Does behaviour affect the dispersal of flatback post-hatchlings in the Great Barrier Reef?

The ability of individuals to actively control their movements, especially during the early life stages, can significantly influence the distribution of their population. Most marine turtle species develop oceanic foraging habitats during different life stages. However, flatback turtles (Natator depressus) are endemic to Australia and are the only marine turtle species with an exclusive neritic development. To explain the lack of oceanic dispersal of this species, we predicted the dispersal of post-hatchlings in the Great Barrier Reef (GBR), Australia, using oceanographic advection-dispersal models. We included directional swimming in our models and calibrated them against the observed distribution of post-hatchling and adult turtles. We simulated the dispersal of green and loggerhead turtles since they also breed in the same region. Our study suggests that the neritic distribution of flatback post-hatchlings is favoured by the inshore distribution of nesting beaches, the local water circulation and directional swimming during their early dispersal. This combination of factors is important because, under the conditions tested, if flatback post-hatchlings were entirely passively transported, they would be advected into oceanic habitats after 40 days. Our results reinforce the importance of oceanography and directional swimming in the early life stages and their influence on the distribution of a marine turtle species.

The ability of individuals to actively control their movements, especially during the early life stages, can significantly influence the distribution of their population. Most marine turtle species develop oceanic foraging habitats during different life stages. However, flatback turtles (Natator depressus) are endemic to Australia and are the only marine turtle species with an exclusive neritic development. To explain the lack of oceanic dispersal of this species, we predicted the dispersal of post-hatchlings in the Great Barrier Reef (GBR), Australia, using oceanographic advection-dispersal models. We included directional swimming in our models and calibrated them against the observed distribution of post-hatchling and adult turtles. We simulated the dispersal of green and loggerhead turtles since they also breed in the same region. Our study suggests that the neritic distribution of flatback post-hatchlings is favoured by the inshore distribution of nesting beaches, the local water circulation and directional swimming during their early dispersal. This combination of factors is important because, under the conditions tested, if flatback post-hatchlings were entirely passively transported, they would be advected into oceanic habitats after 40 days. Our results reinforce the importance of oceanography and directional swimming in the early life stages and their influence on the distribution of a marine turtle species. rsos.royalsocietypublishing.org R. Soc. open  The net, tidal-averaged water circulation in the modelled study area. 'Zone A' is the wind-driven coastal boundary layer and represents areas where the currents favour a neritic dispersal, while 'Zone B' is the Coral Sea Lagoonal current, which flows seawards to the Pacific Ocean.
rookeries, such as Curtis Island, and minor (less than 10 nesting females/season) rookeries, such as Mon Repos. Most flatback rookeries in eastern Australia are located on the continental islands within the southern GBR, and to a limited extent on mainland beaches along the central Queensland coast. There are also major rookeries for green and loggerhead turtles within the same geographical extent of the southern GBR. However, they are distributed on the outer shelf coral cays and islands of the Capricorn Bunker Group and Swain Reefs (figure 1a). The nesting season for marine turtles in eastern Australia begins in mid-October until late January for flatbacks, and early May for green and loggerheads; the peak emergence of hatchlings for all species is in February [27].

