Sea turtles return home after intentional displacement from coastal foraging areas

Abstract

Vulnerable species may be removed from their normal habitat and released at a new location for conservation reasons (e.g. re-establish or augment a local population) or due to difficulty or danger in returning individuals to original sites (e.g. after captivity for research or rehabilitation). Achieving the intended conservation benefits will depend, in part, on whether or not the released animals remain at the new human-selected location. The present study tested the hypothesis that hard-shelled sea turtles along the coast of north-eastern Australia (9–28°S, 142–153°E) would not remain at new locations and would attempt to return to their original areas. We used satellite-tracking data gathered previously for different purposes over several years (1996–2014). Some turtles had been released at their capture sites, inferred to be home areas, while other turtles had been displaced (released away from their inferred home areas) for various reasons. All non-displaced turtles (n = 54) remained at their home areas for the duration of tracking. Among displaced turtles (n = 59), the large majority travelled back to their respective home areas (n = 52) or near home (n = 4). Homing turtles travelled faster and adopted straighter routes in cooler water and travelled faster by day than by night. Our results showed that displacement up to 117.4 km and captivity up to 514 days did not disrupt homing ability nor diminish fidelity to the home area. However, for homing turtles we infer energetic costs and heightened risk in unfamiliar coastal waters. Confirmed homing suggests that moving individuals away from danger might offer short-term benefit (e.g. rescue from an oil spill), but moving turtles to a new foraging area is unlikely to succeed as a long-term conservation strategy. Priority must rather be placed on protecting their original habitat.

Introduction

In diverse situations, wild animals may be removed from their normal habitat and subsequently released at a new location with expectation of a beneficial outcome. The objective may be to establish threatened species in a new area, reintroduce them in an area of local extinction, or augment a locally diminished population (for examples see Griffith et al. 1989; Fischer and Lindenmayer 2000). In addition, a localised environmental catastrophe may prompt the removal of vulnerable animals to a safer location (e.g. Barham et al. 2006). Furthermore, animals that have been temporarily held in captivity, e.g. for research or rehabilitation, may be released at locations distant from their area of origin for logistical feasibility or in expectation of more favourable conditions for the animals.

Biological background knowledge is essential in assessing the feasibility of moving vulnerable wild species (Stamps and Swaisgood 2007; IUCN/SSC 2013). A fundamental question must be considered for a highly mobile species (Stamps and Swaisgood 2007) namely: can the displaced animals be expected to remain at the new location? Clearly a positive answer is necessary to meet most conservation goals, yet a negative answer must be assumed if animals are expected to return home after displacement for research or rehabilitation. For many species, no clear answer is available.

For hard-shelled sea turtles (Cheloniidae), there is evidence of long-term fidelity to foraging sites, long-term fidelity to breeding sites, and the capacity for migration between these sites at irregular intervals (Miller 1997; Plotkin 2003). However, these common patterns can be subject to variation. For example, seasonal and ontogenetic shifts in foraging habitat have been reported for some species at some locations (e.g. Musick and Limpus 1997; Morreale and Standora 2005; Shimada et al. 2014). Consequently, inference from natural behaviour offers uncertain guidance about potential responses of sea turtles to unnatural displacement.

Direct studies of displaced turtles have predominantly investigated the ocean navigation ability of adult female turtles after experimental displacement from breeding sites (e.g. Luschi et al. 1996, 2001; Hays et al. 2003a; Lohmann et al. 2008). Information about turtles displaced from coastal foraging areas tends to be sparse and site-specific (e.g. Limpus 1992; Avens et al. 2003) and largely reliant on recapture of marked animals. Although some displaced turtles in the Avens study (Avens et al. 2003) were radio-tracked briefly, that technology was unsuitable for continuous tracking over long duration and distance.

With satellite-linked devices, wild animals can be very effectively tracked over extended time periods and almost unlimited geographic range (Godley et al. 2008; Hazen et al. 2012). Platform transmitter terminals (PTTs) allow long-duration tracking with remote delivery of estimated positions, but location accuracy is relatively low (Hays et al. 2001; Hazel 2009). More accurate and more frequent locations can be obtained from Fastloc GPS (FGPS) receivers (Hazel 2009) although these must be linked with the Argos PTT system to allow remote data delivery. Despite the technical capacity of satellite-linked systems, research is typically limited by logistical and funding constraints. For the present study, these two factors, as well as ethical considerations, precluded a large-scale displacement experiment. Instead, we sought insight from tracking data that had been gathered for diverse purposes at diverse times during prior work with Cheloniidae in coastal foraging areas of Queensland, Australia.

