R. typus individuals were tagged between 2007 and 2012 at several locations in Pacific Panama with near real-time tether/towed satellite transmitters (model SPOT 253C; Wildlife Computers, Redmond, WA, USA). The tag model used specifies a battery life of 280 d, assuming 250 Argos transmissions per day, which occur only when the fish is swimming near the surface and the transmitter is exposed. To maximize battery life, transmitters were programmed to limit transmissions to a time block from 01:00 to 22:00 h every 2 d and to slow the repetition rate after 10 successive dry transmissions. The float tag was attached to the fish ca. 10 cm below the first dorsal fin by a 1.5 m tether made of stainless-steel cable and a 3.5 cm stainless steel dart using a 3.0-m Manny Puig™ pole spear. Visibility and sea conditions precluded obtaining the sex and size estimates, important demography variables “to ascertain the conservation status and ensure persistence of the specie” (sensu Sequeira et al., 2016). The Animal Care and Use Committee of the Smithsonian Tropical Research Institute approved this sampling procedure.

ARGOS transmissions were filtered by deleting all transmissions in land (dry), quality “Z” and movements representing speeds over 100 km/day (6.4 km faster than the maximum speed reported by Eckert and Stewart, 2001). Two comparable analyses were performed: one using complete filtered transmission data (CD) of the animals and one with partially trimmed or removed trajectories (see l Figures S1 ) for suspicious of earlier detachment or freely floating tags (sensu Hearn et al., 2013). A Bayesian state-space random walk (SSRW) model for animal movement was used to optimize the obtained R. typus tracks based on ARGOS satellite tracking data error, which Vincent et al. (2002) calculated to be around 226 m for location class (LC) LC3, 372 m for LC2, and 757 m for LC1 and LC A. This model was used because it is useful when observations occur irregularly in time and when working with ARGOS transmission estimation errors (Jonsen et al., 2005). The model was run using the “fit_ssm” function in the “bsam” package (Jonsen, 2016) of R software (version 3.4.4; R Core Team, 2018). The model was set to have four time-steps per day (tstep = 0.25), 5,000 samples during the adaptation and update phases, a thinning of 10 to minimize within-chain sample autocorrelation, and a span of 0.2 for the degree of smoothing.

Bayesian SSRW-corrected R. typus tracks were modeled with a hidden Markov model (HMM) to identify two different behavioral states—foraging and migratory behavior. The environmental variables described in Table 1 were used as the model covariates. The HMM model was run using the “fitHMM” function in the “moveHMM” R package (Michelot et al., 2016). The initial values used to differentiate between foraging and migrating behavioral states were set to 5 ± 2 m and 30 ± 10 m for the step mean, respectively, and pi was set to 0 for the turning angle.

Foraging areas were identified and mapped by creating a kernel density plot (in ArcGIS v10.6) with all the modeled locations identified by the HMM model for that behavioral state. Low-use, medium-use, and high-use areas were determined based on the first, second, and third quartiles, respectively.

Hearn et al. (2018) determined detachment point of towed SPOT-5 satellite tags by comparing the trajectories of four tagged whale sharks (one individual with a SPLASH) that transmitted between 31 and 167 days and two deliberately floating tags. They concluded that detachment points can be usually identified by interpreting daily transmission times, class quality along with depth and temperature of the transmitted data. In addition, this interpretation can be facilitated with available information on general ocean circulation patterns (Hearn et al., 2018).

We used daily sequences of transmission times (sensu Hearn et al., 2013) to preliminary test those tags were attached to the sharks and not freely floating along the currents and winds ( Figure S1 ). Shark traveling speed was calculated using filtered but unmodeled locations. Shark tracks (consecutive list of locations) were converted to ArcGIS line shapefiles using the Data Management tool “XY to Line”, which created a line along chronological locations. Each track line was then split at each location point by using the ArcGIS tool “Split Line at Point”, resulting in segmented tracks with individual lines between locations. The length (km) of each segment of the track was then calculated by using the “Calculate Geometry” tool available in the Attribute Table menu in ArcGIS. The difference in time (hr) between consecutive locations was calculated and used to estimate traveling speed (TS) between locations as: TS= Segment length (km)/time difference between consecutive points (hr).

