Atlantic Leatherback Migratory Paths and Temporary Residence Areas

Background Sea turtles are long-distance migrants with considerable behavioural plasticity in terms of migratory patterns, habitat use and foraging sites within and among populations. However, for the most widely migrating turtle, the leatherback turtle Dermochelys coriacea, studies combining data from individuals of different populations are uncommon. Such studies are however critical to better understand intra- and inter-population variability and take it into account in the implementation of conservation strategies of this critically endangered species. Here, we investigated the movements and diving behaviour of 16 Atlantic leatherback turtles from three different nesting sites and one foraging site during their post-breeding migration to assess the potential determinants of intra- and inter-population variability in migratory patterns. Methodology/Principal Findings Using satellite-derived behavioural and oceanographic data, we show that turtles used Temporary Residence Areas (TRAs) distributed all around the Atlantic Ocean: 9 in the neritic domain and 13 in the oceanic domain. These TRAs did not share a common oceanographic determinant but on the contrary were associated with mesoscale surface oceanographic features of different types (i.e., altimetric features and/or surface chlorophyll a concentration). Conversely, turtles exhibited relatively similar horizontal and vertical behaviours when in TRAs (i.e., slow swimming velocity/sinuous path/shallow dives) suggesting foraging activity in these productive regions. Migratory paths and TRAs distribution showed interesting similarities with the trajectories of passive satellite-tracked drifters, suggesting that the general dispersion pattern of adults from the nesting sites may reflect the extent of passive dispersion initially experienced by hatchlings. Conclusions/Significance Intra- and inter-population behavioural variability may therefore be linked with initial hatchling drift scenarios and be highly influenced by environmental conditions. This high degree of behavioural plasticity in Atlantic leatherback turtles makes species-targeted conservation strategies challenging and stresses the need for a larger dataset (>100 individuals) for providing general recommendations in terms of conservation.


Introduction
Many species show considerable behavioural plasticity in terms of foraging and habitat use in response to fluctuations in environmental conditions and prey availability [1][2][3][4][5], or to changes in energetic requirements associated with the different stages of the annual cycle (e.g., reproduction, migration [6][7][8]). In addition, a high degree of phenotypic plasticity usually exists between geographically separate populations experiencing different ecological conditions. For instance, rockhopper penguins Eudyptes chrysocome from three different colonies in the Indian Ocean have been reported to show significant differences in diving behaviour and foraging effort with consequences on life history traits such as chick growth [9]. Similarly, gravid green turtles Chelonia mydas have been shown to exhibit contrasted, probably food-mediated, patterns of depth utilisation between Ascension Island (mid-Atlantic) and northern Cyprus (Mediterranean Sea) [10].
High degree of behavioural plasticity within a species may make species-targeted conservation strategies more difficult to implement. For instance, Cape gannets Morus capensis from two colonies off South African coasts show contrasted foraging strategies: birds from one colony feed on natural prey, i.e. pelagic fish targeted by fisheries, while occupants of the second colony feed mainly on fishery wastes [11]. Therefore some fisheries may increase food availability for gannets through waste while other fisheries compete directly with the birds when harvesting their main natural prey, making the implementation of any conservation policies in this area particularly challenging [12][13][14]. This example highlights the difficulty of implementing efficient conservation strategies at a species level without taking into account interpopulation variability in terms of foraging and dispersal behaviour.
Sea turtles are long-distance migrants that exhibit a high variability in migration destination among individuals of a same population and among populations [15]. The potential determinants of migration destination have recently been investigated in the loggerhead turtle Caretta Caretta from a major rookery in the Mediterranean [16]. It appeared that the pattern of adult dispersion from the breeding area closely matched the different drift scenarios that would have been experienced by hatchlings as they first left their natal beach. In their early lives as they passively drift in ocean currents, turtles may explore different habitats and potential future foraging areas. Then, as adults, they may use this initial experience to migrate to predictable foraging sites. This hypothesis of ''hatchling drift scenarios'' has also been suggested to explain the genetic connectivity between geographically distant populations of green turtles [17].
Here, we investigated the movements and diving behaviour of both north and south Atlantic leatherback turtles during the postbreeding migration of 12 individuals from three different nesting sites and 4 individuals captured at one foraging site to assess the potential determinants of intra-and inter-population variability in migratory patterns. We particularly focused on oceanographic conditions encountered by the turtles during the migration in order to test potential hatchling drift scenarios at the Atlantic Ocean scale.

