Big and Fast: GPS Loggers Reveal Long-Range Movements in a Large, Riverine Turtle

Abstract

Rio Grande Cooters (Pseudemys gorzugi) occupy the Rio Grande watershed and have among the smallest ranges of all North American freshwater turtles. Anthropogenic dewatering is considered to have caused range contractions and population declines. We sought to facilitate management recommendations by determining the extent of movement and potential associations with extrinsic cues. We conducted a GPS-enabled telemetry study from August 2015 to May 2017 on the Devils River in Texas, USA. We included Capture–Mark–Recapture data from 2011, 2014, and 2015–2018 to determine population status in conjunction with movement ecology. Turtles showed increased movement as streamflow and water depth increased. Larger movements were also made mid-year, coincident with the peak nesting season. We speculate that seasonality and increases in streamflow facilitate switches from slower, localized movement to transiting modes. We observed individual heterogeneity in transitory movements. Such movements led us to maintain our population estimate of 726 to 1219 individuals is representative of the entire Devils River. The extent of movement in P. gorzugi has been previously underestimated and long-range movements could explain observed genetic structure. Future efforts to re-establish natural flow regimes in the Rio Grande basin could potentially be the most effective management approach for this range-restricted chelonian.

1. Introduction

Movement in individual organisms is a fundamental life feature that plays a pivotal role in shaping biodiversity patterns via its effect on species range and population persistence [1,2]. Individual movements at multiple spatiotemporal scales are driven by the intrinsic motivation and ability of an individual organism to respond to external or environmental changes [1,3]. For instance, individual animal movements may vary spatially and temporally in response to seasonality—i.e., periodic fluctuations in environmental resources [4]. However human-mediated land-use and climate change may affect the rate and duration of such environmental fluctuations and, therein, the landscape structure [5]. The ability of individual animals to respond to alterations in the landscape they occupy determines whether movement parameters are adaptive [5]. Movements may be commonly categorized as foraging, dispersal, and migration, with each of these movement modes relating to a different spatiotemporal scale [6]. Studies that seek to elucidate environmental drivers of individual movement and shifts between movement modes [4,7] are, therefore, important, especially given the effects of broad-scale anthropogenic changes on threatened taxa.

Turtles constitute a major vertebrate group in which 47.9% of constituent species are listed as Threatened per the International Union for the Conservation of Nature (IUCN) criteria—i.e., categorized as Vulnerable, Endangered, or Critically Endangered [8]. These long-lived vertebrates show delayed sexual maturity and are prone to population collapses following increases in adult mortality [9,10]. The intrinsic vulnerability of such life-history traits to anthropogenic impacts may be further exacerbated when the species has a restricted range. With the 13th smallest range among 49 non-marine turtles in the United States, Rio Grande Cooters (Pseudemys gorzugi) are an example of rarity by geographic distribution [11,12].

These large riverine turtles are endemic to the Rio Grande (=Rio Bravo Del Norte) drainage system [13]. Specifically, the species is known from the lower Rio Grande watershed in Texas and northeastern Mexico and the Pecos River drainage in northwestern Texas and southeastern New Mexico [13,14]. The species has an apparent gap in its distribution on the Pecos River, from south of the New Mexico border to Independence Creek in Terrell County, Texas (Figure 1). This gap has been previously reported to extend 161 km in length and was attributed to nonpoint source pollution [13,14,15]. More recently, Mahan et al. [16], surveyed the lower Pecos River and failed to detect P. gorzugi over a ~390 km stretch of the river, wherein this absence was attributed to heightened conductivity levels. Additionally, this watershed has several dams and diversion structures to meet water supply needs for agriculture, expanding urban areas, and the largest oilfield in Texas. These water diversions have led to an intermittent flow and thereby reductions in the extent of the aquatic and riparian habitat [17,18]. Rio Grande cooters are suspected to have undergone reductions in range and declines in population and were petitioned (USFWS Docket No. FWS-R2-ES-2015-0061) to be listed as an endangered or threatened species under the federal Endangered Species Act [17,19]. However, the proposal was ultimately rejected [20].

P. gorzugi has been among the most poorly-documented turtles in North America [21]. However, P. gorzugi has gained research attention since the proposal for their federal listing under the Endangered Species Act [22,23,24]. Previous studies on movement [13,25] continue to be limited in number and indicate sedentary behavior in these turtles, with maximum recorded movements of 300 m. However, a range-wide genetic study indicated a panmictic population with several shared common alleles among study populations, indicating moderate-to-high levels of gene flow. Further, the study implied recent connectivity across the distributional gap between northern (i.e., New Mexico) and southern (i.e., Texas) populations [17]. More recently, genome-wide data revealed that populations of P. gorzugi in New Mexico and Texas are distinct from each other. However, a lack of genetic structure was observed in the species up to an extent of ~200–300 km [26]. Additionally, during CMR surveys conducted opportunistically at a single site on the Devils River in Texas, we observed a considerable number of new entrants, i.e., unmarked turtles, from one sampling occasion to the next, which indicated movement between sites. This finding led us to test for the occurrence of long-range movements which could begin to explain the lack of population differentiation in P. gorzugi. The overall goal of this work is to describe the movement patterns of Pseudemys gorzugi within the Devils River with concomitant insights on species demography.

Collecting high-resolution data on the movements of small vertebrates is more feasible now due to advances in battery performance and circuitry that have reduced the size and cost of data loggers enabled with global positioning systems (GPS). We expected most movements of P. gorzugi to cover a short distance with long-distance movements occurring less frequently [27]. Therefore, we used very-high-frequency (VHF) transmitters equipped with GPS data loggers to increase the spatial and temporal resolution of our location data and attempt to fulfil our primary objective of potentially detecting long-distance movements in adults. Additionally, we collected CMR data concurrent with visits to obtain telemetry data.