The primary objective for our study was to investigate whether or not free-living sea turtles tend to remain at a new location after displacement from their foraging areas. Based on evidence of strong site fidelity at Australian coastal foraging sites (Limpus 2008), we hypothesised that the majority of our study turtles would not remain at their new locations and would attempt to make their way back to their original areas. However, we suspected that distance of displacement or duration in captivity might reduce a turtle’s motivation or ability to return to its original area. We therefore wanted to investigate environmental variables that could influence speed of travel and whether direct or indirect routes were adopted. In combination, speed and straightness of track would determine the overall duration of successful return journeys.

We accepted that an opportunistic study sample would not be comprehensive for all species or balanced for all variables of interest. However, the present study encompassed multiple species and a wide range of displacement situations that had occurred during prior work. By drawing on existing tracking data, we aimed to gain new insights while avoiding new deployment costs and additional intervention in the lives of turtles.

Materials and methods

Study turtles

We assembled 113 tracks of turtles that had been captured in shallow water (<10 m) in various tropical and subtropical foraging habitats of north-eastern Australia between 1996 and 2014 (Fig. 1; Table S1). Our complete data set comprised 79 green turtles Chelonia mydas, 30 loggerhead turtles Caretta caretta (one of them tracked twice), two olive ridley turtles Lepidochelys olivacea and one hawksbill turtle Eretmochelys imbricata. Turtles were captured for research by the rodeo method (Limpus 1978) (n = 105) and captured incidentally on a baited drum-line set by the Queensland Shark Control Program (n = 1). Other turtles were taken into care after being found debilitated on or near the shore (n = 6), hereafter termed rescued turtles. The study turtles included adult and immature individuals of both sexes as identified by laparoscopic examination of the gonads, by curved carapace length (CCL), or by combination of CCL and tail length (Limpus and Reed 1985; Limpus and Limpus 2003; Limpus 2008). Turtle sizes ranged from 38.1 to 121.2 cm CCL, median 98.0 cm (interquartile range 91.1–106.1 cm). Research turtles were released within 5 days of original capture. Rescued turtles were released after 69–514 days in rehabilitation centres (Table S1).

Fig. 1

Release sites of study turtles were dispersed widely along the coast of Queensland, Australia

Before release, each turtle was fitted with a tracking device attached to the carapace with epoxy glue and fibreglass (e.g. Shimada et al. 2012). Some turtles received a PTT (n = 27), while the majority (n = 86) received an Argos-linked FGPS device that provided PTT locations in addition to FGPS data. Turtles were released at <0.1 to 431.2 km from their capture locations. Tracking periods ranged from 5 to 915 days (Table S1).

Data preparation

Preliminary screening was applied to all tracks (i.e. both FGPS and PTT data), using the R package SDLfilter (available from https://github.com/TakahiroShimada/SDLfilter), to remove temporal and spatial duplicates and retain only a single fix (latitude/longitude pair) per time and location. For concurrent FGPS fixes, the fix derived from the highest number of satellites was retained (Hazel 2009; Shimada et al. 2012). For concurrent PTT fixes, the fix with highest location class (LC) was retained (CLS 2011). When concurrent fixes had the same quality index, the fix with the shortest summed distances to the previous and subsequent fix was retained. We excluded any locations acquired during breeding migrations. We also excluded any locations on land (above high tide line) because, in eastern Australia, foraging sea turtles rarely ascend beaches above the high tide line, although some individuals may rest on intertidal substrate where they become exposed at low tide (Limpus et al. 2005). All analyses were conducted using R software (R Core Team 2015).

Classification of displaced and non-displaced turtles

To determine (a) whether a turtle had been displaced from its original area and (b) whether displaced turtles returned to their original areas, we used PTT locations because these were available for all tracks (n = 113) and in some instances the PTT data provided a longer tracking duration than the corresponding FGPS data (in a device that used both tracking systems, the PTT component could remain functional after FGPS operation was halted by diminishing battery power or by epibiont growth on the GPS receiver).

