Introduction
Whale sharks Rhincodon typus were first described from the western Indian Ocean (Smith, Reference Smith1829). Relatively little is known about regional population sizes and trends, although local declines linked to targeted fisheries have been recorded from some areas worldwide (Alava et al., Reference Alava, Dolumbaló, Arnel, Yaptinchay, Trono, Fowler, Reed and Dipper2002; Pine et al., Reference Pine, Alava, Yaptinchay, Irvine and Keesing2007). In the Indian Ocean declines attributed to fisheries have been recorded in India (Hanfee, Reference Hanfee2001), Maldives (Anderson & Ahmed, Reference Anderson and Ahmed1993) and Thailand (Theberge & Dearden, Reference Theberge and Dearden2006). Although there has been no targeted fishery in Western Australia there is some indication of declining numbers, hypothesized to result from fishing in other areas of this population's range (Bradshaw et al., Reference Bradshaw, Mollet and Meekan2007, Reference Bradshaw, Fitzpatrick, Steinberg, Brook and Meekan2008); however, aspects of the decline are debated (Holmberg et al., Reference Holmberg, Norman and Arzoumanian2008). Whale sharks have been recorded from the waters around Seychelles since 1756 (Lionnet, Reference Lionnet1984) and, although there has never been a commercial whale shark fishery there, the potential effects of targeted fisheries in other areas within the region give cause for concern, as evidenced by the species’ categorization as Vulnerable on the IUCN Red List (IUCN, 2008).
Capture–mark–recapture methods are effective in characterizing population dynamics of mobile populations (Cormack, Reference Cormack1964; Jolly, Reference Jolly1965; Seber, Reference Seber1965). These techniques are based on marking a sample of individuals from a population and the frequency of their subsequent resighting, enabling estimates of population size, apparent survival and recapture probability (Seber, Reference Seber1982; Pollock et al., Reference Pollock, Nichols, Brownie and Hines1990). However, quantifying tag loss rates is usually required to correct parameter estimates, and this can be estimated by tagging animals simultaneously with two different markers (Stobo & Horne, Reference Stobo and Horne1994; Diefenbach & Alt, Reference Diefenbach and Alt1998). With the advent of inexpensive digital photography and photo-identification software, capture–mark–recapture techniques can now be applied using naturally occurring markings on some species. The unique spot patterns on whale sharks provide good reference marks for photo-identification (Arzoumanian et al., Reference Arzoumanian, Holmberg and Norman2005; Speed et al., Reference Speed, Meekan and Bradshaw2007). We used the accessibility and distinctiveness of whale sharks for photo-identification around Seychelles from 2001 to 2007, combined with conventional tagging, to estimate population size and apparent survival rates of this aggregation. The combination of photo-identification with conventional tagging allowed the estimation of probability of tag loss and provided data on how conventional tags could contribute to monitoring methods for this species.
Methods
The study area around the island of Mahe, Seychelles, has been described previously (Rowat & Gore, Reference Rowat and Gore2007; Fig. 1). Mahe is a mountainous island situated centrally on a shallow continental plateau. The area of study is the coastal zone area extending to a maximum of 4 km offshore (Fig. 1c). A micro-light aircraft was used to guide the survey team to sharks for in-water identification. Whenever possible the animals were sexed by the presence (males) or absence (females) of pelvic claspers (Meekan et al., Reference Meekan, Bradshaw, Press, McLean, Richards, Ouasnichka and Taylor2006) and details of prominent marks or scars were recorded. Marker tags were also attached opportunistically to sharks on the longitudinal keel (carina) to the left of the dorsal fin with a titanium dart on a 5 cm-long stainless-steel wire tether, using a fibreglass pole spear. The first tags attached in 2001 were hard, fibreglass-reinforced, plastic placard-type tags (Floy Tags, Seattle, USA; Plate 1a). Damage to these tags observed on resighting of some tagged animals in subsequent years made identification uncertain (Plate 1b). These tags accumulated a layer of bio-fouling organisms, which made the tags brittle and led to tag fracture and the loss of large amounts of the tag, including the identifying number (Plate 1c). An analysis of fouled tags was not done because only two were collected; however, barnacles (Balanus spp.) and gooseneck barnacles (Lepas sp.) were the most common species found attached. Re-tagging was therefore done from 2003 onwards using flexible synthetic rubber tags that helped prevent fouling adhesion (Aquasign, Champion Technologies, Aberdeen, UK; Plate 1d).