We attempted to explain variation in movements as a function of individual and environmental factors [2,28,29,30]. Specifically, we expected that increases in the streamflow and water depth would enable connectivity among sites on the river and, consequently, tagged turtles would be more likely to undertake longer, exploratory movements. Furthermore, the species exhibits size dimorphism with females attaining a larger body size relative to males [13]. We expected sex-biased movements with females moving greater distances than males, potentially to meet the energetic demands of a larger body size and egg generation. Additionally, we expected to observe longer movements associated with nesting forays during the peak nesting season in June [31,32]. Furthermore, we constructed home ranges for individual turtles by incorporating an approach that modeled space- and time-use simultaneously. This potentially enabled a characterization of space-use while also shedding insights on underlying behavioral mechanisms. Finally, we sought to provide information on the demography of the species using CMR data, in association with observed dispersal patterns. With this study, we sought to address knowledge gaps in the life-history of this rare species and collect data applicable to informing management as well as adding to the understanding of the ecology and habitat requirements of P. gorzugi across its entire range.

2. Materials and Methods

2.1. Study Site

Our study was conducted on the Devils River in Texas, USA (Figure 1; https://www.gbif.org/, accessed on 2 January 2021). The Devils River is listed as an Ecologically Significant Stream Segment by regional water planning groups in Texas, in part due to the threatened and endangered species and unique communities that inhabit this watershed (https://tpwd.texas.gov/landwater/water/conservation/water_resources/water_quantity/sigsegs/regionj.phtml, accessed on 2 January 2021). This river originates in Sutton County and flows intermittently southward into Val Verde County until its confluence with the Amistad Reservoir, an impoundment on the Rio Grande River near Del Rio, Texas. The Devils River lies almost entirely in an incised limestone canyon with spring discharges accounting for a higher baseflow [18].

Surface water flow on the Devils River is mostly maintained along a 64 km stretch from Baker’s Crossing to the Amistad Reservoir [17,18,33]. Wide, shallow riffles are often followed by deep pools with gravel and cobble bottoms, wherein larger deep pools may contain silt composed of detritus and fecal matter from sheep and goats washed in by rainfall. Such nutrient runoff may add to the productivity of pools. At approximately the confluence of the Devils and Dry Devils rivers, the river assumes lentic characteristics as a result of water impoundment in Amistad Reservoir [33].

The upper portions of the Devils River commonly have submergent and emergent aquatic vegetation. These include pondweed (Potamogeton sp.) and yellow cow-lily (Nuphar lutea) in pools, while horsetail (Equisetum sp.) and cat-tail (Typha sp.) are common marginal plants seen in the lower portions. Water-nymph (Najas sp.) and green algae such as Chara sp., Spirogyra sp., and Cladophora sp. are seen in spring areas, especially those frequented by livestock. Intermittent groves of pecan (Carya illinoiensis), willow (Salix sp.), live oak (Quercus virginiana), and sycamore (Platanus occidentalis) constitute riparian vegetation [33,34].

The Dolan Falls Preserve (DFP) and the Devils River State Natural Area (DRSNA) were our primary study sites on the Devils River (Figure 1). The DFP is a 1942 ha property owned and managed by The Nature Conservancy (TNC). The DRSNA consists of 8090 ha of mostly unmanaged land [34]. Dolan Creek, which is 20.1 km in length and dry most of the year, passes from the north end of the DRSNA and exits at the southwestern corner of the property. Dolan Creek flows through the DFP where it contains water year-round due to spring outflows [34]. While the river passes over many waterfalls that do not exceed 1.5 m in height, Dolan Falls spans the entire width of the river, and at a height of 4.6 m is considered the single greatest barrier to fishes found in the lower Devils River [33].

2.2. Telemetry Study

2.2.1. Turtle Capture and Transmitter Attachment

We used FLR V-Ultra-lightweight GPS backpack telemeters (Telemetry Solutions, Walnut Creek, CA, USA) to track movements of adult P. gorzugi from August 2015 to May 2017. The large and deep plunge pool below Dolan Falls and spring-fed macrophyte-rich pools at the DRSNA have year-round water with relatively high visibility that enabled us to snorkel and capture turtles by hand. P. gorzugi are medium-to-large sized turtles, that show sexual size dimorphism. Female turtles are larger than male turtles while males have elongated foreclaws and longer, thicker tails than females [13,14,15,17]. We used these traits to diagnose male and female turtles. The dimensions of our transmitters were 72 × 43 × 37 mm and weight was 90.2 g. These devices consist of a conventional VHF transmitter integrated with a data logger capable of taking a pre-programmed daily number of GPS position fixes. We attached transmitters to the six largest females and four largest males captured during sampling in August 2015 (

Table 1. Six female (F) and four male (M) Rio Grande Cooters (Pseudemys gorzugi) were tagged with GPS-enabled VHF transmitters and monitored from August 2015–May 2017 in the Devils River, Texas. We provide a summary of body size, weight, tagging duration, number of positional fixes, and maximum movement for each tagged individual. We did not recover data from two (ID three and ten) of the ten tagged individuals. NA—Not Available.