To improve the relatively low accuracy of raw PTT locations, we fitted hierarchical Bayesian state space models (hSSM) following Jonsen et al. (2006). This technique provides more accurate location estimates by accounting for observation error and heterogeneity using tracking data from multiple animals. Because the process involves highly intensive computation, we balanced processing time against the benefits gained from multiple tracks as follows: our PTT data set was divided into 12 smaller portions with each subset containing 9836–12,903 observations acquired from 9 to 13 turtles. The model was fit to each subset of PTT data via two Markov chain Monte Carlo (MCMC) chains using the R package bsam, provided by Jonsen et al. (2013). Each MCMC chain was run for 300,000 iterations, excluding the first 200,000 samples as a burn-in. Every 100th of the last 100,000 samples was retained to reduce autocorrelation. Convergence and autocorrelation for hSSM were examined using diagnostic plots generated by the bsam package. The hSSM locations were estimated at six hourly intervals (mean interval of the raw Argos fixes). We dropped hSSM locations that fell on land and locations for periods when raw Argos fixes were absent for more than 5 days, the latter because error of hSSM locations appeared to inflate if 20 or more consecutive positions were missing (Bailey et al. 2008). Finally, the high-quality PTT locations (LC 3, 2, 1) were merged with the hSSM data. These locations, with an expected mean error of 2.2 km (Hoenner et al. 2012), are hereafter referred to as post-processed hSSM data.

We used the post-processed hSSM data to calculate the utilisation distribution (UD) for each turtle. To avoid problems of autocorrelation, we applied the movement-based kernel density method of Benhamou (2011) as implemented in the R package adehabitatHR (Calenge 2006, 2015a). To define the resettlement area of each turtle, we used the 95 % contour of the UD, with a buffer of width 2.2 km (expected mean error of our post-processed hSSM data). A turtle was deemed to have settled in the buffered 95 % UD provided the turtle did not move outside the 95 % UD for longer than 1 day. In cases where the 95 % UD comprised two or more disjunct polygons, an earlier polygon was excluded if the turtle had moved out of it and did not return to it.

A turtle was classified as displaced if its release location was outside its resettlement area, and classified as non-displaced if its release location was within its resettlement area (Fig. 2). Provided the capture location was contained within the resettlement area, the resettlement area was deemed to represent the original area of that turtle. Thus, a displaced turtle that subsequently returned to its original area was regarded as returning home (Fig. 2b). If a turtle did not return to its original area, the distance between capture and resettlement was measured to the periphery of the resettlement area. In the special case where transmission ceased while a turtle was still travelling (n = 2), the resettlement area could not be estimated. In this situation, we classified the turtle as non-displaced if the distance between its capture and release locations was shorter than 95th percentile diameters of circularised resettlement areas of all other turtles (16.8 km, n = 111). If the distance was greater than this, the turtle was classified as displaced.

Fig. 2

At site 1, a loggerhead turtle T53800 was tracked twice. a On the first occasion in 1998, the turtle was not displaced. After release it remained in its original foraging area. b On the second occasion in 2010, the turtle was displaced by 18.3 km from its capture location. It travelled back to its original area and thus was regarded as a homing turtle. Square capture location, triangle release location, cross-hatched polygon resettlement area. Grey line is the travelling path after displacement. Empty circle location of relatively low residency, filled circle location of relatively high residency

