Studies of habitat selection can reveal important patterns to guide habitat restoration and management for species of conservation concern. Giant gartersnakes Thamnophis gigas are endemic to the Central Valley of California, where >90% of their historical wetland habitat has been converted to agricultural and other uses. Information about the selection of habitats by individual giant gartersnakes would guide habitat restoration by indicating which habitat features and vegetation types are likely to be selected by these rare snakes. We examined activity patterns and selection of microhabitats and vegetation types by adult female giant gartersnakes with radiotelemetry at a site composed of rice agriculture and restored wetlands using a paired case-control study design. Adult female giant gartersnakes were 14.7 (95% credible interval [CRI] = 9.4–23.7) times more likely to be active (foraging, mating, or moving) when located in aquatic habitats than when located in terrestrial habitats. Microhabitats associated with cover—particularly emergent vegetation, terrestrial vegetation, and litter—were positively selected by giant gartersnakes. Individual giant gartersnakes varied greatly in their selection of rice and rock habitats, but varied little in their selection of open water. Tules Schoenoplectus acutus were the most strongly selected vegetation type, and duckweed Lemna spp., water-primrose Ludwigia spp., forbs, and grasses also were positively selected at the levels of availability observed at our study site. Management practices that promote the interface of water with emergent aquatic and herbaceous terrestrial vegetation will likely benefit giant gartersnakes. Given their strong selection of tules, restoration of native tule marshes will likely provide the greatest benefit to these threatened aquatic snakes.

The study of habitat selection, or the use of habitat disproportionate to its availability (Manly et al. 2002), plays an important role in our understanding of animal ecology and management. Habitat selection is often thought of as a hierarchical process (Johnson 1980), by which individuals select both their home ranges and the locations in which to allocate their time within the home range. Heuristically, we often attribute a series of choices to individual animals (Cooper and Millspaugh 1999), with the resulting patterns indicating selection for or against habitats or habitat attributes. Knowledge about habitat selection is an essential component of managing habitat for animal populations (Cooper and Millspaugh 1999). Studies of habitat selection can unveil the resource requirements of a species, quantify or predict the effects of habitat changes on populations, and, if linked to individual fitness, imply consequences of habitat change (Cooper and Millspaugh 1999; Manly et al. 2002). Furthermore, habitat selection studies can lead to a more mechanistic understanding of animal–habitat relationships (Cooper and Millspaugh 1999), which might be critical to the success of habitat restoration and reintroduction efforts (Swaisgood et al. 2007).

Giant gartersnakes Thamnophis gigas (Figure 1) are endemic to the Central Valley of California, where they have lost >93% of their wetland habitat within the past 150 y (Frayer et al. 1989; Garone 2007; Huber et al. 2010). Giant gartersnakes have been nearly extirpated from the southern two-thirds of their range (U.S. Fish and Wildlife Service 2015), but persist in the rice agricultural landscape of the Sacramento Valley (northern Central Valley; Halstead et al. 2010). Habitat restoration is an important component of the draft Giant Gartersnake Recovery Plan (U.S. Fish and Wildlife Service 2015), but most of what is known about giant gartersnake habitat is based on mark–recapture studies at a few sites and detection–nondetection surveys at a large spatial scale (Wylie et al. 2010; Halstead et al. 2014). Information on the selection of habitat components by individuals is needed to guide habitat restoration and management. The objective of our study was to examine the selection of microhabitats and vegetation composition by adult female giant gartersnakes in a population occurring among restored marshes and rice agriculture.

Figure 1.

Adult female giant gartersnake Thamnophis gigas. U.S. Geological Survey photo by Matt Meshriy.

Figure 1.

Adult female giant gartersnake Thamnophis gigas. U.S. Geological Survey photo by Matt Meshriy.

Close modal

Study Site

We conducted our study in and around Gilsizer Slough, Sutter County, California. The site consisted of a remnant slough used as a water supply and drain for surrounding agriculture. Agriculture around the slough was dominated by rice, although tomatoes, melons, other row crops, and alfalfa have occasionally been grown in the area. From 2003 to 2009, approximately 229 ha of marsh habitat were constructed from these agricultural fields as mitigation for take of giant gartersnakes or their habitat. Our study primarily occurred in the created marshes and the drains and canals associated with surrounding rice agriculture.

Field methods

We captured giant gartersnakes using modified floating funnel traps (Casazza et al. 2000) or by hand in 2009 and 2010. We retained individuals >180 g for intracoelomic implantation of a radio transmitter (9–11 g Model SI-2; Holohil Systems Ltd., Carp, Ontario, Canada). Radiotransmitters were <5% of each individual's preoperative mass. Because giant gartersnakes are sexually dimorphic for size (Wylie et al. 2010), most individuals large enough for a transmitter were females. We tracked 19 females, and obtained 12–233 (mean = 94) observations per female.