Turtle ID	Transmitter Frequency (MHz)	SCL (mm)	Weight (in g)	Sex	Number of Positional Fixes	Duration (in Days)	Maximum Net Displacement (km)
One	150.400	319	4210	F	215	330	3.13
Two	150.420	328	4430	F	382	606	10.28
Three	150.440	283	3290	F	NA	NA	NA
Four	150.460	279	2370	M	220	594	2.67
Five	150.480	291	3130	F	188	245	35.5
Six	150.500	320	3900	F	336	601	1.18
Seven	150.520	291	3240	F	386	640	5.31
Eight	150.540	251	2170	M	207	517	16.15
Nine	150.560	264	2310	M	171	565	10.97
Ten	150.580	267	2040	M	NA	NA	NA

).

Positional fixes were only possible at the water surface, so we constrained the number of times a GPS position was logged to optimize battery life. We programmed the data logger to take three positional fixes daily. Two fixes were scheduled during daylight hours at 10:00 and 16:00 h, when most aquatic or atmospheric basking was observed [35], while a third occasion was at 23:00 h to potentially capture nocturnal activity such as nesting.

We attached transmitters using a modified protocol that combined the ‘hole in shell’ and ‘adhesive’ methods [36,37]. The net weight of each transmitter including the weight of ancillary materials approximated 150 g, equivalent to 4.8% of mean body weight of tagged turtles ($\overline{x}$ = 3137 g; N = 10). Although this exceeds the generally accepted rule of 2% body weight, an extension of this ratio to 6–12% has been empirically supported [38].

2.2.2. Tracking

We visited our primary study sites at least four times a year between 2015 and 2017 to download positional fixes from telemetered turtles (Table 1). The timing of our visits was predicated on site access enabled by TNC. We used a TRX-1000S scanning receiver and a 3-element folding Yagi directional antenna (Wildlife Materials, Murphysboro, IL, USA) to manually locate animals. Access to high-ground overlooks were limited, so we actively tracked turtles from canoes. When within range, we attenuated transmitter signals to precisely locate turtles and subsequently snorkeled to capture turtles by hand. GPS data were then downloaded wirelessly when the transmitter was in range with a FLR-V base station (Telemetry Solutions, Walnut Creek, CA, USA). We also obtained passive downloads of positional fixes by securing our FLR-V base station on rocks that were known basking areas, potentially used by tagged turtles. Additionally, we conducted two telemetry surveys using a small fixed-wing aircraft to fly over the length of the river at low altitude to locate turtles suspected of moving large distances. Turtles discovered by aircraft surveys were subsequently located and captured for download of their positional fixes using canoes or motorized boats. We removed transmitters from turtles when the strength of the VHF signal appeared to decrease.

2.2.3. Movement Analysis

All statistical analyses were conducted using R v3.6.0 [39]. We removed positional fixes with precision dilution of position > 10, since this threshold offers a practical alternative between removing erroneous outliers and an acceptable amount of data reduction [40]. We sought to examine whether turtles made long-distance movements and relate these movements to environmental cues. We used the ‘adehabitatLT’ package to generate descriptive parameters of individual turtle trajectories [41]. For each turtle, we summarized the total of movements (hereafter, movements) between all successive relocations for each date using the ‘data.table’ package [42]. These movements were straight-line distances between consecutive points on the river channel. Turtles are likely to have taken longer routes, but we use straight-line distances as a minimum approximation. Unequal time intervals (i.e., number of days) underlie these movements, but we were interested in capturing the magnitude of movements between relocations rather than a daily metric [30]. These movements are ignorant to directionality, i.e., the summed distance across two potential movement steps may overestimate the net displacement. However, at a maximum of three daily positional fixes, we believe our large movements (≥95 percentile value) would be correlated with large net individual displacements. We used streamflow (cu. ft/s) and gage height (ft) data from a United States Geological Survey (USGS) instream gaging station in Dolan Creek above the Devils River as potential predictors of these long-distance movements. Missing values accounted for 8% of the USGS dataset as improbable values were automatically screened until they could be verified. We used the ‘ggplot2’ package to examine daily activity by individual turtles [43]. We visually inspected the relation of activity with ambient air temperature by plotting the number of monthly positional fixes with long-term monthly average air temperature [44]. We viewed the frequency distribution of movements and the relationship of these movements with streamflow by plotting the movement and streamflow data by date.

We conducted principal component analyses (PCAs) using the ‘stats’ package in R [39] to visually inspect the relationship of large total movements (i.e., ≥95 percentile value) with extrinsic cues such as streamflow (cu. ft/s), gage height (ft), day of year (1 to 365), and change in streamflow (i.e., difference in streamflow between consecutive positional fixes; cu. ft/s). Since we expected individual-level heterogeneity, we partitioned data by each turtle ID and ran a PCA on eight individual datasets. Biplots for PCAs were color-coded using large movements versus all other movements (1 = less than 95th percentile value; 2 = equals or greater than 95th percentile value). We centered all continuous predictors such that means were 0 and scaled these by dividing each predictor by its standard deviation [45].