(a) The location of Seychelles relative to East Africa, (b) the shallow Mahe plateau and (c) the study area around the island of Mahe, extending to 4 km offshore.
Marker tags used in this study. (a) The fiberglass reinforced plastic Floy tag, prior to deployment, (b) Floy tag after 10 months and (c) after 1 year's deployment on a whale shark showing marine bio-fouling and break-up, and (d) the flexible polymer Aquasign tag used from 2003.
Photo-identification
Photographs were taken of whale sharks from 2001 to 2007. However, digital photo-identification increased the efficiency of collecting usable photographs from 2004 onwards, resulting in a larger number of photo-identities post-2004. The focal area for photo-identification was the unique spot patterns behind the gill slits (Arzoumanian et al., Reference Arzoumanian, Holmberg and Norman2005; Meekan et al., Reference Meekan, Bradshaw, Press, McLean, Richards, Ouasnichka and Taylor2006; Speed et al., Reference Speed, Meekan and Bradshaw2007). Printed photographs from the earlier years were digitally scanned and digital images were captured from videos where available. We used the software I3S (Interactive Individual Identification System), which was developed for matching the spot patterns of ragged-tooth sharks Carcharias taurus (Van Tienhoven et al., Reference Van Tienhoven, Den Hartog, Reijns and Peddemors2007) and is an effective tool for photo-identification of whale sharks (Speed et al., Reference Speed, Meekan and Bradshaw2007).
The images were separated into left or right side and the spot pattern was converted into a digital fingerprint using I3S. Fingerprints were then matched using I3S for identification of sharks (1) within 1 day's encounters, (2) among different days and (3) among years. An annual database of unique I3S identities recorded at least once during the course of that year were combined with the sighting data from conventional tags to create an inter-annual history for capture–mark–recapture models to estimate population size. Not all individuals had photo-identities of both left and right side; thus, images collected at different times from the opposite sides of two apparently different sharks may in fact belong to the same shark (Meekan et al., Reference Meekan, Bradshaw, Press, McLean, Richards, Ouasnichka and Taylor2006). The capture history was therefore compiled from conventional marker tag sightings and photo-identities of sharks with left-side images because these were more common than right-side images (287 cf. 222).
Tag retention
To predict the probability of tag retention an analysis of the number of days of known tag attachment was compared with the period when a tagged shark was first resighted without a tag (assessed via photo-identification; Arnason & Mills, Reference Arnason and Mills1981; Bradshaw et al., Reference Bradshaw, Barker, Harcourt and Davis2003). A simulation using the software R (Appendix; R Development Core Team, 2007) predicted the probability of tag retention after 1 year. Our approach was based on combining information of known periods of tag retention and tag loss from 21 deployed tags. We used a bootstrap sampling (with replacement) approach for each day for a simulated tag deployment of up to the maximum number of days observed with confirmed tag retention (773 days) to estimate the probability of a tag being retained for each day. We first created an equivalent-sized sample of tags and randomly determined whether they were retained or lost on each day of the simulation. If the tag was retained in the simulation and the total known tag retention duration was greater than the number of days elapsed, the probability of retention was set at 1.0. If the number of days elapsed exceeded the known duration of retention, then the probability of retention was set at 0.5 (that is, an equal probability of losing or retaining the tag after the final observation). For tags in the simulated sample chosen as ‘lost’, we assigned a probability of 0.5 if the number of days elapsed was less than the known duration of tag loss (equal probability of losing or retaining a tag on that day), and 0.0 if the known duration of loss exceeded the number of days elapsed. This process was repeated for each day for 1,000 iterations to build a 95% confidence interval (CI) for the decaying tag retention probability-time function.
