Study Area

Our study site was located along a 13.7-km reach of a river within the eastern panhandle of West Virginia. Agricultural fields, consisting of hay fields, cornfields, and cattle pastures, bordered most of the river along this reach. Although our study area was primarily agriculture, the watershed was mostly forested with scattered agriculture and sparse residential development. West Virginia's Ridge and Valley region averages 76 cm of precipitation annually (Kozar and Mathes 2001).

Sampling

We surveyed 5 primary sites corresponding with a concurrent river restoration project (Selego et al. 2012). Study sites were representative of the overall conditions of the river, and search effort was consistent across sites. Our surveys varied from 600 to 1100 m of river length and included up to 150 m perpendicularly from the river's edge on both sides of the river; the sites were 300 to 1000 m apart from their neighboring sites. The lengths and distances matched survey constraints from the restoration project (e.g., 250-m separation of bird surveys), and the site widths encompassed 95% of freshwater turtle migration distances (Bodie 2001). We conducted surveys from May 2009 to August 2011. We opportunistically captured turtles by hand or dipnet as the sites were intensively surveyed on foot or by canoe. We marked adult turtles with a triangular file (Cagle 1939); juveniles < 2 yrs old were similarly marked with white enamel to avoid early physical alteration of the young turtles by use of a marking file. We determined sex of the turtles by observation of male secondary sexual characteristics (i.e., longer precloacal tail lengths and a concave plastron) in the smallest males (Lovich et al. 1990; Brooks et al. 1992) and in females of equivalent size (≥ 160 mm in carapace length; Harding and Bloomer 1979). To determine age, we counted annual carapace scute rings (Harding and Bloomer 1979), up to about 20 yrs, after which the rings begin to fade. For every capture, we recorded date, time, location (± 4 m), weather conditions, observed activity when captured, identification marks, and perpendicular distance from the river. We released all wood turtles at their points of capture after they had been marked.

Radiotelemetry

We tracked 31 wood turtles (15 males, 10 females, and 6 juveniles) by radiotelemetry to provide specific movement information. We tracked relatively equal proportions of turtles at each site (i.e., reference sites, n = 9; control sites, n = 9; and restoration reach, n = 13) with 3-element Yagi antennas and R2000 receiver (ATS, Isanti, MN). We applied transmitters (ATS R1860, 15 g, well below 10% turtle mass) with epoxy to the back right edges of the carapaces. We observed all tagged turtles initially to make certain that they could move without hindrance and that transmitters did not interfere with their regular activities. We conducted tracking 1 to 2 times per week between 0800 to 1930 hrs from March to October and once a month from November to February each year.

Vegetation Surveys

We conducted vegetation surveys within 10 × 10-m plots centered on a random sample (n = 110) of the overall turtle captures. During primary plant growth and production in spring and summer, 2009 to 2011, we randomly generated a number between 1 and 31 (n = 31 tracked turtles) each radio-telemetry day to determine which turtle capture of the day would have an associated vegetation survey, regardless of the turtle captured being marked, unmarked, or radio-tagged. We captured unmarked turtles and changed the beginning tracking site almost every telemetry day, which reduced the probability of the same turtles being associated with the vegetation plots. Within the plots, we identified and measured all trees with diameters > 5 cm for their diameter at breast height (dbh). We identified all shrubs, including saplings, taller than 1 m and < 5 cm in diameter and counted their number of stems. We surveyed the field layer (all plants, woody and herbaceous, < 1 m tall) in 1-m² subplots in each corner of the 10 × 10-m plot. We identified and estimated percent cover of each species in the field layer, as well as leaf litter, woody debris, and bare ground and rock. At the center of each subplot, we used a Robel pole (Robel et al. 1970) to determine vegetative vertical density from visual obstruction readings taken 4 m from the pole (1 m height) in all 4 cardinal directions and averaged for the plot. From the center of the 10 × 10-m plot, we documented canopy cover (%) and a description of the local cover type (e.g., pasture, crop field, wetland, forest). We paired random vegetation plots (n = 110) with each turtle plot, randomly located up to 300 m in a random cardinal direction from the turtle plots.

