Quantifying movements and home ranges of an estuarine turtle: the effects of urbanization and boundaries

Abstract

Background

Tracking small-bodied animals in estuarine environments entails significant technological and analytical challenges. Diamond-backed terrapins are small (max 1.4 kg) turtles that inhabit salt marshes of the eastern U.S. and the Gulf of Mexico. Terrapin movements have been studied with VHF radio telemetry, acoustic telemetry, and mark and recapture methods, which have indicated maximum straight-line movement distances < 10 km and mean home ranges < 1 km².

Results

We deployed 21 Argos satellite tags on adult female terrapins at two sites on Long Island, New York to better understand the spatial ecology of this imperiled species, and to test newly available tracking technology. We processed the location data three ways: (1) we used a location data filter to remove unlikely terrestrial and oceanic locations and applied a state-space model to account for Argos location errors, (2) we applied the state-space model to unfiltered data to determine the effects of not removing unlikely locations, and (3) we used only the highest quality location class 3 (LC 3) locations. We used the data resulting from each of these approaches to calculate four different movement metrics: summer home range size (95% minimum convex polygons (MCPs) and kernel density estimates (50% and 95% KDE, with both reference [href] and least squares cross validation [LSCV] bandwidths)), the total distance traveled from June to August, maximum distance traveled in one day, and daily movement rates. Home ranges estimated from the three processing techniques were similar in size and covered the same spatial areas. Estimates for total distance traveled, daily movement rates, and maximum distance traveled were similar between the state-space modeling techniques, but LC 3 estimated distances were twice as long. Movement metrics and home ranges were similar between the two study sites, despite differences in urbanization and bay size.

Conclusions

These results suggest that most movement metrics and home range estimates are fairly insensitive to these different analytical techniques, even at relatively smaller spatial scales. Additionally, our study indicates substantially larger home ranges and longer straight-line movements than VHF telemetry or sonic tag studies, highlighting the utility of satellite tags to improve our understanding of terrapin ecology and conservation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40850-026-00273-3.

Background

Studying spatial ecology can reveal a wide variety of animal behaviors, including habitat preferences, movement patterns, and home range sizes [1]. Spatial ecology is also used to test basic ecological principles (e.g., [2, 3]). When this involves rare or imperiled species, the goal of such studies can be the identification of appropriate actions that can be taken to protect habitats and implement management plans [4–6]. Turtles (order Testudines) are among the highly endangered taxa [7] and a wide variety of techniques have been used to track their movements based on body size, habitats used, and available technology [8–10]. Until recently, the use of satellite telemetry (GPS and Argos tags) to track turtles has been limited to larger-bodied species including adult sea turtles and giant tortoises, because such tags have been too large and heavy for smaller turtle species [11]. However, technological advances have produced smaller and lighter tags that are now available to use on juvenile sea turtles, diamond-backed terrapins, smaller tortoises, and freshwater turtles (e.g., [12–19]).

The diamond-backed terrapin (Malaclemys terrapin) is the only North American estuarine turtle species; they range along the coastlines of the eastern U.S. and Gulf of Mexico from Cape Cod, Massachusetts to Corpus Christi, Texas and Bermuda [20]. Terrapins inhabit salt marshes dominated by Spartina alterniflora and Spartina patens and mangrove swamps in Florida [21]. Terrapins may be considered a keystone species for their role in top-down control of Spartina herbivores such as the periwinkle snail (Littoraria irrorate), seed dispersal, and nutrient movement [22–25]. During the 18th and 19th centuries, terrapins were over-harvested for human consumption, resulting in steep population declines [20]. Populations have partially recovered but are still decreasing and terrapins are listed as vulnerable on IUCN Red List [26]. Reasons for recent declines include habitat loss, exploitation by the pet trade, vehicle and boat strikes, drowning in crab pots, and subsidized predation [27–29].

Previous diamond-backed terrapin home range studies have reported female terrapins moving < 10 km from capture sites, and 95% minimum convex polygon (MCP) and 95% kernel density estimate (KDE) home ranges < 1 km² [28–34]. These studies used VHF radio telemetry, acoustic telemetry, and mark and recapture methods, which have limitations such as radio wave attenuation in saltwater and requiring extensive boat time, direct line of sight between transmitter and receiver, and close proximity. Terrapins have been radio tracked > 9 months but may go missing for extended periods of time and if located again, whereabouts of the individuals for that timeframe can only be speculated [28, 29]. Butler (2002) located one terrapin 8 days after it was last seen nesting at a beach 6.4 km away and a second > 10 days after it was last seen 4.8 km away; another six terrapins were released with transmitters and were never located again [28]. In these examples, routes traveled between sightings or after release cannot be determined. Conversely, satellite telemetry tracks turtles continuously for > 9 months, collects fixes 24 h/day, in all weather conditions, providing more detailed movement paths. Continuous tracking with satellite telemetry can thus reveal short-term movements that radio telemetry may fail to detect [35]. Due to the recent development of smaller satellite tags, Lamont et al. (2021, 2023) used Argos-based satellite tags to track female terrapins in Florida, revealing considerably larger terrapin home range estimates of 9.4 km² (95% MCP) and 8.1 km² (95% KDE) compared to prior radio telemetry studies [14, 18].

Differences in home range size between radio and satellite tracking studies may be caused by differences in terrapin movements, tracking range, surface time required for location fix, accuracy of each fix, number of fixes possible, etc. Argos satellite tags only require one satellite to collect a location fix, which can take < 1 s, but have large error radii (250-1,500 m) that can lead to overestimation in home ranges [36–38]. Terrapins likely remain within the range of Argos error estimates, making it difficult to determine fine-scale movements in a small area (< 5 km²) [14, 18, 31, 39]. This is not a problem for studies investigating animals that make long-distance movements (e.g. sea turtles, sharks, whales, hundreds to thousands of km); in such instances Argos errors are likely negligible [40]. Alternatively, GPS satellite tags collect more precise locations (~ 10 m) but require connection to at least four satellites, taking 30 s to several minutes to obtain a fix. The inability of GPS tags to collect fixes underwater has limited their use to track freshwater turtles but they have been used to enhance terrestrial radio-tracking studies [41–43]. Fastloc-GPS tags that obtain fixes within < 1 s with greater accuracy are available, but these tags are more expensive and are not yet small enough to use on terrapins [44, 45]. Therefore, despite their relatively low accuracy, Argos tags are currently the most appropriate tool for tracking species that only surface briefly [9].