We attempted to locate each individual ≥5d/wk from April through October of each study year. This period encompasses breeding (April through May), gestation (May through August), and parturition (July through September). Foraging and ecdysis (shedding of the skin) occur intermittently throughout the active season. We did not assess habitat selection during November through March, when individuals are brumating (in a dormant, hibernation-like state) in mammal burrows, riprap, or debris that provides a thermally stable environment. At each location, we characterized behavior of each snake (resting, moving [slithering or swimming], foraging, mating, or unknown) and indicated whether it was in terrestrial or aquatic habitat. We visually estimated the percent cover of microhabitats in eight categories (bare ground, open water, rock, litter [dead vegetation], terrestrial vegetation, emergent vegetation, submerged vegetation, and rice) within a circular quadrat of 1-m radius centered on the individual's location. We also collected data on the percent cover of >12 major vegetation types (species or higher taxonomic category, including tules [bulrushes; Schoenoplectus acutus], cattails [Typha spp.], water-primrose [Ludwigia spp.], sedges [Cyperaceae], rushes [Juncaceae], duckweed [Lemna spp.], mosquito fern [Azolla spp.], algae, grasses [Poaceae], forbs, shrubs and vines, cultivated rice, and others) within the same quadrat. We defined percent cover of vegetation types as the proportion of total vegetation composed of each vegetation type, rather than the proportion of the total area. For each individual snake location, we selected an available location at a random Uniform(minimum = 1, maximum = 360) azimuth and Uniform(2, 50) m distance from the snake location. The 50-m maximum distance represents the 0.75 quantile of daily movement distances at this site based on previous telemetry work (G.D. Wylie, U.S. Geological Survey, unpublished data). We collected the same data for each random location as for each observed location. In 2009, we limited each random location to the aquatic or terrestrial environment in which the individual was located; in 2010, we relaxed this restriction and considered all locations (both terrestrial and aquatic) to be available to the individual.

Analytical methods

We used five different analyses to examine the habitat relationships of giant gartersnakes. To describe the mean used and available percent cover of each microhabitat and vegetation type, we used a hierarchical linear regression model that quantified the variance within and among individuals. We used a hierarchical Bernoulli model with random intercepts to describe the probability that giant gartersnakes would be in terrestrial, rather than aquatic, habitats. To examine whether the probability of activity varied between terrestrial and aquatic environments, we used a hierarchical logistic regression model with random individual slopes and intercepts. We also used two separate hierarchical case-control logistic regression models to evaluate the selection of microhabitats and vegetation types by giant gartersnakes. Each of these models is described in detail below.

We used a Bayesian analysis of hierarchical linear models to describe the mean percent cover of habitats and vegetation types used by snakes and available to snakes. These models included a random intercept for each individual snake to account for correlation within individuals, and decomposed the variance into within- and among-individual components for each predictor variable. This model had the form

where yi was the percent cover of a microhabitat or vegetation type at observation i, α was the population mean percent cover of a microhabitat or vegetation type, ηj was the deviation of the mean percent cover for snake j from the population mean, and σ was the residual variation among observations. Values of ηj were distributed as Normal(mean = 0, SD = σind), and accounted for correlation of observations within individual snakes. Priors for this model were α ∼ Uniform(0, 100), σ ∼ Uniform(0, 50), and σind ∼ Uniform(0, 20).

We described the proportion of giant gartersnake locations in terrestrial vs. aquatic habitats using a hierarchical Bernoulli model. We estimated the probability of an average individual female giant gartersnake being in the terrestrial environment using a hierarchical Bernoulli model as follows:

where yi was the habitat type (1 = terrestrial, 0 = aquatic; Data S1, Supplemental Material) at observation i, pi was the probability that observation i would be of a snake in the terrestrial environment, α was the logit-scale intercept for the population average probability, and ηj ∼ Normal(0, σind) was the logit-scale random intercept for snake j. We used an uninformative (on the probability scale; Lunn et al. 2013) Normal(0, 1.648) prior for α, and gave σind an uninformative Uniform(0, 5) prior.

We expanded the above Bernoulli model to estimate differences in the probability of activity in terrestrial and aquatic environments. This model took the form

where yi (1 = active, 0 = inactive; Data S1, Supplemental Material) was the observed activity status of the snake at observation i, pi was the probability that observation i was of a snake engaging in activity (moving, foraging, mating, etc.), α was the logit-scale intercept for the population average probability of activity, βj ∼ Normal(μβ, σβ) was the individual-specific effect of location xi (1 = terrestrial, 0 = aquatic) on the probability of being active, and ηj ∼ Normal(0, σind) was the logit-scale random intercept for snake j. Priors for this model were selected to be uninformative, with priors for α and μβ specified as Normal(0, 1.648), and priors for σind and σβ specified as Uniform(0, 5). From this model, we calculated the probability of activity by an average adult female giant gartersnake in terrestrial and aquatic environments, and calculated the odds ratio for activity in the aquatic environment relative to the terrestrial environment as . We used a posterior predictive check for goodness-of-fit of an aggregated (binomial) version of this model based on counts using a χ2 discrepancy measure, and derived ĉ and a Bayesian P-value from the χ2 discrepancy measures (Kéry and Royle 2016).

To assess microhabitat and vegetation selection of adult female giant gartersnakes, we used a Bayesian analysis of hierarchical case-control logistic regression models, with giant gartersnake locations (“used”) as the cases and random points (“available”) as controls. Our model was hierarchical in that it included an individual-level random effect for model coefficients. The inclusion of these random effects properly places the individual as the sample unit, from which observations are considered subsamples (Gillies et al. 2006). Perhaps equally important, allowing model coefficients to vary among individuals allows selection to vary among individuals, and quantifies the amount of among-individual variation in the selection of each habitat attribute (Gillies et al. 2006). Relative to treating the individual as a fixed effect, treating the individual as a random effect weights individuals by the number of locations, resulting in individuals that contribute fewer data to be pulled toward the population mean (“shrinkage;” Gelman and Hill 2006). The output of this model is a resource selection function, which is the relative probability that a given resource unit will be used, given its covariate profile (Keating and Cherry 2004), for both the population and individual giant gartersnakes. We also used a random effect for groups of observations for instances when an individual did not change locations between successive observations. This allowed for correlation in habitat attributes in a paired case-control study design while using all available data to characterize the surrounding environment available to the individual. The structure of the model was therefore:

where yi is a vector of ones for each observation pair i, βl,k is a matrix of variable l and location k specific coefficients, and xi,l is a matrix of the differences between used and available habitats or vegetation types for each observation pair i and each variable l (Data S2, Supplemental Material). The coefficient for location k within snake j was distributed as βl,k ∼ Normal(μl,j, σl,j), with the individual mean coefficients in turn distributed as μl,j ∼ Normal(μl, σl). This parameterization of the hierarchical case-control logistic regression model, which uses differences between measurements of used and available points for each variable as predictor variables, does not contain an intercept.

