Year-round movements of a small seabird and oceanic isotopic gradient in the tropical Atlantic

: Despite the proliferation of seabird tracking studies, there is a relative paucity of studies on small tropical seabirds. We present for the first time the distribution and movements of the little-known Boyd’s shearwater Puffinus boydi , a Procellariiform endemic to the Cape Verde Islands. We tracked 28 birds from 2 breeding sites (Ilhéu Raso and Ilhéu de Cima) with geolocator loggers from 2007 to 2012. We also analysed stable isotopes of carbon and nitrogen in the 1st primary (P1), the 6th rectrice (R6) and the 1st (S1) and 8th (S8) secondary feathers to reveal moulting pattern and oceanic isotopic gradients. Birds migrated on average 1450 km westward, to the central Atlantic Ocean (5 to 15° N, 30 to 40° W), where they stayed on average 114 d, from May to August. Boyd’s shearwaters exploited oceanic waters year-round and showed δ 13 C values similar to other oceanic seabird species and δ 15 N values indicating the lowest known trophic level among all central Atlantic seabirds. Isotope values in flight feathers suggest most animals moult their P1 and R6 around the breeding ground, whereas all birds moult S1 and S8 at the non-breeding quarters. Correlations of δ 13 C and δ 15 N values from S8 with the longitude of the non-breeding area indicate the existence of large-scale isotopic gradients matching those known at baseline levels. Combining geolocator tracking and stable isotope analyses in feathers not only allowed us to describe in detail the annual life cycle and distribution of the species, but also the oceanic isotopic gradients in the tropical Atlantic.


INTRODUCTION
Over recent decades, studies on the biology and ecology of tropical seabirds have been mainly fo cused on diet, foraging and performance at the breeding colonies (Ashmole & Ashmole 1967, Ballance & Pitman 1999, Spear et al. 2007). More re cent ly, studies have been extended to include the relationship between breeding performance and en viron mental features (Surman et al. 2012, Catry et al. 2013). In contrast, at-sea distribution of many tropical seabirds remains poorly known and the sparse information available is mostly based on shipboard and coastal observations (Jaquemet et al. 2004, Ballance 2007. Despite the standardized approaches used in ship surveys (Tasker et al. 1984, Camphuysen & Garthe 2004, unreliable at-sea identification of some species (Ainley et al. 2012) and usually unknown origin and breeding status of observed individuals make these counts difficult to interpret.
In the last 2 decades, the rise in the use of extrinsic and intrinsic markers has underpinned an exponential increase in studies on the pelagic ecology of seabirds. Regarding extrinsic markers, the light level logger (geolocator) has become an essential device for studying year-round movements in much more detail than ever before, improving our understanding on the ecological needs and constraints of seabirds at sea (e.g. González-Solís et al. 2007, Guilford et al. 2012. However, the increasing use of geolocators to study seabird distribution and behaviour has been clearly biased towards species from temperate and subantarctic waters. Thus, there is still a clear lack of knowledge about the year-round at-sea ecology and distribution of tropical seabirds, with only a few species well studied (Catry et al. 2009, Pinet et al. 2011, Dias et al. 2015, Precheur 2015. Similarly, intrinsic markers, such as stable isotope analysis (SIA) of δ 13 C and δ 15 N of various tissues have been widely used to study seabird trophic ecology. Typically, δ 13 C values have been used to determine the diet of seabirds whereas δ 15 N values reflect trophic level in a general manner (Hobson et al. 1994, Cherel et al. 2008. However, isotopic values of δ 13 C and δ 15 N at baseline are also known to vary geographically in the marine environment (McMahon et al. 2013a,b). Spatial maps of isotopic landscapes, so called 'isoscapes', reflecting this variability, are just now beginning to emerge, mostly based on largescale studies on plankton (Somes et al. 2010, McMahon et al. 2013a. Whether this spatial isotopic variability propagates up to the food chain and can provide insights into the foraging movements or wintering areas of predators is still a matter of study (Quillfeldt et al. 2005, Cherel & Hobson 2007, Navarro et al. 2013. In this regard, combining SIA with tracking studies can help validate the relationship be tween isotope values and foraging movements (Jaeger et al. 2010). Despite increasing interest in linking isotope values of feathers to seabird movements, especially during the less known non-breeding season, only few studies showed a correspondence between δ 13 C and δ 15 N in feather isotope values and non-breeding distribution of seabirds tracked with geolocators , Hedd et al. 2012.
