Sympatry or syntopy? Investigating drivers of distribution and co‐occurrence for two imperiled sea turtle species in Gulf of Mexico neritic waters

Abstract Animals co‐occurring in a region (sympatry) may use the same habitat (syntopy) within that region. A central aim in ecology is determining what factors drive species distributions (i.e., abiotic conditions, dispersal limitations, and/or biotic interactions). Assessing the degree of biotic interactions can be difficult for species with wide ranges at sea. This study investigated the spatial ecology of two sea turtle species that forage on benthic invertebrates in neritic GoM waters: Kemp's ridleys (Lepidochelys kempii) and loggerheads (Caretta caretta). We used satellite tracking and modeled behavioral modes, then calculated individual home ranges, compared foraging areas, and determined extent of co‐occurrence. Using six environmental variables and principal component analysis, we assessed similarity of chosen foraging sites. We predicted foraging location (eco‐region) based on species, nesting site, and turtle size. For 127 turtles (64 Kemp's ridleys, 63 loggerheads) tracked from 1989 to 2013, foraging home ranges were nine to ten times larger for Kemp's ridleys than for loggerheads. Species intersected off all U.S. coasts and the Yucatán Peninsula, but co‐occurrence areas were small compared to species' distributions. Kemp's ridley foraging home ranges were concentrated in the northern GoM, whereas those for loggerheads were concentrated in the eastern GoM. The two species were different in all habitat variables compared (latitude, longitude, distance to shore, net primary production, mean sea surface temperature, and bathymetry). Nesting site was the single dominant variable that dictated foraging ecoregion. Although Kemp's ridleys and loggerheads may compete for resources, the separation in foraging areas, significant differences in environmental conditions, and importance of nesting location on ecoregion selection (i.e., dispersal ability) indicate that adult females of these species do not interact greatly during foraging and that dispersal and environmental factors more strongly determine their distributions. These species show sympatry in this region but evidence for syntopy was rare.


| INTRODUC TI ON
Understanding how factors such as abiotic conditions, dispersal limitations, and biotic interactions (Soberón, 2007) influence species distributions is a central aim in the field of ecology. Abiotic conditions are often used as indicators of habitat suitability and may set the "fundamental" or "Grinnellian" niche (Hutchinson, 1957;Soberón, 2007), be used in species distribution modeling (Austin, 2002), and predict future impacts of environmental changes (Franklin, 2010). However, it has long been recognized that environment alone cannot fully explain species distributions and that biotic interactions (e.g., predatory/prey relationships, competition) determine the "realized" or "Eltonian" niche within the wider environment-defined niche (Hutchinson, 1957;Soberón, 2007). Species distribution models are beginning to be improved by incorporating biotic interactions (Kissling et al., 2011;Pollock et al., 2014) but assessing the degree of sympatric biotic interactions at a site can be more difficult than classifying environmental variables, especially for species with wide ranges at sea.
Loggerheads and Kemp's ridleys, however, both forage primarily in shallow waters on benthic invertebrates (Bjorndal, 1997;Shaver, 1991), making competition or a syntopic overlap between these two species plausible. Additionally, Kemp's ridleys and loggerheads are found throughout GoM neritic waters, and both are found in the GoM during every stage of their lives (Lamont, Putman, Fujisaki, & Hart, 2015). They nest on sandy GoM beaches: Kemp's ridleys nest from approximately April to July primarily on the coasts of Mexico and Texas but also sporadically in other Gulf states including Alabama and Florida, and loggerheads nest from April to September along the entire GoM coast from Mexico to the Dry Tortugas, Florida.
Here, we use satellite tracking data for Kemp's ridley and loggerhead sea turtles to assess how they distribute themselves across GoM neritic waters and what drivers may be important to their distribution. We delineate home ranges, quantifying the degree of species spatial and temporal co-occurrence. To assess abiotic drivers, we characterize habitat variables and foraging ecoregions and compare them across species. Together, these tools help us understand and compare how these species utilize nearshore GoM waters, what factors influence this distribution, and whether these two imperiled species show evidence for sympatry or syntopy in this region.

