charles Maisonneuve6, Junior A. Tremblay†,7 and Todd Katzner‡,1,8

1Division of Forestry and Natural Resources, West Virginia University, 322 Percival Hall, Morgantown, West Virginia 26506, USA; 2Cellular Tracking Technologies LLC, 2405 North Center Avenue, Suite B, Somerset, Pennsylvania 15501, USA; 3Department of Civil and Environmental Engineering, Acopian Engineering Center, Lafayette College, Easton, Pennsylvania 18042, USA; 4Virginia Department of Game and Inland Fisheries, 1320 Belman Road, 5 Fredericksburg, Virginia 22401, USA; West Virginia Division of Natural Resources, 1 Depot Street, Romney, West 6 Virginia 26575, USA; Ministere du Developpement durable, de l’Environnement, de la Faune et des Parcs, 92, 2e Rue 7 Ouest, Rimouski, Quebec G5L 8B3, Canada; Ministere du Developpement durable, de l’Environnement, de la Faune et des Parcs, 880 chemin Sainte-Foy, Quebec City, Quebec G1S 4X4, Canada; and 8United States Department of Agriculture, Forest Service, Timber and Watershed Laboratory, Parsons, West Virginia 26287, USA

Summary 1. Animals respond to a variety of environmental cues, including weather conditions, when migrating. Understanding the relationship between weather and migration behaviour is vital to assessing time- and energy limitations of soaring birds. Different soaring modes have different efficiencies, are dependent upon different types of subsidized lift and are weather dependent. 2. We collected GPS locations from 47 known-age golden eagles that moved along 83 migration tracks. We paired each location with weather to determine meteorological correlates of migration during spring and fall as birds crossed three distinct ecoregions in north-east North America. 3. Golden eagle migration was associated with weather conditions that promoted thermal development, regardless of season, ecoregion or age. Eagle migration showed age- and season- specific responses to weather conditions that promoted orographic lift. 4. In spring, adult eagles migrated earlier, over fewer days, and under more variable weather conditions than did pre-adults, suggesting that adults were time limited and pre-adults made choices to conserve energy. In fall, we found no difference in the time span of migration or when each age class migrates; however, we saw evidence that pre-adults were less efficient migrants than adults. 5. The decision by soaring birds to migrate when thermals developed allowed individuals to manage trade-offs between migratory speed and migratory efficiency. When time was limited (i.e. spring movement of adults speeding towards nesting territories), use of whatever lift was available decreased the time span of migration. When migration was not time limited (e.g. spring movements by pre-adults, all movements in fall), eagles avoided suboptimal flight conditions by pausing migration, thus increasing the time span of migration while reducing energetic costs.

Key-words: Aquila chrysaetos, energy-limited, golden eagle, migration, orographic lift, soaring, thermal lift, time-limited, weather

Introduction

*Correspondence author. E-mail: [email protected] Animals respond to cues about environmental condition † Present address. Environment Canada, 801-1550, avenue d’Esti- or habitat quality when they move. Habitat quality, mea- mauville, Quebec City, G1J 0C3 Quebec, Canada. sured as forage availability, forage quality or habitat per- ‡Present address. Forest & Rangeland Ecosystem Science Center, meability, inﬂuences movement through terrestrial U.S. Geological Survey, 970 Lusk St., Boise, Idaho 83706, USA.

© 2014 The Authors. Functional Ecology © 2014 British Ecological Society 2 A. E. Duerr et al. environments (Lyons, Collazo & Guglielmo 2008; Avgar some cases, arriving before food is abundant (van Wijk et al. 2013). Movement through air, such as flight by birds, et al. 2012). Arrival of non-territorial birds onto breeding often is related to weather conditions that support efficient grounds is not constrained by time because they do not flapping flight (Mitchell et al. 2012) or soaring flight need to arrive early to claim a breeding site. These inexpe- (Shamoun-Baranes et al. 2006). Thus, for birds, airspace is rienced non-breeders also are likely less efficient at obtain- habitat (Diehl 2013) and weather conditions define habitat ing food, thus may be energy limited. quality. Several predictions follow the time- and energy-limited The flight of large birds is restricted by weather condi- hypotheses for soaring migrants. First, time limitation tions and topography that favour development of supple- should be correlated with decreased migratory efficiency mental lift (Leshem & Yom-Tov 1996; Yates et al. 2001; because time-limited animals would migrate whenever con- Shannon et al. 2002; Shamoun-Baranes et al. 2003; Katz- ditions permit (e.g. use both orographic and thermal lift; ner et al. 2012a). Supplemental lift for soaring land birds Duerr et al. 2012). This would allow time-limited migrants primarily is in the forms of thermal and orographic lift, to reduce the number of days spent migrating between which corresponds to flight modes of thermal soaring and wintering and breeding grounds, although they would gliding and orographic soaring (Kerlinger 1989; Alerstam spend a large proportion of each season migrating. The & Hedenstrom 1998). Thermal lift occurs when differential consequence of such migratory shifts for time-limited ani- heating of the earth’s surface warms air, causing it to rise, mals would be a change in migration timing (e.g. migrate sometimes to thousands of metres (type I thermal cells; earlier in calendar year), especially relative to energy-lim- Hardy & Ottersten 1969; Kerlinger 1989). Orographic lift ited migrants. Animals that conserve energy during migra- depends upon upward deflection of horizontal winds by tion should migrate only during conditions that promote structures, such as trees or ridgelines, and occurs at rela- the most efficient flight modes (i.e. thermal soaring) and tively low altitudes above ground level (Kerlinger 1989; pause migration when less optimal conditions exist (i.e. Alerstam & Hedenstrom 1998). Other flight modes that not use orographic soaring). Pausing migration to wait for depend on supplemental lift are also weather dependent optimal weather would increase the total number of days and include dynamic soaring (e.g. Sachs et al. 2012) and spent in migration while decreasing the proportion of time use of thermal streets (thermals aligned such that circling spent migrating each season. is not needed to regain lift; type II thermal cells; Hardy & Golden eagles (Aquila chrysaetos) in eastern North Ottersten 1969; Kerlinger 1989). Although general weather America are a useful model both to understand soaring patterns associated with each flight mode are understood, flight responses to changes in weather conditions during knowledge is nascent of how soaring birds respond to migration and to test hypotheses about time- and energy- variation in weather. limited migration. This species is the largest soaring Understanding sources of variation in flight behaviour migrant in eastern North America (Allen, Goodrich & requires pairing weather conditions with animal locations Bildstein 1996), and they take advantage of a variety of lift throughout flight (Shamoun-Baranes, Bouten & van Loon types (Duerr et al. 2012; Lanzone et al. 2012). Lift types 2010; Mandel et al. 2011). Soaring birds show different used by eagles may vary throughout their migratory range responses to weather conditions in different regions, which as they cross several ecological regions with marked differ- show how continental-scale differences in weather affect ences in topography (Commission for Environmental soaring (Shamoun-Baranes et al. 2003; Klaassen et al. Cooperation 1997) from the Appalachian Mountains to 2011; Mandel et al. 2011). Studies that pair animal loca- Quebec, Canada (Katzner et al. 2012b). Golden eagle pop- tions recorded at high frequencies (up to two locations per ulations are age structured, with breeding generally minute) with high-resolution environmental data sets pro- delayed until at least their fifth year (Watson 2010); thus, vide detailed understanding of variation of flight modes. we expect that individuals of different ages will have differ- For example, flight modes shift from thermal soaring and ent levels of experience and be under different time and gliding to orographic soaring as wind speeds increase energy constraints. (Lanzone et al. 2012). We also know that for some species, We tested hypotheses about time- and energy-limited thermal soaring and gliding is faster, more direct and ener- migration of golden eagles with a several step (objectives) getically more efficient than orographic soaring (Duerr process. To do this, we first (Objective 1) determined mete- et al. 2012). Observations of active selection of one of orological correlates of migration and described how these these flight modes over the other suggest soaring birds correlates change with season, eagle age and region. Mete- may make trade-offs during flight in response to ecological orological correlates allow inference about flight modes and evolutionary constraints. used by eagles. Second, we placed meteorological corre- Birds face age-specific time and energy constraints that lates into the context of selection of migratory conditions may be especially apparent during migration (Hedenstrom and flight modes by (Objective 2) describing prevailing 1993; Hedenstrom & Alerstam 1995, 1997). Adults often weather patterns experienced by golden eagles in spring are time limited as they face strong selective pressures to and fall. Third, we (Objective 3) determined how the time arrive at breeding sites early, so they can acquire and span of migration (number of days spent migrating), pro- defend territories (Sergio et al. 2007; Newton 2008); in portion of season spent in active migration and timing of