Ethics statement
This study adhered to the legal requirements of the countries in which the work was carried out, and to all institutional guidelines.  (5.7uN-53.9uW) and three in Gabon at Kinguere beach (0.2uN-9.2uW). One turtle was equipped in Uruguay at Kiyu (34.7uS-56.7uW) after it was incidentally captured by an artisanal bottom-set gillnet, and three were equipped in international waters of the Southwestern Atlantic (29.5uS-41.7uW; 28.3uS-44.0uW and 28.2uS-44.3uW respectively) after they were incidentally captured by Uruguayan pelagic longliners. Among these 16 turtles, 14 were mature females, one was a mature male (UR06-2) and one a subadult (UR06-1; Table 1). Most of the tagged animals were females as, for logistical reasons, fieldwork mainly occurred at the nesting sites. Some of these tracks have been previously published [20,26,36] but not the post-breeding migrations of the turtles nesting in Gabon, which are described for the first time in the present study. For all turtles, SRDLs were attached on the pseudocarapace using custom-fitted harness systems except for two turtles (FG05-4 and FG05-5) for which SRDLs were directly attached to the carapace [36].

Turtle movement analysis
Turtle movements were reconstructed using the Argos satellite location system (www.cls.fr). Inter-nesting tracks occurring during the nesting season were not included in the analysis. All tracks were processed in a similar way as in Gaspar et al. [37]: all locations of all accuracies were analysed, however Argos locations implying an apparent speed above 2.8 m.s 21 (i.e. .10 km.h 21 ) were discarded as travel rates above this threshold are considered as biologically unlikely [32]. Tracks were then smoothed and resampled every 3 hours. This sampling interval provides a spatial resolution sufficient for sampling the mesoscale variations of the ocean current fields and thus correctly estimating the currents along the tracks (see below). A local linear regression with a time window of two days was used to re-sample the tracks. Epanechnikov kernel was used to weigh observations in that window, and eventually adjust the size of the window according to the quality of the data in order to avoid over-smoothing the tracks. Re-sampled tracks (hereafter referred as apparent path) were analysed in three ways, as described below.
First, thanks to the regular re-sampling interval used, we calculated the time spent in 1u latitude by 1u longitude areas along the apparent paths in order to distinguish sections where turtles spend significantly more or less time, hereafter referred as Temporary Residence Areas (TRAs) and transit areas, respectively. When considering the cumulative frequency distribution of the time spent per 1u * 1u area, the curve reveals an inflection at the ypoint corresponding to 90 hours (i.e. 76.1%). Accordingly, we considered that for each turtle, a TRA could be defined as 1u * 1u area where the animal spent at least 90 h. All tracks were thus divided into several sections (TRA vs transit) for which behavioural parameters were calculated (see below).
Secondly, due to the impact that ocean currents may have on an animals' movements [37][38][39] we estimated the surface currents experienced by each individual in order to distinguish the animal's apparent path (including a current drift component) from its own swimming motion (hereafter referred as motor path). In short, this consisted of computing surface velocity fields on a daily basis, by summing the geostrophic and Ekman components deduced from altimetry and wind stress data, respectively (www.aviso.oceanobs. com). Then, at each 3-h re-sampled location, we calculated (1) an apparent velocity, (2) a local surface current velocity and (3) a swimming velocity, corresponding to the difference between the apparent and the current velocities. This current correction was performed for all turtles except those remaining at low latitudes (,10u) where geostrophic approximations break down [37].
Last, we considered that an animal could stay in any given TRA either by decreasing its travel rate or by modifying the spatial structure of its apparent path, i.e. its apparent path straightness. Straightness variations can be detected along a path by successively measuring the ratio D/L for path sections with a constant length L. Consistently, each apparent path was resampled in a form of a sequence of n steps with a constant length l (l = 15 km in the present study, corresponding to the average distance between our successive Argos locations), and the ratio D i / L was successively calculated for each location (x i , y i ) at the centre of a 10-steps (L = 150 km) window, i.e. between location (x i-5 , y i-5 ) and location (x i+5 , y i+5 ). To further investigate the relation between the apparent path and the swimming behaviour of the turtle, the same procedure was applied to the motor paths.