Daily marine currents (surface and 15 m depth at ¼ ° resolution) were obtained from Copernicus database named: MULTIOBS_GLO_PHY_NRT_015_003 (Global Total Surface and 15m Current (COPERNICUS-GLOBCURRENT) from Altimetric Geostrophic Current and Modeled Ekman Current Processing. Current Speed (CS) was calculated from V and U current vectors using the Pythagorean theorem in where: CS=square root(V²+U²). Current speed, at ¼° resolution was overlapped spatially with whale shark locations by using the “Spatial Join” function in ArcGIS Pro software (set to a search radio of 30 km for each location). A temporal overlap was done by merging the two datasets (current speed and shark locations) by date in R software. Current speed (mean speed within 30 km around each location during the transmission day) and whale shark traveling speed (for the segment between two consecutive locations) were correlated using Pearson method Figures S1, 2 ). This correlation was analyzed independently when whale shark transmissions occurred inside an eddy as explained earlier in the manuscript.

Solitary individuals were tagged sporadically between September 2007 and March 2012. A total of 32 R. typus were tagged along the Pacific coast of Panama; 25 were tagged in the Gulf of Chiriquí and 7 were tagged in the Las Perlas Archipelago, Gulf of Panama ( Table 2 ). Two tags failed to transmit. A total of nine sharks showed dubious transmission times or detached shortly after tagging ( Figure S1 ): five with long migrations and high frequency of transmission rates in morning hours (54741, 54745, 54764, 54878, 107719) were isolated from the analyses, and four tags with relative short movements were partially trimmed when transmission rates changed (107708, 54747, 54749, 54870). Those nine full trajectories, however, were modeled and analyzed for detachment by correlating sharks traveling speed with real-time marine current speed and wind current (see below). Accordingly, to compare analyses, results are presented for trimmed/removed data followed by full untrimmed data (30 tags) in parenthesis. Tagged sharks transmitted 51 d (498) with up to 307 (1,053) transmissions ( Table 2 ). Rhincodon typus migrated north (to the Gulf of Tehuantepec, Mexico), south (to Ecuador), and to the open ocean ( Figure 1 ).

A Pearson correlation test showed that the number of transmission days did not correlate with the number of transmissions (r² = 0.05). Transmitters need to be at the surface for a certain amount of time to send a correct location, and some R. typus may only surface sporadically or under rough sea conditions, meaning that while an individual shark’s tag may be transmitting, received transmissions may be infrequent or sporadic.

The random walk state-space model was run for 18 (24) of the 30 tagged R. typus. Four sharks had information gaps > 100 without transmission, which impeded the ability of the model to run properly. For these four cases, the original data were split into ‘before’ and ‘after’ the data gap; these periods were modeled separately and then bound together to create a single modeled track. Seven sharks had too few transmission days or transmissions to run the model. These seven sharks were not included in the MHH model but were included in the estimation of % of locations inside MPAs, and in calculating the spatial and temporal overlap with vessel traffic and fishing grounds (see Table 2 ). From a total of 1,053 Argos locations, the Bayesian state-space random walk model (programed to estimate the location of each whale shark at four locations per day) resulted in 3,118 (9,968) modeled locations, now on referred to as “modeled locations”. Based on these modeled locations sharks traveled at a mean speed of 8.5 (10.4) km/day with a maximum speed of 76.9 km/day ( Table 3 ).

