Multiple satellite tracking datasets inform green turtle conservation at a regional scale

Satellite tracking studies of marine megafauna have grown over the last few decades. The number of these individual datasets are now at levels which if combined, can infer on population level movement and spatial use. Here we use this approach to quantify distribution and important areas for one of the largest populations of green turtles (Chelonia mydas) in the Indo‐Pacific.


| INTRODUC TI ON
Populations of marine megafauna around the world are declining due to a range of threats (Bowler et al., 2020;Halpern et al., 2008;Maxwell et al., 2013). A key first step to impact assessment is to have accurate species distribution maps and knowledge of their important areas of use. For many marine migratory species, telemetry is one of the main ways to obtain this spatial data. However, in order to map population or stock level distribution, large numbers of individuals tracked from representative sub-populations are needed over relevant time scales, which individual tracking projects rarely achieve.
The green turtle (Chelonia mydas) is a highly migratory species of marine turtle with a circum-tropical distribution, listed by the International Union for Conservation of Nature (IUCN) as endangered (Seminoff, 2004) and listed as vulnerable in Australia by the Environment Protection and Biodiversity Conservation Act 1999. In the Indo-Pacific region, some of the largest rookeries of green turtles are found in Australia (Limpus, 2008;Prince, 1993) but there is only fragmented data published on the spatial distribution of some of these stocks (see Fitzpatrick et al., 2012;Limpus, 2008;Shimada et al., 2019). This is especially the case for the Western Australian stocks (but see Waayers et al., 2011;Whiting et al., 2008) where many large rookeries are located and where there is expanding coastal and offshore activities related to hydrocarbon and mineral resources exploitation. This includes large amounts of vessel traffic, especially around bulk export port facilities, dredging for shipping channels and the presence of artificial light at night (Kamrowski et al., 2014). Recognition of the potential impact of these activities on threatened marine turtles has led to a number of satellite tagging projects over the last decade, some of them mandated through the environmental approval process of new projects Waayers et al., 2019). However, most of this information remains unpublished.
In Western Australian waters, three genetically distinct green turtle stocks occur: the Northwest Shelf stock (extending from Ningaloo Reef in the Pilbara to the Maret Islands in the Kimberley), Scott-Browse (includes Scott Reef on the edge of the continental shelf of Western Australia and Browse Island on the mid shelf) and the isolated Ashmore Reef (edge of the continental shelf) (Dethmers et al., 2006;FitzSimmons & Limpus, 2014;Waayers, 2014). Stocks are genetically isolated on the rookeries but individuals from multiple genetic stocks overlap at foraging grounds Jensen et al., 2016). Although most of the major rookeries have been identified (Commonwealth of Australia, 2017;Limpus, 2008;Pendoley et al., 2016), the spatial extent of the area used by female turtles while at sea during the nesting season (i.e. inter-nesting) has not yet been quantified for the majority of rookeries, neither has the area used post-nesting; for migration and foraging. This uncertainty hampers the designation and protection of these important areas, critical for their survival, and needed for their conservation and management. In the absence of this information, the Australian Government, in the Recovery Plan for Marine Turtles in Australia (Commonwealth of Australia, 2017) has designated inter-nesting habitat within a 20 km radius around representative and important turtle rookeries as habitat critical for the survival of green turtles (as an inter-nesting buffer). It has also defined so called Biologically Important Areas (BIA) (https://envir onment.gov.au/marin e/marin e-speci es/bias) (https://www.envir onment.gov.au/webgi s-frame work/apps/ncva/ncva.jsf) which are often based on limited information and reliance on expert elicitation where peer-reviewed published literature are unavailable. Similar in concept to Ecological or Biologically Significant Areas (EBSAs), this approach of identifying the areas used for breeding, foraging, and migration is a useful tool for conservation managers to inform marine spatial planning and assist decision-making (Corrigan et al., 2014;Dunn et al., 2014).
Although a qualitative approach is valid in the absence of quantitative data, in many cases it may be considered inadequate for basing decisions on. This is especially pertinent given that robust scientific data from satellite tag deployments is available and could be combined and analysed to support decision-making.
Here we assess all the satellite tracking data available for nesting female green turtles in Western Australia (see Tucker et al., 2020;Waayers et al., 2019) and use it to identify rookeries where transmitters have not been deployed. We use a combination of new satellite transmitter deployments (n = 20) and existing datasets we were able to access (n = 76) to quantify the inter-nesting and post-nesting movements and distribution of green turtles in Western Australia, comprising 10 rookeries and two of the three genetic stocks. We hypothesize that migratory routes and foraging areas will differ between the two stocks due to the disparate nesting locations. In addition, we assess the effect of grid cell size on the results and the representativeness of the distribution and important areas we calculate, based on the available sample sizes at both the rookery and stock scale.