Mean water circulation in the Great Barrier Reef
The southern GBR has two areas with distinct oceanography (figure 1b): Zone A is located inshore and is the wind-driven coastal boundary layer where the longshore northward current is highly macroturbulent with numerous eddies, jets and stagnation zones; Zone B is located further offshore and has a general southward current trend, which is the Coral Sea Lagoonal current formed by the intrusion of the East Australian Current (EAC) on the GBR continental shelf. This current is permanent but it waxes and wanes with the periodic formation and disappearance of the Capricorn Eddy, a cyclonic eddy which forms on the seaward side of the Capricorn channel [42][43][44][45]. The currents and the seaward water export are larger during the austral spring and summer months, in which the turtle nesting season takes place [27] Polygons represent the probability of distribution of s-turtles, with dark and light colours representing 50% and 95% probability, respectively. S-flatbacks were released from Wild Duck and Peak islands (red triangles from north to the south) and were modelled for 120 days until 00 h on 2 May 2012. S-green/loggerheads were released from North West, Heron and Wreck islands (purple triangles from west to east) and were modelled for 12 days until 00 h on 14 January 2012. After 120 days of simulation, the s-flatbacks under active swimming remained inside the GBR, displaying a neritic distribution, similar to that of observed flatbacks (panels a and c). In contrast, most drifting s-flatbacks were advected seawards into the Pacific Ocean by the East Australian Current (EAC).
the s-turtles accounting for the effect of the geographical location of nesting beaches, and the tidal phase (spring/neap) when the s-turtles entered the sea. In a later step, we added a swimming behaviour component to our simulations, and created a sensitivity analysis based on changing swimming speeds and directions, as well as the length of the passive dispersal phase.
Our dispersal simulations were based on releasing s-turtles in January 2012. This date was chosen as the climatic and oceanographic features in January 2012 were representative of the average hatching season, and the currents and wind values we used were within 1 s.d. of the average long-term conditions for the month. The variability in current velocity/direction (electronic supplementary material, figure S1a) and wind speed/direction (electronic supplementary material, figure S1b) in the southern GBR during the months that hatchlings disperse from beaches (December-February) fluctuates within a narrow range across years, with some exceptions due to extreme weather events. Thus, our modelling approach is based on assessing the variability in the dispersal due to (i) a selection of geographical/oceanographic parameters under typical climatic conditions, and (ii) a range of hypothetical swimming behaviours of the animals. The main focus of our study was to assess how behaviour influences dispersal patterns, and we acknowledge that different oceanic conditions such as monthly/yearly variability, especially those related to atypical weather events, might provide different dispersal predictions and it would be a worthy hypothesis to test in further studies.