| Turtle capture and tracking
We tagged Kemp's ridley and loggerhead females ( Figure 1) with satellite transmitters after they nested on beaches along the GoM (Table 1, Figure 2). Tagging, including attaching platform terminal transmitters (PTTs), followed our previous studies (Hart et al., 2015;Shaver et al., 2016)

| Determining foraging behavior with statespace modeling
Switching state-space modeling estimates location and behavioral mode at regular time intervals, accounting for satellite positional F I G U R E 1 Satellite-tagged loggerhead (Caretta caretta) female heading back to the water after nesting (left panel) and a female Kemp's ridley (Lepidochelys kempii) during nesting (right panel). Images taken with permission (MTP 176 issued to KH and USFWS permit TE840727-3 issued to DS) under conditions not harmful to turtles errors and dynamics of animal movement patterns (Jonsen, Flemming, & Myers, 2005). We conducted SSM following our previous studies (see Hart et al., 2014;Shaver et al., 2016 and Supporting Information Appendix S1 for additional information on this technique) to determine dates of each turtle at its foraging ground(s). The behavioral mode was binary, defined as "migration" or "foraging" in earlier applications (Breed, Jonsen, Myers, Bowen, & Leonard, 2009;Jonsen et al., 2005;Jonsen, Myers, & James, 2007); here, we deemed the modes "transiting" or "area-restricted search" (ARS; Kareiva & Odell, 1987). Since we tagged animals during nesting seasons, we split ARS into "foraging" or "internesting" and considered "transiting" to be movements between these ARS modes. We considered foraging to occur after a migration, unless high-quality locations on land indicated the turtle was still in the internesting period.
TA B L E 1 Tagging locations and years for Kemp's ridley (Lepidochelys kempii) and loggerhead (Caretta caretta) sea turtles in the Gulf of Mexico  (Fritts et al., 1983;Seney & Landry, 2011;Shaver & Rubio, 2008) and loggerheads generally stay within waters of the continental shelf (−200 m, Hawkes et al., 2011) so we also filtered out locations in waters deeper than −200 m (neritic zone cutoff, but see Supporting Information Appendix S1 for exceptions). Some turtles were recaptured and tagged twice. When two tracking periods were available, we used only the longest tracking period.
All Kemp's ridleys entered a migration mode, but two loggerheads did not, despite being tracked for more than 150 days (i.e., their foraging area was near their internesting area). We averaged the first month and day of the first foraging period for the other loggerhead turtles in order to calculate a mean foraging start date, 30 July, which we used as the beginning of the foraging period for these two turtles. fixed-kernel least-squares cross-validation smoothing factor (h cv ) for each KDE (Seaman & Powell, 1996;Worton, 1995). We used aRcgIs 9.3 (ESRI, 2007) to cal cul ate the in-water area (km 2 ) within each kernel density contour (50% for core-use and 95% for home range; Hooge, Eichenlaub, & Hooge, 2001) and to plot the data. Similar to KDEs in our previous studies, we also tested for site fidelity within foraging areas using the anImal movemenT analysIs exTensIon foR aRcvIew 3.3 (ESRI, 2002; see Supporting Information Appendix S1). We bounded the range for random walks from −200 m to 0 m bathymetry to include only the realistic extent of the in-water habitat for our animals during the study period; however, we smoothed out the shoreline with a 5 km buffer to account for many small bays and points close to land. We used a 10 km grid to show the number of individual KDEs per cell and where Kemp's ridleys and loggerhead KDEs co-occurred.

| Home range comparisons and overlaps
We also separated the loggerhead KDEs into their two nesting subpopulations to show any spatial overlap with where these turtles forage.
Further, to summarize where and when the two species co-occurred, we determined how many 95% KDEs and individual turtles overlapped in both space and time. Overlaps in time were calculated using the entire date ranges of spatially co-occurring 95% KDEs and counting any days of overlap, regardless of the percent of spatial overlap of the individual KDEs.
To characterize at-sea foraging areas selected by individual tur-

