Skip to main content
Proceedings of the National Academy of Sciences of the United States of America logoLink to Proceedings of the National Academy of Sciences of the United States of America
. 2015 Nov 2;112(46):E6331–E6338. doi: 10.1073/pnas.1503381112

Fat, weather, and date affect migratory songbirds’ departure decisions, routes, and time it takes to cross the Gulf of Mexico

Jill L Deppe a,1, Michael P Ward b, Rachel T Bolus b,c, Robert H Diehl c, Antonio Celis-Murillo b, Theodore J Zenzal Jr d, Frank R Moore d, Thomas J Benson e, Jaclyn A Smolinsky f, Lynn N Schofield a, David A Enstrom e, Eben H Paxton g, Gil Bohrer h, Tara A Beveroth e, Arlo Raim e, Renee L Obringer h, David Delaney i, William W Cochran e
PMCID: PMC4655507  PMID: 26578793

Significance

Bird migration has captivated the attention of scientists and lay people for centuries, but many unanswered questions remain about how birds negotiate large geographic features during migration. We tracked songbirds across the Gulf of Mexico to investigate the factors associated with birds’ departure decisions, arrival at the Yucatan Peninsula (YP), and crossing times. Our findings suggest that a bird’s fat reserves and low humidity, indicative of favorable synoptic weather patterns, shape departure decisions. Fat, date, and wind conditions predict birds’ detection in the YP. This study highlights the complex decision-making process involved in crossing the Gulf and its effects on migratory routes and speeds. A better understanding of the factors influencing migration across these features will inform conservation of migratory animals.

Keywords: migration, ecological barrier, Gulf of Mexico, songbirds, weather

Abstract

Approximately two thirds of migratory songbirds in eastern North America negotiate the Gulf of Mexico (GOM), where inclement weather coupled with no refueling or resting opportunities can be lethal. However, decisions made when navigating such features and their consequences remain largely unknown due to technological limitations of tracking small animals over large areas. We used automated radio telemetry to track three songbird species (Red-eyed Vireo, Swainson’s Thrush, Wood Thrush) from coastal Alabama to the northern Yucatan Peninsula (YP) during fall migration. Detecting songbirds after crossing ∼1,000 km of open water allowed us to examine intrinsic (age, wing length, fat) and extrinsic (weather, date) variables shaping departure decisions, arrival at the YP, and crossing times. Large fat reserves and low humidity, indicative of beneficial synoptic weather patterns, favored southward departure across the Gulf. Individuals detected in the YP departed with large fat reserves and later in the fall with profitable winds, and flight durations (mean = 22.4 h) were positively related to wind profit. Age was not related to departure behavior, arrival, or travel time. However, vireos negotiated the GOM differently than thrushes, including different departure decisions, lower probability of detection in the YP, and longer crossing times. Defense of winter territories by thrushes but not vireos and species-specific foraging habits may explain the divergent migratory behaviors. Fat reserves appear extremely important to departure decisions and arrival in the YP. As habitat along the GOM is degraded, birds may be limited in their ability to acquire fat to cross the Gulf.


During migration, animals encounter ecological barriers, inhospitable environmental features that prevent or impede movement due to increased risk of mortality from starvation, predation, collision, and severe environmental conditions (e.g., weather for aerial migrants, aquatic temperature or chemical gradients for aquatic migrants) (15). Because barriers can have important consequences on survival and future reproductive success (6), animals have evolved behavioral, morphological, and/or physiological means to safely negotiate them (79). Barriers can include large geographic features (e.g., large water bodies, deserts, mountains), inhospitable land cover types (e.g., agricultural “deserts”), anthropogenic structures (e.g., tall buildings, towers, dams, weirs), and unfavorable weather and aquatic conditions (e.g., droughts, storms, strong temperature gradients), although the extent to which any of these functions as a barrier to migration varies (4, 5, 1012).

Approximately two thirds of all songbird species and millions of individuals breeding in eastern Canada and the United States encounter the Gulf of Mexico (GOM) while migrating to tropical or subtropical wintering grounds in the Caribbean, Mexico, and Central and South America (13). Unfavorable weather conditions combined with a lack of resting and refueling opportunities over open water can be lethal (14, 15). Accounts of thousands of songbirds washing ashore (16), exhausted songbirds alighting on offshore structures or boats (17), terrestrial birds in the stomachs of sharks (18), and flights away from the coast in seasonally inappropriate directions (19, 20) reinforce the view that crossing the GOM presents considerable risk. However, though the GOM is often considered a barrier (17, 20), large numbers of birds routinely cross it (2125) and arrive on the opposite coast in good energetic condition (26), suggesting that the Gulf is not inherently a barrier. Rather, the risks of crossing the GOM and the extent to which it functions as a barrier appear to be determined by intrinsic and extrinsic factors—notably, weather (10, 19, 27) and fat reserves (20, 28). By coordinating the timing and orientation of departure with favorable conditions, crossing large bodies of open water can be quick, energy efficient, and safe (2). Under such conditions, crossing features like the GOM should be preferred to circumnavigating them, because crossing can substantially reduce travel distances and time while reducing exposure to predators and pathogens (8, 29). However, how departure decisions and conditions affect the fate of songbirds crossing large geographic features is unknown, hindering our understanding of how songbirds negotiate ecological barriers.

The risk of crossing the GOM is dynamic and unpredictable due to spatiotemporal changes in atmospheric conditions over water as well as the variable energetic condition of birds. Small songbirds that incorrectly assess risk incur large fitness consequences; therefore, natural selection presumably has favored flexible migration strategies (4, 24). Under such strategies, animals may use decision rules to assess risk by considering intrinsic and extrinsic factors at the time and location of departure. For birds crossing the GOM, an ∼1,000-km nonstop flight that can take more than 24 h, maintaining a positive energy balance is likely the primary consideration when assessing risk (30). After prolonged and/or energetically inefficient flights, migrants may run out of energy reserves and die. Longer flights also may increase exposure to inclement weather. Therefore, migrants should make decisions that minimize both time and energy expenditure.

Intrinsic factors are expected to influence birds’ decisions about whether and when to migrate across the GOM; such factors include fat reserves, age, and wing aerodynamics (e.g., wing length). The role of these factors is likely species specific. Energetic reserves (primarily fat but also protein) contribute to reducing the risks of migrating where foraging options are scarce and/or energetic demands are elevated (31, 32). Age may affect migratory decisions because older individuals might use previous experience to optimize their travel and assess and manage risk (33, 34). Individuals, or species, with longer, more-pointed wings and lower wing loading benefit from more energy efficient flight and are more likely to attempt longer, nonstop flights (35). Furthermore, evidence suggests that species with wing morphology adapted for long-distance flight can fly in a broader range of weather conditions than birds with morphologies less adapted for long-distance flight (9).

Atmospheric conditions are the primary extrinsic factors influencing decisions regarding migratory flights, particularly over water bodies with limited opportunities to land (36). Wind plays a critical role, affecting departure date and migratory directions, routes, speeds, flight durations, and energy consumption (35, 3740). Decisions to depart stopover sites and initiate flight over large water bodies also are influenced by barometric pressure, temperature, relative humidity, and short-term trends in these variables, which are indicative of synoptic weather patterns and may provide information about future weather conditions (14, 19, 36).

Studies of migratory departure decisions of small birds along edges of large geographic features, including the GOM, have contributed substantially to our understanding of how they assess and manage risk at the onset of flight (20, 28, 31, 41), but the consequences of these decisions on migratory routes, crossing times, and arrival on the other side are unknown. Light-geolocator studies have advanced our understanding of general migration routes of small birds in relation to geographic features and overall rates of migration (42), but the low spatial and temporal precision of these studies do not permit analyses of how behaviors are informed by dynamic intrinsic and extrinsic variables at fine scales. Though GPS loggers and satellite transmitters have become smaller (43), nonarchival units with fine temporal resolution are not nearly small enough for small songbirds. Because of differences in flight mechanics, physiology, behavior, and time of migration (day vs. night), we do not expect songbirds to follow the same behavioral rules to guide their migration as the shorebirds, waterfowl, and raptors that have been tracked with these larger nonarchival technologies (2, 4, 11). Automated radio telemetry, however, allows for the collection of precise spatial and temporal data on animal movements without the need to retrieve the device (20), and transmitters can be deployed on animals as small as 0.3 g (44). Automated telemetry coupled with favorable geography facilitates detection over a large area, providing a tool to link departure decisions, arrival status, and crossing times with intrinsic and extrinsic conditions.

To understand how small Neotropical migratory songbirds negotiate the GOM, we established two networks of automated radio telemetry systems to record departure behavior of three species from coastal Alabama (AL) and detect birds in the northern Yucatan Peninsula (YP) following passage across the Gulf (Fig. 1, SI Methods, Fig. S1). For species with winter distributions in southern Mexico and Central and South America, birds departing AL under prevailing winds should arrive at the YP (45), a landmass known for its importance to Neotropical migrants (22). Using this network, we first examined birds’ departure decisions (over water, over land, or no departure, i.e., stopover) from coastal AL in relation to intrinsic (species, age, fat, wing length) and extrinsic (weather, date) variables to identify the conditions that favor initiation of flights across the GOM. Second, using information on known arrivals at the YP and crossing times from AL to the YP, we identified the conditions suitable for flights across the GOM and examined whether departure decisions from AL accurately predicted arrival at the YP. Third, we evaluated factors influencing crossing times to the YP. By detecting songbirds in the YP, ∼1,000 km away from their departure site in coastal AL, we provide unique insight into the factors and conditions that minimize risk of crossing the GOM and influence its role as an ecological barrier.

Fig. 1.

Fig. 1.

Locations of automated telemetry towers around the Gulf of Mexico. (A) Locations of our capture site (blue arrow) and tracking towers in AL (black circles). (B) Locations of tracking towers along the northern Yucatan Peninsula (black circles). The distance between the two regions ranges from 950 to 1,040 km.