© 2014 The Authors. Functional Ecology © 2014 British Ecological Society, Functional Ecology Responses of a soaring bird to thermal generation 3 migration differed by eagle age and season. Addressing Bildstein 2007) and released. Both types of telemetry collect GPS these objectives provides ecological and evolutionary con- data, but CTT-1100s transmit data through the Global System for texts for differences in migratory behaviour that we identify. Mobile Communication (GSM) network, whereas platform terminal transmitters (PTTs) transmit data through ARGOS weather satellites. Cellular terminal transmitters (CTT) transmitters collected data at 15-min intervals, and PTT transmitters collected Materials and methods data at 1-h intervals. We removed data points when CTTs recorded GPS locations, but other telemetry data (e.g. heading, STUDY AREA altitude, speed) were not recorded. We collated and analysed eagle location data from spring and fall migration from November 2006 Golden eagles are found in five ecoregions of eastern North Amer- through December 2012 (Fig. 1). ica (Fig. 1; (Commission for Environmental Cooperation 1997; Katzner et al. 2012b). Their winter range includes mainly eastern temperate forest, despite some winter in northern forest, and their DATA CLASSIFICATION breeding range includes northern forest, taiga, tundra and Arctic We classified each eagle location first by migration period and cordillera. Eastern temperate forest is dominated by mixed broad- then as part of migratory or non-migratory flights. We identified leaf and evergreen forests. It has a mild climate with warm humid migration periods as the series of GPS data points from each indi- summers and cool to mild winters and a variety of landforms that vidual eagle that connected wintering and breeding ranges. Golden range from the Appalachian Mountains, large plateaus, ridge and eagles in eastern North America have distinct wintering and valley systems and coastal plains (Commission for Environmental breeding territories that encompass their southernmost and north- Cooperation 1997). Northern forest is dominated by cover of bor- ernmost GPS locations (Miller 2012). Within each migration per- eal evergreen trees; it has many lakes, long cold winters and short iod, we grouped locations from the start to the end of each warm summers and is characterized by hilly terrain of the Cana- nominal hour (e.g. 08.00–09.00) and calculated the average tag- dian Shield interspersed with shallow to deep moraine deposits. recorded flight speed for each nominal hour. ‘Migratory’ locations Taiga is dominated by rolling uplands and lowlands and has long À included average hourly flight speeds >10 km h 1, and ‘non- and very cold winters and short cool summers, and vegetation var- À migratory’ locations included flight speeds <5kmh 1 (Katzner ies from bogs, wetlands and boreal forests to open shrublands and et al. 2012a). Hours in which eagles averaged speeds between 5 sedge meadows. Tundra, with low forms of vegetation and similar À and 10 km h 1 were classified as intermediate locations. Thus, on topography to taiga, and Arctic cordillera, with bare rocky moun- a given day, GPS locations from a single eagle could have been tains, both have similar climate to taiga. classified as migratory, non-migratory and intermediate locations. We used migration classifications of eagle locations to calculate DATA COLLECTION FROM EAGLES three measures of time related to migration. The first was time span of migration, which was calculated as the number of days We used cannon or rocket nets and bow nets set over bait to cap- between the first and last day of migration. The second was the ture 47 golden eagles during migration and winter November– proportion of season spent migrating, calculated as the number of March 2006–2012 in Pennsylvania, Virginia and West Virginia, migratory GPS locations divided by the total number of GPS USA and in May 2009 at a nest on the Gaspe Peninsula, Quebec, locations (flying and perched, daytime and night-time) in a migra- Canada. Each eagle was aged by plumage and moult characteris- tion season. The third, the migration mid-point, was a measure of tics (Jollie 1947; Bloom & Clark 2001) and classified as adult (less when eagles migrated and was calculated as the median calendar than or equal to age 5) or pre-adult (less than age 5). Eagles were day from the time span of migration. outfitted with Cellular Tracking Technologies CTT-1100 GPS- At each eagle location, we used the RNCEP package (Kemp GSM (CTT, Somerset, PA, USA; n = 44) or Microwave Teleme- et al. 2011) in R (R Core Team 2012) to interpolate weather try GPS PTT-100s (PTT, Columbia, MD, USA; n = 3) telemetry conditions over space and time from the National Centers for systems attached as backpacks with Teflon ribbon (Bird & Environmental Prediction/National Centers for Atmospheric

(a) (b) (c)

Fig. 1. Study area depicting (a) elevation and major geologic features and (b) spring and (c) fall locations of golden eagles as they migrated across ecoregions of eastern North America, March 2006 to December 2012. Northern ecoregions (Arctic cordillera, tundra and taiga) are cross-hatched and were combined for analyses.

Research 40-year reanalysis project (Reanalysis II; Kalnay et al. Due to the limited numbers of telemetry locations in tundra and 1996). Reanalysis II is a model that predicts values of weather Arctic cordillera, we combined these ecoregions with taiga and variables (Table 1) at the surface and at four pressure levels above refer to them as northern ecoregions (Fig. 1). the surface (1000, 925, 850 and 700 mb) (see Appendix S1 in Sup- porting Information for definitions of select weather variables). At each eagle location, we modified wind variables by including the CORRELATIONS AMONG METEROLOGICAL VARIABLES absolute values of west–east and south–north wind vectors, wind speed (i.e. without a directional component) and a modified value Meteorological variables are often either positively or negatively of thermal energy (D °C per 100 m; Chevallier et al. 2010), in correlated (e.g. downward solar radiation should be negatively which we substituted geopotential height of pressure levels for correlated with cloud cover). We included only weather variables < Á height of pressure levels in the standard atmosphere. that were not correlated (|R| 0 5) with others in our analyses. Calculating weather conditions at the flight altitude of eagles When variables were correlated, we retained the weather variable required an additional interpolation step because flight altitudes of that we expected would have provided the clearest ecological inter- eagles rarely corresponded exactly with altitudes (geopotential pretation if associated with migration. height) of pressure levels in the Reanalysis II data set. We linearly interpolated weather variables between geopotential height of MIGRATION RESPONSE MODELS pressure levels and the eagle’s flight altitude as recorded by the telemetry unit. If the eagle flew below the lowest pressure level We developed migration response models (so termed to distin- (1000 mb), we interpolated between values of weather variables at guish from other analyses) to determine meteorological correlates 1000 mb and at the earth’s surface. Surface elevation was deter- of migration (Objective 1). We built these models separately for mined from the global Advanced Spaceborne Thermal Emission spring and fall, as prevailing weather patterns differ by season and Reflection Radiometer (ASTER) digital elevation model (pro- (Kerlinger & Moore 1989). These models included a binary duced jointly by US National Aeronautics and Space Administra- response variable of migratory flight or non-migratory flight; we tion and Japan’s Ministry of Economy, Trade and Industry; did not consider intermediate points in this analysis. Weather con- http://gdex.cr.usgs.gov/gdex/). ditions (Table 1) were explanatory variables. We used generalized We also classified eagle locations by the level I ecoregion (Com- linear mixed models (GLMMs) with the binomial distribution, mission for Environmental Cooperation 1997) in which they logit link and empirical sandwich estimators of covariance (PROC RC occurred using A GIS 10.1 (Esri [SIC], Redlands, CA, USA). GLIMMIX; SAS 9.3, Cary, NC, USA). Fixed effects that we included were weather variables, eagle age, ecoregion and interactions of age and ecoregion with weather variables. We included two ran- Table 1. Weather variables from NCEP/NCAR Reanalysis II dom effects in each model. These were a term for eagle individual data set interpolated to space and time of golden eagle locations that accounted for correlation among repeated measures on the in northeast North America during migration, March 2006– same individual and a single-year/month term that accounted for December 2012 temporal autocorrelation. We accounted for multiple comparisons within models using false discovery rates (Benjamini & Hochberg Measurement 1995; Garcıa 2004; Pike 2011). To assess the magnitude of weather Weather variable units Atmospheric level effects on eagle migration, we calculated odds ratios for significant