| Abiotic factors at home ranges
Environmental conditions are highly variable throughout the GoM.
Although our previous studies highlighted foraging hot spots of Kemp's ridley (Shaver et al., 2013) and loggerheads in the GoM (Hart et al., 2014), it has been unclear whether the two species occupy similar habitat in a multidimensional space. To answer this question, we characterized the foraging location (50% KDE centroid) of each turtle using spatially explicit habitat data and we tested habitat similarity between the two species using permutational multivariate analysis of variance (MANOVA).
We derived mean, minimum, and maximum sea surface temperature (SST) for the period outside the nesting season for both species As a preliminary analysis, we first conducted a Wilcoxon rank sum test to compare the two species with respect to six habitat variables including both spatial and environmental factors (latitude, longitude, distance to shoreline, SST, NPP, and bathymetry). This analysis was conducted to eliminate nonsignificant variables if there were any. We then visually examined a plot of the first two principal component scores using a set of selected variables based on the preliminary analysis. We used the "vegan" package of R (Oksanen et al., 2013) to conduct the permutational MANOVA.
We also conducted a Kruskal-Wallis test to compare foraging centroid bathymetry by tagging locations for each species. For each turtle, we used the first satellite location from the SSM input file as a proxy for the tagging locations. Exceptions were made if the first point was not near the known tagging location. If this was the case, the next point that was near the tagging location was used. For turtles with multiple KDEs, we used only the most recent KDE centroid, so that each turtle was represented by one centroid.

| Ecoregion selection
The GoM is comprised of diverse marine ecoregions, each with unique ecological properties. We used foraging centroids for each turtle to determine their foraging ecoregion, based on the marine ecoregions delineated by Wilkinson et al. (2009). To understand the factors that dictated the selection of foraging regions of individuals, we used a categorical regression tree (CART) analysis to predict the foraging ecoregion of individual turtles from biotic (species and SCL) and spatial factors (tagging site).

| Co-occurrence of foraging home ranges
The 95% KDEs of both species showed spatial overlap throughout the GoM, including off the coasts of all U.S. Gulf states, as well as on the western shore of the Yucatán Peninsula ( Figure 3). The co-occurrence areas included 11 individual loggerheads (12 KDEs) and 16 Kemp's ridleys (20 KDEs  July-November (Table 3). For these combinations, the time each KDE overlapped ranged from 2 to 94 days (mean ± SD = 34.3 ± 32.5 days) for a total of 206 days across all combinations. These six unique space and time co-occurrence areas ranged in size from 18 to 191 km 2 (mean ± SD = 78.6 ± 73.0 km 2 ) for a total of 471.7 km 2 . Looking at each turtle pair that showed some overlap, these space and time co-occurrences represented only a portion of the two home ranges involved: 0.1%-12.8% (mean ± SD = 3.9% ± 4.6%; Table 3  13.5 m) (  Figure 6a); therefore, we included all variables in the analysis. There were some moderately high to high bivariate correlations between habitat variables, including correlations between latitude and SST variables (|r| > 0.9) and between NPP, bathymetry, and distance to shoreline (|r| > 0.6).

| Habitat similarity and ecoregion selection
The first two principal component (PC) scores from all habitat variables explained almost 100% of variability contained in the data. The scatter plots of the two PC scores indicated a considerable overlap in ecological niche space between the two species, yet overall loggerheads occupied a larger niche space that also contained the niche  (Figure 7), accounting for 66% of variable importance followed by SCL (19%) and species (17%).

F I G U R E 4 The number of individual turtles with home ranges (95% Kernel Density Estimates [KDEs]) in 10 km grid cells throughout the Gulf of Mexico. Both Kemp's ridley (Lepidochelys kempii)
and loggerhead (Caretta caretta) KDEs are included: If a turtle had more than one KDE, these were merged so that every turtle was only counted once throughout all grid cells F I G U R E 5 The number of subpopulations with intersecting home ranges (95% Kernel Density Estimates) in each 10 km grid cell throughout the Gulf of Mexico: Turtles from the same subpopulation were combined and could include loggerheads (Caretta caretta) tagged in Dry Tortugas National Park, loggerheads tagged in the Northern Gulf, or Kemp's ridleys (Lepidochelys kempii), for a possible maximum of three subpopulations represented per grid cell. However, we did not find all three in any grid cell. Therefore, dark blue cells indicate where the two loggerhead subpopulations co-occur, and red indicates where Kemp's ridleys co-occur with a single loggerhead subpopulation

