Habitat Correlates of the Southern Torrent Salamander ,

-A systematic stratified sampling design was used to quantify the habitat relationships of the southern torrent salamander, Rhyacotriton variegatus, in northwestern California. We sampled 53 first to third order streams, each surrounded by at least 5-7 ha of relatively homogeneous forest or harvested forest habitat. Measurements of 121 attributes of the forest and stream environment were recorded in conjunction with area-constrained aquatic sampling for salamanders. A subset of 68 variables, grouped into 11 ecological components including attributes at the landscape, macrohabitat, and microhabitat scales, was used in a hierarchical analysis of habitat associations. Results from discriminant and regression analyses indicated that this species occurs within a relatively narrow range of physical and microclimatic conditions and is associated with cold, clear headwater to low-order streams with loose, coarse substrates (low sedimentation), in humid forest habitats with large conifers, abundant moss, and HO% canopy closure. Thus, the southern torrent salamander demonstrates an ecological dependence on conditions of microclimate and habitat structure that are typically best created, stabilized, and maintained within late seral forests in northwestern California. The southern torrent salamander (Rhyacotriton variegatus) (previously R. olympicus variegatus) is the southernmost member of the family Rhyacotritonidae, comprised of a single genus with four species endemic to the Pacific Northwest (Good and Wake, 1992). Rhyacotriton variegatus occurs in aquatic habitats in coniferdominated forests at elevations below 1449 m (L. Diller, pers. comm.) in the coast ranges from Mendocino County, California to the Little Nestucca River and Grande Ronde Valley of northwestern Oregon (Nussbaum et al., 1983; Stebbins, 1985; Good and Wake, 1992; Leonard et al., 1993). Welsh and Lind (1992) examined the ‘This article was written and prepared by U.S. Government employees on official time, and it is therefore in the public domain and not subject to copyright. metapopulation structure of R. variegatus in northwestern California and estimated the species occurs in isolated sub-units at a frequency of 0.07 populations per km. Few quantitative data exist on the habitat affinities of R. variegatus (see Corn and Bury, 1989; Bury et al., 1991). Anecdotal and general accounts indicate that R. variegatus occur in springs, seeps, small streams, and margins of larger streams. They avoid open water and seek the cover of moss, rocks, and organic debris in shallow, cold, percolating water (Anderson, 1948; Nussbaum and Tait, 1977; Nussbaum et al., 1983; Stebbins, 1985; Bury, 1988; Bury and Corn, 1988; Corn and Bury, 1989; Welsh, 1990; Bury et al., 1991; Good and Wake, 1992; Leonard et al., 1993). Substrate conditions described for this species consist of water flowing through gravel, pebble, and cobble with little fine sediment. Habitat use differs slightly between the adult and larval life 386 H. H. WELSH AND A. J. LIND stages of R. variegatus. The larvae are entirely aquatic. Adults, although primarily aquatic, will occasionally use adjacent moist riparian and forest microhabitats in the wet season (pers. obs.). Recent research in northwestern California revealed that R. variegatus was found at significantly more sites in late seral forests (oldgrowth) than in early seral stages (Welsh and Lind, 1988; Welsh, 1990), but the number of sites sampled was small and these results require confirmation. Corn and Bury (1989) reported higher densities and biomass of this salamander in streams in uncut forests compared with logged forests in western Oregon. Rhyacotriton variegatus has a naturally patchy distribution in northwestern California, showing a strict association with headwaters and low order tributaries (Welsh and Lind, 1988, 1992). This salamander is a State “species of special concern” (Jennings and Hayes, 1994) due to the following factors: (1) distributional limits imposed by this habitat specificity; (2) an unusually high degree of genetic heterogeneity among sub-populations (Good and Wake, 1992); (3) the apparent association of this species with late seral attributes; and (4) the rapid loss of late seral forests due to timber harvesting (Thomas et al., 1988). A petition recently accepted by the U.S. Fish and Wildlife Service for listing R. variegatus as threatened under the Endangered Species Act, cites these same factors (Federal Register, 1995). A better understanding of the habitat relationships of this salamander might explain their absence or lower abundance