We fit two separate case-control models to the data: one to estimate the selection of microhabitats, and the other to estimate the selection of specific vegetation types. We chose to model microhabitats and vegetation separately to limit the number of predictor variables in a given model and because the vegetation type data provided a detailed description of the vegetated microhabitats, which were a subset of all microhabitats. We chose to model microhabitats and vegetation types that were most common, exhibited the greatest variation, and addressed important management questions. These variables were presumed to be most predictive of relative probability of use and to inform habitat management for conserving giant gartersnakes. For the microhabitat model, we included percent cover of bare ground, rock, open water, litter, emergent vegetation, submerged vegetation, terrestrial vegetation, and rice (Table 1). For the vegetation model, we included percent cover of tules, cattails, water-primrose, mosquito fern, algae, grasses, forbs, duckweed, and rice (Table 1). We scaled data input into the model to units of 10% difference so that parameter estimates were on a scale that could be related to observable differences in the field. We selected priors to be uninformative, with Normal(0, 1.648) priors on model coefficients and Uniform(0, 5) priors on standard deviations. For the selection models, we assessed goodness-of-fit using the standardized Pearson χ2 statistic (Hosmer et al. 2013) using the maximum likelihood fit of each model with random individual slopes for each predictor variable.

Table 1.

Descriptive statistics for percent cover of microhabitats and vegetation types at observed female giant gartersnake Thamnophis gigas locations and associated random points at Gilsizer Slough, California, USA, 2009–2010. Means and standard deviations between individuals (SDInd) and between repeated observations within individuals (SDResid) are based on an intercept-only model with a normally distributed individual random effect. Values for each parameter are presented as posterior mean (SD). Variables are listed in descending order of available percent cover.

Descriptive statistics for percent cover of microhabitats and vegetation types at observed female giant gartersnake Thamnophis gigas locations and associated random points at Gilsizer Slough, California, USA, 2009–2010. Means and standard deviations between individuals (SDInd) and between repeated observations within individuals (SDResid) are based on an intercept-only model with a normally distributed individual random effect. Values for each parameter are presented as posterior mean (SD). Variables are listed in descending order of available percent cover.
Descriptive statistics for percent cover of microhabitats and vegetation types at observed female giant gartersnake Thamnophis gigas locations and associated random points at Gilsizer Slough, California, USA, 2009–2010. Means and standard deviations between individuals (SDInd) and between repeated observations within individuals (SDResid) are based on an intercept-only model with a normally distributed individual random effect. Values for each parameter are presented as posterior mean (SD). Variables are listed in descending order of available percent cover.

Bayesian analysis of models used Markov chain Monte Carlo sampling carried out in JAGS (Plummer 2014). We selected model burn-in, sampling iterations, and thinning rates to ensure convergence was achieved and effective sample sizes were >5,000 for all monitored parameters. For the linear models describing used and available microhabitats and vegetation types, we based posterior inference on five chains of 200,000 iterations each, after a burn-in period of 10,000 iterations. We used five chains of 100,000 iterations each, following a burn-in period of 10,000 iterations, for posterior inference for the Bernoulli models of the probability of observations in terrestrial vs. aquatic environments and the probability of activity in these environments. For the microhabitat and vegetation selection models, we based posterior inference on five chains of 1,000,000 iterations each, after a burn-in period of 100,000 iterations. In all cases, we thinned the Markov chain Monte Carlo output so posterior inference was based on 100,000 samples from the stationary posterior distribution (used and available description thinning rate = 10, Bernoulli model thinning rate = 5, and selection thinning rate = 100). We fit each model with JAGS 3.4.0 (Plummer 2014) called from R 3.2.1 (R Core Team 2015) using the package runjags (Denwood 2016). We diagnosed convergence with visual examination of history plots and with the Gelman–Rubin statistic (Gelman and Rubin 1992); we observed no evidence for lack of convergence (R-hat < 1.01 for all monitored parameters).

We obtained 1,789 (1,093 in 2009; 696 in 2010) active season locations of 19 female giant gartersnakes. On average, 76.3% (95% CRI = 68.8–83.5%) of female giant gartersnake observations occurred in terrestrial habitats. No evidence for lack of fit of the hierarchical binomial model examining probability of being active across terrestrial or aquatic habitats was apparent (ĉ = 0.78, 95% CRI = 0.29–1.53; Bayesian P-value = 0.74). When in terrestrial environments, an average individual was actively moving, mating, or foraging on the surface during 2.4% (CRI = 1.0–4.6%) of its observations. In contrast, an average individual was actively swimming or slithering on vegetation above or floating on the water during 28.0% (CRI = 15.2–42.4%) of its aquatic observations. The odds of an average individual being active was therefore 14.7 (CRI = 9.4–23.7) times greater in aquatic habitats than in terrestrial habitats.

Microhabitats and vegetation types associated with cover were positively selected by adult female giant gartersnakes during the active season (Table 2; Figure 2). The null hypothesis that the models adequately described the case-control habitat use and availability data was not rejected (microhabitat model, χ2 = −0.035, P = 0.486; vegetation type model, χ2 = −0.246, P = 0.402). Litter, emergent vegetation, terrestrial vegetation, and submerged vegetation microhabitats were positively selected, and rock and rice were avoided (Table 2; Figure 2). Among microhabitats, individuals varied least in their selection of water and most in their selection of rice and rock (Table 2). Among vegetation types, aquatic vegetation types were typically selected more strongly than terrestrial vegetation types. Tules, duckweed, water-primrose, forbs, and grasses were positively selected, and rice was avoided (Table 2; Figure 3). Individuals varied least in their selection of grasses, and most in their selection of algae (Table 2).