effects. Odds of an eagle migrating (OA) during average weather Air temperature °C Surface, 1000, 925, conditions were calculated by applying average values of weather 850, 700 mb variables to the solutions for fixed effects (beta estimates) from the Barometric pressure* mm mercury Surface linear model. We also calculated the odds of an eagle migrating ° Best-lifted index K Surface (OS) when the significant weather effect was increased by 1 SD À2 Downward solar Wm Surface and all other weather variables were held at seasonal averages. † radiation flux The odds ratio for each significant effect was OS/OA and was Geopotential height m 1000, 925, 850, 700 mb interpreted as the multiplicative change in odds of migrating. † À2 Latent heat flux Wm Surface We developed three alternative migration response models for À1 Omega (vertical Pascal s Surface, 1000, 925, each of the two migration seasons (spring and fall, six models velocity) 850, 700 mb total). Our three models in each season incorporated weather vari- À2 Precipitable water kg m Surface to 300 mb ables from different altitudes above ground level, to understand À2 À1 Precipitation kg m s Surface eagle response to weather at the earth’s surface, at eagle flight alti- † rate (>0) tude, or somewhere in between. Each of the three alternatives used Relative humidity % Surface, 1000, 925, a set of weather variables (temperature, omega, relative humidity, 850, 700 mb south–north wind vector, west–east wind vector, thermal energy † À2 Sensible heat flux Wm Surface and modified wind variables) from different pressure levels (see À1 South–north wind ms Surface, 1000, 925, 850, Table 1). In the first model, all weather variables were from the vector 700 mb earth’s surface. In the second model, the set of variables were all À2 Water equivalent kg m Surface from the 925-mb pressure level. In the third model, measurements of snow of the set of variables were interpolated linearly to the eagle’s À1 West–east wind ms Surface, 1000, 925, 850, flight altitude. Other variables from the Reanalysis II data set vector 700 mb were from the earth’s surface for all models (Table 1). We used ‡ À1 Thermal energy °C 100 m Average from 1000, 925, pseudo-Akaike’s information criteria (PAIC) to compare the three 850, 700 mb alternative models in separate model sets for spring and fall.

*Units for barometric pressure converted from Pascals. † Variables reported as 6 h averages and not interpolated over SEASONAL WEATHER COMPARISON MODELS time. ‡Modiﬁcation from Chevallier et al. (2010) using geopotential We developed seasonal weather comparison models to determine height at pressure levels instead of from standard atmosphere. whether and how weather encountered by eagles diﬀered by

© 2014 The Authors. Functional Ecology © 2014 British Ecological Society, Functional Ecology Responses of a soaring bird to thermal generation 5 season (Objective 2). In these models, we only considered those weather variables that were associated with migration in spring or Results in fall (as determined in Objective 1). When an interaction with age or ecoregion entered migration response models for either DATA COLLECTED FROM EAGLES spring or fall (Objective 1), we also included these interactions in our seasonal weather comparison models (Objective 2). We com- We tracked 47 eagles moving along 83 migration tracks bined migratory, intermediate and non-migratory locations to (Table 2). Telemetry devices collected 87 029 GPS loca- define weather conditions encountered by eagles in spring and in tions transmitted from CTT/GSM systems (n = 44) and fall. Season and season–age and season–ecoregion interactions 2233 GPS locations transmitted from PTT/ARGOS sys- were explanatory variables. We used GLMMs with a Gaussian = distribution, identity link and empirical sandwich estimators of tems (n 3). Birds contributed data exclusively for spring covariance (PROC GLIMMIX; SAS 9.3). We included explanatory vari- migration (n = 27), exclusively for fall migration (n = 1) or ables as fixed effects. Random effects were eagle individual and for both seasons (n = 19). Eagle ages included 20 pre- month/year term. adults, 24 adults and three eagles that provided data as both a pre-adult and an adult. TIME RESPONSE MODELS

We developed three sets of time response models to determine DATA CLASSIFICATION whether and how migration timing differed by age or season. = Measures of time were time span of migration (number of days), In spring, CTTs (n 44 birds) transmitted 44 680 GPS proportion of season spent migrating (proportion of GPS loca- points from non-migratory locations (flight speed tions) and migration mid-point (calendar day), and each included <5kmhÀ1), 19 617 points from migratory locations (flight migratory, intermediate and non-migratory locations. speed >10 km hÀ1) and 2418 points from intermediate The first model compared time span of migration between À (flight speed 5–10 km h 1) locations. PTTs (n = 3 birds) adults and pre-adults in spring and in fall. The response variable was time span of migration, and explanatory variables were age, transmitted 372 non-migratory, 226 migratory and 29 season and age–season interaction. We used GLMMs with a intermediate points. In fall, CTTs (n = 17 birds) transmit- Gaussian distribution, the identity link and empirical sandwich ted 13 484 non-migratory, 5688 migratory and 778 inter- estimators of covariance (PROC GLIMMIX; SAS 9.3). We included mediate locations and PTTs (n = 3 birds) transmitted explanatory variables as fixed effects and random effects of eagle 1248, 260 and 98 points from non-migratory, migratory individual and year. The second time response model compared proportion of sea- and intermediate locations, respectively. son spent migrating between adults and pre-adults in spring and in fall. The response variable was proportion of season spent CORRELATIONS AMONG METEOROLOGICAL migrating and explanatory variables were age, season and age– season interaction. We used GLMMs with a binomial distribu- VARIABLES tion, logit link and empirical sandwich estimators of covariance (PROC GLIMMIX; SAS 9.3). We included explanatory variables as Several weather variables were correlated. Downward solar fixed effects and random effects of eagle individual and year. radiation was correlated with sensible heat flux and latent The third time response model compared the migration mid- heat flux (r > 0Á57). We retained downward solar radiation point between adult and pre-adult eagles. The response variable and excluded sensible heat flux and latent heat flux as was migration mid-point, and the explanatory variable was eagle downward solar radiation is a complete measure of inci- age. We built separate models for spring and fall as calendar days differ between seasons. We used GLMMs, a Gaussian distribu- dent solar radiation onto the earth’s surface, whereas sen- tion, identity link and empirical sandwich estimators of covariance sible and latent heat fluxes are partial measures of energy (PROC GLIMMIX; SAS 9.3). The fixed effect was age, and random transferred back to the atmosphere from the ground. Best- effects were eagle individual and year. lifted index was correlated with precipitable water

