Vegetation and elevation influence the timing and magnitude of soil CO 2 efflux in a humid , topographically complex watershed

Vegetation and elevation influence the timing and magnitude soil CO2 efflux in a humid, topographically complex watershed J. W. Atkins, H. E. Epstein, and D. L. Welsch University of Virginia, Department of Environmental Sciences, Charlottesville, Virginia American Public University System, Charles Town, West Virginia In topographically complex watersheds, landscape position and vegetation heterogeneity can alter the soil water regime through both lateral and vertical redistribution, respectively. These alterations of soil moisture may have significant impacts on the spatial heterogeneity of biogeochemical cycles throughout the watershed. To evaluate how landscape position and vegetation heterogeneity affect soil CO2 efflux (FSOIL) we conducted observations across the Weimer Run watershed (373 ha), located near Davis, West Virginia, for three growing seasons with varying precipitation. An apparent soil temperature threshold of 11 °C at 12 cm depth on FSOIL was observed in our data—where FSOIL rates greatly increase in variance above this threshold. We therefore focus our analyses of FSOIL when soil temperature values were above this threshold. Vegetation had the greatest effect on FSOIL rates, with plots beneath shrubs at all elevations, for all years, showing the greatest mean rates of FSOIL (6.07 μmol CO2 m -2 s) compared to plots beneath closed-forest canopy (4.69 μmol CO2 m -2 s) and plots located in open, forest gaps (4.09 μmol CO2 m -2 s) plots. During periods of high soil moisture, we find that CO2 efflux rates are constrained and that maximum efflux rates occur during periods of average to below average soil water availability. While vegetation was the variable most related to FSOIL, there is also strong inter-annual variability in fluxes determined by the interaction of annual precipitation and topography. These findings add to the current theoretical constructs related to the interactions of moisture and vegetation on biogeochemical cycles within topographically complex watersheds.


