Do silvicultural practices to restore oaks affect salamanders in the short term?

Salamanders are an important ecological component of eastern hardwood forests and may be affected by natural or silvicultural disturbances that alter habitat structure and associated microclimate. From May to August in 2008 (pretreatment) and 2011 (post-treatment), we evaluated the response of salamanders to three silvicultural practices designed to promote oak regeneration — prescribed fire, midstory herbicide application and shelterwood harvest — and a control. We trapped salamanders using drift fences with pitfall traps in five replicates of the four treatments. Only the southern gray-cheeked salamander Plethodon metcalfi and the southern Appalachian salamander P. teyahalee were captured in sufficient numbers for robust statistical analysis. We analyzed data for these species using single-species dynamic occupancy models in statistical software program R. We allowed changes in four covariates to influence extinction probability from pre- to post-treatment implementation: 1) percent leaf litter cover; 2) percent understory cover; 3) percent CWD cover; and 4) percent canopy cover. The final combined model set describing extinction probability contained four models with &Dgr;AIC < 2 for P. metcalfi and nine models with &Dgr;AIC < 2, including the null model, for P. teyahalee. For both species, the 95% confidence intervals for model-averaged extinction probability parameter estimates overlapped zero, suggesting none were significant predictors of extinction probability. Absence of short-term salamander response in midstory herbicide and prescribed burn treatments was likely because of minor or transitory changes to forest structure. In shelterwood harvests, any potential effects of reduced canopy and leaf litter cover may have been mitigated by rapid post-treatment vegetation sprouting. Additionally, climatic conditions associated with high elevation sites and high amounts of rainfall in 2011 may have compensated for potential changes to microclimate. Continued monitoring of Plethodon salamanders to assess responses at longer time scales (e.g. > 3 years post-treatment) is warranted.