Table 2. Number of migratory golden eagles, migration tracks and global positioning system locations recorded by cellular terminal transmitters (CTT) and platform terminal transmitters (PTT) from north-eastern North America, March 2006 to December 2012

Number of locations by ecoregion

Transmitter Number of Number of Eastern Northern Northern Season Age type individuals tracks temperate forest forest regions*

Spring Adult CTT 24 25 6990 11 549 2602 PTT 2 3 201 159 64 Pre-adult CTT 22 28 9976 20 856 14 742 PTT 1 1 0 92 111 Fall Adult CTT 8 10 1687 4025 1749 PTT 3 5 421 1000 185 Pre-adult CTT 9 11 2499 7115 3239 PTT 0 0 0 0 0

*Taiga, tundra and Arctic cordillera were combined into northern regions.

(r = À0Á73) and air temperatures at all but the highest Table 3. Migration response model. Spring weather variables and pressure level (r < À0Á69). We included best-lifted index interactions with eagle age and ecoregion used in generalized lin- and excluded precipitable water and air temperatures ear mixed model (logit link) to compare locations of migratory and non-migratory movements for golden eagles in eastern North because best-lifted index is a measure of atmospheric sta- America, March 2006 to December 2012. Weather variables meability that more directly relates to lift experienced by birds sured at the earth’s surface fit data better than variables measured than either precipitable water or temperature. The modi- at eagle flight altitude (DPAIC = 508) or at 925 mb pressure level fied wind variables were correlated (i.e. wind speed was (DPAIC = 4579). We used false discovery rates to correct for mul- correlated with absolute values of both the south–north tiple comparisons and identify variables associated with migration (significant P-values are bold) wind vector and the west–east wind vector; R > 0Á64). We retained absolute values of south–north and west–east vec- Effect F-value P tors of wind as these provided detail on wind effects along two axes. Unmodified wind variables (south–north and Downward solar radiation 252Á46 <0Á0001 Thermal energy 74Á25 <0Á0001 west–east wind) were not correlated with other weather Barometric pressure 9 Ecoregion 12Á74 <0Á0001 variables and were retained in all models. Ecoregion 9Á59 <0Á0001 South–north wind vector 9Á5 0Á0021 Thermal energy 9 Age 8Á56 0Á0034 MIGRATION RESPONSE MODELS South–north wind vector 9 Age 5Á13 0Á0235 Relative humidity 9 Ecoregion 5Á05 0Á0064 The best migration model for spring included variables Precipitation rate 9 Ecoregion 4Á8 0Á0083 measured only at the earth’s surface (Table 3), while the Downward solar radiation 9 Ecoregion 3Á62 0Á0269 best fall migration model included measurements at the Downward solar radiation 9 Age 4Á58 0Á0324 925-mb pressure level (mean altitude above ground level Thermal energy 9 Ecoregion 3Á09 0Á0454 for the 925 mb pressure level during fall was 325 m Æ 222; Best-lifted index 9 Ecoregion 3Á03 0Á0484 West–east wind vector 9 Age 2Á31 0Á1283 SD). Fit of the data for alternative models in spring was Best-lifted index 2Á09 0Á148 poor when weather variables were measured at eagle flight Age 1Á93 0Á1647 altitude (DPAIC = 508) or at the 925-mb pressure level Precipitation rate 1Á77 0Á1835 (DPAIC = 4579). Likewise, fit of the data for alternative Barometric pressure 9 Age 1Á59 0Á207 fall models was poor when weather was measured at the Omega 1Á47 0Á2247 Water equivalent of snow 9 Age 1Á26 0Á2617 earth’s surface (DPAIC = 1015) or at eagle flight altitude West–east wind vector 9 Ecoregion 1Á06 0Á3468 D = ( PAIC 1847). Alternative models for spring and fall Omega 9 Ecoregion 1Á04 0Á3543 are not presented. Relative humidity 9 Age 0Á83 0Á3614 The most important factors that influenced migration Water equivalent of snow 0Á82 0Á3664 of golden eagles in spring and fall were those associated Omega 9 Age 0Á73 0Á3934 |South–north wind vector| 9 Ecoregion 0Á68 0Á5058 with increased thermal formation – downward solar radi- Barometric pressure 0Á54 0Á464 ation and thermal energy (ranked 1 and 2 in spring and |West–east wind vector| 9 Ecoregion 0Á50Á6074 fall migration response models; Tables 3 and 4; Fig. 2; |West–east wind vector| 0Á39 0Á5301 Tables S1 and S2). Furthermore, this relationship was Water equivalent of snow 9 Ecoregion 0Á37 0Á6939 always positive, indicating that probability of an eagle |West–east wind vector| 9 Age 0Á33 0Á5666 Precipitation rate 9 Age 0Á28 0Á5939 being recorded as engaged in migratory flight in a given Relative humidity 0Á23 0Á6297 hour (hereafter probability of migration) increased as South–north wind vector 9 Ecoregion 0Á21 0Á8111 downward solar radiation and thermal energy increased. Best-lifted index 9 Age 0Á18 0Á6748 This pattern held regardless of season, age or ecoregion |South–north wind vector| 0Á15 0Á7004 (Tables 3–5). |South–north wind vector| 9 Age 0Á13 0Á7227 West–east wind vector 0 0Á9886 In spring, adult and pre-adult eagles responded differently to variation in weather (age interactions in migration PAIC, pseudo-Akaike’s information criteria. response models; Table 3). Although probability of migration increased with increasing values of the south–north wind vector and increasing thermal energy, these responses probability of migration was negatively correlated with were stronger for pre-adults than for adults (Table 5). barometric pressure. These responses suggest that in spring, pre-adults track Eagles also responded differently to relative humidity these weather conditions more tightly than do adults and precipitation rates in each ecoregion. As humidity lev- (Table 5). els increased, probability of migration by eagles increased Eagles responded to certain spring weather conditions in northern regions but decreased in northern and eastern differently in each ecoregion through which they migrated temperate forests. In temperate forests, eagles increased (ecoregion interactions in migration response models; probability of migration as precipitation rates increased, Table 3). Probability of migratory flight was positively which contradicts with the response to relative humidity. correlated with barometric pressure in eastern temperate In other ecoregions, probability of migration decreased as forests and northern forests. In northern regions in spring, precipitation increased.