The lack of basic knowledge regarding year-round distribution, phenology and trophic ecology becomes a matter of conservation concern in polytypic species difficult to identify at sea and with unclear taxonomic status. The Little−Audubon's shearwater complex (Puffinus assimilis−lherminieri, Procellariiformes), small-sized seabirds spread within tropical and temperate waters, is a particularly poorly known seabird complex as shown by the various taxonomic revisions that occurred over recent decades (reviewed in Austin et al. 2004). After many years of controversy, Audubon's shearwater is now suggested to include 3 subspecies, the Audubon's shearwater P. l. lherminieri, the Barolo shearwater P. l. baroli and the Boyd's shearwater P. l. boydi, with a conservation status of 'Least Concern' (BirdLife International 2015, Carboneras et al. 2016a), although in the present study we preferred to follow a precautionary principle and maintain the specific status of the taxon P. boydi (Hazevoet 1995, Robb & Mullarney 2008. Indeed, the conservation status and taxonomy of several closely related seabird taxa still remain controversial partly due to our lack of knowledge on their spatial ecology, since this is important for understanding migratory connectivity, reproductive isolation mechanisms, and therefore potential for lineage divergence . Therefore, studies on the phenology and year-round distribution of species within seabird complexes with controversial taxonomic relationships are particularly timely. Recent geolocation and stable isotope studies on Barolo shearwater breeding on the Macaronesian archipelagos of the Azores and the Salvagens (Neves et al. 2012 showed this subspecies to disperse in the surroundings of the breeding colonies outside the breeding period. However, there is very little knowledge about detailed biology of the close ly related Boyd's shearwater P. boydi Mathews, 1912, endemic to the Cape Verde Islands, especially those aspects related to phenology, year-round distribution and trophic ecology. Roscales et al. (2011) revealed the distribution and trophic position of Boyd's shearwaters only at the end of the breeding season, when animals foraged close to the colony. Away from breeding grounds, Boyd's shearwater has only been seen in small numbers off the Senegal coast in October (Hazevoet 1997, Dubois et al. 2009), in 1976 1 bird was trapped on St. Helena (Bourne & Love ridge 1978) and a suspected observation of 1 individual was reported from the Canary Islands in December 2012 (Velasco 2013). However, the majority of the observations of individuals of this species have been reported all year round in Cape Verde and surrounding waters (Bourne 1955, Hazevoet 1995, Dubois et al. 2009), which suggests a non-migratory behaviour, even though non-breeding grounds remain unknown.
To fill in this gap, we provide the first detailed study on the year-round movements and distribution of the Boyd's shearwater, based on geolocation and SIA of feathers over multiple years. We aim to (1) reveal main foraging areas during breeding and non-breeding seasons, the detailed phenology of their life cycle and, in particular, clarify whether Boyd's shearwater performs dispersal movements or oriented migration to a specific non-breeding area; and (2) to bring new insights into the existence of isoscapes and their potential use to study the movement of tropical top predators by linking the isotopic values of feathers with individual non-breeding areas.

Study site and species
We conducted fieldwork during the breeding seasons of 2007 to 2012 in the Cape Verde Islands, on Ilhéu Raso (16°36' N, 24°35' W) and Ilhéu de Cima (14°58' N, 24°38' W), 2 islets 180 km apart. We visited the colonies during the incubation period, from February to early April, depending on the year. Additionally, we visited Raso in November 2009. The Boyd's shearwater is a taxon within the Little− Audubon's shearwater complex (Puffinus assimilis− lherminieri, Procellariiformes) (reviewed in Austin et al. 2004). Traditionally, 'assimilis' and 'lherminieri' were recognized as 2 species groups, but with numerous taxa within each group (Cramp & Simmons 1977, Warham 1990, Carboneras 1992, Brooke 2004). In the last decade, a molecular study by Austin et al. (2004) proposed 3 geographically discrete clades of the complex identified in the North Atlan tic, Southern (Australasia) and tropical Pacific and Indian oceans. A recent revision (Carboneras et al. 2016a,b) has suggested the separation of little shearwater Puffinus assimilis, distributed in the southern hemisphere, from the Audubon's shearwater Puffinus lherminieri, distributed in the North At lan tic Ocean and Caribbean Sea. Particularly, 2 North-Atlantic taxa, Barolo shearwater Puffinus baroli (breeding in the Azores, Madeira, and the Canary Islands) and Boyd's shearwater Puffinus boydi (breeding in the Cape Verde Islands) have been switched between 'assimilis' and 'lherminieri' groups by various authors over the years. Hazevoet (1995) considered P. boydi as an independent species. The Boyd's shearwater is endemic to the Cape Verde Islands, where it is thought to breed on most islands and islets (not known from Maio and extinct on Sal) (Hazevoet 1995), with a population estimation of ca. 5000 pairs (BirdLife International 2015). Birds (body mass ≈ 160 g) nest in burrows in soft soil or in rocky cavities. Both parents share incubation of a single white egg that may take 44−60 d to hatch (Carboneras et al. 2016a) and breeding lasts from January to June. Birds are thought to disperse after breeding; however, some were reported visiting the breeding colony at the end of August (Bourne 1955) and throughout the year (Hazevoet 1995). Diet is not well known; the few stomachs examined by Bourne (1955) contained fish and cephalopods up to 8 cm. A closely related species (P. baroli) from the Azores feeds mostly on cephalopods and fish (Neves et al. 2012, J. A. Ramos et al. 2015.