Table 2.

Parameter estimates describing population-level selection by, and individual variation in, selection of microhabitats and vegetation types of adult female giant gartersnakes Thamnophis gigas at Gilsizer Slough, California, USA, 2009–2010. Population mean selection effects indicate the change in log-odds of use by an average adult female with a 10% additive increase in percent cover (e.g., 5% to 15%), and correspond to estimates of the μl parameters from case-control regressions (values are on the logit scale). Similarly, estimated standard deviations of variation in selection among individuals correspond to the σl parameters from case-control regressions and are on the logit scale. Positive log-odds indicate selection (variables in bold); negative log-odds indicate avoidance (variables in italics). Values are posterior medians (95% credible intervals). Selection for variables with the same superscript was not statistically different (95% credible interval of the difference overlapped zero). Variables are listed from most to least selected within models.

Parameter estimates describing population-level selection by, and individual variation in, selection of microhabitats and vegetation types of adult female giant gartersnakes Thamnophis gigas at Gilsizer Slough, California, USA, 2009–2010. Population mean selection effects indicate the change in log-odds of use by an average adult female with a 10% additive increase in percent cover (e.g., 5% to 15%), and correspond to estimates of the μl parameters from case-control regressions (values are on the logit scale). Similarly, estimated standard deviations of variation in selection among individuals correspond to the σl parameters from case-control regressions and are on the logit scale. Positive log-odds indicate selection (variables in bold); negative log-odds indicate avoidance (variables in italics). Values are posterior medians (95% credible intervals). Selection for variables with the same superscript was not statistically different (95% credible interval of the difference overlapped zero). Variables are listed from most to least selected within models.
Parameter estimates describing population-level selection by, and individual variation in, selection of microhabitats and vegetation types of adult female giant gartersnakes Thamnophis gigas at Gilsizer Slough, California, USA, 2009–2010. Population mean selection effects indicate the change in log-odds of use by an average adult female with a 10% additive increase in percent cover (e.g., 5% to 15%), and correspond to estimates of the μl parameters from case-control regressions (values are on the logit scale). Similarly, estimated standard deviations of variation in selection among individuals correspond to the σl parameters from case-control regressions and are on the logit scale. Positive log-odds indicate selection (variables in bold); negative log-odds indicate avoidance (variables in italics). Values are posterior medians (95% credible intervals). Selection for variables with the same superscript was not statistically different (95% credible interval of the difference overlapped zero). Variables are listed from most to least selected within models.
Figure 2.

Odds ratios that adult female giant gartersnakes Thamnophis gigas select one quadrat relative to another with a 10% increase in percent cover of the given microhabitat at Gilsizer Slough, California, USA, 2009–2010. Black circles represent the posterior median of the population-level odds ratio; error bars represent the 95% credible interval. Gray dots represent posterior mean odds ratios for individual female giant gartersnakes. The horizontal dashed gray line represents an odds ratio of one, which indicates no selection for or against a microhabitat. Error bars for individuals are omitted for clarity. Note that error bars indicate uncertainty in the estimation of the population-level odds ratio; they are not directly related to the dispersion of individual means.

Figure 2.

Odds ratios that adult female giant gartersnakes Thamnophis gigas select one quadrat relative to another with a 10% increase in percent cover of the given microhabitat at Gilsizer Slough, California, USA, 2009–2010. Black circles represent the posterior median of the population-level odds ratio; error bars represent the 95% credible interval. Gray dots represent posterior mean odds ratios for individual female giant gartersnakes. The horizontal dashed gray line represents an odds ratio of one, which indicates no selection for or against a microhabitat. Error bars for individuals are omitted for clarity. Note that error bars indicate uncertainty in the estimation of the population-level odds ratio; they are not directly related to the dispersion of individual means.

Close modal
Figure 3.

Odds ratios that adult female giant gartersnakes Thamnophis gigas select one quadrat relative to another with a 10% increase in percent cover of the given vegetation type at Gilsizer Slough, California, USA, 2009–2010. Black circles represent the posterior median of the population mean; error bars represent the 95% credible interval of the population mean. Gray dots represent posterior mean odds ratios for individual female giant gartersnakes. The horizontal dashed gray line represents an odds ratio of one, which indicates no selection for or against a vegetation type. Error bars for individuals are omitted for clarity. Note that error bars indicate uncertainty in the estimation of the population-level odds ratio; they are not directly related to the dispersion of individual means. Vegetation types are tules Schoenoplectus acutus, cattails Typha spp., water-primrose Ludwigia spp., duckweed Lemna spp., mosquito fern Azolla spp., grass (Poaceae spp., other than rice), and rice Oryza sativa.

Figure 3.

Odds ratios that adult female giant gartersnakes Thamnophis gigas select one quadrat relative to another with a 10% increase in percent cover of the given vegetation type at Gilsizer Slough, California, USA, 2009–2010. Black circles represent the posterior median of the population mean; error bars represent the 95% credible interval of the population mean. Gray dots represent posterior mean odds ratios for individual female giant gartersnakes. The horizontal dashed gray line represents an odds ratio of one, which indicates no selection for or against a vegetation type. Error bars for individuals are omitted for clarity. Note that error bars indicate uncertainty in the estimation of the population-level odds ratio; they are not directly related to the dispersion of individual means. Vegetation types are tules Schoenoplectus acutus, cattails Typha spp., water-primrose Ludwigia spp., duckweed Lemna spp., mosquito fern Azolla spp., grass (Poaceae spp., other than rice), and rice Oryza sativa.