| D ISCUSS I ON
Recent meta-analyses have found that species co-occurrence is generally less than expected by chance, although presence-absence matrices specifically for herpetofauna found less structure than for homeotherms (Gotelli & McCabe, 2002;Ulrich & Gotelli, 2010). For decades, it has been known that Kemp's ridley and loggerhead turtles co-occur throughout their range in the Gulf of Mexico (Fritts et al., 1983;Hildebrand, 1982;Márquez, 1990;Rabalais & Rabalais, 1980) and that both species inhabit and forage in nearshore waters (Lewison, Crowder, & Shaver, 2003;Márquez, 1994;Plotkin, Wicksten, & Amos, 1993;Shaver, 1991). We found areas of co-occurrence existed for two sea turtle species in GoM neritic waters, but the majority of foraging KDEs were located in separate locations. Kemp's ridley foraging home ranges were concentrated in the northern GoM, whereas loggerhead foraging home ranges were concentrated in the eastern GoM. In addition, areas where the two turtle species co-occurred made up a small proportion of the total home range area (1.1%) and for the six unique KDE combinations that co-occurred in both space and time, co-occurrence was minimal (mean 3.9% area overlap and mean 34.3 days). Our results fall in line with the meta-analyses, as we did not find a high degree of species overlap.
Factors determining the co-occurrence (or lack thereof) of species in natural communities have long been a subject of debate. Diamond (1975) introduced "assembly rules" that considered competition to be a major driving force of species distributions. However, these rules have been challenged as null models were able to predict similar habitat distributions in the absence of interspecific competition (Connor & Simberloff, 1979;Gotelli, 1999),

| Prey resources
Although studies have shown that Kemp's may forage opportunistically (Witzell & Schmid, 2005), those studies focused almost exclusively on immature Kemp's ridleys and differences may occur among life-stages with adults becoming more specialized on crab consumption (Shaver, 1991 Shaver, 1991). Shaver (1991) suggested that Kemp's ridleys forage within −50 m water depth based on gut contents and the distribution of crabs in the GoM: Both of these species are common in the northern GoM and found almost exclusively in shallow waters less than 70 m deep (Powers, 1977). Our foraging centroid locations support this as- Loggerheads also forage on crabs, but they are considered generalists (Vander Zanden, Bjorndal, Reich, & Bolten, 2010) therefore they may utilize deeper water to forage on species such as whelks or fish. Additionally, Kemp's ridley home ranges were significantly and substantially (~10×) larger than loggerhead home ranges. The wider breadth of prey on which loggerheads forage may allow them to remain in a narrower area, whereas Kemp's ridleys may search for their specific prey item resulting in a larger home range than loggerheads.
Another consideration in reviewing the prey resources of these species is that the northeastern GoM represents a dividing F I G U R E 6 (a) Box plots of six habitat variables (latitude, longitude, distance to shoreline, daytime mean sea surface temperature, and net primary production) at an identified foraging location ( Kemp's ridleys are found throughout the GoM (Eaton et al., 2008), indicating that crabs may play an important role in regulating adult distribution (Shaver, 1991).

| Dispersal effects
We found that the nesting beach was the most important variable we tested in predicting foraging ecoregion for turtles in the show continual switching between migration and foraging modes as they move along the GoM coast (Hart, Lamont, et al., 2012;Shaver et al., 2013). Perhaps dispersal limitations are more important for loggerheads that do not repeatedly interrupt migration to replenish energy reserves. Additionally, dispersal (i.e., migration) to and from nesting grounds may impact temporal co-occurrence on foraging grounds. Kemp's ridleys complete their annual nesting season earlier than loggerheads (July vs. August, respectively), so the timing of Kemp's arrival at the foraging grounds would be expected to occur earlier than loggerheads.
The distribution of sea turtle foraging areas in the Mediterranean is influenced strongly by the timing and distance of hatchling dispersal based on passive drift (Hays, Fossette, Katselidis, Mariani, & Schofield, 2010), and this type of dispersal may also play a role in the patterns we observed in the GoM. However, genetic evidence demonstrated that directed swimming by surface-pelagic juvenile green turtles in the GoM, and not just passive drift, contributed to their ultimate foraging destinations (Shamblin, Witherington, Hirama, Hardy, & Nairn, 2018 juveniles showed that these turtles were not passively drifting but actively swimming (Putman & Mansfield, 2015

| Environmental effects
Establishing important environmental factors for foraging habitat can help explain species distribution (the fundamental niche; Hutchinson, 1957;Soberón, 2007)