in early seral forests and provide a basis for management alternatives that could reverse the reduction in its numbers due to forestry practices. The objectives of our study were: (1) to examine and quantify the habitat associations of R. variegatus at multiple spatial scales throughout the mixed conifer-hardwood forests of northwestern California; (2) to clarify the nature of the apparent relationship with forest succession; and by meeting the first two objectives, to (3) provide information critical for evaluating the potential impacts of continued habitat alterations on this species. MATERIALS AND M E T H O D S The general sampling design and the strategy of analysis used here have previously been described (Welsh and Lind, 1995). However, this study differs considerably from our earlier work in the details of both sampling and analysis. Here we provide a general outline of methods, with particular emphasis on those details that pertain to our study of R. variegatus. While this research is exploratory and correlative, and not designed to demonstrate cause and effect, such an approach is vital for developing testable hypotheses, and can, by itself, produce strong and useful inferences about real habitat relationships (see controlled experience studies; Waters and Erman, 1990). Site Selection. -Sites were distributed systematically across the range of R. variegatus within northwestern California using a stratified design, with a random component, at four nested levels: (1) biogeographic, (2) geographic (township and section), (3) seral stage, and (4) minimum essential microhabitat, Level one was defined by the known range of the species in California based on published accounts (Stebbins, 1985) and included portions of two bioregions: the North Coast and Klamath (Welsh, 1994). Levels two and three consisted of systematic selection of alternating townships, randomly choosing sections therein, and selection of forest stands in from one to four seral stages within each section. These criteria were instituted across most of the range described at level one, and up to 1115 m based on elevational limits known to us at the time (1989) (Stebbins, 1985; unpubl. data). All sampling occurred on public lands (state and federal parks and Forest Service) within the drainages of the Smith, Klamath, Trinity, Mad and Van Duzen Rivers of the Siskiyou, Klamath, and Coast Range Mountains. Sites occurred from near the Oregon border as far south as southern Humboldt and Trinity counties, California (latitude 40°22’ ), and east from the Pacific Ocean to western Siskiyou and western Trinity counties (longitude 123’25’). We sampled in mixed conifer-hardwood forests dominated by Douglas-fir (Pse i dotsuga menziesii) and redwood (Sequoia sempervirens). We used forest age (the mean of three corings from the dominant size class of conifers) to represent seral stage. Up to four stands were selected in each section when available (one each clearcut [0-30 yr], young [31-99], mature [l00-200], and old-growth [ +200]). Stand ages sampled ranged from one year old clearcuts to a 941 yr old redwood stand. Sampling sites were located in at least 5-7 ha of contiguous forest or clearcut (no edge habitat) with uniform forest structure and tree species composition (relatively homogeneous stands). Our sampling design for R. variegatus differed from the design previously described (Welsh and Lind, 1995) primarily at the fourth level, that of selecting sites within stands with minimum essential microhabitat (MEM). The intent was to maximize time, effort, and the usefulness of data sets by not sampling for salamanders at sites with an extremely low probability of occurrence. Therefore our sampling universe at the fourth level was limited to first to third order streams. An acceptable site had to contain at least 10 m2 of perennial aquatic habitat in a RHYACOTRITON VARIEGATUS HABITAT USE 387 natural watercourse. This 10 m2 could be any configuration of seep, spring, or stream channel, but the entire area had to consist of aquatic microhabitat. We used two hydrophilic plant species (California spikenard [Aralia californicus] and chain fern [Woodwardia fimbriata]), or the presence of populations of macroinvertebrates such as stoneflies (order Plecoptera) and Dobson fly larvae (Corydalus sp.), as evidence of perennial water. Final site placement was determined by the most direct approach from the nearest trail or road access. All sites were located at least 75 m from any high contrast forest edge. We selected sites with a variety of stream microhabitats (McCain et al., 1990) but avoided centering the 10 m2 sites on deep pools