The ecological connectivity of whale shark aggregations in the Indian Ocean: a photo-identification approach

Genetic and modelling studies suggest that seasonal aggregations of whale sharks (Rhincodon typus) at coastal sites in the tropics may be linked by migration. Here, we used photo-identification (photo-ID) data collected by both citizen scientists and researchers to assess the connectedness of five whale shark aggregation sites across the entire Indian Ocean at timescales of up to a decade. We used the semi-automated program I3S (Individual Interactive Identification System) to compare photographs of the unique natural marking patterns of individual whale sharks collected from aggregations at Mozambique, the Seychelles, the Maldives, Christmas Island (Australia) and Ningaloo Reef (Australia). From a total of 6519 photos, we found no evidence of connectivity of whale shark aggregations at ocean-basin scales within the time frame of the study and evidence for only limited connectivity at regional (hundreds to thousands of kilometres) scales. A male whale shark photographed in January 2010 at Mozambique was resighted eight months later in the Seychelles and was the only one of 1724 individuals in the database to be photographed at more than one site. On average, 35% of individuals were resighted at the same site in more than one year. A Monte Carlo simulation study showed that the power of this photo-ID approach to document patterns of emigration and immigration was strongly dependent on both the number of individuals identified in aggregations and the size of resident populations.


SA, 0000-0002-9929-7312
Genetic and modelling studies suggest that seasonal aggregations of whale sharks (Rhincodon typus) at coastal sites in the tropics may be linked by migration. Here, we used photo-identification (photo-ID) data collected by both citizen scientists and researchers to assess the connectedness of five whale shark aggregation sites across the entire Indian Ocean at timescales of up to a decade. We used the semi-automated program I 3 S (Individual Interactive Identification System) to compare photographs of the unique natural marking patterns of individual whale sharks collected from aggregations at Mozambique, the Seychelles, the Maldives, Christmas Island (Australia) and Ningaloo Reef (Australia). From a total of 6519 photos, we found no evidence of connectivity of whale shark aggregations at ocean-basin scales within the time frame of the study and evidence for only limited connectivity at regional (hundreds to thousands of kilometres) scales. A male whale shark photographed in January 2010 at Mozambique was resighted eight months later in the Seychelles and was the only one of 1724 individuals in the database to be photographed at more than one site. On average, 35% of individuals were resighted at the same site in more than one year. A Monte Carlo simulation study showed that the power of this photo-ID approach to document patterns aggregation sites for several months or years. These researchers did not speculate on the proportion of sharks within a site that might make such large migrations and the frequency with which such journeys might occur. Furthermore, they noted that the average maximum duration of attachment of satellite tags was likely to be too brief to document cross-ocean movements [1].
Here, we examined the ecological connectivity of aggregations of whale sharks across the Indian Ocean on a timescale of up to a decade. We used a photo-identification (photo-ID) approach to achieve this objective, because photographs of the stable white spot and stripe patterns that cover the dorsal surface and flanks of the body enable individual whale sharks to be uniquely identified [27][28][29][30][31]. By searching for matches in large libraries of photo-IDs collected at sites spread across the entire ocean basin we were able to investigate patterns in the spatial scale and frequency of movements among aggregations as well as of residency within aggregations [27][28][29][30][31]. One of the major issues in using our approach was the power of these analyses to detect migration given variation in sampling effort, the timing of sampling and in relative population sizes at different aggregation sites. We used a simulation study based on population sizes, resight rates and likely sampling errors to estimate the ability of our analyses to detect different levels of migration among populations across the Indian Ocean. The implications of these models for our results are discussed.

Study sites and data collection
Photographs used in this study were collected by a number of research organizations, ecotourism operators and tourists from the following key aggregation sites in the Indian Ocean: Ningaloo Reef (Australia) and Christmas Island (eastern Indian Ocean); the Maldives (central Indian Ocean); and the Seychelles and southern Mozambique (western Indian Ocean; figure 1). The years for which photographs were obtained differed among sites and ranged from one (Christmas Island) to 15 years (Ningaloo;  table 1). Details of photo-ID sampling at each locality can be found in the electronic supplementary material, appendix S1.
Photographs were taken by researchers, tourism operators and tourists, who swam alongside individual whale sharks and recorded the natural marks along their flanks with still or video cameras. Ideally, photos were taken on both of the right and left flanks of the shark although this was not always possible as sharks dived or swam away before both photographs could be completed. Where possible, observers recorded the sex of the shark by determining whether claspers on the pelvic fins were present (males) or absent (females) [27]. Sex was classified as indeterminate in some cases due to small size of the shark, poor visibility, or where the shark swam away before pelvic fins could be examined. Images were entered into a database and grouped according to site of capture, sex (male, female or indeterminate), and whether the photo was of the left or right flank of the whale shark.

