Corresponding author: J. Karlsson, Climate Impacts Research Centre (CIRC), Department of Ecology and Environmental Science, Umeå University, Box 62, 981 07 Abisko, Sweden. (Jan.Karlsson@emg.umu.se)
 The winter period is seldom included in the estimates of aquatic-atmospheric carbon exchange. In this study we quantified the flux of carbon dioxide (CO2) and methane (CH4) over 3 years from 12 small subarctic lakes. The lakes accumulated consistent and high amounts of CO2 and CH4 (56–97% as CO2) over the winter, resulting in a high flux during ice thaw. The CO2 flux during ice thaw increased with increasing mean depth of the lakes, while the CH4 flux was high in lakes surrounded by mires. The ice thaw period was quantitatively important to the annual gas balances of the lakes. For nine of the lakes, 11 to 55% of the annual flux occurred during thaw. For three of the lakes with an apparent net annual CO2 uptake, including the thaw period reversed the balance from sink to source. Our results suggest that the ice thaw period is critically important for the emissions of CO2 and CH4 in small lakes.
 Lakes are important components of the global carbon cycle by releasing CO2 and CH4 to the atmosphere [Battin et al., 2009]. The emission may result from both internal degradation of organic matter and input of supersaturated water from land [Karlsson et al., 2007; McCallister and del Giorgio, 2008; Stets et al., 2009]. The emission are believed to be partly controlled by climate-related factors [Jansson et al., 2008; Tranvik et al., 2009], acting on both loading of carbon from land to water [Curtis, 1998] and on respiration rates and net ecosystem production in the aquatic systems [Ask et al., 2012; Gudasz et al., 2010].
 Assessing the importance of lakes for land-atmospheric carbon exchange in the landscape requires a solid knowledge of the temporal and spatial variability in CO2 and CH4 fluxes, which is not present to date. In boreal, arctic, and many mountainous regions, lakes are covered by ice for a major part of the year. Although carbon turnover occurs at low rates compared to during the summer period, the winter period is long and winter metabolism could be significant on an annual scale [Karlsson et al., 2008; Welch and Bergmann, 1985], resulting in high accumulation of CO2, and sometimes also CH4, under ice in lakes [Kortelainen et al., 2006; Michmerhuizen et al., 1996; Striegl et al., 2001]. Still, only a few studies on carbon emission from lakes include fluxes of both CO2 and CH4 during the ice thaw period in spring [Karlsson et al., 2010; Striegl and Michmerhuizen, 1998], implying underestimation of the carbon emissions from lakes with a seasonal ice cover.
 In this study we test the hypothesis that the total carbon emission (CO2 + CH4) during ice thaw is of generally high importance for the annual carbon emission in lakes with long ice cover period. We sampled CO2 and CH4 in 12 relatively small subarctic lakes over a 3 year period including under ice and during the ice-free period. We estimated the emission of CO2 and CH4 during ice thaw as the difference in mass of carbon before and after ice thaw and finally compared this emission to that of the rest of the ice-free period.
2 Materials and Methods
 Twelve small (area: 0.6–14.5 ha, mean depth: 0.7–2.4 m, max depth: 1.5–8.5 m) lakes in subarctic northern Sweden (68°N, 18°E) were studied. The Dissolved Organic Carbon (DOC) in the lake water, measured monthly during the ice-free period of 2009, is 9.1 ± 1.2 mg L−1 (mean ± 1 SD) (E. Lundin, unpublished data, 2009). The catchments of the lakes are dominated by mountain birch forest (six lakes), by mire vegetation (four lakes), by a mix of birch forest and low-alpine tundra vegetation (one lake), or by a mix of all three vegetation types (one lake). Sampling of the lakes was carried out from 2009–2011 (2009: 7 lakes, 2010: 12 lakes, 2011: 11 lakes). All lakes were sampled as late as possible before ice thaw, around 1 week after ice thaw and then approximately monthly during the ice-free period (May–October). Two of the lakes were also sampled monthly during one winter (October 2009 to April 2010). We sampled the stream A2 (around 2 m wide) entering lake 1. The stream were sampled 100 m upstream of the lake every second week during the ice-free period (June–September) of 2009 and 2010 (May–October).
 Water samples in winter were collected from each 1 m depth interval at three locations (shallow, intermediate, and deep) in the lakes (except for the deep point in lake 1 which was sampled at every 2 m) with a Ruttner sampler and carefully transferred into 1 L bottle from which samples were collected for CH4 (22 ml glass vial) and CO2 (60 ml syringe). The ice thickness was measured on each sampling occasion. During ice-free season lakes were sampled at the deepest part at 1 m depth. Water samples from streams in both summer and winter were collected from the surface by hand into a 1 L polyethylene bottle and subsampled as above.
 For determination of CH4 concentrations, 5 ml of lake water was injected (one sample per depth and date) into a 22 ml exetainer. During each sampling occasion exetainers containing CH4 gas of two different concentrations (three replicates per concentration) were also prepared as above. The CH4 in the headspace of the exetainers were analysed using a gas chromatograph (Perkin Elmer Clarus 500 equipped with a headspace sampler) within 24 h of sampling.
