Global Biogeochemical Cycles

Winter carbon dioxide effluxes from Arctic ecosystems: An overview and comparison of methodologies


  • Mats P. Björkman,

    1. Department of Plant and Environmental Sciences, University of Gothenburg, Gothenburg, Sweden
    2. Now at Norwegian Polar Institute, Polar Environmental Centre, Tromsø, Norway.
    Search for more papers by this author
  • Elke Morgner,

    1. Department of Arctic and Marine Biology, Faculty of Biosciences, Fisheries and Economics, University of Tromsø, Tromsø, Norway
    2. Department of Arctic Biology, University Centre in Svalbard, Longyearbyen, Norway
    Search for more papers by this author
  • Elisabeth J. Cooper,

    1. Department of Arctic and Marine Biology, Faculty of Biosciences, Fisheries and Economics, University of Tromsø, Tromsø, Norway
    Search for more papers by this author
  • Bo Elberling,

    1. Departments of Arctic Biology, Arctic Geology, and Arctic Technology, University Centre in Svalbard, Longyearbyen, Norway
    2. Department of Geography and Geology, University of Copenhagen, Øster, Copenhagen, Denmark
    Search for more papers by this author
  • Leif Klemedtsson,

    1. Department of Plant and Environmental Sciences, University of Gothenburg, Gothenburg, Sweden
    Search for more papers by this author
  • Robert G. Björk

    1. Department of Plant and Environmental Sciences, University of Gothenburg, Gothenburg, Sweden
    2. School of Science and Technology, Örebro University, Örebro, Sweden
    Search for more papers by this author


[1] The winter CO2 efflux from subnivean environments is an important component of annual C budgets in Arctic ecosystems and consequently makes prediction and estimations of winter processes as well as incorporations of these processes into existing models important. Several methods have been used for estimating winter CO2 effluxes involving different assumptions about the snowpack, all aiming to quantify CO2 production. Here, four different methods are compared and discussed: (1) measurements with a chamber on the snow surface, Fsnow, (2) chamber measurements directly on the soil, Fsoil, after snow removal, (3) diffusion measurements, F2-point, within the snowpack, and (4) a trace gas technique, FSF6, with multiple gas sampling within the snowpack. According to measurements collected from shallow and deep snow cover in High Arctic Svalbard and subarctic Sweden during the winter of 2007–2008, the four methods differ by up to two orders of magnitude in their estimates of total winter emissions. The highest mean winter CO2 effluxes, 7.7–216.8 mg CO2 m−2 h−1, were observed using Fsoil and the lowest values, 0.8–12.6 mg CO2 m−2 h−1, using FSF6. The Fsnow and F2-point methods were both within the lower range, 2.1–15.1 and 6.8–11.2 mg CO2 m−2 h−1, respectively. These differences result not only from using contrasting methods but also from the differences in the assumptions within the methods when quantifying CO2 production and effluxes to the atmosphere. Because snow can act as a barrier to CO2, Fsoil is assumed to measure soil production, whereas FSF6, Fsnow, and F2-point are considered better approaches for quantifying exchange processes between the soil, snow, and the atmosphere. This study indicates that estimates of winter CO2 emissions may vary more as a result of the method used than as a result of the actual variation in soil CO2 production or release. This is a major concern, especially when CO2 efflux data are used in climate models or in carbon budget calculations, thus highlighting the need for further development and validation of accurate and appropriate techniques.