Turtle diving behaviour
SRDLs provided measurements of diving behaviour from a pressure sensor, which sampled depth every 4 seconds with an accuracy of 0.33 m. Data were statistically summarised onboard over 6-h collection periods providing the number of individual shallow (between 2 and 10 m) and deep (.10 m) dives performed during the period, their mean (6 SD) duration and mean (6 SD) maximum depth, as well as the proportion of time spent at the surface and diving (in shallow or deep waters). SRDLs continuously logged summaries but only a sample of these data was relayed by satellite because of the limited bandwidth of the Argos link. For each temporary residence/transit area identified as above, the above mentioned dive parameters were averaged for statistical analyses.

Satellite-derived oceanographic data
In addition to the estimation of the surface current fields (see above), the oceanographic regions crossed by the turtles were characterised using bathymetry, chlorophyll a data and altimetry. Bathymetry data were issued from the National Geophysical Data Center, National Oceanic and Atmospheric Administration, U.S. Department of Commerce, at a spatial resolution of 1/30u (ETOPO2v2; www.ngdc.noaa.gov). The seafloor regimes were subdivided as follows: neritic (i.e. continental shelf waters (,200 m) and shelf slope (200 to 2000 m)) and oceanic (.2000 m). Chlorophyll a surface concentration was described using monthly grids produced by the SeaWiFS project (spatial resolution of 9 km; http://web.science.oregonstate.edu/ocean.productivity/). Altimetry data obtained from AVISO (www.aviso.oceanobs.com) provided weekly maps of sea level anomaly (MSLA) and maps of absolute dynamic topography (MADT) on a 1/3 * 1/3u Mercator grid. Both MSLA and MADT data underwent a time linear interpolation to obtain daily gridded fields.

Drifter data
To assess the potential drift scenarios of passive particles from our different tagging sites, we used the Global Lagrangian Drifter Center (DAC) at NOAA's Atlantic Oceanographic and Meteorological Laboratory (AOML) assembles these raw data, applies quality control procedures and interpolates them via kriging to regular 6-h intervals. Here we selected satellite-tracked buoys that have passed within a window with 65u of amplitude in longitude and latitude (1) centred on each tagging site or (2) centred on a particular TRA.

Migration patterns
Tracking duration of the sixteen turtles ranged from 103 days (FG05-4) to 715 days (SU05-1) for recorded distances ranging from 2834 to 17 614 km ( Table 1). Distinct dispersal patterns were observed according to the tagging location and 22 Temporary Residence Areas (TRAs) were identified (Fig. 1).
Suriname -French Guiana complex. The six females which left French Guiana and Suriname between June and July 2005 dispersed widely but remained into the North Atlantic. Four females dispersed north-eastward (FG05-1, FG05-2, FG05-3 and FG05-4), reaching the Azores Front (between 34uN and 41uN, TRA1) at the end of summer/beginning of autumn. They spent between several weeks to several months in this oceanic area before three of them headed south at the end of autumn/ beginning of winter towards the Cape Verde islands. One female headed north-westward (FG05-5) and reached the Eastern continental shelf of USA (TRA2) in October 2005 where she remained until transmission stopped one month later. The last female (SU05-1) dispersed eastward reaching the Guinea Dome area (between 10uN -14uN and 23uW -   .4 months in the estuary, she headed northeast towards tropical waters before transmissions ceased in July 2008.

Drifter trajectories
Buoys travelling off the French Guiana-Suriname coasts have been shown to drift in different directions (Fig. 3). First, northwest towards the North American coasts (B1) and then possibly drift into the Gulf Stream until they reach the Azores (B2). From the Azores, the buoys can travel northward to the Irish Sea and the Bay of Biscay (B3), eastward to the Iberian coasts (B4), or southward to the Cape Verde islands, via the Canaries Islands (B5). Secondly, buoys can travel broadly northward to the Gulf Stream area (B6 and B7) and then drift to the east (B2). Last, they can travel eastward to the African coasts reaching the Guinea Dome area (B8 and B9). Buoys travelling off the Panama coasts (Fig. 3) can travel first northward to the Gulf of Mexico, and then possibly disperse either to the east (B10) or to the west into the Gulf (B11) or travel eastward by drifting into the Gulf Stream (B2). Buoys travelling off the Gabon coasts (Fig. 3) can travel westward into the South Atlantic Gyre (B12), from where they can end up on the South American continental shelf (B13), they can then travel south-eastward along the Brazilian coasts (B13). Buoys travelling off the Uruguay coasts (Fig. 3) can travel southward to the Brazil-Malvinas confluence area (B14). Although such data should be taken with caution as they were collected at different periods, they suggest that passive objects may drift from our different tagging sites and reach all the leatherback TRAs identified in this study, in approximately 1 to 3 years.