Do silvicultural practices to restore oaks affect salamanders in the short term?
Salamanders are an important ecological component of eastern hardwood forests and may be aff ected by natural or silvicultural disturbances that alter habitat structure and associated microclimate. From May to August in 2008 (pretreatment) and 2011 (post-treatment), we evaluated the response of salamanders to three silvicultural practices designed to promote oak regeneration -prescribed fi re, midstory herbicide application and shelterwood harvest -and a control. We trapped salamanders using drift fences with pitfall traps in fi ve replicates of the four treatments. Only the southern gray-cheeked salamander Plethodon metcalfi and the southern Appalachian salamander P. teyahalee were captured in suffi cient numbers for robust statistical analysis. We analyzed data for these species using single-species dynamic occupancy models in statistical software program R. We allowed changes in four covariates to infl uence extinction probability from pre-to post-treatment implementation: 1) percent leaf litter cover; 2) percent understory cover; 3) percent CWD cover; and 4) percent canopy cover. Th e fi nal combined model set describing extinction probability contained four models with Δ AIC Ͻ 2 for P. metcalfi and nine models with Δ AIC Ͻ 2, including the null model, for P. teyahalee . For both species, the 95% confi dence intervals for model-averaged extinction probability parameter estimates overlapped zero, suggesting none were signifi cant predictors of extinction probability. Absence of short-term salamander response in midstory herbicide and prescribed burn treatments was likely because of minor or transitory changes to forest structure. In shelterwood harvests, any potential eff ects of reduced canopy and leaf litter cover may have been mitigated by rapid post-treatment vegetation sprouting. Additionally, climatic conditions associated with high elevation sites and high amounts of rainfall in 2011 may have compensated for potential changes to microclimate. Continued monitoring of Plethodon salamanders to assess responses at longer time scales (e.g. Ͼ 3 years post-treatment) is warranted.
Ecologically, salamanders are important components of eastern hardwood forests in the United States. In some areas of the Appalachian Mountains, biomass of salamanders is twice that of birds and equal to small mammals (Burton and Likens 1975). Salamanders are an important food source for reptiles, birds and mammals and can act as regulators of forest fl oor invertebrates (Whitaker and Rubin 1971, Pough 1983, Davic and Hartwell Jr. 2004, Walton et al. 2006. Members of the genus Plethodon are ectothermic, lungless terrestrial salamanders with direct development (no aquatic larval stage) and require moist substrates and high relative humidity for dermal respiration (Spotila 1972, Feder 1983, Petranka 1998. Forest disturbances that substantially reduce canopy and leaf litter cover may increase risk of salamander desiccation by elevating temperatures and reducing moisture on the forest fl oor (Ash 1988, Petranka et al. 1994, Harpole and Haas 1999. Additionally, reduced cover may increase the risk of predation and alter abundance and composition of invertebrate prey (Ash 1997, Knapp et al. 2003, Homyack and Haas 2009, Matthews et al. 2010. Terrestrial salamanders may respond to habitat changes by reducing aboveground activities (e.g. foraging and mating) or by trading growth and reproduction for increased basic maintenance costs (Homyack et al. 2011). Th erefore, silvicultural practices that substantially alter forest vegetation and microclimate could have negative eff ects on Plethodon populations.
Widespread oak regeneration failure -the failure of oak seedlings or saplings to attain canopy status -is a focal issue of forest land managers. Oak regeneration is especially problematic on intermediate and highly productive sites where competition from faster-growing species such as yellowpoplar Liriodendron tulipifera is fi erce after canopy release (Aldrich et al. 2005). Several silvicultural methods for regenerating oak by altering light conditions have been proposed, but few have been experimentally tested on intermediate Table 1. List of a priori candidate models used in dynamic occupancy of Plethodon metcalfi and P. teyahalee on Cold Mountain Game Land, Haywood County, NC. Covariates are changes in estimates from pre-(i.e. 2008) to post-(i.e. 2011) oak regeneration treatments. ψ ϭ pretreatment year, Î ϭ extinction probability, p ϭ detection, γ ϭ colonization probability, canopy ϭ percent canopy cover, CWD ϭ percent coarse woody debris cover, leaf litter ϭ percent leaf litter cover, understory ϭ percent understory cover.
We conducted a large-scale replicated study to determine short-term (i.e. Ͻ 3 years) response of salamanders to three oak regeneration practices (prescribed fi re, midstory herbicide application, and a shelterwood harvest) and a control. Terrestrial salamanders are largely subterranean, with only a small proportion of the population captured on a given sampling occasion (Hairston 1987). Detection probability likely varies spatially and temporally (DeMaynadier and Hunter Jr. 1998, Hyde and Simons 2001, Petranka and Murray 2001. Th erefore, using pretreatment (baseline) and post-treatment data, we used the spatial and temporal replication inherent in a before -after control -impact experimental design and analyzed data with hierarchical dynamic occupancy models to determine the short-term response of Plethodon salamanders to the environmental changes associated with the oak regeneration treatments (MacKenzie et al. 2003).

Study area
Our study was conducted in Haywood County, North Carolina on Cold Mountain Game Land (CMGL) located in the Blue Ridge Physiographic Province. CMGL encompassed 1333 ha of second growth, upland mixedoak forests with elevations ranging from 940 -1280 m and was managed by the North Carolina Wildlife Resources Commission for diverse vegetation conditions attractive to a variety of wildlife species. Terrain was mountainous with gentle to steep slopes with an overstory of oak Quercus spp., hickory Carya spp. and yellow-poplar. Composition in the midstory consisted primarily of shade-tolerant species, including sourwood Oxydendrum arboreum , fl owering dogwood Cornus fl orida , silverbell Halesia tetraptera , blackgum Nyssa sylvatica and red maple Acer rubrum . Th e precipitation averaged 1200 mm annually.
We established fi ve, 5-ha units of the four treatments for a total of 20 units. Treatment units were separated by a Ͼ 10-m buff er and contained mature ( Ͼ 70 years old), fully stocked, closed-canopied stands where oaks comprised at least 10% of the overstory tree basal area ( Ն 25.0 cm dbh). We selected stands that contained Ͼ 1000 oak seedlings ha Ϫ1 , few ericaceous shrubs, ∼ 2 m 2 ha Ϫ1 of basal area beneath the main canopy, and no substantial disturbance within the last 15 -20 years (Keyser et al. 2008). All treatment units were between 940 -1240 m in elevation. Aspect varied within and among treatment units because of complex topography. We randomly assigned the prescribed fi re, midstory removal using herbicide, and shelterwood harvest treatments to units, resulting in a completely randomized design.