Fig. S1.

Fig. S1.

(A) Locations of automated tracking towers (solid black circles) around Mobile Bay in coastal Alabama. (B) Inset from A showing the location of tracking towers 1–4 (solid black circles) and the capture location (yellow asterisk) on the Fort Morgan Peninsula.

SI Methods

Focal Species.

We considered three species that vary in their breeding and wintering distributions and flight morphology (wing loading and aspect ratio) (59). SWTH (Catharus ustulatus) and REVI (Vireo olivaceus) are long-distance Nearctic–Neotropical migratory songbirds that breed in Canada and the United States and winter in southern Mexico and Central and South America (77, 78). WOTH (Hylocichla mustelina), however, is a shorter-distance migrant that breeds in the United States and winters in southern Mexico and Central America (79). All three focal species are common around the GOM in the fall, and their migratory behavior and physiology have been well-studied (26, 37, 39, 51, 56, 58, 8083). Based on differences in geographic distribution and flight morphology between the long-distance migrants (SWTH, REVI) and shorter-distance migrant (WOTH), we expected them to differ in departure decisions, proportion of birds arriving at the YP, and crossing times.

Alabama Capture and Departure Site Description.

We captured and radio-tagged birds of the three focal species at a long-term, passive banding and research station in the Bon Secour National Wildlife Refuge (30° 13′ 49″ N, 88° 0′ 13″ W) on the FTM Peninsula, Alabama. The banding station has been in operation and supervised by F.R.M. since 1990. The FTM Peninsula, a known stopover site for migratory songbird species in the fall (20, 51), is 26 km long, ∼0.7 km wide, and located directly south of Mobile Bay along the northern coast of the GOM (Fig. S1).

The vegetation at the banding site is a combination of sandy scrub and pine forest (for detailed description of the vegetation, see refs. 51, 84, 85). Dominant understory species of both vegetation types include Sand Live Oak (Quercus geminata), Myrtle Oak (Quercus myrtifolia), Inkberry (Ilex glabra), Yaupon (Ilex vomitoria), Greenbrier (Smilax spp.), Muscadine (Vitis rotundifolia), Wax Myrtle (Myrica cerifera), Red Bay (Persea borbonia), Saw Palmetto (Serenoa repens), and Sand Heath (Ceratiola ericoides). Most of these species reach a maximum of 4 m in height, but tend to form a dense vegetation layer. The canopy is exclusively Slash Pine (Pinus elliottii), which may be as tall as 13 m, but generally occurs in sparse to moderate densities throughout the study area.

Capture and Tagging Methods.

We captured birds from September 2 to October 28, 2009–2013, with the exception of October 1–13, 2013, when the US government shutdown prevented activities within the Bon Secour National Wildlife Refuge. We generally operated mist-nets between sunrise and noon Central Daylight Time, unless unfavorable weather prohibited the safe operation of nets. We banded each radio-tagged bird with a uniquely numbered United States Geological Survey aluminum band. We classified birds as hatch year or after-hatch year based on plumage characteristics and skull pneumatization (86). Hatch-year birds were born during the summer and were migrating for the first time, whereas after-hatch year birds had migrated at least twice (fall and spring) before capture. We measured birds’ unflattened wing chord length (mm) and assigned them a fat score ranging from 0 to 5 based on a visual inspection of fat reserves (76). We used each bird’s fat score as an index of its physical condition in our models. To assess how within-species variation in wing length affected departure decisions, arrival at the YP, crossing time (time between departure from FTM and arrival at the YP), and trans-Gulf (direct) flight durations, we computed the deviation of each individual’s wing length from the mean wing length of that species’ FTM population (indexed as a z-score) based on long-term banding records for each species.

We tagged birds with analog, pulse transmitters from JDJC Corp., Holohil Systems Ltd. (model BD-2), and Lotek (PIP31 beeper transmitters). We tagged vireos with transmitters from JDJC Corp. in 2009–2010, Holohil Systems in 2011, and Lotek in 2012–2013. All thrushes (2009–2013) were tagged with transmitters from JDJC Corp. Each transmitter had a unique frequency ranging from 163.828 to 166.060 MHz with pulse widths of 14, 22, or 29 ± 2 ms (±SE, manufacturer specified) and pulse intervals of 350, 780, and 900 ± 10 ms. The mean weight of vireo transmitters was 0.73 ± 0.01 g (n = 42), including cloth and thread for attachment, and transmitters had a minimum life span of ∼21 d. Thrush transmitters weighed 1.13 ± 0.03 g (±SE, n = 25) with cloth and thread, and had a lifespan of ∼28 d. Transmitters weighed less than 5% of birds’ body mass, and in most cases were less than 3%. We attached transmitters to birds’ backs between the wings with an eyelash adhesive (Revlon brand) and a small amount of cyanoacrylate glue (Loctite brand) using a modification of a method developed by Raim (68) and described in Smolinsky et al. (20). This approach allowed the transmitters to fall off after 3–4 wk, coinciding with the expected life span of the transmitter battery to remove the weight and aerodynamic drag impacts of the transmitter on birds’ future movements.

Automated Tracking of Birds in Alabama.

We established a network of automated radio-tracking systems in Alabama to record the departure behavior of radio-tagged birds. Each year of the study, three tracking towers were established and operated on the FTM Peninsula to estimate birds’ departure date, time, and direction (Fig. S1). The precise location of the towers on the FTM Peninsula varied slightly among years depending on availability of receivers and site access, although two towers within 1.8 km of the banding station were operational each year. From 2009 to 2011, towers 1, 3, and 4 were operational on the peninsula, and from 2012 to 2013 towers 2–4 were operational. From 2012 to 2013 we operated an additional tracking tower on Dauphin Island (7.5 km west of FTM banding station, tower 5 in Fig. S1).

Towers 1, 3, and 4 in 2009–2011 and towers 2, 3, and 5 in 2012–2013 were equipped with six directional Yagi antennas positioned at 60° intervals to allow us to estimate each bird’s azimuth upon departure from FTM (63, 64); antenna azimuths were 0, 60, 120, 180, 240, and 300°, and towers were 9.2 m tall. From 2009 to 2010 we used three-element directional yagi antennas (PCTel Inc.), and from 2011 to 2013 we used three-element directional yagi folded-dipole antennas (JDJC Corp.). Although these two antenna designs perform similarly, the latter ensures better antenna input impedance, which enhances tuning across towers.

In 2013 we operated a tower along the northern extent of Mobile Bay (50 km north of FTM, tower 6 in Fig. S1). Tower 6 as well as tower 4 from 2012 to 2013 was equipped with four high-gain stacked directional antennas (designed by W.W.C.). Each stacked antenna consisted of two 3-element yagi antennas separated by 125 cm, connected by a Y connector, and tuned to 164.5 MHz. In comparison with the standard single three-element yagi directional antennas used in FTM, this antenna design increases directionality and gain (87). These high-gain towers maximized our detection range, improving our detection of birds that made reverse movements (oriented northward upon departure) and subsequently departed south, but reduced our ability to precisely estimate azimuths.

We used automated receiving units (ARUs) designed by JDJC Corp. to autonomously monitor transmitter frequencies. Each ARU was programmed to tune to each frequency at 0.5- to 4.0-min intervals in Alabama. Sampling intervals increased with the number of transmitters programmed in the ARU at any particular time and the number of antennas on the tower. ARUs recorded signal strength, noise, pulse width, and pulse interval for each transmitter on each antenna.

Automated Tracking in the Yucatan Peninsula.

We established a telemetry fence along the northern coast of the YP using seven towers, each equipped with 2–4 high-gain stacked yagi antennas, identical to the ones used in Alabama (towers 4 and 6), to detect the arrival of birds in Mexico (Fig. 1). In 2009–2011, each tracking tower in Mexico was equipped with two high-gain antennas oriented toward 90° and 270°, roughly parallel to the northern YP coastline. From 2012 to 2013, each tower was equipped with two additional high-gain antennas oriented toward the north (0°) and south (180°) to improve detection probability as well as provide coarse information on arrival direction.

We positioned all antennas above the vegetation and surrounding objects (e.g., buildings) to improve the detection capabilities of our tracking towers; antennas were placed an average of 10.6 ± 2.4 m (SE) above ground level. The remarkably flat topography of the northern YP favored the detection of arriving birds by our tracking towers, and altitude of a radio-tagged bird was the most important factor in determining detection distance. Theoretically, a bird flying at an altitude of 500 m could be detected 80 km away with our high-gain antennas, whereas a bird flying at an altitude of 1,000 m could be detected at a distance of ∼115 km.

Various factors, such as electromagnetic noise, can restrict the operational detection distance of antennas in the field by masking the transmitter signal. We selected sites that had minimal electromagnetic noise of a relatively continuous nature (e.g., powerline transformers or wireless internet communications) to minimize this effect; however, other sources of short, but irregular electromagnetic interference were unavoidable (e.g., lightning strikes and rain). An evaluation of our telemetry data revealed these sporadic noise events were of limited impact temporally and spatially. Given the factors that can limit the detection distance of our tracking towers, we spaced our towers an average of 57 ± 7 km (SE) apart along the entire northern coast of the YP to maintain continuous detection along the entire coast (Fig. 1). ARUs sampled each transmitter frequency at 3- to 6-min intervals in the YP and recorded the same parameters (signal strength, noise, pulse width, and pulse interval) for each transmitter on each antenna.

Processing Automated Radio Telemetry Data.

We developed R and Python scripts (M.P.W. and L.N.S.) to search the millions of lines of data for radio-tagged birds in AL and the YP. Based on Ward et al. (71), we took a conservative approach to detecting radio-tagged individuals in AL and the YP based on six criteria, or filters: (i) signal strength needed to be > −130 dBm; (ii) background noise level < −130 dBm; (iii) pulse width within 2 ms of the manufacturer-specified width; (iv) pulse interval within 50 ms of the specified interval or a multiple thereof; (v) frequency (MHz) within two one-thousandths of the specified frequency; and (vi) the bird had to be detected for three consecutive sampling periods. We manually verified all detections in AL and the YP to eliminate false positives, such as those occurring as a result of broad-spectrum noise, which was characterized as occurring on multiple frequencies simultaneously from the same direction.