Detailed analyses for homing turtles

Displaced turtles that returned to their original areas were classified as homing turtles. For these turtles, we merged FGPS locations with high-quality PTT locations (LC 3, 2, 1) and then used the R package SDLfilter to apply additional filtering as follows. In order to remove locations above the high tide line, the water depth at track locations was estimated using bathymetry models and tidal data. Horizontal resolution of the bathymetry models was 110 m for one release site (site 11, see Fig. 1) (Daniell 2008) and 100 m for the other release sites (Beaman 2010). Tidal data were obtained from the Australian Bureau of Meteorology and Queensland Department of Transport and Main Roads. Filtering according to water depth was applied to the high-quality PTT locations and to the FGPS fixes derived from four satellites. Filtering by water depth was deemed unnecessary for FGPS fixes derived from >4 satellites because these fixes had estimated accuracy <64 m at site 1 and <33.1 m at site 8 (Hazel 2009; Shimada et al. 2012), thus higher accuracy than the bathymetry models. After filtering by water depth, we applied a data-driven filter following the method of Shimada et al. (2012). Location fixes were removed if the speed both from a preceding location and to a subsequent location exceeded the maximum realistic swimming speed (V _max) or if all of the following applied: (a) fixes were derived from only four GPS satellites or from the PTT system, the inner angle (180° minus the animal’s turning angle) was <90°, and the speed either from a preceding location or to a subsequent location exceeded a maximum “loop trip” speed (V _lp) estimated for each species (Table 1). Estimated error (mean ± SD) for high-quality data filtered by this method was 47.1 ± 61.0 m (Shimada et al. 2012).

Table 1 Data preparation for detailed analyses of homing turtle behaviour: threshold speed (km h⁻¹) used in the data-driven filter

To investigate homing behaviour in detail, our analyses focused on the homing segment of the track, that is, from point of release to the first location of relatively high residency within the resettlement area (Fig. 2b). We used the residence time method (Barraquand and Benhamou 2008), implemented in R package adehabitatLT (Calenge 2006, 2015b) to distinguish locations of relatively high and low residency. We excluded from our detailed analysis any homing tracks that included locations <100 m from land, other than during the first 3 h after release. This was necessary because very close proximity to land would restrict direction of travel and introduce a confounding effect on straightness of the track. We calculated straightness index (Batschelet 1981) equal to straight-line distance from first to last location (beeline distance) divided by summed track length. Summed track length was simply the sum of distances between successive locations along the track. Thus, a turtle swimming in a straight line all the way would have straightness index = 1 and a turtle swimming along a more circuitous path would have a straightness index <1.

We used generalised linear models (GLMs) to model travelling speed and straightness index during the overall homing trip as functions of displacement distance, sea surface temperature (SST) at release, season, latitude, and species. We also checked correlations between travelling speed and straightness. We obtained SST as daily estimates derived from satellite data at a resolution of 0.1° (NASA Earth Observations 2014). The Australian seasons were defined as: spring September–November, summer December–February, autumn March–May, and winter June–August (Bureau of Meteorology 2015). Potential effects of bathymetry were not considered because estimated water depths were consistently shallow (mean ± SD = 7.3 ± 4.0 m, n = 1046) and the bathymetry models (resolution 100–110 m) would not identify small features in the complex substrate at our study sites.

We also evaluated travelling speed and straightness index during diurnal and nocturnal periods, using track segments between the first and last fixes of each day and night. To differentiate day and night periods, we estimated time of sunrise and sunset at each location using the R package StreamMetabolism (Sefick 2015). We again examined factors affecting travelling speed and straightness index in generalised linear mixed effects models (GLMMs) and included day/night as a fixed effect, together with other significant effects identified in the preceding analyses. Individual turtles were treated as random effects because some turtles required multiple day/night periods for their journey.

We chose distributions for response variables in the GLMs and GLMMs as follows: travelling speed (continuous, positive, and skewed to right) was fitted with the gamma distribution, and straightness index (proportion) was fitted with the beta distribution. We used the R package stats to fit gamma GLMs (R Core Team 2015), package betareg to fit beta GLMs (Cribari-Neto and Zeileis 2010), and package glmmADMB to fit both gamma and beta GLMMs (Fournier et al. 2011; Skaug et al. 2015). For each model, we computed the variance inflation factors (VIF) among the covariates using the R package car (Fox and Weisberg 2011). We considered collinearity was not an issue if the values were <3 (Zuur et al. 2010). Homogeneity of variance was assessed by plotting residuals versus fitted values. Transformations were applied to data when necessary to meet assumptions of the models. Response variables were centred to have a mean of zero for analyses with GLMs and GLMMs (Becker et al. 1988). We used the R package MuMIn (Barton 2015) to rank all possible models by second-order Akaike information criterion (AICc). We selected a set of models within two AICc units of the best-ranked model to identify models with similar explanatory power (Burnham and Anderson 2002). AICc model weights (ω _i ) were computed as the weights of evidence in favour of each model i within the “best subset”. We compared each model in the “best subset” to a null model by likelihood ratio test using the R package lmtest (Zeileis and Hothorn 2002).