One possible explanation for larger terrapin home ranges found in satellite tag based studies vs. traditional studies is the increased number of location fixes, which can increase the accuracy of home range estimates [46, 47]. Satellite tags can produce large datasets (> 1,000 locations/individual) (Lamont et al. 2023), whereas radio telemetry and mark and recapture studies typically have much smaller datasets (< 300 locations/individual) [27–29]. Christensen & Chow-Fraser (2014) discovered Blanding’s turtle (Emydoidea blandingii) movement corridors, nesting habitat, and fine-scale movements that would have been missed if only radio telemetry was used; home ranges were on average 24.7% larger when GPS tag fixes were included [41]. GPS tags also yielded larger and more accurate American alligator (Alligator mississippiensis) home ranges than VHF telemetry by identifying marsh habitat use and collecting more spatial data [48]. Differences in home range analytical technique, study site ecoregion, and the sex of the turtles studied can also influence home range size estimates [8].

Terrapin movements between populations are not well studied, however, nesting research shows that most terrapins tend to stay within their isolated populations and are generally philopatric [49]. Average female dispersal for nesting lies between 10 and 100 m [50, 51]. Sheridan et al. (2010) found that terrapins displayed sex-biased dispersal in Barnegat Bay, NJ, with mature females having a larger mean dispersal distance (722 m) than mature males (104 m) [33]. Baldwin et al. (2005) found only two individuals out of 2000 marked individuals moved between two Florida islands only 1000 m apart [52]. Sheridan (2010) used patterns in microsatellite marker variation to detect dispersal barriers to terrapin movement and found that open water and developed landscapes were barriers to movement while shallow water and emergent vegetation habitat facilitated movement [53].

In this study, we explored techniques to account for relatively large Argos location errors, including filtering unlikely locations with boundaries and state-space models (SSMs). Our objectives were to calculate and compare terrapin home ranges, total distance traveled, maximum distance moved in a day, and daily movement rates using three different data processing techniques. We used data from 21 Argos based satellite tags deployed on adult female terrapins at two study sites at nearly identical latitudes on Long Island, New York, one highly urbanized and one rural site. The differences in urbanization between these two sites affected how easily unlikely locations were filtered out with boundaries.

Methods

Study sites and capture

Adult female diamond-backed terrapin movement data were collected at two sites on Long Island New York (U.S.A.). The westernmost site, Jamaica Bay (JB), is an estuarine embayment in extreme western Long Island, southeastern New York City, covering approximately 25 km². Jamaica Bay is heavily urbanized; > 70% of surrounding land cover is developed medium or high intensity building [54] with population densities of 22,108 people/km² (Queens) and 39,368 people/km² (Brooklyn) [55]. Jamaica Bay terrapins have been studied since 1998 (e.g., [22, 56–58]), the two known major JB nesting sites are at Ruler’s Bar Hassock (RBH) (40.62141, -73.82706) and John F. Kennedy Airport (JFK) (40.64329, -73.78281) [59]. Although RBH and JFK are only 4.5 km apart, the two nesting population differ: the number of nests and nesting females at RBH has been declining steadily since 1998 while the population at JFK has been growing rapidly [59]. The JB emergent plant community is dominated by smooth cordgrass (Spartina alterniflora) and common reed (Phragmites australis) [60, 61]. In 1924, JB had approximately 10 km² of salt marsh; less than 4 km² remain today [62, 63]. There are ongoing salt marsh restoration projects to increase salt marsh habitat and mitigate the effects of human development surrounding JB [60, 62]. In addition to saltmarsh decline, terrapins in JB face heavy boat traffic, subsidized predation, and poor water quality [58].

The second study site was Orient Beach State Park (OBSP) (41.13064, -72.26398) at the extreme eastern end of Long Island’s North Fork, 173 km ENE of JB. Orient Beach State Park is located on a long narrow sand peninsula that protects Long Beach Bay. The park, bay, and bordering tidal salt marshes are approximately 5 km². The surrounding area is lightly urbanized; 25% of land in Orient, NY is developed open space or low density building and 40% is pastures, cultivated cropland or barren land [54], population density is 196 people/km² (~ 1% of JB) [55]. Orient Beach State Park is in the Peconic Estuary, a 640 km² system between the north and south forks of Long Island. The Peconic Estuary has experienced less salt marsh loss caused by Phragmites australis intrusion and urban development than JB [64]. The terrapin population at OBSP is poorly studied, but anecdotal information suggests that terrapins are abundant and widespread [65–67].

Adult female terrapins were captured on land as they attempted to nest at these two sites, satellite tags were attached (as per Lamont et al. 2021), and terrapins were released immediately after attachment epoxy had hardened. A total of 22 satellite tags were deployed on female terrapins in 2023 and 2024; four on RBH terrapins in June 2023, four on JFK terrapins in June 2023 (one failed immediately), five on RBH terrapins in June 2024, five on JFK terrapins in June 2024, and 4 on OBSP terrapins in June 2024. In 2023, both Wildlife Computers SPOT-387 (39 g, 59 mm x 29 mm x 23 mm) and Lotek K2g173A (34 g, 60 mm x 27 mm x 17 mm) were used, in 2024, only Wildlife Computers SPOT tags were used. Tags were programmed to be active 24 h/day and limited to ≤ 250 transmissions/day. Tag deployment was limited to female terrapins only due to the smaller size of males, which have an average mass of about 200–300 g. Adult female terrapins ranging from 800 to 1,000 g were tagged following the recommended 5% standard, such that the mass of the transmitter and epoxy was 40–50 g, less than 5% of the terrapin’s mass [68]. Animal work was pre-approved by Hofstra University IACUC (permit #22.23-09) and appropriate state and federal permits.