To validate the custom code, we searched for frequencies of transmitters that had been programmed in the ARUs but never attached to birds (fully functional but inactive transmitters that were remaining at the end of the season) and for test transmitters known to be functional and active and placed at known locations around tower sites in AL and the YP. Using the six criteria described above, our detection program failed to detect any of the frequencies that were programmed but not attached to birds, and was able to reliably detect all test transmitters. On several occasions, however, we detected frequencies that matched three criteria (signal strength, background noise, and frequency). Close examination of those detections suggested they were radio-tagged birds from other studies. The transmitters we used in our study had shorter pulse intervals and pulse widths than typical transmitters, allowing us to differentiate our radio-tagged birds from those of other studies. We manually verified all positive detections of birds in AL and the YP by plotting signal strength and noise. Positive detections revealed a peak in signal strength, but no corresponding change in noise, as birds departed AL or arrived at the YP, followed by a tapering off in signal strength as birds departed the site (AL) or passed by the tower (YP). We also examined adjacent frequencies in our ARU worklist to verify that they did not show a similar pattern.

Estimating Departure Date, Time, and Direction from Alabama.

When a sufficient amount of high-quality data (e.g., signals > −122 dBm) were available from multiple towers, we used biangulation or triangulation to estimate an individual’s departure track and direction following the approach described in Smolinsky et al. (20). We used Location of a Signal Software (LOAS; Ecological Software Solutions, LLC) to estimate series of Universal Transverse Mercator (UTM) locations defining the track of each bird, and then used a locally weighted regression (LOESS) to create a smoothed, predicted departure track. We used the last five predicted coordinates of the track to estimate the bird’s departure bearing and time. When an insufficient number of high-quality signals were available to permit the use of the LOAS-LOESS approach, we estimated birds’ vanishing bearings based on data from the strongest tower and used them as estimates of departure direction in our analyses. Vanishing bearings were estimated for some birds in 2009–2010 that had rapid departures (too few signals) or partially obscured signals (poor-quality sequence of signals) upon departure, and for all birds in 2011–2013 when transmitters were sampled at longer intervals due to a larger number of frequencies programmed into the ARUs. We estimated birds’ vanishing bearings based on changes in signal strength across the six antennas of the strongest tower, usually tower 3, adjacent to the banding site (Fig. S1), and we used the last detection to estimate the departure bearing and time. For birds detected on multiple towers, departure bearings and vanishing bearings were highly correlated (Pearson correlation r = 0.98, P < 0.0001, n = 17). Smolinsky et al. (20) estimated the accuracy of departure bearings from the six-antenna towers at FTM to be 2.7 ± 2.4° (range 0–13°) by locating test transmitters at known bearings from each tower.

Estimating Arrival Date and Time at the Yucatan Peninsula.

When a bird was detected by the tracking towers in the YP, we estimated the arrival time as the hour (Central Daylight Time) of the peak signal strength on the east or west antenna of the strongest tower. The east and west antennas were generally oriented along the coast, and the highest gain of the antenna is directly in front of the antenna; therefore, the peak signal strength correlated with the arrival of the bird over land. Locations were conservatively estimated as being at one of the seven towers or midway between adjacent towers based on which towers and antennas detected the birds. Forty-four percent of birds arriving at the YP were detected by multiple towers (2–4 towers). We archived departure and arrival date in Movebank (72).

Weather Variables.

We used weather variables from the National Center for Environmental Prediction North American Regional Reanalysis (NARR) dataset (73). NARR has 3-h temporal resolution, 32 km horizontal resolution, and ∼250 m vertical resolution. We obtained NARR variables from the Env-DATA service available on www.movebank.org, a website repository of animal movement data (73, 74, 88). The Env-DATA service provides interpolated variables to the nearest time, location, and altitude using an inverse-distance weighted method.

For the departure and arrival analyses, we considered weather variables that have been found to influence migrant departure decisions in avian systems, including (i) surface-level humidity; (ii) surface-level barometric pressure; (iii) 24-h change in surface humidity; (iv) 24-h change in surface pressure; (v), wind speed; and (vi) wind direction. We used surface-level humidity and pressure, and 24-h change in these variables, because they are predictive of larger, synoptic weather systems that influence the availability of favorable winds or probability of inclement weather over open water (9, 36, 46, 89). We used wind speeds and directions to calculate a wind-profit index. Wind profit was defined as the speed (m/s) of wind toward 180° (see formula for wind profit in ref. 46), or how favorable winds were crossing to the YP. Because wind profits vary by altitude, and we were unsure of the altitudes of the birds, we averaged wind profits calculated at 1-, 2-, 3-, and 4-km altitudes at FTM on the night of departure. These four altitudes represent the range of wind conditions available to most songbird migrants in the airspace (90). We excluded temperature due to its significant correlation with ordinal date and its small daily variation (Tables S1 and S2).

Table S1.

Correlation matrix among weather variables used in CART analysis of departure decisions

Variable Wind profit Temperature Humidity Pressure Humidity change Pressure change
Calendar day r = −0.11 r = −0.81 r = −0.35 r = 0.15 r = −0.08 r = −0.16
P = 0.08 P < 0.0001 P < 0.0001 P = 0.02 P = 0.20 P = 0.01
Wind profit r = 0.09 r = −0.34 r = 0.02 r = −0.16 r = −0.17
P = 0.14 P < 0.0001 P = 0.73 P = 0.009 P = 0.006
Temperature r = 0.27 r = −0.37 r = −0.09 r = 0.07
P < 0.0001 P < 0.0001 P = 0.12 P = 0.27
Humidity r = −0.15 r = 0.51 r = −0.09
P = 0.02 P < 0.0001 P = 0.13
Pressure r = 0.34 r = 0.19
P < 0.0001 P = 0.002
Humidity change r = −0.22
P = 0.72

Although many variables demonstrated statistically significant correlations, correlation coefficients were generally small. Rather, statistical significance appeared to be driven by our large sample size (n = 244). Therefore, we removed variables with |r| > 0.6 (deleted variables denoted by boldface text). Temperature was strongly correlated with calendar date, so we removed it from our analysis.

Table S2.

Correlation matrix among weather variables used in CART analysis of arrival status at the YP

Variable Wind profit Temperature Humidity Pressure Humidity change Pressure change
Calendar day r = −0.04 r = −0.83 r = −0.29 r = 0.13 r = 0.017 r = −0.12
P = 0.68 P < 0.0001 P = 0.005 P = 0.16 P = 0.87 P = 0.27
Wind profit r = −0.05 r = −0.32 r = 0.16 r = 0.05 r = −0.24
P = 0.66 P = 0.002 P = 0.13 P = 0.62 P = 0.02
Temperature r = 0.24 r = −0.40 r = −0.26 r = 0.14
P = 0.02 P = 0.0001 P = 0.01 P = 0.18
Humidity r = 0.10 r = 0.56 r = 0.16
P = 0.32 P < 0.0001 P = 0.13
Pressure r = 0.39 r = 0.12
P = 0.0001 P = 0.27
Humidity change r = −0.19
P = 0.07

Similar to our departure weather data, many variables demonstrated statistically significant correlations, but correlation coefficients were generally small. Rather, statistical significance appeared to be driven by our large sample size (n = 90). Therefore, we removed variables with |r| > 0.6 (deleted variables denoted by boldface text). Temperature was strongly correlated with calendar date at the time of departure, so we removed it from our CART assessing arrival status.

In the analysis of trans-Gulf (i.e., direct) flight times of over-water departing birds, we created a model to estimate the wind profit that an individual would encounter during a direct flight across the GOM to the YP. This model simulated flight tracks from the FTM Peninsula in Alabama to the YP each night that a bird made a direct flight. Each simulated track began at civil twilight at 1 km altitude with a track direction of 180°. The bird’s heading, which remained constant over the entire track, was calculated using vector addition from the airspeed (10 m/s) and the wind vector at FTM at the time of departure; wind data were estimated using inverse-weighted distance interpolation of the four nearest points and two nearest times in the NARR dataset. New locations along the track were updated on hourly time steps by calculating the groundspeed and track direction using vector addition of the bird’s airspeed and heading with wind speed and wind direction. We then determined the next location by using the pyproj module in Python 3.3.3 to project the hourly travel distance along the loxodrome (or shortest distance along the earth’s surface without a change in heading) in the updated track direction, using a WSG84 projection and an estimated earth’s radius of 6,378,137 m. Tracks ended when they reached land, which for 180° tracks was the YP. The location of land was defined by the land–water mask used by NARR available in raster format, which also has a 32-km precision. A simulated bird “reached land” if one of the four nearest points on the raster grid was land. At each hourly time-step, wind speed and direction were recorded and wind profit was calculated as defined above. Each track’s wind-profit values were averaged across hourly time-steps and across four altitudes (1, 2, 3, and 4 km) to create single nightly profit values, which were each used in the analysis.

Statistical Analyses.

We used a GLMM with a multinomial distribution and generalized logit link function to examine differences in departure decisions among species and a GLMM with a binomial distribution and logit link function to assess differences in probability of arrival in the YP in relation to species and departure decision. We modeled departure day as a random effect to account for the potential nonindependence of multiple individuals departing on the same day. GLMMs were run two separate ways, with thrush species considered separately and as a single group. The outcome of the results was not affected by whether we combined thrush species or considered them separately.

To determine which extrinsic and intrinsic factors predicted departure decision from AL, we performed a CART analysis (75). We classified birds (n = 244) into three groups based on whether they departed and if they departed, their departure direction: over-water departure, over-land departure, and no departure (i.e., stopover). Stopover duration was conservatively estimated as the difference between capture and departure date and time. Over-land migrants departed in a northward direction (≥270° or ≤90°) indicative of reverse movements or the adoption of a circum-Gulf route. Over-water migrants included all birds that departed toward the south (>90° and <270°) over the Gulf.