The frequency distribution of movements was not normal and positively skewed, i.e., most observed movements were clustered to the left side of the x-axis and the right side of the distribution had a longer tail. Furthermore, individual movement distances constituted non-independent, repeated measures. For this reason, we sought to explain variation in our non-normal response variable with a model that incorporated fixed and random effects. We fit generalized linear mixed models (GLMMs) in the R package ‘glmmADMB’, using Poisson and Negative Binomial response distributions to explain observed variation in movements in relation to environmental cues and individual-level variables [46]. We rounded down movements to the nearest integer to enable use of these discrete error distributions. Since we specifically sought to elucidate drivers of large-magnitude movements, we did not remove outliers from our data. We used body size (straight-line carapace length [SCL] in mm) and sex (male or female) as individual-level predictors. We included streamflow, gage height, change in streamflow, and day of year as extrinsic predictors. Additionally, we included day of year and gage height as quadratic terms to account for potential peaks and troughs in movement patterns [30]. We accounted for non-independence of positional fixes by using individual turtle as a random factor. We centered and scaled all predictors. We back-transformed these variables for plots of model output. Candidate models included random variation in intercepts and random variation in slopes and intercepts. We evaluated model fit using Akaike Information Criterion adjusted for small sample size (AIC_c) values in package ‘MuMIn’ [47]. Models with a $\Delta$AIC_c ≤ 2 were considered similar to models that best supported the data. Once we had chosen the optimal random effects structure, we assessed error distributions and evaluated our fixed factors. A subset of candidate models (Table S1) specified higher-order interactions to account for among sex differences in response to increased streamflow and day of year. The fixed effect structure that best supported the data was chosen using AIC_c values.

2.2.4. Home Range Analysis

We viewed tracks of individual turtles and measured maximum net displacement (i.e., the total river length between the two most widely spaced locations) using Google Earth (Google, Mountain View, CA, USA). We characterized space use in tagged turtles by constructing Time Local Convex Hulls in the package ‘tlocoh’ [48]. This method constructs Utilization Distributions (UDs) by aggregating local minimum convex polygons (MCPs) constructed around each point. Nearest neighbor selection for construction of local hulls uses a time-scaled distance metric, which essentially pushes apart points that may be close in two-dimensional space but far apart in time. Local MCPs constructed by this method are local in space and time and can provide metrics of temporal use. We used the adaptive $a$ method in identifying nearest neighbors for each point. This method uses all points within a cumulative distance $\le$ $a$. We first examined the scaling factor $s$ which determines the degree to which hulls are local in time and space. We chose a value of $s$ for each turtle, such that 60% of hulls were time selected [48]. We determined the upper and lower bounds of $a$ for each turtle as the minimum $a$ value, which includes every point as a nearest neighbor in a hull with 3, 5, and 10 points, respectively [48]. We then plotted isopleths from unionized hulls using the range of $a$ as above and selected the $a$ value that preceded jumps in isopleth area curves and which corresponded to local minima in isopleth edge area ratio curves for the lower isopleths [48]. We exported individual locations and isopleths generated for selected $a$ values as shapefiles to plot and symbolize individual home range plots as thematic maps in ArcGIS Pro. We did not expect P. gorzugi to inhabit terrestrial areas beyond the shoreline, so we clipped the UD contours to the river [30,49]. We compared home range areas at 95% (i.e., total) and 50% (i.e., core) isopleth densities among male and female turtles using two sample t-tests in the ‘stats’ package in R [39].

We calculated two time-use metrics for turtles, the number of separate visits to a hull (i.e., revisitation), and the mean number of locations per visit (i.e., duration). For these time-use metrics, we used an inter-visit gap of 24 h, barring two individuals (i.e., individual IDs 4 and 9) for which we used an inter-visit gap of 36 h, as a duration of 24 h was lower than the median sampling frequency. We inspected the distribution of hulls in time-use space using scatterplots of hull revisitation rates and duration. We manually selected regions of low, medium, and high duration and revisitation rates in individual scatterplot space. Hull parent points were then symbolized by category (i.e., low, medium, and high) of duration and revisitation rate in thematic maps of individual home range. We were particularly interested in identifying hull parent points that were associated with high duration and revisitation or slower, localized movements as well as low duration and revisitation or transitory movements [48].

2.2.5. CMR Study

We conducted systematic CMR surveys from May 2015 to September 2018. We also included opportunistic one-day CMR surveys that were conducted in May and August 2011 and August 2014. We maintained consistent sampling over this duration at DFP and so include the data from this primary site alone. Of the 22 sampling visits, 3 were one-day surveys, 5 were two-day surveys, and 14 were three-day surveys. We snorkeled for and captured turtles by hand. We captured many turtles that were freely swimming, but had greater success actively searching underwater rock crevices and vegetation patches for turtles that may be hidden. Captured turtles were measured (SCL in mm; weight (W) in g) and classified as males, females, and juveniles. We classified turtles as male and female based on secondary sexual traits as above, and turtles with SCL $\le ~$100 mm that lacked secondary sexual traits as juveniles. We marked turtles for identification by cohort/year by notching or drilling a unique combination of marginal scutes [50]. We used Passive Integrated Transponder (PIT) tags or Monel flipper tags to mark turtles for individual identification. We injected PIT tags using an applicator either into the abdominal cavity via the inguinal pocket or subcutaneously into either hind limb. Flipper tags were placed between the toes of either hind limb; however, we primarily used these for individual identification when PIT tags were unavailable at the onset of the study. We photographed dorsal and ventral views of each turtle and their respective cohort and individual IDs prior to release. We scanned recaptured turtles for PIT tags using an Avid Power Tracker VI tag reader (Avid Identification Systems, Inc., Norco, CA, USA) and generated individual capture histories over 22 sampling occasions. The redundancy of cohort- and individual-level identification along with the catalog of images allowed us the opportunity to potentially identify individuals with PIT tag loss and replace these with new PIT tags/individual IDs. For any individuals that exhibited tag loss, we retained the older PIT tag ID in our analysis.