Table 4. Migration response model. Fall weather variables and models; Tables 4 and 5). Golden eagles increased probabil- interactions with eagle age and ecoregion used in generalized lin- ity of migration as the west–east wind vector measured at ear mixed model (logit link) to compare locations of migratory the 925-mb pressure level decreased in eastern temperate and non-migratory movements for golden eagles in eastern North America, March 2006 to December 2012. Weather variables mea- forests and northern regions, but the opposite was true in sured at the 925-mb pressure level fit data better than variables northern forests. Probability of migration varied by ecore- measured at eagle flight altitude (DPAIC = 1847) or at the earth’s gion and age for barometric pressure. In fall, adult eagles surface (DPAIC = 1015). We used false discovery rates to correct increased probability of migration as barometric pressure for multiple comparisons and identify variables associated with increased for all ecoregions, with the greatest effect migration (significant P-values are bold) observed in northern forests. Probability of pre-adult Effect F-value P migration increased with decreasing pressure in all ecoregions, with the magnitude of this effect increasing as they Thermal energy 39Á82 <0Á0001 migrated south. Downward solar radiation 32Á32 <0Á0001 |South–north wind vector at 925 mb| 9Á58 0Á0020 Age 9Á08 0Á0026 SEASONAL WEATHER COMPARISON MODELS Barometric pressure 9 Age 8Á57 0Á0034 Best-lifted index 8Á19 0Á0042 Certain weather conditions that eagles encountered and Ecoregion 7Á36 0Á0006 that were identified as important from migration response Omega at 925 mb 6Á76 0Á0093 models differed between seasons (Table 6). Eagles encoun- Barometric pressure 9 Ecoregion 6Á23 0Á002 West–east wind vector at 4Á45 0Á0117 tered more than twice the levels of downward solar 925 mb 9 Ecoregion radiation in spring than in fall (Tables 6 and 7; Fig. 2). West–east wind vector at 925 mb 4Á37 0Á0365 Best-lifted index encountered by eagles was twice as high in Precipitation rate 3Á82 0Á0507 fall as in spring (atmosphere more stable in fall) and omega Relative humidity at 925 mb 9 Age 3Á80Á0514 measured at 925 mb was slightly higher in fall than in Thermal energy 9 Ecoregion 2Á77 0Á0625 – Water equivalent of snow 2Á72 0Á0992 spring. The south north wind vector encountered by eagles |South–north wind vector at 2Á63 0Á0722 was greater in fall in spring. Seasonal differences in the 925 mb| 9 Ecoregion south–north wind vector corresponded to shifts in prevail- Precipitation rate 9 Ecoregion 2Á41 0Á0897 ing wind patterns (Table 7; Fig. 2) with predominant winds |West–east wind vector at 2Á37 0Á0932 from the north in spring and from the south in fall. 925 mb| 9 Ecoregion Relative humidity at 925 mb 2Á26 0Á1325 Other weather conditions experienced by eagles differed Downward solar radiation 9 Ecoregion 1Á89 0Á1505 among ecoregions by season (season–ecoregion interac- Downward solar radiation 9 Age 1Á76 0Á1845 tions), but did not differ seasonally between adults and South–north wind vector at 1Á51 0Á2207 pre-adults (season–age interactions; Table 6). Eagles 925 mb 9 Ecoregion encountered lower precipitation rates in eastern temperate Omega at 925 mb 9 Ecoregion 1Á30Á2719 Barometric pressure 0Á77 0Á3807 forests and northern regions, but higher precipitation rates Water equivalent of snow 9 Ecoregion 0Á73 0Á4828 in northern forests in fall compared to spring. The west– |South–north wind vector at 925 mb| 9 Age 0Á64 0Á4235 east wind vector measured at the 925-mb pressure differed Omega at 925 mb 9 Age 0Á63 0Á4288 in magnitude between seasons and among ecoregions; Water equivalent of snow 9 Age 0Á60Á44 however, prevailing wind in all ecoregions and during both Precipitation rate 9 Age 0Á46 0Á4988 |West–east wind vector at 925 mb| 9 Age 0Á29 0Á5925 seasons was from the west (Fig. 2). Barometric pressure Best-lifted index 9 Ecoregion 0Á27 0Á7642 experienced by tagged eagles was higher in fall than in |West–east wind vector at 925 mb| 0Á22 0Á6406 spring in forested regions but lower in fall than spring in Best-lifted index 9 Age 0Á20Á6529 northern regions. Relative humidity at 925 mb 9 Ecoregion 0Á19 0Á8254 South–north wind vector at 925 mb 0Á08 0Á774 West–east wind vector at 925 mb 9 Age 0Á04 0Á8362 TIME RESPONSE MODELS Thermal energy 9 Age 0Á03 0Á8644 South–north wind vector at 925 mb 9 Age 0 0Á9962 Time span of migration (number of days from start to end of migration) differed by season (first time response model; PAIC, pseudo-Akaike’s information criteria. F1,32 = 11Á07, P = 0Á002) but not by age (F1,32 = 1Á84, P = 0Á185) or by the age–season interaction (F1,32 = 0Á03, In fall, along with downward solar radiation and ther- P = 0Á870). Eagles spent less time migrating in spring mal energy, probability of migratory flight increased as (26Á1 Æ 2Á6, SE, days) than in fall (70Á9 Æ 17Á3 days). best-lifted index (atmospheric stability), omega (vertical The proportion of season spent migrating (proportion velocity) measured at 925 mb and absolute value of the of the GPS locations classified as migratory) differed by – vector for south north wind at 925 mb increased (Tables 4 age (first time response model; F1,50 = 5Á43, P = 0Á02) and and 5). season (F1,50 = 25Á62, P < 0Á001), but not by the age–sea-

Fall migration diﬀered by ecoregion for two weather son interaction (F1,50 = 2Á45, P = 0Á124). Adults spent a variables (ecoregion interactions in migration response greater proportion of each season migrating that did

(a) (b)

Fig. 2. Average weekly values (SE bars) of weather variables for 2012 that differed for migratory (dark triangles) and non-migratory locations (grey circles) or between seasons (spring February to July, fall September to January) for golden eagles migrating in eastern North America. Downward solar radiation was measured as 6 h averages, while thermal energy and wind variables were interpolated to times when eagle locations were recorded (15-min to 1-h intervals). For downward solar radiation (a) migration > non-migratory locations and spring > fall (see Tables 3–5). For thermal energy (b) migratory > non-migratory locations. Prevailing winds measured by the west–east wind vector (c) were from the west regardless of season. Winds measured by the south–north wind vector (d) were greater in spring than fall, with a shift in prevailing direction between seasons. pre-adults (both seasons combined, adults 0Á296 Æ 0Á016, birds use orographic lift to travel along ridgelines (Duerr SE; pre-adults 0Á241 Æ 0Á014). Eagles spent a greater pro- et al. 2012). The response of eagles choosing to migrate portion of each season migrating in spring (0Á314 Æ 0Á013) preferentially when conditions existed for thermal develop- than in fall (0Á210 Æ 0Á013). ment shows that eagles were seeking to maximize flight Migration mid-point (middle day of time span of migra- efficiency and minimize energetic expenditures whenever tion) differed by age in spring (third time response model; possible. However, migratory responses to weather condi-

F1,9 = 52Á08, P < 0Á001) but not fall (F1,4 = 1Á28, tions that promote orographic lift differed by age and sea- P = 0Á321; Fig. 3). During spring, median calendar day for son. This contrast highlights the contradictory temporal adult migration was 82 Æ 2 (SE) days (22 March) and and energetic constraints faced by eagles of different ages 116 Æ 4 days (25 April) for pre-adults. During fall migra- and different reproductive statuses. tion, median calendar day for migration was 315 Æ 5 days Many breeding raptors are time limited during spring (10 November). migration (Sergio et al. 2007; Newton 2008). Timing of migration in spring is constrained by age-specific breeding behaviour. During spring, adult golden eagles migrated Discussion earlier, when solar radiation levels were lower, and they Migratory flight of a model soaring migrant was positively showed a weaker response (measured as increased proba- associated with weather conditions that promoted thermal bility of engaging in migration) to thermal energy than did lift. This response was prominent during both spring and later migrating pre-adults. Adults migrated when condi- fall migration, across all ecoregions, and for all age classes. tions were conducive to both thermal lift and less efficient Migration is faster and more direct when birds soar and orographic lift, and a greater proportion of their season glide in and out of thermals, and migration is slower when was spent migrating, compared to pre-adults. These