Model predictors included species, fat score, age, relative wing length (z-score), ordinal date, relative humidity, 24-h change in humidity, barometric pressure, 24-h change in pressure, and wind profit. To objectively compare weather available to birds that departed the banding site on the evening of capture and those that did not (i.e., those that stopped over), we used weather conditions at civil twilight on the evening of capture. Most migrants depart at civil twilight and likely decide whether or not to depart based on conditions at that time (20) (Fig. S3).

Fig. S3.

Fig. S3.

Circular plots illustrating the distribution of departure times (Central Standard Time) for (Left) Red-eyed Vireo, (Center) Swainson’s Thrush, and (Right) Wood Thrush.

We performed a second CART analysis to determine which factors are related to the detection of birds at the YP. We restricted this analysis to over-water departures (n = 90), because they were the only birds for which we had reliable fat scores at the time of departure. We classified over-water migrants into one of three groups: direct flight, indirect flight, and not detected. We considered birds that were detected at the YP in ≤35 h to have arrived following a direct trans-Gulf flight from AL. We considered birds detected ≥70 h after their southward departure from coastal Alabama as having adopted an indirect route through unknown stopover sites on their way to the YP. No birds detected in the YP had crossing times between 35 and 70 h. All predictors were the same as in the departure decision analysis, except that we used weather at the actual time of a bird’s departure.

We explored the correlation matrix for each set of weather variables used in our CART analyses (Tables S1 and S2). Though some variables showed statistically significant correlations with other variables, likely because of our large sample size, the amount of variance explained in one variable by another was generally small. Temperature was the exception, which was significantly correlated with calendar day (|r| = 0.81). We excluded variables from our analysis if |r| ≥ 0.6, because cases below this cutoff represent less than 36% of the variance in one variable being explained by another.

CART is a recursive partitioning method that uses tree-building algorithms to determine a set of if–then logical split conditions/statements that allow sampling units (in this case, birds) to be accurately classified (75). CART is a nonparametric and nonlinear analysis that allows for continuous ordinal and nominal predictor variables in predicting group membership and was appropriate for the types of nonlinear relationships evident in our data. We performed both CARTs using the Gini index, evaluated using v-fold cross-validation, and pruned to the minimal cp error rate. To evaluate the results of each CART, we used CCR, Cohen’s kappa, and a Monte Carlo analysis comparing the tree to random trees. Our CART analyses used functions from the R package rpart and an additional Monte Carlo analysis, wrapped and written by Brad Compton (University of Massachusetts, Amherst, MA) available in the cartware package on www.umass.edu/landeco/teaching/multivariate/labs/multivariate_labs.html. For a full explanation of classification and regression trees, see ref. 75.

To determine if potential nonindependence of multiple individuals departing on the same day might have affected the results of our CART analysis, we also analyzed the relationships between the predictor variables and departure decision (stopover, over-water departure, over-land departure) and arrival status using GLMMs (multinomial distribution with generalized logit link function). In each model we included date as a random effect to account for the potential nonindependence. We compared the outcome from GLMM predicting departure decisions and arrival categories as a function of explanatory variables to the outcome from our CART models, and we found qualitatively the same patterns; the same variables came out as important predictors in our GLMM as in our CARTs. Although the numbers differed slightly, this is expected because CART is better able to deal with nonlinearities, whereas a GLMM constrains relationships to being linear (on the link function scale). We present our analyses in the form of CART models to provide a straightforward and intuitive interpretation of the decision-making process used during departure and affecting arrival at the Yucatan Peninsula.

We performed a GLMM with a negative binomial distribution and a log-link function to examine differences in stopover duration among species. We used a similar GLMM to assess the relationships between crossing time (time between last detection in AL and arrival at the YP) and intrinsic variables. This GLMM included crossing times of all birds that arrived in the YP as the dependent variable, and species, age, and wing length (z-score) as independent variables. The GLMM included over-land departures and birds that stopped over in AL in addition to over-water departures. We used a negative binomial distribution because stopover duration and crossing time were not normally distributed, and attempts to transform the variables did not improve their fit. We converted all times to integers (number of hours) to conform to the assumptions of the analysis, and used a negative binomial rather than Poisson distribution to account for potential overdispersion. We modeled departure day as a random effect to account for the potential nonindependence of multiple individuals departing on the same day.

For trans-Gulf flight duration, we used a GLM with a negative binomial distribution and a log-link function to examine the relationship between trans-Gulf (i.e., direct) flight durations and intrinsic and extrinsic variables. We omitted the random effect of date because of difficulty fitting the full model due to the small sample size. Fitting single-variable GLMM for trans-Gulf flight duration with the random effect of date confirmed that the effect of date was minimal and qualitative results were identical to the full model without date. We included species, age, wing length (z-score), fat, and wind profit over the GOM. We calculated a Gulf-wide wind-profit index by averaging conditions along a simulated track beginning at civil twilight at 1 km altitude with a track direction of 180° and assuming a constant heading (Weather Variables). We restricted our analysis to the subset of arrivals that made direct flights to the YP after departing AL over water on the evening of capture, because this was the only subset of individuals for which we had reliable data on fat and wind conditions during the time of crossing.

We performed CART analyses in R (package rpart), and all remaining statistical tests were performed in SAS 9.4 (PROC GLIMMIX, 91). Departure and arrival data and bird characteristics are archived in Movebank.

Results

Departure Decisions.

We recorded departure behavior and arrival status of 119 Swainson’s Thrushes (SWTH; Catharus ustulatus), 25 Wood Thrushes (WOTH; Hylocichla mustelina), and 100 Red-eyed Vireos (REVI; Vireo olivaceus; SI Methods). Eighty-five percent of birds departed coastal AL on the evening of capture; of these, 43% departed southward (over water; bearing >90° and <270°) and 57% departed over land (bearing ≥270° or ≤90°) in directions consistent with reverse movement or circum-Gulf routes. Fifteen percent of radio-tagged birds did not depart the evening of capture and stopped over at the site for more than 24 h.

Departure decisions varied significantly between vireos and the two thrush species [generalized linear mixed model (GLMM): F2,140 = 8.84, P = 0.0002, thrush species combined]. Most vireos departed over land, whereas most thrushes departed over water (Fig. 2, SI Methods, Fig. S2). The percentage of birds that stopped over (i.e., did not depart the day of capture) was similar among species, although the stopover duration of vireos was more than twice as long as that of either thrush species (median: REVI = 7.5 d, SWTH = 2.6 d, WOTH = 3.2 d; GLMM: F2,10 = 7.60, P = 0.0098). Regardless of species and departure direction, most individuals departed the capture site within an hour after sunset (SI Methods, Fig. S3).

Fig. 2.

Fig. 2.

Percentage of birds selecting each departure decision at our coastal AL study site within 24 h of capture.

Fig. S2.

Fig. S2.

Circular plots illustrating the distribution of departure bearings of (A) Red-eyed Vireo, (B) Swainson’s Thrush, and (C) Wood Thrush.

A combination of fat, species, atmospheric humidity, and 24-h change in humidity predicted departure decisions from AL [Fig. 3; correct classification rate (CCR) = 68.9%, κ = 0.474, P < 0.0001]. Our classification and regression tree (CART) analysis accurately predicted 76.1% of birds that selected over-water departures. On nights when humidity was low (<62%), a condition that generally occurs following the passage of a cold front (36), vireos and thrushes with high fat reserves (scores of 4–5) departed south over the GOM. When humidity was >62%, thrushes only departed over water when they had maximum fat reserves (score of 5). In contrast, vireos generally did not depart over water when humidity was >62%, regardless of how much fat they carried.

Fig. 3.

Fig. 3.

(A) Classification and regression tree illustrating predicted classification of birds’ departure decisions: stopover at coastal AL site for ≥24 h, over-land departure, or over-water departure. Box plots illustrate median and quartiles for variables predicting departure decision from AL: (B) fat scores, (C) humidity, and (D) 24-h change in humidity at sunset on the evening of capture. Negative values for change in humidity denote a drop in humidity, whereas positive values denote an increase in humidity. Sample sizes refer to number of individuals predicted in each class and the proportion reflects the accuracy of classification to each departure group.

Migrants that departed over land or stopped over in AL were correctly classified 66.9% and 62.5% of the time, respectively. Birds of all species with moderate fat reserves (scores of 2–3) and vireos with high fat reserves (scores 4–5) departing on days with humidity >62% were most likely to depart over land. Lean birds (fat scores 0–1) were likely to depart over land if humidity dropped by at least 5% over the previous 24 h, but stopped over if humidity increased.

Arrival at the YP.

The percentage of birds arriving at the YP varied among species; a significantly larger percentage of SWTH and WOTH (31% and 28%, respectively) were detected in the YP than REVI (16%; GLMM: species F2,185 = 27.96, P < 0.0001; Fig. 4). Though there was no overall effect of departure decision on arrival at the YP (GLMM: departure group F2,185 = 1.42, P = 0.2431), there was a significant interaction between species and departure decision (GLMM: species × departure group F3,185 = 99.07, P < 0.0001). Thrushes that departed over water had a greater probability of being detected in the YP than those that did not depart over water (t = 3.08, P = 0.0024); this was not true for Red-eyed Vireos (t = 0.19, P = 0.8479).

Fig. 4.

Fig. 4.

Percentage of radio-tagged birds in each departure category that were detected in the Yucatan Peninsula. Sample sizes for each departure category within each of the three focal species are noted above bars.

Considering only the 90 birds that departed over water the evening of capture, 26 traveled directly to the YP (arrived <35 h after departure from AL), 7 traveled indirectly to the YP (arrived >70 h after departure), and the remaining 57 were not detected. No birds detected in the YP had crossing times between 35 and 70 h. Three variables predicted whether birds initiating over-water flights were detected in the YP: departure date, fat, and wind profit (Fig. 5; CCR = 77.6%, κ = 0.462, P = 0.0068).