Analysis

Both SPOT 387 and K2G 173 A tags utilize the Argos satellite system to generate locations, which are assigned to location classes (LCs) using Kalman filtering. There are seven LCs (0, 1, 2, 3, A, B, Z) based on accuracy estimates. The most accurate is LC 3 in which the true location of the tagged animal is estimated to fall within a radius of < 250 m. Radii for LC 2 and LC 1 are 250–500 m and 500–1500 m, respectively. The least accurate is LC 0 with a radius of > 1500 m. Location class A and B have unknown accuracies but are still useful in some data analyses [38]. Location class Z points fail the Argos plausibility test and were removed from the dataset [69]. Tags only transmit when terrapins surface under appropriate conditions and when satellites are within range, so locations are collected at irregular time intervals and often with large temporal gaps. In addition to inevitable errors associated with the technology, we expected some low-quality data points because little is known about terrapin surfacing behavior and, especially in the northern portions of their range, little is known about their overwintering behavior. Satellite tags can only achieve accurate location fixes when the terrapins are at the surface for a sufficient time, and that time requirement varies according to atmospheric conditions and satellite locations. Terrapins in the region overwinter in a variety of underwater conditions from about November to May, often covered in mud [70, 71], which also reduces the number and probably the accuracy of location fixes.

For our first analytical method (state-space model), we used a Bayesian hierarchical state-space model (SSM) to generate daily mean locations for each terrapin using the ‘fit_ssm’ function from the “bsam” R package [72, 73] to improve location estimates and account for temporal gaps (R version 4.3.0; R Core Team, 2023). A hierarchical first difference correlated random walk (hDCRW) was fit to terrapin tag datasets simultaneously to improve estimations by calling JAGS (package rjags) to run the Markov Chain Monte Carlo (MCMC) algorithm [72]. The MCMC arguments were set following Lamont et al. (2023) terrapin study with 10,000 adaptive samples, 30,000 samples from the posterior distribution, a time step of one day and thinning by 20 to reduce MCMC autocorrelation. We used the ‘diag_ssm’ function from the ‘bsam’ package to assess convergence of the fitted ‘fit_ssm’ model via Gelman and Rubin’s shrink factor [74]. Any locations collected after a temporal gap of more than five days were removed prior to running the SSM because large temporal gaps cause uncertainty in SSM location estimates. These temporal gaps typically occurred during November and December, perhaps coinciding with brumation.

Daily mean locations generated from the SSM (above) were used to estimate summer (June-August) home ranges. Minimum convex polygons (95% MCP) and kernel density estimates (50%, 95% KDE) were calculated using the ‘adehabitatHR’ R package [75]. We calculated 95% KDE and 50% KDE two ways, once with the reference bandwidth (href) and again with least squares cross validation (LSCV), because bandwidth selection greatly influences home range size [76]. Smaller bandwidths, such as LSCV, may underestimate home range size while larger bandwidths, such as href, may overestimate home range. The 50% KDE was used to represent the core area of each terrapin’s activity. Although the SSM accounts for most of the location errors, we used 95% MCPs and 95% KDEs to account for additional outliers. Distances between subsequent daily mean locations were calculated in ArcGIS to determine the total distance traveled from June to August, maximum distance traveled in one day and daily movement rates.

For our second analytical method (boundary filtered + state-space model), we explored the effect of removing unlikely locations on movement metrics. Both unlikely terrestrial and oceanic locations for terrapins in JB and OBSP were removed using boundaries digitized in ArcGIS. Boundary polygons extended approximately 100 m inland of shorelines where terrapins could potentially nest and excluded hardened shorelines (e.g. bulkheads, roads). All location fixes within the boundary were selected to create a “boundary filtered” dataset, while fixes outside boundary lines were excluded. The JB boundary was 25.8 km², including all inlets and extending west to the end of the Rockaway Peninsula where the bay meets the Atlantic Ocean (Fig. 1A). The OBSP boundary was 34 km² and included Long Beach Bay and portions of Gardiner’s Bay (Fig. 1B). Other bays west of OBSP were excluded from the boundary because of a lack of LC 2 and LC 3 points. We repeated home range calculations and other movement metrics as described above.

Fig. 1.

Boundaries used for removing unlikely location fixes on land and in the ocean outlined in white and Argos error radii. LC 3 error radius of 250 m in green, LC 2 error radius of 500 m in yellow, LC 1 error radius of 1,500 m in red. (A) Jamaica Bay boundary (25.8 km²) (B) Orient Beach State Park boundary (34.0 km²)

We estimated our third set of movement measures (home range, total distance traveled from June to August, maximum distance traveled in one day and daily movement rates) using the LC 3 data (Fig. 2) without using SSM. We then calculated percent change in home range area, distance between home range centroids and percentage of spatial overlap using ArcGIS among the three approaches (SSM without boundary filtering, SSM with boundary filtering, LC3 data) to compare methods of estimating home ranges. We did not estimate centroids and overlap for KDEs with LSCV due to fragmentation of home ranges into multiple small “islands” of space use. We used repeated measures ANOVAs to test for differences in home range size, total distance traveled from June to August, maximum distance traveled in one day, and daily movement rates amongst the three data processing techniques for each terrapin. We used paired t-tests to compare home range sizes between the two SSM methods. Shapiro-Wilk tests were used to confirm data normality; nonparametric data was normalized with logarithmic transformations.

Fig. 2.