Image analysis
Spot and stripe patterns present on the flanks of the shark in the area forward of the dorsal fin and behind the last gill slit were used to identify individual sharks (electronic supplementary material, figure S1). To date, no changes in these natural markings have been found to occur over time (up to 12 years) [27,30]. The public domain pattern-recognition software I 3 S (Interactive Individual Identification System) was used to look for matches of ID-photos both within and between sites [32]. Details of use of the image analysis software can be found in the electronic supplementary material, appendix S2. Image analysis was undertaken by one trained observer to minimize the effect multiple users may have had on the selection criteria for matching. Any matches were confirmed by at least one additional observer to reduce the chances of a false positive.

Data analysis
The total number of unique individuals photographed over the sampling period was determined in addition to the number of sightings per year for each aggregation site. The proportion of new individuals to resights in a year for each aggregation was also calculated, with resights defined as individuals recorded by photo-ID in a previous year of sampling. As Christmas Island was sampled for 1 year, photos from this site were only analysed for among-site matches and were not used in any statistical analyses.   A number of metrics were used to compare the sampled aggregations. Permutational analysis of variance (ANOVA) was used in PRIMER 6 statistical package [33] to test for differences between the average numbers of unique individuals sampled per year among sites [34]. In order to balance the design of the analysis, the database was restricted to sharks observed from 1999 to 2012. A Euclidean distance matrix with unrestricted permutation of the raw data and a total of 999 permutations were used to obtain the p-value, which was set at a significance level of 0.05 for all analyses. The same test was then used to identify differences between the average proportions of resights per year for each site from 1999 to 2012. For this analysis, first year of sampling was deleted from the dataset for each site, as there was no possibility of resights in that year. Lastly, permutational ANOVA was used to compare the average number of years individual whale sharks were observed at each site over the entire study period. Where significant results were found, pair-wise tests were conducted. The number of sightings per year as a proportion of the estimated reported population was also examined. Population estimates were all derived from previous studies using Jolly-Seber open population models and included 95% confidence intervals (table 1). As Davies et al. [4] gave estimates derived from each of researcher and public data, the average between these two estimates was used in this comparison.
Chi-square contingency tests [35] were used to compare the number of years individual sharks returned to sites. As the minimum number of years a site was sampled was six (Mozambique), six consecutive years of data collection were used in each site; however, these years were not identical. . Sharks observed returning for 5 and 6 years were pooled into one group in order to meet the assumptions of χ 2 that required more than 80% of cells with expected values greater than five.
As spot and stripe patterns differ between the left and right sides of a whale shark, there was a risk that single individuals could have been double-counted if each side had been photographed separately between sampling periods and/or sites [27,30]. Since the number of images for each flank was relatively similar for the Indian Ocean database, analyses were repeated for all sharks on both the left and right flanks. This meant that all sharks were considered in the analyses and the probability of identifying any migrants was maximized. Where results were of the same significance level, results were reported for the left images as these had a slightly larger sample size (table 1).

Sampling effort
Sampling effort at each site could not be established, because photographs were sourced from both researchers and tourists and for the latter, no effort data were available. Furthermore, techniques varied among sites, with spotter planes used on a daily basis for locating sharks at Ningaloo Reef and the Seychelles during the whale shark season, but not at Christmas Island, Maldives and Mozambique [36][37][38]. The database also contained photographs taken during chance encounters with whale sharks by the public and researchers.