Fig. 5.

Fig. 5.

(A) CART illustrating classification of birds among three arrival groups: birds that were not detected in the YP and those that arrived via indirect (>70 h following departure) and direct flights (<35 h following departure). Our CART of arrival at the YP includes only birds that departed over water on the evening of capture. Box plots illustrate median and quartiles for variables predicting arrival at the YP: (B) departure ordinal date, (C) fat score, and (D) wind profit. Positive wind-profit values denote headwinds, whereas negative values denote tailwinds. Larger-magnitude values indicate stronger wind speeds. Weather variables were retrieved for the date and time of departure.

The CART model accurately predicted the arrival of 73.9% of over-water departing birds at the YP following direct, trans-Gulf flights. Birds had a high probability of completing trans-Gulf flights to the YP if they departed after September 24, carried large fat reserves (score of 5), and had a wind profit greater than −2.4 at the time of departure (light headwinds or tailwinds). The model was unable to accurately classify birds that arrived >70 h following departure, likely because conditions at their actual departure location and time were unknown.

The CART correctly classified 76.1% of migrants that departed over water but were not subsequently detected in the YP. Birds that departed before September 24 or departed afterward but with fat scores less than 5 were unlikely to be detected at the YP.

Crossing Time.

When considering all birds that arrived at the YP, REVI took significantly longer to travel between the AL capture site and the YP (median = 208.6 h, n = 16) than SWTH (26.3 h, n = 37) and WOTH (28.4 h, n = 7; GLMM: F1,28 = 4.50, P = 0.0429), because more vireos departed over land and flew indirectly to the YP. Eighty-one percent of all REVI flew indirectly to the YP compared with 41% of SWTH and 29% of WOTH (GLMM: F1,30 = 3.77, P = 0.0616; thrush species combined for analysis). Age and wing length were not related to crossing time when we considered all arrivals (GLMM; age: F1,28 = 0.36, P = 0.5534; wing length: F1,28 = 2.06, P = 0.1620).

Birds that made direct flights across the Gulf on the evening of capture (79% of arrivals), for which we had data on intrinsic and extrinsic variables at the time of crossing, had a significant negative relationship between trans-Gulf flight duration and mean wind profit (GLM: F1,21 = 20.96, P = 0.0002; SI Methods, Fig. S4). There was no relationship between trans-Gulf flight duration and species (F2,21 = 0.10, P = 0.9025), age (F1,21 = 0.09, P = 0.7712), fat (F1,21 = 0.47, P = 0.4986), or wing length (F1,21 = 0.12, P = 0.7378). Trans-Gulf flight times ranged from 14.9 to 34.6 h (mean ± SE: 22.4 ± 5.1 h).

Fig. S4.

Fig. S4.

Relationship between flight time (direct flights only) and mean wind-profit index. Tailwinds are denoted by positive values, whereas headwinds are denoted by negative values. Larger values indicate stronger winds.

Discussion

Our findings support the hypothesis that flexible strategies are adaptive for mitigating the dynamic conditions and risks of ecological barriers during migration (4, 24, 46). This study also advances our understanding of the interactions among intrinsic and extrinsic factors influencing decisions made by small songbirds to navigate potentially risky flights across the GOM. Songbirds departing from coastal Alabama were able to assess departure conditions to take advantage of favorable circumstances for crossing the Gulf safely and quickly.

Previous songbird data in support of this hypothesis along the edges of large geographic features have only considered either departure or arrival behavior (20, 31, 40, 41, 47), and the proximate cues and decision rules used to identify favorable conditions have been unclear (40). By detecting songbirds after negotiating ∼1,000 km of open water, we identified conditions affecting both departure and arrival, providing strong empirical support for the hypothesis that conditions that affect departure also influence crossing behaviors. We also identified criteria, or cutoff values, which may define birds’ decision rules. In particular, fat, humidity, and wind profit are the proximate factors associated with movements across the GOM to the YP. The tight relationship between trans-Gulf (direct) flight durations and wind profit over the GOM further emphasizes the advantages accrued by exploiting favorable environmental conditions.

For the species we studied, fat score was the primary variable related to departure decision and arrival at the YP, suggesting that the amount of fat a bird carries is a key determinant of successful crossing. Birds departing the northern Gulf coast with large fat reserves have a larger buffer for dealing with en route exigencies, such as deteriorating weather conditions over open water, thereby improving their probability of arrival at the YP (31). Thus, birds may exert some control over their ability to mitigate risk associated with crossing the Gulf by increasing fat reserves before departure, provided they can locate and acquire food resources (48, 49). The importance of fat in shaping departure behavior (direction and stopover duration) at other coastal sites and islands suggests that our findings are applicable to migration across other geographic features (31, 47).

The importance of fat in shaping both departure decisions and detection of over-water departures at the YP also underscores the value of high-quality habitat along the edge of geographic features that offer few, if any, refueling opportunities. In coastal areas, where human impacts are high, foraging opportunities may be reduced, limiting birds’ ability to gain sufficient fat for nonstop flights over water (50). Birds do not gain mass at our capture site in AL (51), and a large percentage of lean birds depart the site over land (20), supporting the hypothesis that limited foraging opportunities along the coast may affect birds’ likelihood of crossing the Gulf. Though interior sites may offer suitable refueling options for birds arriving at the coast with insufficient fat for crossing, our findings suggest that birds departing over land require significantly longer to arrive at the YP (vireos) or have a lower probability of arriving at the YP (thrushes) than birds departing over water.

Although extremely important, energetic reserves alone did not explain differences in migratory behaviors. Low humidity, wind profit (tailwinds or light headwinds), and ordinal day also were associated with songbirds’ departure decisions from coastal AL as well as whether and when over-water departing birds arrived at the YP. Low (and dropping) humidity, clear skies, cooler temperatures, and southward winds are typical following the passage of a cold front (36). The tailwinds associated with this synoptic weather pattern are particularly valuable to migrants crossing the Gulf, because they increase ground speed and reduce travel time and energy expenditure (36, 38, 52). Additionally, the clear skies of low-humidity nights provide good visibility for departure orientation; clear skies correlate with more departures and less orientation scatter (53). Ordinal day is likely important for predicting the arrival of over-water departures because synoptic weather systems favoring trans-Gulf flights (i.e., strong cold fronts moving into the GOM) become more common later in the fall (late September through October) (19, 27, 54). The fate of birds departing over water before September 24 is unknown; they may have died due to unfavorable conditions or passed through other areas of the southern Gulf Basin.

Our study capitalized on our ability to detect small songbirds in the YP following their departure from coastal AL to provide insight into the factors and conditions that minimize risk of crossing the GOM and influence its role as an ecological barrier. We were unable to determine the fate of birds that were not detected in the YP, but it is expected that this subset of birds includes some radio-tagged individuals that successfully arrived elsewhere in the southern Gulf Basin, e.g., Cuba, and do not solely represent birds that died. This conclusion is supported by the overlap in the range of departure dates, fat, and wind profit between birds that arrived at the YP following trans-Gulf flights and those not detected in the YP (Fig. 5 B–D). It is reasonable to expect that similar conditions (e.g., large fat reserves, wind profit in the direction of travel to those areas) contribute to their ability to negotiate the GOM to arrive elsewhere.

Against our prediction, age appeared unrelated to the decision-making process. Provided young birds have sufficient fat and depart over water under favorable conditions, their likelihood of arrival at the YP, crossing times, and trans-Gulf flight durations are indistinguishable from those of adult birds, demonstrating that they are able to manage the risks associated with the GOM the first time then encounter it. Young and adult birds also are equally likely to cross the GOM in the spring (34). In contrast to our findings, young songbirds making short (average 35–65 km) migratory flights across water and along coastlines in New Brunswick, Canada, had longer flight durations than adults, because they departed under a broader range of wind conditions, including less-supportive winds (55). The lack of selectivity in departure conditions by young birds in New Brunswick may allow them to depart the breeding area quickly, providing benefits such as reduced exposure to predators, cold temperatures, and resource competition that outweigh energy and time savings associated with waiting for favorable winds (55). However, for birds crossing large water features like the GOM, the energy costs and mortality risk of crossing under less-profitable winds likely exert a much greater selection pressure on departure behavior, restricting the range of weather conditions under which migrants depart. The similar response of young and adult birds in the GOM may be due to innately programmed decision rules, experience gained before arrival at on the northern coast of the GOM, or both.

Species differed in their migratory behaviors. Thrushes that initially departed over water had a higher probability of detection in the YP than birds departing over land or stopping over in coastal AL. In contrast, vireos had a similar likelihood of detection in the YP regardless of whether they departed the capture site over water or over land. This finding suggests that vireos did not have suitable conditions for traveling to the YP directly from the AL site, but were able to locate appropriate conditions elsewhere that allowed them to arrive at a later date. An important implication of this finding is that departure direction from a stopover site may not be a reliable indicator of a bird’s endpoint or route, as many vireos departed to the north, a seasonally inappropriate direction, yet were detected in the YP.

In general, REVI negotiated risk associated with the GOM differently than the two thrush species, which responded similarly. The GOM may function more as a barrier to vireos. Vireos departed over land more frequently than thrushes, likely in response to greater constraints in the weather conditions permitting over-water departure (specifically humidity), had a lower overall likelihood of arrival at the YP, took almost seven times longer to cross the Gulf, and had arrival patterns that were less influenced by departure decisions. The differences between vireos and thrushes are not explained by initial fat reserves, biogeography, or flight morphology. Mean fat scores of vireos were significantly higher than those of SWTH but similar to those of WOTH (SI Methods, Fig. S5). Previous studies have shown that REVI and SWTH captured in coastal AL with large fat reserves orient toward South America (28, 56), and geolocator studies confirm that both species moving through the northern coast of the GOM arrive at South American winter grounds (23, 57). Based on differences in species’ flight morphologies (wing loadings and aspect ratios), vireos are expected to be more energy-efficient during long-distance flights than either thrush species, particularly WOTH (58, 59), and less susceptible to weather effects (9); thus we expected the GOM to represent less of a barrier to vireos. An alternative explanation is that the smaller body size of vireos translates into slower airspeeds (38) and consequently longer flights and a greater probability of exposure to inclement weather; however, our comparison of trans-Gulf flight times demonstrated no difference among species.