Introduction
Soil respiration (R SOIL ) is a major component of the terrestrial carbon cycle (Raich and Potter, 1995;Schimel, 1995) and is 30-60 % greater than net primary productivity globally (Raich and Potter, 1995).Estimates of annual soil carbon emissions range from 68 to 100 Pg of carbon per year (Schlesinger, 1977;Raich and Schlesinger, 1992;Bond-Lamberty and Thomson, 2010).Temperate systems contribute approximately 20 % of the annual global R SOIL (Bond-Lamberty and Thomson, 2010) but have been shown to be recent carbon sinks, averaging 0.72 Pg of C uptake per year from 1990 to 2007 (Pan et al., 2011).R SOIL can be estimated in the field by measuring soil CO 2 efflux (F SOIL )the direct rate of CO 2 crossing the soil surface over a period of time (Raich and Schlesinger, 1992).F SOIL can vary spatially and temporally within and across systems as a result of the varied and complex interactions of controlling mechanisms (Drewitt et al., 2002, Trumbore, 2006;Vargas et al., 2010).The edaphic controls on F SOIL on the landscape scale include soil temperature, soil moisture, root biomass, microbial biomass, soil chemistry, and soil physics (Fang et al., 1998;Davidson et al., 1998;Kang et al., 2000;Xu and Qi, 2001;Epron et al., 2004).These factors do not simply elicit additive or monotonic responses but, rather, create complex responses of F SOIL across spatial and temporal scales (Dilustro et al., 2005;Pacific et al., 2009).
Soil temperature is quite commonly a primary driver of F SOIL (e.g., Fang and Moncrieff, 2001), and in complex terrain, temperature regimes can be mediated by elevation, slope, and aspect (Wu et al., 2013).The effects of elevation and topography on soil temperature can in turn affect Published by Copernicus Publications on behalf of the European Geosciences Union.carbon cycling (Schindlbacher et al., 2010) either directly or through indirect processes (Murphy et al., 1998).Soil water content (SWC), however, often serves as an important secondary control on F SOIL .At high SWC values, CO 2 transport through the soil pore space is limited (Davidson and Trumbore, 1995;Jassal et al., 2005).Production of soil CO 2 can also become limited at high SWC values due to anoxia and decreased microbial aerobic respiration (Oberbauer et al., 1992).At low SWC values, F SOIL is decreased as well due to microbial desiccation and concomitantly reduced microbial activity (Van Gestel et al., 1993), resulting in decreased CO 2 production (Scanlon and Moore, 2000).
In topographically complex landscapes, precipitation gradients that exist as a function of elevation affect decomposition rates, CO 2 production, and movement of CO 2 through the soil (Schuur and Matson, 2001).The complex landscape structure and heterogeneity of mountain catchments also directly affect local soil moisture regimes through the lateral redistribution of soil water, adding to the spatial heterogeneity of these biogeochemical and physical processes.F SOIL therefore varies across landscape positions as a function of this soil water redistribution (Riveros-Iregui and McGlynn, 2009).In subalpine forested systems for example, soil water content has been shown to be a strong driver of the spatial (Scott-Denton et al., 2003) and temporal (Pacific et al., 2008) variability of F SOIL .
In addition to meteorological variables, vegetation (itself controlled by the spatial heterogeneity of micrometeorology) can influence carbon cycling within a watershed.Vegetation affects carbon cycling directly through photosynthesis (Raich and Schlesinger, 1992;Ekblad and Högberg, 2001;Högberg et al., 2001), above-and belowground tissue allocation (Chen et al., 2013), and litter production (Prevost-Boure et al., 2010).Vegetation therefore controls the quantity and quality of soil organic matter (SOM) within systems, which in part will determine decomposition rates and soil CO 2 production (e.g., Berg, 2000).However, the role of belowground plant and microbial processes in the dynamics of SOM has become increasingly more apparent, showing that root and rhizosphere contributions to SOM are substantive (e.g., Schmidt et al., 2011).Vegetation also exerts controls on production of CO 2 through root respiration in the soil and through complex mycorrhizal associations that can mediate the response of soil CO 2 production to rain pulse events (Vargas et al., 2010).Finally, vegetation also elicits feedbacks on the abiotic aspects of a system, including the soil moisture and soil temperature regimes, further impacting biogeochemical cycling (Wullschleger et al., 2002;Metcalfe et al., 2011;Vesterdal et al., 2012).
Interannual variation in R SOIL within systems can be high and exceed the interannual variation in net ecosystem exchange (NEE) of carbon (Savage and Davidson, 2001); this interannual variation can be driven in large part by the dynamics of precipitation (Raich et al., 2002).Current climate models project potentially dramatic changes in precipitation in the coming years (Kirtman et al., 2013), and presently the controls on interannual variation in R SOIL in response to changing precipitation regimes are poorly understood on spatial scales ranging from landscapes to regions.The interactions among topography, vegetation cover, and climate are therefore an important and complicated area of study.
Interannual climate variability in mountainous, subalpine catchments, however, has been shown to alter the spatiotemporal heterogeneity of carbon dynamics within those systems (Riveros-Iregui et al., 2011, 2012).In a subalpine watershed in Montana, Riveros-Iregui et al. (2012) found that areas with low upslope accumulated area (generally uplands and drier areas) showed F SOIL increases during wet years, while poordrainage areas (riparian areas) showed F SOIL decreases during wet years.This resulting bidirectional response is a function of the landscape heterogeneity of the system, soil biophysics, and interannual climate variability (Riveros-Iregui et al., 2012).
Given the possible interactions among precipitation, topography, and vegetation, we examined how F SOIL varies as a function of landscape position and vegetation cover in response to interannual variation in precipitation within a complex, humid watershed.To do this we used a plot-based approach with repeated measures sampling to account for spatial and temporal variation in the biophysical controls on F SOIL within our study watershed.The empirical nature of this study design, coupled with the use of portable infrared gas analyzers (IRGAs) to measure soil CO 2 efflux, is a robust and proven way of quantifying the seasonal dynamics of F SOIL and allows for greater consideration of the spatial variability of F SOIL (Riveros-Iregui et al. 2008;Riveros-Iregui and McGlynn, 2009) on the watershed scale.We attempted to answer the following questions: 1. How does F SOIL respond to interannual variation in precipitation in a humid, complex watershed?
2. How do landscape position and vegetation heterogeneity affect F SOIL , and how do they interact with interannual variation in precipitation?July temperature is 18.8 • C, and the mean daily maximum January temperature is −3.9  5a).

Site description
The Weimer Run watershed is adjacent to the Canaan Valley in West Virginia, which exists in a transitional zone between the Appalachian Valley and Ridge and the Appalachian folded plateau (Matchen, 1998).The surrounding ridge tops and the study site are underlain by Pennsylvanian sandstone from the Pottsville formation (Allard and Leonard, 1952).The overstory vegetation within the watershed is a mixed northern hardwood coniferous forest, consisting of yellow birch (Betula alleghaniensis), red maple (Acer rubrum), red spruce (Picea rubens), and black cherry (Prunus serotina; Allard and Leonard, 1952;Fortney, 1975).The understory is comprised of Rhododendron max-imum, Kalmia latifolia, Osmundastrum cinnamomeum, and Osmunda claytoniana (Fortney, 1975).