| Data collection and compilation
Satellite tracking data from 76 adult female green turtles comprising island (nearshore and offshore) and mainland rookeries (n = 10) covering all Western Australian stocks except Ashmore Reef (Spring & Pike, 1998) were identified from Waayers et al. (2019) and compiled ( Figure 1). One of the datasets is from a peer-reviewed paper (Waayers et al., 2011)  Did not include turtles with post-nesting local residence in the vicinity of the nesting grounds (e.g. turtles that did not migrate and used the same or similar area post-nesting as they did during inter-nesting).
b Migratory movements were recorded for only two turtles at Rosemary Is, with remaining turtles (4) displaying local residency, thus the values displayed are the median for each turtle.

| Turtle movement model and metrics
A Bayesian switching state-space model (SSM) (Jonsen et al., 2003(Jonsen et al., , 2005 was applied to the raw ARGOS location data to account for location error and estimate behavioural states for each individual track using the R (R Core Team, 2019) package bsam (Jonsen et al., 2017).
The SSMs were fitted using a switching first-difference hierarchical correlated random walk model (hDCRWS) with a total of 100,000 Markov Chain Monte Carlo (MCMC) iterations after an initial burn in of 10,000 samples with every 10th iteration kept after burn in.
For tracks that did not converge, the model was re-run with an additional 10,000 iterations until it converged. We used a 6-hr time step for the model to generate 4 location estimates per day. Tracks with large gaps (>7 days) were split and analysed with a hierarchical model that analysed each portion of data separately. One of two behavioural modes (b) was assigned to each location estimate; either transient (locations with a SSM behaviour value b < 1.5) or resident (locations with a SSM behaviour value b > 1.5) (Jonsen et al., 2005). (transit between foraging areas, where more than one foraging area occurred -which was rare). Some individuals (n = 19) did not migrate and used the same/similar area post-nesting as they did during internesting for the duration of the deployment and therefore all the resulting SSM behaviour values were b > 1.5 (resident). In order to split the tracking data into inter-nesting and foraging for these individuals, we used the median date of the end of inter-nesting behaviour for all other turtles in the same nesting area as a cut-off. For transmitters that also provided FastlocGPS data (

| Time-weighting
Time-weighting was applied to the SSM tracks based on a modified version of the method described by Block et al. (2011). This was done to account for bias due to differences in deployment duration and the associated decrease in the number of individuals tracked with time due to premature detachment, or tag failure not uncommon with satellite tracking datasets (Hays et al., 2007).

| Quantification of spatial distribution and important areas
The Scott-Browse stock (Table 1). We also combined NWS stock-Kimberley and Scott-Browse stocks for the analysis of post-nesting movements as both these stocks use the Kimberley region predominantly ( Figure 1).
In order to identify spatial distribution and important areas (i.e. areas of high use of importance for life-history stages, such as breeding, migration and foraging and similar to the concept w t = 1∕n t defined for EBSAs and BIAs), we gridded the study area and calculated the time spent in each grid cell using the time-weighted tracks for each turtle for the entire track duration and for inter-nesting, foraging and migration behaviours separately using the R package trip (Sumner & Luque, 2015). Time spent was calculated using a 3 × 3 km square grid for inter-nesting and a 20 × 20 km square grid for migration and foraging. These grid cell sizes were selected based on the scale of the movements; very small (100 s of m to km) for inter-nesting and very large (100-1,000 s of km) for post-nesting and to be in line with grid cell sizes used for other turtle species distributions quantified in this region .
We To represent spatial use for inter-nesting, migration and foraging, we ranked the summed grid cell values (both occupancy index and number of turtles) from largest to smallest and determined the spatial distribution as the cells encompassing the top 95% (for inter-nesting) and 75% (for migration, foraging and the entire track) of the cumulative frequency distribution based on the method described by Soanes et al. (2013) and akin to 75% and 95% utilization distributions. The 75% distribution was used instead of the 95% for foraging, migration and the entire track to exclude rare large migrations with low/no individual overlap.
In order to understand which rookeries and stocks were mixed on the foraging grounds we plotted the 75% foraging distribution and named all the areas on the coast where a discrete 75% distribution polygon occurred (e.g. Shark Bay, Exmouth, etc) or applied a name that was indicative of the broad area of use (i.e. Barrow group).
Where the foraging distribution was continuous along the coast, we identified general geographical boundaries (using similar methods to the above) to assist with spatial planning and management of foraging green turtles. Each turtle's individual 75% distribution was then overlaid on this map and we summed the number of turtles having their individual 75% distribution within each of the bounded areas. In order to understand the effect of grid size on this process, we repeated the process using a range of other grid sizes (5, 10, 50, 100 km).
We also calculated the proportion of the spatial distributions for each behaviour inside marine protected areas (MPAs), including

