Globally, many turtle species are in decline and their distribution, richness, and abundance are projected to be further altered under many future climate scenarios (Ihlow et al. 2012; Böhm et al. 2013). Detection is crucial to understanding current and projected status and distribution, and cryptic species or those with cryptic life stages, such as turtles, require proper survey methodology (Kellner and Swihart 2014; Monk 2014; Tesche and Hodges 2015). A significant weakness in understanding turtle population trends is the ability to identify appropriate species- and site-specific survey methods. Turtles are problematic study subjects that require an appreciation of their phenotypic plasticity, which can create conflicting monitoring strategies even within a single watershed (Willson et al. 2011; Snover et al. 2015).

The western pond turtle (Emys marmorata) embodies this species conservation challenge. Native from Mexico and extending north through Washington, E. marmorata is on many state and international conservation lists (Bury and Germano 2008); it is currently being reviewed for listing under the US Endangered Species Act after an unsuccessful petition in 1993 (US Fish and Wildlife Service [US FWS] 2015). As a habitat opportunist that requires land and water, E. marmorata occupies a variety of aquatic environments (riverine, palustrine, and lacustrine) that border diverse terrestrial landscapes such as urban natural areas, pasture, and protected mountain forests (Reese and Welsh 1997; Spinks and Shaffer 2005; Bury and Germano 2008; Pilliod et al. 2013). As a result of their wide geographic distribution, their breeding phenology, diet, habitat availability, and use vary considerably (Bury and Germano 2008). This diversity in ecology and behavior suggests a need to develop narrow, site-specific monitoring strategies that are conducive to the behavior of the target population. Thus, it is important to understand the environmental triggers that elicit detection-dependent behavioral and life-history responses to guide management. However, this problem has only recently garnered attention (Zaragoza et al. 2015).

It is difficult to determine how specific environmental variables influence turtle behavior and activity patterns and which of those variables are most important, but in many studies the overarching factor is temperature (Crawford et al. 1983; Millar et al. 2012; Currylow et al. 2013; Akins et al. 2014). Other variables also known to influence turtle behavior include increased cloud cover interfering with sense of direction (Bowne and White 2004), the onset of rain prompting increased activity (Tucker et al. 2015), foraging availability modifying migration routes (Godley et al. 2002), and intermittent or permanent hydrological conditions (Bondi and Marks 2013). Furthermore, the time of day and location of the sun (Crawford et al. 1983) and forest management practices (Currylow et al. 2013) can dramatically influence where turtles bask or hibernate–overwinter, respectively. As ectotherms, turtles modify their behavior and select habitats to reach optimal temperatures at which physiological processes operate most efficiently while balancing this need with predator avoidance, foraging, and mating (Crawford et al. 1983). For many aquatic turtles, microhabitat temperatures dictate where they are most likely to be found, which can have important repercussions for management and conservation strategies (e.g., Sternotherus odoratus, Picard et al. 2011; Malaclemys terrapin, Akins et al. 2014).

Failure to consider the species–habitat interaction can lead to an often overlooked source of error, that of imperfect detection (Kellner and Swihart 2014; Tesche and Hodges 2015). For example, if visual basking surveys result in negative detections for E. marmorata in areas lacking stereotypical basking habitats (i.e., large woody debris), investigators may falsely conclude turtle absence when turtles are instead basking on shorelines, rocks, and other substrates not visible or targeted in the surveys. Further, applying an incorrect survey method over a broad geographic region with similar habitat relationships could imply an exaggerated decline or widespread extirpation.

The primary purpose of our case study was to investigate the behavior of E. marmorata at 2 long-term monitoring sites in Sequoia National Park, California, that have a history of highly variable population estimates over nearly 20 yrs (Jeffcoach 2014). Jeffcoach (2014) found high variability in population estimates at both monitoring sites despite increasing within-season sampling efforts that were atypical of historic 1-day annual sampling. As a follow-up to these findings, we used high-resolution temperature data from turtle shells to obtain behavioral data and investigated the environmental factors that determine whether turtles are in aquatic or terrestrial habitat. Moreover, we summarize the diel behavioral patterns at locations representative of typical southern Sierra Nevada foothill stream habitat.

