Correlating seabird movements with ocean winds: linking satellite telemetry with ocean scatterometry
- First Online:
- Cite this article as:
- Adams, J. & Flora, S. Mar Biol (2010) 157: 915. doi:10.1007/s00227-009-1367-y
Satellite telemetry studies of the movements of seabirds are now common and have revealed impressive flight capabilities and extensive distributions among individuals and species at sea. Linking seabird movements with environmental conditions over vast expanses of the world’s open ocean, however, remains difficult. Seabirds of the order Procellariiformes (e.g., petrels, albatrosses, and shearwaters) depend largely on wind and wave energy for efficient flight. We present a new method for quantifying the movements of far-ranging seabirds in relation to ocean winds measured by the SeaWinds scatterometer onboard the QuikSCAT satellite. We apply vector correlation (as defined by Crosby et al. in J Atm Ocean Tech 10:355–367, 1993) to evaluate how the trajectories (ground speed and direction) for five procellariiform seabirds outfitted with satellite transmitters are related to ocean winds. Individual seabirds (Sooty Shearwater, Pink-footed Shearwater, Hawaiian Petrel, Grey-faced Petrel, and Black-footed Albatross) all traveled predominantly with oblique, isotropic crossing to quartering tail-winds (i.e., 105–165° in relation to birds’ trajectory). For all five seabirds, entire track line trajectories were significantly correlated with co-located winds. Greatest correlations along 8-day path segments were related to wind patterns during birds’ directed, long-range migration (Sooty Shearwater) as well as movements associated with mega-scale meteorological phenomena, including Pacific Basin anticyclones (Hawaiian Petrel, Grey-faced Petrel) and eastward-propagating north Pacific cyclones (Black-footed Albatross). Wind strength and direction are important factors related to the overall movements that delineate the distribution of petrels at sea. We suggest that vector correlation can be used to quantify movements for any marine vertebrate when tracking and environmental data (winds or currents) are of sufficient quality and sample size. Vector correlation coefficients can then be used to assess population—or species-specific variability and used to test specific hypotheses related to how animal movements are associated with fluid environments.
Marine vertebrates, including fish, turtles, pinnipeds, cetaceans, and seabirds, exist within a turbulent environment of ocean currents and near-surface winds (Schneider 1991). At sea, seabirds of the order Procellariiformes (e.g., petrels: gadfly petrels, shearwaters, and albatrosses) dwell between the wave-roughened surfaces of the oceans and the lower ~30 m of the atmosphere (Wilson 1975; Alerstam et al. 1993; Pennycuick 2002). Because petrels have evolved flight characteristics that allow them to extract energy from prevailing winds over the ocean’s surface—variability in wind speed and direction encountered by birds is of fundamental importance (Ainley 1977; Furness and Bryant 1996; Ballance et al. 1997; Spear and Ainley 1998; Weimerskirch et al. 2000; Shaffer et al. 2001a, b; Phillips et al. 2004; Spear et al. 2007). Wind is particularly important for petrels that have large ranges at sea (millions of square kilometers), foraging trips that can exceed thousands of kilometers during the breeding season (Jouventin and Weimerskirch 1990; Phalan et al. 2007; MacLeod et al. 2008), and long-distance trans-oceanic migrations (Spear and Ainley 1999; Croxall et al. 2005; Shaffer et al. 2006). Such impressive ranging behaviors are determined in part by wing morphology (Warham 1977; Pennycuick 1982; Spear and Ainley 1997a; Weimerskirch et al. 2000) and flight behavior according to prevailing wind conditions (Pennycuick 2002; Sachs 2005).
In recent decades, meteorologists and oceanographers have developed remote sensing technologies (e.g., scatterometry, radar altimetry) and models to depict global atmospheric and oceanographic circulation patterns (Kalnay et al. 1996; Kistler et al. 2001). At the same time, ecologists increasingly have used tracking devices attached to seabirds to measure movements over large regions of the world’s oceans (e.g., satellite transmitters, global positioning systems, and archival geolocation sensors; Burger and Shaffer 2008; Phillips et al. 2008). Given advances in device miniaturization, it is now possible to track even the fine-scale movements of small-bodied (<1,000 g) seabirds throughout the world’s oceans. Obtaining linked environmental information at sea, however, remains challenging (Wilson et al. 2002).