Oceanographic advection-dispersal model with animal behaviour
The model domain covered the southern extent of the GBR Lagoon (figure 1). We used the unstructuredmesh SLIM model (2D Second-generation Louvain-la-Neuve Ice-ocean Model; www.climate.be/slim) to simulate the hydrodynamics of the region and the dispersal of s-turtles [48]. The mesh size varied between 150 m near reefs and islands to 10 km in offshore waters (figure 1a). The SLIM model has been calibrated for the GBR by Andutta et al. [49][50][51] and Thomas et al. [52], and used for ecological applications including the study of the accumulation and movement of marine debris [53,54], the nearshore dispersal of flatback hatchlings [29], and the fate of fine sediment in two dimensions and three dimensions [55,56]. Forcing parameters for the model included wind stress (wind speed and direction), water circulation in the Coral Sea, and the tides and tidal currents at the shelf edge following the model calibrations for the GBR [49][50][51][52]. Daily wind data between January and April, from the period 1999-2014, were provided by the Australian Bureau of Meteorology (BOM) and averaged to calculate the monthly mean wind stress during the hatching season.
The dispersal simulation allows for directional swimming of the s-turtles, following the method of Wolanski & Kingsford [5]. Scenarios were designed to test the hypothesis of passive drift (the null scenario) and active swimming dispersal (table 1). In all scenarios, the initial period of swimming frenzy was determined from published data, namely during the first hour the s-turtles swam eastwards at 0.3 m s −1 , and then swam for 72 h at 0.08 m s −1 [29,30,[57][58][59]. After the simulated swimming frenzy, either their swimming behaviour was stopped and they dispersed passively, or they swam directionally; the parameters behind each scenario are listed in table 1. The simulations lasted for 120 prototype days.
The passive drift was the null scenario (scenario PD-0), in which s-turtles were left to drift with currents with no additional behaviour. To comprehensively test this hypothesis, we designed five scenarios to account for the influence of the geographical location (GL) of rookeries and the tidal phase (TP) on the dispersal of turtles (table 1). To test whether the location of rookeries of the different species results in distinctive dispersal trajectories, we compared the passive dispersal of s-turtles released from a subset of rookeries where only flatbacks (n = 16) and only green/loggerheads (n = 3) nest (figure 1a; electronic supplementary material, table S1). We then analysed the s-turtle dispersal within the nesting range of the EA flatback population. We tested the differences in the passive dispersal of s-flatbacks (i) between regions (n = 8 rookeries in each region: Broad Sound and Capricornia), and (ii) within each region, comparing the major rookeries (n = 1 in each region; Wild Duck Island for Broad Sound and Peak Island for Capricornia) to a subset of minor/intermediate rookeries (n = 7 in each region) and non-nesting beaches (proximal beaches with no records of flatback nesting, n = 7 in each region) (figure 1a; electronic supplementary material, table S1). All subsets were randomly selected with the r. sample function in GME 0. 7.3.0 [60]. In each case 5000 s-turtles were released near the selected rookeries, on 3 January 2012 at 00.00. We used the major rookeries as a proxy for each region, namely Wild Duck Island for Broad Sound and Peak Island for Capricornia. To quantify the importance of the springneap tidal cycle at the time when turtles entered the sea at Wild Duck and Peak Islands (figure 1a, electronic supplementary material, table S1), we simulated the dispersal of 5000 s-turtles from each rookery released on two different dates: (a) at neap tide (3 January 2012 at 15.00) and (b) at spring tide (7 January 2012 at 12.00).
The effect of behaviour on the dispersal of s-flatbacks was tested through a sensitivity analysis by analogy with modelling fish swimming dispersal [5], where we tested different directions and speeds based on the currently known behaviour of flatback post-hatchlings [22,29,61] and expert opinion. A standard run (scenario DS-1-SR) was arbitrarily designed based on our best guess of what is known about the biology and swimming behaviour of flatback post-hatchlings, namely that the s-flatbacks swam north-westward at 0.02 m s −1 for 75% of the time (the remaining 25% of the time s-turtles drift passively). To test the sensitivity of this solution to other parameters, all other scenarios were arranged by changing one swimming parameter from the standard run (table 1). The parameters that defined the swimming behaviour of s-turtles were: (i) direction: northwest, north, south, east; (ii) speed: 0.01 m s −1 , 0.02 m s −1 , 0.04 m s −1 ; (iii) proportion of time s-turtles spent swimming per day: 25%, 50%, 75%, 100%; and (iv) day the post-frenzy swimming behaviour started: day 4 (immediately after the end of swimming frenzy), day 30, 60 or 90 (table 1). For all scenarios that tested the influence of the swimming behaviour we released 15 000 s-turtles near the major rookeries (Wild Duck and Peak islands), on 3 January 2012 at 00.00.

Outputs and data analysis
We calculated the area where 95% of the s-turtles were located (this is called the 95% distribution area), as well as core dispersal zones (area where 20% of the s-turtles were located) using the Kernel                To quantify the influence of behaviour in the dispersal, we calculated the dispersal success (percentage of s-flatbacks dispersed into inshore habitats of the GBR) for each individual scenario of the sensitivity analysis (26 scenarios in total; one passive drift and 12 directional swimming scenarios for each releasing location) (table 1). Then, for each scenario we averaged the dispersal success over the two releasing locations to obtain the overall dispersal success, which we ranked based on a quantile classification as very low (≤25%), low (25.01-50%), medium (50.01-75%) and high (>75%). The inshore region was delimited by the wind-driven coastal boundary layer (see 'Zone A' in figure 1b). S-turtles located in the Coral Sea Lagoonal current (see 'Zone B' in figure 1b) were considered to get swiftly advected outside the GBR, and thus counted as distributed into offshore waters.