Environmental characteristics of temporary residence areas
For two turtles (FG05-1 and FG05-3) no temporary residence areas were identified possibly due to the relatively short duration of their tracks (,4 months) and/or the low quality of the data towards the end of the tracks. For the 14 remaining turtles, TRAs were located both in the neritic (e.g. TRA7, 10, 21 Figs. 1, 2) and the oceanic zone (e.g. TRA1, 11, 13; Figs. 1, 2) and were characterised by a high diversity of oceanographic conditions. Amongst the neritic TRAs, one (TRA21) was located in the estuary of the Rio de la Plata characterised by a high chlorophyll a surface concentration whereas others (e.g. TRA2, 7, 10) were located on the edge of continental shelves with a steep slope. Amongst oceanic TRAs, two were located in highly dynamic areas characterised by important mesoscale eddy activity: the Gulf Stream (TRA11, Fig 4a) and the Brazil/Malvinas Confluence (TRA22), others were located in the Azores Current (TRA1), the Guinea Dome area (TRA3) and the South Equatorial Current (TRA12, 13, 16) characterised by oceanic fronts clearly highlighted in maps of absolute dynamic topography (MADT, Fig. 4b). All TRAs of Gabonese turtles were situated in the South Equatorial Current characterised by high chlorophyll a surface concentrations (Fig. 4c).

From the nesting site to the first temporary residence area
All turtles satellite-tagged on their nesting beach reached their first TRA after 21 to 99 days of transit with a high mean swimming and apparent velocities (typically .45 cm.s 21 , i.e.  (Table S1). Turtles from Gabon spent a lower percentage of time between 0-10 m deep compared to other turtles and performed shallower dives (Table S1).

From transit areas to temporary residence areas
As turtles reached a TRA, there were marked changes in their vertical and/or horizontal behaviour depending on the type of habitat they exploited.
The passage from a neritic transit area to a neritic TRA (FG05-5, PA05-5, UR06-2, UR06-3) was associated with a decrease in swimming velocity (Kruskal-Wallis followed by a post-hoc Bonferroni test, p,0.05 in all cases, Table S1, Fig. 5) and in the mean straightness index for the motor path while dive parameters remained similar except for UR06-2 and UR06-3 for which dive depth decreased.
The passage from a neritic transit area to an oceanic TRA occurred only once (PA05-4) and was associated with an increase in dive duration (Table S1).

Within neritic temporary residence areas
Within neritic TRAs, the mean swimming and apparent velocities were typically low (,45 cm.s 21 , i.e. 39 km.day 21 , Table S1, Fig. 5) with a lower straightness index along the motor and apparent paths than before reaching the TRA (mean D/L typically ,0.8). Within neritic TRAs, turtles spent a majority of their time in the upper water column with more than 40% of their time spent between 0-10 m (up to 69% for SU05-1, Table  S1) while dives were typically shallow (,50 m) and short (,20 min, Table S1, Fig. 5). Turtles PA05-4 and PA05-5 as they mostly remained along the continental slope of the Gulf of Mexico performed deeper (between 60 and 140 m) and longer (typically .20 min) dives. Compared to transit areas, the diving effort in term of total number of dives per hour increased regardless the initial domain (neritic or oceanic) they came from.