Flowchart depicting the three methods of data filtering and analysis used to estimate home ranges from satellite tag locations

Results

Twenty-one adult female terrapins were captured and tagged in June 2023 and June 2024 at JB and OBSP on Long Island, NY (Additional file 1: Table S1). The tags collected 31,063 locations between June 15, 2023 and when the last tag stopped working on May 13, 2025, excluding 6 LC Z points. Individual tag productivity fell into two clear categories, 16 “high performance” tags collected > 1,000 locations each and > 100 LC 3 each; five “low performance” tags collected < 400 locations each and < 10 LC 3 each. The low performance tags (20% of deployed tags) consisted of two Lotek tags deployed at RBH in 2023 and three Wildlife Computers tags deployed at JFK in 2024. Location class B fixes dominated the dataset (54%) followed by LC A (14%) fixes and LC 3 (11%) fixes. The average tracking duration was 215 days (range 79–358). The JB boundary removed 1,239 fixes of 13,174 fixes (9.4%) collected between tag deployment in June and August 31 with an average of 60 fixes removed per high performance tag (range 24–119) and 105 fixes removed per low performance tag (range 54–168) (Fig. 3; Additional file 1: Table S2). An average of 6% (2.2–11.7%) of the total number of fixes were removed per high performance tag and the average number of fixes/day decreased from 14.2 before filtering to 13.3 after filtering. Boundary filtering removed an average of 50.4% (23.1–73.7%) of the total number of fixes per low performance tag and caused temporal gaps > 5 days. As a result of this, home ranges for low performance tags were estimated using the longest duration of uninterrupted data collection (26–57 days). Boundary filtering did not produce temporal gaps for the high performance tags. The OBSP boundary removed 1,135 fixes out of 6,551 (17.3%) between June to August 2024 with an average of 284 fixes removed per tag (235–351); an average of 17.6% (13.4–21.4%) of the total number of fixes were removed per tag (Fig. 3). The average number of fixes/day decreased from 20.2 before filtering to 16.7 after filtering.

Fig. 3.

Raw locations and boundary filters for JB (left) and OBSP (right). A-B) LC 0, 1, A, and B points. C-D) LC 2 and 3 points. E-F) Locations selected for boundary filtering. All LCs are shown

Home range

Jamaica Bay

Shapiro-Wilk tests confirmed normality of all JB home range sizes except for LC 3 50% and 95% KDE-LSCV for high performance tags and unfiltered SSM 95% KDE-href for low performance tags (Additional File 1: Table S5). Unfiltered SSM mean home range sizes for the 12 high performance tags were 5.7 ± 4.3 km², 11.5 ± 7.9 km², 5.8 ± 3.0 km², 1.9 ± 1.3 km², and 0.9 ± 0.6 km² for 95% MCP, 95% KDE-href, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively. Unfiltered SSM mean home range sizes for the 5 JB low performance tags were 32.2 ± 33.7 km², 28.7 ± 14.5 km², 14.8 ± 15.4 km², and 5.3 ± 3.7 km² for 95% MCP, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively. Median 95% KDE-href size for low performance tags was 61.8 km².

Boundary filtered SSM mean home range sizes for the 12 high performance tags were 5.5 ± 4.0 km², 11.0 ± 7.4 km², 5.2 ± 2.8 km², 1.9 ± 1.2 km², and 0.8 ± 0.6 km² for 95% MCP, 95% KDE-href, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively. Boundary filtering reduced home range sizes for high performance tags an average of 3.4%, 2.8%, 8.6%, 3.2%, and 11.4% for 95% MCP, 95% KDE-href, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively (Table 1). Boundary filtered SSM mean home range sizes for low performance tags were 8.4 ± 7.0 km², 20.5 ± 13.3 km², 17.0 ± 9.1 km², 4.9 ± 3.5 km², and 4.0 ± 2.2 km² for 95% MCP, 95% KDE-href, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively. Boundary filtering reduced home range sizes for low performance tags by 55%, 56%, 38%, 47%, and 12% for 95% MCP, 95% KDE-href, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively (Table 2).

Table 1.

Home ranges (km²) estimated from unfiltered SSM and boundary filtered SSM data for JB high performance tags

ID	95% MCP			95% KDE-href			50% KDE-href			95% KDE-LSCV			50% KDE-LSCV
	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change
243573	10.7	10.3	-4.3	19.9	16.6	-16.6	2.2	2.0	-12.1	3.1	2.1	-33.6	0.2	0.2	7.2
243574	10.7	10.8	0.7	17.3	17.9	3.4	2.8	2.6	-6.1	9.9	9.9	-0.7	1.5	1.4	-7.6
243575	1.4	1.2	-18.3	2.8	2.5	-10.1	0.7	0.7	-3.5	2.3	2.7	14.6	0.6	0.7	16.8
243576	3.6	2.5	-30.6	8.5	7.2	-15.4	1.2	1.0	-13.3	3.5	4.1	17.3	0.3	0.2	-29.3
243577	2.8	2.4	-15.3	9.4	8.3	-11.2	1.4	1.3	-8.3	5.5	2.9	-47.0	0.3	0.2	-35.4
266878	4.1	5.0	21.9	9.7	11.2	14.7	1.3	1.5	15.9	4.5	4.3	-6.1	0.5	0.4	-27.4
266879	14.6	12.5	-14.0	30.3	28.7	-5.0	5.1	4.8	-5.3	10.1	8.2	-18.6	0.8	0.6	-33.5
266880	4.3	4.8	13.1	8.5	9.2	8.5	2.6	2.5	-4.0	6.8	7.3	6.6	1.7	1.8	5.0
266882	1.3	1.3	-4.1	3.2	3.4	6.4	0.6	0.6	4.8	2.5	2.4	-2.5	0.5	0.5	-1.5
266883	6.1	6.8	12.1	11.1	10.4	-5.9	1.9	1.8	-3.9	8.3	5.6	-31.8	1.3	0.9	-29.2
266884	7.3	6.9	-4.8	13.1	12.8	-1.9	2.7	2.7	-1.1	9.4	9.5	1.2	1.7	1.7	0.8
266885	1.8	1.8	2.3	4.0	4.0	-0.4	0.9	0.9	-1.2	3.9	3.8	-2.3	0.9	0.8	-2.9
Mean	5.7	5.5	-3.4	11.5	11.0	-2.8	1.9	1.9	-3.2	5.8	5.2	-8.6	0.9	0.8	-11.4

Percent change shows difference in home range size after boundary filtering

Table 2.