Based on the MHH model, R. typus spent equal amount of time foraging and migrating, with an average of 50% ± 6.5% (± SE) of their time foraging (range = 0%–100%) and 49.7% ± 6.5% of their time migrating (range = 0%–100%), contrary to the full trajectory data with an average of 76% ± 0.04% (± SE) foraging (range = 17%–100%) and 24% ± 0.04% migrating (range = 0%–83%) ( Table 4 , Figure 2 top). Tagged R. typus foraged mainly inside the Panamanian Exclusive Economic Zone (EEZ) or the territorial waters in both the Gulf of Panama and the Gulf of Chiriquí and, to a lesser extent, parallel to 4°N, north and east of the Galapagos islands ( Figure 2 bottom).

The MHH model results showed that environmental conditions were different for each behavior. Specifically, chlorophyll-a concentrations and primary productivity were significantly higher when R. typus were foraging (p < 0.001 in both cases). The SST did not differ significantly between the two behavioral states ( Table 5 ). Mean water depths were significantly (p < 0.05) deeper when the tagged sharks were migrating -1,534.4 ± 41.3 m (-2,366.5 ± 20.9 m) than when they were foraging -765.6 ± 26.8 m (-1,015.5 ± 18.2 m) ( Table 5 ), and most foraging locations 76.9% (66.5%) occurred in coastal waters (< 50 km from the shore). Oceanic areas (50 km away from shore) shallower than 1,000 m were scarce within the study area, covering only 2.05% (3.64%) of the territory, however, up to 5.5% (19%) of the modeled locations occurred within these “shallow” areas indicating a significant use of seamounts and ridges by R. typus but only for the complete data (Resource Selection model, p < 0.005). The probability of changing between behavioral states (i.e., from migrating to foraging and vice versa), as well as the probability of staying in a particular state under the stationary distribution ( Figure 3 ), was correlated with chlorophyll-a, productivity, and SST ( Table 6 ).

There was a total of 24,437 eddies registered inside the study area between September 2007 and December 2012 (study period), only five of the analyzed shark tracks spend some time within an eddie, ranging from 0.5% to 11% of their time, especially in anticyclonic eddies (65%) ( Figure 4 top right, Table 4 ). However, full analyzed trajectories showed 10.25% of R. typus tracks overlapped temporal and spatially within them. On average, R. typus spent 6.2 ± 0.02% of their time within an eddy, although there was a high degree of variation among individuals. Six animals spent a significant amount of time (up to 50.4% in the case of individual 107719) within eddies. When foraging, an average of 10.9 ± 2.0% of the locations were within an eddy compared to 17.7% when migrating. When whale sharks were located inside an eddy, the eddy speeds (measured within the amplitude of an eddy) were significantly higher when R. typus were migrating (p < 0.001). There was not an apparent preference for staying in cyclonic or anticyclonic eddies, the frequency of whale sharks inside each these was 50%, both when calculated with the number of modeled locations or time spend inside the eddy ( Table 4 , Figure 4 top left).

R. typus visited a total of 10 (17) MPAs within three (five) countries (Mexico, Costa Rica, Panama, Colombia, and Ecuador) with 3.4% (22.6%) of the modeled locations occurring within an MPA, which accounted for a mean of 35.8 ± 5.4% (48 ± 6.3%) of their time; 97% (86%) of these locations occurred in Panama with higher rates in Parque Nacional Coiba and the recently expanded oceanic MPA Cordillera Coiba with nearly 68,000 km² ( Figure 4 bottom, Tables 2 , 7 ).

Of the areas occupied by tagged R. typus, 26.2% (50%) of all modeled locations occurred within high-effort fishing areas, 15% in medium-effort and 58% in low effort areas ( Figure 5 top ). In these areas, 86% of the fishing was tuna purse seines with a monthly average (mean ± SE) of 4,025 ± 1,417 fishing hours (503,987 fishing events between 2012 and 2018), and 13% were drifting longlines with a monthly average of 621.4 ± 157 fishing hours (174,701 fishing events). Purse seine effort did not differ significantly between the first, second, or fourth quarters (p > 0.05) but was lower during the third trimester (16%). Most of the drifting longline effort in the area occurred during the last quarter (53.2%). Vessels with a Colombian flag were the most common (25.2%, equivalent to an average of 1,170 fishing hours per month) followed by vessels from Mexico (24.26%), Venezuela (14.9%), and Panama (14.3%). The majority 70.9% (88.8%) of R. typus modeled locations occurred within high-impact vessel traffic areas (areas with a higher density of vessels moving at higher speeds), 22% (11.2%) of the modeled locations occurred with medium impact areas, and 6.9% (0.0%) modeled locations occurred within low-impact areas ( Figure 5 bottom ).