Vegetation and elevation classes
Three elevation classes were established along the northeastern aspect of the watershed to form an elevation gradient: "Low" (975 m), "Mid" (1050 m), and "High" (1100 m).Site elevations were determined using a digital elevation map (DEM) derived from 1/9 arc second elevation data from the Shuttle Radar Topography Mission (SRTM) (USGS 2006) processed with ArcGIS ® software (ESRI; Redlands, CA).In order to address the effects of vegetation cover on F SOIL , three vegetation cover classes were established: "Canopy"closed canopy, forest interior with no shrub layer; "Shrub"closed canopy, forest interior, with dense shrub layer; "Open" -forest gap with no canopy closure, within the forest interior.Differences among vegetation classes were confirmed using a plant area index (PAI) which was measured for each plot in June 2010 with a LAI-2000 plant canopy analyzer (LI-COR Lincoln, Nebraska).PAI was strongly statistically significantly different among vegetation cover types (F = 13.39;p value = 0.0003).Shrub plots were the greatest (3.46 m −3 m 3 ), followed by Canopy plots (2.14 m −3 m 3 ) and then Open plots (1.75 m −3 m 3 ; Table A1 in the Appendix).
At each elevation level in the watershed, three 2 × 2 m plots of each vegetation class were established -for a total of 27 plots across the entire watershed (Fig. 1).One of the Open replicate plots at the Low elevation had to be removed from analysis due to inundation during the summer of 2011.Data from the remaining 26 plots were analyzed.

Soil CO 2 efflux
An EGM-4 portable infrared gas analyzer (IRGA) with an attached SRC-1 soil chamber (PP Systems, Amesbury, MA) was used to measure soil CO 2 efflux rates.The EGM-4 has a measurement range of 0-2000 ppm (µmol mol −1 ), with an accuracy of better than 1 % and linearity better than 1 % throughout the range.The SRC-1 has a measurement range of 0-9.99 g CO 2 m −2 h −1 .Plots were sampled approximately weekly (every 5-10 days) from the middle of May until the end of September, from 2010 to 2012.For March until mid-May, and during October and November, plots were measured approximately every 2 weeks (12-21 days) during times when they were snow-free.F SOIL was measured 1-3 times at different locations within the plot at each measurement interval and averaged for a plot-level estimation of F SOIL .Plots were sampled between 09:00 and 16:00 EST, and the sequence of plot measurements was varied to avoid a time-of-day bias in the results and account for diurnal variation in soil CO 2 flux over time.Our sampling followed a rotating scheduling where for one sampling period we would start at, say, the High elevation, then proceeding to work down the mountain (Mid, then Low), and the next week we would start at the Mid, then working down to the Low finishing with the High, and the next week we would then start at the Low, then High, then Mid, and so on.This method was followed through the experiment.

Volumetric water content
Volumetric water content ( field ) was measured using a Campbell HydroSense CD 620 (Campbell Scientific) set to water content measure mode with 12 cm probes (Campbell Scientific; ±3.0 % m −3 m 3 , with electrical conductivity < 2 dS m −1 ; sampling volume using 12 cm rods was ca.650 cm 3 ).A minimum of three measurements was taken in each plot per sampling event and averaged to make a plotlevel estimation of field .
Measurements taken by the Campbell HydroSense CD 620 have a known bias in soils where bulk density is outside of the 1-1.7 g cm 3 range, where organic matter is > 10 %, and where clay content is > 40 %.(Campbell Scientific).In order to calibrate field measurements, a calibration procedure from Kelleners et al. (2009) was followed where P , the period, which is the square wave output from the probe in milliseconds, is converted to K a , the relative soil permittivity (unitless).P is related to field as shown in Eq. ( 1): Equation ( 2) converts P to K a .
K a = (P − P air ) (P water − P air ) where P air is the period in air and P water is the period in deionized water.P air was calculated empirically at 0.79 ms.P water was calculated at 1.37 ms following the procedure outlined in Kelleners et al. (2009) by placing the probes of the Campbell Hydrosense CD 620 in deionized water in an 18.92 L acid-washed container, with total vessel conductivity measured at 0.47 µS.Soil samples were taken in conjunction with Hy-droSense measurements in 2012 (depth: 12 cm; volume: 56.414 cm 3 ; n = 37), and actual VWC (volumetric water content, m −3 ) m 3 )) ( lab ) was calculated using Eq.(3) from Rose ( 2004), where w is the gravimetric water content of the soil sample (g −3 g 3 ), ρ b is the soil bulk density (g cm −3 ), and ρ (g cm −3 ) is the density of water: In order to calibrate field measurements of VWC ( field ), √ K a values were then regressed against lab to create an Eq. ( 4), relating √ K a to (R 2 = 0.74) such that field measurements of VWC ( field ) could be converted to in order to account for discrepancies in organic matter, soil bulk density, and clay content: was then converted to water-filled pore space (WFPS; m −3 m 3 ) using the soil porosity ( ; m −3 m 3 ): (5) WFPS provides a more mechanistic variable that takes into account the bulk density and porosity of the soil, which influence the transport and storage capacity of the soil with regard to soil CO 2 .

