A total of 178 adult female Kemp’s ridley sea turtles were outfitted with satellite transmitters after nesting at each of the major nesting beaches in the western GoM (NMFS and USFWS, 2015) between 2010 and 2017: Padre Island National Seashore, Texas, United States (PAIS, n = 76, 2010–2017); Rancho Nuevo, Tamaulipas, Mexico (RNMX, n = 44, 2010, 2011, 2014–2016); Tecolutla, Veracruz, Mexico (VCMX, n = 58, 2012–2017) (Table 1 and Supplementary Table S1). Each individual received a platform terminal transmitter (PTT) manufactured by Wildlife Computers (SPOT: n = 32, 2013–2015 or SPLASH: n = 24, 2010–2013) or Sirtrack (Kiwisat 101: n = 29, 2011–2013 or Kiwisat 202: n = 93, 2014–2017) (Table 1). Straight carapace lengths (nuchal notch to posterior tip, SCLs) were obtained at the time of PTT attachment for all turtles using straight, metal calipers or converted from curved carapace lengths (CCLs) using the equation developed by Schmid and Witzell (1997). PTTs were attached using the methods described in Shaver et al. (2013).

Platform terminal transmitters were programmed in the following ways: continuously on (n = 44, 2010–2014), 24-h on/24-h off (n = 10, 2013) and 6-h on/6-h off (n = 75, 2010–2017). PTTs transmitted data using the ARGOS system, which estimates each location using the following classes: 3, <250 m; 2, 250 to <500 m; 1, 500 to <1500 m; 0, >1500 m; A and B, unknown; Z, failed plausibility tests (CLS, 2011). Six PTTs deployed in 2010 were processed using least-squares analysis. The remaining PTTs were processed using the Kalman filter, which provides improved accuracy and increases the number of estimated positions (Lopez and Malardé, 2011). Kalman filtering was unavailable for the six PTTs deployed in 2010 processed using least-squares analysis. To ensure there were enough locations to identify primary foraging regions, only individuals tracked for at least 30 days after the mean last date of nesting for each specific nesting beach were used in analyses. Any tracks eliminated from analysis were confirmed to still be within or near the inter-nesting area at the last transmission. Mean last date of nesting was calculated separately for each of the three beaches using the last nesting date of each beach for each of the PTT deployment years to account for temporal differences in nesting seasons for each nesting beach. Eleven individuals were recaptured and tracked up to three times. In these instances, all tracking data were used if tracking durations were long enough. All activities were carried out according to protocols approved by the National Park Service Institutional Animal Care and Use Committee.

A switching state-space model (SSM) was used to estimate locations and behavioral modes of each individual to determine area-restricted-search-type movements (ARS; i.e., “foraging” or “inter-nesting”) and migratory-type movements (i.e., “exploratory” or “transit”). Because satellite-location data are often received at irregular time intervals and can contain large positional errors and temporal gaps, SSMs provide a means to estimate the most likely movement patterns of an animal and account for these errors while using the specific dynamics of a species’ movement patterns (Jonsen et al., 2005). Location data for each individual were first fit with a continuous-time correlated random walk (CTCRW) model to predict temporally regular locations in R using the “crawl” package (Johnson, 2018; R Core Team, 2019) with an initial value centered on their release location. The CTCRW model was fit using a prior distribution and estimated the location error from ARGOS estimates. For PTTs processed using least-squares, variance parameters were fixed using ARGOS provided error estimates for the three highest quality location classes (3, 2, and 1) to estimate the error parameters. For PTTs processed using Kalman filtering, ARGOS provided error ellipse information was used (McClintock et al., 2015). Locations for each individual were simulated at 6-h time-steps over 1,000 imputations. SSMs were then modeled using a two-state continuous-time hidden Markov model using the R package ‘momentuHMM’ from the simulated CTCRW tracks over 1,000 imputations (McClintock and Michelot, 2018). Model parameters were set using the gamma distribution for step lengths and the wrapped Cauchy distribution for turning angles. Parameter values were set at 750 ± 200 m (ARS) and 2,000 ± 750 m (migratory) for step lengths and π ± 0.1 rad (ARS) and 0 ± 0.7 rad (migratory) for turning angles.

After defining movement types, migratory and inter-nesting movements, locations on land and those interpolated during tracking gaps ≥ 7 days were removed from the SSM-derived locations for foraging area analyses. Inter-nesting locations were determined from the ARS locations using the mean last date of nesting for each nesting beach. For individuals that remigrated back to the nesting beach, inter-nesting was considered to begin after a migration that coincided with the nesting season (March–July) (NMFS and USFWS, 2015). For dispersal analyses, migratory movements after the mean last date of nesting were retained in SSM-derived location data.