1. Introduction

[2] Since the late 1960s, it has been known that snow can act as a barrier to carbon dioxide (CO2) from the soil as it moves to the atmosphere; consequently, CO2 concentrations in soil and snow can increase [Kelley et al., 1968]. This finding and subsequent research [e.g., Sommerfeld et al., 1993; Zimov et al., 1993; Fahnestock et al., 1998; Welker et al., 2000; Grogan and Jonasson, 2006; Elberling, 2007; Koch et al., 2008] up to the present day [e.g., Liptzin et al., 2009; Björkman et al., 2010; Morgner et al., 2010] have changed our understanding of winter from it being considered a dormant season to one that can have a great impact on annual global carbon budgets [e.g., Zimov et al., 1993]. Recently, the role of winter snow in high-latitude ecosystems and the possible change in snow abundance due to a changing climate [Arctic Climate Impact Assessment (ACIA), 2005] have been included in the global climate agenda [Intergovernmental Panel on Climate Change, 2007]. This has further been highlighted by recent work on expected climate feedback mechanisms, where higher CO2 production due to an increasing snow depth has been found [e.g., Welker et al., 2000; Nobrega and Grogan, 2007]. The basic understanding of this feedback loop is that a cover of snow insulates the soil environment from cold air temperatures [Brooks et al., 1997] and, thereby, markedly extends the period when soil and water are not frozen. These physical parameters allow soil microbial activity to occur during winter, with CO2 production as a result [Brooks et al., 1997].

[3] A wide range of methods has been used to evaluate winter CO2 effluxes, and previous comparisons of techniques have claimed that different methods are preferable [McDowell et al., 2000; Schindlbacher et al., 2007]. Winter measurements of CO2 effluxes have commonly been conducted using CO2 respiration chamber measurements in snow pits [e.g., Elberling and Brandt, 2003] or as floating chambers on top of the snow cover [e.g., Winston et al., 1995]. Different approaches of snowpack gradient measurements have also been conducted, with air sampling within and above the snow [e.g., Sommerfeld et al., 1993; Schindlbacher et al., 2007; Sullivan et al., 2008; Liptzin et al., 2009; Björkman et al., 2010]. In addition, CO2 traps installed at the soil-snow interface [e.g., Grogan and Chapin, 1999; Welker et al., 2000] have been used, and eddy correlation has been performed [e.g., Lloyd, 2001]. However, the base trap method lacks temporal resolution and only provides information on mean winter CO2 production for a period of months. Eddy correlation, in turn, gives wider ecosystem-based results, but a homogenous and flat terrain is essential for eddy covariance measurements. Furthermore, this method lacks the resolution needed when dealing with small-scale experimental designs. These last two methods were, therefore, not evaluated in this study.

[4] The aim of this study was to compare four different techniques for evaluating soil CO2 production and effluxes during the winter in high-latitude ecosystems and to illustrate the range of values thus obtained for winter CO2 efflux from the soil and through the snowpack. Measurements were collected at two Arctic locations, Adventdalen, Svalbard and Latnjajaure, Sweden, from shallow and deep snow regimes.

2. Method

2.1. Site

[5] The two locations used in this study are both located within the Arctic region defined by ACIA [2005]. The subarctic Latnjajaure site is situated in a U-shaped valley, 15 km west of Abisko, Sweden (68°20′N, 18°30′E, 980 m above sea level (asl)). The other site, in Adventdalen, Svalbard (78°10′N, 16°06′E, 40 m asl), is a High Arctic valley with underlying continuous permafrost. Both study locations have similar vegetation and soil conditions described in detail by Björk et al. [2007] and Morgner et al. [2010].

[6] In this study, two different snow regimes were examined at each location: shallow and deep snow cover. Within each snow regime and location there were six replicates, giving a total of 24 study plots. The shallow snow plots in Latnjajaure and Adventdalen had a snow cover ranging from 10 to 50 cm. In Adventdalen, deep snow cover was created using six snow fences (600 × 150 cm), resulting in a snow depth of 90–150 cm. In Latnjajaure, two naturally occurring snow beds of 110–180 and 170–320 cm were used, with three replicates in each. Furthermore, the measurements included two different vegetation types, heath and meadow. Earlier studies found no difference in winter effluxes between these vegetation types [Björkman et al., 2010; Morgner et al., 2010], and therefore, no distinction was made between them in this study.