or high gradient/discharge habitats (e.g., waterfalls or cascades) because the use of such habitats by Rhyacotriton conflicted with literature accounts. However, these microhabitat types did occur peripherally at many sites and were sampled and included in our analysis. Animal Sampling. -Fixed area aquatic searches (Welsh, 1987; Bury and Corn, 1991) (see also quadrat sampling [Jaeger and Inger, 1994]) were conducted during daylight at each site to determine numbers of animals present. All but two sites were sampled during the summer (June to October) of 1989 to minimize seasonal effects. One or two searchers worked side by side to thoroughly search a 10 m2 plot. Each plot was systematically searched, from downstream up, with all pebbles, cobbles, and boulders turned and finer substrates carefully hand sifted down to the armoured streambed or to a depth of 15 cm, unless we saw a salamander escape deeper, in which case we pursued it. We captured both larvae and adults. We assume our capture rates are correlated to absolute densities and provide us relative densities that are valid for comparing R. variegatus numbers amoung sites. Measuring Biotic and Abiotic Parameters.-We selected habitat variables for measurement using three criteria: (1) parameters that reflected structural, compositional, and microclimatic aspects of the forest and stream environment relevant to R. variegatus as indicated by previous research; (2) parameters that would indicate change in structure and composition of the forest resulting from common management practices such as timber harvesting and reforestation, or natural successional events such as fire, landslide, or flood; and (3) variables that incorporated aspects of the forest and stream environment reflecting three scales of spatial organization: landscape, macrohabitat, and microhabitat. Our approach was to initially estimate a wide range of parameters, and then eliminate highly redundant variables using correlation analysis prior to multivariate analyses. Variables with a high number of zero values across sites (~70%) were also removed because they could not be normalized, and we believe that they were not likely to affect salamander distribution or abundance. Initial measurements of general locator variables (landscape scale), forest structure (macrohabitat scale), and microhabitat variables, resulted in a total of 121 variables (see Welsh [1993] for the complete list and details on measurements). Aquatic microhabitat categories followed McCain et al. (1990) except we combined all pool types into two categories and defined two additional categories: seep and splash zone (Table 1). Seeps were defined as shallow (<2 mm), slow-flowing water through rock substrates; splash zones were defined as wetted areas with no measureable flow or depth. Aquatic substrate composition was characterized in two ways: (1) as a visual estimate of the percent of the search area in ten substrate categories (e.g., gravel, pebble, cobble; sizes follow Platts et al., 1983); and (2) as a percent of a 4000 cm3 sample of streambed substrate, taken from a representative site just above each 10 m2 search area, dried, sifted into five size classes and weighed (listed by actual size, see Table 1). We performed preliminary descriptive analyses to assess the normality of the distributions of all variables, and deviations were corrected by appropriate transformations (Sokal and Rohlf, 198 1). Variable reduction procedures resulted in 68 independent variables for our multivariate analyses (Table 1). Statistical Analyses. -We employed discriminant analysis and regression for statistical analyses. Used together, these techniques can reveal aspects of the habitat that may be limiting for a species and also indicate those aspects that might be managed to maintain or increase animal numbers (this complimentary approach is discussed in more detail by Welsh and Lind [1995]). Life stages were combined for both analyses. In our multivariate analyses we assumed that univariate normality implied multivariate normality (we did not test multivariate normality directly). For the multivariate analyses we grouped variables into ecologically meaningful subsets (ecological components) on the basis of similarity of spatial scale and vertical stratum of the forest environment (Table 1) (cf. Bingham and Sawyer, 1991; Welsh and Lind, 1995). We then ran separate analyses on each component. This approach is a biologically sound and methodologically valid way to increase the sample size to variables ratio, which promotes a more substantive analysis when dealing with a large number of independent variables (James and