Bird tracking and spatial data analysis
During the study period (2007−2012) we deployed a total of 90 geolocators on 68 individuals of Boyd's shearwaters. We captured breeding birds by hand in the burrow and deployed geolocators, which we retrieved after ≥1 yr. Over the course of the study, we used 3 different types of loggers from the British Antarctic Survey (BAS): Mk9 (n = 32), Mk13 (n = 15) and Mk18-H (n = 43). Each logger was attached with a cable tie to a plastic ring, which was deployed on the tarsus of the bird; weight of equipment was approximately 2 g (1.25% of body mass). We de ployed only 1 geolocator per breeding pair.
Geolocators recorded ambient light intensity, time and immersion in seawater. Light levels were measured every 60 s and, depending on the type of device, the maximum value within each 5 min (Mk18-H logger) or 10 min (Mk9 and Mk13 loggers) interval was recorded. We processed raw light data and visually supervised each transition using TransEdit from BASTrack software (British Antarctic Survey). The sunrise and sunset times were estimated applying the light threshold value of 20. To estimate sun elevation angle, we calibrated the loggers before deployment and after recovery on an open site without shading. The value of sun elevation angle was calculated and applied for each logger, ranging from −5.82 to −3.49 (mean −4.54). Light level data were converted into latitude derived from day length and longitude derived from the time of local midday with respect to Greenwich Mean Time, using BASTrack software. This process results in estimation of 2 positions of the animal per day (Delong et al. 1992, Hill 1994, Afanas yev 2004, with a mean error ± SD of 186 ± 114 km (Phillips et al. 2004). Furthermore, as the latitude estimates are highly sensitive to errors and changes in day length, positions in equatorial regions may present lower accuracy (Hill & Braun 2001). In addition, cloudy weather at sunrise and/or sunset may lead to error estimated to 340 km in latitude and 105 km in longitude (Nisbet et al. 2011).
It is worth mentioning that interpretation of geolocation positions especially in equatorial latitudes should be accepted with caution, especially the latitude estimations around equinoxes (Hill & Braun 2001, Ekstrom 2004, Lisovski et al. 2012. Detailed exa mination of error in latitude estimation is necessary to avoid the possible misleading interpretation of geolocation positions. Particularly in this study, previous visual examination of positions showed a clear pattern (Figs. S1 & S2 in Supplement 1 at www.int-res. com/ articles/ suppl/ m579 p169_ supp/) re -peated in all individuals during the breeding period and resulting from a shift in the latitudinal error between inter equinoctial intervals: positions before spring equinox -reflecting mostly incubation and the early chick-rearing period -were distributed northerly from the colony, whereas the positions after the spring equinox -reflecting the chick-rearing period -were distributed southerly from the colony. To avoid possible misleading interpretation that during incubation animals forage in the north and during chick rearing in the south of the colony, we pooled together prelaying, incubation and chickrearing as a breeding period.
Obtained positions were filtered for each logger separately applying a 3-level filtering method, removing positions (1) 15−30 d before and after equinoxes, (2) with obvious interference at dawn or dusk, and (3) when flight speeds sustained over a 48 h period were higher than 30 km h −1 applying iterative backward/forward speed filtering (Mc -Connell et al. 1992). The speed threshold was defined after visual examination of distributions of flight speeds. We also excluded positions from the day of deployment and recovery of the logger. Overall, 66% of original locations were retained for further analysis.
Kernel density utilization distribution (UD) estimates were generated from filtered locations (projection: Lambert Equal-Area Azimuthal, centred to the centroid of all locations) during different periods of the life cycle separately for each bird and year of tracking using package adehabitatHR (Calenge 2006) in R (R Core Team 2016). Kernel contours of 50% ('core-area') were calculated using a smoothing parameter (h) equivalent to the mean error of the geolocators (Phillips et al. 2004). We examined various spatial parameters for each track: (1) the area exploited during the breeding and non-breeding periods (50% UD; in km 2 ); (2) location of the centroids of breeding and non-breeding areas (50% UD), which were calculated using 'centroid' function from package geosphere (Hijmans et al. 2012); (3) the total distance (great-circle) from the breeding colony to the centroid of the non-breeding area and (5) the accumulated distance covered within the nonbreeding area (without migration), which were estimated using the functions 'distance' and 'distance-Track' from the argosfilter package (Freitas 2012), respectively.
