Exploratory Modeling Indicates Red-Backed Salamander Detections are Sensitive to Soil pH at C. F. Phelps Wildlife Management Area, Virginia

Abstract

Red-backed salamanders represent an important component of Virginia ecosystems, but there are few habitat models that can reliably predict the presence/absence of this species. We surveyed the habitats of red-backed salamanders at one site in the Piedmont region of Virginia and collected data on an array of habitat variables with which this species is normally associated. We used logistic regression to develop a model predicting the presence or absence of the species at a given 50m-transect. Our final model incorporated soil organic layer pH variability and mineral layer average pH, and accounted for 30% of the variation in our data. We conclude that soil pH is a limiting determinant of habitat use for this study site, and that it may affect adaptive behaviors for highly acidic soils. INTRODUCTION As researchers address the issues of amphibian decline, there is an increasing need to better understand how salamanders in terrestrial ecosystems interact with their habitat. Greater understanding of the habitat ecology of these species would likely improve our ability to manage and conserve amphibian diversity in local watersheds, thereby reducing the ecosystem damage that would result from the loss of these species (Cushman 2005, Wyman 1990). In the Rappahannock River watershed of Northern Virginia, both Mitchell (1998) and McGhee and Killian (2010) have surveyed amphibian, and specifically, salamander diversity, but little has been done to assess the habitat relationships of commonly detected species. To address this need, we conducted a preliminary study of salamander habitat for a single site in the Rappahannock River drainage at the C. F. Phelps Wildlife Management Area (WMA) concurrent with a species diversity survey and developed a simple habitat model for our most commonly detected terrestrial salamander, Plethodon cinereus Green 1818 (red-backed salamander). Corresponding author: Jay D. McGhee jmcghee@nwmissouri.edu Virginia Journal of Science, Vol. 63, No. 3, 2012 http://digitalcommons.odu.edu/vjs/vol63/iss3 138 VIRGINIA JOURNAL OF SCIENCE The red-backed salamander is common to Virginia forests and the Rappahannock River watershed, and is considered an important component of the local ecosystems in which they occur (Burton and Likens 1975, Davic and Welsh 2004). While several studies have noted particular habitat features associated with this species, such that a hypothetical niche-gestalt can be conceptualized (James 1971), only a few studies have actually developed predictive models of habitat use, primarily to compare the effects of silviculture treatments (Demaynadier and Hunter 1998, Morneault et al. 2004, McKenney et al. 2006). The red-backed salamander occurs in the leaf-litter and well-drained soil underlying deciduous, northern conifer, and mixed deciduous-coniferous forests with numerous cover objects (logs and rocks) and little underbrush (Burger 1935, Petranka 1998, Richmond and Trombulak 2009). This lungless salamander is dependent on gas exchange through the skin for respiration, and is sensitive to moisture and temperature shifts, typically adjusting to these changes by moving vertically through the soil column (Taub 1961, Heatwole 1962, Spotila 1972). They tend to prefer a neutral soil pH, cooler temperatures and ready access to lower soil layers as predation refugia (Bogert 1952, Heatwole 1962, Spotila 1972, Wyman and Hawksley-Lescault 1987). Females attach eggs within natural crevices or beneath embedded rocks or decaying logs (Petranka 1998). We wished to determine whether we could successfully predict red-backed salamander occurrence at a given site using variables associated with these general habitat features known to be key components in their ecology. We hypothesized that red-backed salamanders would be detected in leaf litter associated with cover objects and moist, cool soil conditions of neutral pH. We predicted that a logistic regression model would include variables measuring the amount of coverage by cover objects, soil moisture, and soil ph. METHODS We used transect sampling to locate salamanders (Jaeger 1994, Jaeger and Inger 1994, Mitchell 2000). We randomly selected the starting location of transects using a GPS. We sampled transects by searching five 1-m quadrats placed randomly within 2 10m increments (Jaeger 1994, Jaeger and Inger 1994, Mitchell 2000). We searched quadrats by removing large cover objects (rocks and decaying wood) and searching leaf litter (Mitchell 2000). We identified captured salamanders to species, and measured snout-vent length and total length to estimate and assign age-classes (Petranka 1998, Moore and Wyman 2010). We collected habitat data at both the transect-level and the quadrat-level. Transectlevel data included air temperature, air pressure, relative humidity, vapor pressure deficit (vapor pressure deficit represents the difference between the actual moisture in