Statistical Analyses

To estimate population size, we used the POPAN model in Program Mark (Version 8; Cooch and White 2016). To avoid violating the assumption of equal probability of capture, we excluded the 31 radio-tracked turtles from the population size analysis. We performed all other statistical analyses in R 2.10.1 with α = 0.05 (R Development Core Team 2008). We used independent t-tests, comparing sample means, to determine whether the adult male:female sex ratio and juvenile:adult ratio differed significantly from 1:1.

We calculated home range sizes using 50% and 95% minimum convex polygons (MCP; Adehabitat package, R) for each turtle tracked > 10 times over the study period. We used a 1-way ANOVA to determine whether there was a difference among the home range sizes of males, females, and juveniles. After conducting residual diagnostics, we discovered one outlier (female, 95% home range: 36.97 ha) in the data set; after removal of the outlier, the data satisfied test assumptions required for statistical analyses. Using ArcMap software (Environmental Systems Research Institute, Inc., Redlands, CA), we uploaded the radiotelemetry capture locations and measured the distances from the furthest terrestrial point of each tracked turtle to the river's edge. We used a 1-way ANOVA to determine whether the maximum overland distance traveled from the river by males, females, and juveniles differed; transformation of the response was not needed to meet assumptions. We used Tukey tests at 5% significance level to determine how the sexes differed.

We used multivariate analysis of variance (MANOVA) with Pillai test statistic to determine whether structural (i.e., canopy cover, bare ground and rock cover, leaf litter cover, woody debris cover, and vertical density) differences existed in vegetation among plot type (i.e., turtle and random), seasons (i.e., spring and summer), years (i.e., 2009, 2010, and 2011), and plot type × year or season interactions. MANOVA assumes joint multivariate normality, equal variances, and independent observations (Finch 2005). We square-root transformed percent cover of bare ground and rock, leaf litter, and woody debris; log-transformed vertical density; and inverse log-transformed canopy cover to approximate normality. If the global MANOVA indicated significance, we used the structural variables in univariate ANOVAs (Bonferroni correction: α = 0.05/5 structural variable tests = 0.01) with Tukey HSD tests to determine where the significance occurred.

To determine whether overall and native community composition differed among plot types, seasons, years, and their interactions, we used permutational multivariate analysis of variance (PerMANOVA, 1000 permutations; adonis function, vegan package, R) with Bonferroni correction (α = 0.05/3 tests = 0.017) on the field, shrub, and tree layers separately. PerMANOVA is robust to departures from parametric distribution assumptions and suitable for community composition analysis (Walters and Coen 2006; Lorion and Kennedy 2009). For significant main effects (i.e., plot type, season, or year), we used indicator species analysis (ISA, indval function, labdsv package, R) to determine which species were more likely to occur, calculating the indicator values for each species by taking the product of its relative frequency and its relative average abundance within the plot types (Dufrêne and Legendre 1997). An indicator value threshold of 0.25 and α = 0.05 (p-values generated through randomization procedures; Dufrêne and Legendre 1997) were used to determine which species characterized the plot types, seasons, and years. We plotted significant main effects and species with nonmetric multidimensional scaling (NMDS) ordination (metaMDS and envfit functions, vegan package, R).

Population Size and Sex Ratio

We captured 284 unique wood turtles (males = 137, females = 88, and juveniles = 59). Total captures numbered 1,443 (1,159 recaptures, 80%). We estimated the population size to be 793 (SE = 88.75) individuals. Males > 20 yrs old accounted for 76.6% (n = 105) of all male captures. Females > 20 yrs old accounted for 58% (n = 51) of all female captures. Because of wearing of the annual rings in wood turtles > 20 yrs of age, an average age could not be determined for the adult turtles. Juveniles ranged in age from 0 (hatchling) to 8 yrs, the average being 4 yrs (SE = 0.29). The adult male:female sex ratio was 1.6:1 and did not differ statistically from 1:1 (t_5.95 = 0.91, p = 0.399). The juvenile:adult ratio (1:3.8) varied from 1:1 (t_13.46 = 2.76, p = 0.016).