Close modal

During the active season, female giant gartersnakes selected a variety of habitat types that provided some form of cover, such as emergent, terrestrial, and submerged vegetation and litter. Concealment from predators and shelter from environmental extremes are likely important components of giant gartersnake habitat in California's Central Valley. The wariness of giant gartersnakes (Van Denburgh and Slevin 1918; Fitch 1940; Wright and Wright 1957) has been attributed to the general lack of shelter and abundance of native predators, such as raptors, otters, and wading birds, in the Central Valley (Fitch 1940; Hansen 1980). Similarly, the large size of giant gartersnakes has been attributed to the combination of the long, hot growing season in the Central Valley with giant gartersnakes' aquatic habit, which provides thermal stability (Fitch 1940; Hansen 1980). Although the aquatic environment is thermally stable, a variety of microclimates offered by habitats that provide cover likely enhance thermoregulatory opportunity and physiological performance (Huey et al. 1989; Huey 1991; Blouin-Demers and Weatherhead 2001; Row and Blouin-Demers 2006). Emergent vegetation, terrestrial vegetation, and litter, in particular, likely provide both concealment from predators and protection from environmental extremes, and appear to be important components of giant gartersnake habitat. The neutral selection for open water and lack of variation among individuals for selection of this microhabitat likely indicates consistent selection of the interface of open water with microhabitats that provide cover (Valcarcel 2011). This juxtaposition of cover and water might also concentrate fish, frogs, and anuran larvae that serve as prey for giant gartersnakes.

The interpretation of the selection of terrestrial vegetation by giant gartersnakes is hampered by their propensity to be found underground while in the terrestrial environment. On average, adult giant gartersnakes are underground at >60% of observations in terrestrial environments (Halstead et al. 2015). Whether giant gartersnakes are selecting surface characteristics or characteristics of their chosen refuge is unknown. Nonetheless, herbaceous terrestrial vegetation likely provides important cover for giant gartersnakes as they move between aquatic foraging habitat and subterranean refuges.

Whether in terrestrial or aquatic environments, giant gartersnakes have a variety of vegetation types from which to select among the vegetated habitats. As was indicated from microhabitats, vegetation that provides the greatest concealment and protection while providing thermoregulatory opportunity is likely an important component of giant gartersnake habitat. Tules were likely the most strongly selected vegetation type for a number of reasons. Tules were the dominant native emergent vegetation in the Central Valley prior to European settlement (Frayer et al. 1989; Garone 2007), and giant gartersnakes therefore share a long evolutionary history with tules. Tules also provide a unique habitat structure that differs from that of other vegetation within the range of giant gartersnakes. Tules have round, rigid stems that emerge singly from rhizomes, providing structure amenable to the locomotion of giant gartersnakes, but not their predators. Both live and dead stems are often blown over and provide horizontal basking surfaces for giant gartersnakes from which they can readily drop into the water below to escape predators. Tules therefore likely provide excellent opportunity for thermoregulation and escape from predators. Whether the structure of tules also promotes successful foraging warrants further attention (Mullin and Mushinsky 1995; Mullin and Gutzke 1999).

In contrast to tules, giant gartersnakes used cattails in proportion to their availability. Although cattails are native to California marshes, giant gartersnakes might prefer tules to cattails because of the different structural attributes of these emergent wetland plants. Although similar in size and habit to tules, cattail leaves are flat and emerge from rhizomes in a dense, fan-shaped cluster. Cattails do not appear to provide horizontal basking habitat of the same high quality as that of tules, nor do they provide the same quality of water-level structure for locomotion and escape from predators. Cattails often form dense monocultures, the interiors of which are less likely to be used by giant gartersnakes than are patch edges (Valcarcel 2011). Without thoughtful construction of marshes or management to provide cattail-free areas, we suspect that the giant gartersnake response to cattails could become negative in dense monocultures. Additional studies that include experimental treatments or potential nonlinear responses to cattails are warranted.

Some of the vegetation types in our study are considered invasive. Water-primrose consists of a complex of native species, nonnative species, and hybrids (Okada et al. 2009). Two species, Ludwigia hexapetala and L. peploides montevidensis, in particular, are considered highly invasive by the California Invasive Plant Council (http://www.cal-ipc.org/ip/management/plant_profiles; accessed September 2016) and are likely the dominant species at our study site. Because of the propensity for these species to form dense mats that clog waterways, they are the target of intensive control efforts (Okada et al. 2009). The effects of water-primrose on giant gartersnakes do not appear to be entirely negative, however. The positive selection of water-primrose at our study site was likely related to active control by herbicides and mechanical removal in canals; average water-primrose percent cover in random quadrats was <8% throughout our study (Table 1). Control of water-primrose generally results in the restriction of this plant to canal or wetland edges, where giant gartersnakes concentrate their activity (Valcarcel 2011). The response of giant gartersnakes to water-primrose would likely be negative in the absence of management to control the formation of dense monocultures.

In our study, duckweed was positively selected by giant gartersnakes. Duckweed floats and can form dense mats, but at our site these species generally occurred as a thin diffuse layer along the margins of canals and marshes or along the interface of open water and emergent vegetation, and averaged <1.5% cover in random quadrats. We rarely observed giant gartersnakes among dense mats of duckweed or other floating vegetation, and such mats likely alter the thermal profile, water chemistry, prey communities, and plant communities occurring below them. Thus, we would expect the response to duckweed and other floating vegetation to be negative when it becomes dense enough to form floating mats.

Rock, present at our study site as riprap armoring canal banks and water control structures and as a base for a road at the crest of an adjacent levee, was avoided by giant gartersnakes. Rocky substrates were likely limited in the Central Valley marshes in which giant gartersnakes evolved, and early accounts of the species indicate that it was confined to areas with a mud substrate (Fitch 1940). Nonetheless, riprap and decaying concrete along canal banks likely provide shelter to a few individuals (gray dots in Fig. 2) and serve as brumation sites during the inactive season (G.D. Wylie, unpublished data).