We originally wanted to examine the relationship with speed and straightness of travel for all variables of interest (species, displacement distance, season, SST, latitude, and day/night period). However, it emerged that the relevant portion of our data set (i.e. displaced turtles that returned home and had tracks not restricted by very close proximity to land) suffered from collinearity and was highly unbalanced with respect with season, species, and latitude; for example, none of these turtles were released during the summer months; season was highly correlated with SST (VIF > 3); green turtle tracks began at six different sites spread over a wide latitudinal range, but all loggerhead tracks began at one site. We were therefore obliged to analyse different combinations of variables for separate subsets of the homing turtles. See “Homing behaviour of displaced turtles” section for details of turtle subsets and the variables addressed for each subset.

Results

Outcomes for displaced turtles

Fifty-nine turtles were classified as displaced. They comprised 44 green turtles (including two rescued), 13 loggerhead turtles (one rescued), and two olive ridley turtles (both rescued). These displaced turtles had been retained for <1 to 514 days (median = 1 day, interquartile range 0.9–1.5 days), and they had been displaced from their capture locations by 6.6–432.1 km (median = 17.5 km, interquartile range 13.3–21.3 km) (Table S1).

Most displaced turtles (n = 52 or 88 %), including two rescued turtles, returned home and resettled in their original areas (e.g. Figs. 2b, 3a, b). Another four displaced turtles moved towards their respective capture areas and settled within 1.8–14.1 km of their capture location, but their resettlement areas (95 % UD) did not include the capture location: these comprised two green turtles (one research, one rescued), an olive ridley turtle (rescued), and a loggerhead turtle (rescued) (e.g. Fig. 3c).

Fig. 3

Representative tracks of turtles after displacement. Square capture location, triangle release location, grey line travelling path, grey filled polygon resettlement area. a, b At site 1, these green turtles were displaced and returned to their areas of capture. c At site 2, this olive ridley turtle was found debilitated and displaced after rehabilitation. The turtle moved towards its capture area but its resettlement area did not include the capture location. d At site 5, this green turtle was displaced and resettled away from its capture location

Two displaced turtles (research) travelled towards their capture locations, but satellite transmission ceased before these turtles reached their area of capture. For one individual (K89296 green turtle, displaced by 19.2 km, Fig. S1j red), the transmissions abruptly ceased on the 31st day, at which time the turtle had reached a point 10.7 km from the capture location. The other individual (QA12903 loggerhead turtle, displaced by 432.1 km, Fig. S1h) had moved 53.6 km towards its capture location when transmission ceased on the 54th day. Detail of this turtle’s track showed that during the first 20 days after release it moved 44.5 km towards its capture location. For the next 5 days, its movements were localised along the coast. For the last 29 days, most locations indicated a nearby beach. There was a notable change in the PTT data quality: during the first 20 days of travel, only 9 % of the data were high-quality fixes (LC 3, 2, 1) whereas during the last 34 days, 86 % of the data were high-quality fixes.

Only one displaced turtle (research) did not move towards its capture location during its tracking period of 120 days. Instead this turtle (QA45689 green turtle, displaced by 7.8 km) settled in an area 35.1 km from its capture location (Fig. 3d).

Outcomes for non-displaced turtles

Fifty-four turtles were classified as non-displaced. They comprised 35 green turtles, 18 loggerhead turtles and one hawksbill turtle (one rescued green turtle, all others research). The non-displaced turtles had been held for <1 to 170 days (median = 1 day, interquartile range 0.9–1.9 days) and released at locations <0.1 to 8.9 km (median = 4.2 km, interquartile range 1.9–6.1 km) away from their capture locations. After release, all non-displaced turtles remained in their original areas (95 % UD) (e.g. Fig. 2a).

