2.1. Site Descriptions
 Continuous measurements of atmospheric and subnivean CO2 concentrations were made during the winter of 2006–2007 in acidic tussock tundra near, Toolik Lake, Alaska (68°37′N, 149°36′W, 740 m asl), and in upland boreal forest on the University of Alaska campus in Anchorage, Alaska (61°12′N, 149°49′W, 42 m asl). Vegetation at the tussock tundra site is nearly equally distributed among graminoids, deciduous shrubs, evergreen shrubs and mosses (Shaver and Chapin, 1991) and the soil contains approximately 10.0 kg SOC/m2 in the upper 20 cm of the profile (Schimel et al., 2006). The overstory of the upland boreal forest site is a near even mixture of Picea glauca and Betula papyrifera. The forest has approximately 6000 stems/ha, which amount to a basal area of 170 m2/ha. The stems are approximately 50% P. glauca and 50% B. papyrifera, although the B. papyrifera individuals are generally much larger, such that the species accounts for approximately 70% of the basal area. Cores taken following 2007 radial growth at 25 cm above the ground surface revealed 65 annual increments in the largest P. glauca and 80 annual increments in the largest B. papyrifera individual, providing a minimum time since the last stand-clearing disturbance. Dominant species in the understory include Ledum palustre, Empetrum nigrum, Vaccinium vitis-idaea, Cornus canadensis, Lycopodium clavatum and the mosses Hylocomium splendens and Pleurozium schreberi. Soils at the site hold approximately 4.5 kg SOC/m2 in the upper 20 cm of the profile.
2.2. Continuous Measurements
 In tussock tundra, micrometeorological sensors and sensors for atmospheric and subnivean [CO2] were mounted on a 3.5 m tower. Power for the system was delivered by a 1 kW wind turbine (Southwest Windpower, Flagstaff, AZ) and a battery bank consisting of 16 200AH/6V absorbent glass mat batteries wired in series (Trojan Battery Co., Santa Fe Springs, CA). At the upland boreal forest site, sensors were mounted on a 10 m meteorological tower and supplied with AC power from the Ecosystems and Biomedical Laboratory on the University of Alaska campus in Anchorage. Micrometeorological stations consisted of sensors for air temperature and relative humidity, wind speed, barometric pressure, snow depth, snow temperatures, soil temperatures and soil water contents, in addition to sensors for atmospheric and subnivean [CO2] (Table 1). Sensors were read every 30 s and hourly averages were logged, with the exception of barometric pressure and soil water sensors, which were read and logged once per hour.
Table 1. Description of Micrometeorological Stations Established in Arctic Tussock Tundra Near Toolik Lake, Alaska, and in Upland Boreal Forest in Anchorage, Alaskaa
|Variable||Tussock Tundra||Upland Boreal Forest|
|Air temperature and RH at 3 m||CS215b||CS500b|
|Wind speed at 3 m||014Ab||014Ab|
|Atmospheric [CO2] at 3 m||GMP343c||GMP343c|
|Snow temperature 0.1 m intervals||Hobo H8d||CS107b|
|Subnivean [CO2] at 0.02 m||GMP343c||GMP343c|
|Soil temperature at 0.05 m||NA||CS107b|
|Soil temperature at 0.10 m||CS107b||NA|
|Soil temperature at 0.25 m||CS107b||CS107b|
|Soil water content at 0.25 m||CS616b||CS615b|
|Data logger||CR3000b||CR10 and CR800b|
|Power supply||WHI-200 wind turbinee||AC line power|
 Atmospheric CO2 probes at both sites were housed in modified radiation shields, while subnivean CO2 probes were suspended 0.02 m above the soil surface using laboratory scaffolding. In tussock tundra, the subnivean CO2 probe was installed in an area visually determined to be microtopographically intermediate between a tussock and an inter-tussock area. In the upland boreal forest, the subnivean CO2 probe was installed approximately 30 cm beyond the canopy of a white spruce tree. At both sites, subnivean CO2 probes were placed approximately 1.0 m from the site where soil temperature and soil water probes were installed in a qualitatively similar microenvironment. CO2 probes were checked against a series of calibration gases before and following the winter measurement period. Linear corrections were applied when there was evidence of sensor drift.
 Measurements of snowpack density were made using snow pits (n = 2 or 3) excavated on monthly intervals at both sites during the winter of 2006–2007. Density measurements were made in continuous 10 cm increments using a stainless-steel RIP 1 density cutter [Elder et al., 1991] (Snowmetrics, Fort Collins, CO) and a 600 g capacity spring scale with 5 g resolution (Pesola AG, Baar, Switzerland). Snowpack density was taken as the average of the 10 cm measurements.
2.3. Point Measurements
 Point measurements of atmospheric and subnivean CO2 concentrations were made in tussock tundra on 5 December 2006, 15 March 2007 and 9 May 2007 and in upland boreal forest on 17 November 2006, 22 December 2006, 6 February 2007, 24 February 2007, 21 March 2007 and 13 April 2007. Measurements were made using a hollow stainless steel probe, equipped with a perforated tip and plumbed with 3.2 mm ID bev-a-line tubing [Fahnestock et al., 1998, 1999; Brooks et al., 1999; Welker et al., 2000; Hubbard et al., 2005]. The tubing was attached to a LI-800 NDIR CO2 analyzer (LI-COR Biosciences, Lincoln, NE), which was equipped with a 5 cm optical bench, a micro-diaphram pump downstream of the optical bench (850 mL/min, KNF Neuberger Inc., Trenton, NJ) and a digital multimeter (Fluke Corporation, Everett, WA). The LI-800 was allowed to warm-up for 1.5 h and checked against a series of calibration gases before each measurement campaign. Measurements were made by inserting the probe to the base of the snowpack and drawing air through the analyzer until CO2 concentrations stabilized.