Turtle dispersal under the passive drift hypothesis
The passive dispersal of s-turtles differed greatly between rookeries of the different species. Assuming passive transport for the s-green and s-loggerheads under typical oceanographic conditions, the simulations resulted in 92.5% of the s-turtles leaving the model domain and entering the EAC within 14 days. In contrast, under typical oceanographic conditions s-flatback turtles remained in lagoonal waters for 40 days before starting to leave the domain. Under most passive drift scenarios for s-flatbacks, the core dispersal zones after 120 days of simulation were located within the Coral Sea Lagoonal current flowing towards oceanic waters outside of the GBR (see 'Zone B' in figure 1b).
In addition, s-flatback post-hatchling dispersal differed significantly between the two regions (BAI = 0.590); s-flatbacks released from Capricornia under typical conditions reach Zone B sooner and are more likely to get advected into oceanic currents (figure 3). Despite the differences between the regions, the distribution area of s-flatbacks released from non-rookeries, minor rookeries and major rookeries within each region was similar (Broad Sound BAI minor/major = 0.906, BAI non/major = 0.993; Capricornia BAI minor/major = 0.942, BAI non/major = 0.933) ( figure 4).
Under the conditions that were modelled, the phase of the tidal sea surface elevation (neap/spring) at the time of release had an effect on the short-term (less than 7 days of dispersal) distribution of passive

The importance of directional swimming
The standard run (DS-1-SR) scenario displayed a reasonable fit with the sparse data on post-hatchling distribution ( figure 2b). Under the passive drift scenario, 67   In contrast, the lowest dispersal success (low proportion of s-flatbacks in inshore habitats) was observed when s-flatbacks were entirely passive (scenario PD-0) or maintaining headings towards the east (leading s-turtles directly into the EAC, scenario DS-4) or north (leading s-turtles into the Coral Sea Lagoonal current, scenario DS-6). Low dispersal success was also related to low swimming efforts by s-flatbacks; namely, scenarios in which s-flatbacks began to swim after more than 90 days (DS-12) or spent only 25% of the day swimming (DS-7).

Discussion
This study focused on simulating, with a high-resolution hydrodynamic model, the effect of behaviour on the dispersal of s-flatback turtles in the GBR, by assessing different scenarios of passive drift and active swimming. We suggest potential mechanisms that could drive the neritic development of flatback turtles. In particular, our models indicate that s-flatback post-hatchling distribution during the first months of their life is likely to be defined by the combination of (i) inshore location of nesting beaches, (ii) local tides

Table 2.
Dispersal success (percentage of s-flatbacks in inshore waters) of each scenario of the sensitivity analysis. The dispersal success for Wild Duck Island, Peak Island, the average over the two locations and the overall dispersal success are shown for each scenario. Overall dispersal success is based on the average, and ranked as very low (≤25%), low (25.01-50%), medium (50.01-75%) and high (> 75%). S-flatbacks were modelled for 120 days until 00. 00         It is also important to consider the oceanographic processes in this region, as well as their interaction with variable turtle behaviour. If the post-hatchlings were spending most of the time drifting near the surface, they would be swept directly into the EAC and dispersed into the Coral Sea. In contrast, if they were developing prolonged dives near reefs, the internal currents could increase their retention, as it happens with fish larvae and other small organisms [64]. However, if this was the case, we would expect to find more records of predated or stranded flatback turtles within the adjacent coral cays and islands of the Capricorn Bunker Group and Swain Reefs (figure 1b)-both of which are areas with high rates of reef-based dive and fishing operations.
Our results indicate that the advection rate of s-flatback post-hatchlings was similar among inshore major, minor and non-rookeries, as well as between tides. Combined, these results suggest that some of the evolutionary drivers that have favoured the selection of the larger flatback rookeries appear to be related to broad regional-scale differences (e.g. selection of inshore versus outer edge beaches). A previous study in the GBR suggested that nesting distribution might be related to the exposure of rookeries to wind and wind-generated waves that assist hatchlings' movement during the swimming frenzy, and indicated that flatback turtles seem to prefer sheltered beaches [65]. At a local scale, the nesting distribution of flatback turtles could also be influenced by other biophysical features (e.g. access to beach, presence of fringing reefs) which are not accounted for in our simulations.
Local water circulation patterns seem to have an important effect on the dispersal of post-hatchlings in coastal waters [29]. Based on our models, spring tides promote a nearshore dispersal during the first week and might prolong the retention of s-flatbacks in the GBR (figure 5a; electronic supplementary material, figure S2a). This finding is consistent with a previous study [29] which revealed that the tides and currents around nesting islands can influence the direction and spread of flatback hatchling dispersal during the first two weeks at sea. Distribution in nearshore areas might affect the probability of survival of post-hatchlings given that marine predators are likely to be more abundant in these areas [66,67], yet flatback post-hatchlings seem to have phenotype and behavioural adaptations which potentially decrease the risk of predation in coastal waters [22].
The region of origin also had an effect on the export/retention rate of s-flatbacks (figure 3). There was an increased seaward export of s-flatbacks from rookeries in the Capricornia region, likely to be related to the closer location of these rookeries to the Capricorn Eddy (situated around the Capricorn Bunker Group, figure 1b). While it has been suggested that the nesting distribution of other species of marine turtles is strongly associated with the proximity to favourable ocean currents [68], it is possible that, conversely, the nesting distribution of flatback turtles has been shaped by the selection of beaches with favourable local water circulation patterns [29], but distant from the main currents leading into the ocean.