Monte Carlo simulation
We used a simple Monte Carlo simulation approach to examine the sample sizes that would be required to reliably detect migration rates (MR) of varying levels among four aggregations of whale sharks (Christmas Island was excluded due to low sample sizes). The parameter we estimated was the number of unique individuals required to be observed at the sink population in a single year (N sink ) given a range of MR (1-20% of the population, expressed as a proportion of the source population) and a desired detection probability (the desired certainty that a migration of a given rate would be observed) of 80 and 95%, following the equation: where N source was the observed number of unique individuals at the source population per year; P source was the estimated yearly size of the source population; and P sink was the estimated yearly size of the sink population. Because of uncertainty and year-to-year variation in values of N source , P source and P sink , these were entered into the Monte Carlo simulation by sampling a normal distribution with the appropriate mean and standard error based on the empirical data (

Whale shark abundance and sightings
Over the 15 years of sampling, a minimum of 1724 whale sharks were recorded and uniquely identified using photo-ID (table 1)   whale shark was resighted at an aggregation was seven, which occurred at both the Seychelles and Ningaloo sites.
On average, the proportion of resightings to new individuals within a year of sampling did not differ significantly among sites (F 3,29 = 1.82, p = 0.18). All sites followed similar general patterns of a consistent proportion of resights to new individuals after the first few years of sampling (electronic supplementary material, figure S4). The Maldives had the highest proportion of resights relative to new individuals in the later years of sampling.

Monte Carlo simulation
Our simulation showed that we had better chances of detecting migrants (if they were present) to sink populations at Ningaloo Reef and the Maldives than the Seychelles and Mozambique given current numbers of individuals identified each year (figure 6). At Ningaloo, a migration rate of at least 12-13% could be detected from the Maldives, and approximately 5% per year from source populations in the Seychelles and Mozambique. In the Maldives, we had 80% confidence of detecting migrants at immigration rates of at least 2-5% from all three other source populations ( figure 6). In comparison, the simulation shows that immigrants into the Seychelles and Mozambique aggregations would be more difficult to detect with 80% confidence. For these sites, MR as high as 20% per year might not be detected given current rates of identification ( figure 6). On average, an immigration rate of 10% of sharks from Mozambique into the Seychelles site would have an 80% chance of being detected. Notably, immigrants from Ningaloo Reef would be detected at the lowest MR for each sink population when compared with other potential source sites.