Treatments
We evaluated the initial phase (1 -2 years post-treatment) of three long-term oak regeneration practices: prescribed burns at ∼ 4-year intervals, midstory removal using herbicide with re-application after three years, and a shelterwood harvest followed by a prescribed fi re after three years. Specifi cally, our study evaluated the response of Plethodon salamanders to four treatments: 1) one prescribed burn, 2) one midstory herbicide treatment, 3) a shelterwood harvest with 30 -40% basal area retention, and a 4) control.
Two of the fi ve prescribed burn replicate units were burned in April 2009 with the remaining three units burned in April 2010. In midstory herbicide replicate units, competing midstory trees were treated with Garlon 3A using the hack-and-squirt method in late summer 2008 to reduce total basal area by 25 -30% without creating new canopy gaps (Loftis 1990). Shelterwood harvest replicate units were harvested from winter 2009 to early summer 2010 with the goal of leaving approximately 30 -40% of the original stand basal area and enhancing light conditions on the forest fl oor (Brose et al. 1999). Trees felled with standard chainsaws and grapple cutters were dragged by rubber tire skidders to log landings where loaders fi lled haul trucks. Some units required skid trails due to steep areas. During this study, no silvicultural manipulation occurred in the control plots (Keyser et al. 2008, Raybuck et al. 2012).

Salamander sampling
In 2008 (pre-treatment) and 2011 (post-treatment), we sampled salamanders from mid-May to mid-August in all 20 units. We established six randomly oriented single-arm drift fence arrays in 2008 (Todd et al. 2007) with three fences installed at a lower slope site (e.g. lower one-third of each unit) and three fences at an upper slope site (Greenberg and Waldrop 2008). In two of the treatment replicates (one herbicide and one control), we were unable to establish an upper site because of steep and rocky terrain. A fourth fence was installed at each lower and upper site that had fences by 2011. Drift fences were Ͼ 10 m apart and constructed of 7.6-m sections of aluminum fl ashing with a 19-l bucket buried at each end, fl ush with the ground. We placed a moist sponge in each bucket to provide moisture for captures. We had fi ve observers in 2008 and four in 2011. Each drift fence was checked at least every two days; we checked fences daily after rain events. We recorded each captured individual to species and released it immediately at the capture location.
All procedures used in our study were in accordance with the Society for the Study of Amphibians and Reptiles guidelines for fi eld research, and approved by the Institutional Animal Care and Use Committee of North Carolina State University (permit no. 08-035-O).

Habitat data
At the middle drift fence in the lower and upper slope site of each unit, we recorded aspect (1 ϭ northern to eastern aspects [mesic]; 0 ϭ southern to western aspects [xeric]) and elevation. We estimated elevation using a handheld GPS. Pre-and post-treatment, we measured percent canopy cover with a spherical densiometer at each drift fence (Lemmon 1956). Pre-and post-treatment, we measured percent leaf litter cover, percent understory cover, and percent cover of coarse woody debris (CWD) along a 15-m randomly oriented transect line at each drift fence at the bucket furthest uphill. We measured percent leaf litter cover, percent understory cover (i.e. plants shorter than Ͻ 0.9 m in height), and percent CWD cover by summing the total distance for each category along each transect. Th en, in each unit, we divided the total distance for each category by 90 m (six transects per unit).