Traveling speed was not correlated to current speed ( Figures S1, 2 ) at two different depths for any of the five suspicious whale shark trajectories: 107719 (R= 0.078, p>0.05 at surface; R=0.15, p>0.005 at 15 m), 54741(R= -0.032, p>0.05; R=0.025, p>0.005), 54745 (R= -0.004, p>0.05; R=-0.1, p>0.005), 54764 (R= -0.2, p>0.05; R= 0.045, p>0.05), and 54878 (R= -0.2, p>0.05; R= -0.059, p>0.05). In addition, tag speed was tested for correlation with wind diffusivity current speed and marine currents, no correlation was found between these three variables, except for a significant correlation between tag speed and wind speed for tag 54764 ( Table 8 ). These analyses suggest that those five tags were not floating on the surface, carried away by marine currents and winds, or detached from the animal.

We analyzed and plotted the individual tracks for all 30 animals. However, we paid particular attention to those five tags with the longest record (PTTS 107719, 54741, 54745, 54764, and 54878) that seem to suggest that they separated from the animal at some point and continued transmitting with the ocean currents. Hearn et al. (2013) recognized the small sample size used in the analyses, the “considerable variation among individuals in the length and time spent at the surface”, and that data recorded by Argos satellite were intermittently with gaps of several days in some of the tagged animals. We cannot explain technically why transmissions in our longest records were at the same time on every day, and recognize this potential caveat. However, our analyses open the door to an alternative interpretation that could take more into account the animal’s diving behavior (surface, vertical) while migrating or provisioning (feeding), and the lack of information on the movement and habitat use of whale sharks away from shallow coastal areas (Brunnschweiler et al., 2009). Graham et al. (2006) demonstrated that whale sharks could display “ultradian, diel and circa-lunar rhythmicity of diving behavior” during feeding. In addition, daily provisioning patterns change significantly during the day, apparently in response to changes in temperature and perhaps the size of the individual (Araujo et al., 2020).

Consequently, we argue that it is not easy to discern from the number and frequency of ARGOS transmissions alone the movement of whale sharks in an open ocean highly influenced by current, eddies, and sea surface height. Especially if it is considered that our data was properly filtered to correct potential satellite errors, went through two modeling processes (Bayesian and Markov models) to understand and interpret their trajectories and behaviors, and these trajectories, in turn, were analyzed with each individual’s velocity information, current velocity and current velocity within the eddies. Hearn et al. (2018) showed floating tags drifting into open ocean at very different speeds and following an unmeasured current pattern into an eddy; we instead, measured animal movement in real-time and no relationship with surface currents and surface wind were observed, as previously reported for the species (Sleeman et al., 2010).

Previous studies have shown that the distribution and aggregation of R. typus are associated with physical and biological oceanographic features relating to productivity (Stevens, 2007). Indeed, our results show that most inferred foraging occurred in shallower coastal waters, with a significant use of seamounts and ridges, and was significantly related to productivity and chlorophyll-a concentrations. This is in accordance with findings from other locations, where seasonal R. typus aggregations are typically associated with temporary increases in food resources (Stevens, 2007). In the tropical eastern Pacific, productivity is strongly influenced by the upwelling systems. Ryan et al. (2017) examined R. typus movements in relation to thermo-biological frontal systems in the eastern Pacific and found a strong association between individual movements and the northern equatorial upwelling front, with 80% of shark positions occurring within the front. The same authors observed that seasonal movements between the equatorial and eastern boundary upwelling systems linked to productivity. Thus, our data indicate that R. typus in Panamanian waters exhibit similar foraging behaviors to individuals inhabiting the Pacific.

