Ethics statement

This study was performed according to national Portuguese laws for the use of vertebrates in research. All procedures followed the guidelines for the use of fishes in research of the American Fisheries Society. No animals were manipulated or sacrificed. All data were obtained during regular commercial fishery operations under the Azorean Fisheries Observer Program (POPA) jointly run by IMAR- UAz and the Regional Government of the Azores (SRRN/DRP).

Sightings

The Azores archipelago consists of nine islands that are broadly grouped into the western, central and eastern island groups (Figure 1A) and is located along the mid-Atlantic ridge in the Atlantic Ocean (Figure 1b). We used the pole-and-line tuna fisheries database obtained from the observation programme for fisheries of the Azores (POPA) to look for whale shark sightings in the wider Azores region. The program was launched in 1998 and annually deploys observers in around 20 of the large tuna vessels which equates to approximately 50% of the tuna fleet. It operates with trained on-board observers who record georeferenced data on all tuna fishing activities and other scientifically relevant information including fishing effort, tuna catch, and sighting of different associated and non-associated species such as turtles, marine mammals and whale sharks. The tuna season typically extends from May to November during which boats actively seek for tuna and come into port every 2–8 days, depending on the amount of fish caught. The observers are permanently deployed and observations may occur whenever to boat is out of the harbour. A fishing event with a whale shark associated to tuna will always result in the sighting of the animal very close to the boat's deck (<15 m). The average speed of the tuna vessels when searching at daytime is approximately eight knots (∼15 km h⁻¹), meaning that a vessel will cover over 1300 linear km over a week. The presence of a whale shark swimming at or very close to the surface resulting in a fishing event will be detected by an observer covering 360° from the boat's flybridge. Telemetric studies done elsewhere [18] all found that whale sharks are known to spend the majority (∼80%) of their daytime very close to the surface, even if this behaviour is frequently interspersed with deep dives or permanence at deeper layers of the epipelagic domain. Therefore, we can reasonably assume that the probability of daytime whale shark detections by observers should be high and unbiased by their nighttime behaviour.

Fishing effort is not evenly distributed across the Azores Economic Exclusive Zone (EEZ, ca. 1 million km²). No approach can guarantee the removal of all double sightings in a location without further verification from photographic databases, therefore we propose a filtering technique to minimize potential duplicate sightings of the same individuals within each surveyed month. A screening process was applied to the raw data for the study area, whereby the region was divided into a regular grid of 4 km×4 km sampling units (SU; n = 24642), which corresponds to the highest spatial resolution available for environmental (remotely sensed) data. These SUs were used to filter the raw data per month by aggregating all events within each SU for each year/month combination. A whale shark was present within a SU if there was at least one sighting made within a calendar month. Although this does not remove the possibility of multiple detections of the same individual entirely, the monthly resolution between successive detections is expected to remove the vast majority of these occurrences. Pseudo-absence generation follows the recommended guidelines outlined by [19] where a large number of pseudo-absences are randomly generated with respect to each presence with a ratio of (1∶100), which allows for a greater sampling of the background conditions. For each presence, one hundred random pseudo-absences were generated for that month (centred on the 15th of each month) by randomly resampling the SU grid with replacement. This procedure ensured that the temporal distribution of presences and pseudo-absences retain the same proportion across all months and years.

Environmental variables

Satellite derived estimates of both sea surface temperature (SST) and chlorophyll (chl-a) were extracted for the area for the years 1998–2013. In order to cover this full period, the satellite data for both SST and chl-a for this period had to originate from different sources: AVHRR1 (1998–2009) and MODIS2 nightime SST (2002–2013) for SST, and Globcolour database (1998–2013) for chl-a. The merging of both SST datasets required a regression of overlapping values in order to produce a full length dataset for each variable (Table 1). The North Atlantic Oscillation (NAO) index is a basin-scale alternation of atmospheric mass between the Subtropical and the Arctic Atlantic [19]. The index used here is the station based monthly index obtained from the NCAR database (http://www.cgd.ucar.edu/cas/jhurrell/indices. html). Bathymetry was derived from the General Bathymetric Charts of the Oceans (GeBCO) dataset [20], which is a global topographic relief model with a resolution of 0.086^o from which the average slope (in angular degrees) within each SU⁻¹ was derived. The distance from the shore (m) was calculated for each sampled SU by using the distance from the SU midpoint to the nearest island shoreline. The level of fishing effort was also incorporated by using the frequency of fishing events to have occurred within each SU across all years. The sum of the fishing events within each month (events SU month⁻¹) is used to verify if returning to the same area would have an effect on the probability of whale shark detection rates.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Predictor variables used for GAM modelling of whale shark detections in the Azores EEZ.