Home ranges (km²) from boundary filtered and unfiltered SSM for JB low performance tags

ID	95% MCP			95% KDE-href			50% KDE-href			95% KDE-LSCV			50% KDE-LSCV
	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change
243578	5.9	3.6	-39.0	22.1	9.8	-55.7	2.8	1.3	-53.6	7.7	5.8	-24.7	0.7	0.8	14.3
243579	26.8	18.9	-29.5	61.8	41.1	-33.5	14.9	10.3	-30.9	48.1	32.4	-32.6	11.6	7.4	-36.2
266881	7	3.7	-47.1	23.5	15.9	-32.3	4.5	3.9	-13.3	20.1	16	-20.4	3.8	3.9	2.6
266886	88.8	3.3	-96.3	258.7	9.7	-96.3	40.9	2.8	-93.2	26.3	10.8	-58.9	3.5	3	-14.3
266887	32.5	12.2	-62.5	65.5	25.8	-60.6	11.1	6.4	-42.3	41.2	19.9	-51.7	6.7	4.8	-28.4
Mean	32.2	8.4	-54.9	86.3	20.5	-55.7	14.8	4.9	-46.7	28.7	17	-37.7	5.3	4	-12.4

Percent change shows difference in home range size after boundary filtering

Using unfiltered LC 3 data points alone, JB terrapin mean home range sizes were 6.9 ± 4.0 km², 10.5 ± 5.5 km²,1.8 ± 0.8 km² for 95% MCP, 95% KDE-href, and 50% KDE-href, respectively (Table 3). Median home range size was 7.8 km² for 95% KDE-LSCV and 0.9 km for 50% KDE-LSCV. Location class 3 home ranges were not estimated for the five low performance tags because < 10 LC 3 fixes were collected.

Table 3.

LC 3 home ranges (km²) for JB high performance tags

ID	Year	Site	# of LC 3 locations	95% MCP	95% KDE-href	50% KDE-href	95% KDE-LSCV	50% KDE-LSCV
243573	2023	JFK	142	5.9	11.7	1.4	9.1	0.7
243574	2023	RBH	166	9.6	15.8	2.5	11.0	1.7
243575	2023	RBH	184	2.8	4.4	1.1	3.9	0.9
243576	2023	JFK	106	3.9	7	1.1	3.6	0.5
243577	2023	JFK	167	3.1	5.4	1.1	3.5	0.6
266878	2024	RBH	165	8	11.5	1.8	6.8	1.1
266879	2024	RBH	156	15.4	23.3	2.7	8.4	0.7
266880	2024	JFK	177	8.5	12.1	2.8	10.4	2.4
266882	2024	JFK	196	2.9	4.6	0.7	3.5	0.5
266883	2024	RBH	168	9.8	11.9	2.2	10.4	1.9
266884	2024	RBH	180	10	12.8	2.6	11.3	2.2
266885	2024	RBH	169	3.2	5.8	1.2	4.3	1.0
Mean			173	6.9	10.5	1.8	7.2	1.2

The average distance between JB high performance tag unfiltered and filtered SSM home range centroids was 231 ± 187 m (95% MCP), 88 ± 76 m (95% KDE), and 40 ± 27 m (50% KDE). The average distance between LC 3 and unfiltered SSM home range centroids was 283 ± 253 m (95% MCP), 236 ± 182 m (95% KDE), and 111 ± 115 m (50% KDE). The average distance between LC 3 and boundary filtered SSM home range centroids was 354 ± 428 m (95% MCP), 266 ± 234 m (95% KDE), and 103 ± 98 m (50% KDE). Home ranges estimated from the three methods shared high proportions of spatial overlap with averages between 80 and 100% (Fig. 4). Paired t-tests comparing 95% MCP, 95% KDE, and 50% KDE showed no significant differences between home range areas estimated from boundary filtering and those estimated without filtering (Table 6). Repeated measures ANOVAs revealed no significant differences between sizes of 95% MCP, 95% KDE, and 50% KDE between the three methods except for log transformed 50% KDE-LSCV (Table 7). Boundary filtered (X = 0.79 km², SD = 0.54) and unfiltered (X = 0.87 km², SD = 0.53) 50% KDE-LSCV were not significantly different (p = 0.0968), but both were significantly smaller compared to LC 3 (X = 1.18 km², SD = 0.66, p = 0.0035, 0.0197).

Fig. 4.

Example of home ranges for one terrapin estimated from unfiltered SSM (blue), boundary filtered SSM (green), and LC 3 (red). Black dots indicate raw LC 3 location fixes only. (A) 95% MCP (B) 95% KDE-href (C) 95% KDE-LSCV (D) 50% KDE-href (E) 50% KDE-LSCV

Table 6.

Paired t-test results comparing terrapin home range sizes estimated from unfiltered SSM and boundary filtered SSM

Home range	Location	Tag performance	t-statistic	df	p-value
95% MCP	JB	High	-0.85036	11	0.4132
95% KDE-href	JB	High	-1.2589	11	0.2341
95% KDE-LSCV	JB	High	-1.784	11	0.102
50% KDE-href	JB	High	-2.137	11	0.05589
50% KDE-LSCV	JB	High	-1.9398	11	0.07847
95% MCP	JB	Low	-1.5151	4	0.2043
95% KDE-href *	JB	Low	-2.1533	4	0.09762
95% KDE-LSCV	JB	Low	-3.1548	4	0.03436
50% KDE-href	JB	Low	-1.3933	4	0.236
50% KDE-LSCV	JB	Low	-1.5988	4	0.1851
95% MCP	OBSP	High	-2.1154	3	0.1247
95% KDE-href	OBSP	High	-1.5512	3	0.2187
95% KDE-LSCV	OBSP	High	-6.1225	3	0.008759
50% KDE-href	OBSP	High	-1.4054	3	0.2545
50% KDE-LSCV	OBSP	High	0.474	3	0.6679

* indicates data was log transformed

Table 7.