Homing behaviour of displaced turtles

Of the displaced turtles that returned home, 29 qualified for detailed analyses because they were tracked with Argos-linked FGPS devices and their homing tracks were unrestricted by very close proximity to land (e.g. Figs. 2b, 3b). Some of these turtles took a very direct route while others swam along a relatively circuitous path. Table 2 contains summary statistics for variables of interest associated with the homing track of each turtle. The effects of these variables were addressed separately for different subsets of the homing turtles (see “Detailed analyses for homing turtles” section). The effects of latitude were examined only for green turtles (homing turtle subset 1). The effects of species were examined only for turtles associated with site 1 (homing turtle subset 2). Relocation distance, SST, and travelling speed/straightness index were included as possible explanatory variables in both cases. We omitted season as an explanatory variable in all models because of its strong correlation with SST. This means that SST may act as a surrogate variable for other environmental attributes which change seasonally.

Table 2 Summary statistics for 29 turtles that were included in detailed analyses of homing behaviour

The first homing turtle subset comprised 22 green turtles, for which we tested the effects of latitude and other relevant variables using the following two global models.

• 1. Global model: Travelling speed ~ Displacement distance + SST + Straightness index + Latitude

Latitude did not appear in the best-ranked model and the model selection process resulted in only one model being included in the “best subset”. This model had SST as its only predictor (Table 3). Neither latitude, displacement distance, nor straightness index provided any improvement in prediction of travelling speed.

Table 3 Detailed analyses for homing turtle tracks: overall movements

• 2. Global model: Straightness index ~ Displacement distance + SST + Travelling speed + Latitude

Four models were included in the “best subset” of models but the “best subset” included the null model, that is, a simple estimate of the mean straightness index with no explanatory variable as predictor (Table 3). This result, together with likelihood ratio tests, indicates that none of the variables including latitude had any perceptible influence on the straightness index.

The second homing turtle subset comprised green turtles (n = 12) and loggerhead turtles (n = 7) that were released in the same area (site 1). For this subset, we tested the effects of species and other relevant variables as expressed in the following third and fourth global models.

• 3. Global model: Travelling speed ~ Displacement distance + SST + Straightness index + Species

Two models were included in the “best subset” of models (Table 3). The best-ranked model used both SST and species as predictors of travelling speed, and the second-best model included only SST. Neither displacement distance nor the straightness index appeared to affect travelling speed (Table 3).

• 4. Global model: Straightness index ~ Displacement distance + SST + Travelling speed + Species

Only one model was included in the “best subset”: the model included SST as a solo predictor (Table 3). Neither species, displacement distance nor travelling speed appeared to influence the straightness index (Table 3).

• 5. We re-analysed the data using all qualified homing turtles (subset 1 + subset 2, n = 29), omitting latitude as a covariate because our results for the first homing data subset indicated latitude had no effect on travelling speed or straightness. We used the same global models (3 and 4 above) that we had applied to our second homing data subset. The inclusion of additional green turtles from different sites did not change the results of the model selection with the second homing subset (Table 3). That is, cooler SST values were in general associated with faster travelling speed (Fig. 4a) and with straighter (less circuitous) routes (Fig. 4b). The result also indicated that green turtles tended to travel faster than loggerhead turtles (Fig. 4a).

  Fig. 4

  

  a Travelling speed and b straightness index for tracks of homing turtles that were released at various sea surface temperatures (SST). Green turtles = filled circle, loggerhead turtles = empty circle. The solid line is model fit; grey band denotes 95 % confidence interval

Day/night movement

Among the 29 homing turtles analysed in detail, there were large variations in travelling speed and straightness index by day and by night (Table 2). Day/night effects on homing behaviour were tested with SST and species as explanatory variables. Our selection of these two variables was determined by results of preceding analyses of overall movements. We used data for all qualified homing turtles (subset 1 + subset 2, n = 29) in the following two global models.

• 6. Global model: Day/night travelling speed ~ SST + Day/night + Species

Two models were selected in the “best subset” (Table 4). Day/night was an important variable since it occurred in both models. Turtles tended to travel faster during the day than the night (Fig. 5). SST also occurred in both models as expected. Species was retained in the best model which had considerable support relative to the other model: the AICc model weights were more than double when species was included (Table 4). Day/night travelling speed decreased approximately 0.06 km h⁻¹ per 1 °C increase, and in general, green turtles travelled faster (fit = 0.85 km h⁻¹) than loggerhead turtles (fit = 0.60 km h⁻¹).