Within oceanic temporary residence areas
Within oceanic TRAs, mean swimming and apparent velocities were highly variable among individuals depending on the actual oceanic dynamics assessed through current velocity (Table S1, Fig. 5). Accordingly turtles showed variable spatial structure of their path (i.e. path straightness) while remaining within an oceanic TRA: (1) in fast-current TRAs such as the Brazil/ Malvinas Confluence and the Gulf Stream, turtles UR06-3 and PA05-2 had relatively fast swimming and apparent velocities (typically .45 cm.s 21 , i.e. 39 km.day 21 ,) but a relatively lower straightness index for both the motor and apparent paths (typically ,0.8).
(2) Yet, in similar fast-current oceanic TRAs such as the Loop Current, turtle PA05-4 showed a high straightness index for its motor path, a high swimming velocity opposite to the main current resulting in a slow apparent velocity and a low straightness index for the apparent path. (3) Conversely, in low-current oceanic TRAs, such as the South Equatorial Tropical Gyre, turtle UR05-1 showed low swimming and apparent velocities (typically ,30 cm.s 21 , i.e. 26 km.day 21 ) but a high straightness index for both motor and apparent paths (typically .0.8) whereas turtles SU05-1, FG05-2 and FG05-4 showed a low straightness index for the motor path with similar low swimming and apparent velocities (typically ,35 cm.s 21 , i.e. 30 km.day 21 ). (4) Finally, all three Gabonese turtles showed low apparent velocities (typically ,30 cm.s 21 , i.e. 26 km.day 21 ) in the South Equatorial Tropical Gyre with either low (GA06-1) or high (GA06-2 and GA06-3) straightness index for the apparent paths.

Discussion
For the last ten years, many studies have investigated in detail the diving behaviour and movements of leatherback turtles during their migration cycle in the Atlantic Ocean [19][20][21][24][25][26][27][28][29][30][31][32][33][34][35]. For instance, in the North Atlantic, Ferraroli et al. [19] and Hays et al. [29] tracked females from their nesting sites in French Guiana and Grenada, respectively, while James et al. [31,32] tracked male and female leatherback turtles from an important foraging site in Nova Scotia. Evans et al. [26] described the migration patterns in the Gulf of Mexico of females nesting in Panama whereas in the South Atlantic, the recent study of López-Mendilaharsu et al. [20] focused on the behaviour of turtles captured in the Southwestern Atlantic Ocean. Yet to date, only one study concurrently investigated the migratory behaviour of leatherback turtles from both nesting and foraging sites in the North Atlantic basin [27]. The present study similarly brings together individual tracks but from three major nesting sites and one recently identified foraging area over the North and South Atlantic Ocean to identify temporary residence areas and associated environmental determinants. As such this study provides a new point of view on leatherback migration patterns and complements previously published works.

Atlantic migratory paths and TRAs
By monitoring 16 leatherback turtles from three nesting sites and one foraging area over the Atlantic ocean, this study clearly illustrates that the general dispersal patterns and TRAs used by the turtles may vary among individuals of a same nesting population and among populations. For instance females tracked from the nesting sites in French Guiana and Suriname only dispersed through the North Atlantic basin heading broadly northwest, northeast, or east (this study and [19,27]) whereas two of the three females tracked from their nesting beach in Panama dispersed in the Gulf of Mexico and the third one reached the Gulf Stream area (this study and [26]). To date, no satellite-tracked females from the Caribbean, French Guiana or Suriname nesting populations have ever entered the Gulf of Mexico or travelled south to the South Atlantic. In the Southern hemisphere, all three females tracked from Gabon dispersed through the South Atlantic basin mainly remaining within the South Equatorial Current while the turtles captured in coastal and oceanic waters off South America remained in the Southwestern Atlantic (this study and [20]). So within nesting populations, there is a tendency for migratory paths to be broadly similar (i.e. remaining within the same ocean body such as North Atlantic or Gulf of Mexico) but with large variation existing between the extreme paths taken (e.g. FG05-5 and FG05-3). Yet, there is a much greater variability of migratory paths between populations.
We identified 22 TRAs distributed throughout the Atlantic Ocean, 9 in the neritic domain and 13 in the oceanic domain. This corroborates previous studies suggesting that leatherback turtles are both oceanic and neritic foragers [20,25,40]. As a consequence, these TRAs did not share a common oceanographic determinant but on the contrary were associated with mesoscale surface oceanographic features of different types (i.e. altimetric features and/or surface chlorophyll a concentration). Several TRAs were located in distinct oceanic frontal zones and eddies. The importance of oceanographic fronts to this species, but also to marine birds and mammals (review in [41]) has already been described [19,24,34,42]. Other TRAs were located in estuaries and along coastal shelf breaks that constitute sharp water density discontinuities where biomass concentrates, including gelatinous zooplankton, the leatherback prey [43][44][45]. Slope waters seem indeed of important use for leatherback turtles. For instance, turtles PA05-4 and PA05-5 spent most of their time along the continental slope of the Gulf of Mexico, maybe foraging on gelatinous zooplankton aggregated along the shelf-break front [43]. All TRAs used by the turtles have been previously described as productive areas: e.g. the Mauritania upwelling [46], the Gulf of Mexico [47], the Gulf Stream [48], the Brazil/Malvinas Confluence [49], and the estuary of Rio de la Plata [50,51] suggesting that TRAs may indeed be associated with foraging. In addition, several TRAs identified in this study closely match the high-foraging success areas previously identified for leatherback turtles during their pluri-annual migration in the North Atlantic [27]. Interestingly, individuals from a same nesting area may show contrasting patterns in habitat use such as PA05-5 only exploiting oceanic TRAs and PA05-2 only neritic ones. Migratory paths and habitat use patterns in the leatherback turtle thus are both characterized by high intra-and inter-population variation.