Linking flight behaviors and movements at sea to wind speed and direction is required for understanding aspects of foraging ecology especially among far-ranging seabirds. Both the direction and velocity of wind must be considered to determine the amount of power (and thus energy) required in the various modes of flight. Previous studies have focused on direct observations and mathematical modeling of flight behaviors. For example, Pennycuick (1982, 1997, 2002) quantified flight behaviors among a variety of petrels and calculated ‘minimum power’ speeds and ‘maximum range’ speeds. And in the southern Atlantic, Alerstam et al. (1993) observed albatrosses adjusted their mean flight directions with respect to wind direction according to wind speed. Presently, however, we are not aware of any previous avian study that has applied a technique that can generate a singular mathematical test statistic (e.g., correlation coefficient) that relates seabird flight directions and speeds to ocean winds. Such application could potentially allow more refined statistical hypotheses testing or improve modeling approaches (Felicísimo et al. 2008; González-Solís et al. 2009). A better understanding of how seabirds travel in association with winds also is important for the interpretation of distributional patterns at sea recorded from ships (Abrams 1985; Spruzen and Woehler 2002) and may prove useful, following sufficient ground-truthing, for correction of at-sea densities based on movements relative to winds and ship direction (Spear et al. 1992).
Herein, we apply vector correlation (Crosby et al. 1993) to quantify the association of seabird movements at sea with ocean winds. First, we provide an example of vector correlation applied to satellite-bourn, scatterometer-derived winds (SeaWinds on QuikSCAT, hereafter QSCAT), and continuous winds recorded by six meteorological buoys in the North Pacific Ocean. In this example, we expect vector correlations to be relatively high because the two platforms are independent measures of the same phenomenon, ocean winds. This example primarily is for proof of concept and secondarily to address how our purpose-built programs attempt to confront certain aspects of anticipated variability in QSCAT data such as near-coastal measurements of winds and potential effects of small-scale weather phenomena. Second, based on previous observations (Alerstam et al. 1993; Weimerskirch et al. 1993; Spear and Ainley 1997a, b), we hypothesize that movements (i.e., path segments) that comprise individual track lines among five satellite-tracked petrels are correlated with wind speed and direction measured by QSCAT over the open ocean. Furthermore, the petrels in this study should fly predominantly with quartering tail-winds (i.e., 105–165° with respect to birds’ trajectories).1
Meteorologists and oceanographers seeking to quantify wind and current patterns developed vector correlation to measure the degree of association between two independent sets of vectors. Detailed reviews and mathematical proofs for vector correlation are presented in Breckling (1989) who reviewed angle and direction correlation; Hanson et al. (1992) provided a meteorological application; and Crosby et al. (1993) present the necessary background describing the basic mathematical properties of vectors, properties of the product-moment correlation coefficient, and a brief history of vector correlation. Crosby et al. (1993) suggested a “universally acceptable” definition of vector correlation for geophysics with an example evaluating coherence among marine surface winds. Recently, vector correlation has been used to cross-validate winds measured at oceanographic buoys with winds derived using satellite scatterometry (Freilich and Dunbar 1999), patterns and dynamics of coastal breezes off the west coast of India (Aparna et al. 2005), and the relationship between synoptic winds and currents in the Santa Barbara Channel, CA (Breaker et al. 2003). We are aware of only three biological studies that have used vector correlation: Acosta et al. (1997) used complex vector correlation (Kundu 1976) to assess post-larval recruitment of spiny lobster (Panulirus argus) as related to wind forcing; Queiroga (2003) examined the same dynamic for crab megalopae (Carcinus maenas); and Brooks (2005) evaluated tidally influenced movements of green sea turtles (Chelonia midas) in a channelized estuary following Crosby et al. (1993).