2.2. Efflux Measurements

[7] Four commonly used methods for measuring CO2 efflux in winter were compared (Figure 1). Two methods are based on direct measurements from closed chambers and using portable infrared gas analyzers (IRGA). A LI-COR 6400-09/6262 Soil CO2 Flux Chamber (LI-COR, Lincoln, Nebraska) was used in Adventdalen (as described by Elberling [2007]), whereas an IRGA based on the PP System SBA-4 OEM CO2 Analyser (PP System, Hertfordshire, United Kingdom) was used at Latnjajaure (following an approach similar to that of Björkman et al. [2010]). Both systems included chambers with a diameter of 10 cm and were equipped to cope with the low temperatures expected during field sampling campaigns. The two other methods are based on air sampling in the field and subsequent gas chromatographic analysis (Varian 3800, Varian Inc., Palo Alto, California, United States) [Klemedtsson et al., 1997] and calculations based on Fick's first law of diffusion [Sommerfeld et al., 1993].

Figure 1.

Overview of different methodologies used in this study. (a) Direct measurements with an infrared gas analyzer (IRGA) and floating chamber, Fsnow, or (b) from snow pits, Fsoil. (c) Air sampling at two points, F2-point, or (d) in time series to follow a trace gas through the snow, FSF6.

[8] Effluxes of CO2 from the snow surface (Fsnow) were measured using a floating chamber technique and IRGA [Winston et al., 1995]. The chamber was placed on top of an undisturbed snowpack to allow CO2 concentration to increase within the chamber as a result of effluxes through the snow. Effluxes of CO2 from snow pits (Fsoil) were obtained by digging down to the bottom of the snowpack and fitting the IRGA-connected chamber to collars, inserted into soil before snow cover built up [Elberling and Brandt, 2003]. Measurements were taken starting 20–45 min after snow removal to allow ventilation [Grogan and Jonasson, 2006]. This is to let bursts and chimney effects [McDowell et al., 2000] associated with trapped CO2 within the soil and snow to equilibrate. After measurements were taken, snow pits at Latnjajaure were manually refilled, whereas Adventdalen snow pits typically were refilled by wind. The data used here for Adventdalen have been reported in the work of Morgner et al. [2010]. The flux of CO2 through one-dimensional diffusion, F2-point, was calculated on the basis of Fick's first law, following Sommerfeld et al. [1993]; the porosity and tortuosity were calculated according to Albert and Shultz [2002] and Prieur Du Plessis and Masliyah [1991], respectively; and the CO2 diffusion constant, 0.1381 cm2 s−1 [Massman, 1998], was corrected to snowpack temperature and pressure [Massman, 1998]. Effluxes were then calculated using concentration of CO2 at the soil-snow interface and in the ambient air. Samples were collected in gas-proof syringes from the ambient air and through 1.6 mm thick stainless steel tubing attached to an avalanche probe, which was inserted through the snow to the snow-soil interface. Five samples were taken at each level and syringes were also checked before use to ensure that they remained gas-proof at low temperatures. Samples were transferred to headspace bottles for later analysis with gas chromatography. Basic snow measurements (density, temperature, depth, and profile) were taken in snow pits nearby to avoid disturbing the snowpack used in further measurements (data are presented to some extent by Björkman et al. [2010] and Morgner et al. [2010]). Fluxes of CO2 through diffusion (FSF6) were also evaluated using a trace gas, SF6, to examine the actual diffusion rate through snow [Björkman et al., 2010]. Air samples were withdrawn every 7.5 cm (shallow sites) or 25 cm (deep sites) within the snowpack and in ambient air using the same approach as in F2-point, but with more air inlets on the avalanche probe. The air samples were collected so that they formed a time series following the release of the trace gas at the soil-snow interface, in order to follow the temporal transport of SF6 within the snowpack. The diffusion rate given by the trace gas was then used to estimate the diffusion coefficients, which subsequently were used in Fick's first law to quantify CO2 diffusion across the snowpack. Data, diffusion coefficients used, and further details about the method are reported in Björkman et al. [2010]. Total winter emissions of CO2 were calculated based on interpolation of the data to cover the periods between measurements. Winter was considered to have started when a continuous snow cover had been established, and the end of winter was determined by an abrupt increase of soil temperature, indicating thawing of the soil and the snow cover.