Table 4 Detailed analyses for homing turtle tracks: day/night movements

Fig. 5

Estimated travelling speed of homing turtles by day and by night. Filled circle is model fit; error bars denote 95 % confidence interval

• 7. Global model: Day/night straightness index ~ SST + Day/night + Species

None of the variables was associated with straightness of day/night segments of homing tracks: the best-ranked model was the null model (Table 4).

Discussion

This study presented substantial evidence that highly mobile marine species like Cheloniidae cannot be expected to remain at new human-selected locations after the animals have been intentionally displaced from their original coastal foraging grounds.

Confirmation of homing behaviour

The results provided strong support for our initial hypothesis: most displaced turtles attempted to return home, and furthermore, most of them succeeded. For our study turtles, homing ability was not limited by distance of displacement (up to 117.4 km) or by captivity duration (up to 514 days). The successful homing animals included green turtle adults and juveniles of both sexes and loggerhead turtle adults of both sexes. In addition, one olive ridley turtle returned home and the other resettled near its capture area. The single hawksbill turtle was not displaced.

A few turtles did not return home according to strict study criteria, but did not conclusively fail to return. Most of the non-returned turtles travelled to areas near their respective capture areas. The single displaced turtle that adopted a resettlement area far (35.1 km) from its capture location appears to indicate a rare instance of failure to return home. However, we noted that this turtle was tracked for 120 days, a period shorter than the median tracking duration (157 days) and there remains a highly speculative possibility that the turtle could have completed a homing journey after the cessation of tracking.

Our results showed no evidence of impaired homing capacity for rescued turtles that had spent 69–514 days in rehabilitation centres. Our one apparent failure to home was not a rescued turtle. Of the five rescued turtles that were displaced, two returned home and the other three resettled near home. It was plausible that the near-home rescued turtles had actually returned to their true original areas. A rescued turtle may have drifted beyond its home area while it was in a debilitated state, in which case its capture location (where it was found and rescued) would have been outside its true original area.

Potential fitness benefits and costs

Almost all the displaced turtles showed a strong homing tendency, and all non-displaced turtles remained in their original areas after release. This finding was consistent with long-term site fidelity, a widely reported phenomenon in groups as diverse as Chiropterans (Lewis 1995) and Elasmobranchs (Knip et al. 2012) albeit with intra-taxon variation. The development and persistence of site fidelity would imply this behaviour is associated with a fitness benefit in terms of evolutionary adaptation (Parker and Smith 1990).

Details of the potential fitness benefit accruing to Cheloniidae through their fidelity to foraging areas have not been determined experimentally. The benefit might be explained in broad terms by site familiarity. This intuitively relevant concept has seldom been included in habitat selection models and remains difficult to measure (Piper 2011). We surmise that, through long familiarity with a particular area, sea turtles would discover where to find food efficiently, where to find shelter for resting, where predators typically occur, and where they can best be evaded. Such site familiarity could enable individuals to adjust their foraging behaviour to balance food acquisition and predation risk, as has been observed in sea turtles in Western Australia (Heithaus et al. 2008). Thus, we infer that each turtle derives a fitness benefit by remaining faithful to its home foraging area and conversely, we infer fitness costs will accrue for a displaced turtle. It must necessarily expend energy in travelling back to its home area after unnatural displacement, and it may face greater risk and forage less efficiently while it is in unfamiliar habitat.

Factors influencing homing travel

Sea surface temperature (SST) was the key factor identified as influencing homing behaviour: in cooler water, the study turtles travelled faster and followed straighter routes. Greater speed in cooler water was an unexpected finding for Cheloniidae. They are ectothermic animals that are affected by ambient water temperature (Spotila et al. 1997). Cooler water has been found to slow the metabolic rate of green turtles (Southwood et al. 2003, 2006) and reduce their activity. For example, green turtles within the southern part of our study area were found to make notably longer resting dives at cooler temperatures than at warmer temperatures (Hazel et al. 2009). Similarly, slower travel could be expected at cooler temperatures, yet our results indicated the converse. In the scientific literature, we could find no plausible explanatory principle. Insight regarding this surprising finding might be gained through future research involving systematic displacement experiments.