Table 5. Odds ratios are multiplicative changes in the odds of patterns indicate time limitation and a potential advantage migration of golden eagles during spring or fall due to an increase to adults to return to northern breeding territories as in the value of that variable by 1 SD from migration response quickly as possible. In contrast, pre-adult migration in models. Eagle data were collected in eastern North America from March 2006 to December 2012 spring was correlated with increasing thermal energy and south to north winds (i.e. tailwinds), both of which con- Odds ratios tribute to migratory efficiency. Pre-adults also spent a smaller proportion of their season migrating, but the time Effect Spring Fall span of migration (number of days) was greater, suggest- Downward solar 2Á89 1Á26 ing that they waited for weather conditions that promoted radiation efficient flight modes. These two factors together provide Thermal energy 1Á69 strong evidence that pre-adults, in contrast to adults, act – Á |South north wind vector 1 26 to minimize energetic expenditures during spring migra- at 925 mb| tion. Best-lifted index 1Á15 Omega at 925 mb 1Á15 Evaluating temporal and energetic limitations is rela- tively more complex during fall migration. Adult eagles Age interactions Adult Pre-adult typically begin migration in October and November, after – South–north wind 1Á00 1Á26 successful breeders complete the 18 20-week breeding cycle vector 9 Age (Watson 2010). Breeding activities did not delay south- Thermal energy 9 Age 1Á19 1Á62 bound migration by adults, and there was no difference in Ecoregion interactions ETF NF NR ETF NF NR time span that adults and pre-adults spent in migration (although adults appeared to spend a greater proportion Precipitation 1Á04 0Á73 0Á99 of their season migrating in both spring and fall than did 9 rate Ecoregion pre-adults). Thus, adults were not time limited. Adults Barometric pressure 1Á34 1Á28 0Á85 encountered lower levels of downward solar radiation in 9 Ecoregion Relative humidity 0Á86 0Á98 1Á46 fall than in spring. This likely reduced their migratory effi- 9 Ecoregion ciency by forcing them to use orographic soaring or flap- West–east wind vector at 0Á63 1Á07 0Á90 ping flight. However, they were more likely to migrate as 9 925 mb Ecoregion barometric pressure increased, when type I thermals Age and ecoregion develop (Kerlinger 1989), and solar radiation and thermal interactions ETF NF NR ETF NF TAI energy are on the rise. Therefore, adults appeared to have migrated more efficiently than pre-adults and were not Barometric pressure 1Á01 1Á33 1Á02 time limited during fall migration. 9 Ecoregion (adult)* Barometric pressure 0Á64 0Á84 0Á98 In fall, pre-adults migrated during conditions of lower 9 Ecoregion (pre-adult) barometric pressure than adults, which better corresponds to the development of orographic lift than of thermal lift. ETF, Eastern temperate forest; NF, Northern forest; NR, north- They appeared to have used less efficient lift types and ern regions (Taiga/tundra/Arctic cordillera combined) are ecoregion-specific odds ratios. flight modes than did adults. This pattern might suggest *Adult and pre-adult are age-specific odds ratios. pre-adults were time limited. However, there was no

Table 6. Tests to determine whether weather variables (including interaction terms) that were associated with migratory movements of golden eagles diﬀered between spring and fall from seasonal weather comparison models. Data were collected in eastern North America, March 2006 to December 2012. We used false discovery rates to correct for multiple comparisons (signiﬁcant P-values are bold)

Variable Eﬀect F-value d.f. P

Downward solar radiation Season 296Á77 1 <0Á001 Precipitation rate Season 9 Ecoregion 12Á62<0Á001 Omega at 925 mb Season 9Á71 0Á002 South–north wind vector Season 5Á61 0Á018 West–east wind vector at 925 mb Season 9 Ecoregion 5Á51 2 0Á004 Best-lifted index Season 5Á38 1 0Á020 Barometric pressure 9 ecoregion Season 9 Ecoregion 5Á32 0Á005 Thermal energy Season 3Á49 1 0Á062 Thermal energy Season 9 Age 2Á11 0Á147 Relative humidity Season 9 Ecoregion 2Á08 2 0Á125 |South–north wind vector at 925 mb| Season 1Á27 1 0Á260 South–north wind vector Season 9 Age 0Á46 1 0Á499 Barometric pressure Season 9 Age 0Á36 1 0Á546

© 2014 The Authors. Functional Ecology © 2014 British Ecological Society, Functional Ecology 10 A. E. Duerr et al. 88 Á 078 74 92 Á Á Á 129 0 3 4 26 Á 0000823 56 Á Á Æ Æ Æ Æ 6 0 10 63 33 18 Æ Æ Æ Á Á Á 027 Á 1 Á 65 Fall Á ng and fall migra- 0000292 Á 43 724 Á 44 4 0000825 0 Á Á Fig. 3. The per cent of GPS tagged golden eagles that were 6 0 10 Æ Æ Æ migrating for each day of the year. This illustrates the timing of spring (February to July) and fall (August to January) migration 93 01 Á Á for adult and pre-adult ages. Eagles were tracked from March 2006 to December 2012 in eastern North America. 0000334 Á

difference in time span of migration or migration mid- point between adults and pre-adults. That pre-adults used 91 000077 0 76 730 Á Á Á less efficient flight modes but took similar amounts of time 5 0 5 Æ Æ Æ as adults suggest that pre-adults responded to similar tem- 8 Á 65 poral pressures as adults but were simply less efficient Á migrants than were adults. 0000151 Á In spite of the near-universal response to thermal lift, adult and pre-adult eagles showed different responses to temporal and energetic constraints that reflected seasonal 72 162 Á differences in the weather conditions that they encountered 07 0 78 0 73 11 81 736 Á Á Á Á 45 3 000056 0 236 0 3 5 Á Á 5 0 12 in spring but not in fall. In spring, levels of downward Æ Æ Æ Æ Æ Æ Æ solar radiation were lower when adults migrated than 78 01 26 44 Á Á Á Á 56 77 0 0 9 Á Á when pre-adults migrated. The lower solar radiation À Spring encountered by adults may be a consequence of differences 0000243 Á in weather relative to conditions when pre-adults migrate, or it may be an artefact of shorter day length when adults migrate. In either case, lower levels of solar radiation corresponded to a weaker migratory response to thermal 32 1 0000688 0 16 730 Á Á Á energy in spring by time-limited adults than by pre-adults. 5 0 9 Æ Æ Æ In contrast, pre-adults showed a greater dependence upon 33 41 Á Á orographic lift in fall than did adults, which reflected prevailing weather conditions during that period. 0000307 Á Seasonal differences in weather showed that eagles encounter less thermal development in fall than in spring throughout migration. The subsidized lift that fall-migrating eagles use primarily occurred in the form of orographic 15 2 0001009 0 43 728 Á Á Á lift. Probability of migration in the fall increased as winds 5 0 6 along the south–north vector became stronger (i.e. greater Æ Æ Æ 6 Á

22 north or greater south winds measured at 925 mb). Á Migrating into headwinds may seem counterintuitive, but 0000414 Á because of the topography along the migration route (e.g. in New York and eastern Pennsylvania), such winds are

SD) of weather variables that were associated with migration (identified in migration response models) of golden eagles and that differed between spri frequently deflected upward along ridgelines and provide Æ supplemental lift (Kerlinger 1989). Although not well doc- umented, birds also may use dynamic soaring to gain altitude over land; this form of lift also requires turning into the wind (Kerlinger 1989; Sachs & Mayrhofer 2001). The combination of these multiple forms of subsidized lift cre- Average values ( ates a system where each lift type could be used by eagles individually or together to provide a synergistic effect to Downward solar radiation 346 Omega at 925 mb West to east wind at 925 mb 3 South to north wind vector Best-lifted index Precipitation rate 0 Barometric pressure 733 Table 7. tions (from seasonal weather comparisons) Variable Ecoregion interactions ETF NF NR ETF NF NR ETF, Eastern temperate forest; NF, Northern forest; NR, Northern regions (taiga/tundra/Arctic cordillera combined). reduce the costs of migration.