For our purposes, consider the two-dimensional vector W in earth-oriented Cartesian coordinates such that component u increases to the right (eastward) along the x-axis and v increases upward (northward) along the y-axis. Analytical methods required for vectors need to account for both magnitude and direction. The generalized correlation coefficient (ρv2) of Crosby et al. (1993) is based on mathematics described in Hooper (1959) and the complex correlation coefficient (ρT) of Kundu (1976). Both coefficients (ρv2 and ρT) provide simple, scalar statistical measures of the degree of agreement between two, two-dimensional vector series. Because covariances of both scalar components (u and v) of each vector are considered, the single generalized correlation coefficient accounts for the contribution of both components toward the overall correlation. The coefficient ρv2 ranges from 0 to 2, with 2 indicating perfect agreement between the two vector sets. The technique is invariant to coordinate system transformations and scaling applied to either vector data set by a constant magnitude or angular shift. Empirical tests by Crosby et al. (1993) showed that for sample sizes >64, the distribution of nrv2, where n = the sample size used to calculate rv2 (the corresponding sample parameter for the population parameter ρv2) approximates the theoretical chi-square distribution with four degrees of freedom. This property and graphical presentation of bootstrap estimates for smaller sample sizes (as few as n = 8) allow for vector correlation significance tests at various sample sizes (Crosby et al. 1993; Breaker et al. 1994).
Wind speed and direction: oceanographic buoys and QuikSCAT scatterometry
Summary of wind data at six National Data Buoy Center (NDBC) locations throughout the northeast Pacific and co-located synoptic winds derived from the SeaWinds scatterometer onboard the QuikSCAT spacecraft
Days available (%)
Total # co-locations
# rain flagged (%)
# wind flagged (%)
# co-locations retained (%)
Point Conception, CA
Southern Aleutians, AK
Monterey Bay, CA
Satellite telemetry of five petrel species
We arbitrarily selected one individual each from five different species’ tracking data sets for analyses: Sooty Shearwater (SOSH, Puffinus griseus), Pink-footed Shearwater (PFSH, Puffinus creatopus), Hawaiian Petrel (HAPE, Pterodroma sandwichensis), Grey-faced Petrel (GFPE, Pterodroma macroptera gouldi), and Black-footed Albatross (BLAL, Phoebastria nigripes). We outfitted individuals with satellite-linked platform terminal transmitters (PTTs) programmed with 60-s repetition rates and duty cycles that ranged from continuous (i.e., no off period; HAPE, SOSH, BFAL), 2 h on:2 h off (GFPE), and 10 h on:48 h off (PFSH). We attached PTTs along the dorsal midline between the scapulae following MacLeod et al. (2008) and Hyrenbach et al. (2002; BFAL). Bird locations were generated by CLS-ARGOS (CLS 2007), and we retained data for location classes (LC) 3, 2, 1, 0, A, and B and excluded all unclassified locations (i.e., LC Z). We used satellite tracking and analyses tools (STAT) online software (Coyne and Godley 2005, www.seaturtle.org/STAT, accessed October 14, 2009) to manage and pre-filter ARGOS data. First, STAT flagged potential “swapped locations” (for each location generated, ARGOS reports two sets of latitude and longitude coordinates [“mirror locations”], and occasionally set two contains the more accurate location). We examined mirror locations and these were manually swapped if needed. Second, when ARGOS returned two sets of location data for the same time, we selected the location with the more accurate LC, or greater number of uplinks received by satellites. We exported pre-filtered data from STAT for final filtering using the publically available R package argosfilter version 0.5 (R Development Core Team 2007; Freitas et al. 2008, http://cran.r-project.org/web/packages/argosfilter/index.html, accessed October 14, 2009). We used SDA filter (Freitas et al. 2008) with speed, distance, and angle (SDA) criteria to remove imprecise ARGOS location data. We modified the scripting to allow a maximum threshold speed for filtering of 70 km h−1, and applied default minimum and maximum angle and distance values set at 15° and 2,500 m, and 25° and 5,000 m, respectively, for all individuals. We then interpolated filtered data using Matlab to hourly locations using the linear technique of Tremblay et al. (2006). We avoided generating interpolated locations between ARGOS locations separated by >8-h. We used the interpolated, hourly bird location data to generate a bird-movement vector set (W2; i.e., bearing and speed, oceanographic convention) and decomposed these to u and v scalar components of bird movements.