Repeated measures ANOVA results comparing home ranges estimated from unfiltered SSM, boundary filtered SSM, and LC 3

	Location	df	F-statistic	ges	p-value
95% MCP	JB	1.16, 12.73	3.55	0.025	0.078
95% KDE-href	JB	1.01, 11.13	0.94	0.04	0.355
50% KDE-href	JB	1.03, 11.28	0.47	0.004	0.511
50% KDE-LSCV*	JB	1.41, 15.46	13.43	0.095	0.001
Total Distance Traveled	JB	1.03, 10.28	171.44	0.852	< 0.001
Average Distance Traveled Per Day	JB	1.03, 10.28	171.52	0.852	< 0.001
Maximum Distance Traveled	JB	1.14, 11.40	86.16	0.690	< 0.001
95% MCP*	OBSP	1.10, 3.29	3.45	0.097	0.153
95% KDE-href*	OBSP	1.01, 3.03	1.00	0.196	0.372
95% KDE-LSCV	OBSP	1.01, 3.02	43.52	0.164	0.007
50% KDE-href*	OBSP	1.22, 3.66	1.96	0.034	0.248
50% KDE-LSCV	OBSP	1.03, 3.09	0.04	0.008	0.859
Total Distance Traveled	OBSP	1.00, 2.00	0.43	0.048	0.580
Average Distance Traveled Per Day	OBSP	1.19, 2.38	0.15	0.019	0.768
Maximum Distance Traveled	OBSP	1.12, 2.24	33.94	0.752	0.022

A Greenhouse-Geisser (GG) correction was utilized because the sphericity assumption was violated. * Indicates data was transformed with a square root function. JB 95% KDE-LSCV data could not be normalized and was excluded

The average distance between JB low performance tag unfiltered and filtered SSM home range centroids was 1,534 ± 2,111 m (95% MCP), 1,844 ± 2,247 m (95% KDE), and 530 ± 683 m (50% KDE). Boundary filtered home ranges were located completely within larger unfiltered home ranges and covered an average of 44% of unfiltered home range area (95% MCP, 95% KDE, 50% KDE). Paired t-tests comparing 95% MCP, 95% KDE, and 50% KDE showed no significant differences between home range areas estimated from boundary filtering and those estimated without filtering except for 95% KDE-LSCV (Table 6). Unfiltered 95% KDE-LSCV (X = 28.67 km², SD = 14.49) were significantly larger than boundary filtered 95% KDE-LSCV (X = 16.98 km², SD = 9.06).

Orient Beach State Park

Shapiro-Wilk tests confirmed normality for all OBSP home range sizes except for LC 3 95% MCP, 95% KDE-href and 50% KDE-href (Additional File 1: Table S5). Unfiltered SSM mean home range sizes were 8.4 ± 4.9 km², 19.9 ± 15.3 km², 10.2 ± 3.2 km², 4.6 ± 3.8 km², and 1.7 ± 0.5 km² for 95% MCP, 95% KDE-href, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively. Boundary filtered SSM mean home range sizes and ranges were 6.7 ± 3.4 km², 14.6 ± 8.6 km², 8.7 ± 2.9 km², 3.7 ± 2.5 km², and 1.9 ± 0.7 km² for 95% MCP, 95% KDE-href, 95% KDE-LSCV, 50% KDE-href, and 50% KDE-LSCV, respectively. Removing unlikely locations using the OBSP boundary reduced home range sizes an average of 17%, 21%, 15%, and 16% for 95% MCP, 95% KDE-href, 95% KDE-LSCV, and 50% KDE-href, respectively (Table 4). Removing locations increased home range size an average of 8% for 50% KDE-LSCV. Orient Beach State Park mean home range sizes from LC 3 were 11.8 ± 2.5 km² and 1.8 ± 0.8 km² for 95% KDE-LSCV and 50% KDE-LSCV, respectively (Table 5). Median home range size was 3.4 km² for 50% KDE-href, 15.5 km² for 95% KDE-href, and 7.7 km² for 95% MCP.

Table 4.

OBSP home ranges (km²) estimated from boundary filtered SSM and unfiltered SSM data

ID	95% MCP			95% KDE-href			50% KDE-href			95% KDE-LSCV			50% KDE-LSCV
	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change	No Filter	Filter	% Change
263461	3.2	3.1	-3.1	8.3	6.5	-21.7	1.9	1.6	-15.8	6.6	5.4	-18.2	1.5	1.3	-13.3
263462	8.6	6.1	-29.1	15.4	13.0	-15.6	3.6	3.4	-5.6	11.4	9.6	-15.8	2.6	2.4	-7.7
263463	14.9	11.3	-24.2	42.4	26.8	-36.8	10.3	7.3	-29.1	14.9	12.9	-13.4	1.7	2.7	58.8
263466	7.0	6.3	-10.0	13.4	11.9	-11.2	2.7	2.3	-14.8	7.8	6.8	-12.8	1.3	1.2	-7.7
Mean	8.4	6.7	-16.6	19.9	14.6	-21.3	4.6	3.7	-16.3	10.2	8.7	-15.1	1.7	1.9	7.5

Percent change shows difference in home range size after boundary filtering

Table 5.

LC 3 home ranges (km²) for OBSP

ID	Year	Site	# of LC 3 locations	95% MCP	95% KDE-href	50% KDE-href	95% KDE-LSCV	50% KDE-LSCV
263461	2024	OBSP	115	6.6	12.4	2.8	8.9	2.1
263462	2024	OBSP	82	7.4	16	3.5	12.3	2.6
263463	2024	OBSP	102	15.3	51.9	7.3	15.6	0.6
263466	2024	OBSP	98	7.9	14.9	3.2	10.4	2.0
Mean			99	9.3	23.8	4.2	11.8	1.8

The average distance between OBSP filtered and unfiltered SSM home range centroids was 269 ± 109 m (95% MCP), 283 ± 111 m (95% KDE), and 106 ± 47 m (50% KDE). The average distance between unfiltered SSM and LC 3 home range centroids was 364 ± 223 m (95% MCP), 705 ± 586 m (95% KDE), and 227 ± 113 m (50% KDE). The average distance between boundary filtered SSM and LC 3 home range centroids was 543 ± 121 m (95% MCP), 761 ± 610 m (95% KDE), and 425 ± 192 m (50% KDE). Home ranges estimated from the three methods shared high proportions of spatial overlap, typically ranging from 70 to 100%. Paired t-tests comparing 95% MCP, 95% KDE, and 50% KDE showed no significant difference between home ranges estimated from boundary filtering and those estimated without filtering except for 95% KDE-LSCV (Table 6). Unfiltered 95% KDE-LSCV (X = 10.15 km², SD = 3.24) were significantly larger than boundary filtered 95% KDE-LSCV (X = 8.66 km², SD = 2.86). Repeated measures ANOVAs revealed that there were no significant differences between sizes of 95% MCP, 95% KDE, and 50% KDE between the three methods except for 95% KDE-LSCV (Table 7). Location class 3 95% KDE-LSCV (X = 11.8 km², SD = 2.52) were significantly larger than boundary filtered 95% KDE-LSCV (X = 8.66 km², SD = 2.86, p = 0.0019). Filtered and unfiltered 95% KDE-LSCV were also significantly different (p = 0.0178).