METHODS

Study Sites

This study was conducted at 2 locations in Sequoia National Park in Tulare County, California. Sycamore Creek (36°29′14.8″N, 118°50′48.8″W, NAD83; elevation 482 m) is a steep, intermittent stream that typically flows from December to June, with some residual pools persisting perennially during years with more precipitation. The North Fork Kaweah River (36°32′50.5″N, 118°53′56.3″W, NAD83; elevation 523 m) is a low-gradient, perennial river with deep, stratified pools (Fig. 1). At both study sites, the dominant riparian vegetation is California sycamore (Plantanus racemosa), white alder (Alnus rhomifolia), and red willow (Salix laevigata). The upland forests are composed of interior live oak (Quercus wislizeni), California buckeye (Aesculus californica), and blue oak (Quercus douglasii), with brome (Bromus spp.) as the predominant ground cover (National Park Service [NPS] 2009).

View Figure

• Download as Powerpoint
• Download Figure

Field Sampling of Turtles

From March 2014 through October 2015, turtles were captured by hand via snorkeling in pool habitat, but we also opportunistically captured some turtles when they were visible out of the water as we were walking along the stream. Hand capture while snorkeling involves extensive searching of the underwater habitat (see Meyer et al. 2013). For each turtle, we recorded maximum carapace length (mm), body mass (g), sex, and age class using secondary sex characteristics (Ashton et al. 2012). Large juveniles and adults were then selected for the thermal behavior study.

Temperature dataloggers (iButtons; Thermochron iButton, Maxim Integrated, San Jose, CA) were affixed to the posterior portion of the turtle carapace, scute C3 or C4, using nonexothermic waterproof epoxy and superglue. The total mass of the iButton and epoxy was kept at 7 g or less and was not placed on turtles when the mass was > 10% of the turtle's body mass. Upon retrieval, there was no noticeable damage to turtle skin or scutes. Recaptured turtles that had lost their iButton had very little residual epoxy. Additional iButtons were paired in 3 regions, distributed within each search area at both study sites. For each pair, one measured air temperature 1 m above ground and one was attached to a rock in the water, at a maximum of 1 m deep, in an area that best represented where turtles were being captured. We set all dataloggers to record temperature (°C) every 15 min, resulting in a maximum of 21 continuous days of measurements.

Each location was revisited several times to continue placing iButtons on turtles and to retrieve deployed dataloggers. At Sycamore Creek, 5 iButtons were retrieved from turtles, 11 fell off, and 5 turtles were not captured again during the study period. At the North Fork Kaweah River, 3 iButtons were retrieved, 2 fell off, and 13 turtles were not captured again. The several iButtons that were placed on turtles likely fell off owing to abrasive contact with rocky substrates.

Data Processing

After retrieval of iButtons, the raw data were organized so that each turtle's temperature data corresponded with the geographically closest set of air and water dataloggers. The raw data from the turtle, air, and water iButtons were plotted together to determine when the turtle was in the air, water, or in a burrow. The turtle was deemed to be in the air or basking when the turtle temperature was equal to or exceeded that of the measured air temperature. When the turtle temperature was equal to or less than water temperature it was determined to be in the water. For many of the records, the turtle's temperature corresponded very well with the air or water, but there were cases where the turtle temperature did not clearly align with either air or water temperatures. When the turtle temperature remained stable for a prolonged period of time, but in-between the air and water temperatures, it was classified as “burrow” behavior (Supplemental Fig. S1; all supplemental material is available at http://dx.doi.org/10.2744/CCB-1240.1.s1).

Statistical Analysis

Temperatures collected from turtle shells were grouped into hourly intervals and the dominant behavior during that hourly period was converted to a binary response variable—in water (1) or out of water (0; land, burrow). Generalized linear mixed-effects models via penalized quasi-likelihood were used to determine how environmental variables influenced whether turtles were in or out of their aquatic habitat (glmmPQL in package MASS; Ripley et al. 2013). Air and water temperatures (°C) were obtained using iButtons (previously described). Relative humidity (%) and solar radiation (W/m²) were obtained from the National Park Service Ash Mountain Meteorological Station (NPS 2016). Barometric pressure (in Hg) was obtained from the nearest station at the Visalia Airport (National Oceanic and Atmospheric Administration [NOAA] 2016). Finally, day length (hr) was calculated using the Maptools package by using the location of each study site (Bivand and Lewin-Koh 2015). All continuous explanatory variables were joined to turtle behavior by date and time.