Fig. S5.

Fig. S5.

Distribution of fat scores upon initial capture and tagging for the three focal species. A GLM with a multinomial distribution and generalized logit link function and capture data included as a random effect showed that Swainson’s Thrush had significantly lower average fat scores (mean ± SE: 2.89 ± 0.19) than both Red-eyed Vireos (3.47 ± 0.26) and Wood Thrush (4.02 ± 0.48; F2,191 = 4.29, P = 0.0151). Average fat scores were similar for REVIs and WOTH. We attempted to tag birds across fat classes in the three species.

Wintering ecology, diet, and habitat requirements may explain the differences observed in the species migratory behavior. REVI winter in mixed-species flocks and are not territorial (60, 61); thus, their arrival time may be less constrained, allowing them to wait for more favorable conditions and avoid the risks of crossing a large body of water. Conversely, SWTH and WOTH defend winter territories (62, 63), and the benefits of taking quicker, more direct routes to acquire higher quality territories may outweigh the potential risk of encountering inclement weather over water. Additionally, although all three species are known to consume fruit during fall migration and on the wintering grounds, REVIs typically glean insectivorous prey from broad-leaved canopy foliage (48, 64, 65), whereas Swainson’s Thrush and Wood Thrush forage on the ground and in low vegetation (66, 67). The scrub-shrub and pine vegetation along the Fort Morgan (FTM) Peninsula may be more at odds with vireos’ broad-leaved canopy foraging preferences than the understory preferences of the thrush species. REVIs may have departed over land in search of vegetation better suited to their foraging habits.

During migration, birds encounter and respond to spatiotemporally fluctuating landscapes. Consequently, they are constantly assessing risk by comparing alternative behaviors, whether it be when or where to land, in which vegetation type to settle, which food resources to consume, or when and in what direction to depart from a stopover site to minimize fitness costs. A major finding of this study is that songbirds encountering the GOM appear to mitigate risk in relation to crossing the Gulf by departing with large fat reserves when weather variables signal favorable flight conditions. Birds have some control over fat gain and can decide when and in which direction to depart when facing different weather conditions. By adjusting their behaviors, they can exert some control over the extent to which the GOM functions as a barrier and inhibits or facilitates safe, timely, and energy-efficient movement toward their wintering destinations.

Materials

Capture and Tagging Methods.

We captured, radio-tagged, and gathered intrinsic measurements (age, wing chord length, fat score) on Red-eyed Vireos, Swainson’s Thrushes, and Wood Thrushes at a long-term banding station in the Bon Secour National Wildlife Refuge (30° 13′ 49″ N, 88° 0′ 13″ W) on the FTM Peninsula, AL, from September 2 to October 28, 2009–2013. The FTM Peninsula is located directly south of Mobile Bay along the northern coast of the GOM (Fig. 1, SI Methods, Fig. S1). We fitted analog pulse transmitters to birds using a modified adhesive approach (20, 68) (SI Methods).

Automated Tracking in the GOM.

We operated three automated radio-tracking towers along the coast within 7.5 km of our AL capture site each season (SI Methods, Fig. S1). We mounted six 3-element directional Yagi antennas at 60° intervals on each tower to estimate birds’ departure direction in degrees and classify departures as over water or over land (69, 70). In 2012 we added a tracking tower to Dauphin Island (7.5 km west of capture site), and in 2013 we added a tower to the north of Mobile Bay; the latter was equipped with four high-gain, stacked directional antennas to improve detection of birds that departed over land. We used automated receiving units (JDJC Corp.) to autonomously monitor transmitter frequencies at 0.5- to 4-min intervals.

We established a “telemetry fence” along the entire northern coast of the YP using seven tracking towers, each equipped with high-gain stacked antennas identical to the ones used in AL (Fig. 1). We oriented two high-gain antennas toward 90° and 270°, roughly parallel to the northern YP coastline, to detect radio-tagged birds as they arrived. We spaced towers an average of 57 ± 7 km (±SD) apart along the YP coast to maintain continuous detection. We sampled each transmitter frequency at 3-to 6-min intervals in the YP.

Departure and Arrival Date, Time, and Direction.

We conservatively detected radio-tagged birds in AL and the YP based on signal strength, background noise, pulse width, pulse interval, measured frequency, and temporal pattern of detections (71) (SI Methods). In AL, when sufficient data were available from multiple towers, we used biangulation or triangulation to estimate each bird’s departure direction and time based on the last five track coordinates (20). When insufficient data were available for track estimation, we estimated vanishing bearings based on changes in signal strength across the six antennas of the strongest tower. We used the last detection to estimate departure bearing and time. Stopover duration was conservatively estimated as the difference between capture date and date of departure determined by our telemetry data. When a bird was detected in the YP, we estimated arrival time as the time of peak signal strength on the east or west antenna of the strongest tower. Departure and arrival data are archived in Movebank (72).

Weather Data.

We retrieved weather variables from the National Center for Environmental Prediction North American Regional Reanalysis dataset (73) via the Environmental Data Automated Track Annotation System (Env-DATA) service available on www.movebank.org (72, 74). Variables were interpolated to the nearest time, location, and altitude. We examined correlations among weather variables and only retained variables with |r| < 0.60 (Tables S1 and S2). We retrieved the following variables predictive of synoptic weather systems: (i) surface-level humidity; (ii) surface-level barometric pressure; (iii) 24-h change in surface humidity (negative values indicate a drop in humidity, positive values an increase); (iv) 24-h change in surface pressure (sign denotes direction of change); (v) wind speed at four altitudes (1, 2, 3, and 4 km); and (vi) wind direction at four altitudes. We used wind speeds and directions to calculate an average wind-profit index. Wind profit was defined as the speed (m/s) of wind toward 180° (i.e., how favorable winds were for crossing to the YP; see formula for wind profit in ref. 46). Positive wind-profit values denote favorable winds for crossing to the YP (i.e., tailwinds), whereas negative values denote unfavorable winds (i.e., headwinds). Larger-magnitude values indicate stronger wind speeds. For the CART analyses we calculated an average wind profit across the four altitudes at the capture site at civil twilight or the time of birds’ departure. For the analysis of trans-Gulf flight duration we calculated a Gulf-wide wind-profit index by averaging conditions across the four altitudes along a simulated track beginning at civil twilight with a track direction of 180° and assuming a constant heading (SI Methods).

Statistical Analyses.

We used GLMM with a multinomial distribution and generalized logit link function to examine differences in departure decisions among species and to assess patterns of arrival (direct flight, indirect flight, not detected) in relation to species and departure decision. We modeled departure day as a random effect to account for potential nonindependence of multiple individuals departing on the same day.

We performed a CART analysis to determine which extrinsic and intrinsic factors predicted each of the three departure decisions in AL: over-water departure, over-land departure, or stopover (75). Model predictors included species, fat score (0–5) (76), age, relative wing length (z-score, calculated within each species), ordinal date, relative humidity, 24-h change in humidity, barometric pressure, 24-h change in pressure, and wind profit at civil twilight on the date of capture to objectively compare weather available to birds that departed on the evening of capture and those that did not (SI Methods). Most migrants depart at civil twilight and likely decide whether to depart based on conditions at that time (20).

We performed a second CART analysis to determine which factors were related to the detection of over-water departing birds in the YP, distinguishing between birds that arrived following a direct, trans-Gulf flight and those arriving >70 h after departure from AL, either via a delayed trans-Gulf flight or a detour around the Gulf. We restricted this analysis to birds that departed over water on the day of capture, because they were the only birds for which we had reliable fat scores and weather conditions at the time of departure and because the departure behavior of these birds suggested they selected a trans-Gulf route. We classified over-water departures into one of three groups: direct flight, indirect flight, and nonarrival. All predictors were the same as in the departure decision analysis, except that we used weather conditions at the actual time of birds’ departure.

We performed a GLMM with a negative binomial distribution and a log-link function to compare stopover duration among species and predict crossing time (represented as number of hours) as a function of intrinsic and extrinsic variables. We modeled departure day as a random effect to account for multiple individuals departing on the same day. For trans-Gulf flight duration, we used a GLM with a negative binomial distribution and log-link function but omitted the random effect of date because of difficulty fitting the full model due to the small sample size. Single-variable GLMM for trans-Gulf flight duration with the random effect of date produced qualitative results identical to the full model without date. We computed all statistics in R 2.15.2 (package rpart; SI Methods) or SAS 9.4 (PROC GLIMMIX).

Acknowledgments

We thank our numerous field technicians at The University of Southern Mississippi Fort Morgan Peninsula Banding Station, and Melgar Tabasco and Waldemar Santamaria for assistance in the Yucatan Peninsula. We also thank our many site partners around the Gulf for granting permission to install and operate telemetry equipment. Thanks to Sarah Davidson (Movebank curator) and Rolf Weinzierl who assisted us with archiving our movement data and accessing atmospheric data. Janet Ruth and three anonymous reviewers provided valuable feedback on earlier drafts of this paper. This work was supported by the National Science Foundation (NSF) (IOS Awards 1147096, 1145952, and 1147022), National Geographic Society Committee on Research and Exploration (Award 8971-11), American Ornithologists’ Union (J.A.S.), Birmingham Audubon Society (J.A.S.), Eastern Illinois University (Research and Creative Activity Awards to J.L.D. and L.N.S.), University of Illinois Urbana‐Champaign, and The University of Southern Mississippi. G.B. was supported by NASA Award NNX11AP61G. T.J.Z. was supported by a NSF GK-12 Program Award (0947944). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the US Government.