Mean total distance traveled

Shapiro-Wilk tests confirmed normality of total distances traveled between June 26 and August 30 and a repeated measures ANOVA was conducted (Additional File 1: Table S5; Table 7). Differences in JB mean total distance traveled between unfiltered SSM (X = 50.3 ± 10.4 km, n = 11) and boundary filtered SSM (50.6 ± 10.4 km, n = 11) were not significant (p = 0.9298), but both were significantly lower than the mean total distance traveled calculated from LC 3 data points (109.7 ± 15.2 km, n = 11, p = < 0.0001) (Additional file 1: Table S3). Differences in OBSP mean total distance traveled between unfiltered SSM (69.6 ± 16.3 km, n = 3, 55.4–87.4 km), boundary filtered SSM (64.8 ± 14.9 km, n = 3), and LC 3 (70.0 ± 2.6 km, n = 3) were not significantly different (p = 0.580) (Additional file 1: Table S4).

Average distance traveled per day

Shapiro-Wilk tests confirmed that daily movement rates were normally distributed and repeated measures ANOVAs were conducted (Additional File 1: Table S5; Table 7). The difference in average distance traveled per day in JB between unfiltered SSM (X = 0.76 ± 0.16 km) and boundary filtered SSM (X = 0.77 ± 0.16 km) was not significant (p = 0.9291), but both were significantly lower than LC 3 (X = 1.7 ± 0.23 km, p = < 0.0001) (Additional File 1: Table S3). The difference in average distance traveled per day in OBSP between unfiltered SSM (X = 1.1 ± 0.25 km), boundary filtered SSM (X = 0.98 ± 0.23 km), and LC 3 (X = 1.1 ± 0.04 km) was not significantly different (p = 0.768) (Additional File 1: Table S4).

Maximum distance traveled in one day

The estimates for maximum straight-line distance traveled in one day for JB were 5.6 km using unfiltered SSM, 5.3 km using boundary filtered SSM, and 8 km using LC 3 (Additional File 1: Table S3). Estimates for OBSP were 4 km for both unfiltered and boundary filtered SSM and 7 km using LC 3 (Additional File 1: Table S4). A Shapiro-Wilk test confirmed that data was normally distributed and repeated measures ANOVAs were conducted (Additional File 1: Table S5; Table 7). Jamaica Bay maximum distance traveled was not significantly different between filtered and unfiltered (p = 0.9893), but both were significantly different than maximum distance estimated from LC 3 (p = < 0.0001). Orient Beach State Park maximum distance traveled was not significantly different between unfiltered and filtered (p = 0.2515), but unfiltered (p = 0.0488) and filtered (p = 0.0499) were slightly significant compared to maximum distance estimated from LC 3.

Discussion

This study demonstrates that satellite tags are suitable for tracking terrapins and have significant advantages over other telemetry methods such as producing larger datasets, operating through multiple seasons, and no time required to relocate tagged individuals. Disadvantages of using Argos satellite tags to track terrapins is the generation of unlikely fixes such as those on land in NYC and in the Atlantic Ocean and the inability to track male and juvenile terrapins due to their smaller size. Radio telemetry can also produce erroneous locations, but these are often recognized and filtered out in the field if the tagged animal can be located. Although Argos satellite tags have relatively large error estimates, techniques are available (boundary filtering and the SSM) to identify and remove unlikely locations and estimate the most probable locations. Other terrapin tracking studies could utilize boundary filtering to prevent home range overestimations when sufficient high-quality locations (LC 1–3) are obtained. However, caution is needed while filtering smaller datasets where a greater proportion of data may be excluded. Although boundary filtering removed ~ 50% of data from low performance tags, this method produced more biologically accurate home ranges that did not include urbanized areas and the open ocean; therefore, we recommend this approach to obtain as much valuable information as possible from low quality location data. Similar approaches for removing unlikely or poor-quality locations have been used in satellite tracking studies on other taxa [13, 77–80].

Determining how to draw boundaries to filter out low-quality or unlikely locations depends on research goals and the researcher’s knowledge of the study organism’s habitat and behaviors. Boundaries used in this study included approximately 100 m of shoreline for nesting and excluded the Atlantic Ocean. Removing unlikely locations in urban environments is a simpler process than removing locations from non-urban environments. To construct the JB boundary, large portions of shorelines could be traced along solid features such as bulkheads. Poor quality locations outside of the boundary were in Queens, Brooklyn, Manhattan, and Nassau County and obviously inaccessible for terrapins. In contrast, OBSP shorelines are mostly undeveloped and include more tidal wetlands, thus, more care was required drawing the OBSP boundary. Additionally, designing the OBSP boundary was further complicated as terrapins utilize both Long Beach Bay and Gardiner’s Bay, unlike terrapins restricted to JB.

Boundary filtering for OBSP removed an average of 18% of fixes per tag compared to 6% at JB, probably because JB is approximately 5x larger than Long Beach Bay where terrapins were located in OBSP. Due to JB’s larger area, more fixes were located within the bay rather than on land and thus were included with the boundary filtered dataset. The ability to determine which salt marshes terrapins utilized in JB was likely enhanced because of JB’s larger area (25 km²). We were unable to determine specific areas utilized by OBSP terrapins likely because Long Beach Bay is only about 5 km² and Argos error radii cover a higher proportion of Long Beach Bay than JB (Fig. 1). This suggests that tracking species with Argos satellite tags in smaller areas is generally less informative and may not be appropriate if the study’s goal is to determine fine-scale movements and space use patterns. Another possibility to consider is that OBSP terrapins are using most of the habitat available in Long Beach Bay whereas individual JB terrapins are not utilizing the entire bay because its area is larger.