2.3. Statistical Analysis

[9] The methodologies were evaluated on the basis of total winter mean effluxes and as two specific periods, one for each site (28 November and 3 December 2007 for Adventdalen and 19–21 April 2008 for Latnjajaure, the two measurements with the most complete data set for all methods evaluated). The data were analyzed using a one-way analysis of variance (ANOVA), with methodology as factor; this was followed by Tukey's HSD post hoc test. Data were log-transformed to achieve a normal distribution, and a constant was added to eliminate skewness and ensure variance homogeneity (for further details, see Økland et al. [2001]). The untransformed CO2 efflux measured by each method was also investigated in relation to snow depth using curve estimation regression models, comparing 11 different models. Statistical analyses were conducted using SPSS 14.0 (SPSS Inc., Chicago, Illinois).

3. Results

[10] Mean winter soil effluxes of CO2 (mg CO2 m−2 h−1) using the Fsoil method were 3–75 times higher (p < 0.001; n = 101 and 112, respectively) than any other method for deep snow regimes at both Adventdalen and Latnjajaure (Figure 2). Furthermore, Fsnow was 18 times higher than FSF6 at the deep snow sites in Adventdalen; this difference was significant (p = 0.013). For the shallow snow regime, there were also significant differences (p = 0.034; n = 85) between methods at Adventdalen, although no difference was found at the shallow sites in Latnjajaure (n = 37). At the Adventdalen shallow sites, the Fsoil effluxes were 9 times higher than those for Fsnow; this difference was significant (p = 0.025). Fsoil effluxes were also higher, by a factor of seven, than the FSF6 values, although not significantly so (p = 0.090).

Figure 2.

Mean winter CO2 efflux measured using the different methodologies in deep and shallow snowpacks at (a) Adventdalen, Svalbard, and (b) Latnjajaure, Sweden. Error bars represent standard errors of the means. Mean values are significantly (p < 0.05) different among methodologies if letters (lowercase letters for deep snow and capital letters for shallow snow) differ within the same snow regime.

[11] In December 2007, the CO2 efflux at Adventdalen showed more or less the same significant differences (p = 0.001; n = 24) between methodologies in the deep snow as for the mean winter soil effluxes. The Fsoil method showed significantly higher efflux than F2-point and FSF6: 4 times (p = 0.013) and 32 times (p = 0.001), respectively. Furthermore, the Fsoil efflux was about 2.5 times higher than the Fsnow rate (Figure 3); however, it was not statistically different. In contrast, the shallow snow sites at Adventdalen (n = 17) did not show any significant difference between methods during this period (Figure 3). In Latnjajaure, the April 2008 measurements showed the same differences as the mean winter efflux, with a significantly higher (p = 0.008; n = 23) efflux from the Fsoil than from all the other methodologies under deep snow. Again, there were no significant differences found under shallow snow conditions (n = 17; Figure 3). The significantly higher effluxes given by the Fsoil method can also be seen for the total winter emissions of CO2 given by the different methods (Table 1).

Figure 3.

CO2 efflux measured during individual periods (with the most complete data set) using the different methodologies in deep and shallow snowpacks at (a) Adventdalen, Svalbard (28 November and 3 December 2007), and (b) Latnjajaure, Sweden (19–21 April 2008). Error bars represent standard errors of the means. Mean values are significantly (p < 0.05) different among methodologies if letters (lowercase letters for deep snow and capital letters for shallow snow) differ within the same snow regime.

Table 1. Mean (±SE) Total Winter Emissions of CO2 Estimated by the Four Different Methodsa
 Release From Snow Surface Fsnow (g CO2 m−2 yr−1)Release From Soil Surface Fsoil (g CO2 m−2 yr−1)Diffusive Flux Through Snow (g CO2 m−2 yr−1)
  • a

    The winter period for Adventdalen deep and shallow snowpacks was 2 October 2007 to 29 May 2008 (241 days). For Latnjajure, deep and shallow snowpacks were examined in the periods 8 November 2007 to 1 July 2008 (237 days) and 8 November 2007 to 30 April 2008 (174 days), respectively.