In fall, pre-adult eagles were more likely to migrate as high-quality habitat for migration in which eagles can pressure levels dropped and adult eagles were more likely conserve both time and energy (Duerr et al. 2012). Use of to migrate as pressure levels increased. Age-specific migra- thermal lift increases flight speed and reduces the time span tory response under different pressure regimes may appear of migration, which is especially important when migration to contrast with patterns observed at Hawk Mountain is time limited. When eagles are conserving energy on Sanctuary, Pennsylvania, where eagle counts tend to dou- migration, they can wait out periods of bad weather and ble within 3 days following the passage of a cold front return to migration when conditions are more conducive to (Allen, Goodrich & Bildstein 1996). However, cold fronts flight. Knowing responses to optimal migratory conditions are followed by high-pressure air masses, with correspond- is important when assessing the trade-off between time- and ing increases in barometric pressure, increases in type I energy limitations. However, to understand this problem, it thermal lift and, in our study, increased adult migration. is also vital to consider migratory responses to less optimal Most cold fronts passed by Hawk Mountain within 5 days weather conditions (i.e. orographic lift is lower-quality hab- of one another (Allen, Goodrich & Bildstein 1996), with a itat for migration) and time span of migration. By doing so, warm front, decreasing barometric pressure, and pre-adult we able to provide evidence that (i) adult eagles are more migration passing between cold fronts. Therefore, it is strongly time limited in spring, (ii) pre-adult eagles show likely that increased counts in the days following a cold behaviours consistent with energy limitation in spring, (iii) front at Hawk Mountain actually include adults migrating adults and pre-adults attempt to conserve energy in fall and within high pressure behind a cold front and pre-adults (iv) pre-adults appear to be less efficient migrants than migrating within low pressure with a cold front. By adults in fall. Thus, we conclude that golden eagles show migrating following a cold front, adults appear to supple- age- and season-specific responses to variation in habitat ment orographic lift, the prevailing lift type in fall, with quality of airspace that corresponds directly to time and thermal lift to a greater degree than pre-adults. These pat- energy conservation. terns strongly suggest that adult eagles respond more rap- idly to changing weather conditions, providing additional Acknowledgements evidence that adults migrate more efficiently than pre- adults. Funding for this work was received from PA SWG grants T-12 and T47- Differences in pressure systems also account for differ- R-1, US DoE grant DE-EE0003538, NASA Climate and Biological Response Grant #NNX11AP61G, Charles A. and Anne Morrow Lind- ences in migration patterns among ecoregions. This seems bergh Foundation, Hydro-Quebec, and the authors’ organizations. This intuitive because weather patterns and topography can dif- publication was completed in part with funds provided by the Virginia fer dramatically among regions (Commission for Environ- Department of Game and Inland Fisheries through a Federal Aid in Wild- life Restoration grant from the US Fish and Wildlife Service. This work is mental Cooperation 1997; Strandberg et al. 2009). In Scientific Article No. 3222 of the West Virginia Agricultural and Forestry spring, migration in forested ecoregions was associated Experiment Station, Morgantown. Use of golden eagles for this research with higher pressure, but migration in northern regions was approved by the West Virginia University Institutional Animal Care and Use Committee (IACUC) protocol #11-0304 and Animal Care Com- was associated with lower pressure. This is a logical conse- mittee of the Ministere du developpement durable, de l’Environnement, de quence of trends in prevailing weather patterns, as subpo- la Faune et des Parcs du Quebec [MDDEFP] (permits CPA-FAUNE-07- lar low pressure generally prevails to the north, and is 00-02 and CPA-FAUNE-2009-06). TK and ML are both part owners of Cellular Tracking Technologies, LLC that manufacture GPS-GSM trans- separated from southerly higher pressure by the jet stream mitters used in this project. We thank many people and organizations for (Moran 2009). their assistance with fieldwork and data collection. These include T. Golden eagles appeared to be generalists with regard to Anderson, the Bonta Family, B. Baillargeon, P. Beaupre, D. Brauning, J. Buchan, E. Estell, R. Faubert, R. Hartzel, B. Knight, D. Kramer, S. selection of migratory conditions. Raptors, including Jones, A. Mack, C. McCurdy, R. Mc Nicoll, B.D. Miller Family, P. eagles, are regularly observed from hawk-watch sites Reum, B. Sargent, D. Smith, R. Tallman and M. Wheeler, the National located on mountain peaks or ridgelines (Allen, Goodrich Aviary, the Carnegie Museums of Natural History. We thank three anon- ymous reviewers for comments that greatly helped improve this & Bildstein 1996). These observations create an appar- manuscript. ently false impression that soaring birds depend on orographic lift to migrate. Although eagles showed a consistent response to thermal development, they readily References migrated under conditions supporting both thermal and Alerstam, T. & Hedenstrom, A. (1998) The development of bird migration orographic lift. Eagles also showed season- and ecore- theory. Journal of Avian Biology, 29, 343–369. Allen, P.E., Goodrich, L.J. & Bildstein, K.L. (1996) Within- and among- gion-specific differences in migratory responses to weather year effects of cold fronts on migrating raptors at hawk mountain, Penn- conditions. Generalist behaviour in response to migratory sylvania, 1934–1991. The Auk, 113, 329–338. conditions is very likely a result of evolutionary pressures Avgar, T., Mosser, A., Brown, G.S. & Fryxell, J.M. (2013) Environmental and individual drivers of animal movement patterns across a wide geo- related to variable time and energy constraints to migra- graphical gradient. Journal of Animal Ecology, 82,96–106. tion that eagles experience over the course of their long Benjamini, Y. & Hochberg, Y. (1995) Controlling the false discovery rate: a lives. practical and powerful approach to multiple testing. Journal of the Royal Statistical Society. Series B (Methodological), 57, 289–300. A consistent response to thermal development allows Bird, D.M. & Bildstein, K.L. (2007) Raptor Research and Management soaring birds to both reduce time spent migrating and to Techniques. Hancock House, Surrey, British Columbia and Blane, Wash- conserve energy during migration. Thermal lift represents ington.