We applied a POPAN formulation of the Jolly–Seber model to individual capture histories in Program MARK to estimate abundance (N_s), capture probability (p), survival probability (φ), and probability of net new entrants (b) to the population [51,52]. Our abundance estimate (N_s) is the superpopulation size, which refers to every individual that was alive during the study period, and the entry probability (b) could result from either birth or immigration. Entry probabilities sum to 1 to ensure all N_s enter the population sometime during the study duration [51,53].

We used information–theoretic methods to select the best-supported model estimates of N_s, φ, p, and b. Candidate models stipulated fixed time-dependent b, φ, and p as well as constant φ and p. We adjusted parameter counts for models such that fully time-dependent models had $K$ parameters each for p and b, $K$-1 parameters for φ, and 1 parameter for N_s (

Table 2. Movement data were collected during a telemetry study conducted in the Devils River, Texas, from 2015–2017. Our best-fit model specified a type-two negative binomial error distribution. Turtles moved larger distances with increasing streamflow and gage height (i.e., water depth). Female turtles moved more than male (SexM) turtles, with larger movements made mid-year. Smaller turtles showed larger movements than larger turtles across both sexes (SCL—Straight Carapace Length).

Variable	Estimate	SE	Z	p Value
Intercept	5.664	0.176	32.23	<0.001
SCL	−0.663	0.190	−3.48	<0.001
SexM	−0.900	0.404	−2.23	0.026
(Day of Year)2	−0.533	0.071	−7.49	<0.001
Day of Year	0.218	0.116	1.89	0.059
Gage Height	0.127	0.045	2.84	0.005
Streamflow	0.304	0.050	6.39	<0.001

). We estimated the overdispersion parameter ($\widehat{c}$) by dividing the global χ² by its corresponding degrees of freedom [54]. The overdispersion parameter was incorporated to provide Quasi-likelihood Akaike Information Criterion scores corrected for a small sample size (QAIC_c) to evaluate model fit. Models with a $\Delta$QAIC_c ≤ 2 were considered similar to models that best supported the data.

3. Results

3.1. Movement Analysis

We recovered data from eight of the ten tagged turtles (Table 1). VHF signals were never recovered for one female (ID three) and one male (ID ten) turtle immediately following the initial tagging effort. We do not know whether this is due to device failure or due to these turtles moving beyond the range of our study area.

The data from eight of the ten tagged turtles accounted for a total of 2105 positional fixes. We obtained 13.7% of a maximum 15,360 positional fixes that would have been possible from eight transmitters over this study duration. The average number of positional fixes per individual turtle was 263 ($\pm$89 SD) with a range from 171 to 386 (Table 1). The average number of positional fixes was 301 ($\pm$94 SD) for females and 199 ($\pm$25 SD) for male turtles.

We observed that 78.4% of our location data included one positional fix per turtle per date, 21.5% included two positional fixes per turtle per date, and 0.1% included three positional fixes per turtle per date (Figure S1). The duration between the first and last positional fix for each turtle averaged 512 ($\pm$145 SD) days with a range from 245 to 640 days (Table 1). The average duration between the first and last positional fixes among female and male turtles was 484 ($\pm$183 SD) and 559 ($\pm$39 SD) days, respectively. Across all turtles, we obtained 474 positional fixes from August to December 2015, 1179 from January to December 2016, and 452 from January to May 2017. The number of monthly positional fixes across all turtles averaged 175 ($\pm$118 SD), with the highest number of positional fixes in February and the lowest number of positional fixes in June and July (Figure 2). For all turtles, movements between consecutive positional fixes averaged 159 (±492 SD) m, with female and male turtles moving an average distance of 148 (±470 SD) and 185 (±544 SD) m, respectively. Movements ranged from a minimum of 0 m to a maximum of 7874 m, with the time difference between movements averaging 2 days and ranging from a minimum of 24 s to a maximum of 45 days. Of a total 1729 observations, 25 movements exceeded a magnitude of 1634.5 m (i.e., Mean $+$ 3SD; Figure 2). The variation in our data, cumulatively accounted for by the first and second principal axes of our PCAs, ranged between 62 and 71% across individuals. We observed positive correlations between large movements with gage height (i.e., water depth), day of year, streamflow, and change in streamflow. We observed individual-level variation in the influence of predictors on large movements (Figure S2).

Our top-ranked model (Table S1) fit movement data to a type-two negative binomial distribution (i.e., quadratic mean–variance relationship) with intercepts and slopes randomly varying among individual turtles. This model showed that movement increased as the streamflow ($\beta$ = 0.304, SE = 0.05, Z = 6.39, p < 0.001; Table 2; Figure 3) and gage height or water depth ($\beta$ = 0.127, SE = 0.05, Z = 2.84, p < 0.001; Table 2; Figure 3) increased. Larger movements were seen from day 150 to 220—i.e., between May and August ($\beta$ = −0.533, SE = 0.07, Z = −7.49, p < 0.001; Table 2; Figure 3). Male turtles moved shorter distances than females ($\beta$ = −0.900, SE = 0.40, Z = −2.23, p < 0.026; Table 2; Figure 3) while across both sexes, larger turtles moved shorter distances than smaller turtles ($\beta$ = −0.533, SE = 0.07, Z = −7.49, p < 0.001; Table 2; Figure 3). The ‘unexplained’ variation in our dataset was the largest (SD_residuals = 0.38), followed by heterogeneity in individual response to predictors (SD_slopes = 0.30), while heterogeneity in individual intercepts (SD_intercepts = 0.28) was the smallest.