   Deep111 ± 58231 ± 4953 ± 186 ± 3
   Shallow12 ± 48128 ± 1947 ± 4214 ± 11
   Deep13 ± 211365 ± 54679 ± 16124 ± 37
   Shallow2 ± 26 ± 641 ± 4116 ± 16

[12] Furthermore, Fsoil was the only method that showed a significant CO2 efflux response to changes in snow depth in all tested statistical models (Figure 4). The models showing the best fit were the compound, logistic, growth, and exponential regression models, which all gave the same degree of correlation (p < 0.001; r2 = 0.484). Although no statistical influence of snow depth on the CO2 effluxes was found for Fsnow and F2-point, both these methods showed a pattern suggesting stabilization of the CO2 effluxes with increased snow depth (Figure 4). The FSF6 method showed stable values with similar effluxes in all snow regimes. The concentration of CO2 at the soil-snow interface showed a significant positive linear relationship (p < 0.001; r2 = 0.367) with increasing snow depth (Figure 5).

Figure 4.

Relationship between snow depth and CO2 efflux using different methodologies at Adventdalen, Svalbard, and Latnjajaure, Sweden. (a) Air sampling in time series to follow a trace gas through the snow, FSF6, or (b) at two points, F2-point. (c) Direct measurements with an IRGA and floating chamber, Fsnow, or (d) from snow pits, Fsoil. Note the deletion of two outliers in Figure 4a and the scale differences on the y axis in Figure 4d.

Figure 5.

Relationship between snow depth and CO2 concentration at the soil-snow interface at Adventdalen, Svalbard, and Latnjajaure, Sweden.

4. Discussion

[13] The most striking result from this study is that depending on the method used, the estimated CO2 effluxes can differ by up to two orders of magnitude (Table 1 and Figures 2 and 3). Data from the literature (Table 2) for other Arctic sites also indicate that estimates of winter CO2 emissions might vary more as a result of the method used than as a result of actual variation in soil CO2 production or release. This is a major concern, especially when CO2 efflux data are used in climate models and in carbon budget calculations.

Table 2. Mean Total Winter Emissions of CO2 From Arctic Tundra According to Method and Regiona
Winter Efflux (g CO2 m−2 yr−1)MethodologyHabitat TypeLocationLatitudeLongitudeSource
  • a

    Values presented here are given as g CO2 m−2 yr−1 and have been recalculated from the original publication when necessary. Values given in parentheses indicate measurements in experimentally increased snow depths (snow fence).

  • b

    Estimated from figure.

  • c

    Measured from mid-March to snowmelt.

88F2-pointTussock tundraUnited States, Alaska, Toolik68°37′N149°36′WSullivan et al. [2008]
4.5 (8.6)–36 (48)F2-pointDry/moist tundra (fence)United States, Alaska, Toolik68°38′N149°38′WWelker et al. [2000]
5–40F2-pointDry to wet tundraUnited States, Alaska68°38′–70°23′N148°45′–149°38′WFahnestock et al. [1998]
2.0–96.7F2-pointDry to wet tundra and natural snow driftsUnited States, Alaska68°38′–70°23′N148°45′–149°38′WFahnestock et al. [1999]
110–733F2-pointForest tundraRussia, Cherskiy69°N161°EZimov et al. [1993]
304–344F2-pointForest tundraRussia, Cherskiy69°N161°EZimov et al. [1996]
11–235FSF6Heath and meadow tundraSweden, Latnjajaure68°20′N18°30′EBjörkman et al. [2010]
6 (4)–12 (5)FSF6Heath and meadow tundra (fence)Svalbard, Adventdalen78°10′N16°04′EBjörkman et al. [2010]
69.6–251FsnowWet and tussock tundraUnited States, Alaska68°38′–70°23′N148°45′–149°38′WOechel et al. [1997]
∼255–770bFsoilHeath tundraSweden, Abisko68°20′N18°50′EGrogan and Jonasson [2006]
245 (249)cFsoilHeath tundra (fence)Sweden, Abisko68°20′N18°50′ELarsen et al. [2007]
76–194FsoilProstrate-shrub tundraSvalbard, Endalen78°N19°EElberling [2007]
154 (198)–147 (271)FsoilHeath and meadow tundra (fence)Svalbard, Adventdalen78°10′N16°04′EMorgner et al. [2010]
99 (158)Ftrap (soda lime)Mesic birch hummock (fence)Canada, Northwest Territories64°50′N111°38′WNobrega and Grogan [2007]
440–700Ftrap (soda lime)Tussock tundraUnited States, Alaska68°37′–69°25′N149°35′–148°24′WGrogan and Chapin [1999]
93–324Ftrap (soda lime)Mesic and wet tundraSvalbard, Adventdalen78°10′N16°06′ESjögersten et al. [2008]
0.7(1.4)–1.1(1.3)Ftrap (NaOH)Dry/moist tundra (fence)United States, Alaska, Toolik68°38′N149v38′WWelker et al. [2000]
24.2–43.6Ftrap (NaOH)Dry/mesic/wet tundraCanada, Alexandra Fiord78°54′N75°55′WWelker et al. [2004]