| Spatial dynamics
A metacommunity can be defined as a group of local communities with interacting species linked by dispersal (Leibold, et al., 2004). While we primarily saw loggerheads and Kemp's ridleys on east and west sides of the GoM respectively, a few individuals did forage on "opposite" ends, indicating that they can be linked by dispersal and that appropriate "patches" occur across the GoM for both species. Interaction in the foraging areas is also plausible because of their similar dietary requirements and some empirical evidence of spatial overlap. The species sorting perspective of metacommunity theory states that environmental heterogeneity influences species responses and that habitat (i.e., patch) quality affects local composition in combination with dispersal (Leibold, et al., 2004). We are drawing a link between turtle presence and assumed community type based on general knowledge of their spatial locations and similar ecological roles in the GoM. That community types occupied by each species may be ecologically similar for turtles is also supported by their niche space overlaps. However, we did find that environmental variables were an important factor in determining occurrence across foraging sites, indicating patch quality may influence local composition. Only limited, coarse-scale benthic habitat data are available for these foraging areas (i.e., "mud," "sand," "gravel," or "rock"); however, we saw evidence that there may be differences in patch quality for the different species such that loggerheads may prefer sand while Kemp's ridleys may prefer mud. We feel that this example involving marine megafauna represents one of a few empirical tests of this organizing concept. Additionally, the few areas of overlap may provide sites with a heterogenous environment allowing for co-occurrence and further habitat studies in these areas are warranted. We observed this at least for sediment types, with the intersect area having both sand and mud. Future habitat modeling with finer-scale benthic resolution as well as prey abundance surveys would be valuable for understanding habitat differences for these species. Regardless, these areas represent clear long-term and persistent locations of high-use zones for key imperiled species.

| CON CLUS ION
We build on previous work delineating foraging areas for Kemp's ridleys and two subpopulations of loggerheads across seven nesting sites in the GoM and demonstrate locations where two imperiled sea turtle species co-occur during their foraging periods. However, our analysis was restricted to adult females. Determining habitat-use and whether similar drivers affect the distribution for males and juveniles will be important for the conservation of these species.
Unless habitats and dispersal for two species are similar, parsing out the importance of competition may not be possible, as distribution may be due more to changes in habitat or dispersal ability (Ulrich & Gotelli, 2013). It is possible that Kemp's ridleys and loggerheads may compete for resources; both species have been documented to forage extensively on crabs (Burke, Standora, & Morreale, 1993;Wallace, Avens, Braun-McNeill, & McClellan, 2009). However, the separation in foraging areas (based on centroid locations and low level of direct overlap), the significant differences in environmental conditions at foraging grounds, and the importance of nesting location on ecoregion selection (i.e., dispersal ability) indicate that adult females of these species do not interact greatly during foraging, that they show little syntopy, and that dispersal and environmental factors more strongly determine their distributions within their shared GoM region.
Analyses on plant and mammal communities across geological time have shown that aggregated species were more common before the increase in human population during the Holocene, suggesting that current evidence for sympatry and syntopy may also be impacted by human activity (Lyons et al., 2016). As these two sea turtle species evolved about 3-6 million years before this Holocene shift (Bowen, Meylan, & Avise, 1991), it is possible that the distributions we see today might be more constricted than in the past.
Anthropogenic impacts in the GoM are ongoing and across 19 modeled stressors, the highest were for ocean acidification, sea surface temperature, sea level rise, ultraviolet anomalies, pollution, and shipping (for 2013; Halpern et al., 2015). Therefore, continuing to investigate how anthropogenic pressures influence species' habitat selection will be a key addition to understanding drivers of distribution for these imperiled sea turtles. Kemp's ridley turtles. We thank those that aided with formatting data for spatial analyses. We acknowledge the use of the Satellite Tracking and Analysis Tool (STAT) and telemetry data generated as part of the ongoing Deepwater Horizon NRDA (publicly available from www.seaturtle.org). Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

CO N FLI C T O F I NTE R E S T
None declared. I. led the writing. All authors contributed critically to the drafts and gave final approval for publication.

DATA ACCE SS I B I LIT Y
Raw data are exempt from publication due to sensitivity of endangered species location information. All other data used for analyses are presented in the paper or in the Supporting Information Appendix S1.