For PTT data that failed to converge using SSM, locations were filtered by removing locations on land, that exceeded 5 km/h, had turning angles greater than 25° or were in depths greater than 100 m. The 100 m isobath was used as greater depths have been shown to be biologically implausible for Kemp’s ridleys (Shaver and Rubio, 2008; Seney and Landry, 2011). Filtered locations were used to include those individuals whose data failed to converge using SSM, but met all other criteria (e.g., tracking duration), in dispersal analyses. Mean daily locations (MDLs) were then calculated from SSM-derived and filtered locations using a custom script in R to normalize the data for all analyses. MDLs represented the centroid location of all SSM-derived or filtered locations for each day locations were available.

To determine primary foraging regions, the GoM was divided into two major areas: northern GoM (nGoM) and southern GoM (sGoM). The Exclusive Economic Zone (EEZ) boundary between the United States and Mexico demarcated the division between the nGoM and sGoM. The EEZ was chosen as the boundary as it provided a known, defined border between different environmental, anthropogenic, and political factors that may affect marine turtle turtles in the GoM. Each region included all marine waters from the coast out to the 100 m isobath in the corresponding direction from the EEZ boundary. The Atlantic coasts of Florida and Georgia were included as part of the nGoM to simplify the region.

A 25 km hex-bin grid was used to calculate the number of foraging days and identify areas of increased use throughout the Gulf using SSM-derived MDLs of ARS locations identified as foraging. This grid size was chosen as a compromise between improved data visualization and matching the spatial error of MDLs. A 10 km hex-bin grid was also investigated for comparisons to previous studies of female Kemp’s ridley movements (Shaver et al., 2016a, 2017; Hart et al., 2018a, b) and is presented in the Supplementary Material. The ratio of SSM-derived or filtered MDLs in each GoM region was also calculated for each individual to determine regional dispersal of females from the primary and major secondary nesting beaches throughout the Gulf. Individuals with ≥70% of filtered or SSM-derived MDLs within a region were assigned to that region. The percentage of the total adult female Kemp’s ridley population that forages in the nGoM and sGoM was then calculated using the relative proportion of the nesting population that each nesting beach represented. Relative proportions were calculated using the mean percentage of nests laid in each nesting region (Texas, Tamaulipas, and Veracruz states) between 2010 and 2017 as a proxy for adult female population size.

For individuals tracked multiple times and those tracked remigrating to the nesting beach, and then returning to foraging areas, tracks were compared between deployments and migrations to determine fidelity to Gulf regions. Utilization distributions (UDs) were then calculated using 95% kernel-density estimates (KDEs) for PTTs with SSM-derived foraging MDLs for all deployments, or during remigrations, to assess fidelity to specific foraging areas. A fixed-kernel least-squares cross-validation smoothing factor (h_cv) was used to calculate each 95% KDE. KDEs were compared for each turtle using the UD overlap index (UDOI) (Fieberg and Kochanny, 2005) and Bhattacharyya’s affinity (BAs) (Bhattacharyya, 1943). The UDOI and BA are statistics used to determine the amount of spatial overlap in three-dimensional UDs. Values of zero describe UDs with no overlap, while values of one indicate 100% overlap. The UDOI value can be greater than one if UDs overlap significantly, but are not uniformly distributed (Fieberg and Kochanny, 2005). The UDOI has been found to be the best estimator for describing the degree of space sharing and the BA is better suited for comparing the overall similarity between UDs (Fieberg and Kochanny, 2005); thus, both are reported here. KDEs, UDOIs, and BAs were calculated in R using the package “adehabitatHR” (Calenge, 2006). All other spatial analyses were conducted in ArcGIS 10.7.

Individuals measured between 58.0 and 75.2 cm SCL (mean ± SD: 63.9 ± 2.7 cm). Turtles were tracked from three to 1,554 days (mean ± SD: 368 ± 313 days) for a total of 65,466 tracking days. Overall, 150 PTTs (PAIS: n = 68; RNMX: n = 37; VCMX: n = 45) deployed at the primary and major secondary nesting beaches between 2010 and 2017 provided enough data to identify regional dispersal of individuals using SSM-derived MDLs or, for datasets that did not converge using SSM, filtered MDLs. Of these 150 PTTs, foraging areas were modeled for 117 PTTs (PAIS: n = 55; RNMX: n = 27; VCMX: n = 35) using SSM (Figure 1). Data from 33 PTTs did not converge using SSM and 28 PTTs had deployments too short in time for analyses. During the study period, 10 turtles were tracked twice, and one turtle was tracked three times from their respective nesting beaches. Eight of these individuals tracked multiple times had tracking durations long enough (≥30 days) for analyses. In addition, seven turtles were tracked migrating back to their nesting beaches during deployments. All remigrant turtles had tracking durations long enough for analyses. No individuals remigrated to nesting beaches other than the beach they were initially tracked from.