Footnotes

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. M.W. is a Guest Editor invited by the Editorial Board.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1503381112/-/DCSupplemental.

2Deceased August 20, 2010.

References

  • 1.Lank DB, Butler RW, Ireland J, Ydenberg RC. Effects of predation danger on migration strategies of sandpipers. Oikos. 2003;103(2):303–319. [Google Scholar]
  • 2.Gill RE, Jr, et al. 2009. Extreme endurance flights by landbirds crossing the Pacific Ocean: Ecological corridor rather than barrier? Proc R Soc B Biol Sci 276(1656):447-457.
  • 3.Cryan PM. Wind turbines as landscape impediments to the migratory connectivity of bats. Environ Law. 2011;41:355–370. [Google Scholar]
  • 4.Mellone U, López-López P, Limiñana R, Urios V. Weather conditions promote route flexibility during open ocean crossing in a long-distance migratory raptor. Int J Biometeorol. 2011;55(4):463–468. doi: 10.1007/s00484-010-0368-3. [DOI] [PubMed] [Google Scholar]
  • 5.Caudill CC, et al. Indirect effects of impoundment on migrating fish: Temperature gradients in fish ladders slow dam passage by adult Chinook salmon and steelhead. PLoS One. 2013;8(12):e85586. doi: 10.1371/journal.pone.0085586. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Winkler DW, et al. Cues, strategies, and outcomes: How migrating vertebrates track environmental change. Mov Ecol. 2014;2(1):10. [Google Scholar]
  • 7.Faraci FM. Adaptations to hypoxia in birds: How to fly high. Annu Rev Physiol. 1991;53:59–70. doi: 10.1146/annurev.ph.53.030191.000423. [DOI] [PubMed] [Google Scholar]
  • 8.Alerstam T. Detours in bird migration. J Theor Biol. 2001;209(3):319–331. [Google Scholar]
  • 9.Saino N, et al. Spring migration decisions in relation to weather are predicted by wing morphology among trans-Mediterranean migratory birds. Funct Ecol. 2010;24(3):658–669. [Google Scholar]
  • 10.Lowry GH., Jr Trans-gulf spring migration of birds and the coastal hiatus. Wilson Bull. 1945;57(2):92–101. [Google Scholar]
  • 11.Felicísimo AM, Muñoz J, González-Solis J. Ocean surface winds drive dynamics of transoceanic aerial movements. PLoS One. 2008;3(8):e2928. doi: 10.1371/journal.pone.0002928. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Nunn AD, Cowx IG. Restoring river connectivity: Prioritizing passage improvements for diadromous fishes and lampreys. Ambio. 2012;41(4):402–409. doi: 10.1007/s13280-012-0281-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Rappole JH. Ecology of Migrant Birds: A Neotropical Perspective. Smithsonian Inst Press; Washington, DC: 1995. [Google Scholar]
  • 14.Newton I. The Migration Ecology of Birds. Academic; London: 2008. [Google Scholar]
  • 15.Diehl RH, Bates JE, Willard DE, Gnoske TP. The hazards of nocturnal migration over Lake Michigan: A case study. Wilson J Ornithol. 2014;126(1):19–29. [Google Scholar]
  • 16.Wiedenfeld DA, Wiedenfeld MG. Large kill of neotropical migrants by tornado and storm in Louisiana, April 1993. J Field Ornithol. 1995;66(1):70–80. [Google Scholar]
  • 17.Paynter RA. Autumnal migrants on the Campeche Bank. Auk. 1953;70(3):338–349. [Google Scholar]
  • 18.Springer S. 1938. Notes on the sharks of Florida. Proc FL Acad Sci 3(1):9-41.
  • 19.Able KP. Fall migration in coastal Louisiana and the evolution of migration patterns in the Gulf region. Wilson Bull. 1972;84(3):231–242. [Google Scholar]
  • 20.Smolinsky JA, Diehl RH, Radzio TA, Delaney D, Moore F. Factors influencing the movement biology of migrant songbirds confronted with an ecological barrier. Behav Ecol Sociobiol. 2013;67(12):2041–2051. [Google Scholar]
  • 21.Gauthreaux SA. A radar and direct visual study of passerine spring migration in southern Louisiana. Auk. 1971;88(2):343–365. [Google Scholar]
  • 22.Deppe JL, Rotenberry JT. Temporal patterns in fall migrant communities in Yucatan, Mexico. Condor. 2005;107(2):229–245. [Google Scholar]
  • 23.Delmore KE, Fox JW, Irwin DE. 2012. Dramatic intraspecific differences in migratory routes, stopover sites and wintering areas, revealed using light-level geolocators. Proc R Soc B Biol Sci 279(1747):4582–4589.
  • 24.Stanley CQ, MacPherson M, Fraser KC, McKinnon EA, Stutchbury BJM. Repeat tracking of individual songbirds reveals consistent migration timing but flexibility in route. PLoS One. 2012;7(7):e40688. doi: 10.1371/journal.pone.0040688. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Fraser KC, et al. Consistent range-wide pattern in fall migration strategy of purple martin (Progne subis), despite different migration routes at the Gulf of Mexico. Auk. 2013;130(2):291–296. [Google Scholar]
  • 26.Deppe JL, Rotenberry JT. Scale-dependent habitat use by fall migratory birds: Vegetation architecture, floristics, and geographic consistency. Ecol Monogr. 2008;78(3):461–487. [Google Scholar]
  • 27.Buskirk WH. Influence of meteorological patterns and trans-gulf migration on the calendars of latitudinal migrants. In: Keast A, Morton ES, editors. Migrant Birds in the Neotropics: Ecology, Behavior, Distribution and Conservation. Smithsonian Inst Press; Washington, DC: 1980. pp. 485–491. [Google Scholar]
  • 28.Sandberg R, Moore FR. Migratory orientation of red-eyed vireos, Vireo olivaceus, in relation to energetic condition and ecological context. Behav Ecol Sociobiol. 1996;39(1):1–10. [Google Scholar]
  • 29.Ydenberg RC, Butler RW, Lank DB. Effects of predator landscapes on the evolutionary ecology of routing, timing and molt by long-distance migrants. J Avian Biol. 2007;38(5):523–529. [Google Scholar]
  • 30.Alerstam T. Optimal bird migration revisited. J Ornithol. 2011;152(Suppl1):5–23. [Google Scholar]
  • 31.Schmaljohann H, Naef-Daenzer B. Body condition and wind support initiate the shift of migratory direction and timing of nocturnal departure in a songbird. J Anim Ecol. 2011;80(6):1115–1122. doi: 10.1111/j.1365-2656.2011.01867.x. [DOI] [PubMed] [Google Scholar]
  • 32.Jenni L, Jenni-Eiermann S. Fuel supply and metabolic constraints in migrating birds. J Avian Biol. 1998;29(4):521–528. [Google Scholar]
  • 33.Thorup K, Alerstam T, Hake M, Kjellén N. 2003. Bird orientation: Compensation for wind drift in migrating raptors is age dependent. Proc R Soc B Biol Sci 270(Suppl1):S8–S11. [DOI] [PMC free article] [PubMed]
  • 34.McKinnon EA, Fraser KC, Stanley CQ, Stutchbury BJM. Tracking from the tropics reveals behaviour of juvenile songbirds on their first spring migration. PLoS One. 2014;9(8):e105605. doi: 10.1371/journal.pone.0105605. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Bowlin MS, Wikelski M. Pointed wings, low wingloading and calm air reduce migratory flight costs in songbirds. PLoS One. 2008;3(5):e2154. doi: 10.1371/journal.pone.0002154. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Richardson WJ. Timing of bird migration in relation to weather: Updated review. In: Gwinner E, editor. Bird Migration: Physiology and Ecophysiology. Springer; Berlin: 1990. pp. 78–101. [Google Scholar]
  • 37.Cochran WW, Kjos CJ. Wind drift and migration of thrushes: A telemetry study. IL Nat Hist Surv Bull. 1985;33(1):297–330. [Google Scholar]
  • 38.Pennycuick CJ. Bird Flight Performance. Oxford Univ Press; New York: 1989. [Google Scholar]
  • 39.Cochran WW, Wikeski M. Individual migratory tactics of New World Catharus thrushes: Current knowledge and future tracking options from space. In: Greenberg R, Marra PP, editors. Birds of Two Worlds: The Ecology and Evolution of Migratory Birds. Johns Hopkins Univ Press; Baltimore: 2005. pp. 274–289. [Google Scholar]
  • 40.Shamoun-Baranes J, van Gasteren H. Atmospheric conditions facilitate mass migration events across the North Sea. Anim Behav. 2011;81(4):691–704. [Google Scholar]
  • 41.Woodworth BK, Mitchell GW, Norris DR, Francis CM, Taylor PD. Patterns and correlates of songbird movements at an ecological barrier during autumn migration assessed using landscape- and regional-scale automated radiotelemetry. Ibis. 2015;157(2):326–339. [Google Scholar]
  • 42.McKinnon EA, Fraser KC, Stutchbury BJM. New discoveries in landbird migration using geolocators and a flight plan for the future. Auk. 2013;130(2):1–12. [Google Scholar]
  • 43.Hallworth MT, Marra PP. Miniaturized GPS tags identify non-breeding territories of a small breeding migratory songbird. Sci Rep. 2015;5:11069. doi: 10.1038/srep11069. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Wikelski M, et al. Simple rules guide dragonfly migration. Biol Lett. 2006;2(3):325–329. doi: 10.1098/rsbl.2006.0487. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Rappole JH, Ramos MA. Factors affecting migratory bird routes over the Gulf of Mexico. Bird Conserv Int. 1994;4(2-3):251–262. [Google Scholar]