Soil temperature
During each field sampling session, soil temperature (T SOIL ; • C) was measured at 12 cm using a 12 cm REOTEMP soil thermometer (REOTEMP San Diego, CA) at a minimum of two locations within the plot.These measurements were averaged to create a plot mean temperature for each sampling event.

Soils
Soil pH was determined using a 1 : 1 measure of soil (from 0 to 5 cm depth) with deionized water and measured with a Fieldscout SoilStik pH Meter (Spectrum Technologies, Inc. Plainfield, IL) with an accuracy of ±0.01 pH and ±1 • C. Soil samples were taken from 0-5, 0-12, and 0-20 cm profiles within the soil.Soil bulk density (ρ s ), total bulk density (ρ t ), soil particle density, and soil porosity ( ) were also calculated for each sample (Grossman and Reinsch, 2002;Flint and Flint, 2002).Soil bulk density (ρ s ) is defined as the bulk density of the soil fraction, where the soil fraction consists of soil that has been sieved to less than 2 mm and all gravel and root material has been removed.Total bulk density (ρ t ) is defined as the absolute density of the sampled soil, including soil, roots, and gravel, and is simply the sample dry mass over the sample volume.Total soil carbon and nitrogen were assessed using an NA 2500 elemental analyzer (CE instruments; Wigan, United Kingdom).Soil organic matter (SOM) content was estimated using the loss-on-ignition method at 500 • C (Davies, 1974).

Data analysis
We chose to parse our data at 11 • C rather than strictly by growing or dormant seasons in order to develop a more functional understanding of the controls on F SOIL .The 11 • C threshold was chosen for multiple reasons.(1) Mean measured soil temperature at 12 cm across our watershed during our three years of observations exceeded 11 • C for the period 6 May to 13 October.This period coincides with the growing season and allows for slight variance with a buffer on either end.(2) Piecewise regression (using the segmented package in R) identifies an estimated break point of 11.58 • C ± 0.47 standard error when the ln(F SOIL ) is regressed against soil temperature.Based on our observations, we opted for the more conservative threshold of 11 • C. (3) Below 11 • C, the F SOIL values are tightly coupled to temperature, while above 11 • C there is increasing variance in F SOIL that we feel warrants exploration.All analyses and means presented are for measurement periods where soil temperatures are above 11 • C, unless otherwise noted.
We employed a mixed-model analysis of variance (ANOVA) with repeated measures to identify main and interactive effects of elevation and vegetation on soil CO 2 efflux, soil temperature, and water-filled pore space using the proc mixed procedure in SAS 9.3 (SAS Institute, North Carolina, USA).All means presented are least-squares means calculated using a Tukey-Kramer adjustment.
To decouple the effects of soil temperature and soil moisture on F SOIL , linear regressions of soil temperature against the natural log of F SOIL were done by year (2010,2011,2012), by vegetation cover type (Open, Canopy, Shrub), by elevation (Low, Mid, High), by year and vegetation (Open 2010, Canopy 2010, etc.), and by year and elevation (Low 2010, Mid 2010, etc.).The residuals from each model were then regressed against WFPS by each combination.All linear regressions use the lm function in R 3.0.1 (R Core Team, 2013).
Differences in soil organic matter (SOM) were examined with a Kruskal-Wallis rank sum test using the kruskal.test in R 3.0.1 (R Core Team, 2013).A two-way mixed-model ANOVA using the proc mixed procedure in SAS 9.3 was used to examine main and interactive effects of elevation, vegetation, and soil depth on soil bulk density and total bulk density.Soil bulk density, soil organic matter, total soil carbon, total soil nitrogen, and the plant area index were individually regressed against the mean plot-level soil CO 2 efflux for each corresponding plot (e.g., High-Canopy 1, High-Open 2).Means were calculated from all flux data above 11 • C for all 3 years (2010-2012).

Results
Exponential regression of F SOIL measurements against soil temperature at 12 cm (T SOIL ; Fig. 2a) shows a positive relationship (R 2 = 0.316; y = 0.829 + e (0.1149x) ) with increases in temperature resulting in increased efflux rates.The amount of variance explained by T SOIL lessens above 11 • C (R 2 = 0.104), with F SOIL measurements below 11 • C showing a much tighter relationship with temperature (R 2 = 0.434).To explore this variance, all data above 11 • C were isolated and examined in order to parse out controls above this apparent temperature threshold for this system.The natural log of flux measurements above 11 • C for all years was regressed against T SOIL (Fig. 2b), showing a significant positive relationship with soil temperature (R 2 = 0.119; y = 0.096x − 0.010).From this linear model, the residuals were then regressed against WFPS.The residuals from the ln(F SOIL ) values above 11 • C show a significant negative relationships with WFPS (Fig. 2c), but this explains only marginally more of the variance (R 2 = 0.019).