Connectivity of Indian Ocean aggregations
No evidence of cross-ocean migration by individual whale sharks was found by our comparison of photo-ID databases from sites spread across the Indian Ocean. However, a regional connection was Seychelles in September of the same year provided evidence of an ecological connection between these two aggregation sites. This is the second observation of movement between these sites; the first was a shark tagged in the Seychelles in November 1996 that was resighted 11 months later in Mozambique [2]. In our study, the shark matched by photo-ID must have swum a minimum of 13.6 km d −1 in order to reach the Seychelles eight months after it was photographed in Mozambique. This value is well within the average swimming speeds recorded by earlier studies of 30 km d −1 [1], implying either that the animal did not take a direct route between these sites or was moving at less than average speeds.
Our evidence for limited dispersal must, however, be considered in the context of our simulation studies that examined the power of the photo-ID technique to detect MR. These simulations showed that we had a relatively high chance of detecting migrants to sink populations in the Maldives and at Ningaloo Reef. In the case of the former, the resident population size was very small (only 80-100 sharks), while in the latter our database of IDs was very large (at least 738 uniquely identified sharks; table 1). Conversely, the database of IDs in both the Seychelles and Mozambique was smaller relative to the likely size of resident populations, with models suggesting that we had relatively low power to detect migration from other sources to these sink populations. Despite this problem, it is notable that we were able to detect a migrant moving between the Seychelles and Mozambique. If our simulation studies are indeed representative, it seems possible that rates of migration between these sites may be relatively high, with at least 10% of individuals per year (or more) moving between locations. Overall, our simulations suggest that given present sample sizes, effort and rates of resighting, the power of the photo-ID approach to detect low (less than 5%) rates of migration among localities is relatively weak, except where resident populations are very small, as is the case in the Maldives. It is also important to note that our simulation generated large confidence intervals around estimates of medians, suggesting that variation in factors such as sampling effort and environmental conditions could alter the potential for resighting.
The lack of evidence for movements at ocean-basin scales in photo-ID records combined with the results of our Monte Carlo simulation suggests that either these are likely to be rare events, or that they are occurring in parts of the population (such as mature animals) that are not present at aggregation sites and are thus inaccessible to the photo-ID approach. So where do these juvenile males go when they depart their seasonal aggregations? Similar to our study, satellite tagging studies have revealed evidence for migrations of whale sharks at regional scales, both in the Indian Ocean and elsewhere. For example, Brunnschweiler et al. [22] tracked a shark off the coast of Mozambique in February 2006 to the southern tip of Madagascar over a period of three months. Both photo-ID and conventional and satellite tags were used by Hueter et al. [20] to record movements of individual whale sharks that formed a summer feeding aggregation off the northeastern Yucatan Peninsula, Mexico into other regions of the Gulf of Mexico, the northwestern Caribbean Sea and the Straits of Florida. In the Galapagos Islands, a whale shark was tracked by a satellite tag to a distance 1650 km to the west before returning to the Islands four months later [19]. At Ningaloo Reef, movements of six whale sharks were tracked using satellite tags for up to seven months and these animals travelled a maximum straight line distance of 1500 km to the northeast [23]. More recently at the Ningaloo Reef aggregation, a combined satellite tracking, photo-ID and public survey approach revealed an extended distribution of sharks along the Western Australian coastline, with sightings in this region observed in all months of the calendar year [39]. Despite these results, identification of migration patterns at annual scales within this region remains incomplete, largely because of the inability of researchers to deploy tags on sharks for more than six to eight months. It is also possible that some juvenile males do not depart Ningaloo Reef and may instead maintain year-round residence, but are simply using deeper habitats in the seasons when they are assumed to be absent. For example, Cagua et al. [40] compared photo-ID and passive acoustic telemetry data of sharks at an aggregation site off the coast of Tanzania. Results from photo-ID displayed a clear seasonal pattern of sightings, with no whale sharks observed between March and September, whereas acoustic data demonstrated a year-round residence of the same animals, with sharks swimming deeper and further away from shore during this time and thus being inaccessible to data collection using photo-ID [40]. It is also possible that such clandestine behaviour may mask movements at larger spatial scales rather than just within aggregation sites.
In contrast to the results of our photo-ID comparison, genetic studies suggest that whale sharks across the Indian Ocean form a single, panmictic population [12]. Our simulation showed that low, but potentially significant (at least in terms of gene flow) rates of long-distance dispersal at ocean-basin scales are unlikely to be detected by the photo-ID technique. The weight of evidence from both tagging and our photo-ID study does suggest, however, that regional movements may be far more common than movement at large scales. If this is the case, panmixis can still occur in populations through a form of genetic 'hopscotch', whereby sharks breed with others in neighbouring regions, and their offspring subsequently move to more distant areas [14]. Alternatively, juvenile males may not be representative of the rest of the population and the long-distance movements of mature sharks may be responsible for the lack of genetic structure of populations across the Indian Ocean. Generally, home range size and reported satellite tracks increase with shark size, with some of the longest whale shark tracks to date coming from mature females [20,41,42]. Only a small number of these long-distance migrations and subsequent breeding would need to occur in order to produce panmixis [16]. If these movements were rare, they might be largely irrelevant to the management of aggregation sites on ecological (as opposed to genetic) timescales.