Vector correlations of (1) winds measured at buoys and (2) seabird movements with winds measured by QuikSCAT scatterometry
First, we used vector correlation following Crosby et al. (1993) to evaluate the degree of association between buoy and co-located QSCAT winds. Similar to other studies having analogous objectives (i.e., validation and assessment of QSCAT winds), this exercise allowed us to evaluate our purpose-built programs within a context where we expected high and statistically significant vector correlations. We determined statistical significance among vector correlation coefficients rv2 by randomly sampling (with replacement) paired vector sets from buoy data and shuffled QSCAT data. For each of 10,000 permutations, we calculated the vector correlation coefficient rv2 and determined probability values for rv2 by summing the number of permuted rv2 values ≥ the empirical value, divided by 10,000. Statistical significance for rv2 was determined by setting alpha = 0.05 (Breaker et al. 1994).
Secondly, we used vector correlation to evaluate the degree of association between seabird flight trajectories (speed and direction) and co-located QSCAT winds. We matched bird-movement data to daily QSCAT wind data (W1) according to the bird’s synoptic position with QSCAT satellite over-passes (using QSCAT swath-coverage-times as described earlier). Nicholls and Robertson (2007) contributed lengthy discussion regarding the estimation of speeds using ARGOS data. In our analyses, bird speed refers to ground speed, which in our case, is the point-to-point speeds calculated along a bird’s interpolated track line. Although, point-to-point speeds estimated from ARGOS data have the tendency to underestimate the true speeds of petrels that travel along zig-zag flight paths (Pennycuick 1982; Alerstam et al. 1993), our ARGOS filtering and relatively short off-periods in duty-cycling would remove unrealistic maximum flight speeds such as those generated from locations estimated over short time series (i.e., <2 h; Nicholls and Robertson 2007). For our analyses, point-to-point values (and interpolated values between points) provided averaged speeds and directions for correlation with winds. We correlated entire track lines and 8-days serial segments from each bird with QSCAT winds. For the serial segment analyses, the 8-days window provides a minimal time window with sufficient co-location sample size (~10–16 co-locations twice daily) to calculate meaningful vector correlation coefficients along an individual’s track line (Breaker et al. 1994). Temporal and spatial autocorrelation effects are expected to be minimal because co-locations are separated by a minimum of ~12 h. Furthermore, because the frequency of co-locations is set by the satellite orbit, and this remains consistent for any tracked seabird, autocorrelation effects are incorporated into the calculation of the single vector correlation coefficient. We determined statistical significance for entire track line correlations among individuals using the resampling technique described earlier for buoy–QSCAT correlation. Because QSCAT winds potentially are autocorrelated over short time-scales (several hours to several days), bootstrapped probability estimates should be examined with caution when near alpha = 0.05 or at when co-location sample size is appreciably large (i.e., bootstrapped P values would tend to underestimate the true values). Lastly, we calculated isotropic, vector-averages and directional standard deviations for incidental wind angles (α ± SD) experienced by birds following Yamartino (1984). All graphical representations of winds and bird trajectories are depicted using the oceanographic convention.
Vector correlation of buoy and QSCAT winds
RSS rain flagging applied to QSCAT data removed 3–17% of the co-located data (greatest flagging at 46,042, Monterey Bay; Table 1). Our wind flagging applied to QSCAT data removed an additional 6–13% of the co-located data (greatest flagging at 46,063, Point Conception; Table 1). After rain and wind flagging, there remained from 75% of co-located data at station 46,042 (Monterey Bay) to 92% of co-located data at station 51004 (southeast Hawaii; Table 1).