3.2. Home Range Analysis

The average maximum net displacement for all turtles was 10.7 ($\pm$11.3 SD) km, varying from 1.2 to 35.5 km. The average maximum net displacement among female and male turtles was 11.1 ($\pm$14.1 SD) and 9.9 ($\pm$6.8 SD) km, respectively. For all turtles, the average home range size (i.e., 95% UD) was 45.5 ($\pm$91.4 SD) ha, varying from 0.4 to 258.0 ha. The average home range size of female turtles was 53.1 ($\pm$114.6 SD) ha and 33.0 ($\pm$51.2 SD) ha in males; however, home range size was not significantly different among sexes (t = −0.340, df = 5.82, p = 0.746). For all turtles, average size of the core area was 3.8 $(\pm$4.8 SD) ha, varying from 0.10 to 9.2 ha. The average core area size of male turtles was 4.7 ($\pm$4.5 SD) ha and 3.2 $(\pm$5.4 SD) ha in females; however, core area size was not significantly different among sexes (t = 0.428, df = 5.09, p = 0.686).

Among our tagged turtles, we make special note of the maximum net displacement and home ranges of one female [29] and two male turtles (i.e., IDs five, eight, and nine) that showed the greatest maximum net displacements of 35.5, 16.2, and 11 km, respectively (Table 1; Figure 4, Figure 5 and Figure 6). We observed that the female turtle (ID five) showed two discrete utilization areas, one that extended 3.4 km downstream from Dolan Falls and the second in the Devils River arm of Lake Amistad. The intervening river length between these two areas was 22.5 km. This female showed a high number of high-duration revisits within the core area in the Dolan Falls pool and a low number of low-duration revisits in the intervening river length and Lake Amistad (Figure 4). Similarly, one male turtle (ID eight) showed two discrete core areas, one along the DRSNA and the second core area was 7 km upstream. This male showed a high number of low-duration revisits in both core areas, while a low number of low-duration revisits were indicated in the river stretch intervening these core areas (Figure 5). The core areas for the second male (ID nine) included river stretches at the DRSNA, 1.8 km downstream of Dolan Falls pool, as well as the spring-fed stretch of Dolan Creek within Dolan Falls Preserve (Figure 6). Positional fixes obtained ~11 km downstream for this turtle are not included in the home range. This male turtle showed a high number of high-duration revisits in the core area at the DRSNA site, a medium number of low-duration revisits at Dolan Creek, and a low number of low-duration revisits at Dolan Falls pool. A low number of high-duration revisits were seen 11 km downstream of the core areas, while the riverine stretch intervening this downstream point and upstream core areas indicated a low number of low-duration revisits (Figure 6).

Three female turtles, IDs two, six, and seven, that were tagged at Dolan Falls pool showed home ranges restricted to a 1 km extent of the pool over the study duration. Two of these females, IDs six (Figure S3) and seven, (Figure S4) showed core areas within Dolan Falls pool and smaller discrete home range areas ≤ 0.5 km downstream of Dolan Falls pool. These female turtles showed a high number of high-duration revisits in their core areas and showed a low-to-medium number of low-duration revisits at the edge of their core areas and smaller, separate, 95% utilization patches (Figure S3, Figure S6, Figure S7). Additionally, a female (ID one; Figure S8) and male (ID four; Figure S9) turtle revealed home ranges restricted to within a 1 km extent of their tagging site at the DRSNA. The female turtle at the DRSNA showed two discrete utilization areas that were 0.6 km apart and showed a high number of high-duration revisits in its core area and a low number of low-duration revisits at the upstream home range boundary (Figure S8). Finally, the male turtle (ID four) at the DRSNA exhibited a high number of high-duration revisits in the core area and a low-to-medium number of low- and high-duration visits at the upstream boundary of its home range (Figure S10).

3.3. CMR Study

Over 22 sampling occasions, we caught 293 turtles once, 90 turtles twice, 47 turtles three times, 22 turtles four times, nine turtles five times, 2 turtles six times, and one turtle seven times, constituting a total 766 captures of 464 turtles. We caught 91 female turtles (SCL 251.1 $\pm$ 45.4 mm; W 2263.1 $\pm$ 994.2 g), 357 male turtles (SCL 200.9 $\pm$ 36.0 mm; W 1094.4 $\pm$ 539.5 g), and 16 juvenile turtles (SCL 96.0 $\pm$ 30.5 mm; W 170.9 $\pm$ 193.8 g). The model that best fit our live recapture data and received the highest QAIC_c weight specified constant φ and N_s and time-dependent p and b (φ_(.)p_(t)b_(t)N_s(.);

Table 3. Live recapture data were collected over 22 sampling occasions in 2011, 2014, and 2015–2018 at the Dolan Falls pool on the Devils River. We sought to estimate abundance of adult P. gorzugi using POPAN formulations that made no assumptions about geographic closure. The model specifying constant apparent survival (φ), time-dependent capture probability (p), and probability of net new entrants (b) best fit our data; CMR—Capture–Mark–Recapture, QAIC_c—Quasi-likelihood Akaike Information Criterion scores corrected for a small sample size, K—Number of Parameters, w—Model weight, N—Superpopulation estimate.