Population size estimate
Open- and closed-population estimates based on capture–mark–recapture assume that the samples taken are representative of the population and that the likelihood of mortality, emigration, immigration and the probability of recapture are the same for all individuals (Cormack, Reference Cormack1964; Jolly, Reference Jolly1965; Seber, Reference Seber1965). For models that assume demographic closure (no net immigration or emigration), we used the software CAPTURE (Otis et al., Reference Otis, Burnham and White1978) to examine variants of the basic Lincoln–Petersen model. CAPTURE provides a goodness-of-fit test for each model and selects the most probable model(s) for the dataset (Otis et al., Reference Otis, Burnham and White1978). Some models allow heterogeneity of the likelihood of capture (h) and others allow for variation in time between capture (t), or apparent survival (ϕ; Pollock, Reference Pollock1991). To estimate population size in an open population model that does not assume demographic closure we used the Cormack–Jolly–Seber model (Schwarz & Arnason, Reference Schwarz and Arnason1966) based on the POPAN option in software MARK (White & Burnham, Reference White and Burnham1999). The model provides estimates of apparent survival (ϕ) and of capture probability (p) assuming that the animal is alive and available for capture, as well as probability of entry into the population per occasion (β) and population size (N). Bootstrap goodness-of-fit is not available for the POPAN option in MARK, so we constructed a second, recaptures-only analysis using the same data to estimate ϕ and p explicitly and to estimate goodness-of-fit of the model.
Results
Marker tagging and tag retention
We tagged a total of 211 whale sharks between 2001 and 2006 (Table 1), an average of c. 35 tags per year. The resighting of tags deployed in the previous year varied among years (0–52.3%) with an overall mean of 17.1% (Table 1). Photo-identification provided information on tag retention by identifying a number of previously tagged sharks that had lost their tags. Two periods in particular had poor tag retention, the initial period of 2001–2002 when tags were fracturing (four observed cases) and 2005–2006 when seven sharks were resighted with tether wires but no tags. The mean period of attachment was 380 ± SD 216 days from date of deployment to date of resighting (range 104–733 days). The simulated tag retention probability decay function, derived from the simulation based on verified tag retention or loss, estimated an annual tag retention probability of 0.43–0.69 (Fig. 2). Retention probability was relatively stable until approximately day 300, after which time it dropped markedly and assumed a new asymptote (Fig. 2).
Tag retention probability derived from the R simulation (see text for details and Appendix) based on known tag retention and loss, showing the rapid decline in the probability of tag retention after 300 days.
Tagging summary showing new tag deployments, number of resightings of tags deployed in previous years and the percentage of the resightings for each year relative to the number of tags deployed in the previous year.
Photo-identification
From 11,681 images of whale sharks collected from different sources, 1,149 were suitable for the photo-matching software (I3S). A total of 360 individual sharks had I3S identities of one or both sides in the image database. I3S identified 80 individual sharks that were resighted a total of 116 times in subsequent years. Using left-side-only matches to avoid possible duplications, 64 previously identified sharks were resighted a total of 79 times (Table 2). Over 2001–2007 resighting rates using I3S were variable (0–23.7%), with an overall rate of 21.9%. In the later years (2005–2007), when more photo-identities were taken, there was a continuing increase in the rate of resightings, with an average of 23.9% over these 3 years (Table 2).
I3S (Van Tienhoven et al., Reference Van Tienhoven, Den Hartog, Reijns and Peddemors2007) photo-identification summary for each of the 7 years of the study, giving annual number of left and right identities, the annual total number of whale sharks identified by either or both sides and the number of identity matches by left side only.
* Includes sharks that only have one side recorded and so may include duplicates (see text)
Population size
The closed population estimates using CAPTURE on the data from marker-tagged individuals only for 2001–2007 violated the tests for closure (Z = 2.98, P = 0.001). Therefore, all models examined gave highly variable and unreliable estimates. The open-population models implemented in MARK (POPAN option) estimated a population of 354–611 (95% CI, SE = 64). The model estimated a low annual probability of capture (p = 0.12, SE = 0.04) based on the conventional marker tagging data alone. These low estimates are due in part to the high annual tag loss probability.