Soil CO 2 efflux (F SOIL )
Repeated measures ANOVA analyses show no significant differences in F SOIL among years when data are pooled.Significant differences among years do occur when data are parsed by elevation (F 4, 633 = 3.17 Across all data above 11 • C, there was a significant effect of elevation (F 2, 633 = 3.44; p = 0.032), with plots at High elevation sites showing the highest F SOIL rates and High sites statistically differing from Low sites, with Mid elevation sites not differing from either (Fig. 3a).2010 was the only year to show a statistically significant difference in F SOIL among elevation classes within a year, with Low elevation sites exhibiting significantly lower F SOIL rates (F 2, 633 = 3.17; p = 0.013).
Differences among vegetation classes were stark (F 2, 633 = 37.58; p = < 0.001).Shrub classes across all elevation classes and all years had higher rates of F SOIL (6.07 ± 0.42 µmol CO 2 m −2 s −1 ) than Canopy (4.69 ± 0.42 µmol CO 2 m −2 s −1 ) or Open (4.09 ± 0.42 µmol CO 2 m −2 s −1 ) plots.This Shrub effect was most notable during 2010, the driest year during the study, when Shrub plots showed the highest rates of F SOIL recorded during the study (7.48 ± 0.674).Statistical differences among vegetation classes among years were complex.Shrub 2010 and Open 2011 were uniquely different among all combinations (Fig. 3b).

Water-filled pore space (WFPS)
WFPS tracked well with precipitation across years, with 2010 having the lowest values of WFPS and 2011 having the highest rates of WFPS.WFPS in 2011 was significantly greater than either 2010 or 2012 (F 2, 633 = 17.27; p = < 0.001; Table 2).During 2010, when precipitation was lower than average, an apparent elevation effect on WFPS is observed, with High elevation plots exhibiting significantly lower WFPS measurements than either Low elevation or Mid elevation plots (Fig. 3e).During 2011 and 2012, under extreme and moderate moisture regimes, this elevation effect is not evident.During 2010, vegetation treatment types are not significantly different, but in 2011, when there is more moisture in the system, statistical differences among vegetation classes are apparent, as Shrub and Canopy plots exhibit higher WFPS values than Open plots (Fig. 3f).

Soil temperature (T SOIL )
Data for all years showed a significant effect of elevation on T SOIL across elevation classes for all data above 11 • C (F 2, 633 = 170.76;p = < 0.001).Low elevation sites were warmer (15.99 ±0.35 • C) than Mid sites (14.71 ± 0.35 • C) and High (14.94 ± 0.35 • C) elevation sites.There was no statistical difference in soil temperature by elevation within years (Fig. 3c).
Vegetation (Fig. 3d) had a statistically significant effect on T SOIL (F = 52.79;p = < 0001).Shrub plots were the coolest (14.93 ±0.35 • C), Open plots the warmest (15.62 ± 0.35 • C), and Canopy plots were in between (15.10 ± 0.35 • C).No within-year comparisons were statistically significant.There were also no differences in temperature among years when data were pooled and compared by year alone.
Soil porosity from 0-12 cm ranges from 0.58 to 0.82 m −3 m 3 and is correlated with vegetation cover, with higher values in the Shrub plots (0.77 ± 0.03 m 3 m −3 ), medial values in Canopy plots (0.68 ± 0.03 m 3 m −3 ), and lower values in Open plots (0.65 ± 0.03 m 3 m −3 ; Table E1).Shrub plots also show the highest concentrations of total soil carbon (9.35 %), significantly greater than other vegetation types (F = 9.79; p = 0.0002).Vegetation also influences total soil nitrogen, with Shrub plots exhibiting higher proportions of total soil N than other plots (Table E1; F = 6.36; p = 0.0029).Total soil carbon also differed by elevation, with Low and High classes showing greater proportions of total soil carbon in samples than Mid elevation sites (Table D1; F = 6.28; p = 0.0031).Mid level plots also showed lower proportions of total soil nitrogen than other elevation levels (Table D1) (F = 6.45; p = 0.0027).
Regressions of mean plot-level soil F SOIL ) against soil bulk density, soil organic matter, total soil carbon, total soil nitrogen, and plant area index only yielded a statistically significant relationship between F SOIL ) and soil bulk density (R 2 = 0.302; p = 0.003; Fig. 5).

Discussion
The threshold approach employed in this paper allows for a quantification of the controls on soil CO 2 efflux during periods when fluxes are not temperature limited.This threshold was chosen empirically after analyzing the data.While the exact threshold of 11 • C may not be applicable to all water- sheds, if similar or related methods for threshold determination (e.g., piecewise regression or Bayesian change-point analysis) are used, this approach offers potential for comparisons and insights into controls on fluxes.If varying thresholds are found, it would be of research interest to examine the variance.