  • 46.Erni B, Liechti F, Bruderer B. The role of wind in passerine autumn migration between Europe and Africa. Behav Ecol. 2005;16(4):732–740. [Google Scholar]
  • 47.Goymann W, Spina F, Ferri A, Fusani L. Body fat influences departure from stopover sites in migratory birds: Evidence from whole-island telemetry. Biol Lett. 2010;6(4):478–481. doi: 10.1098/rsbl.2009.1028. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Loria DE, Moore FR. Energy demands of migration on red-eyed vireos, Vireo olivaceus. Behav Ecol. 1990;1(1):24–35. [Google Scholar]
  • 49.Alerstam T. Ecological causes and consequences of bird orientation. Experientia. 1990;46(4):405–415. doi: 10.1007/978-3-0348-7208-9_10. [DOI] [PubMed] [Google Scholar]
  • 50.Barrow W, Jr, Chen CC, Hamilton RB, Ouchley K, Spengler TJ. Disruption and restoration of en route habitat, a case study: The Chenier Plain. In: Moore FR, editor. Stopover Ecology of Nearctic-Neotropical Landbird Migrants: Habitat Relations and Conservation Implications. Allen; Lawrence, KS: 2000. pp. 71–87. [Google Scholar]
  • 51.Woodrey MS, Moore FR. Age-related differences in the stopover of fall landbird migrants on the coast of Alabama. Auk. 1997;114(4):695–707. [Google Scholar]
  • 52.Liechti F. Birds: Blowin’ by the wind? J Ornithol. 2006;147(2):202–211. [Google Scholar]
  • 53.Åkesson S, Walinder G, Karlsson L, Ehnbom S. Reed warbler orientation: Initiation of nocturnal migratory flights in relation to visibility of celestial cues at dusk. Anim Behav. 2001;61(1):181–189. doi: 10.1006/anbe.2000.1562. [DOI] [PubMed] [Google Scholar]
  • 54.Gauthreaux SA, Michi JE, Belser CG. The temporal and spatial structure of the atmosphere and the evolution of bird migration. In: Greenberg R, Marra PP, editors. Birds of Two Worlds: The Ecology and Evolution of Migratory Birds. Johns Hopkins Univ Press; Baltimore: 2005. pp. 222–234. [Google Scholar]
  • 55.Mitchell GW, Woodworth BK, Taylor PD, Norris DR. Automated telemetry reveals age specific differences in flight duration and speed are driven by wind conditions in a migratory songbird. Mov Ecol. 2015;3(1):19. doi: 10.1186/s40462-015-0046-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Sandberg R, Moore FR, Bäckman J, Löhmus M. Orientation of nocturnally migrating Swainson’s Thrush at dawn and dusk: Importance of energetic condition and geomagnetic cues. Auk. 2002;119(1):201–209. [Google Scholar]
  • 57.Callo PA, Morton ES, Stutchbury BJM. Prolonged spring migration in the Red-eyed Vireo. Auk. 2013;130(2):240–246. [Google Scholar]
  • 58.Yong W, Moore FT. Flight morphology, energetic condition, and the stopover biology of migrating thrushes. Auk. 1994;111(3):683–692. [Google Scholar]
  • 59.Andrews CB, Mackenzie SA, Gregory TR. 2009. Genome size and wing parameters in passerine birds. Proc R Soc B Biol Sci 276(1654):55–61. [DOI] [PMC free article] [PubMed]
  • 60.Morton ES. Adaptations to seasonal changes by migrant land birds in the Panama Canal zone. In: Keast A, Morton ES, editors. Migrant Birds in the Neotropics: Ecology, Behavior, Distribution, and Conservation. Smithsonian Inst Press; Washington, DC: 1980. pp. 437–453. [Google Scholar]
  • 61.Ridgely RS, Tudor G. The Birds of South America. Vol 1 Univ of Texas Press; Austin: 1989. [Google Scholar]
  • 62.Winker K, Rappole JH, Ramos MA. Population dynamics of the Wood Thrush in southern Veracruz, Mexico. Condor. 1990;92(2):444–460. [Google Scholar]
  • 63.Blake JG, Loiselle BA. Habitat use by Neotropical migrants at La Selva Biological Station and Braulio Carrillo National Park, Costa Rica. In: Hagan J, Johnston DW, editors. Ecology and Conservation of Neotropical Landbird Migrants. Smithsonian Inst Press; Washington, DC: 1992. pp. 257–272. [Google Scholar]
  • 64.Holmes RT, Black CP, Sherry TW. Comparative population bioenergetics of three insectivorous passerines in a deciduous forest. Condor. 1979;81(1):9–20. [Google Scholar]
  • 65.Cohen EB, Moore FR, Fischer RA. Experimental evidence for the interplay of exogenous and endogenous factors on the movement ecology of a migrating songbird. PLoS One. 2012;7(7):e41818. doi: 10.1371/journal.pone.0041818. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 66.Holmes RT, Bonney JRE, Pacala SW. Guild structure of the Hubbard Brook bird community: A multivariate approach. Ecology. 1979;60(3):512–520. [Google Scholar]
  • 67.Yong W, Moore FR. Long-distance bird migrants adjust their foraging behavior in relation to energy stores. Acta Zool Sinica. 2005;51(1):12–23. [Google Scholar]
  • 68.Raim A. A radio transmitter attachment for small passerines. Bird Banding. 1978;49(4):326–332. [Google Scholar]
  • 69.Larkin RP, Raim A, Diehl RH. Performance of a non-rotating direction-finder for automatic radio tracking. J Field Ornithol. 1996;67(1):59–71. [Google Scholar]
  • 70.Kays R, et al. Tracking animal location and activity with an automated radio telemetry system in a tropical forest. Comput J. 2011;54(12):1–18. [Google Scholar]
  • 71.Ward MP, Sperry JH, Weatherhead PJ. Evaluation of automated radio telemetry for quantifying movements and home ranges of snakes. J Herpetol. 2013;47(2):337–345. [Google Scholar]
  • 72.Wikelski M, Kays R. Movebank: Archive, analysis and sharing of animal movement data. 2014 Available at www.movebank.org.
  • 73.Mesinger F, et al. North American Regional Reanalysis. Bull Am Met Soc. 2006;87(3):343–360. [Google Scholar]
  • 74.Dodge S, et al. The environmental-data automated track annotation (Env-DATA) system: Linking animal tracks with environmental data. Mov Ecol. 2013;1(1):3. doi: 10.1186/2051-3933-1-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 75.De’ath G, Fabricius KE. Classification and regression trees: A powerful yet simple technique for ecological data analysis. Ecol. 2000;81(11):3178–3192. [Google Scholar]
  • 76.Helms CW, Drury WH. Winter and migratory weight and fat field studies on some North American buntings. Bird Banding. 1960;31(1):1–40. [Google Scholar]
  • 77.Cimprich DA, Moore FR, Guilfoyle MP. Red-eyed Vireo (Vireo olivaceus) In: Poole A, editor. The Birds of North America Online. Cornell Lab of Ornithology; Ithaca, NY: 2000. [Google Scholar]
  • 78.Mack DE, Yong W. Swainson’s Thrush (Catharus ustulatus) In: Poole A, editor. The Birds of North America Online. Cornell Lab of Ornithology; Ithaca, NY: 2000. [Google Scholar]
  • 79.Evans M, Gow E, Roth RR, Johnson MS, Underwood TJ. Wood Thrush (Hylocichla mustelina) In: Poole A, editor. The Birds of North America Online. Cornell Lab of Ornithology; Ithaca, NY: 2011. [Google Scholar]
  • 80.Cochran WW. Long-distance tracking of birds. In: GalIer SR, Schmidt-Koenig K, Jacobs GJ, BelIeville RE, editors. Animal Orientation and Navigation. US Government Printing Office; Washington, DC: 1972. pp. 39–59. [Google Scholar]
  • 81.Cochran WW. Orientation and other migratory behaviors of a Swainson’s Thrush followed for 1500 km. Anim Behav. 1987;35(3):927–929. [Google Scholar]
  • 82.Fuchs T, Haney A, Jechura TJ, Moore FR, Bingman VP. Daytime naps in night-migrating birds: Behavioural adaptation to seasonal sleep deprivation in the Swainson’s thrush, Catharus ustulatus. Anim Behav. 2006;72(4):951–958. [Google Scholar]
  • 83.Owen JC, Moore FR. Relationship between energetic condition and indicators of immune function in thrushes during spring migration. Can J Zool. 2008;86(7):638–647. [Google Scholar]
  • 84.US Fish and Wildlife Service . Bon Secour National Wildlife Refuge Comprehensive Conservation Plan. US Fish and Wildlife Service; Atlanta: 2005. [Google Scholar]
  • 85.Zenzal TJ, Fish AC, Jones TM, Ospina EA, Moore F. Observations of predation and anti-predator behavior of Ruby-throated Hummingbirds during migratory stopover. Southeast Nat. 2013;12(4):N21–N25. [Google Scholar]
  • 86.Pyle P. Identification Guide to North American Birds, Part 1. Slate Creek; Bolinas, CA: 1997. [Google Scholar]
  • 87.Ying L, Liu H, Wei M, Gong S. A novel slot yagi-like multilayered antenna with high gain and large bandwidth. IEEE Antennas Wirel Propag Lett. 2014;13:790–793. [Google Scholar]
  • 88.Dodge S, et al. Environmental drivers of variability in the movement ecology of turkey vultures (Cathartes aura) in North and South America. Philos Trans R Soc Lond B Biol Sci. 2014;369(1643):20130195. doi: 10.1098/rstb.2013.0195. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 89.Åkesson S, Hedenström A. Wind selectivity of migratory flight departures in birds. Behav Ecol Sociobiol. 2000;47(3):140–144. [Google Scholar]
  • 90.Bellrose FC. The distribution of nocturnal migrants in the airspace. Auk. 1971;88(2):397–424. [Google Scholar]
  • 91.SAS Institute Inc . What’s New in SAS 9.3. SAS Institute, Inc.; Cary, NC: 2012. [Google Scholar]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences

RESOURCES