Directional swimming in flatback post-hatchlings
Swimming behaviour appears to play a major role in the post-hatchlings' distribution during the first few months of their life. As mentioned before, our results provide evidence that the location of flatback nesting beaches favours the retention of s-flatback post-hatchlings during the first 40 days of dispersal. However, our simulations also suggest that under typical climatic conditions, after 40 days s-flatback post-hatchlings with no swimming are likely to be swiftly advected towards the Coral Sea into oceanic habitats where they have not been historically encountered. We tested a variety of hypothetical swimming scenarios, and conclude that directional swimming greatly decreased the seaward export of s-flatbacks, and further increased the likelihood of retention of s-flatbacks in inshore waters. This result was qualitatively validated with long-term field evidence given that, like our results, they also show that immature and adult flatbacks in eastern Australia occur within the Great Barrier Reef and coastal waters of southern Queensland (figure 2) [27,69]. Furthermore, the similarity among the modelled dispersal paths of s-flatbacks and the actual migration routes of existent tracked adult turtles (figure 2b,c) supports the theory that post-hatchlings might be able to imprint the location of habitats they pass through during their development, which can potentially influence the later migrations and distribution of juvenile and adult turtles [70][71][72].
In particular, s-flatbacks with active and directed swimming behaviour during the first months after entering the sea are more likely to disperse into neritic habitats. The range of horizontal swimming that favours s-turtle dispersal within the GBR (table 2; electronic supplementary material, figure S3) reveals that the more time and effort s-flatbacks invest in swimming, the higher is their chance of maintaining a neritic dispersal. Overall, our results suggest that, under typical oceanographic conditions, flatback post-hatchlings can have a short initial drifting phase, but afterwards they need to actively swim to increase the likelihood of a neritic distribution. According to our simulations, sustained speeds (more than 0.02 m s −1 ) and time spent swimming (more than 75% of the day) are influential parameters that increase the inshore dispersal success. In addition, our simulations show that accounting for directionality causes substantial differences in the dispersal outputs (table 2). For turtles released from the two major rookeries, a northwest heading (DS-1-SR) resulted in the best fit with the observed posthatchling distribution ( figure 2a,b). The scenario with a south heading (DS-3) also favoured a coastal retention of s-flatbacks, leading them to southern areas where reports of wild post-hatchlings occur but are scarce. The results of our study are based on hypothetical swimming parameters and provide on September 1, 2017 http://rsos.royalsocietypublishing.org/ Downloaded from valuable baseline criteria to experimentally test the orientation and swimming efforts of flatback turtles in the GBR.
While our study highlights the importance of considering swimming behaviour when simulating the early dispersal of marine turtles, numerous studies have shown that annual variability in oceanic conditions can greatly influence dispersal predictions [35,36,73]. As an example of this, simulations of the dispersal of young loggerheads in eastern Florida evidenced that not only the spread in dispersal patterns, but also the magnitude of the influence of navigational behaviour varied greatly among different years [36]. Future extensions of our modelling would be greatly enhanced by considering the influence of seasonal and inter-annual variability of the oceanographic conditions in the dispersal of post-hatchlings in the GBR.