Bloom, P.H. & Clark, W.S. (2001) Molt and sequence of plumages of Mandel, J.T., Bohrer, G., Winkler, D.W., Barber, D.R., Houston, C.S. & golden eagles and a technique for in-hand ageing. American Bird Bander, Bildstein, K.L. (2011) Migration path annotation: cross-continental 26,97–116. study of migration-flight response to environmental conditions. Ecologi- Chevallier, D., Handrich, Y., Georges, J.Y., Baillon, F., Brossault, P., cal Applications, 21, 2258–2268. Aurouet, A. et al. (2010) Influence of weather conditions on the flight of Miller, T.A. (2012) Movement ecology of golden eagles (Aquila chrysaetos) migrating black storks. Proceedings of the Royal Society B: Biological in eastern North America. PhD Dissertation, Pennsylvania State Univer- Sciences, 277, 2755–2764. sity, State College, Pennsylvania. Commission for Environmental Cooperation (1997) Ecoregions of North Mitchell, G.W., Newman, A.E.M., Wikelski, M. & Ryan Norris, D. (2012) America, Level I, II, and III Maps. Available at [http://www.epa.gov/ Timing of breeding carries over to influence migratory departure in a wed/pages/ecoregions/na_eco.htm] from the United States Environmen- songbird: an automated radiotracking study. Journal of Animal Ecology, tal Protection Agency (EPA), Office of Research and Development, 81, 1024–1033. National Health and Environmental Effects Research Laboratory, Wes- Moran, J.M. (2009) Weather Studies Introduction to Atmospheric Science, tern Ecology Division, Corvallis, Oregon, USA and at [http://www.cec. 4th edn. American Meteorological Society, Boston, Massachusetts. org/pubs_docs/documents/index.cfm?varlan = english&id = 344] from Newton, I. (2008) The Migration Ecology of Birds. Academic Press, Lon- the Commission for Environmental Cooperation (CEC), Montreal don. (Quebec), Canada. Pike, N. (2011) Using false discovery rates for multiple comparisons Diehl, R.H. (2013) The airspace is habitat. Trends in Ecology & Evolution, in ecology and evolution. Methods in Ecology and Evolution, 2, 278– 28, 377–379. 282. Duerr, A.E., Miller, T.A., Lanzone, M., Brandes, D., Cooper, J., O’Malley, R Core Team (2012) R: A Language and Environment for Statistical Com- K. et al. (2012) Testing an emerging paradigm in migration ecology puting. R Foundation for Statistical Computing, Vienna, Austria. shows surprising differences in efficiency between flight modes. PLoS Sachs, G. & Mayrhofer, M. (2001) Shear wind strength required for ONE, 7, e35548. dynamic soaring at ridges. Technical Soaring, 25, 209–215. Garcıa, L.V. (2004) Escaping the Bonferroni iron claw in ecological studies. Sachs, G., Traugott, J., Nesterova, A.P., Dell’Omo, G., Kummeth,€ F., Oikos, 105, 657–663. Heidrich, W. et al. (2012) Flying at no mechanical energy cost: disclos- Hardy, K.R. & Ottersten, H. (1969) Radar investigations of convective pat- ing the secret of wandering albatrosses. PLoS ONE, 7, e41449. terns in the clear atmosphere. Journal of the Atmospheric Sciences, 26, Sergio, F., Blas, J., Forero, M.G., Donazar, J.A. & Hiraldo, F. (2007) 666–672. Sequential settlement and site dependence in a migratory raptor. Behav- Hedenstrom, A. (1993) Migration by soaring or flapping flight in birds – ioral Ecology, 18, 811–821. the relative importance of energy-cost and speed. Philosophical Transac- Shamoun-Baranes, J., Bouten, W. & van Loon, E.E. (2010) Integrating tions of the Royal Society of London Series B-Biological Sciences, 342, meteorology into research on migration. Integrative and Comparative 353–361. Biology, 50, 280–292. Hedenstrom, A. & Alerstam, T. (1995) Optimal flight speed of birds. Philo- Shamoun-Baranes, J., Liechti, O., Yom-Tov, Y. & Leshem, Y. (2003) Using sophical Transactions of the Royal Society of London Series B-Biological a convection model to predict altitudes of white stork migration over Sciences, 348, 471–487. central Israel. Boundary-Layer Meteorology, 107, 673–681. Hedenstrom, A. & Alerstam, T. (1997) Optimum fuel loads in migratory Shamoun-Baranes, J., van Loon, E., Alon, D., Alpert, P., Yom-Tov, Y. & birds: distinguishing between time and energy minimization. Journal of Leshem, Y. (2006) Is there a connection between weather at departure Theoretical Biology, 189, 227–234. sites, onset of migration and timing of soaring-bird autumn migration in Jollie, M. (1947) Plumage changes in the golden eagle. The Auk, 64, 549– Israel? Global Ecology and Biogeography, 15, 541–552. 576. Shannon, H.D., Young, G.S., Yates, M.A., Fuller, M.R. & Seegar, W.S. Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., Gandin, (2002) American White Pelican soaring flight times and altitudes relative L. et al. (1996) The NCEP/NCAR 40-year reanalysis project. Bulletin of to changes in thermal depth and intensity. Condor, 104, 679–683. the American Meteorological Society, 77, 437–471. Strandberg, R., Klaassen, R.H.G., Olofsson, P. & Alerstam, T. (2009) Katzner, T.E., Brandes, D., Miller, T., Lanzone, M., Maisonneuve, C., Daily travel schedules of adult Eurasian Hobbies Falco subbuteo – vari- Tremblay, J.A. et al. (2012a) Topography drives migratory flight altitude ability in flight hours and migration speed along the route. Ardea, 97, of golden eagles: implications for on-shore wind energy development. 287–295. Journal of Applied Ecology, 49, 1178–1186. Watson, J. (2010) The Golden Eagle, 2nd edn. Yale University Press, New Katzner, T., Smith, B., Miller, T., Brandes, D., Cooper, J., Lanzone, M. Haven, Connecticut. et al. (2012b) Status, biology, and conservation priorities for North van Wijk, R.E., Kolzsch,€ A., Kruckenberg, H., Ebbinge, B.S., Muskens,€ America’s Eastern Golden Eagle (Aquila chrysaetos) Population. The G.J.D.M. & Nolet, B.A. (2012) Individually tracked geese follow peaks Auk, 129, 168–176. of temperature acceleration during spring migration. Oikos, 121, 655– Kemp, M.U., Emiel van Loon, E., Shamoun-Baranes, J. & Bouten, W. 664. (2011) RNCEP: global weather and climate data at your fingertips. Yates, R.E., McClelland, B.R., McClelland, P.T., Key, C.H. & Bennets, Methods in Ecology and Evolution, 3,65–70. R.E. (2001) The influence of weather on golden eagle migration in north- Kerlinger, P. (1989) Flight Strategies of Migrating Hawks. University of western Montana. The Journal of Raptor Research, 35,81–90. Chicago Press, Chicago, Illinois. Kerlinger, P. & Moore, F.R. (1989) Atmospheric structure in avian migra- Received 4 December 2013; accepted 25 November 2014 tion. Current Ornithology, 6, 109–142. Handling Editor: Craig Franklin Klaassen, R.H.G., Hake, M., Strandberg, R. & Alerstam, T. (2011) Geo- graphical and temporal flexibility in the response to crosswinds by migrating raptors. Proceedings of the Royal Society B-Biological Sci- Supporting Information ences, 278, 1339–1346. Lanzone, M.J., Miller, T.A., Turk, P., Brandes, D., Halverson, C., Maison- Additional Supporting information may be found in the online neuve, C. et al. (2012) Flight responses by a migratory soaring raptor to version of this article: changing meteorological conditions. Biology Letters, 8, 710–713. Leshem, Y. & Yom-Tov, Y. (1996) The use of thermals by soaring Appendix S1. Definitions of select weather variables. migrants. The Ibis, 138, 667–674. Table S1. Spring migration response model. Lyons, J.E., Collazo, J.A. & Guglielmo, C.G. (2008) Plasma metabolites and migration physiology of semipalmated sandpipers: refueling perfor- Table S2. Fall migration response model. mance at five latitudes. Oecologia, 155, 417–427.