The generally negative response of adult female giant gartersnakes to rice was unexpected, but could be caused by several factors. Perhaps the most limiting aspect of rice as habitat for giant gartersnakes is the short time for which the annual rice crop provides emergent cover for giant gartersnakes. In the Sacramento Valley, cultivated rice generally emerges from shallowly flooded fields in late May or early June, but a canopy that provides cover does not mature until late June. Rice fields generally remain flooded until sometime in late August or early September, when water is drawn off the fields to allow them to dry enough for harvest. The giant gartersnake active season extends from April through September (Wylie et al. 2009); rice therefore only provides appropriate emergent aquatic habitat for approximately one-third of the active season. For the remainder of the active season, rice field microhabitats could be alternatively classified as bare ground, open water, or litter. The response of individual giant gartersnakes to rice was highly variable, with some individuals remaining in canals for the entire active season, others selecting rice and venturing into it, and still others remaining in restored wetlands where no rice was available to them. For those individuals located in rice or canals adjacent to rice, the high proportion of rice cover might result in underestimation of the selection for this habitat type because random locations within a rice field nearly always resulted in >75% rice cover. Several aspects of the relationship of giant gartersnakes to rice agriculture remain interesting and relevant to conservation in an agro-ecosystem, including 1) the degree to which individuals occurring in canals depend on productivity from neighboring rice fields; 2) the effect of rice fields, if any, on the carrying capacity of canals for giant gartersnakes; 3) the potential for rice fields to serve as habitat for neonate, juvenile, or male giant gartersnakes; 4) whether potential pools formed from the flooding and drying regime typical of rice fields in the Sacramento Valley results in ecological traps, where availability of prey is temporarily high but individuals are exposed to mortality from agricultural equipment; and 5) the demographic and behavioral responses of giant gartersnakes to rice farming practices, including fallowing, crop substitution, and pesticide use. Despite these unknowns, the cultivation of rice in the Sacramento Valley has allowed giant gartersnakes to persist (Halstead et al. 2010), whereas cultivation of other crops, such as cotton, corn, and grains, has very nearly extirpated the species from the San Joaquin Valley (U.S. Fish and Wildlife Service 2015).

Inference from our study applies to those habitats in the immediate vicinity of adult female giant gartersnakes. Our decision to limit availability to locations within the 75th percentile of daily movement distances has several implications. Much of the available environment was proximate and potentially usable by giant gartersnakes; therefore, many of the points considered available to giant gartersnakes likely were also used by giant gartersnakes. Because of this unknown degree of contamination of our available points, we could not estimate a resource-selection probability function (Keating and Cherry 2004; Rota et al. 2013). Our results therefore represent the relative likelihood of use of locations with a given covariate profile, rather than the absolute probability of use (Boyce and McDonald 1999; Manly et al. 2002; Keating and Cherry 2004; Rota et al. 2013). Nonetheless, our results provide useful information that may aid habitat management for giant gartersnakes.

The selection of habitats that provide cover adjacent to water indicates that maintaining a mosaic of cover and water is likely beneficial to giant gartersnakes during the active season. Promoting the growth of clumps of emergent vegetation and maintaining emergent vegetation along canal and wetland margins are likely the most effective means to achieve this goal. Tules are particularly valuable as giant gartersnake habitat, and managing for this species should provide the greatest benefit to giant gartersnakes. Water-primrose and cattails also can provide habitat for giant gartersnakes, provided that management of these species prevents the formation of monocultures. Although our study site is representative of restored marshes managed for giant gartersnakes, studies of giant gartersnake habitat selection at additional sites with different microhabitat and vegetation availability would be valuable to evaluate the evidence for some of the hypotheses suggested herein, including nonlinear responses to cattails, water-primrose, and duckweed. Although individual responses to rice were variable, examination of the behavior and demography of giant gartersnakes inhabiting rice fields and their associated canals warrants further study because the extent of rice habitat greatly exceeds that of managed marsh within the current range of the species. Experimental studies that examine giant gartersnakes' response to habitat manipulation would provide the most reliable inference about giant gartersnake habitat selection.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Data S1. Microsoft Excel file containing data used in the analysis of giant gartersnake Thamnophis gigas terrestrial vs. aquatic locations and surface activity at Gilsizer Slough, Sutter County, California, USA, 2009–2010. id = individual snake identification number (used as a random effect); unique_loc_id = unique location number (i.e., repeated numbers represent sequential observations when the individual did not move); loc_in_id = unique location number within each individual snake; terrestrial = 1 if the observation was in a terrestrial location, 0 if the individual was in an aquatic location, NA if the location could not be determined; active = 1 if the individual was observed in any mobile or active behavior, 0 if the individual was still for the entire observation, NA if activity could not be determined.

Found at DOI: 10.3996/042016-JFWM-029.s1; (50 KB XLSX).

Data S2. Microsoft Excel file containing data used in the analysis of giant gartersnake Thamnophis gigas microhabitat (microhab_select_hab_data_0910) and vegetation selection (microhab_select_veg_data_0910) at Gilsizer Slough, Sutter County, California, USA, 2009–2010. y = vector of response variables (all 1s for this parameterization of paired case-control logistic regression); id = individual snake identification number (used as a random effect); unique_loc_id = unique location number (i.e., repeated numbers represent sequential observations when the individual did not move); loc_in_id = unique location number within each individual snake. Remaining data are the differences in percent cover between observed and random locations of giant gartersnakes for each predictor variable (ground = bare ground; rock = rock (riprap or rocks >10 cm diameter); tv = terrestrial vegetation; litter = litter; water = open water; submerged = submerged aquatic vegetation; emergent = emergent aquatic vegetation; ricehab = cultivated rice [as a habitat]; tule = tule [Schoenoplectus acutus]; cattail = cattail [Typha spp.]; primrose = water-primrose [Ludwigia spp.]; dw = duckweed [Lemna spp.]; algae = algae; mf = mosquito fern [Azolla spp.]; grass = grasses [Poaceae]; dicot = forbs; riceveg = cultivated rice [as a species]).