Combining the marker tag and left-side photo-identity datasets gave a total of 512 identified sharks for 2001–2007 (Table 3). During this period 115 resightings were made with a sighting rate of 0.9–23.2% and an overall rate of 22.4% (Table 3). Using capture histories derived from the combined marker tag and photo-identity datasets, the closed population models also violated the assumption of closure (Z = -6.09, P = 0.0001). The open population model showed consistent increases in recaptures (p) after 2004 and fluctuations in apparent survival (ϕ; Table 4), which confound population estimates. This largely resulted from the introduction of intensive photo-identification in the last 3 years of the study, resulting in many more sightings for 2004–2007 than in previous years. Of the 360 individual photo-identities, only 34 were from before 2004. Removing the three earlier years and reanalysing the combined marker tag and photo-identity data for 2004–2007 provided a more reliable dataset. The closed population estimates using CAPTURE indicated that the time-dependent model {m(t)} was the most parsimonious with the Darroch population estimator. This gave a population estimate of 476–672 (95% CI, SE = 50). The tests for closure of the population were, however, still violated (Z = -4.557, P < 0.001). With the reduced data series the open-population POPAN models gave an estimated population size of 348–488 (95% CI, SE = 34) based on the constant model {ϕ(.)p(.)β(.)N(.)}(Table 5).
Number of individual whale sharks verified by tag or left side I3S (Van Tienhoven et al., Reference Van Tienhoven, Den Hartog, Reijns and Peddemors2007) photo-identification for each year and the number of resightings of identified sharks from previous years.
Parameters from Cormack–Jolly–Seber open population model for the combined photo-identification and tag data for 2001–2007, showing estimates of apparent survival (ϕ) and probability of capture (p) between each year.
Population estimate and parameters from Cormack–Jolly–Seber open population model (POPAN option) for the combined photo-identification and tag data for 2004–2007, giving estimates of apparent survival (ϕ), capture probability (p), probability of entry into the population (β), and population size (N).
Apparent survival, recapture probability and goodness-of-fit
The constant recaptures-only model {ϕ(.)p(.)} for 2004–2007 was the highest ranking with an AICc weighting (wAICc) of 0.437; for the three remaining models ϕ(t)p(.), ϕ(.)p(t) and ϕ(t)p(t), wAICc = 0.298, 0.158 and 0.106, respectively. The top-ranked model provided estimates of ϕt = 0.578 ± SE 0.095 and an overall p = 0.317 ± SE 0.072. The bootstrap goodness-of-fit simulation would not run on the reduced dataset because of too few resightings (i.e. many simulations produce zero capture histories that are not biologically or statistically valid) but on the full data it gave no evidence of overdispersion (p = 0.259). As such overdispersion was unlikely to be an issue with the reduced dataset, this indicates that the data fit both the recaptures-only and POPAN models.
Discussion
In view of the apparent decreases in whale shark populations in the Indian Ocean (Anderson & Ahmed, Reference Anderson and Ahmed1993; Hanfee, Reference Hanfee2001; Theberge & Dearden, Reference Theberge and Dearden2006; Bradshaw et al., Reference Bradshaw, Mollet and Meekan2007) it is important to verify whether the decrease is a change in the population size or a change in distribution, with possible corresponding increases in other areas. There is some evidence to support this from comparisons of aerial survey data from South Africa (Cliff et al., Reference Cliff, Anderson-Reade, Aitken, Charter and Peddemors2007) and Seychelles (Rowat et al., Reference Rowat, Gore, Meekan, Lawler and Bradshaw2009). In 2001–2005, when sightings per hour were relatively low in South Africa, they were higher in Seychelles and vice versa, with the exception of 2003 when both areas had many sightings. However, a more detailed comparison between other areas in the region is hampered by the lack of sufficient data to allow reliable population estimation in many locations. Tag resighting results from other areas are scarce. Over 1999–2003, 70 whale sharks were tagged with marker tags off Belize, with variable resighting rates of 0–18.8% between years and a high incidence of tag failure and bio-fouling (Graham & Roberts, Reference Graham and Roberts2007). Tagging studies around Holbox Island in the Mexican Caribbean deployed 556 marker tags on whale sharks over 2004–2006 but resighting data have not yet been published (B. Hueter, pers. comm.). The results of the between-year resighting of conventional tags in the Seychelles (an average of 17% over 6 years) initially suggested great potential for this technique; however, the high variability in tag resighting rates (0–52%) probably reflects the low tag retention. For this reason, the technique is likely to produce few reliable data.