https://doi.org/10.1371/journal.pone.0102060.t001

Model selection

A model was developed to test which environmental variables might explain the probability of whale shark sightings across the Azores EEZ within the 16 year period of study. Monthly whale shark sightings (presence/absence) were set as the response variable and the spatial and non-spatial environmental and climatic variables outlined in Table 1 were used as predictors. Pearson correlations were first used to examine potential co-linearity between predictor variables. Highly correlated variables (r>0.7) were not included within the same models. Exploratory biplots revealed that the relationship between the response variable and several of the predictor variables were non-linear so a binomial generalized additive model (GAM) was chosen to model the whale shark data. GAM are regression models where smoothing splines are used instead of linear coefficients for covariates. The estimated degrees of freedom (edf) for each of the variables were limited to ≤3. This allowed for either linear, curvilinear or sigmoid relationships between whale shark presence/absences and the explanatory variables, which represent a reasonable range of relationships that could be expected and interpreted. Presences and absences across all years are initially split and are randomised separately to ensure an adequate number of presences in both datasets. A pseudo-random algorithm is implemented to randomly assign each presence and absence into the training and testing dataset in a ratio of 75∶25 respectfully. The model derived from the training GAM will be used to evaluate the predictive capabilities of the model using the receiver operating characteristic curve (ROC – e.g., [21]) on the test dataset. The ROC is used to determine the accuracy of the model by evaluating the models capability of discriminating between sites where a species is present versus sites where a species is absent with values approaching 1 indicating a greater ability for the model to discriminate between presences and absences. A stepwise procedure was used to select the most parsimonious model using the Akaike's Information criterion (AIC) function dropterms. There are concerns associated with stepwise procedures due to the potential of introducing variable bias and model overfitting, and several alternatives have been suggested to help counteract this problem [22]. We addressed this issue by following the recommended procedure outlined by [23] in which a double penalty approach is used together with a maximum likelihood (ML) estimator of the regression splines. By adding the extra penalty to the functions in the null space of the model, a variable that does not contribute significantly to the model can be completely removed and is a robust and powerful way of removing variable selection bias introduced through stepwise procedures [24]. The partial deviance explained (%) is then calculated for all variables present in the final model. All statistical analysis and model development were performed using packages (gam; mgcv) developed for the R programming environment [25].

Regional patterns, migrations and the role of fringe habitats

Over the 16 year period of this study we found that the probability of occurrence of the whale shark in the wider Azores region increased drastically in 2008. Prior to this, and for a full decade, these large animals had only been sighted sporadically in spite of the large fishing effort across the Azorean EEZ. Thereafter, they became frequent summer visitors to the region. In addition, we found that SST helps predicting the probability of whale shark occurrence in the region across the years, providing a plausible explanation for the post 2007 shift at the regional scale.

As with most fisheries datasets, the survey effort in this study was not uniformly distributed, which can be problematic when discerning the effects of environmental cues on the patterns of the fishery-dependent occurrences. The pole-and-line tuna fishery in the Azores operates all over the EEZ but the vessels will often shift between areas where higher catches are expected based on recent catches (i.e., within the last couple of weeks). However, we believe that this variation did not significantly impact sightings of whale sharks because there is a very strong association of whale sharks with tuna in the region, as ascertained by the fact that nearly all sightings coincided with a successful fishing event. In addition, the probability of whale shark occurrences within a given 4 km×4 km SU did not increase with a concomitant increase in fishing effort above a small threshold (3 events SU⁻¹). It should be noted that the highest distribution of fishing effort was concentrated around the central islands, however sightings in this area were rare, with only 10 sightings made within 20 km of the coastlines of the central islands. If whale sharks were present anywhere in the region in substantial numbers prior to 2008 one could expect larger whale shark associated catches and, concomitantly, higher whale shark sightings. Therefore, the inter-annual variation found in sightings should largely reflect a true variation in whale shark visits to the region.