| Effect of sample size on calculated spatial distributions
We assessed the effect of sample size on the spatial distribution extent for each behaviour using the 95% distribution using the R package SDLfilter (Shimada, 2019). Please see Shimada et al. (2020) for details but to summarize, we took the cell values of each turtle raster layer (i.e. relative proportion of time spent per cell) and for increasing sample size from 2 to the maximum number of individuals, we merged all areas identified by existing data (e.g. n-1 individuals) and calculated the proportion of time spent by an additional individual (n) over the collective area. We then calculated mean proportion of overlap for each sample size (2 to the maximum) from 1,000 random permutations. We considered an asymptote was attained once the mean proportion of overlap reached 0.95. After this point, the sample size was deemed sufficient to describe the general distribution of the population because on average a new additional individual will only spend 5% of time outside the identified area.

| Effect of sample size on calculated spatial distributions
The plots of the cumulative proportion of overlap for the inter-

| Movement behaviour
Satellite tag deployments (all data combined) lasted between 19.5 and 537.5 days (median = 131.6 days) with a mean F I G U R E 2 Effect of sample size (number of turtles tracked) on the estimation of the extent of the spatial distributions (calculated as proportion of overlap of individual distributions, i.e. overlap reaching 1.0 (100%) indicates the addition of more individuals to the sample will not increase the size of the distribution) of green turtles during the inter-nesting period at each rookery ( After accounting for the short tracks (n = 6, ~30 day duration) and non-migratory turtles (13) Figure 1a; Figure S3). After accounting for the short tracks (6)

| Inter-nesting
Higher values of occupancy index and high overlap of turtles per grid cell occurred adjacent to nesting beaches for all rookeries (Figure 3; Figure S2). The area of use (95% distribution) for each site varied between 52.4 and 618.2 km 2 when calculated with the occupancy index, and between 78.5 and 1,890.3 km 2 when calculated with the number of turtles (Table 2). Using the depth of the water column obtained from bathymetry data underlying the inter-nesting locations from SSM tracks, inter-nesting turtles occupied mostly shallow waters (median depth = 9 m, range = 4-62 m) (Table 2) Figure S5), highlighting these regions as potential areas for protection.

| Foraging
Foraging turtles were largely concentrated in shallow waters (median of 9 m, ranging from 1 to 104.5 m) although the two turtles with foraging behaviour at Rowley Shoals located near the 300 m bathymetric contour used deeper water (82.6 ± 18.4 m) (Figures 1 and 5c,f). The considerably larger when using number of turtles than using occupancy (69,122 km 2 and 23,798 km 2 , respectively), likely because the occupancy distribution was influenced by the many discrete, small foraging areas with high occupancy that were used by only one or low numbers of individuals (Figure 5a-c,g,h). Whereas the 75% foraging distribution using number of turtles indicated all grid cells used by 1 or more turtles irrespective of the intensity in which areas were used resulting in a larger extent. Given this, and that our distribution is likely to be an underestimate (based on our sample size analysis), we considered the distribution using the number of turtles per grid cell as a more conservative measure of the overall spatial distribution of foraging turtles. The 75% foraging distribution included Shark Bay, a near continuous polygon from Ningaloo to Roebuck Bay, the southern (Buccaneer Archipelago and Adele Island) and northern Kimberley (off the Bougainville Peninsula) and one site in the Northern Territory (Tiwi Islands and Coburg Peninsula). To assist with management, we identified 13 geographical areas ( Figure S6) that made up the 75% distribution where multiple turtles from either one or both stocks cooccurred (Table 3; Figure S7, Table S1). The size of grid used for the calculation of foraging distribution did not have a strong influence on the identification of these foraging areas although it clearly has an influence on the size of the area ( Figure S8). Almost all foraging regions defined with a 20 km grid cell were also identified with increasing and decreasing grid sizes, with the exception of Rowley Shoals that was not identified with a 50 or 100 km grid ( Figure S8).
There was a reasonable match between the spatial extent of the 95% inter-nesting distribution (occupancy index and number of turtles) and the Inter-nesting Habitat Critical Areas (defined as a 20 km buffer around known rookeries) for the Lacepede Islands, Maret Islands and Scott Reef ( Figure 6; Table S2). Although the 95% inter-nesting distribution for the NWS stock-Pilbara rookeries was encompassed by the 20 km buffer, it was much smaller (Figure 6). We identified a large overlap (40.4%) between the 75% migration distribution for the  (Table S2).
The 75% foraging distribution overlapped with existing foraging BIAs recognized by the Australian Government in Western Australia in the Barrow Group, Dampier Archipelago, Port Hedland and Roebuck Bay, but there was only 5% ovelap overall ( Figure 6; Table S2). Some of the designated foraging BIAs were not validated by our analysis (e.g. Joseph Bonaparte Gulf although it encompasses some of the migration distribution from WA to NT) and many of the foraging areas quantified in this study are not formally recognized as BIAs (Figure 6e,f).
Interactive maps of the distributions delimited here can be accessed in the North West Atlas (https://north westa tlas.org/nwa/ nws2s -megaf auna).