Vertical and horizontal behaviours within TRAs
Despite highly variable oceanographic conditions among TRAs, turtles interestingly rather exhibited relatively similar horizontal and vertical behaviours when in TRAs. First, when taking into account the influence of surface currents on the horizontal behaviour of the animals, it appears that, in general, turtles slowed down their swimming velocity as they reached TRAs and exhibited highly sinuous motor and apparent paths. This may be associated with area-restricted searching (ARS) patterns that other marine predators display when foraging [52][53][54]. However, in certain cases this general behaviour was shaped by local current conditions. This was revealed by the method used in this study which assesses the contribution of both the animal and the environmental cues to the way an animal remains in TRAs. For instance, within zones of high mesoscale activity (presence of many eddies) turtles rather increased their swimming velocities while performing sinuous movements to remain in the productive patch (e.g. turtles UR06-3 and PA05-2). An interesting case is the turtle PA05-4 that remained at the edge of the Loop Current for several months showing a highly sinuous apparent path and a low corresponding velocity but a straight motor path and high swimming velocity. This suggests that during several months, the turtle headed in a direction opposed to the Loop Current while she apparently remained in a restricted area looping within the flow. This behaviour might be an original strategy by which turtles feed at counter-current. Indeed, swimming at countercurrent allows an animal to prospect water mass and thus potentially a prey patch without moving with respect to the sea bottom. Such behaviour may provide some benefits, as, for example, in terms of orientation by limiting extensive drifts throughout the oceanic basin, or in terms of foraging by maintaining the animal in an area where surface resources availability may be driven by deep, bathymetricmediated, oceanic processes. This behaviour has been previously suggested for a leatherback turtle foraging in the Azores Current [37]. Different horizontal tactics seem thus to be used by the turtles to remain in a productive patch according to local oceanographic conditions. This highlights the necessity to cautiously interpret horizontal movement patterns in marine predators in relation to contemporaneous environmental dynamics [22,37]. Novel tracking technologies such as fastlocH GPS loggers by improving accuracy in tracking marine species [55] may help resolving the underlying patterns of movement in great details and allow a better understanding of relationships with environmental parameters.
Shallow diving behaviour was observed in all TRAs at all latitudes in a relatively homogenous way among individuals. In oceanic TRAs, dives were longer (.20 min) than in neritic TRAs and mainly concentrated in the epipelagic layer (50-80 m). This suggests that the diving behaviour was shaped by local prey distribution and density, as described for other marine vertebrates (e.g., [56,57]). Periods of very short shallow dives and high use of surface waters have previously been reported for leatherback turtles foraging at high latitude [24,28,33] where gelatinous plankton is available at shallow depths [58,59]. Similar pattern was described in basking sharks (Cetorhinus maximus) foraging on continental shelves [52,57]. Higher variability in diving behaviour was observed in oceanic TRAs. Such variability in oceanic areas has also been observed in other marine species, particularly sea birds [60] and is likely driven by the stochastic nature of the oceanic environment resulting in less predictable and patchily distributed prey. This suggests that in neritic and geographically well-delimited TRAs, such as the Rio de la Plata estuary, where turtles exhibit relatively consistent diving patterns, spatio-temporal fishing regulations to mitigate bycatch may be more easily designed than in oceanic TRAs.