Future directions and management implications
Photo-ID techniques may be useful for tracking the movement patterns of a large sample size of whale sharks at regional spatial scales, in which case the data obtained from the approach could be improved by more intensive sampling at this scale within the Indian Ocean of eastern Africa. An expansion of photo-ID sampling efforts in the region of Southeast Asia would also greatly benefit our understanding of regional movements. Here, a whale shark photographed in Taiwan in 2012 was resighted in the Philippines in 2013 [43]. A combined satellite telemetry and photo-ID approach might also be valuable, as has shown to be the case for whale sharks that aggregate seasonally off the Yucatan Peninsula [20] and Ningaloo Reef [39] and for other species such as white sharks (Carcharodon carcharias) in South Africa [44], and would likely enhance our understanding of whale shark movements in the Indian Ocean.
Although whale sharks and white sharks occupy very distinct ecological niches, photo-ID methods have been applied for both species and demonstrated similar patterns of site fidelity. This demonstrates the potential applicability of photo-ID approaches for other elasmobranchs that possess body patterns allowing them to be individually identifiable [30]. Standardization of methods and variables recorded when sampling would allow for cross-species comparison of site fidelity and movement patterns.
Regional patterns of connectivity of whale shark aggregations suggests that conservation and management approaches should focus at this scale [45,46]. Cross-jurisdictional management will be required, although to a much lesser degree and involving far less complexity than would be the case if cross-ocean movements were commonplace. Even at this smaller scale, regional approaches to management are probably to face substantial hurdles, given the track record of initiatives such as the management of fish stocks that migrate through, or occur in, more than one exclusive economic zone [47]. These are known as straddling stocks and the catch by one country can influence the fisheries of others [48]. According to international law, these stocks should be managed cooperatively through regional fisheries management organizations (RFMOs). Such cooperation and enforcement can be very difficult to achieve, especially as membership to RFMOs is voluntary and non-members can undermine cooperative efforts [47,48]. An understanding of the successes and challenges faced by RFMOs could be used to assist in the development of regional management and conservation plans for whale sharks.

Limitations of photo-identification and the effect of population size
The semi-automated process of photo-ID has the potential to match two unique individuals (false positive) or fail to find a match even though it does exist (false negative). By examining matches by eye, we reduced the possibility of a false positive and previous studies have found I 3 S to be a useful tool in minimizing the rate of false negatives [49]. The use of only one trained observer in image analysis further reduced the chances of a false negative by minimizing the variation in subjectivity when fingerprinting images. Although a majority of individuals had photo-ID images of both sides of their flanks, there was still the possibility that a match may have been missed as opposite sides of the shark could have been photographed in different locations. The growing size of photo-ID databases and resights within locations will further reduce the likelihood of this event.
A lack of records of sampling effort for each site also hindered the interpretation of data. Our findings of year-to-year variation in abundance and resight rates were most likely due to differences in sampling effort. Citizen science programmes could assist in this regard by reporting measures of effort such as number of days sampled per season and number of whale shark encounters per trip. However, such data will always be complicated by the variation in techniques (such as the use of spotter planes to locate sharks) among sites.
Although all four aggregations displayed similar proportions of new and resighted sharks, the numbers of years that individual sharks were resighted differed significantly among aggregations. This was probably due to the varying population sizes of the aggregations, so that in the Maldives, where the population was only around 90 sharks, the greatest proportion of the population was sampled on a year to year basis. Consequently, there was an enhanced probability of resighting resident sharks in successive years and new migrants. In the Seychelles, where population size was much larger, a relatively low proportion of the population was sampled and there was thus less chance of returning whale sharks being resighted or the identification of migrants from other source populations. The large confidence limits around the curves produced by our simulation reflect the impact this type of variability had on the chances of identifying migrants.

Conclusion
Despite the results of genetic studies, any evidence for the cross-ocean connectivity of whale shark aggregations on ecological timescales remains elusive. Our study suggests that photo-ID techniques are unlikely to be of use for the understanding of movement patterns at such scales because these broadscale movements, if they occur, are likely to be relatively rare events that we may not have the power to detect. At smaller, regional scales, photo-ID studies may be of greater use but considerable effort will be required to build libraries of identifications in order to detect movement patterns where population sizes are relatively large. Moreover, better coverage of sampling sites at regional scales will also aid understanding of the frequency, extent and degree of residency of whale sharks at aggregation sites.
Ethics. Photo collection by researchers at Ningaloo was carried out under permit numbers SF006870, CE002833 and CE003171 from the Western Australia Department of Parks and Wildlife (previously Department of Environment and Conservation). No permissions were required for photo collection in the Maldives. The study in Seychelles was conducted by the Marine Conservation Society Seychelles under a Memorandum of Understanding with the Government of Seychelles, represented by the Department of Environment, under the UNDP-GEF Project 'Strengthening Seychelles' protected area system through NGO management modalities'. The study in Mozambique was conducted under approval from the provincial government and was operated in compliance with all national laws. Christmas Island photos were collected entirely by ecotourists.
Data accessibility. All data are provided in the electronic supplementary material.