Vector correlation (Crosby et al. 1993) for winds measured at Pacific NDBC buoys and co-located winds derived from the SeaWinds scatterometer onboard the QuikSCAT spacecraft
All wind speeds
Binned wind speeds
Point Conception, CA
Southern Aleutians, AK
Monterey bay, CA
Vector correlation of seabird trajectories and QSCAT winds
Summary satellite tracking data for five procellariiform seabirds
Duty cycle, hours (on:off)
Locations retained (%)
Summary of QSCAT wind data and procellariid co-locations
Co-locations retained (%)
rv2 max (n)b
118 ± 31
100 ± 34
120 ± 24
129 ± 23
115 ± 30
Pink-footed Shearwater #64378 (post-breeding adult) departed its colony in the Juan Fernandez Islands, Chile on April 26, 2006. It traveled northward along the Chile–Peru coastline en-route to a wintering site along the southern coast of Baja California Sur, Mexico (Fig. 2b). Overall, this individual’s movements were significantly correlated with QSCAT winds (rv2 = 0.49, P = 0.0054; Fig. 2b; Table 4). PTT duty-cycling (48-h off-periods) precluded sufficient co-located data (i.e., >10 co-locations per 8-days) to calculate rv2 for path segments. This shearwater traveled with a variety of quartering tail-winds, crossing winds, and occasional head-winds (100° ± 34°; Fig. 3b).
Grey-faced Petrel #67596 (colony-attending adult, breeding status unknown) departed its colony in the Alderman Islands, New Zealand on July 17, 2006, to travel throughout portions of the southwest Pacific, Tasman Sea, and East Australian coastline (Fig. 2d). Overall, this individual’s movements were significantly correlated with QSCAT winds (rv2 = 0.55, P < 0.0001; Fig. 2d; Table 4). Maximum 8-days rv2 = 1.36 occurred August 20–27, while it transited counterclockwise around an anticyclone (Table 4, Fig. 4b). This petrel traveled predominantly with quartering tail-winds, occasionally with crossing winds, and occasionally with tail-winds (129° ± 23°; Fig. 3d).
Buoy–QSCAT wind correlations
Vector correlations (range in rv2 = 1.28–1.90) between co-located winds measured at buoys and by SeaWinds on QuikSCAT in our study are consistent with Freilich and Dunbar (1999) who report vector correlations rv2 of <1.4 (coastal buoy locations), 1.6–1.8 (offshore California, Washington, and Hawaii), and >1.8 for offshore buoys located in the NE Pacific. Vector correlations in their study were computed using winds (within 50 km) measured by the similar NASA scatterometer during the brief NSCAT-mission preceding the launch of the SeaWinds scatterometer. Sánchez et al. (2007) observed rv2 ranging from 1.33 to 1.61 among land-based anemometers along the Iberian Peninsula, Portugal and co-located QSCAT data (within 80 km of station).
The greatest relative percentage of rain flagging at Monterey Bay buoy 46042 likely resulted from the dense marine layer (coastal fog) off the central California coastal region and the conservative flagging employed by RSS, rather than actual rainfall. The greatest relative wind flagging and lowest buoy–QSCAT vector correlation occurred at the Point Conception Buoy 46063 located 27 km from shore and in a region with a wind field that is modified by coastal mountain topography. Freilich and Dunbar (1999) also reported lowest vector correlations among coastal buoy stations where small-scale wind variability was expected to be greatest. We suspect that lower relative vector correlation at the southeast Hawaii buoy 51004 results from increased small-scale variability in wind speeds and directions detected by buoys with associated downdrafts resulting from virga in the wake of localized microbursts. Unlike the case in Monterey Bay, the rain flagging method employed by RSS for QSCAT data may not be as sensitive to such small-scale events that would be detected more accurately by buoy-based anemometers.
Petrel–QSCAT wind correlations
Our method for integrating petrel movements with QSCAT winds provides a convenient way to measure how seabirds travel in association with ocean winds. Movements constituting the entire track lines among all five petrels examined were significantly correlated with co-located winds measured by satellite-based scatterometry. Vector correlation (rv2) values calculated over an individual’s entire track line generally were less than the greatest values calculated along serial multi-day sections (i.e., 8-days windows used in this study).
Both the non-breeding GFPE and BFAL displayed lesser rv2 8-days values within spatially concentrated areas separated by path segments characterized by relatively greater rv2 values (Fig. 2d, e). The GFPE traveled from a colony in northeastern New Zealand to concentrate activity within a well-defined hotspot area for this species off the east coast of Australia (Fig. 2d; MacLeod et al. 2008). The BFAL displayed aggregated locations and lesser rv2 values within two regions: off central California and in the central northwestern Pacific; both areas were separated by a path segment with relatively greater rv2 values (Fig. 2e).