TRA fidelity and hatchling drift hypothesis
On one occasion, two individuals, one from the Southeast Atlantic and one from the Southwest Atlantic, stayed in the same TRA suggesting a potential connection between turtles from both sides of the South Atlantic. Leatherback turtles flipper-tagged on the beaches of Gabon have indeed previously been recovered in the waters of Argentina and Brazil [18] suggesting that turtles captured in international waters of the Southwest Atlantic likely belong to the West African nesting populations. Among the 16 turtles tracked in this study, several of them showed strong fidelity to TRAs (Fig. 2). Fidelity to a specific area has already been described in leatherback turtles foraging in Nova Scotia and in the Rio de la Plata estuary [20,32] but also in other sea turtle species [61]. Such behaviour is counterintuitive considering the high variability in post-breeding migration destinations observed among turtles of a given nesting population or among nesting populations. Yet, both may be linked to initial hatchling drift patterns [16,17]. The possible drift scenarios of hatchling turtles dispersing from their nesting sites may be inferred by looking at passive drifter trajectories. Here most of the individual dispersal patterns observed in the North Atlantic, the South Atlantic and the Gulf of Mexico showed interesting similarities with the trajectories of some satellite-tracked drifters (Figs 1, 3), although such data should be taken with caution as they were collected at different periods. In addition, most of the TRAs used by adult turtles during their post-breeding migrations were located along the drifter trajectories corroborating the ''hatchling drift scenario'' hypothesis [16]. Indeed, it has been suggested that hatchling turtles may imprint on several possible future and predictable foraging sites during the years when they are passively carried by ocean currents. Then, as adults they may make the decision to go to the preferred site(s) based on that initial experience and may follow the same routes [16,17]. Clearly, not all hatchling drift patterns generate possible scenarios for adult migration because of differential mortality rate between oceanographic areas (Gaspar et al. submitted). In addition, not all adult migration patterns match a hatchling drift scenario. For instance, in this study, some females left French Guiana and crossed the North Atlantic Gyre in a southwest-northeast direction heading towards the Azores. In this area, ocean currents are very weak and such trajectory could not occur by passive drift. Many other drifter trajectories end up however around the Azores which indeed represent a TRA used by many turtles (this study and [24,27]). This suggests that adult leatherback turtles may return to specific sites previously explored in their early lives without, however, always following the same routes as hatchlings but rather use shortcuts.

Conclusion
Identification of habitat use and associated diving behaviour is the first step for effective conservation of marine vertebrates. In this study, 22 temporary residence areas that may correspond to foraging areas have been identified in contrasted oceanographic environments ranging from neritic to oceanic domains for 16 Atlantic leatherback turtles. The observed migratory paths and TRAs distributions appear to be related to multiple oceanographic conditions, and may be linked with initial hatchling drift scenarios [16]. This study thus highlights the importance but also the difficulty of implementing spatio-temporal fishing regulations over a large geographical scale and suggests that modification of fishing gears and fishing behaviours might be more efficient to protect such highly migratory species. Despite the sample size and diversity of study sites used in this study, it also appears that a larger multi-year dataset (at least .100 individuals) is needed through international collaborative efforts for providing general recommendations in terms of conservation of this critically-endangered species.

Supporting Information
Table S1 Summary of diving behaviour, swimming/apparent/ current velocities and time spent in transit area/temporary residence area (TRA)/inter-TRA in oceanic (O) or neritic (N) domains in 16 Argos tracked leatherback turtles during their migration between 2005 and 2008 (see Fig. 1). Transit areas correspond to the time turtles spent from their nesting beach to their first TRA. TRAs correspond to 1u * 1u areas where turtles spent more than 90 hours. Inter-TRAs correspond to the time turtles spent between two TRAs (see M&M for details). * for PA05-2, the 35 days at the end of the track were not taken into account due to the very few numbers of locations obtained during this period. Differences between areas were statistically tested using Kruskal-Wallis test followed by a post-hoc Bonferroni test. Different letters indicate significant (p,0.05) differences among areas. Values are expressed as mean 6 SD.