F I G U R E 6
Inter-nesting, migration and foraging distribution of green turtles in the Northwest of Australia (a), 95% inter-nesting distribution at rookery scale (b-c), 75% migration distribution at stock/sub-stock scale (d) and 75% foraging distribution (e-f) overlaid with Marine Protected Areas (a-f), the defined inter-nesting Habitat Critical Areas (Commonwealth of Australia, 2017) (b-c), and the Foraging BIAs formally recognized by the Australian Government (in the Conservation Values Atlas, https://www.envir onment.gov.au/topic s/marin e/marin e-biore giona l-plans/ conse rvati on-value s-atlas) (e-f). Note that there are some designated habitat critical inter-nesting areas for which we did not have any tracking data (where inter-nesting critical habitat is indicated on the maps with no corresponding inter-nesting distribution) TA B L E 3 Number of individual turtles (calculated using the 75% foraging distribution of each individual turtle) from each rookery and stock/sub-stock at each region along the coast (see Figure 5).

NWS stock-Kimberley
Outlined rows indicate areas used for foraging by all stocks/sub-stocks

| D ISCUSS I ON
Compiling and analysing a large satellite tracking dataset allowed us to quantify the spatial extent of inter-nesting for a representative portion of adult female green turtles from Western Australian rookeries (n = 10) comprising 2 genetic stocks. We have provided the first spatial data on the areas used post-nesting, identifying two main migratory corridors (Pilbara and Kimberley) to many disparate, primarily neritic foraging areas dispersed throughout inshore waters along the northwestern Australian coastline. Despite some limitation with sample size, our quantification of foraging distribution provides more robust data to support spatial management and conservation of green turtles at local and regional scales than is currently available.
Our results show that the existing delineation of a 20 km radius around green turtle rookeries as an inter-nesting buffer in the Recovery Plan for Marine Turtles in Australia (Commonwealth of Australia, 2017) encompassed the inter-nesting distribution we defined here, supporting its effectiveness in protecting nesting green turtles. For the NWS stock-Pilbara rookeries, the 95% distribution area was, however, considerably smaller than the 20 km buffer.
The migration patterns we found were in accordance with a review of turtle migration strategies by Godley et al. (2008) suggesting there are 4 general strategies, with green turtles showing two of these: Type A1; oceanic and/or coastal movement to neritic foraging grounds and Type A3; local residence. We found a Type A1 movement pattern for the majority of WA turtles (86%) and 14% Type A3 as has been found for green turtles nesting at the Cocos Keeling Islands (Whiting et al., 2008). Even turtles from the Scott-Browse stock nesting on Sandy Islet at Scott Reef (an oceanic atoll at the edge of the Australian continental shelf) showed mostly Type 1 movement, as has also been found for other oceanic islands such as the Galapagos Islands and Ascension Island Hays et al., 2002;Seminoff et al., 2008). These oceanic island nesters often perform directed movement crossing oceanic regions to reach foraging grounds in mainland coastal locations despite apparent availability of foraging habitat at these islands or potentially other locations along the way Shimada et al., 2019).
Indeed, we found only 1-3 individuals that had movement behaviour classified as foraging on route to foraging grounds, with the majority having only one main foraging ground.
However, in contrast to green turtles in Galapagos and Ascension islands that migrate over extensive (thousands of km) oceanic areas Seminoff et al., 2008), the migratory behaviour of green turtles in Western Australia was largely dominated by less extensive movements (804.6 ± 778.2 km) and mostly coastal routes, with many even resident (14%), although some large distances (>2,000 km) and oceanic movements were recorded. This may be related to the fact that the majority of the studied rookeries were on the continental shelf, relatively close to the mainland shore. This predominance of coastal travel was also observed for green turtles in the Mediterranean Sea , Equatorial Guinea (Mettler et al., 2019) and Costa Rica (Blanco et al., 2012) with the distribution of foraging habitat availability and quality (Christiansen et al., 2017) a suggested reason for this migration strategy. Regional oceanographic setting and ocean currents also play a significant role in hatchling dispersal and are thought to largely determine adult foraging grounds (Gaspar et al., 2012;Hays et al., 2010;Scott et al., 2014). The dispersal of simulated particles from different locations along the coast of Western Australia showed that release locations on wide continental shelves, as is present on most of the Northwest Shelf, resulted in particles being retained on the shelf (Robson et al., 2017).