Geolocators also recorded salt-water immersion data sampled every 3 s and registered summary value every 10 min (varying from 0, when the logger was dry the entire 10 min period, to 200, when the logger was permanently wet). This information was used to help define some phenological parameters (see next subsection).

Phenology
Dates defining the phenology of species were identified visually from geographical positions, light and immersion data. During equinox periods, when latitude estimation is not accurate (Hill & Braun 2001), we used only changes in longitude and in immersion data to detect changes in movements and estimate dates of arrival to and departure from the breeding colony.
We estimated various phenological parameters: last night spent at the colony (continuous dry record over prolonged period of time during darkness), departure from the breeding and non-breeding area (the first day that the bird's location was outside the cluster of previous day's positions and was followed by directed movement away from this area), duration of the non-breeding period and migratory movements, arrival to the breeding and non-breeding area (the first day the bird entered the cluster of positions after a directed movement towards that area), the first day and night an individual spent in the burrow (detected by a continuous dry record over a prolonged period of time during daylight and darkness), first day of incubation (min. 2 consecutive days spent in the burrow), duration of the incubation period (from the first day of incubation until the return from the last foraging trip, including time spent outside on foraging trips between incubation shifts), and, finally, incubation shift and foraging trip duration.
Parameters referring to incubation duration were estimated only for individuals with 2 or more continuous years of tracking data (with the same geolocator or the geolocator that was replaced during incubation and recovered the following year). For those individuals we could estimate the onset, duration and end of incubation from light and immersion data. As some loggers failed to collect data for the entire deployment period or some animals did not breed, sample sizes for different phenological parameters vary somewhat between analyses. Based on these parameters, we identified and considered 4 periods of the life cycle: (1) breeding, period between logger deployment and departure on migration and period between the arrival to the colony from migration and recovery of the logger, (2) postnuptial migration, (3) non-breeding, period between arrival to non-breeding area and start of prenuptial migration and (4) prenuptial migration. One individual did not migrate and spent the non-breeding season in the vicinity of the Cape Verde Islands, so we considered the last night the animal spent at the colony (burrow) as the end of the breeding period. Similarly, the start of the subsequent breeding season was assigned as the first night the animal visited the burrow.
We used repeated measures ANOVA with individual as an error term (to account for pseudo-replication as few individuals were tracked >1 yr) to test for differences between the duration of the post-and prenuptial migration and the size of the core range areas between the breeding and non-breeding periods.

Stable isotope analysis
Boyd's shearwater is expected to moult the first primary feather at the end of the breeding period, just before migration, reflecting the isotopic composition of the breeding area (Cramp & Simmons 1977, Roscales et al. 2011. Known primary moult patterns of similar shearwater species are described as descen dent, i.e. from the innermost to the outermost primary feather, with a duration of 3−5 mo, while the outermost rectrice feather is among the last to be moulted (Monteiro et al. 1996, Bridge 2006, Ramos et al. 2009). Moult of secondary feathers of shearwaters has been previously linked with the non-breeding area (Neves et al. 2012. In this study, carbon (δ 13 C) and nitrogen (δ 15 N) isotope ratios were examined in different wing-feather types: 1st primary (the innermost), 1st and 8th secondary and 6th rectrices (the outermost) feathers (hereafter named as P1, S1, S8 and R6, respectively). All feathers were sampled when we recovered the geolocator and they were stored in plastic bags before the analysis. For birds with the same logger recovered after ≥ 2 yr, feather sampling also occurred at the point of logger recovery, but these feathers are only related with the last year of tracking. In total, the dataset for statistical analysis consisted of 32 sets of 4 feathers from 28 individuals (4 individuals with feathers from 2 different years).
To avoid any possible contamination, feather samples were washed in 0.25 M sodium hydroxide solution, rinsed with distilled water and oven dried at 40°C for 24 h. Subsequently, we manually cut each feather to small fragments using stainless steel scissors and weighed a sample of 0.30−0.32 mg on a precision scale. Stable isotope values are expressed in delta notation (δ) as parts per thousand (‰) according to the following: δX = [(R sample /R standard ) − 1], where X is 15 N or 13 C and R is the corresponding ratio 15 N/ 14 N or 13 C/ 12 C, respectively. R standard values for 15 N and 13 C were based on atmospheric N 2 and the Vienna Pee Dee Belemnite standard, respectively. Replicate measurements of laboratory standards (2 standards for every 12 unknowns) indicated measurement errors of approximately 0.2 and 0.1 ‰ for nitrogen and carbon, respectively. The analysis of stable isotopes was carried out at Scientific-Technical Services of the University of Barcelona.