Gulf of Mexico regions were assigned to all PTTs that provided enough data for analyses (n = 150). In total, 41,591 foraging days were modeled with 86.1% in the nGoM and 13.9% in the sGoM. Both regions of the Gulf contained areas with high numbers of foraging days with hotspots near the Yucatán Peninsula, northern and southern Gulf coasts of Florida, including the Florida Keys, the Mississippi River Delta, and the Texas-Louisiana Shelf (Figure 2 and Supplementary Figure S1). Almost all PAIS turtles foraged in the nGoM with 99.8% of SSM-derived foraging days in the region and 98.5% of individuals dispersed to northern Gulf waters (Figure 3, Supplementary Figure S2, and Table 2). Only one individual tracked from PAIS (PTT 117512) migrated to the sGoM with its track ending offshore of the Yucatán Peninsula. Two turtles (PTTs 117515/152803 and 152804), one of which was tracked twice, foraged offshore of PAIS for the entirety of their tracking. One turtle (PTT 152808) foraged along the Atlantic coasts of Florida and Georgia (Figure 4). Individuals tracked from RNMX also primarily foraged in the nGoM with 90.7% of SSM-derived foraging days in the region and 83.8% of individuals dispersed to the nGoM (Figure 3, Supplementary Figure S2, and Table 2). Conversely, individuals tracked from VCMX were almost evenly split with 57.8% dispersing to the nGoM and 65.5% of SSM-derived foraging locations were located there (Figure 3, Supplementary Figure S2, and Table 2). No turtles from RNMX or VCMX foraged near the nesting beach they were tracked from.

All SSM-derived turtle tracks remained within the nearshore waters (≤100 m depth) of the GoM and Atlantic coast of the United States (Figure 1). In general, individuals foraged in either the nGoM or sGoM, but not both, and did not transition between regions over time to forage. However, there were two exceptions to this: (1) some individuals who remained within the coastal waters of southern Texas to forage and (2) individuals displaying short-duration (≤10 days) ARS-type behavior during migrations from nesting beaches in Mexico. One individual (PTT 152804) tracked from PAIS, which primarily foraged offshore of PAIS in the nGoM during its 387-day tracking period, briefly foraged in sGoM waters for 45 days before returning to its main foraging area off the south Texas coast. This individual had a maximum incursion distance into sGoM waters of 76.2 km (mean ± SD: 46.0 ± 25.7 km) from the EEZ border. Additionally, two individuals (PTTs 126252 and 126257) tracked from RNMX displayed ARS-type behavior within the sGoM near the United States/Mexico border (mean ± SD: 47.2 ± 24.4 km south of the United States/Mexico EEZ) during their migrations to nGoM foraging areas. These individuals undertook ARS-type behaviors in the sGoM during periods just after the completion of nesting seasons in Mexico (August–September). However, these individuals displayed ARS-type behavior for ≤10 days before continuing their migrations north into the nGoM where they remained for the duration of their tracking periods.

The majority of Kemp’s ridley nests (90.8%) were laid on the beaches of Tamaulipas, Mexico, concentrated at RNMX, between 2010 and 2017. The beaches of Veracruz, Mexico and Texas, United States accounted for relatively minor proportions of the nesting population with 8.2 and 1.0% of nests, respectively. Nesting in both of these states was concentrated at VCMX and PAIS. Using these proportions to weight the dispersal of tracked females from their nesting grounds to foraging areas indicated that 81.8% of the adult female population may use the nGoM as their primary foraging area (Table 2).

Eight of the 11 turtles that were tracked more than once (2X: n = 7, 3X: n = 1) had tracking durations long enough (≥30 days) during each deployment to assess regional fidelity over time. All eight individuals used the nGoM, following similar migratory corridors each time they were tracked, displaying strong fidelity to the region (Figure 5). One individual foraged offshore from PAIS during both deployments. All other turtles migrated away from PAIS. The mean time between repeat tracking was 4.0 ± 1.3 (mean ± SD) years.

Seven turtles were tracked remigrating to the nesting beach and then returning to forage during their deployments, including one of the turtles that was tracked twice. All individuals used the same region of the Gulf during successive migrations to their preferred foraging grounds, using the same migratory corridor each time (Figure 6). The mean time between remigrations to the nesting beach was 1.4 ± 0.5 (mean ± SD) years, but remigration intervals reported here only represent those individuals whose transmitters remained active long enough to capture a remigration during their tracking period. Seventy-six females were tracked ≥ 1 year, with seven of those tracked through a remigration, and 38 females were tracked for ≥ 1.5 years, with five of those remigrating within that period (Supplementary Table S1). Remigrating turtles in the nGoM initiated returns to the nesting beach during November and December while individuals returning from the sGoM began their migrations in February and March.