Model	K	QAICc	ΔQAICc	w	∑ w
φ(.)p(t)b(t)N(.)	46	2022.06	0.00	0.979	0.979
φ(t)p(t) b(t)N(.)	66	2029.77	7.71	0.021	1.00
φ(.)p(.)b(t)N(.)	25	2067.05	44.99	0.00	1.00
φ(t)p(.) b(t)N(.)	45	2081.51	59.45	0.00	1.00

). Annual apparent survival was estimated at 0.804 $\pm$ (0.007 SE). Catchability ranged from a minimum of 0.007 $\pm$ (0.003 SE) to 0.999 $(\pm$0.009 SE) and entry probability ranged from 0 to 0.298 $(\pm$0.166 SE). The superpopulation, i.e., the estimated sum of observable and unobservable individuals that have ‘passed through’ the site over the study period consisted of 973 ($\pm$ 126 SE) turtles, with an estimated range of 726 to 1219 turtles. Additionally, in April 2018, we recaptured a male turtle (SCL 226 mm; W 1380 g) at the DRSNA which was originally captured in 2015 at Dolan Falls pool. This net displacement of 1.4 km was the sole individual movement between the DRSNA and Dolan Falls pool observed over the study duration.

4. Discussion

4.1. Movement Analysis

The movement of an organism has a major role to play in the fate of individuals and in determining the structure and dynamics of populations, communities, and ecosystems [1]. Riverine turtle species have been observed to move long distances, with the Amazon Turtle (Podocnemis expansa) moving distances of 45–422 km between foraging areas and nesting beaches in the Amazon River basin [55,56]. We have observed the largest movements reported for P. gorzugi, identified environmental cues that drive these movements, and elucidate the implications of these findings on the abundance, population genetic structure, and potential conservation approaches for the species.

A relatively small fraction of total programmed fixes was realized during our study, most likely due to the inability of GPS data loggers to obtain positional fixes while submerged [28,57]. Despite the low fix success rate, our study yielded nearly six times as many positional fixes [28], over 1.1 to 4 times the duration [28,58,59,60] relative to previous semi-aquatic and freshwater turtle GPS telemetry studies. Similarly, the number of relocations from our study is 1.1 to nearly 5 times greater than conventional VHF freshwater turtle telemetry studies [30,61]. However, we also acknowledge that the number of tagged turtles that we retrieved data from may not be representative of population-level dispersal patterns.

We obtained year-round positional fixes across tagged turtles with a higher number of positional fixes at lower ambient air temperatures. Since positional fixes were only possible at the surface, we speculate that the higher number of positional fixes may be a result of higher surface and atmospheric basking activity that is inversely proportional to air temperature [62]. Additionally, nocturnal surface activity was rare to the point of being non-existent among tagged turtles.

The location data predominantly consisted of one diurnal positional fix per turtle per date. For this reason, we maintain that our movement estimates between successive relocations by date are an appropriate proxy for net individual movement. Tagged P. gorzugi showed an increased magnitude of movement as the streamflow and water depth increased. Previous studies have documented individual turtles dispersing downstream during high-flow events [63,64,65,66] with subsequent upstream movements to counteract such dispersal occurring as streamflow decreases [63,66,67]. An increased flow has also been observed to enable site access in turtles [68] and flooding has been considered an important mechanism of population connectivity, particularly in stream-dwelling turtle species [64]. We observed long-range movements that occurred downstream and upstream. Individual heterogeneity in response to an increased streamflow and water depth is best exemplified by the two turtles (ID five and nine) that dispersed downstream and one other (ID eight) that moved upstream. Further, one of the turtles (ID nine) that dispersed downstream following an increased streamflow showed a subsequent upstream movement to counteract this dispersal. In addition, a previous study examining the response of Devils River fish assemblage to flooding observed that habitat types became less distinct and more riffle-like subsequent to flooding [33]. On this basis, we speculate that an increased streamflow potentially enabled among-site connectivity and facilitated movement in tagged turtles, with individual heterogeneity in the directionality and magnitude of such movement. Additionally, the higher sampling frequency and duration of our GPS telemetry study revealed an individual movement that was two orders of magnitude greater than previous estimates of the maximum movement in P. gorzugi. However, these previous estimates relied on CMR and radio telemetry studies that sampled less frequently and over a relatively short duration [13,29]. The maximum net downstream movement observed during this study is among the largest reported for North American freshwater turtles, only exceeded by an observation from a female Suwannee Cooter (Pseudemys suwanniensis). This female P. suwanniensis showed a downstream movement of 130 km from the Santa Fe River to the Suwannee River estuary in response to high flow conditions followed by an upstream movement back to the original capture site during CMR surveys that spanned 2011–2015 [63]. Although the magnitude of movement observed in this female P. suwanniensis was considerably larger than our study, the environmental cues driving such movement appear to be consistent.

Larger movements occurred between the months of May and August and female P. gorzugi moved longer distances than male P. gorzugi. Reproductive studies of female P. gorzugi in New Mexico observed 86% of female turtles with shelled eggs or follicles between the late-May and mid-June [32]. Similarly, year-round observations during a study on the female reproductive cycle of P. gorzugi in Texas revealed peak numbers of females with shelled eggs in late-June [31]. Higher cumulative movements by females and increased movement during peak nesting season appear to be consistent with the reproductive strategies hypothesis, where higher female movements are associated with nesting forays [30,66]. However, variation among female and male turtle movement may also be, in part, attributed to an unbalanced sample among sexes, with the number of tagged female P. gorzugi and the resulting number of positional fixes exceeding the positional fixes yielded from a smaller number of tagged male turtles. Across both sexes, we observed that larger turtles moved shorter distances than smaller turtles. This contrasts to the greater mobility observed with a larger body size in previous studies of turtle movement, where higher swimming speeds and ability to handle strong currents at larger body sizes were considered to confer greater mobility [30,66,69,70].

4.2. Home Range Analysis

We estimated two-dimensional space-use for P. gorzugi by calculating home range area. The individual heterogeneity in net displacement was also observed in home range size, with larger turtles across both sexes appearing to have relatively static home ranges. We did not find statistical support for differences in core area and home range size among male and female turtles, likely due to a small sample size of eight individuals.