Statistical analyses of isotopic data
We could not directly test for differences between the 2 colonies as they were sampled in different years (Ilhéu Raso: 2007 andIlhéu de Cima: 2009. After visual comparison, there was no indication in systematic differences in isotopic values between colonies; therefore, all the statistical analyses were performed with isotopic data of both colonies pooled together. To test for differences among feathers in isotopic data, we first checked δ 13 C and δ 15 N values for normal distribution using Q-Q plots and Shapiro-Wilks' test. We used linear mixed models (LMM, R package lmerTest; Kuznet sova et al. 2015) to compare isotopic values among feathers (fixed factor) and we accounted for pseudo-replication including individual and sampling year as random factors. The p-values were calculated from Type 3 F-statistics with Satterthwaite's approximation for degrees of freedom, while pairwise comparisons were calculated based on differences of least squares means (function 'difflsmeans' package lmerTest) and adjusted using Bonferroni correction.
Based on the isotopic differences found among feathers (see 'Results'), we inferred that the S1 and S8 were moulted during the non-breeding period. Because both showed similar isotopic values, and to allow a comparison with the isotopic data of previous studies on a closely related species (Neves et al. 2012, we used S8 for subsequent analyses. To link isotopic values of S8 feathers with the non-breeding area of each individual we determined the individual centroids of 50% kernel of non-breeding area. We used LMM to examine whether the variation in S8 feather isotopic values could be explained by the location of their non-breeding area (latitude and/or longitude of centroid as fixed, individual and year as random factors; R package lmerTest; Kuz netsova et al. 2015). The best-supported model was selected using the Akaike Information Criteria cor-rected for small sample sizes (AIC c ) (R package Mu-MIn; Bartoń 2016). To understand the possible influence of the sampling year we verified its importance based on the likelihood ratio test (function 'rand' from package lmerTest) and by calculating the variance explained by the sampling year.
Statistical analyses were carried out using the R software version 3.2.1 (R Core Team 2016). All values are presented as means ± SD, and we assumed a significance level p < 0.05.

Recovery of loggers
We retrieved 43 loggers (recovery rate 47.8%) from 32 unique individuals. Most loggers were recovered in the year following deployment; however, 7 loggers were retrieved after 2 consecutive years. Eight other individuals (n = 6 and n = 2, respectively) were tracked over 2 and 3 consecutive years, by recovering and deploying a new logger each year. Eleven loggers failed or did not contain enough data for further analysis. Overall, the final dataset contained 38 year-long tracks of 28 unique individuals (9 from Ilhéu Raso, 19 from Ilhéu de Cima), including 10 individuals with 2 yr of tracking. We calculated kernel UD density for 38 tracks for non-breeding (2007, 9 tracks; 2008, 3 tracks; 2009, 6 tracks; 2010, 12 tracks; 2011, 8 tracks) and 35 tracks for breeding season, as 3 tracks from 2010/2011 did not contain enough locations for kernel estimation.

Phenology of annual cycle
Boyd's shearwaters presented some variability in their phenological parameters, especially in the timing of the first day and the first night in the burrow (Table 1) and on the duration of the non-breeding period (Table 1, Fig. 1 with individual phenologies). Furthermore, the duration of the prenuptial migration was statisticaly longer than the postnuptial migration (repeated-measures ANOVA, F 1, 45 = 7.463, p = 0.009), with birds travelling for 7.2 ± 6.0 d to reach the breeding colony on their prenuptial migration in contrast with 4.9 ± 2.6 d to reach the non-breeding area on their postnuptial migration (Table 1).

Seasonal changes in at-sea distribution
During breeding, birds dispersed in different directions around the breeding colony and in proximity to Cape Verde Islands. With 1 exception (bird ID 2007_ 047), which foraged in the neritic area of the African coast in November-December, the tracked birds did not forage in neritic waters but north of the breeding colonies, reaching up to 30°N (Fig. 2, Fig. S3 in Supplement 1 and the animation in Supplement 2 at www. int-res. com/ articles/ suppl/ m579 p169_ supp/). The estimated individual core range area during the breeding season (50% kernel UD) ranged from 292 000 to 764 400 km 2 (470 600 ± 111 500 km 2 , n = 35).