This spatiotemporal pattern of regional occurrence linked to the thermic regime at Azorean latitudes in the summer must be put into context with the highly seasonal migratory behaviour of the species [5], [8]. If whale sharks occurring in the Azores during the summer come from any of the North Atlantic tropical areas where they are known to occur or aggregate annually, such as the Caribbean/Gulf of Mexico and Gulf of Guinea/West Africa [16], [26]–[27] (Figure 1), it would presumably take them some time to reach the higher latitudes of the Azores, as those areas are in the region of 2,500 to 6,000 km away from the region. Whale sharks are able to cover distances larger than this across the open ocean [9]–[10] but they possibly do so at considerable energetic costs. Additionally, the summer peak SST in the Azores region is always below the narrow optimal thermic envelope for whale shark (<26.5°C) described for other oceans [11], [16]. Therefore, periodic migrations up north and the concomitant expansion of whale shark oceanic habitat to the Azores might be worthwhile only if migrating individuals can access important local resources (e.g. food, access to mates) while keeping within temperatures warm enough to reduce the metabolic costs of migration [11]. This would explain the clear association seen between the annual occurrences and the 22°C temperature threshold. Such a constraining environmental effect is further noticeable when comparing the match between the geographic location and shape of the 22°C isotherm and the whale shark occurrences across the EEZ within a given season. In particular, there was a striking match between the change from a south-north to a west-east orientation of the isotherm in 2012 and the geographic dislocation of whale shark sightings from the southeast to the south central and western areas of the archipelago in that year. The fact that the interannual change in orientation of the SST gradients in the region appears to be a common mesoscale oceanographic feature [28] clearly indicates that, once whale sharks reach the region, their patterns of pelagic habitat use also responds to these features.

This relationship and the sharp post-2007 increase highlight the possibility that broadscale shifts in the regional biological and physical dynamics driven by climate change may be causing a shift in whale shark oceanic habitat use. In particular, it is most likely that we are witnessing a bottom-up response by the Atlantic whale shark population to changes in lower trophic level organisms (plankton, small fish), which have been evident in the pelagic ecosystem of the North Atlantic for some time [29]. Tropical Atlantic SSTs have shown a steady rise in recent years in accordance with trends found throughout the North Atlantic basin [30]. Changes in the SST, which represent the principal manifestation of climate change in our oceans, have been linked to large changes in the pelagic ecosystem, particularly in the North Atlantic [31]. Poleward extensions in the distribution of plankton [32] and fish species [4] of southern, more tropical affinities across Europe, as a result of the warming of coastal and oceanic waters, have been observed. This warming has led to a widening of the “tropical belt”, forcing an increase in stratification intensity and a decrease in the depth of the mixed layer, which has seen a ten-fold decrease in zooplankton biomass in the tropical ocean [33] and a four-fold decrease in regionally productive upwelling zones in the Gulf of Guinea [34]. This may be viewed as a trophic response to strong bottom-up regulation, as decreases in mean chl-a levels in tropical regions have been positively linked with a decrease in mixed-layer depth caused by continued ocean warming [35]–[36].

Finally, although there is no size information in our observer dataset, sightings of the whale shark appeared to be comprised almost entirely of adult (≥8 m) and large-sized (>10 m) individuals. This is supported by the high consistency of reports from tuna boat skippers (personal observation), local sharkwatching operators (Nuno Sá, personal communication; personal observations) and by our own observations in the field during three years of tagging operations around Santa Maria. If smaller, juvenile individuals (2–7 m) were frequent in the region, their occurrence would have been detected given their characteristic near-surface behaviour at their aggregation sites [10]. This pattern contrasts with the typical juvenile composition of the whale shark aggregations at tropical sites [8] and strengthens the hypothesis that the Azores region is part of the wider oceanic habitat in the North Atlantic for adult whale sharks, of which little is known globally [8]–[9], [14], [16].

Local scale patterns: what brings whale sharks up north?

Within the region, there was a clear spatial pattern in whale shark sightings, namely the disproportionate amount of individuals appearing around the island of Santa Maria. This accumulation of sightings appears to reflect a true aggregation of individuals because our filtering procedure of monthly 4 km×4 km cell sightings is very conservative as to the possibility of repeated individual sightings which would require the animal to stay or repeatedly return to a cell in consecutive months.