Although the straightness of complete homing tracks was strongly associated with SST, the same association was not evident when we evaluated day/night effects. This may reflect imprecise estimates of straightness index for our day/night track segments. These segments were short, and thus, each segment contained relatively few locations (median 4–8 locations used for straightness index of a day/night segment, Table 2).

Inter-specific differences in travelling speed of the homing turtles probably reflect differences in swimming ability. Green turtles generally swim faster than loggerhead turtles (Heithaus et al. 2002), and our results are consistent with that observation. In contrast to travelling speed, straightness indices were similar for green turtles and loggerhead turtles. The similarity in straightness of tracks could suggest both species have similar way-finding ability in coastal waters.

Way-finding ability of homing turtles

The present study was not designed to investigate navigational capacity per se, but our results clearly confirmed the ability of displaced turtles to find the way back to their original areas. For sea turtles, the underlying mechanisms for open ocean navigation are understood to involve predominantly geomagnetic cues at greater distances from the destination, potentially progressing to a hierarchy of other cues at closer range, details of which remain to be elucidated (Åkesson et al. 2003; Avens and Lohmann 2003; Hays et al. 2003a; Benhamou et al. 2011; Lohmann et al. 2013). It seems plausible that a similar hierarchy of cues guided our study turtles, although they did not undertake oceanic travel and generally travelled within a few kilometres of the mainland shore.

Our finding that displaced turtles travelled faster during the day might imply greater availability of way-finding cues during daylight and hence might suggest that visual information could be important for way-finding. This difference is not necessarily related to way-finding; for example, turtles that are not travelling also appear to be more active during the day, as reported for foraging turtles within our study area (Hazel et al. 2009). Furthermore, the findings of Åkesson et al. (2003) suggest that sea turtles do not use celestial cues for orientation. Nevertheless, additional insight might be gained if future studies were to include day/night information when analysing way-finding and navigational behaviour of sea turtles.

Premature disruption of tracking

Transmission from a tracking device may cease for diverse reasons (Hays et al. 2007), and we speculated about the cause of two transmission failures during homing travel. For turtle QA12903, the sudden and concurrent changes in movement pattern and in quality of PTT fixes suggested the turtle became debilitated or died in the area where movement became localised. We suspect this turtle probably became stranded on the shore, given the unusually large proportion of high-quality fixes acquired around the intertidal area during the last period of transmission. The tracking period was relatively short for this turtle (54 days), and there was no apparent sign of degradation in device performance prior to cessation. We were unable to confirm turtle death or investigate possible causes because the site of suspected stranding was inaccessible.

For turtle K89296, signals stopped abruptly after only 31 days, while the turtle was travelling slowly close to shore. There was no evidence of a change in turtle behaviour. Detachment of the tracking device seemed more likely than an early technical failure. Perhaps, the adhesive bond had been gradually weakened by the turtle rubbing its carapace on rocky outcrops that were potentially available en route. A similar explanation might apply for the two non-displaced turtles that had similarly short tracking durations (≤31 days) and no apparent change in behaviour. Rare events like boat strike or attack by a very large predator could disrupt tracking, but we remain cautious about over-interpreting the cessation of tracking. In our study, the tracking data offered persuasive evidence for morbidity or mortality in only one case, turtle QA12903 described above. The wide temporal and geographic range of our study precluded using this single case to derive a quantitative estimate of mortality, as has been done in different circumstances (Hays et al. 2003b).

Conservation implications

Our findings suggest that displacement and periods in captivity do not disrupt a turtle’s ability to find its way back to its original foraging area nor diminish its fidelity to that area. However, there must be an energetic cost for homing turtles and there might be heightened risk of harm in unfamiliar coastal waters. The potential fitness costs of displacement should not be ignored, despite our strong evidence that the majority of displaced turtles can be expected to return home.

Confirmed homing ability suggests that moving individual turtles away from danger could be effective only as a short-term conservation measure, e.g. rescue from temporary threats such as oil spills. The relocation of turtles from their established coastal foraging ground to a new area cannot be expected to succeed as a long-term conservation strategy. Priority must rather be placed on protecting their original habitat.