Vegetation effects
Significantly greater CO 2 fluxes from plots with shrub cover are apparent in our data, despite consistently lower soil temperatures in these plots.We propose that increases in soil CO 2 efflux from beneath shrubs are related to the observed differences in soils beneath plots with shrub cover compared to our other vegetation plots in this watershed.Soil bulk density, soil porosity, soil carbon, and other soil properties have been shown to drive the spatial variability of carbon fluxes (Jassal et al., 2004;Fiener et al., 2012;Luan et al., 2011).
Here we see shrubs decrease soil bulk density (Fig. 4b; Table E1) and increase soil porosity (soil porosity ( ) for Shrub plots averaged 0.77 m 3 m −3 from 0-12 cm depth, compared to 0.65 m 3 m −3 for Open plots and 0.68 m 3 m −3 for Canopy plots; Table E1), allowing for greater diffusivity within the soil matrix and increased transportation potential of soil CO 2 through the soil.While soils in Shrub plots have higher concentrations of SOM and soil C, soil bulk density is lower, which results in overall lower values of SOM and comparable values of soil C by volume.The increased soil porosity in soils beneath shrub cover likely results in increased oxidation of labile soil C. It should be considered that Shrub plots, to 20 cm soil depth, had the highest mean values of SOM (18.13 %), higher soil C (9.35 %), higher soil N (0.47 %), higher C : N ratios (19.36), and lower ρ s (0.39 g cm −3 ) compared to Canopy (SOM = 12.48 %; soil C = 6.35 %; soil N = 0.37 %; soil C : N = 16.30) and Open plots (SOM = 12.48 %; soil C = 5.14 %; soil N = 0.31; soil C : N = 15.76;Table D).The high C : N ratios for Shrub plots possibly indicate lower amounts of available, labile carbon and lower rates of decomposition than other areas of the watershed.This is corroborated by early results from a 2-year litterbag experiment conducted in this watershed (Atkins et al., 2015).This indicates that root respiration contributions from shrubs may be substantive and may also be influenced by varying soil moisture and precipitation regimes.The effect of the soil microbial community on the temperature sen-sitivity of soil respiration can also be enhanced in soils with high soil C : N ratios (Karhu et al., 2014).

Interactions of vegetation and interannual climate variability
While Shrub plots exhibit greater rates of soil CO 2 fluxes than other classes in this watershed during the course of this study, the magnitude of these fluxes is also influenced by the interannual variability in precipitation.Across the 3 study years, there is evidence of an intrinsic link between the movement of carbon and water in this watershed in response to landscape heterogeneities (i.e., vegetation and elevation) and interannual climate dynamics.During 2010, our comparatively dry year, we see increased rates of F SOIL across the watershed but more pronounced increases in fluxes from Shrub plots.Conversely, during 2011, the relatively wet year, vegetation-level differences in F SOIL are statistically unapparent.When changing precipitation regimes are considered, along with future projections of warming and carbon dynamics, the importance of this coupling among water, carbon, and vegetation within humid watersheds cannot be understated.Changes in the distribution, variability, and amount of rainfall, as a result of climate change, are expected to have a major effect on carbon cycling (Borken et al., 2002).The magnitude of this effect, however, remains uncertain (Wu et al., 2011;Ahlström et al., 2012;Reichstein et al., 2013).

Interactions of interannual climate variability and topography
During 2010 (driest year), we see a strong effect of elevation on water-filled pore space (WFPS).During 2011 and 2012, however, there is no apparent effect of elevation on WFPS.When precipitation decreases across the watershed, as is the case during 2010, a different soil moisture regime manifests itself at higher elevations, with lower values of WFPS that contribute, in the case of this watershed, to increased rates of F SOIL .During periods of increased precipitation, the watershed exhibits a more uniform soil moisture regime.The difference in the magnitude of carbon fluxes across elevation levels decreases during years with higher precipitation.During periods of higher precipitation and increased soil moisture, air space within the soil remains filled and transportation of CO 2 through the soil is limited, resulting in decreased rates of F SOIL .The production of CO 2 in the soil is also decreased due to the increased incidence of anoxic conditions as a function of increased WFPS.Our Low elevation plots were statistically similar in wetness to the Mid plots, both of which were wetter than the High plots during the study.The Low elevation plots were also the warmest for each year of the study, yet exhibited the lowest rates of F SOIL for the entire study period.One consideration not explicitly detailed in our study is the effect of topographic aspect on soil water redistribution as plots in our study all had an east-northeasterly aspect.Landscape positions with varying aspect can have differing soil water contents while having similar soil temperature regimes (Kang et al., 2003) that still result in varied soil carbon fluxes.Another contributor to the magnitude of carbon fluxes can be the amount of upslope accumulated area or the connectivity of varying landscape positions to flow paths within watersheds (McGlynn and Seibert, 2003;Pacific et al., 2011).During our wet year, however, we see a diminished effect of these topographic heterogeneities.Enhanced fluxes during years of decreased precipitation suggest that soil respiration in humid mountain watersheds is strongly controlled by soil water and, to a lesser extent, soil temperature.During average and above-average precipitation years, soil respiration values are lower due to limited CO 2 production and/or diffusion through the soil.During years where precipitation is below average, soil respiration values increase.However, what is not considered here are the cumulative effects of interannual variability in precipitation.Would consecutive dry or consecutive wet years result in increases or decreases following the second year?