Turtles either showed distinct shifts in home range, with large movements enabling these shifts, or had relatively established home ranges but with occasional, large homing movements. A high number of high-duration revisits to local hulls may indicate slower and localized movements that are associated with Area Restricted Searching (ARS) behavior while a low number of low-duration revisits likely indicate longer and rapid movements that are associated with transiting behavior [48,71]. One female (ID five) moved from a core area in Dolan Falls pool and exhibited a shift in home range to the Devils River arm of the Amistad Reservoir. Movements within the core area were indicative of ARS behavior while downstream dispersal to Lake Amistad seemed to indicate transiting behavior. The latter movements transited a heterogenous mix of riffle, run, and pool habitats [33] with several rapids that are listed as navigational hazards to paddlers (https://tpwd.texas.gov/state-parks/devils-river/river-trips, accessed on 2 January 2021). We replaced the transmitter on this female after entry into Lake Amistad on March 2016 as the VHF signal was failing. However, we did not subsequently detect this female which truncated continued monitoring. Similarly, we observed a male turtle (ID eight) show a shift in its core area from upstream of the DRSNA to an area 7 km upstream. Movements within each of these core areas appeared to be slower and localized, with the movement between these core areas indicative of transiting behavior. The intervening stretch between these core areas consisted of alternating riffle and pool habitats connected by runs [33], with one waterfall of navigational note (https://tpwd.texas.gov/state-parks/devils-river/river-trips, accessed on 2 January 2021). Further, another male (ID nine) showed distinct core areas in the Devils River alongside the DRSNA, in the adjoining Dolan Creek, and at Dolan Falls pool. The turtle seemed to show ARS behavior alongside the DRSNA; however, movements between Dolan Falls pool and ~9 km downstream as well as the subsequent return to core areas appear to be transitory. Turtles with smaller home ranges (linear extent ~1 km) also showed separate (IDs six and seven) utilization areas or utilization areas that were resolved from each other (ID four). Movements at the edges of home ranges or smaller, discrete utilization areas indicated a low number of short visits which may be indicative of temporary sojourns or exploratory sallies [72]. We speculate that these observations may indicate a common pattern in tagged turtles of area-restricted searching with switches to transitory or exploratory modes [71]. Possibly, changes in the movement mode occurred in association with extrinsic cues such as streamflow, water depth, and day of year with individual heterogeneity in the linear extent of these transitory movements. Wandering movement or the shifting of activity areas in turtles has been previously reported to be associated with changing water levels [73,74]. Some long-range movements may have been escape reactions following capture [30,72], but nonetheless speak to the dispersal capacity of P. gorzugi.

4.3. Population Estimate

A strength of this study was using movement data from our telemetry study to aid in analyzing and interpreting CMR data. Modeling live recapture data while considering the population open to births, deaths, immigration, and emigration was appropriate relative to large net displacements by tagged turtles over relatively short periods [51]. Sparse recapture rates led to a lack of identifiability of catchability and entry probability over sampling occasions with concomitantly large error margins for parameter estimates [51]. We estimated that the number of observed and unobserved turtles that passed through our sampling site at Dolan Falls pool ranged between 726 and 1219 turtles. However, in context with the nature of transitory and exploratory movements exhibited by seven of the eight tagged turtles, we speculate that turtles captured at Dolan Falls pool need not necessarily have been residents. Thus, while this estimate was derived at a single site, we believe it is likely representative of the number of P. gorzugi in the Devils River. Further, the net displacement observed during the CMR study was one order of magnitude lower than revealed by GPS telemetry observations. We maintain that individual movements have likely been underestimated at relatively coarse scales of observation [13,60] with a failure to detect individuals that transited to upstream and downstream sites or detecting individuals at the end of their homing movements. Further, individual movements indicate a violation of geographic closure. Failing to acknowledge this by interpreting the estimate of abundance in isolation would restrict our estimate to the sampled area, indicating a much higher density of turtles than represented in our study.

5. Conclusions

The extent of movement in P. gorzugi has been previously underestimated. We acknowledge that the sample size incorporated in this study may not be representative of population-level dispersal patterns. However, new, more extensive data show individual turtles capable of longer-distance temporary sojourns or directed movements which were correlated with environmental variables such as streamflow, water depth, and seasonality. We recommend GPS telemetry studies be extended to populations of P. gorzugi populations in the New Mexico portion of the Rio Grande drainages to potentially corroborate our findings or provide a contrast in behavior in the northern portion of the current distribution. Vandewege et al. [26] characterized range-wide genetic structure in the species using genome-wide sequence data. Genetic structure within the Rio Grande watershed indicated that P. gorzugi likely use the Rio Grande as a movement corridor, navigating between localized, preferred habitats. Our study corroborates this notion, having revealed individual movement phases that virtually spanned the entire reach of the Devils River which maintains surface water. The Rio Grande river system is heavily regulated and continuously increasing demands on water resources have dramatically reduced its streamflow [18]. However, our study leads us to infer that the restoration of natural, regionally specific streamflow patterns [75,76] is required to maintain historic connectivity across the range of P. gorzugi. A complete restoration of the streamflow may not be politically, socially, and economically feasible. However, a quantitative understanding of the extent to which reductions in streamflow and disruption of flood events have impacted the species in the last century followed by efforts to partially restore natural flow regimes could potentially be the most effective approach in ensuring P. gorzugi persist across their restricted range.