We broadly identified two migratory corridors, one used by the NWS stock-Pilbara and another used by the NWS stock-Kimberley and the Scott-Browse stock with some overlap at the northern and southern extents respectively (Figures 4 and 6). We also identified areas within each corridor that might be acting as Although many of the foraging areas that we define have been previously documented in the literature Heithaus et al., 2008;Pendoley, 2005;Preen et al., 1997;Prince et al., 2012;Richards et al., 2006) (Prince, 1998;DBCA unpublished data) and genetic mixed stock analysis shows that Northwest Shelf turtles are also found in the Gulf of Carpentaria and Indonesia (Dethmers, 2010;Dethmers et al., 2010;Jensen, 2010;Joseph et al., 2014  . Deploying transmitters on turtles on the foraging grounds, including males and juveniles, combined with further genetic analysis may also be useful for increasing our ability to determine the total foraging distribution (reach the asymptote in the cumulative overlap curve) and to identify important mixed or single stock foraging areas. In addition, species distribution models and habitat suitability models may also be useful for documenting suitable foraging habitat for green turtles (e.g. Hawkes et al., 2007;Pikesley et al., 2013), and thus potentially being able to account for the uncertainty (lack of an asymptote in our cumulative probability curve) in the extent of the foraging distribution presented here.
Despite these limitations, the broad geographical distribution we defined is still extremely valuable, as it is the first quantitative analysis of green turtle distribution at the stock scale and is a marked improvement on the previous data deficient methods. Importantly, our analysis of sample size is rarely undertaken and provides additional context with which to interpret the results. au/marin e/parks/) ( Figure 6). Unsurprisingly perhaps, given that the BIAs were not based on quantitative analysis, our results show that the foraging BIAs are largely underestimating and missing important foraging areas for green turtles in Western Australia (only 5% overlap). Although the inter-nesting buffer (20 km radius around rookeries) did overlap with some of the foraging areas, especially on the southern Pilbara coast, these protections are likely to be focussed on the nesting season rather than year round.
Our study highlights the value of compiling and analysing multiple tracking datasets for the delineation of distribution and BIA/ EBSAs for marine turtles to assist with assessing and mitigating impacts of industrial development and other pressures in the region. Our results have shown that the foraging distribution of green turtles from two stocks in Western Australia expands throughout northwest and northern Australian coastal waters, including the Northern Territory and Queensland. Some turtles also crossed into international EEZs of Indonesia (n = 2) and Papua New Guinea (n = 1) potentially exposing them to additional threats from illegal harvest (Joseph et al., 2014;Pilcher et al., 2008). This suggests that a future analysis of satellite tracking data would benefit from the incorporation of data from those areas, with tracking datasets for this species in existence more broadly in the Indian Ocean and Southeast Asia (Hays et al., 2018). The inclusion of these datasets would allow for a more complete picture with the potential to offer further insights into the movement ecology of this threatened and migratory species, as has been shown for other similar collaborative studies (e.g Fossette et al., 2014;Hazen et al., 2017;Hindell et al., 2016Hindell et al., , 2020.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13197.

DATA AVA I L A B I L I T Y S TAT E M E N T
Existing tracking data compiled in this study can be made available upon request with consent from data owners. Data from new satellite transmitter deployments included in this study (n = 20) have been made available at ZoaTrack (https://zoatr ack.org/proje cts/534) and the Australian Institute of Marine Science Data Centre at: https://www.aims.gov.au/nw-shoal s-to-shore/ threa tened -speci es-of-the-north -west.

B I OS K E TCH
The

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found online in the Supporting Information section.