High variability among serial vector correlation windows for the individual petrels in this study [ranging from low values (0.05) to relatively high values (1.36)] indicate that correlated wind- and movement-related ‘events’ may exist for seabirds traveling at sea. This pattern in serial trend data was addressed by Crosby et al. (1993) who described ‘wind-event’ scaling as oscillating periods where vector correlation achieves peak values. Many pelagic seabirds that forage in a hierarchical system, where prey resources are patchily distributed at multiple spatial scales (Fauchald 1999), adopt ranging strategies that alternate between spatially concentrated movements (i.e., area-restricted-searching [ARS] and longer-range directed movements between ARS areas). These movement patterns are especially well documented during the breeding season when adults provisioning chicks commute between their colonies and distant foraging areas (Weimerskirch et al. 1997; Hyrenbach et al. 2002; Pinaud and Weimerskirch 2005, 2007; Suryan et al. 2006, 2008). Suryan et al. (2006), however, found interpretation of ARS behavior among non-breeding Short-tailed Albatrosses confusing because albatross movements can be particularly affected by wind speed (see also Jouventin and Weimerskirch 1990). We suggest moving-window vector correlation or comparison between ARS segments and non-ARS segments can provide additional information useful for future interpretations of ARS behaviors and habitat selection among seabirds. For example, to what degree do presumed key foraging areas—based on either reduced flight speeds and increased turning rates (Weimerskirch et al. 1997; Hyrenbach et al. 2002) or ARS behaviors (Pinaud and Weimerskirch 2005, 2007)—also depend on correlated versus non-correlated movements with ocean winds? Further analyses could also examine variable window durations to help resolve temporal scaling in seabird movement–ocean wind correlations.
The five individual petrels in this study all flew predominantly with crossing tail-winds (105–165°). In the eastern tropical Pacific Ocean, Spear and Ainley (1997a,b) observed Procellariiformes as a group primarily flew with crossing winds (70–110°, relative to bird headings) with incidence of headwind flight positively correlated with wing loading and incidence of tailwind flight negatively related to aspect ratio. Within our group, the HAPE and GFPE have high wing aspect ratios and the least wing loading. The shearwaters have intermediate wing loadings and lower aspect ratios. The BFAL deviates from the group in having both greater wing loading and higher aspect ratio. Our results show that the two Pterodroma petrels and the albatross traveled most frequently with quartering tail-winds, and generally avoided headwinds, albeit the albatross occasionally traveled with slightly more crossing to head-winds. With proportionately larger wings, the large gadfly petrels (e.g., HAPE and GFPE) appear to make relatively better use of tail-wind and cross-wind force than larger-bodied albatrosses (Spear and Ainley 1997b). Unlike the petrels and albatross, both shearwaters were engaged in directed movements associated with long-distance migration. The SOSH traveled more so with crossing winds and occasional head-winds and the PFSH traveled more frequently than the others with head-winds and tail-winds. The pattern displayed by the PFSH may have resulted from inclusion in our analyses of an extended wintering period off the coast of Baja California, Mexico where, in order to remain over the narrow continental shelf, the PFSH would either fly into or against the prevailing northwesterly winds.
Our results also agree with previous telemetry-based studies that comment on how seabirds (restricted almost exclusively to Wandering Albatrosses) fly with winds. Using wind direction derived from sea-level pressure models, Jouventin and Weimerskirch (1990), found Wandering Albatrosses (Diomedia exolans) outfitted with satellite transmitters departed their colony with crossing tail-winds (112–157°) and returned generally with crossing head-winds (68–113°). Our observation that the BFAL tended to avoid headwinds may relate to this bird’s non-breeding status when directed movements are not influenced by the need to commute between foraging areas and a colony. Movements during the non-breeding season should be more influenced by wind direction and strength rather than by the need to commute to a specific area or colony local. Both petrels, the SOSH, and BFAL all show directional (and ground speed) changes associated with changing winds encountered (i.e., path segments with relatively great vector correlations, Fig. 2). These results are consistent with Weimerskirch et al. (1993) who observed breeding Wandering Albatrosses changed course when wind directions changed in order to maintain use of crossing tail-winds during long, looping foraging trips. Reinke et al. (1998) used winds derived from modeled sea-level pressures and found similar results as Weimerskirch et al. (1993) for an individual Wandering Albatross; the bird traveled closer in angle to winds of moderate speeds than to high winds (i.e., opposite to the observations of Alerstam et al. 1993). The SOSH in our study (Fig. 2a) traveled closer in angle with relatively greater tail-winds after reaching New Zealand, consistent with observations of albatross by Alerstam et al. (1993). The tail-winds experienced by the SOSH at 40–50°S likely were strong enough to overcome this species’ observed propensity for avoiding relatively weaker tailwinds at subtropical latitudes (Spear and Ainley 1997a).