At the beginning of May, birds started their postnuptial migration consistently in a westward direction along a migration corridor between 7° and 15°N (Fig. S4). The mean distance between the breeding colo ny and non-breeding area (to the centroid of 50% kernel UD) was 1450 ± 398 km (range 106− 2391 km, n = 38). The main non-breeding area of Boyd's shearwaters was in the Central Atlantic Ocean, west of Cape Verde Basin, over the Mid-Atlantic Ocean Ridge, from 5 to 15°N and from 30 to 40°W (50% kernel UD; Fig. 3, Fig. S3 21°N, 36°W (bird ID 2007_007). The estimated individual core range area during the non-breeding season (50% kernel UD) ranged from 300 700 to 795 400 km 2 (467 700 ± 120 000 km 2 , n = 38), which did not significantly differ from the size of core range areas during the breeding season (repeated measures ANOVA, F 1, 41 = 0.027, p = 0.870, n = 35). During the non-breeding period, birds dispersed or steadily moved over a huge area. Total distance covered within the non-breeding area was on average 33 670 ± 5628 km (range 17 440−47 690 km, n = 38), moving on average 253.1 ± 32.8 km over approximately 24 h by a mean velocity of 10.5 ± 1.4 km h −1 . From all tracked birds, only 1 individual (bird ID 2007_040) did not migrate and stayed in the vicinity of Cape Verde Island year-round. The timing of prenuptial migration mostly overlapped with the autumn equinox period, but data for a few individuals suggest that animals use a similar route to return to breeding grounds (Fig. S4).

DISCUSSION
This is the first study on the movements and yearround distribution of the Boyd's shearwater. We showed that Boyd's shearwaters perform oriented migratory movements and exploit oceanic habitats year-round. Furthermore, we revealed the existence of a longitudinal isotopic gradient in the tropical north Atlantic by relating the isotopic values of the feathers moulted during the non-breeding period and the location of the individual non-breeding area.
Boyd's shearwaters showed some variability in various aspects of their breeding phenology. Small species breeding in the tropics may experience relatively constant environmental conditions, which may cause  (Brooke 1990). The few individuals that started the postnuptial migration relatively earlier, in the beginning of April, were presumably failed breeders; however, we do not have breeding success information of each bird to confirm this hypothesis. The longer duration of pre nuptial migration in relation to the postnuptial one is an opposite pattern to many long-distance migrants (Nilsson et al. 2013) and may be a consequence of prevailing trade winds which advantaged shearwaters during post-nuptial migration through a tailwind but disadvantaged them during prenuptial migration through a headwind (Liechti 2006). Birds started to arrive at the colony in early August, which confirms observations of shearwaters visiting Ilhéu de Cima at the end of August (Bourne 1955). After re turning to the breeding colony, birds were asynchronous in terms of the first day spent in the burrow during daylight; these dates were spread over 4 mo. Those differences might be sex-related, with males visiting burrows earlier than females in some shearwater species (Hedd et al. 2012, Müller et al. 2014, prob ably due to their role in nest defence. However, this asynchrony was also observed in Barolo shearwaters, in a study where only males were tracked (Neves et al. 2012). As sex of animals tracked in this study was unknown and only 1 member of the breeding pair was tracked, we could not estimate the laying date and define the first incubation shift. However, we were able to estimate the  Table 2. Isotopic values of δ 15 N and δ 13 C (‰) in the 1st primary (P1), the 1st (S1) and 8th (S8) secondary feathers and the 6th rectrice (R6) of Boyd's shearwaters breeding in the Cape Verde Islands. P1 and R6 feathers showed similar isotopic values but distinct than S1 and S8 beginning of incubation, on average February 9, which is earlier than Barolo's shearwaters in the Azores (Neves et al. 2012, but see Monteiro et al. 1996 (Booth et al. 2000). Incubation shift length and duration of foraging trips during incubation were similar to Barolo shearwaters tracked in the Azores (Neves et al. 2012). However, the foraging-trip durations of Boyd's shearwaters differed from those of Barolo shearwaters foraging mostly within the Canary Current system (being longer than those of birds breeding in Salvagem Grande but shorter than those in Porto Santo) ). Since we would expect incubation be haviour and foraging strategies to be similar among such closely related taxa, this variability most likely reflects differences in environmental conditions across localities, such as differences in the distance to suitable foraging areas and their typically low predictability in tropical waters, which possibly results in differences in egg neglect episodes (and therefore duration of the incubation) and foraging trip length (and therefore duration of incubation shifts) across populations. During the breeding period, Boyd's shearwaters mainly foraged around the Cape Verde archipelago. Individual core ranges seemed to fluctuate north and south of the archipelago, and some geolocator positions may have even reached the Canary Islands or the Azores (Fig. S3 in Supplement 1, and Supplement 2), but this is most likely due to the