Home Ranges

We tracked 31 turtles (15 males, 10 females, and 6 juveniles) to determine home range sizes. The turtles that were located > 10 times (i.e., 13 males, 9 females, and 2 juveniles; n = 24) were tracked 23 to 75 times (x¯ = 45.2, SE = 2.57) during June 2009 to August 2011. The 7 turtles with ≤ 10 locations were potentially attributable to early failure of the transmitters, the turtles walking out of range, depredation, collection and removal by people, or burial and death during early spring floods. The 50% home ranges varied in size from 0.09 to 4.63 ha (x¯ = 0.90 ± 0.19). Males (x¯ = 0.87 ± 0.33), females (x¯ = 1.05 ± 0.19), and juveniles (x¯ = 0.47 ± 0.10) had similar 50% home range sizes (F_2,21 = 0.31, p = 0.737). The 95% home ranges varied in size from 0.62 to 36.97 ha (x¯ = 5.75 ± 1.46). Mean 95% home range sizes of males (x¯ = 4.29 ± 0.78), females (x¯ = 11.03 ± 3.68), and juveniles (x¯ = 4.04 ± 2.39) were similar when the outlier (female, 95% home range: 36.97 ha) was included (F_2,21 = 0.86, p = 0.437) and when it was removed (F_2,20 = 0.06, p = 0.946).

Male home ranges tended to be elongated along the river, whereas female and juvenile home ranges encompassed a greater amount of terrestrial habitat. Agricultural land, including cornfields, active pastures, and hay fields, were traversed on a regular basis by marked and tracked turtles (19%), although most captures occurred within the forested riparian zone (52%) or were aquatic (29%). All individuals returned to their summer home ranges yearly after most individuals hibernated outside of their summer home ranges.

A 300-m buffer around our study site encompassed all but 2 of 1443 capture locations. Considering only tracked turtles, the mean maximum distances (m) traveled overland from the river differed between the sexes (F_2,28 = 7.26, p = 0.003). Mean male distances traveled (x¯ = 85.67 ± 19.67) did not differ from mean juvenile distances (x¯ = 30.5 ± 12.48) (p = 0.244, Tukey HSD test). However, mean distances that females traveled (x¯ = 139.8 ± 25.79) were greater than either mean male distances (p = 0.032, Tukey HSD test) or mean juvenile distances traveled (p = 0.003, Tukey HSD test).

Vegetation

We recorded 142 (73% native) plant species in the field (n = 128), shrub (n = 32), and tree (n = 33) layers; some species were recorded in multiple layers (McCoard 2012). Structural vegetative characteristics differed between turtle plots (n = 110) and random plots (n = 110; Pillai = 0.11, F_5,208 = 4.98, p < 0.001) and years (Pillai = 0.20, F_10,418 = 4.66, p < 0.001) but not between seasons, the plot type × season interaction, or the plot type × year interaction (p ≥ 0.084). Leaf litter (F_{1 or 2,212} ≥ 0.14, p ≥ 0.233) and woody debris (F_{1 or 2,212} ≥ 0.42, p ≥ 0.033) cover were similar among all variables (Table 1). Canopy cover differed by year, decreasing from 2009 to 2011 (F_2,212 = 11.49, p < 0.001). Turtle plots had less bare ground and rock cover (F_1,212 = 18.47, p < 0.001) and higher vertical density (F_1,212 = 8.09, p = 0.005) than random plots. Vertical density was lower in 2011 than in 2009 or 2010 (F_2,212 = 5.73, p = 0.004).