Found at DOI: 10.3996/042016-JFWM-029.s2; (349 KB XLSX).

Reference S1. Frayer WE, Peters DD, Pywell HR. 1989. Wetlands of the California Central Valley: status and trends, 1939–mid-1980s. Portland, Oregon: U.S. Fish and Wildlife Service.

Found at DOI: 10.3996/042016-JFWM-029.s3; (9134 KB PDF).

Reference S2. Hansen RW. 1980. Western aquatic garter snakes in Central California: an ecological and evolutionary perspective. Master's thesis. Fresno, California: California State University, Fresno.

Found at DOI: 10.3996/042016-JFWM-029.s4; (7107 KB PDF).

Reference S3. U.S. Fish and Wildlife Service. 2015. Revised draft recovery plan for the giant garter snake (Thamnophis gigas). Sacramento, California: U.S. Fish and Wildlife Service, Pacific Southwest Region.

Found at DOI: 10.3996/042016-JFWM-029.s5; (3271 KB PDF); also available at https://www.fws.gov/sacramento/outreach/2015/12-22/docs/GGSrevisedDraftRecoveryPlan2015.pdf (3.19 MB PDF).

Reference S4. Valcarcel P. 2011. Giant gartersnake spatial ecology in agricultural and constructed wetlands. Master's thesis. Corvallis, Oregon: Oregon State University.

Found at DOI: 10.3996/042016-JFWM-029.s6; (939 KB PDF); also available at http://ir.library.oregonstate.edu/xmlui/handle/1957/21719 (939 KB PDF).