effect of the equinoxes on the latitudinal errors (Figs. S1 & S2). Since longitudinal errors of the geolocator methodology are relatively small and the African coast is just east of the archipelago, our results clearly showed that birds do not visit the African shelf to forage in neritic waters. With the exception of 1 individual for a few weeks, all birds were largely oceanic during breeding and over the 5 yr of the study (Fig. 2). Similarly, the closely related Barolo shearwaters breeding in Madeira and other small seabird species in Cape Verde also show oceanic distribution during the breeding period (J. A. ). The oceanic be haviour of Boyd's shearwaters is also suggested by the low carbon values in    cales et al. 2011, Neves et al. 2012, J. A. Ramos et al. 2015), but also by Bulwer's and Fea's petrel in Cape Verde (Roscales et al. 2011). These results contrast with the importance of the continental shelf inferred for the Barolo shearwaters breeding on Salvagens (J. A.  and also for Audubon's shearwaters breeding in the Caribbean (Precheur 2015, P. Jodice unpubl. data). Further studies using more accurate loggers are needed to confirm these results as this apparent neritic behaviour may just result from the latitudinal error of the geolocation method (Fig. S1). After breeding, Boyd's shearwaters performed a longitudinal-oriented migration, heading westward to the oligotrophic waters of the central North Atlantic Ocean. Despite their short migration, Boyd's shear waters constantly moved during the non-breeding season, covering on average more than 30 000 km.
These movements may be a foraging strategy to increase the chances of finding prey in tropical oceanic waters, which typically show lower productivity and predictability of resources than upwelling systems (Weimerskirch 2007). However, distance calculations should be treated with caution as they may be overestimated due to the positional error. The longi tudinal-oriented migration was noticeably consistent across years at coarse scale, wintering in the same area of the Atlantic (except 1 individual remaining around the Cape Verde Islands). All birds spending their non-breeding period in this area also showed clear oceanic habits. The lack of direct observations of Boyd's shearwaters from their migration and non-breeding grounds may be due to the lack of ob servers in those areas and/or to the problematic identification of the taxa at sea. To our knowledge, there are just a few sightings of individuals of the little shearwater complex of unknown provenance (RNBWS 2014), illustrating once again the enormous insights geolocation is providing into the spatial ecol- ogy of seabirds, particularly in closely related taxa with few morphological differences and unclear taxonomic status. In contrast with our results, previous tracking studies on Barolo shearwaters in the Azores and Salvagens mainly showed a dispersive behaviour after breeding (Neves et al. 2012). In addition, there is no spatio-temporal overlap in distribution of the different taxa of the complex, pointing out substantial differences in their migratory behaviour and distribution and potential for lineage divergence, which deserves some attention when discussing the taxonomy within the little shearwater complex.
To understand year-round trophic ecology and study the existence of oceanic isoscapes through the analyses of stable isotopes in feathers it is essential to know the moulting patterns of the study model. Unfortunately, there is a lack of information about moult in Boyd's shearwater. Shearwaters usually show simple descendent moult that takes 3 to 5 mo to complete (Bridge 2006), starting with the innermost primary feather (P1), which in some species may be moulted even before the bird leaves the breeding area (Cramp & Simmons 1977, Monteiro et al. 1996. Ac cording to our geolocation data, birds spent on average 114 d outside of the breeding area, which theoretically should leave enough time to complete moult in the non-breeding area. Our SIA supports this hypo thesis. P1 and R6 were isotopically similar, suggesting that they are moulted in the same area, probably near the breeding area at the end of the breeding and non-breeding period, respectively. These 2 feathers differed from the S1 and S8, which we inferred were moulted during the non-breeding period in the North Central Atlantic, since the isotopic values of the S8 showed a high correlation with the longitude of the centroid of the non-breeding area of each individual. We inferred P1 and R6 to be moulted in the same area (surroundings of the breeding colony), so we ex pected to find lower isotopic variability compared to S1 and S8, which were moulted in different non-breeding areas with potentially different baselines. Therefore, the larger range of isotopic values of the P1 and R6 than S1 and S8 may reflect the inter-individual variability in the phenology at the beginning and at the end of the moulting period, with some birds advancing or delaying their moulting patterns in relation to migration depending, for example, on their breeding success. Moreover, moulting pattern of rectrices is typically more asynchronic among and within individuals than the rest of the flight feathers (Ramos et al. 2009), adding variability in the timing of moult and in turn in the standard deviation and range of the isotopic values we found in R6.