Implications of vegetation dynamics
The most dominant shrub species in this watershed is Rhododendron maximum, an ericaceous understory shrub that has been shown to increase SOM and soil N in forests where it is present (Boettcher and Kalisz, 1990;Wurzberger and Hendrick, 2007).R. maximum occurs most commonly in forest coves and on north-facing slopes with mesic to moist soil water regimes (Lipscomb and Nilsen, 1990).Ericaceous litter also contributes to declines in soil fertility, lower N mineralization rates, and lower decomposition rates due to higher concentrations of foliar polyphenols (Hättenschwiler and Vitousek, 2000;DeLuca et al., 2002;Côté et al., 2000;Wurzberger and Hendrick, 2007).Ericaceous plants have ericoid mycorrhizae that provide a competitive advantage to breaking down organic N over ectomycorrhizae associated with many deciduous and coniferous species (Bending and Read, 1997), which leads to the inhibition of overstory species regeneration (Nilsen et al., 2001).
The areal extent of R. maximum has increased in some areas of southern and central Appalachia (Phillips and Murdy, 1985;Rollins et al., 2010;Brantley et al., 2013;Elliott et al., 2014).Shrub cover in the region is expected to continue to increase given fire suppression, lack of grazing, and forest canopy die-off from infestations (Nowacki and Abrams, 2008;Ford et al., 2012).If precipitation increases in this area in accordance with climate projections, the accompanying increase in soil moisture availability may further the expansion of R. maxiumum.The loss of previously dominant foundational species in these systems (e.g., Picea rubens in West Virginia due to logging and fire in the late 1800s and early 1900s; Tsuga canadensis die-off from hemlock woolly adelgid across the Appalachians and eastern USA) may result in possible, multiple stable states (Ellison et al., 2005) Increase in shrub cover has the potential to further impact ecosystem fluxes and biogeochemical cycling and may contribute strongly to future forest community dynamics.However, conversely, if the variance in interannual precipitation continues to increase, drought years may serve as a possible control on shrub expansion.

Implications of dynamic precipitation
Data from the NCDC's station in Canaan Valley, WV (Station ID 461393), show that precipitation in this region of WV is increasing, notably so since 1993 (Fig. 6b).This increase in precipitation appears to be driven by a notable increase in the number of extreme precipitation days (EPDs), defined here as days on which precipitation exceeds 25.4 mm (Fig. 6c).While precipitation is generally increasing in the Weimer Run watershed, and similar areas across West Virginia, the year-to-year variance is increasing as well.A Breusch-Pagan test, which tests for the presence of heteroscedasticity in linear regression models, shows that NCDC precipitation data from Canaan Valley since 1970 exhibit a statistically significant increase in in-terannual variance (BP = 8.58; p = 0.003).This means that the low-precipitation years are trending much lower than the mean, while the high-precipitation years are trending much higher than the mean, with fewer overall "average" precipitation years.This increased variance appears to again be driven by the increased variance in EPDs from year to year (Fig. 6b  and c) and has been attributed to changes in the North Atlantic Subtropical High and anthropogenic climate change (Li et al., 2011).As soils are subject to year-to-year wetdry cycles, cumulative effects on carbon cycling and carbon fluxes are likely.It is beyond the scope of this study to answer the question posed above; however, with the observed dynamics in precipitation for the region, this may be an important line of future research.These relative extremes in rainfall amounts that occurred during this study resulted in significant differences in soil moisture regimes (measured as WFPS) across the entire watershed and between both our elevation and vegetation cover classes (Sect.3.2; Tables 1 and 2).During 2011, there were 34 EPDs, whereas in 2010 there were only 11 and in 2012 only 9. Precipitation also affected the variance in WFPS within the watersheds by year, as measured by the coefficient of varia-