To start with, the accumulation of sightings off Santa Maria should reflect the fact that average seasonal SST is indeed higher and typically above the position of the 22°C isotherm in that area as opposed to the other island groups. However, it also seems to be in agreement with the hypothesis that individuals must use cues other than large-scale oceanographic features (such as temperature) to orient their migrations and patterns of oceanic habitat use, at least once they reach the region. The NAO index is one of the dominant modes of climate variability in this region and captures much of the variability in climate-ocean interactions, including SST. In the tropical Atlantic, strong positive associations with the NAO have been detected in both phytoplankton [37] and zooplankton [38] populations. Positive NAO years shift the timing and abundance of the chl-a maximum further north and much later causing tropical blooms to be later and smaller in magnitude [39]. Whale shark occurrences in Ningaloo Reef, Australia have been found to be closely associated with the Southern Ocean Index [15]. There was a significant effect of the NAO _winter index as a change towards positive NAO phases increase the probability of whale shark appearances in the region. A longer time series would be necessary to verify this effect.

Access to food resources is one major possible cue. At the archipelago scale, there was a general increase in whale shark sightings after a given threshold in chl-a was reached, yet we found no evidence that the area around Santa Maria holds increased primary productivity In fact, there was a general increase in productivity around the other islands but not this one, which lies further south and therefore above what might be the optimal water temperature for that productivity to develop. The timing of whale sharks sightings in the area coincide with a period of highest SST and lowest chl-a values suggesting a potential lagged response between whale shark an chl-a as whale sharks actively feed on zooplankton and small planktivorous fish [5], [7] and, unlike the large grazer basking shark (Cetorhinus maximus), do not graze directly on algae [40]. Whale sharks have been seen and documented feeding on schools of juvenile horse mackerel (Trachurus picturatus) around Santa Maria (Nuno Sá, personal communication). These and other small planktivorous fish such as boarfish (Capros aper) and snipefish (Macrorhamphosus scolopax) are very abundant in the epipelagic layers over the island's slopes and nearby seamounts, and constitute main prey items of other pelagic predators that frequent these habitats in the region including seabirds [41], fishes [42] and sharks [43]. Our results also suggest that whale sharks may orient themselves with respect to seamounts and areas of steeper bottom slope within the area, as both productivity and sightings were indeed higher close to these topographic features. These patterns correlates well with previous findings for the region that report a peak in the biomass of marine top predators near but not over seamount summits, i.e., at their slopes, and support the hypothesis that this behaviour is adaptive by resulting in increased feeding opportunities in these areas [44]. Therefore, it is very well possible that the aggregation of whale sharks at particular areas once they reach the region reflects local abundance of prey, even if this behaviour is constrained at a first level by the larger scale environmental cues, such as temperature, that determine their migration probability.

Regardless of the whale shark feeding ecology in the region, it is worth noting that the Azorean aggregations could serve other functions such as increased opportunities for mature individuals to engage in social interaction, including mating. At present, there is no evidence to support or disproof this hypothesis, other than the anecdotal evidence that they are essentially composed of adult individuals. It is vital that we explore the demographic composition and the fine scale behaviour of the individuals in Azorean waters, given the profound lack of knowledge on the species' adult and mating grounds [8], [41].

Conclusions

Based on a unique dataset spanning 16 years, we have demonstrated that changes in the mean annual summer SST are an important factor in strengthening the probability that whale sharks would migrate into the higher-latitude, oceanic habitat around the Azores, and that this region is at a thermal boundary for this species. As a result, the region has recently attracted seasonal numbers that highlights the significance of the Azores in the global network of well-established areas of seasonal aggregations in other regions [3], [8], [9], [12], [13]. Yet, the relevance of this region as oceanic habitat and its connectivity with other habitats across the Atlantic needs to be clarified. While we can reasonably assume that the recent influx of whale shark individuals into Azorean waters is related to the large-scale temporal changes in the tropical climate, the reasons for the spatial variability within the region and, in particular, the local aggregations remain unknown, although local productivity related to seamounts appears to play a major factor role.

Whale sharks have been actively hunted in some areas of the Indo-Pacific [45], inadvertently captured in tropical pelagic fisheries [46], [ 47], and suffer injuries from boat traffic in certain coastal aggregations [8], prompting growing conservation concerns and leading to the species' IUCN designation as “Vulnerable” [48]. Our findings underline the potential for an increase of the wider Azores region's importance as an oceanic habitat for the whale shark in the North Atlantic in years of exceptionally high water temperature, and for a concomitant shift in the whale shark distribution within the Atlantic Ocean, as predicted by global modelling studies [17]. In the future, such shifts need to be placed in the context of decadal and very long-term changes in this ocean. The use of telemetric studies, the continuation of the fisheries observer program and the establishment of a monitoring study directed at the aggregation areas within the region will be most useful to further address these questions in greater detail.