The inter-annual variability in stable isotope values was low for nitrogen, but relatively high for carbon values. However, the broad pattern found in longitudinal gradients was similar over the years (Fig. S5). Baselines of nitrogen and carbon values are known to vary between seasons and years due to changing environmental factors (temperature) and/or productivity in marine environment (Goering et al. 1990, Rolff 2000, Graham et al. 2010. Inter-annual differences in stable isotopes were also found in Barolo shearwaters, but the origin is difficult to determine, since these differences may result from changes in diet, foraging areas and/or baseline conditions due to environmental factors, or a combination thereof (Neves et al. 2012, J. A. Ramos et al. 2015. Many seabird species cross the equatorial area of the Atlantic Ocean during their trans-equatorial migrations, but do not forage in this area for extended periods (González-Solís et al. 2007, Guilford et al. 2009, Hedd et al. 2012. So far, the only tracked species known to use the equatorial Atlantic waters as one of their main non-breeding areas is the Bulwer's petrel Bulweria bulwerii (Dias et al. 2015, although in a different period than the Boyd's shearwater, since Bulwer's petrels breed during the non-breeding period of the Boyd's shearwaters. Temporal segregation in the breeding cycles of Bulwer's petrel and Boyd's shearwaters may suggest that this is driven by competition for food, but their segregation in trophic level, as indicated by the greater δ 15 N in the former than in the latter (Roscales et al. 2011), would not support this interpretation. Instead, temporal segregation may partly result from competition for nesting sites . Indeed, previous studies on breeding seabirds of the tropical and subtropical Atlantic indicated that the trophic position of the Boyd's shearwater is the lowest among all pelagic seabirds, together with Barolo and Audu bon's shearwaters (Roscales et al. 2011, Neves et al. 2012, Mancini et al. 2014. No conventional dietary analysis of Boyd's shearwaters has been conducted so far (but see Bourne 1955), but its low trophic level indicates the consumption of small juvenile squid and fish and crustaceans, as found in the diet of the Barolo shearwater (Neves et al. 2012, J. A. Ramos et al. 2015. Previous studies have suggested seasonal changes in the diet of the Barolo shearwater (Neves et al. 2012, J. A. Ramos et al. 2015, as indicated by an increase in δ 15 N values in feathers moulted in the non-breeding season compared to those moulted in the breeding season, suggesting that shearwaters targeted prey with higher trophic level during the non-breeding season (Neves et al. 2012). We also found seasonal changes in the isotopic values quite consistent over 5 years, but changes in δ 15 N were the opposite, i.e. we observed a de crease in δ 15 N and an increase in δ 13 C values from feathers moulted in the non-breeding (S1 and S8) compared to those grown in the breeding season (P1 and R6; Fig. 4). However, this opposite trend in δ 15 N values and its significant correlation with longitude suggest that these changes just reflect baseline isotopic gradients in longitude (Fig. 5). Indeed, correlations between longitude of the non-breeding centroids with the isotopic values of feathers grown in this period broadly match iso scapes based on plankton samples from the same area (Somes et al. 2010, McMahon et al. 2013a. Spatial patterns indicating greater values of δ 13 C and smaller in δ 15 N in the central oligotrophic subtropical Atlantic Ocean were confirmed by a recent study on plankton biomass (Mompeán et al. 2013). Knowledge of baselines is also essential in any isotopic studies of trophic ecology, since baseline adjustment allows for the comparison of species from different geographical origin (Navarro et al. 2013). The strong longitudinal gradient in values of nitrogen and carbon found in this study suggests propagation of isotopic variability up to the food chain on a coarse scale. However, failure to find latitudinal gradients may be related to latitudinal error inherent to geolocation methodology. Another constraint in gradient models is the limitation in modelling techniques to incorporate all sources of uncertainty and error of location estimations. Furthermore, care should be taken, as the high isotopic variability among individuals and the reduction of the moulting area to a centroid may hinder the potential use of this isotopic gradient to infer the nonbreeding areas of un tracked birds. A study using data with more precise spatial resolution and more de tailed knowledge about timing of moult would be required to create complex isoscapes and investigate the potential geographic assignment to foraging movements or non-breeding areas of top predators in the tropical Atlantic Ocean using SIA, but our results show some promising potential for this.
Overall, in this study we provided detailed information about the year-round distribution, trophic ecology, phenology and moulting patterns of Boyd's shearwater. The combined use of geolocators and SIA allowed us to bring new insights to the biology and ecology of a poorly known tropical species.