Migratory R. typus behavior has been linked to environmental factors such as water depth, SST, and eddy currents (Heyman et al., 2001; Riley et al., 2010). In our study, R. typus spent more time in eddies while migrating (17%) than foraging (10%), and when inside an eddy, eddy speed was significantly related to migratory behavior. Additionally, the two individual sharks that spent the most time within eddies were also those that spent the highest proportion of their time migrating (83% and 73%). Indeed, mesoscale eddies are important drivers of ocean ecosystem structure and are foraging hotspots for many marine animals in the eastern Pacific (Willett et al., 2006; Della Penna and Gaube, 2020). Eddy formation can drive the vertical movement of nutrients and phytoplankton and trap productive waters (Della Penna and Gaube, 2020). This may encourage R. typus to follow eddies during their migration to find food sources. Both anticyclonic eddies, which are associated with positive sea level anomalies (in the Tehuantepec and Papagayo eddy areas) and cyclonic eddies, which are associated with negative sea level anomalies (in the Humboldt Current), are linked to R. typus tracks (Revelles et al., 2007). Therefore, R. typus movement may not be specific to eddy type but more influenced by eddy speed. It has been suggested that the temperature gradients associated with eddies can be used for navigation by whale sharks (Sleeman et al., 2010). This might explain the observed association between higher eddy speeds and migratory behavior. However, it is important to note that seasonal and inter-annual variability in oceanographic conditions is characteristic of the eastern tropical Pacific, which may account for some of the variation between individuals and between regions.

R. typus visited 17 MPAs within five countries, yet only 22.6% of the modeled locations occurred within an MPA. This may be considered an underestimation due to the ongoing increase of MPAs in the region. While MPAs are a valuable conservation tool, their effectiveness may be somewhat limited for highly mobile species (Norman, 1999; Reynolds et al., 2017). For migratory species such as R. typus, transboundary MPAs and MPA networks are especially important conservation approaches, as are efforts to protect oceanic areas (such as Cordillera de Coiba) and areas beyond national jurisdiction. In some areas, MPAs have been established to protect seasonal and year-round R. typus aggregations, such as South Ari Atoll in the Maldives and Ningaloo Reef in Australia. Additionally, the Marine Conservation Corridor of the Eastern Tropical Pacific (MCCETP) between the islands of Cocos, Galapagos, Malpelo, and Coiba was formally established in 2004 for the conservation and sustainable use of biodiversity in the equatorial eastern Pacific. Our observations highlight the importance of this area for R. typus. Adaptive management approaches and the use of dynamic MPAs that incorporate spatially dynamic ocean features, such as thermal fronts, are also important considerations given the importance of these features for R. typus and ongoing shifts in species distributions (Hooker et al., 2011).

Injury and mortality because of interactions with fishing gear, including long lines (Wang et al., 2021), have emerged as key threat to many marine species, especially sharks and rays (Stevens et al., 2000). In our study, tagged R. typus transmitted from low-, medium-, and high-effort fishing areas, although approximately half of the modeled locations occurred from high-fishing-effort areas. The observed overlap with the tuna purse seine fishery is perhaps unsurprising given that purse seine fishing occurs more frequently in areas with high net primary productivity (Kroodsma et al., 2018), and tuna are often associated with other large species. Such fishing activities close to R. typus can lead to accidental capture and entanglement. Capietto et al. (2014) examined the spatial-temporal distribution of R. typus and fishery interaction hotspots in the Atlantic and Indian Oceans and recorded the number and fate of captured animals. They found high rates of incidental capture but low mortality because of these interactions. In contrast, high rates of mortality associated with fishing activities have been recorded in the Pacific Ocean (Western and Central Pacific Fisheries Commission, 2010). While we assessed fishing effort based on 70,000 out of 2.9 million motorized fishing vessels, small vessels (< 24 m) not equipped with an AIS system were not included (Kroodsma et al., 2018) despite also posing a threat. While there is a clear overlap between R. typus habitat use and fishing areas, long-term assessments of post-release survival are needed to obtain a clearer picture of the true impact of fishing on R. typus populations in the eastern Pacific (Capietto et al., 2014). Bycatch of juveniles and neonates in Peruvian waters is a clear example of the impact of fisheries on the species at any maturity stage (Pajuelo et al., 2018).