Analyses
We analyzed the change in habitat variables from 2008 to 2011 using a one-way analysis of variance (ANOVA) with treatment as a class variable. When models were signifi cant, we used Tukey ' s studentized range test to determine diff erences among treatment means. We conducted statistical tests in SAS (ver. 9.1.3, SAS Inst.).
Because abundance estimation methods using markrecapture are often logistically diffi cult and ineff ective for large-scale monitoring of terrestrial salamanders (Bailey et al. 2004), we used hierarchical dynamic occupancy models in package " unmarked " (Fiske and Chandler 2011) in statistical software program R to fi t models for Plethodon species for which we had Ͼ 70 captures per year (MacKenzie et al. 2003, Ͻ www.r-project.org/ Ͼ ). Dynamic occupancy models allow inference about the occurrence of species at several sites and about how changes in occurrence are driven by colonization and extinction while simultaneously allowing for imperfect detection (MacKenzie et al. 2003, K é ry and Chandler 2012). We assumed each site (i.e. drift fence array) was either occupied or not occupied by a given species before treatment implementation (i.e. 2008) and after treatment implementation (i.e. 2011). All parameters (i.e, initial occupancy, colonization, extinction and detection) can be modeled as functions of covariates (K é ry and Chandler 2012). Colonization was the probability that a site not occupied pre-treatment implementation became occupied post-treatment implementation. Extinction was the probability that a site occupied pre-treatment implementation became unoccupied post-treatment implementation.
We summarized count data for each site (i.e. drift fence array) from 2008 and 2011 and combined data into 10 sampling occasions per year per species. We ran 16 a priori candidate models with covariates on initial occupancy probability during the pre-treatment year ( ψ ), extinction probability ( Î ), and detection ( p ) while keeping colonization probability ( γ ) constant (Table 1). We predicted that forest changes resulting from oak regeneration treatments were most likely to result in local extinctions of salamanders and rarely colonization of locations; therefore, we identifi ed covariates responsible for potential extinctions. We believed local-scale habitat variables allowed us to understand the mechanisms that might be causing salamander response to silvicultural manipulations better than the large-scale treatment classifi cations. Th us, we did not use treatment as a covariate of salamander occupancy dynamics, but instead investigated the habitat metrics at the drift fence level. More specifi cally, we subtracted the 2008 estimates from the 2011 estimates to quantify the change in each covariate. Because of the large number of potential models, we followed a procedure to identify covariates that aff ected extinction probability. We allowed elevation to infl uence initial occupancy because this variable is known to aff ect occurrence of salamander species (Ford et al. 2002, Takahashi andPauley 2010). We allowed change in percent leaf litter cover and date to infl uence detection because these variables were most likely to aff ect salamander activity patterns (Ash 1995, Petranka 1998. We allowed changes in four covariates to infl uence extinction probability from pre-to post-treatment implementation: 1) percent leaf litter cover; 2) percent understory cover; 3) percent CWD cover; and 4) percent canopy cover. Hence, we modeled elevation on the initial occupancy process, date and percent leaf litter cover on the detection process, and every combination of the four covariates on the extinction process for 15 of the 16 candidate models (Table 1). Th e remaining candidate model was the null extinction probability model. We standardized all covariates before analysis. We tested covariates for collinearity using the correlation function in Excel. We removed a covariate if it was Ͼ 0.6 correlated with another covariate. We ranked convergent models using Akaike's information criterion (AIC) (Burnham and Anderson 2002). We used R package "MuMIn" to obtain model-averaged point estimates of the parameters (Barton 2013). We used a parametric bootstrap procedure with a chi-square fi t statistic to assess goodness of fi t of the AIC best model (Hosmer et al. 1997, K é ry and Chandler 2012).