[14] The first CO2 gradient measurements (similar to F2-point) by Sommerfeld et al. [1993] assumed the snowpack to be a homogeneous medium with a porosity based on average snow density. In contrast to the continental climate, as in the work by Sommerfeld et al. [1993], studies conducted in coastal regions, as our sites, exhibit winter melt events giving rise to a very heterogeneous snow cover, both temporally and spatially. For instance, ice lenses or even thick ice layers frequently develop within the snowpack during winter rain or melt events. Due to thawing events on 31 December 2007 to 1 January 2008, when the temperature rose from −12.7°C to +2.8°C during extensive rainfall, an extensive basal ice layer was formed in Adventdalen (discussed in detailed by Morgner et al. [2010]). A representative Latnjajaure snow profile is presented by Björkman et al. [2010]. Thick ice layers are considered very important in ecosystem stability, as they may act as an impermeable layer for element transfer [van Bochove et al., 2001] and also as a barrier that prevents reindeer from feeding [Turunen et al., 2009]. Such layers limit diffusive and convective transport through the snow and, therefore, lead to a higher CO2 concentration below them [van Bochove et al., 2001; Fortin et al., 2007]. When using the F2-point method, such layers are expected to cause an overestimate of both soil CO2 production and the nonsteady state CO2 efflux to the atmosphere [Schindlbacher et al., 2007]. An improvement of the Sommerfeld et al. [1993] method to account for some of these problems was made by incrementally sampling the CO2 concentration profile together with determining ice lenses within the snowpack [Schindlbacher et al., 2007]. However, these improvements do not include all ice lenses and snow structure properties that might affect the variables affecting the diffusion pathway within the snow, such as porosity and tortuosity. Recently, this snowpack gradient technique was automated for continuous measurements of trace gases [Filippa et al., 2009; Helmig et al., 2009; Liptzin et al., 2009; Seok et al., 2009]. Because of high sampling intensities, this seems to overcome some of the problems with ice lenses and layered snow. However, these methods have all been shown to be sensitive to ventilation, for example, pressure pumping by wind and changes in atmospheric pressure [Albert and Hardy, 1995; Massman et al., 1995] or natural convection as described by Sturm and Johnson [1991]. These sensitivities might decrease the calculated fluxes by ∼30% during moderate wind conditions and by as much as 87% during very windy conditions compared with calm conditions [Seok et al., 2009] and would also decrease the measured emissions when using Fsnow and Fsoil due to reduced concentrations within the snowpack. However, without site-specific measurements of wind and air pressure, the importance of ventilation events cannot be quantified for these methods. The FSF6 method is a further improvement of the diffusion methods, by following the diffusion of a trace gas through the snowpack. This gives the advantage of tracking ventilation events and accounts for the complexity of the snowpack and its physical properties, such as the porosity and tortuosity [Björkman et al., 2010]. However, one of the drawbacks to this technique, along with the other diffusion methods, is the difficulty in handling shallow snow cover, where the concentration gradient is influenced by the short diffusion pathway. This difficulty may be overcome with future improvements, such as those suggested by Björkman et al. [2010].