 The pCO2 was measured using a headspace equilibration technique. The 60 ml syringes were filled without creating air bubbles where after 30 ml of water was replaced by 30 ml of air and equilibrated with lake water by vigorously shaking the syringe for 1 min and after which it was left to stand for one additional minute before analysis of the concentration of CO2 in the headspace gas with an infrared gas analyzer (EGM 4, PP Systems Inc, U.S.) was done. The concentration of CO2 in the water was calculated from the concentration of CO2 in the headspace according to the method of Cole et al. , using Henry's law and the fugacity-pressure relationship presented by Weiss  and for correcting CO2 in air.
 All lakes were echo sounded to estimate total and depth specific volume of water in the lakes. The volume interpolations were performed in the Arc GIS 9.3.1 (ESRI, U.S.) geostatistical analysis package using the ordinary Kriging method. The volumes of the lakes covered with ice in spring were determined by subtracting the volume of ice from the total and depth-specific lake volumes. In order to compare CO2 and CH4 across systems and seasons, we calculated the volume weighted concentrations following Riera et al. . Calculation of flux at spring ice thaw was carried out as the difference in mass of CO2 and CH4 before and after (normally 1 week) ice thaw, assuming that a minor fraction of the CH4 storage under ice was oxidized in the water column [Michmerhuizen et al., 1996]. The flux of CO2 and CH4 to the atmosphere during summer was estimated from the concentration of CO2 and CH4 in water according to Cole and Caraco . The comparison of fluxes between spring ice thaw and the ice-free period was carried out using ice-free data from 2009, i.e., the year with data of highest temporal resolution for the lakes.
 The lakes were ice covered from end of October to mid-May. There was a general increase in lake water concentrations of CO2 and CH4 over the winter (Figure 1), with higher values at increasing depths (see Supporting information). The CO2 and CH4 concentrations rapidly decreased after ice off (Figure 1). All lakes had high concentration of CO2 (range: 470–2344 μM) and CH4 (range: 7–581 μM) before ice thaw in spring compared to the mean values of CO2 (range: 25–56 μM) and CH4 (range: 0.1–2.9 μM) during the ice-free period.
 The flux during ice thaw was estimated to between 31 and 983 (mean: 338) mmol m−2 yr−1 for CO2 and between 2 and 202 (mean 67) mmol m−2 yr−1 for CH4 (Table 1). The flux of CH4 (p < 0.001, Student's t test), not CO2 (p = 0.109, Mann-Whitney rank sum test), was higher in mire lakes compared to the other lakes (Figure 2). The total carbon (CO2 + CH4) flux at ice thaw was dominated by CO2 (range: 56–97%, mean: 86%), with lower (p < 0.001, Student's t test) proportion in mire lakes (56–79 and 71%) compared to the other lakes. We tested for factors that could explain the carbon flux from all lakes at ice thaw. The flux of CO2 increased (r2 = 0.57, p = 0.004, linear regression) with the mean depth of the lakes while emission of CH4 was unrelated (p = 0.123) to mean depth (Figure 2). No linear relationship was found between CO2 or CH4 emission and lake area (p = 0.59 and 0.71, respectively) or DOC concentration (p = 0.15 and 0.33, respectively).
Table 1. Flux of CO2 and CH4 at Ice Thaw (Mean and Range for 2009–2011) and During the Ice-Free Period (Data from 2009) from the 12 Lakes. Mean Depth = Zmean
CO2 (mmol m−2 yr−1)
CH4 (mmol m−2 yr−1)
 The flux during the ice-free period of 2009 was estimated between −895 and 3836 (mean: 777) mmol m−2 yr−1 for CO2 and between 30 and 648 (mean: 117) mmol m−2 yr−1 for CH4 (Table 1 and Figure 3). Ice thaw flux of CO2 equaled 12−56% of annual flux in nine lakes. In the other three lakes, with net uptake of CO2 during ice-free season, ice thaw flux was higher than ice-free net uptake, resulting in a net release to the atmosphere on an annual basis. For CH4, ice thaw fluxes equaled 3–84% of the annual CH4 flux (Figure 3). The total carbon (CO2 + CH4) flux during ice thaw equaled 11–55% of the annual total carbon flux in nine of the lakes. For three of the lakes, including the flux during ice thaw, the annual carbon balance switched from an apparent sink (−247, −213, and −350 mmol m−2 yr−1) to an actual source (211, 81, and 209 mmol m−2 yr−1) of atmospheric carbon.
 The stream was covered by ice from November, was frozen to the bottom in March, and was ice-free again from late April. The CO2 concentration in the stream water increased during winter. There were also relatively high CH4 concentration in the stream in winter (Figure 1), but the trend was not as clear as for CO2.