A glmmPQL was run separately for the Sycamore Creek and North Fork Kaweah River sites because of stark differences in sampling season and habitat. Collinearity of explanatory variables was initially investigated by examining correlations among variables. Collinearity was further assessed if parameter estimates were unstable when collinear variables were used together. At both sites, temporal autocorrelation of hourly data was accounted for using the corCAR1 function on the Date/Time variable. For Sycamore Creek, we were able to include the following variables without substantial problems with collinearity: air temperature, water temperature, relative humidity, solar radiation, barometric pressure, and day length were fixed effects and the individual turtle was the random effect. At North Fork Kaweah River, air temperature, water temperature, and relative humidity were highly correlated, making it unsound to include all 3 variables in the model. Water temperature was excluded from the model because the North Fork Kaweah River has deep, stratified pools with water temperatures not well represented by the iButtons that were placed only in riffles. Of the 2 remaining variables, air temperature is a well-established driver of turtle behavior and was therefore selected over relative humidity for model inclusion. Air temperature, solar radiation, barometric pressure, and day length were fixed effects and the individual turtle was a random effect. Finally, each model was tested for discrimination by assessing the area under the curve via receiver operating characteristic (ROC). The ROC was obtained using the RMS package for R (Harrell 2016) and indicates how well the model correctly classified a turtle as being in or out of the water. We also tested model calibration by comparing observed occurrences and modeled probabilities. All statistical analyses were conducted in R statistical software (R Core Team 2015).

RESULTS

Seven of 8 turtles spent more time in water than basking or in a burrow while the largest turtle, a female, spent more time basking. For the 5 turtles in Sycamore Creek, 55% of the monitoring time was spent in water, 37% was spent basking, and 8% was spent in a burrow. For the 3 turtles in the North Fork Kaweah River, 82% of the monitoring time was spent in water, and each turtle spent at least 70% of its time in the water, including a juvenile turtle that spent 490 of 509 monitoring hours (96%) in water (Table 1).

Table 1. Sex, size, total time recorded, and percent of time spent in a given habitat for western pond turtles (Emys marmorata) at Sycamore Creek and the North Fork Kaweah River, California.

View Fullscreen

Sycamore Creek

At Sycamore Creek, the model performed well at identifying when turtles were in the water (ROC = 0.76; Supplemental Fig. S2). The probability of being in the water decreased at higher air (glmmPQL: t₁₈₇₁ = −6.36, p < 0.001; Fig. 2A) and water (t₁₈₇₁ = −6.35, p < 0.001) temperatures. Significant diel and seasonal patterns of movement were measured in Sycamore Creek turtles; lower solar radiation (t₁₈₇₁ = −6.46, p < 0.001) and longer days (t₁₈₇₁ = 6.45, p < 0.001) increased the probability that a turtle was in the water. No significant relationship was found with relative humidity (t₁₈₇₁ = −0.973, p = 0.33) or barometric pressure (t₁₈₇₁ = −1.49, p = 0.14). However, Sycamore Creek turtle temperatures indicated that turtles were leaving the water for burrows approximately 1−2 d prior to substantially lower air temperatures during spring storms (Supplemental Fig. S1). Finally, we examined the random effect (i.e., individual turtle) by plotting the model output with an adjusted intercept for each turtle and found that individual turtle behavior was largely consistent at the site.

View Figure

• Download as Powerpoint
• Download Figure

North Fork Kaweah River

At the North Fork Kaweah River, the model also performed well at identifying when turtles were in the water (ROC = 0.81; Supplemental Fig. S3). The probability of being in the water decreased at higher air temperatures (glmmPQL: t₁₄₉₈ = −5.13, p < 0.001; Fig. 2B). Significant diel and seasonal patterns of movement were also measured in North Fork Kaweah River turtles; lower solar radiation (t₁₄₉₈ = −2.60, p < 0.01) and shorter days (t₁₄₉₈ = −7.51, p < 0.001) increased the probability a turtle was in the water. No significant relationship was found with barometric pressure (t₁₄₉₈ = −0.81, p = 0.42). Finally, we examined the random effect (i.e., individual turtle) by plotting the model output with an adjusted intercept for each turtle and found that individual turtle behavior was largely consistent at the site.