Table 1. Structural vegetative characteristics measured at random plots (n = 110) and wood turtle plots (n = 110) along a 13.7-km reach of river in West Virginia, from spring 2009 to summer 2011. The variables recorded within 1-m² plots at the corners of a 10 × 10-m plot within a 100-m radius survey circle were % canopy cover (CC), % bare ground and rock cover (BGR), % leaf litter cover (LL), % woody debris cover (WD), and vertical density (VD, cm). Measurements were averaged for the whole plot. Means followed by the same letter within a row for each main effect (plot type, season, year) are not significantly different (Bonferroni correction: α = 0.05/5 tests = 0.01).

View Fullscreen

Overall field layer composition was similar between years, and there was no plot × year interaction (both p ≥ 0.109), but field layer composition differed in the plot × season interaction (pseudo-F_1,212 = 4.58, p = 0.001). Native field layer composition was similar between plots, and there were no plot × season or plot × year interactions (all p ≥ 0.074), but native field layer differed by season (pseudo-F_1,212 = 8.16, p = 0.001) and year (pseudo-F_2,212 = 10.13, p = 0.001). Reed canary grass (Phalaris arundinacea; indicator value [IV] 0.38, p = 0.011) differentiated random plots and bedstraw (Galium spp.; IV 0.28, p = 0.002) differentiated turtle plots (Fig. 1). Reed canary grass (IV 0.42, p = 0.002) and ground ivy (Glechoma hederacea; IV 0.37, p = 0.006) characterized the overall field vegetation in spring, whereas bedstraw (IV 0.34, p = 0.001) and Japanese stilt grass (Microstegium vimineum; IV 0.45, p = 0.004) characterized the overall field vegetation in summer; reed canary grass and bedstraw also differentiated the native field vegetation (same seasons and values; Fig. 2). Reed canary grass and bedstraw influenced the overall interaction between plots and seasons. By year, reed canary grass (IV 0.55, p = 0.001) and wingstem (Verbesina alternifolia; IV 0.39, p = 0.022) characterized the native field vegetation in 2011, whereas sedges (Carex spp.; IV 0.37, p = 0.002) characterized the community in 2009. Overall shrub and tree composition was similar among plots, seasons, and years, and there were no plot × season or plot × year interactions (p ≥ 0.26).

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

Population Size and Sex Ratio

The population estimate for our wood turtles in the study was substantially higher than the population estimate nearly 20 yrs ago (287 to 337 individuals; Niederberger and Seidel 1999) from a smaller area in the same watershed. The sex ratios of adult wood turtles were not different from 1:1 in our study, Pennsylvania, New Jersey, Michigan, and Québec (Harding and Bloomer 1979; Farrell and Graham 1991; Kaufmann 1992b; Walde et al. 2003; Daigle and Jutras 2005). In New Jersey, wood turtles exhibited a 1.5:1 female skew (Harding and Bloomer 1979); our statistically nonsignificant result indicated a 1.6:1 male skew. Our juvenile:adult ratio was 1:3.8, a significant difference from 1:1, reflecting the difficulty we had in finding juveniles and possibly indicating a depression in recruitment in this agricultural site; we did not expect the ratio to be equal as juvenile recruitment of wood turtles is often low (Arvisais et al. 2002). In New Jersey, however, hatchling-and-juvenile:subadult-and-adult ratio was 1:1.2 (Farrell and Graham 1991).

Home Ranges

Wood turtle home ranges followed the stream channel (Strang 1983; Remsberg et al. 2006), a result we found to be particularly strong for males, whereas females and juveniles extended outward terrestrially. Home ranges averaged 28.3 ha in Québec (Arvisais et al. 2002), 24.3 ha in Algonquin Park, Canada (Quinn and Tate 1991), 3.3 ha with no significant difference between the sexes in Pennsylvania (Kaufmann 1995), 30.2 ha in Michigan (Remsberg et al. 2006), and 22.7 and 61.25 ha at 2 sites in Virginia (Sweeten 2008). Wood turtles returned to the same home ranges annually, as reported in previous studies (Strang 1983; Quinn and Tate 1991; Arvisais et al. 2002). Our home ranges were most similar to the home ranges observed in Pennsylvania; both studies had considerably smaller home ranges than those observed in other locations.