 We found high and consistent accumulation of CO2 and CH4 in the lakes during winter and that the resulting carbon flux during ice thaw was important for the annual net exchange of carbon between the lakes and the atmosphere. In some lakes the fluxes at ice thaw added significantly to, or dominated over, the outgassing during the ice-free period, and in other lakes ice thaw fluxes counteracted net carbon uptake during the ice-free period and resulted in a switch in annual carbon balance from a small sink to a small source for atmospheric carbon. The results suggest an active carbon cycling during winter in these catchments, causing the lakes to become hotspots for greenhouse gas emissions as these systems reconnects with the atmospheric system at ice thaw.
 The depth of the lakes could explain a major part of the CO2 flux at ice thaw. This is logical since depth-integrated pelagic respiration is likely to increase with increasing depth of the water column. It is also likely that sediment respiration, which is important for winter respiration in shallow lakes [Karlsson et al., 2008], is more constrained by freezing in shallow versus deep lakes since a larger fraction of shallow lakes will freeze to the bottom. It is, however, uncertain to what degree the observed pattern holds for extrapolating to larger lakes since the respiration and carbon accumulation may not change proportionally with depth. Winter respiration is partly driven by organic carbon produced by benthic algae the previous summer [Karlsson et al., 2008], allowing high respiration rates compared to on settled organic matter [Ask et al., 2012; Ask et al., 2009a]. In deep lakes, suboptimal light conditions for benthic photosynthesis result in lower supply of high-quality organic matter and, consequently, relatively low benthic respiration rates in winter. This will likely not be fully compensated by an increase in pelagic respiration. Thus, the increasing mean depth may lead to proportionally lower accumulation of CO2 in winter, i.e., the slope of the emission-depth relationship may decrease.
 Part of the carbon dynamics, at least early in winter, may be explained by carbon input from the catchment. The CO2 concentration was relatively high in the stream during winter, showing similar trends and values as for the lakes (Figure 1). Using published data from our study site on base flow in late October, i.e., at the time of ice on in the lakes, gives 13 days to exchange the water volume of lake [Olefeldt et al., 2012]. This is a very rough estimate since it is unknown how rapidly the water flow decreases during winter, but it suggests that the external input could be important for the carbon dynamics we observe in the lakes at least in the beginning of the ice cover period. Since CO2 concentrations is higher in stream compared to in lake water in autumn (Figure 1), this input would raise lake CO2 concentrations under ice and contribute to the flux-depth relationship (Figure 2).
 The data suggest that the accumulation and flux of CH4 are controlled differently compared to CO2. Previous studies have found increasing CH4 flux in spring with decreasing lake area [Michmerhuizen et al., 1996] and attributed this to the increased importance of littoral areas, favoring CH4 production, as the lake size decreases. Our lakes cover a smaller gradient in lake area and do not vary considerably in littoral fraction since all lakes are relatively shallow with illuminated benthic habitats that would allow photosynthesis [Ask et al., 2009b]. Instead, comparison of the two intensively studied lakes (Figure 1) as well as all lakes (Figure 2) show that high CH4 accumulation and flux from lakes was mainly related to mire catchments surrounding these lakes. Studies on the mires have shown substantial CH4 production in autumn and winter [Backstrand et al., 2008]. Part of the CH4 produced in the mires in winter may be transported to downstream aquatic systems through unfrozen peat layers underneath the seasonally frozen surface peat. Although we found low CH4 concentration in the stream compared to in the mire lakes (Figure 1), these data are probably poor indicators of terrestrial export of CH4 due to potential CH4 oxidation in streams and that the stream is only partly influenced by mires.
 Our data show that the flux of accumulated carbon to the atmosphere during ice thaw in spring was quantitatively important equaling between 12% and 100% of the annual emission from the lakes, the latter when ice-free season flux was negative. The result is in accordance with a few previous studies quantifying the annual importance of emission at ice thaw of both CH4 and CO2 [Karlsson et al., 2010; Striegl and Michmerhuizen, 1998]. Our data also show that the flux to the atmosphere at ice thaw switched the annual net balance for some lakes from a sink to a source for atmospheric carbon. These results suggest that the emission from small lakes with a seasonal ice cover is larger than normally estimated. The existing studies on annual carbon fluxes from seasonally ice-covered lakes include relatively small lakes and it is likely that the gas accumulated under ice in large deep lakes are released over a longer time period after ice off and that this is to a larger extent captured in typical sampling programs than what is the case for smaller lakes. Nevertheless, given the many small lakes with long ice cover period in boreal and arctic regions, the carbon emission at ice thaw in spring is of large significance and needs to be accounted for in estimates of land-water exchange of carbon gas. Our study suggests that recent estimates of global CO2 and CH4 emissions from lakes [Cole et al., 2007; Tranvik et al., 2009] are too low. The ice cover period in lakes is very sensitive to changes in climate conditions, with major decrease in ice coverage being observed in response to warming since the 19th century [Magnuson et al., 2000]. Thus, knowledge about the importance of the winter period for lake carbon fluxes will be important not only for understanding present day conditions but probably also for predicting effects of climate change on aquatic systems and its related feedback effects on the climate system.
 Thanks to Thomas Westin, Tyler Logan, Marina Becher and Daniel Karlsson for field and laboratory assistance. The study was supported by the Swedish Research Council (621-2008-4390) and the Kempe Foundation.