To our knowledge, Suryan et al. (2006, 2008) is the only author to have evaluated QSCAT winds along albatross track lines. More recently, Felicísimo et al. (2008) and González-Solís et al. (2009) used QSCAT winds to calculate and map model-derived hypothetical “minimum wind cost trajectories” for three migrating shearwaters: Cory’s (Calonectris diomedea), Manx (Puffinus puffinus), and Cape Verde (C. edwardsii). Their models, however, are based on passive particle drift and assumed a minimum flight cost (arbitrary units) when birds traveled directly with tailwinds, despite observations that, in most conditions, similar shearwater species tended to avoid tailwinds (e.g., Spear and Ainley 1997a). The propensity for petrels to avoid tailwinds also has been observed by others. When traveling >100 km day−1, Short-tailed Albatrosses (Phoebastria albatrus) traveled downwind (135–235° incidental angle relative to bird’s trajectory), both Laysan (Phoebastria immutabilis) and Black-footed Albatrosses traveled at 90° relative to the wind, but breeding Waved-albatrosses (Phoebastria irrorata) traveled within 45° of the wind (both head- and tail-winds; Suryan et al. 2008).
Given significant correlations with ocean winds in this study and observations described in previous studies, seasonally and spatially predictable winds associated with sea-level pressure patterns over the oceans (e.g., blocking anticyclones; Higgens and Schubert 1994; Wiedenmann et al. 2002) undoubtedly affect seabird movements over large expanses of open ocean. Winds associated with slower-moving, anticyclonic (high-pressure center) and faster propagating, less stationary cyclonic (low pressure center) circulation likely are important for determining in part the at-sea distribution and long-range trajectories among the Procellariiformes. We suggest vector correlation can provide quantitative insight into how ocean winds—determined by the sea-level atmospheric pressure topography (i.e., cyclones and anticyclones)—significantly affect the movement paths of birds transiting large expanses of the world’s oceans. For example, the HAPE tracked during August 2006, achieved a maximum vector correlation while traveling eastward (i.e., clockwise) to circumnavigate a slow-moving North Pacific 1,030-mb anticyclone (high-pressure system; Fig. 4a). We observed a similar pattern in the related GFPE off New Zealand. Maximum vector correlation for this individual occurred also while circumnavigating (i.e., counter-clockwise) a 1,025-mb anticyclone northeast of New Zealand (Fig. 4b). Nicholls et al. (1997) and Murray et al. (2002) speculated that non-breeding Wandering Albatrosses used southern circum-polar winds associated with weather systems between 30 and 50°S to reach any area within this latitudinal band; however, there are no quantitative environmental analyses presented to support their idea. Jouventin and Weimerskirch (1990) observed Wandering Albatrosses occasionally limited their movements and were “trapped” by slack winds (<5 ms−1) for 1–7 days within near-stationary anticyclones. Murray et al. (2003a, b) observed Wandering Albatrosses circumnavigated cyclones when traversing large regions of the Southern Ocean. Blomqvist and Peterz (1984) correlated the occurrence of several seabirds [including two petrels: Sooty Shearwater and Northern Fulmar (Fulmarus glacialis)] off Sweden with eastward-propagating cyclones and proposed a causal model based on cyclonic wind conditions, wave patterns, and flight behavior to explain the generalized clockwise movements of birds over the Baltic Sea. Consistent with the observations of Murray et al. (2003a, b), but different from our observations of a single BFAL (Fig. 5), they suggested that seabirds on the open ocean should avoid flying into cyclone centers.