Vegetation

Wood turtles require a variety of cover types for annual activities (Quinn and Tate 1991; Kaufmann 1992a; Compton et al. 2002), with preference for bottomland areas (Strang 1983). Compton et al. (2002) developed models from a Maine wood turtle study suggesting that the turtles prefer dry, moderately forested habitats at the watershed scale and sparse forests with low canopy cover near water at the local scale. In our study, low bare ground and rock cover and high vegetative vertical density differentiated wood turtle vegetation plots from random plots, indicating a preference for tall, thick herbaceous vegetation that may provide cover, a variety of vegetative food, and draw in a diverse array of invertebrate prey. Wood turtles also were observed in areas with greater tree richness than at random sites (McCoard 2012), additionally indicating a preference for habitat complexity. Agricultural fields were used by the turtles; turtles, primarily females, were in active pastures and hayfields bordering the river (19% of locations), whereas males were found in cornfields (0.6% of locations). These wanderings, however, were usually within the turtles' home ranges and not temporary trips outside of the home range. Wood turtle hatchlings occurred regularly in New Jersey cornfields (Castellano et al. 2008). Sex differences in habitat use were exhibited by wood turtles: males were found in streams more frequently than were females (Kaufmann 1992a; Compton et al. 2002), whereas females spent more time in grass-sedge-forb associations (Kaufmann 1992a). Hatchlings also showed preferences for habitat, choosing stream entry points composed of red maple (Acer rubrum), alder (Alnus rugosa), silky dogwood (Cornus amomum), partridgeberry (Mitchella repens), rough bedstraw (Galium asprellum), and various mosses and grasses in New Hampshire (Tuttle and Carroll 2005). In Pennsylvania, the majority of terrestrial activity occurred in alder stands and grass-sedge-forb associations (Kaufmann 1992a), as well as stands of black birch (Betula lenta), oaks (Quercus spp.), and red maple (Strang 1983). In Canada, wood turtles occupied alder swale (30%), mixed forest (28%), and grassy areas (12%; Quinn and Tate 1991). In Québec, forest stands were typically young (16 yrs), short (1–4 m), had few trees (25%), a moderate upper shrub layer cover (35%), and low canopy cover (0%–50%; Arvisais et al. 2004).

Management Implications

If landowners and farmers are alerted to the presence and needs of turtles on their properties, populations may be better conserved on these lands (Kaufmann 1992a); compensation may be available through federal programs for landowners who establish and protect their riparian zones. We recommend that, in areas with large-scale agriculture along waterways where wood turtles occur, efforts should be made to create, manage, and maintain riparian buffers that can provide essential terrestrial habitat for the turtles to undergo all aspects of their life histories. Establishing protected buffer strips along streams occupied by wood turtles would aid in conservation of the species (Arvisais et al. 2002). Although nearly all of our wood turtle observations were within 300 m of the river's edge, establishing a buffer that large is unrealistic for active farmland; a 150-m riparian buffer would include estimated migration distances traveled from streams for the majority of freshwater turtles (Bodie 2001), with 10 m recommended as the minimum riparian buffer size adjacent to cultivated areas for wood turtles (Saumure et al. 2007). Because this is private property and actively farmed, we recommend 10-m-wide buffers should be maintained such that tall, thick, diverse herbaceous vegetation can grow to provide food and shelter for the turtles and such that farmers can continue to use the majority of their property for farming. Tree diversity should be managed to provide a variety of seed foods, along with woody debris and canopy cover to provide shelter and shade. For conservation to be effective for wood turtles, extensive areas covering all seasonal habitat types should be protected, because wood turtles can move long distances (Quinn and Tate 1991; Behler and Castellano 2005). The selection and chronological use of habitat by wood turtles should be a research focus (Arvisais et al. 2004). Additional studies on riparian zone use by turtles, especially those species of federal or international conservation concern, are necessary and encouraged (Arvisais et al. 2004); this includes wood turtles, a species threatened by riparian zone degradation (Compton et al. 2002).