Comparisons of unfiltered SSM, boundary filtered SSM, and LC 3 home range estimates found similarities in size, locations of centroids, and spatial overlap. These results suggest that it may be feasible to estimate home ranges using any of these three methods, depending on study goals. Other studies have successfully estimated home ranges using only LC 3 data; Fujisaki et al. (2014) tracked alligators with both VHF radio and Argos tags and used only LC 3 fixes to estimate home ranges; home ranges generated from both tags covered the same core-use areas [35]. Estimating home ranges from LC 3 excluded 84% of the total data for JB and 94% for OBSP. We recommend applying a SSM before estimating home range because this avoids data loss and the SSMs are designed for Argos data. We only recommend boundary filtering if tag performance is especially low or if the study area is small (~ 5 km²) in which many location fixes are located on land.

The estimated distances terrapins traveled in this study provide further evidence that unfiltered SSM or boundary filtered SSM may be used to analyze data. Boundary filtered SSM movement tracks produced similar estimates of distance traveled per day as unfiltered SSM tracks at both study sites, suggesting that location removal does not change distance calculations. Our estimates are not very different from median travel distances of 1–2 km/day from June-August for terrapins tracked with acoustic telemetry [30]. Both unfiltered and filtered SSM methods estimated similar average summer distances; furthermore, the date of the maximum distance traveled in a day by a terrapin was the same for SSM methods at each site. Distance estimates from LC 3 fixes at JB were about 2x higher than those produced from the unfiltered and boundary filtered SSMs. This increase is likely because an average of 173 LC 3 fixes were used in the distance calculations compared to 73 daily mean locations from the SSM. Unlike JB, OBSP movement estimates between the three analytical techniques were not significantly different. An average of 99 LC 3 fixes were used to estimate distance in OBSP, compared to 73 daily mean locations, which could cause similar estimates for distance traveled.

Female terrapin home range estimates in this study were 6-8x larger than radio tracking studies that estimated mean 95% MCP and KDE sizes of < 1 km² [28, 29] and 3 km² [34]. The 50% KDE estimates in this study are most similar in size to the home ranges from previous terrapin studies. Our results are generally similar to those of terrapins tracked with satellite tags in Florida, with mean home range sizes of 9.4 km² (95% MCP), 8.1 km² (95% KDE-LSCV), and 1.3 km² (50% KDE-LSCV) [14]. The similarity in home range size between these two sites is especially surprising given that they are in such different habitats and > 1600 km apart. Satellite telemetry has similarly produced larger home range estimates than VHF radio telemetry for yellow-spotted river turtles (Podocnemis unifilis) and gopher tortoises (Gopherus polyphemus) and for green sea turtles (Chelonia mydas) tracked via acoustic telemetry [13, 17, 81–84].

Small numbers of location fixes collected by VHF and acoustic transmitters could cause an underestimation of home range size. Our highest performing satellite tags collected an average of 11–21 fixes/day during the summer, while radio tracking was typically conducted 1–3 times/week in other terrapin studies [28, 29]. Sixteen out of twenty-one of our tags collected > 800 fixes whereas radio tags typically collect < 100 [85]. Increasing the number of fixes used to estimate home range can increase home range accuracy [46, 47, 86–88]. Additionally, VHF radio telemetry may miss occasional exploratory and nesting forays. Skupien et al. (2016) reported that GPS tags showed American alligators using marsh habitat while VHF tags did not, however, VHF tags did detect alligators underground while GPS tags failed to do so [48]. It is important to note that larger error radii must still be considered for an increase in home range size.

Conclusion

Advancements in satellite telemetry have led to the development of small Argos tags that are suitable for tracking adult female diamond-backed terrapins despite large error estimates. State-space analytical models commonly applied to marine tracking data can be applied to terrapin data to account for location errors and estimate true locations. Although SSMs are typically used for marine species traveling hundreds to thousands of kilometers, the model revealed important movement patterns for terrapins that stayed within a 25 km² area in our study. The movement metrics produced by satellite tags are fundamentally different from those produced by previous tracking and analytical techniques and probably should not be directly compared to results of VHF, sonic tag, and mark-recapture studies. Although this new method is available for tracking terrapins, radio telemetry may still be useful depending on research goals, study area and other factors because it can provide more accurate location fixes. Satellite tracking technology will likely continue improving, producing tags with higher accuracy, such as GPS tags, in the near future. Tag size may also continue to decrease, providing an opportunity to use satellite tags to track male and juvenile terrapins. Understanding terrapin spatial ecology is vital as populations are threatened by habitat loss and rising sea levels. Our results indicate larger female terrapin home ranges and distances traveled than previous tracking studies, which may be important to consider when implementing conservation measures such as defining critical habitat or establishing fishing and vessel activity closures.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

MS and RB acquired resources for this project and designed the study. MS, KH, and RB conducted field investigations. KH analyzed data and led writing of the manuscript. All authors contributed to revisions and final submission approval.

Funding

Funding for this work was provided by Hofstra University’s Faculty Research and Development Grants, Presidential Research Awards, and a Regional Conservation Network Grant from The Northeast Association of Fish and Wildlife Agencies Fish and Wildlife Diversity Technical Committee and the Peconic Estuary Partnership mini-grant program. None of the funding sources had any role in the design of the study or collection, analysis, or interpretation of data, or in writing the manuscript.

Data availability

The datasets analyzed for this study are not publicly available because diamond-backed terrapins are an exploited species but are available from the lead author on reasonable request via email at khough1@pride.hofstra.edu.

Declarations

Ethics approval and consent to participate

Ethical approval was provided by the Hofstra University Institutional Animal Care and Use Committee. All methods were carried out in accordance with relevant guidelines and regulations from the Hofstra University Institutional Animal Care and Use Committee (permit #22.23-09), the New York State Department of Environmental Conservation (permit LCP #1640), the New York State Department of Parks, Recreation, and Historic Preservation (permit #24–0239) and the USDI National Park Service (permit GATE-2023-SCI-0020).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.