[15] The Fsnow approach is often used to minimize snowpack disturbance [McDowell et al., 2000] and has two advantages: it is simple to use in field and data are provided directly. However, according to Schindlbacher et al. [2007], this technique may underestimate emissions by 38% because of lateral diffusion within the snow. Further, the Fsnow method might underestimate the soil CO2 production since some of the CO2 being produced will accumulate below ice lenses. In the present study, the Fsnow method experienced problems with efflux detection because of small effluxes and variations in flux directions that could, to some extent, be due to ventilation events and/or lateral diffusion. Thus, this method seems to work best in homogeneous, not wind-packed, snow in sheltered environments.

[16] It is important to keep in mind that the four methods examined in this study measure the CO2 efflux in different ways. The F2-point, FSF6, and Fsnow all produce data on emissions through the snow to the atmosphere, whereas the Fsoil method involves measurements taken directly at the soil surface, as shown in Figure 1. All of these methods have been used to estimate winter CO2 production [e.g., Sommerfeld et al., 1996; Winston et al., 1997; Elberling, 2007; Björkman et al., 2010]. Thus, the Fsoil method is assumed to provide a direct measure of soil CO2 production and uses the same soil collars year-round, including the growing season. This is believed to ensure comparable efflux measurements throughout the year [e.g., Elberling, 2007]. However, the Fsoil approach has been criticized for the chimney effect created when digging the snow pit [McDowell et al., 2000]. Usually, this effect decreases with time after snow removal, and measurements are obtained after a certain period of ventilation, when a steady state flux is assumed to have established [Grogan and Jonasson, 2006]. However, according to Morgner et al. [2010], it can be shown that when ice layers in the snowpack have to be broken to gain access to collars, high and nonsteady state effluxes may occur for several hours; thus, ice layers reduce vertical flow and enhance lateral flow into the open area of the snow pit, strengthening the chimney effect. Measurements in a 2.5 m deep snow bed in Latnjajaure (M. P. Björkman, unpublished data, 2008) showed that ventilation increased for the first 6 h (from 630 to 830 ppm), followed by a rapid decrease, reaching a 20% lower level (510 ppm) than initial emissions after 7 h. Together with the work of Morgner et al. [2010], these data suggest that for this method, ventilation time and perhaps also snow pit size are key factors that potentially affect the CO2 efflux rate when the snowpack has extended thickness or ice layers. CO2 concentration within the soil is enhanced by increased snow depth [Maljanen et al., 2007], and concentration also needs to reach steady state conditions prior to measurements. Moreover, any rearrangements of the snowpack, such as refilling snow pits manually or by wind, may alter its properties and lead to higher soil CO2 concentrations, resulting in even larger chimney effects. Further, the cooling effect of exposing the soil to cold air can lower the actual soil respiration. In this study, near-surface soil temperatures (measured at −2 cm on an hourly basis) in snow pits were not affected by such influences (data not shown), compared with ambient snow conditions in Adventdalen. However, the CO2 effluxes measured using the Fsoil method were still higher than for other methods. It is worth noting that effluxes from this winter study with the Fsoil method are in line with extrapolated values from the summer season [Morgner et al., 2010], based on Q10 values and observations [Elberling, 2007]. By removing the snow or measuring CO2 emission at the soil-snow interface with base traps (Ftrap) (see Table 2) [Grogan and Chapin, 1999; Welker et al., 2000; Nobrega and Grogan, 2007; Sjögersten et al., 2008], an alteration of the physical parameters of the soil-snow system occurs. A shortcut out of the system (Fsoil) or an optimal sink (Ftrap) could be created, resulting in overestimating the actual CO2 production rates. However, for the Ftrap approach, a solution that consumes CO2 at the same rate as it is produced needs to be used if an overestimation is to be avoided.

[17] All methods included in this study are manual, and because of the harsh conditions during the Arctic winter, sampling intensity is restricted and difficult, particularly during windy conditions. Manual sampling can therefore bias the results, leading to underestimates of winter CO2 efflux from the Arctic, as ventilation events are underrepresented. Therefore, to fully understand the soil C dynamics during the Arctic winter, improved automatic systems for continuous measurements of trace gases are needed.