SeaWinds data are available from July 1999 to the present. The relatively long-term availability of scatterometer-derived ocean winds allows researchers to revisit archived seabird satellite tracking data (BirdLife International 2004). Unfortunately, SeaWinds on QuikSCAT has surpassed its operational life (estimated at 5 years from launch in 1999) and a replacement sensor has yet to be identified when the current unit fails (Witze 2007). Analyses presented herein also are suitable for wind data such as those generated by the NOAA National Centers for Environmental Prediction (e.g., NCEP_Reanalysis 2; Kanamitsu et al. 2002; see http://www.esrl.noaa.gov/psd/, accessed October 14, 2009) or other similar data sets. These data extend the ability to examine vector correlation to seabird tracking studies that occurred before 1996 when the first scatterometer was deployed (NSCAT). Such model-derived data, however, although available at greater temporal frequency (i.e., 4-h intervals), do not retain the same spatial accuracy (2.5° pixel) as scatterometer-derived winds (0.25° pixel). Methods presented herein provide a standardized approach for quantifying movements associated with ocean-surface winds that can provide additional insight into ranging behaviors. Researchers, however, should be aware of several potential limitations. Even relatively low correlation coefficients can be statistically significant when co-location sample sizes are appreciably high. Interpretations of comparisons should be based on vector correlation coefficients that are calculated from a sufficient number, and preferably, similar sample sizes of co-located data. Because locally observed winds are determined at 0600 and 1800 local time, analyses involving species or individuals that have strong diurnal behavioral patterns related to feeding or traveling may be limited. Additionally, for species that travel very close to shore (<25 km from the coastline) it may be difficult or impossible for QSCAT to resolve wind speed and direction accurately. Lastly, it is important to recognize that petrels also make use of waves (both wind waves and ground swells) for flight at the ocean’s surface (Blomqvist and Peterz 1984; Pennycuick 2002; Suryan et al. 2008); although we focused here on the correlation of petrel movements with surface winds, future analyses may benefit by examining vector correlations with wave parameters such as amplitude, period, and direction (see http://polar.ncep.noaa.gov/waves/products.html, accessed October 15, 2009). Future analyses relating seabird movements with ocean winds can provide insight into the effects of environmental changes on ocean wind fields (Parrish et al. 2000; Vecchi et al. 2006). Lastly, vector correlation techniques herein are not limited to seabirds and wind—they also can be used to examine relationships and test hypotheses among any marine vertebrate that moves through waters where currents are measured and/or can be derived from models (Brooks 2005), remote measurements (i.e., geostrophic flow derived from radar altimetry: Ream et al. 2005; Hawkes et al. 2006; Lambardi et al. 2008), or drifters (Bentivegna et al. 2007).
We consider the incidental angle of the wind as it touches a bird traveling in the forward direction such that wind angles 105–165° are equivalent to winds 195–255°; throughout, we refer to the incidental angles between 0 and 180° a convention that is independent of left- versus right-lateral incidence. See Fig. 4 for a graphical explanation.
We thank L. Breaker and E. McPhee-Shaw for enthusiastic discussions and useful contributions. P. Hodem and K. D. Hyrenbach provided PFSH data, P. Lyver provided GFPE data; BFAL data were supplied as part of a collaborative effort on behalf of Oikonos Ecosystem Knowledge, HAPE data were supplied as part of a collaborative effort on behalf of US Geological Survey (USGS) Western Ecological Research Center, D. G. Ainley (HT Harvey and Associates), H. Friefeld (US Fish and Wildlife Service), C. Bailey and J. Tamayose (Haleakala National Park), and J. Penniman (Hawaii Department of Land and Natural Resources); and SOSH data were supported in part by Moss Landing Marine Labs, USGS, and UC Santa Cruz Tagging of Pacific Pelagics. C. MacLeod kindly provided assistance with programming in R. QuikSCAT. Data are produced by remote sensing systems and sponsored by the NASA Ocean Vector Winds Science Team. QSCAT data are available at www.remss.com. Previous drafts of this paper benefited from comments and suggestions provided by H. Moller, D. G. Ainley, J. Yee, K. Phillips, and four anonymous reviewers. The use of trade, product, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the US Government.