Results
Elevation (F 4,14 ϭ 2.12, p ϭ 0.12) and aspect (F 4,31 ϭ 1.96, p ϭ 0.13) were similar among treatments. From pre-to post-treatment, shelterwood harvest units showed the most marked habitat changes; percent understory cover increased by 53%, percent leaf litter cover declined by 47%, percent canopy cover declined by 41%, and percent CWD cover increased by 6% (Table 2). Prescribed burns of 2009 showed a 44% increase in percent understory cover (Table 2). Habitat variables in the midstory herbicide treatment and prescribed burns of 2010 were similar to control units from pre-to post-treatment (Table 2).

Salamander sampling
We captured a total of 607 salamanders during the course of the study.      Table 5. Model set including all models with an Akaike's Information Criterion (AIC) Ͻ 2 from the AIC best model with difference in AIC ( Δ AIC), model weight ( w i ) , and number of parameters ( K ) for estimating dynamic occupancy of Plethodon metcalfi following oak regeneration treatments from 2008 (pre-treatment) to 2011 (post-treatment) on Cold Mountain Game Land, Haywood County, NC. Covariates represent changes in habitat estimates from pre-(i.e. 2008) to post-(i.e. 2011) oak regeneration treatments. * indicates the null model, which was not included in the top models or the model-averaged estimates. ψ ϭ pretreatment year, Î ϭ extinction probability, p ϭ detection, γ ϭ colonization probability, ele ϭ elevation, canopy ϭ percent canopy cover, CWD ϭ percent coarse woody debris cover, leaf litter ϭ percent leaf litter cover. viridescens , 17 Ocoee salamanders D. ocoee , seven blackchin red salamanders Pseudotriton ruber schencki , six southern redbacked salamanders P. serratus , and one seepage salamander D. aeneus (Table 3). Only P. metcalfi and P. teyahalee were captured in suffi cient numbers for robust statistical analysis.