Cues for directional swimming in marine species
There is a great array of sensory interactions between marine animals and their environment that probably play an important part in their behaviour, spanning from visual, olfactory or sound stimuli, to the detection of hydrodynamic and magnetic features [3,5,74]. For example, it has been proposed that coral reef fish larvae in the GBR use directional swimming to self-recruit to their natal reef, by orienting themselves using olfactory cues in open water, and auditory cues when within hearing range of their natal reef [5]. This is also the case for coastal post-larvae crab in New Zealand which use sound cues to find suitable environments to recruit [75]. The external cues that marine turtles may use to guide behaviour such as directional swimming during dispersal warrants further attention.
Directional swimming using Selective Tidal Stream Transport (STST) [7] is unlikely for turtle hatchlings because they would need to remain on the bottom for long periods to enable them to detect pressure changes and judge whether the water pressure (i.e. the tide) is rising or falling. There are, however, orientation cues that marine turtles are known to use, including the detection of wave direction during the swimming frenzy [76][77][78] and a combination of visual and magnetic signals to orient themselves in the open ocean [74,76,[79][80][81][82]. A hypothesis that has not been tested so far is the capacity of post-hatchlings to adjust their movements based on the direction of wind-driven waves in coastal waters: for example, this direction is primarily northwestwardly in the GBR, which, if adopted for directional swimming, our simulations show favours self-recruitment of flatback turtles. The use of sound or olfactory cues remains unknown. Thus, our understanding of the early dispersal of flatback turtles would be greatly enhanced by experimentally testing the orientation of hatchlings as a response to wind-driven waves, smell/sound plumes or magnetic fields [33,80,83], or directly tracking neonates with satellite or acoustic tags [18].

Conclusion
We provide a potential explanation to the differences observed historically in the dispersal strategies (oceanic versus neritic) of coexisting marine turtle species in the GBR. The neritic distribution of sflatback turtles was favoured by the inshore location of nesting beaches. In contrast, the outer location of reef cays and beaches, where s-greens and s-loggerheads nest, favoured a swift dispersal of these species into oceanic waters. The local water circulation in the different regions (Broad Sound and Capricornia) also seems to play an important role in the retention/advection of flatback post-hatchlings, especially during the first month of dispersal. S-turtles released during spring tides displayed a nearshore dispersal during the first week, which could have potential effects on their survival rates. Finally, under typical oceanographic conditions the inshore dispersal is further favoured if directional swimming of s-turtles is added to the models. In particular, our models suggest that high swimming efforts (in terms of speed and sustained swimming) and directionality could be important parameters to consider in future studies of flatback turtle dispersal. Our results highlight the value of integrating knowledge and applications from biological and physical sciences to improve knowledge of animal behaviour. High-resolution hydrodynamic models are a valuable tool to understand the influences of physical processes on the distribution of cryptic marine species and to infer potential pathways of dispersal [3,29,46,48]. Further advances in the resolution and numerical functions of hydrodynamic models, coupled with improved knowledge of species biology and navigation, are clearly needed. These kinds of improvements will enhance the understanding of the impact that small-and large-scale oceanographic events might have on the autonomous horizontal and vertical movements of species, and consequently on the behaviour and distribution of species, especially in view of the current changing conditions in the world's seas and oceans.