Funding for this study was provided by CALFED, the California Department of Water Resources, the California Waterfowl Association, and the U.S. Fish and Wildlife Service. We thank Wildlands, Inc. for access to the study site. J. Yee and T. Chambert provided statistical guidance; and we thank C. Overton, D. Olson, J. Rose, an anonymous reviewer, and the Associate Editor for reviews that greatly improved the manuscript. L. Achter, G. Dancourt, L. Heiker, V. Johnson, S. Marczak, C. Massing, L. McCardle, M. Meshriy, W. Meno, and J. Sweeney collected data for this project. Dr. R. Wack of the Sacramento Zoo and University of California – Davis Wildlife Health Center skillfully performed all transmitter implantation and removal surgeries. Snakes were handled in accordance with the University of California, Davis, Animal Care and Use Protocol 9699 and as stipulated in U.S. Fish and Wildlife Service Recovery Permit TE-020548-5.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Blouin-Demers
G,
Weatherhead
PJ.
2001
.
An experimental test of the link between foraging, habitat selection and thermoregulation in black rat snakes Elaphe obsoleta obsoleta
.
Journal of Animal Ecology
70
:
1006
1013
.
Boyce
MS,
McDonald
LL.
1999
.
Relating populations to habitats using resource selection functions
.
Trends in Ecology and Evolution
14
:
268
272
.
Casazza
ML,
Wylie
GD,
Gregory
CJ.
2000
.
A funnel trap modification for surface collection of aquatic amphibians and reptiles
.
Herpetological Review
31
:
91
92
.
Cooper
AB,
Millspaugh
JJ.
1999
.
The application of discrete choice models to wildlife resource selection studies
.
Ecology
80
:
566
575
.
Denwood
MJ.
2016
.
runjags: an R package providing interface utilities, model templates, parallel computing methods and additional distributions for MCMC models in JAGS
.
Journal of Statistical Software
71
:
1
25
.
Fitch
HS.
1940
.
A biogeographical study of the ordinoides artenkreis of garter snakes (genus Thamnophis)
.
University of California Publications in Zoology
44
:
1
150
.
Frayer
WE,
Peters
DD,
Pywell
HR.
1989
.
Wetlands of the California Central Valley: status and trends, 1939–mid-1980s
.
Portland, Oregon: U.S. Fish and Wildlife Service (see Supplemental Material, Reference S1)
. Found at DOI: (9134 KB PDF).
Garone
P.
2007
.
The fall and rise of the wetlands of California's Great Central Valley: a historical and ecological study of an endangered resource of the Pacific Flyway
.
Berkeley, California
:
University of California Press
.
Gelman
A,
Hill
J.
2006
.
Data analysis using regression and multilevel/hierarchical models
.
Cambridge, United Kingdom
:
Cambridge University Press
.
Gelman
A,
Rubin
DB.
1992
.
Inference from iterative simulation using multiple sequences
.
Statistical Science
7
:
457
472
.
Gillies
CS,
Hebblewhite
M,
Nielsen
SE,
Krawchuk
MA,
Aldridge
CL,
Frair
JL,
Saher
DJ,
Stevens
CE,
Jerde
CL.
2006
.
Application of random effects to the study of resource selection by animals
.
Journal of Animal Ecology
75
:
887
898
.
Halstead
BJ,
Skalos
SM,
Wylie
GD,
Casazza
ML.
2015
.
Terrestrial ecology of semi-aquatic giant gartersnakes (Thamnophis gigas)
.
Herpetological Conservation and Biology
10
:
633
644
.
Halstead
BJ,
Wylie
GD,
Casazza
ML.
2010
.
Habitat suitability and conservation of the giant gartersnake (Thamnophis gigas) in the Sacramento Valley of California
.
Copeia
2010
:
591
599
.
Halstead
BJ,
Wylie
GD,
Casazza
ML.
2014
.
Ghost of habitat past: historic habitat affects the contemporary distribution of giant garter snakes in a modified landscape
.
Animal Conservation
17
:
144
153
.
Hansen
RW.
1980
.
Western aquatic garter snakes in Central California: an ecological and evolutionary perspective
.
Master's thesis. Fresno, California: California State University, Fresno (see Supplemental Material, Reference S2)
. Found at DOI: (7107 KB PDF).
Hosmer
DW
Jr,
Lemeshow
S,
Sturdivant
RX.
2013
.
Applied logistic regression. Third edition
.
Hoboken, New Jersey
:
John Wiley & Sons
.
Huber
PR,
Greco
SE,
Thorne
JH.
2010
.
Boundaries make a difference: the effects of spatial and temporal parameters on conservation planning
.
The Professional Geographer
62
:
409
425
.
Huey
RB.
1991
.
Physiological consequences of habitat selection
.
The American Naturalist
137
:
91
115
.
Huey
RB,
Peterson
CR,
Arnold
SJ,
Porter
WP.
1989
.
Hot rocks and not-so-hot rocks: retreat-site selection by garter snakes and its thermal consequences
.
Ecology
70
:
931
944
.
Johnson
DH.
1980
.
The comparison of usage and availability measurements for evaluating resource preference
.
Ecology
61
:
65
71
.
Keating
KA,
Cherry
S.
2004
.
Use and interpretation of logistic regression in habitat selection studies
.
Journal of Wildlife Management
68
:
774
789
.
Kéry
M,
Royle
JA.
2016
.
Applied hierarchical modeling in ecology: analysis of distribution, abundance, and species richness in R and BUGS. Volume 1: prelude and static models
.
Amsterdam, The Netherlands
:
Academic Press
.
Lunn
D,
Jackson
C,
Best
N,
Thomas
A,
Spiegelhalter
D.
2013
.
The BUGS book: a practical introduction to Bayesian analysis
.
Boca Raton, Florida
:
CRC Press
.
Manly
BFJ,
McDonald
LL,
Thomas
DL,
McDonald
TL,
Erickson
WP.
2002
.
Resource selection by animals: statistical design and analysis for field studies
.
Dordrecht, the Netherlands
:
Kluwer Academic Publishers
.
Mullin
SJ,
Gutzke
WHN.
1999
.
The foraging ecology of the gray rat snake (Elaphe obsoleta spiloides). I. Influence of habitat structural complexity when searching for mammalian prey
.
Herpetologica
55
:
18
28
.
Mullin
SJ,
Mushinsky
HR.
1995
.
Foraging ecology of the mangrove salt marsh snake, Nerodia clarkii compressicauda: effects of vegetational density
.
Amphibia-Reptilia
16
:
167
175
.
Okada
M,
Grewell
BJ,
Jasieniuk
M.
2009
.
Clonal spread of invasive Ludwigia hexapetala and L. grandiflora in freshwater wetlands of California
.
Aquatic Botany
91
:
123
129
.
Plummer
M.
2014
.
JAGS 3.4.0 user manual
. .
R Core Team
.
2015
.
R: a language and environment for statistical computing
.
Vienna, Austria
:
R Foundation for Statistical Computing
.
Rota
CT,
Millspaugh
JJ,
Kesler
DC,
Lehman
CP,
Rumble
MA,
Jachowski
CMB.
2013
.
A re-evaluation of a case-control model with contaminated controls for resource selection studies
.
Journal of Animal Ecology
82
:
1165
1173
.
Row
JR,
Blouin-Demers
G.
2006
.
Thermal quality influences habitat selection at multiple spatial scales in milksnakes
.
Ecoscience
13
:
443
450
.
Swaisgood
R,
Stamps
JA,
Swaisgood
RR.
2007
.
Someplace like home: experience, habitat selection and conservation biology
.
Applied Animal Behaviour Science
102
:
392
409
.
U.S. Fish and Wildlife Service
.
2015
.
Revised draft recovery plan for the giant garter snake (Thamnophis gigas)
.
Sacramento, California: U.S. Fish and Wildlife Service, Pacific Southwest Region (see Supplemental Material, Reference S3)
. Found at DOI: (3271 KB PDF);.
Valcarcel
P.
2011
.
Giant gartersnake spatial ecology in agricultural and constructed wetlands
.
Master's thesis. Corvallis, Oregon: Oregon State University (see Supplemental Material, Reference S4)
. Found at DOI: (939 KB PDF);.
Van Denburgh
J,
Slevin
JR.
1918
.
The garter-snakes of western North America
.
Proceedings of the California Academy of Sciences
8
:
181
270
.
Wright
AH,
Wright
AA.
1957
.
Handbook of snakes of the United States and Canada, Volume II
.
Ithaca, New York
:
Comstock Publishing Associates
.
Wylie
GD,
Casazza
ML,
Gregory
CJ,
Halstead
BJ.
2010
.
Abundance and sexual size dimorphism of the giant gartersnake (Thamnophis gigas) in the Sacramento Valley of California
.
Journal of Herpetology
44
:
94
103
.
Wylie
GD,
Casazza
ML,
Halstead
BJ,
Gregory
CJ.
2009
.
Sex, season, and time of day interact to affect body temperatures of the giant gartersnake
.
Journal of Thermal Biology
34
:
183
189
.

Author notes

Citation: Halstead BJ, Valcarcel P, Wylie GD, Coates PS, Casazza ML, Rosenberg DK. 2016. Active season microhabitat and vegetation selection by giant gartersnakes associated with a restored marsh in California. Journal of Fish and Wildlife Management 7(2):397–407; e1944-687X. doi: 10.3996/042016-JFWM-029

The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Fish and Wildlife Service.

Supplemental Material