Theoretical contributions
Our findings indicate that for this relatively humid watershed, increased precipitation may result in decreased soil water heterogeneity and decreased fluxes of carbon from the soil surface, while decreased precipitation may result in increased soil water heterogeneity and increased carbon fluxes -especially from areas of higher elevation and/or with greater shrub coverage.This study adds to a growing body of literature that deals theoretically with the effects of topography and vegetation on water and carbon cycling, and more specifically on carbon cycling across watersheds with varying degrees of moisture availability.
Similar studies in drier watersheds have found that increases in soil water availability largely result in increases in soil carbon fluxes.Pacific et al. (2008) showed that for the Stringer Creek watershed, a subalpine, montane watershed in Montana, the spatial variability of soil CO 2 efflux was controlled by the input of soil water driven by seasonal snowmelt.Fluxes at riparian areas lower in the watershed were suppressed at high levels of soil water early in the growing season, but as soil water decreased, fluxes increased.Pacific et al. (2009) further compared a wet and a dry year in the same watershed, finding that cumulative fluxes were 33 % higher in riparian areas during the dry year but 8 % lower at landscape positions higher in the watershed.Decreased moisture inputs for Stringer Creek resulted in significant responses in fluxes across landscape positions, but the riparian areas respond similarly to the entirety of the Weimer Run watershed in our study, with dry years resulting in increases in carbon fluxes.It has been shown in previous studies (Clark and Gilmour, 1985;Davidson et al., 2000;Sjogersten et al., 2006;Pacific et al., 2008) that a production optimality of surface CO 2 efflux exists in response to soil water content such that peak rates of surface CO 2 efflux coincide with medial values of soil water content, with soil water varying both temporally and spatially (with elevation).Our study adds the dimension of vegetation to this model, demonstrating that vegetation heterogeneity can have significant effects on surface CO 2 efflux within humid watersheds, particularly during periods of below-average soil water availability.
There are other possible avenues of carbon loss not considered here that may be affected by interannual climatic variability.It is possible that dissolved organic carbon (DOC) and dissolved inorganic carbon (DIC) fluxes from the watershed are increased during wet years due to increased flow in the system.Fluxes from these pools may be significant, but are difficult to measure and often carry a high degree of uncertainty.DIC and DOC fluxes are highly variable spatially, coinciding with preferential flow paths within watersheds as a function of runoff (McGlynn and McDonnell, 2003;Kindler et al., 2011).Manipulative experiments have shown that simulated drought decreases DOC leaching across an elevation gradient by as much as 80-100 % (Hagedorn and Joos, 2014), indicating that these fluxes are also responsive to interannual climate variability.

Conclusions
We completed a 3-year plot-based study focusing on evaluating the effects of vegetation cover and elevation on soil carbon cycling in response to interannual variability in precipitation.By looking at data above 11 • C for soil temperature measured at 12 cm depth, we were able to focus on the effects of soil moisture on carbon cycling without having to control for temperature limitation.We found that during a relatively dry year (2010; 1042 mm), the magnitude of soil www.biogeosciences.net/12/2975/2015/Biogeosciences, 12, 2975-2994, 2015 carbon flux was enhanced across the watershed, but the increase was differential due to statistically greater fluxes from plots with high shrub coverage.Greater fluxes of carbon from plots with high shrub cover were due in part to decreased soil bulk density, high quantities of soil organic matter, and possible increased root respiration present beneath shrubs as compared to either closed-canopy or open-area plots.For 2011 and 2012, relatively wetter years, fluxes were decreased, and the effects of vegetation cover on the magnitude and variability of fluxes were statistically insignificant.Elevation had an effect on carbon cycling in the system by exacerbating vegetation effects during dry periods through increased effects on soil water distribution in the system.While soil water was correlated with elevation for all of our data, the effect was more pronounced during our driest year (2010) where areas higher in the watershed were much drier than lower positions.With the expected increase in precipitation as forecast by climate models and the empirical basis of increased interannual variance in precipitation, these findings offer important insights into the relations among landscape, vegetation, soil, and the associated biogeochemical effects for complex, humid watersheds.Given the increased likelihood of greater interannual variance in precipitation in the future, the coupling between carbon movement and vegetation cover is potentially quite crucial and under-considered.Further, the role of ericaceous shrubs and their future in this system are quite complex and may have a profound influence on biogeochemical cycles.

Figure 1 .
Figure 1.Above: Weimer Run watershed (374 ha) with elevation levels indicated on map.Below: conceptual diagram showing vegetation classes.Images courtesy of the Integration and Application Network, University of Maryland Center for Environmental Science (http://ian.umces/edu/symbols).

Figure 4 .
Figure 4. (a, c) Means of soil bulk density (g cm −3 ) and soil organic matter (%) by elevation treatment.(b, d) Means of soil bulk density (g cm −3 ) and soil organic matter (%) by vegetation treatment.Bars indicate standard error.Colors indicate soil depth profiles.

J
. W. Atkins et al.: Vegetation and elevation influence the timing and magnitude of soil CO 2 efflux Appendix C

Table 2 .
Statistical table from repeated measures mixed-model ANOVA.For all comparisons by elevation, vegetation, and year, n = 633 and df = 2633.For elevation by year and vegetation by year comparisons, n = 633 and df = 4633.

Table C1 .
Kruskal-Wallis rank sum test results for the effects of elevation, vegetation, and soil depth on soil organic matter (SOM %), soil C (%), soil N (%), and soil C : N (%).