the air and the amount of moisture the air could hold when saturated at a given temperature: Bellis 1962), degree and direction of slope, general weather (clear, partly cloudy, overcast, light rain, heavy rain), and habitat (coniferous, mixed deciduous, mixed coniferous-deciduous, open-field/Rosa multiflora brush). Quadrat-level data included soil pH, soil moisture, soil temperature, leaf litter depth, and percent cover (bare ground, leaf litter, natural cover, ground vegetation, and woody stem). We determined soil pH and soil moisture of cored soil samples in a laboratory. Soil samples were obtained by taking 31.7 mm diameter soil probe cores from a quadrat Virginia Journal of Science, Vol. 63, No. 3, 2012 http://digitalcommons.odu.edu/vjs/vol63/iss3 Habitat Modeling of Red-Backed Salamander 139 until sufficient soil was obtained to fill two collection tubes (50 mL centrifuge tubes) with separate organic and mineral fractions. In the laboratory each fraction was thoroughly mixed followed by division into two approximately equal parts—one for percent soil moisture and one for ph. Percent soil moisture was determined by massing the wet samples followed by drying for 24 hours in a 50 C oven. Soil pH was determined using a Barnant 20 digital pH meter. The sample (placed in the centrifuge tube) was covered with enough distilled water to keep the pH probe above the sediment. We waited for 20 minutes to allow the more coarse soil particles to settle out. We then measured the pH after the reading stabilized, but not to exceed 1 minute. We measured leaf litter depth using a ruler placed once within a randomly chosen quadrant of the quadrat. We used the Daubenmire (1959) method to estimate ground cover within quadrats. As we had little information from which to base hypotheses regarding habitat selection at this site, we used logistic regression as an exploratory modeling approach to determine which predictor variables were most associated with captured salamanders at the transect level. For variables measured at the quadrat level, we tested both mean values and their standard deviations as predictors. From our data we created new multiplicative variables where synergistic effects seemed likely (synergistic variable 1: soil temperature*organic layer soil moisture*mineral layer soil moisture, synergistic variable 2: organic-layer soil pH*organic layer soil moisture). We used forward stepwise selection (P = 0.05 to enter and 0.10 to remove) in SPSS (SPSS Inc., Chicago IL). Variable coefficients were assessed using the change in -2 loglikelihood (Hosmer and Lemeshow 1989). The explanatory value of the selected model was evaluated using Nagelkerke’s r (Hosmer and Lemeshow 1989, Nagelkerke 1991, Ryan 1997). For all 2 statistical analyses, detection refers to whether a species was captured or not, as opposed to the number of captures; = 0.05. RESULTS From 13 April 2007 – 21 April 2009, we sampled 91 transects and 455 quadrats, locating 42 red-backed salamanders. We found individuals in 26 of 91 transects (29% encounter rate). Mean SVL for captured adults was 40.06mm ± 0.90 SE while mean SVL for captured juveniles was 27.33mm ± 1.40 SE. Our logistic regression selected a model that explained 30% of the variation in the data (r = 0.30) and produced two 2 predictor variables. The first was the standard deviation of organic soil layer pH (SDOrgpH: 6.50 ± 2.38 SE, change in -2 log likelihood = 9.350, df = 1, P = 0.002, Fig. 1). The second was the average mineral soil layer pH (AvgMinpH: -1.80 ± 0.92 SE, change in -2 log likelihood = 6.376, df = 1, P = 0.012, Fig 2). The model defined the probability of predicting the detectable presence of a red-backed salamander within a transect as equal to . It correctly predicted the absence of salamanders in 81% of cases, and correctly predicted their presence in 37% of cases. Soils throughout the study site tended to be acidic. The average organic layer soil pH across all transects was 4.62 ± 0.10 SE, and the average mineral layer soil pH was Virginia Journal of Science, Vol. 63, No. 3, 2012 http://digitalcommons.odu.edu/vjs/vol63/iss3 140 VIRGINIA JOURNAL OF SCIENCE FIGURE 1. Detection of red-backed salamanders as a function of the variability (standard deviation) in pH of the organic layer of soil for transects on C. F. Phelps Wildlife Management Area, Fauquier and Culpeper County, Virginia, April 2007 – April 2009. Detections tend to increase with increased variation in soil acidity. 4.57 ± 0.08 SE. The pH of the organic and mineral layers were highly correlated (r = 0.93) and 77% of our sites had organic fractions with pH 5 and 83% of the sites had mineral fractions with pH 5. All of our captures were in soils with a pH between 3.5 and 6.5 for the organic layer and between 3.9 and 5.3 for the mineral layer. DISCUSSION The model explained a substantial amount of the variation in presence and absence data. Haan et al. (2007) found similar results in their investigation of Aneides hardii Taylor 1941 (Sacramento salamander) where the best of 18 models were able