Discussion
In the short-term ( Ͻ 3 years), Plethodon metcalfi and P. teyahalee populations did not appear to be aff ected negatively by the experimental oak regeneration treatments. Th e weak relationship that suggests extinction probability for P. metcalfi increases with increasing CWD cover is counter to studies that showed salamanders use and benefi t from CWD (Herbeck and Larsen 1999, McKenny et al. 2006, Semlitsch et al. 2009). CWD increased following shelterwood harvests, so the relationship between CWD and extinction probability may have been linked to greater risk of extinction in shelterwood harvests. In fact, P. metcalfi captures following shelterwood harvests were ∼ 50% less than pre-treatment. In this case, other unmeasured environmental changes following shelterwood harvests (e.g. soil moisture) may have infl uenced salamander occupancy dynamics.
Our results concur with several other studies indicating that salamanders generally show no response or slightly increase in abundance following herbicide applications (Cole et al. 1997, Hood et al. 2002, Brunjes et al. 2003, Homyack and Haas 2009, Cantrell et al. 2013 or single prescribed burns (Ford et al. 1999, Keyser et al. 2004, Greenberg and Waldrop 2008. In fact, Keyser et al. (2004) reported no detectable diff erences in abundance of P. cinereus after highintensity prescribed burns that substantially reduced leaf litter. In our study, prescribed fi res were cool and relatively low-intensity with maximum ground level temperatures of 131 ° C. Post-burn occupancy did not decline, indicating short-term environmental changes following single prescribed fi res were not enough to impact P. metcalfi or P. teyahalee . Rapid recovery of leaf litter and understory cover may have mitigated potential stresses on moisture-sensitive Plethodon .
Although shelterwood harvests resulted in substantial reductions in canopy and leaf litter cover, environmental conditions unique to our study sites may have precluded short-term changes in Plethodon occupancy. A concurrent study within the mid-Cumberland Plateau of southern Tennessee examined the short-term response (1 -2 years) of herpetofauna to a shelterwood harvest with 30 -40% basal area retention. Similar to our fi ndings, no negative responses for herpetofauna species, including for the northern slimy salamander P. glutinosus , were detected (Cantrell et al. 2013). Conversely, other studies have documented shortterm Plethodon declines following shelterwood harvests, suggesting that as little as 41% overstory reduction may result in declines similar to clearcuts (Harpole and Haas 1999, Knapp et al. 2003, Homyack and Haas 2013. In our study, rapid stump sprouting and understory recovery, as well as increased CWD following shelterwood harvests may have provided salamanders with suffi cient cover, shade and moisture, travel corridors, and foraging opportunities despite declines in tree canopy cover (Grover 1998, Morneault et al. 2004, Patrick et al. 2006, Rundio and Olson 2007, Semlitsch et al. 2009). Emigration or mortality, which normally begins late in the fi rst year following disturbance and in subsequent years because of starvation and dehydration (Knapp et al. 2003), may have been off set by high levels of precipitation in 2011. Average precipitation from May to August for Haywood County is 488 mm compared to 1090 mm recorded in 2011. High rainfall may have compensated for potential moisture reductions due to canopy removal by maintaining moisture levels within the tolerance limits of Plethodon salamanders. Additionally, the relatively cool climate of our high elevation sites may have mitigated some of the microclimatic conditions that likely were associated with post-harvest salamander declines in other studies (Harper and Guynn Jr. 1999).
Our short-term sampling window ( Ͻ 2 years post-harvest) may have been insuffi cient to detect impacts on terrestrial salamanders in shelterwood units (DeMaynadier and Hunter Jr. 1995). Studies indicate that Plethodon salamanders can show high site fi delity for up to two years following silvicultural disturbance (Ash 1988, Messere and Ducey 1998, Ford et al. 2000, Bartman et al. 2001). However, lower salamander abundance has been documented in forest stands 2 -10 years following Ͼ 50% canopy removal as compared to mature forest (Pough et al. 1987, Ash 1988, Petranka et al. 1993, Reichenbach and Sattler 2007. Similarly, declines in reproductive success may not be evident for more than three years following forest disturbance (Ash et al. 2003, Patrick et al. 2006, Cummer and Painter 2007, Matthews et al. 2010. Ash et al. (2003) noted that P. jordani captures in 10-year old clearcuts had proportionately fewer juveniles and adult males in reproductive condition.
Planned activities associated with the oak regeneration prescriptions at these sites, including repeated prescribed burns in the prescribed fi re treatment units and prescribed burns in the shelterwood harvest treatment units, may have additive eff ects on salamanders by repeatedly or further reducing canopy, understory and leaf litter cover. Repeated prescribed fi res that potentially facilitate growth of oak seedlings to a competitive size (Van Lear and Watt 1993) could compound treatment eff ects (e.g. reduced leaf litter and moisture at the forest fl oor) and aff ect Plethodon salamanders by decreasing relative humidity in burrows (Floyd et al. 2001). Additionally, the combination of shelterwood harvests and prescribed fi res may substantially change habitat conditions compared to each of these disturbances alone. For example, Matthews et al. (2010) captured 68 -72% fewer salamanders in mechanical fuel reduction treatments followed by two prescribed fi res than in twice-burned treatments alone. Th e high severity prescribed fi res in the mechanical and twice burned treatment units led to substantial overstory mortality compared to the lower-severity prescribed burns in the burn only treatment units (Matthews et al. 2010).

Conclusions
Minor changes to habitat structure following midstory herbicide treatments, rapid re-establishment of understory cover in shelterwood harvests and prescribed burns, and microclimatic conditions of high elevation sites may have contributed to maintaining Plethodon populations following oak regeneration treatments in the short-term. However, planned activities associated with the oak regeneration prescriptions, such as prescribed burns following shelterwood harvests, may have additive eff ects on salamanders. Additionally, lags in demographic response may cause delayed declines in Plethodon populations. Th erefore, longer-term studies are necessary to assess salamander response to varying frequencies and combinations of oak regeneration treatments that could aff ect microclimatic conditions.