2.4. Data Processing and Statistical Analyses
 Monthly measurements of snowpack density were interpolated with cubic spline curves to generate daily estimates of snowpack density using the Expand procedure of SAS 9.1 (SAS Institute, Cary, NC). Snow densities typical of cold, fresh snow (arctic: 250 g/L, boreal: 150 g/L) were applied to the first day, while a density value typical of isothermic snow (arctic: 425 g/L, boreal: 375 g/L) was applied to the final day, to enable interpolation to the tails the measurement period at both sites. Different initial and final snow densities were applied to account for differences in wind speeds between the sites.
 Estimates of CO2 efflux from the subnivean environment to the atmosphere were made using Fick's law of diffusion through unsaturated porous media [Mast et al., 1998]:
where q is the rate of CO2 efflux (μmol m−2 s−1), dC is the difference in [CO2] between the subnivean and the atmosphere (μmol/mol), dz is snow depth (cm) and Deff is the effective diffusion constant through the snowpack, estimated as:
where θ is snowpack porosity (estimated as 1-(density/973), with ice density = 973 g/L), τ is snowpack tortuosity (estimated as θ1/3), and DCO2 is the diffusion constant for CO2 through air at STP, corrected for pressure (P) and temperature (T) in the final terms. Estimates of CO2 efflux were restricted the period when soils were consistently covered by at least 15 cm of snow.
 A wide variety of models have been used to describe the correlation between temperature and respiration and investigators have generally concluded that no one model provides the best fit across a full range of soil temperatures [Lloyd and Taylor, 1994; Fang and Moncrieff, 2001]. In the present study, CO2 efflux data were fit with a simple, commonly used exponential model [e.g., Buchmann, 2000; Mikan et al., 2002; Litton et al., 2003]:
where T is soil temperature (°C) and α and β were fit using nonlinear regression in Proc NLIN of SAS 9.1 (SAS Institute, Cary, NC). It is common that respiration-temperature correlations fail to meet the assumptions of regression analysis, in particular the assumption of homogeneity of variance [Lloyd and Taylor, 1994; Fang and Moncrieff, 2001]. Regressions in the present study are not exceptional in this respect and should, therefore, be treated with caution. Q10 values, which describe the increase in CO2 efflux expected for a 10°C increase in temperature, were calculated as follows:
where β was taken from equation (3).
 During the snow-covered season, air and soil temperatures may be strongly decoupled and belowground respiration in forest ecosystems may show a response to both, as trees extend above the snow surface. Multiple linear regression analyses were used in Proc Reg of SAS 9.1 to examine the potential importance of air temperature as a secondary control on winter CO2 efflux. Nonlinear terms were implemented prior to the multiple regression analysis and it was necessary to include intercepts in the models. First, Proc NLIN was used to fit β in equation (3) for both air and soil temperatures, then stepwise selection was employed in multiple regression analyses to fit coefficients and test the significance of two linear (air and soil) and two exponential terms (air and soil, equation (3)). In both tussock tundra and the upland boreal forest, a linear air temperature and an exponential soil temperature term proved significant at alpha = 0.05. The multivariate models were, then, refit in Proc NLIN to update the β values and reexamined using stepwise selection in a multiple regression to generate the final fit statistics. The multivariate models, which use daily mean air and soil temperatures to estimate daily mean CO2 efflux, were used to estimate CO2 efflux on days with relatively high wind speeds, when turbulence-induced pressure pumping [Massman et al., 1997] may have affected diffusion-based estimates of CO2 efflux. This approach was used on 60 days in tussock tundra and on 20 days for the boreal forest site.
2.5. Potential Sources of Error in Estimates of CO2 Efflux
 The continuous estimates of winter CO2 efflux may be influenced by several sources of error. Takagi et al.  discussed the effects of similar CO2 sensors on temperatures in the surrounding snow and concluded that the effects were relatively minor. In their study, the sensors were turned off between measurements while, in the present study, sensors were run continuously to reduce the likelihood of sensor drift. Consequently, there was probably some effect of the sensors on the stratigraphy of surrounding snow. These artifacts were likely greatest when the snowpack was shallow and when snow temperatures were near 0°C.
 A linear gradient in CO2 concentration from the subnivean environment to the atmosphere is required to estimate CO2 flux using the diffusion model (equations (1) and (2)), but there are several phenomena that could disrupt the linear gradient expected under steady state conditions. The presence of relatively impermeable wind slabs and or ice layers within the snowpack may disrupt the linear gradient [Mast et al., 1998; Jones et al., 1999]. Turbulence-induced pressure pumping during high wind events may also disrupt the CO2 concentration gradient and lead to pulses of CO2 efflux that are not addressed by the diffusion model [Massman et al., 1997; Jones et al., 1999]. Data were eliminated from the arctic site on days when wind speeds were greater than 3.5 m/s and from the boreal site when wind speeds exceeded 0.5 m/s, after visually analyzing the graph of subnivean [CO2] as a function of wind speed. Failure to eliminate data from all days when pressure pumping was active would lead to a systematic underestimation of total winter C efflux.
 The model described in equations (1) and (2) describes diffusion of CO2 from the subnivean environment to the atmosphere directly above the snow surface. Atmospheric CO2 sensors were installed at 3 m height, where CO2 concentrations may be lower than those at the snow surface during periods of limited atmospheric mixing. During these time periods, CO2 efflux was likely overestimated. This error was probably greatest during the nighttime at the boreal forest site and was probably less important at the arctic site.