The incorporation of the I3S identification and resighting data increased the number of sharks identified each year but these were concentrated in the latter part of the study. In 2006, 25 sharks identified by I3S were matched to sharks recorded in 2005 but only six matches were made to earlier years. In 2007, 18 matches were made to 2006, 12 to 2005 and six to earlier years. Comparing the sighting rates between marker tagging and I3S is more difficult because the rapid rate of tag loss after 1 year diminishes inter-annual resightings of marker tags. Furthermore, the major change in the capture of photo-identities over the period makes comments about the rates somewhat speculative, at least with respect to earlier years. Comparing the resightings of sharks identified in only the previous year with I3S for 2005–2007 gives resighting rates of 13.7–25% compared to marker tag rates that peaked at 52.3%. The advantage of marker tags is that once the tags are deployed they can be recognized by anyone without the need for a clear photograph or software to aid matching. The best use of marker tags would thus seem to be in short-term studies (< 1 year) where multiple opportunistic resightings are expected; these can be used to provide information on temporal and spatial ranges. The apparent problem of tag break-up and loss found in this study, as well as in Belize (Graham & Roberts, Reference Graham and Roberts2007), precludes the use of marker tags on this species for long-term studies. Photo-identification requires specialist equipment and standardization of the area used for matching, as well as some experience in confirming I3S matches. However, as the marking patterns on this species are apparently stable over time (Arzoumanian et al., Reference Arzoumanian, Holmberg and Norman2005; Meekan et al., Reference Meekan, Bradshaw, Press, McLean, Richards, Ouasnichka and Taylor2006) photo-identification is the most suitable tool for long-term population studies.
The increase in individuals identified in the later years by I3S confounded estimates of population size. This resulted in the need to reduce the time series used for population estimation to 2004–2007 to make the capture effort more equitable. The open population model based on the marker tags only for 2001–2007 estimated a population of 354–611 (95% CI) whereas the combined marker tag and photo-identification data for 2004–2007 estimated a smaller population of 348–488 (95% CI), with reduced standard error. The obvious lack of closure confirms the suggestion that there is extensive migration into and out of this population, even with the increased resightings. This suggests that the population may consist of a number of site-faithful individuals along with a large number of itinerant sharks that are seen only once, given the time frame of the relatively short study (7 years) compared to the > 20-year generation time estimated for this species (Bradshaw et al., Reference Bradshaw, Mollet and Meekan2007). This hypothesis may be confirmed by the addition of several more years of photo-identification data, assuming that similar resighting effort can be maintained.
Having both site-faithful and transient individuals poses challenges for management. In Canadian killer whales Orcinus orca (Baird, Reference Baird2001) site-faithful and transient individuals have different feeding habits, morphology and behaviour. The residents feed mainly on fish and tend to remain inshore whereas transients feed on marine mammals and are generally more common offshore (Baird, Reference Baird2001). This has allowed the differentiation of distinct populations for management (Barrett-Lennard & Ellis, Reference Barrett-Lennard and Ellis2001). Neritic marine turtles in many areas also have resident and transient groups; however, this is complicated by individuals making long-term use of different areas during the course of their lifecycle (Nietschmann, Reference Nietschmann1981; Boulon, Reference Boulon1994; van Dam & Diez, Reference van Dam and Diez1998). The population of whale sharks around Seychelles consists mainly of juveniles (< 8 m; Rowat & Gore, Reference Rowat and Gore2007), and thus management needs to consider potentially different developmental habitats.
The change in data acquisition during the course of the study, in particular the capture of large numbers of reliable photo-identifications, helped to resolve the issue of high tag loss and allowed for tag retention rate to be quantified. There was a large variation in the period of time that tags remained attached to the sharks, from as few as 47 days to > 733 days. The high variability between annual resighting rates in 2001–2002 and 2005–2006 has been partly explained by problems with either the tags or tether construction. The extreme variation in resightings for 2003–2004 (0–52.3%) may be because of the low numbers of sharks found in 2004, when aerial surveys recorded < 6 h-1 compared to > 10 h-1 in 2003 and 14 h-1 in 2005 (Rowat et al., Reference Rowat, Gore, Meekan, Lawler and Bradshaw2009).