As slow-moving surface dwellers, R. typus are vulnerable to collision with marine vessel traffic, and there is a high degree of global overlap between shipping activity and habitat use (Pirotta et al., 2019; Womersley et al., 2021; Womersley et al., 2022). Based on our observations, most modeled locations from tagged individuals occurred within high-impact vessel traffic areas, that is, a high density of vessels moving at high speed. Previous studies have found that individual R. typus spend approximately half of their time in surface waters (< 10 m depth) (Eckert and Stewart, 2001; Rowat and Gore, 2007; Womersley et al., 2021), and interactions with ships can result in behavioral modification, disturbance, displacement, injury, and even mortality (Pirotta et al., 2019; Womersley et al., 2021). Given the spatial overlap between marine vessels and R. typus habitat use, injury and mortality from ship collisions could be a significant factor in observed population declines (Womersley et al., 2022). Additional research is, therefore, needed to elucidate the extent of such interactions and their effects on R. typus fitness, survival, and, ultimately, population status.

Overall, this study shows that the coastal waters of Panama form an important foraging and migratory habitat for R. typus from which nomadic individuals move to other key areas in the eastern Pacific and the Indo-Pacific. Our observations highlight the need for further conservation of these habitats and the R. typus populations they support. Furthermore, we found that oceanographic processes are key determinants of R. typus distribution and behavior, primarily via influences on productivity, making R. typus vulnerable to changing climate conditions that can increase, water temperature, El Niño-Southern Oscillation events, ocean stratification and reduce productivity (Sequeira et al., 2014; Fiedler and Lavín, 2017; Riascos et al., 2019).

The high degree of overlap between R. typus habitat and human activities highlights the need for marine conservation planning approaches beyond MPAs at key aggregation sites, and the insights gleaned from this work can inform conservation planning at appropriate scales. More generally, our study highlights the value of satellite tracking studies for understanding the behavior, habitat use, and potential adaptation to climate change of highly mobile migratory species, such as R. typus.

Geopolitics have an impact in the conservation of transboundary species including marine protected areas where conflicts can be amplified (Harrison et al., 2018; Mackelworth et al., 2019). Our results suggest that this transboundary species could benefit from a regional conservation plan that includes high seas or international waters to improve its protection. Local regulations and management practices are in place in most countries along with the eastern Pacific offshore islands and some coastal areas, including fishing and product commercialization prohibitions and promoting ecotourism. Establishing a functional marine corridor or seaway (Migravia) was suggested (Guzman et al., 2021), and a plan for connecting EZZ waters of Ecuador, Colombia, Panama, and Costa Rica is in progress, but Peru must be included. The only existing corridor politically was established in 2014 along with six Central American countries, including Panama, Costa Rica, Nicaragua, Honduras, El Salvador, and Guatemala (OSPESCA, 2014). However, the protection of large high-seas areas, perhaps used along the shark’s migratory routes for feeding and breeding as observed in this study, is not considered. We recommend the following actions: 1) Create a regional science-policy task force to coordinate the conservation of whale sharks in the eastern Pacific (Mexico-Peru minimum); 2) Develop marine territorial planning identifying critical aggregation areas potentially used for feeding and breeding; 3) Analyze migratory routes coupled with marine currents and vessel traffic routes to identify potential risk areas; 4) Increase satellite tagging efforts in the high-seas and coastal areas to understand the movement ecology of the species better.