Individuals also showed fidelity to specific foraging areas. Twelve individuals that were tracked multiple times (n = 5), remigrated to the nesting beach and then back to foraging grounds during their initial deployment (n = 6) or both (n = 1) had data robust enough for SSM analysis to identify specific foraging areas and determine recurring use. For all of these individuals, overlap indices (i.e., UDOI and BA) indicated spatial overlap in their preferred foraging areas after each migration from the nesting beach (mean ± SD: 0.8 ± 0.7 UDOI; 0.6 ± 0.2 BA) (Figure 7). Turtles that remigrated during tracking displayed fidelity over successive nesting years (mean ± SD: 0.9 ± 0.8 UDOI; 0.6 ± 0.2 BA) and turtles that were tracked more than once displayed fidelity to foraging grounds over time (mean ± SD: 0.7 ± 0.6 UDOI; 0.6 ± 0.2 BA).

Preferential selection of nGoM waters as foraging grounds, fidelity to foraging regions, and narrow migratory corridors across the Gulf indicate that the adult female Kemp’s ridley nesting population may be more susceptible to regional threats and that the nGoM may be the most significant foraging region for females of the species. Importantly, point source events have the potential to significantly affect a high proportion of the nesting population in the foraging grounds or along migratory routes. For example, the Deepwater Horizon oil spill in 2010 may have had a population-wide effect on nesting females due to its centralized location within the nGoM, overlap with key foraging areas and lying directly within the migratory corridor (Beyer et al., 2016; Shaver et al., 2016a; Hart et al., 2018a). Due to its location, any individuals migrating to central or eastern nGoM foraging grounds, or returning to the nesting beaches, would have had to pass directly through the oiled area. In fact, it has been estimated that >50% of Kemp’s ridleys that forage in the region may have been exposed to oil (Reich et al., 2017) and it has been hypothesized that the spill may have affected, in part, the recovery of the species (Caillouet, 2014; Gallaway et al., 2016a, b).

The nGoM region is also used heavily by a variety of anthropogenic activities, including commercial fisheries, commercial shipping traffic, and oil and gas production (Hart et al., 2018a). Shrimp trawling has been linked with high incidents of marine turtle bycatch, with large numbers of mortalities of Kemp’s ridleys in the nGoM (Lewison et al., 2003; Finkbeiner et al., 2011) and the fishery has been plagued with compliance issues in the past (Lewison et al., 2003; Cox et al., 2007). Vessel traffic has been connected with marine turtle mortalities (Hazel and Gyuris, 2006; Casale et al., 2010) and mortalities in a host of other threatened mobile marine megafauna [e.g., North Atlantic right whales (Eubalaena glacialis) (van der Hoop et al., 2012) dugongs (Dugong dugon) (Marsh et al., 2011) manatees (Trichechus manatus) (Lightsey et al., 2006)]. Vessels have also been identified as a significant threat to marine turtles in the nGoM (Hart et al., 2018a). Hypoxic events and harmful algal blooms are becoming more common in the GoM and are predicted to increase (Brand and Compton, 2007; Justić et al., 2007) which can affect individuals, including marine turtles, and their prey resources (Landsberg et al., 2009; Capper et al., 2013; Walker et al., 2018; Foley et al., 2019).

Not only is a high proportion of the nesting population exposed to varying threats within their foraging regions, there are increasing threats along their migratory corridor, both in United States and Mexican waters. Illegal, unreported and unregulated (IUU) fishing using long-line and gill net gear, primarily targeting red snapper (Lutjanus campechanus) and sharks, may be the most significant inadequately addressed threat and marine turtles are frequently incidentally caught and killed by these operations. Large numbers of stranded marine turtles have been reported in south Texas, including adult female Kemp’s ridleys, during recent years, presumably a result of incidental capture from IUU fishing vessels, called lanchas, operating out of Mexico (Oliver and Jacobs, 2019; Donna Shaver, unpublished data). The U.S. Coast Guard reported seizing 74 of the 175 detected lanchas from south Texas waters in 2019, a new record, and have noted a significant increase of vessels since 2017 (USCG, 2019). Additionally, only an estimated 5–15% of total lancha incursions are detected annually, suggesting the total volume of illegal fishing activity in the GoM may be much higher than what has been reported (Frazer, 2020). These interactions have the potential to impact a high proportion of mature females as these operations occur directly in the primary migratory corridor leading from the nGoM to the major nesting beaches (Shaver and Rubio, 2008; Shaver et al., 2016a).