[18] The different methodologies in this study gave measurements of efflux that responded differently to the increased snow depth (Figure 4). The Fsnow and F2-point methods showed a stabilizing effect with an increased snow depth, whereas Fsoil generated exponentially higher emission rates with increasing snow depth. The latter may be related to factors other than soil CO2 production, for instance, the diffusion from the surrounding soil and snowpack, as suggested by McDowell et al. [2000]. For the trace gas technique, the snow cover thickness did not seem to affect the CO2 effluxes and generated more or less constant emission for all snow depths. However, using FSF6, two outlier results were found that had relatively higher divergence than the other methods. The ability of the trace gas method to track ventilation events [Björkman et al., 2010] may explain these outliers. The stabilizing response of F2-point, which also is illustrated by Fahnestock et al. [1999], can be partly explained by the higher CO2 concentration at the soil-snow interface (Figure 5) under deeper snow and the high concentration gradient between the bottom and top of deeper snow beds, which will be less sensitive to moderate changes in concentrations. Furthermore, ventilation of the snowpack is more likely to affect the bottom concentration in shallow snowpacks than in deeper ones [Albert and Hardy, 1995; Albert and Shultz, 2002]. The hoar layers often found close to the soil surface in shallow snow covers are expected to experience a horizontal “pipe” flow during windy conditions [Colbeck, 1997] that might alter the CO2 gradient. Both these mechanisms make flux estimations with shallow snow cover difficult. Even though the trace gas technique was developed for tracking ventilation events, it has the same difficulties as F2-point with shallow snow cover since both techniques use concentration differences in their calculations. This concentration difference can be very small and sensitive to sampling errors in shallow snow.

[19] As in earlier studies [e.g., Welker et al., 2000], our data (Figure 5) show a positive relationship between snow depth and CO2 concentration at the soil-snow interface. However, the low r2 value indicates high uncertainty in the linear response, suggesting that this might result from a combination of factors apart from the impact of ice lenses and layers discussed earlier in this paper. When the snow depth increases, soil temperature increases because of the insulating capacity of the snow [Brooks and Williams, 1999]. Soil microbial activity continues therefore throughout winter with the microbial biomass peaking in late winter and with fungi becoming a more abundant component of the microbial population during winter [Schadt et al., 2003; Björk et al., 2008]. Other physical properties in the soil also influence CO2 production, e.g., water availability. This has recently been shown to be an important driver of respiration in frozen forest soils [Öquist et al., 2009]. Snow thickness and ice layers within the snow have effects on the equilibrium between soil CO2 production and CO2 released into the atmosphere and may also influence the longer-term soil CO2 production since higher CO2 concentration will limit production within the soil [Luo and Zhou, 2006]. Increased snow depth also prolongs CO2 residence time within the snow, resulting in a higher soil-snow interface concentration. Despite the fact that no clear conclusions can be drawn regarding which processes are key to the increased CO2 concentrations at the soil-snow interface, this study highlights the need for improving the accuracy of winter annual CO2 budget estimates.

5. Conclusion

[20] The importance of snow cover in Arctic ecosystems and its influence on annual CO2 budgets has been highlighted during the last decade. However, the complexity of snow and Arctic conditions and the variety of analytical methods that have been used render current efflux estimations questionable. The four methods examined in this study resulted in emission estimates varying by a factor of up to two orders of magnitude at a single sampling site. About 20 published studies have focused on winter measurements in the Arctic, and most of them have used the F2-point and the Fsoil methods, which clearly give different estimates of CO2 effluxes. This study highlights the fact that the different methods are not equally suitable for quantifying the effect of snow thickness and ice layers on soil CO2 production and corresponding release of CO2 from the snow surface. Thus, to gain a better understanding of CO2 release to the atmosphere and CO2 production within the soil, more accurate and appropriate winter measurements are urgently needed. This can be achieved by more coordinated development and validation of techniques and sampling protocols.


[21] We thank Ditte E. Strebel, Allison Bailey, and Louis Delmas for assisting in the field in Adventdalen. We would also thank Sofie Sjögersten for providing data for calculations in Table 2. The manuscript has been modified based on the constructive comments of two anonymous reviewers. We are grateful for funding from Tellus (The Centre of Earth Systems Science, University of Gothenburg; to R.G.B.).