DISCUSSION

Utilizing temperature sensors to elucidate the behavior of turtles is a helpful technique to directly address imperfect detection and to improve site-specific understanding of turtle phenology and daily activity patterns. Prior studies that have utilized temperature sensors affixed to turtles investigated habitat selection (Fredericksen 2014), overwintering habitat requirements (Greaves and Litzgus 2007), the importance of basking for a turtle's energy budget (Bulté and Blouin-Demers 2010), and changes in thermal quality due to habitat alterations (Yagi and Litzgus 2013). When turtles are easier to capture, population data can be more-reliably obtained and survey methodologies are of less concern (Marchand and Litvaitis 2004; Enneson and Litzgus 2008; Reeves and Litzgus 2008). However, cryptic turtle species and those with greater phenotypic plasticity demand attention to site-specific life-history strategies. In the case of E. marmorata, with a wide distribution that results in great variation in habitat use and detectability, generalizing survey methodology across their distribution has the potential to increase false-negative reporting. Our case study highlights this issue by addressing how turtle behavior drives detectability within a single watershed at stream sites that differ in their water permanence.

At Sycamore Creek, the probability of a turtle being in the water decreased as air and water temperatures increased throughout the day. Basking when air temperatures are elevated makes thermal sense so that a turtle may enhance metabolism (Gatten 1974), potentially kill parasites (Cagle 1950), and promote energy retention (Bulté and Blouin-Demers 2010). However, a basking turtle cannot perform other activities such as mating and foraging, which may only be possible underwater (Bury 1986). Solar radiation also had a significant effect on turtle behavior and was related to environmental temperatures. Sycamore Creek turtles were most likely to be in the water at night. Engaging in underwater activities is thermally optimal at night, when water temperatures are warmer than air temperatures. Relative humidity and barometric pressure were not significant factors in determining the probability of a turtle being in the water. However, some turtles at Sycamore Creek burrowed just before and during storm events (Supplemental Fig. S1). Turtles were also more likely to be in the water as days grew longer. Sycamore Creek can be surveyed safely at any time of the year, but it typically dries up in mid- to late summer. This timing coincides with greatest day length and with increased probability of a turtle being in the water. Presumably, turtles are in the last remaining pools consuming food in preparation for aestivation. In summary, our data suggest that the best snorkel sampling strategy at Sycamore Creek is from dusk to dawn, at greater day-lengths, as water levels are declining.

At the North Fork Kaweah River, turtles were almost equally likely to be in the water at any time of the day, although certain environmental variables influenced this probability. Like Sycamore Creek, and for the same proposed reasons, turtles were less likely to be in the water as air temperatures increased. However, while these turtles were also more likely to be in the water at night when air temperatures were lower, the preference was not as pronounced as for Sycamore Creek turtles. Water temperature and relative humidity were highly correlated with air temperature at this site because surveys were conducted during the middle of the summer when humidity was stable and storm events were rare, unlike the spring survey conditions at Sycamore Creek. Thus, while we were unable to assess the effect of humidity on turtle behavior at this site, it is unlikely to be consequential given that surveys at the North Fork Kaweah River occurred when humidity is less variable. Barometric pressure was also not a significant factor in dictating turtle behavior at this site. Day length was the most influential predictor of turtle behavior at the North Fork Kaweah River, and turtles were more likely to be in the water when days were shorter. This result can be explained by the differences in seasonality and survey times between the 2 sites. At the North Fork Kaweah River, spring flows are high and more dangerous than at Sycamore Creek, so surveys do not begin until mid-summer to early fall. Turtles are more likely to be in the water during the early fall when days are shorter because they also need to build up energy reserves for impending hibernation–overwintering. The proposed reason for increased aquatic activity later in the seasons is the same as for Sycamore Creek turtles—to consume more food for torpor—but the differences in time of year and type of torpor reflect the effect that hydrological regimes can have on turtle behavior. Within a single watershed in northern California, E. marmorata exhibited phenotypic variability, with turtles of the intermittent stream reach leaving the drainage to aestivate in response to water availability and turtles of the perennial reach leaving the drainage to hibernate–overwinter in response to changing air and water temperatures (Bondi and Marks 2013). In summary, at the North Fork Kaweah River our data suggest that the best snorkel sampling regime is at any time of the day in late summer and early fall.