The whale shark population estimates derived from Mahe, Seychelles, suggest that there is considerable migration into and out of the population within the time frame considered (6 years) but equivalent results are not yet available for many other Indian Ocean aggregations. The apparent decline in populations in other areas of the Indian Ocean (Anderson & Ahmed, Reference Anderson and Ahmed1993; Hanfee, Reference Hanfee2001; Theberge & Dearden, Reference Theberge and Dearden2006; Bradshaw et al., Reference Bradshaw, Mollet and Meekan2007, Reference Bradshaw, Fitzpatrick, Steinberg, Brook and Meekan2008) may be reflected in the Seychelles aggregation, and continued monitoring will allow us to test the hypothesis of transience among the regional aggregations within the greater Indian Ocean. Such longer-term information will confirm whether the observed and inferred declines are a regional phenomenon or an artefact of change in population distribution.
Satellite tracking data indicate that whale sharks undertake long-distance movements (Eckert & Stewart, Reference Eckert and Stewart2001; Eckert et al., Reference Eckert, Dolar, Kooyman, Perin and Rahman2002; Wilson et al., Reference Wilson, Polovina, Stewart and Meekan2006; Rowat & Gore, Reference Rowat and Gore2007). These studies add weight to the hypothesis that whale sharks found in known aggregations such as in the Sea of Cortéz, the Philippines, Western Australia and Seychelles move away from these sites. However, none of these studies have yet tracked a whale shark returning to the area in which it was tagged. The lack of a return track from sharks in the Seychelles aggregation is contrary to the number of whale sharks found returning by photo-identification or marker tagging. This is probably due more to problems with the retention of satellite tags by the animals than any other factor.
With the increased use of photo-identification for whale sharks the potential for regional and international comparison of identities may provide further detail of the ranges of sharks found in annual aggregations such as in the Seychelles. To date, no whale shark from Seychelles has matched identities captured from either Ningaloo or Mozambique (M.G. Meekan, C.W. Speed & S. Pierce, pers. comm.) but an ongoing collaborative identification project within the Indian Ocean may yet reveal spatial ranges for sharks in these aggregations.
The effective conservation and management of such a wide-ranging species depend on understanding of its population status and its migratory habits in all major aggregations. The results shown here are the first to point to both a site-faithful and transient population of whale sharks, which is valuable information for conservation management. Longer-term study may show how long individuals stay in the site-faithful group.
Acknowledgements
Our thanks to the hard work of the volunteers of the Marine Conservation Society Seychelles who have assisted with data collection. Rachel Graham, at that time a PhD candidate at York University, assisted with the initial implementation of the tagging programme during 2001–2002. Funding was provided by the Save Our Seas Foundation, the Global Environmental Facility (under the Seychelles Marine Ecosystem Management Project) and through public donations. Maps were produced courtesy of the Seaturtle.org MAPTOOL facility. This programme is run with permission from the Ministry of Environment and Natural Resources, Seychelles, and in accordance with the Wild Animals (whale shark) Protection Regulations, 2003.
Appendix
The appendix for this article is available online at http://journals.cambridge.org
Biographical sketches
David Rowat is a member of the IUCN Shark Specialist Group and has been researching whale sharks since 1996. His other research interests include turtle and cetacean ecology and the development of conservation measures for migratory sharks. Conrad Speed is a marine ecologist whose research focuses on the movement patterns, feeding ecology and population biology of sharks. Mark Meekan is a fish biologist whose research examines tropical sharks and reef fishes. Mauvis Gore works on behavioural ecology and conservation management of basking sharks and Pakistan's marine cetaceans, and on evolution of social organization focusing on primates. Corey Bradshaw has research interests in population ecology (regulation, sustainable harvest, control and extinction dynamics), climate change biology, behavioural ecology and invasive species.