At Sequoia National Park, population estimates for these historic monitoring sites are highly variable, despite attempts to identify a sampling regime that would improve population estimates (Jeffcoach 2014). It has long been recognized that sampling methodology may influence population parameter estimates (Ream and Ream 1966), and this continues to be investigated in increasing detail (Willson et al. 2011; Tesche and Hodges 2015), although much work remains (e.g., species-specific temporal behavior patterns). For over 20 yrs, snorkel surveys at our study sites have been conducted between midmorning and late afternoon. At Sycamore Creek, our data suggest that this methodology may lead to underestimations of population size, given that turtles are more likely to be in the water at night and are therefore less likely to be captured during typical survey times. This misalignment between snorkel surveys and turtle activity may help explain our inability to generate reliable population estimates at Sycamore Creek (Jeffcoach 2014). At this site and at other, more-intermittent streams, like those of San Luis Obispo County, California (Rathbun et al. 2002), surveys may be more successful when conducted during crepuscular or nocturnal hours (or both) and when water availability begins to dwindle, just before aestivation. Why these turtles are in the water predominantly at night is unclear, but proposed reasons include evading nocturnal predators such as raccoons, foraging, mating, and resting (Ernst 1976; Hayes 1993).

In contrast, conducting snorkel surveys during midday is suitable for the North Fork Kaweah River, as turtles are likely to be in the water at any time. At perennial streams, like the North Fork Kaweah River and others (e.g., the Trinity River; Reese and Welsh 1997), imperfect detection may be reduced by concentrating surveys during late summer and early fall, before hibernation–overwintering. However, the large and complex underwater habitat of the river makes detection difficult and has been a likely contributor to poor population estimates. Simultaneously placing traps in larger streams and rivers to capture turtles may increase overall detection and improve population estimates (Ream and Ream 1966; Tesche and Hodges 2015).

Our results highlight the need to modify survey methodologies in response to a species' behavioral phenotypic plasticity according to both the time of day and time of season. In addition, there are reasons to expect ideal sampling methods to vary across a species range. While our results are especially pertinent to streams of the southern Sierra Nevada, they may be less applicable in other parts of the western pond turtle's range. For example, it is important to consider how anthropogenic habitat alterations may influence turtle behavior and survey methods. Ashton et al. (2011) found that E. marmorata in the dammed mainstem Trinity River are subject to much cooler water temperatures than those of the undammed South Fork Trinity River. This contrast in thermal regimes resulted in mainstem turtles spending significantly more time basking and caused premature hibernation–overwintering. Thus, turtles of the undammed South Fork Trinity River are more likely to behave similarly to Nork Fork Kaweah River turtles, but turtles of the dammed mainstem Trinity River behave according to a much different thermal regime. Further, in locations where turtles may have multiple clutches per year, the length of time that female turtles spend on land nesting may be much greater than at locations where turtles have only 1 clutch (Bury et al. 2012b). Aquatic capture probability would likely decrease during peak nesting times in these regions. Finally, in southern portions of the turtles' range, where air and water temperatures are high for much of the year, individuals may be able to reach suitable body temperatures strictly via aquatic basking in shallow waters (Bury et al. 2012b). Depending on the hydrology of a given stream, southern California turtles may be more available for aquatic capture if water is consistently present and temperatures are suitable or less available if streams desiccate earlier each year. Each of these natural and anthropogenic variations in habitat result in site-specific phenotypic variation in turtle behavior that must be accounted for when planning surveys.

California's Central Valley was once an important part of the western pond turtle's range, but much of the wetland habitat is now replaced by urbanization and agriculture (Germano and Bury 2001; Bury and Germano 2008). In the Valley's remaining pond and wetland habitat, including man-made habitat such as wastewater treatment ponds, capturing E. marmorata via trapping is reliable and preferred (Germano and Bury 2001; Bury and Germano 2008; Germano 2016). Today, however, many E. marmorata populations of central California are restricted to foothill stream habitat which is more difficult to survey and less understood (Bury et al. 2012a). In these cases, snorkel surveys are the preferred method, and behavioral data can improve survey methodologies and lead to more-informed estimates of population size. Subsequently, management decisions can be made with more reliability (Germano et al. 2012; Jeffcoach 2014).

Currently, reliable regional, statewide, and range-wide population estimates for E. marmorata are lacking (Germano et al. 2012; Jeffcoach 2014). Filling this gap may help guide management actions, given that this species may be threatened by droughts that are projected to become more extreme in the future (Burke et al. 2006; Griffin and Anchukaitis 2014; Leidy et al. 2016) and in light of the recent re-petition to federally list E. marmorata as an endangered species (US FWS 2015). Reliable population estimates may be relevant to the assessment of a species' current status, to actions seeking to improve status, and to determining whether an identified threat (e.g., drought) may severely impact the persistence of specific populations. Management actions for E. marmorata can be taken knowing that population estimates vary and with the understanding that site-specific research providing baseline data can inform such actions (Tesche and Hodges 2015).