The waters of Storfjorden, a fjord in southern Svalbard, were investigated in late April 2002. The temperature was at the freezing point throughout the water column; the salinity in the top 30 m was just above 34.8, then increased nearly linearly to about 35.8 at the bottom. Nutrient and oxygen concentrations showed a minimal trend all through the water column, indicating minimal decay of organic matter. Normalized dissolved inorganic carbon, fCO2, and CFCs increase with depth below the surface mixed layer, while pH decreases. In waters below 50 m, there was an increase in dissolved inorganic carbon, corrected for decay of organic matter using the phosphate profile, corresponding to about 9 g C m−2 relative to the surface water concentration. We suggest this excess is a result of enhanced air-sea exchange of CO2 caused by sea ice formation. This enhancement is suggested to be a result of an efficient exchange through the surface film during the ice crystal formation and the rapid transport of the high salinity brine out of the surface layer.
 Storfjorden is a large Arctic fjord in southern Svalbard (Figure 1). It is about 160 km long, with a sill depth of 120 m and a maximum depth of about 180 m. Atlantic source water enters from the Norwegian Sea. The characteristics of the water in Storfjorden are greatly influenced by the formation of sea ice in winter, with most regions being ice-covered. As the ice becomes consolidated in central regions, air-sea gas exchange will be hampered. Over the shallow areas along the eastern part of the fjord, however, offshore winds export the ice out of the fjord, thereby enhancing ice production and resulting in regions that remain ice-free until late winter [Haarpaintner et al., 2001a]. Most of the ice formed in the winter season is exported, but some remains, melts in summer, and (together with meltwater from the Barents Sea) contributes to a lower salinity surface water in Storfjorden [e.g., Haarpaintner et al., 2001b].
 In winter, ice formation and the resulting brine rejection lead to convection and a homogeneous surface mixed layer. Over the shallow areas, where offshore winds enhance ice production, the convection will reach the bottom, salt will accumulate in the water, and the water over the shallow fjord shelf will gradually increase in density. As it does so, the dense water will flow off the fjord shelves to fill the deeper regions of the fjord [Haarpaintner et al., 2001a]. This shelf-slope flow will carry with it near-surface constituents, including gases that enter the surface waters from the atmosphere. The water in the deeper layers of the fjord formed by this process will eventually overflow the sill and subsequently sink down into deeper regions of Fram Strait [Quadfasel et al., 1988]. In the Arctic Ocean, shelf-slope plumes are an intimate part of the processes that determine the properties of the mid-depth and deep waters of the Arctic Ocean, including their chemical constituents [e.g., Rudels et al., 1994; Jones et al., 1995; Schauer et al., 1997; Anderson et al., 1999a].
Anderson et al.  evaluated the chemical modification by the decay of organic matter in the high-salinity bottom water of Storfjorden as observed in summer 1986. They showed that the regeneration rates of phosphate, oxygen, and total dissolved inorganic carbon (CT) followed the classical P:O:C ratios 1:−135:106 of Redfield et al. . The regeneration rate of CT during the winter of 1986/1987 was estimated to be 3.7 mmol m−2 d−1 over a 6-month time period, or about 8 g C m−2 yr−1. Dense water plumes leaving Storfjorden [Schauer and Fahrbach, 1999; Quadfasel et al., 1988] will eventually transport this dissolved carbon to the deep ocean.
 We discuss the enhanced transfer of CO2 from the atmosphere to the deep waters of Storfjorden and address the carbon “solubility pump” in distinction to the “biological pump,” which was evaluated by Anderson et al. .
2. Data and Methods
 This study is based on data collected in Storfjorden, southern Svalbard (Figure 1), from April 26 to 29, 2002, during the Arctic Ocean 2002 expedition with the icebreaker Oden. Winter conditions prevailed, and air temperature was well below freezing (Table 1). Also, at this time the fjord was 100% sea ice covered along the investigated area north of the mouth (between stations 6 and 7) as a result of strong southeasterly winds over the few days previous. Water was collected using a 24-bottle rosette sampler equipped with 7.5-L Niskin-type bottles and a CTD (Sea Bird 911). When brought back on board, the rosette was moved into a heated double container as quickly as possible to avoid freezing of the samples.
Dates are given as, for example, 2002-04-26 for 26 April 2002.
 Water for the determination of chlorofluorocarbons CFC-11 and CFC-12 (CFCs), oxygen, total dissolved inorganic carbon (CT), pH, total alkalinity (AT), and nutrients was drawn from water samplers directly after the rosette was secured in the heated container and analyzed on board within hours of sampling. Standard methods of analysis were used: gas chromatography with an electron capture detector for CFCs [e.g., Fogelqvist, 1999], precision ∼2%, automated Winkler titration of oxygen, precision ∼1 μmol kg−1, coulometric determination for CT, precision ∼1 μmol kg−1 [Johnson et al., 1987], spectrophotometric determination of pH, precision ∼0.003 pH units, [Clayton and Byrne, 1993; Lee and Millero, 1995], potentiometric titration of AT, precision ∼1 μmol kg−1 [Haraldsson et al., 1997], and determination of nutrients, precision around 1%, using an auto-analyzer according to the WOCE protocol [Gordon et al., 1993]. The given precisions were computed as standard deviations of duplicate analyses. Certified reference materials (CRM), supplied by A. Dickson, Scripps Institution of Oceanography (USA), were used to calibrate the analyses for CT and AT. The accuracy was set by correcting the measured samples by the factor achieved when dividing the given concentration of the CRM with that determined on board, correction maximum 0.5%.
 The fugacity of CO2 (fCO2) was computed from CT and pH, using the computer program CO2SYS, version 01.03 [Lewis and Wallace, 1998]. The carbonate dissociation constants (K1 and K2) used were those of Roy et al.  and KSO4 determined by Dickson . On the basis of the standard deviations of the individual values of the stability constants of the reactions involved and the precision of the measured CT and pH, the precision of the computed fCO2 was better than 3 μatm [Anderson et al., 1999b].
 At the time of investigation the fjord was filled with high-salinity water at, or very close to, the freezing point. The salinity in the surface water was about 34.8. This increased nearly linearly from a depth of about 30 m to a maximum value of about 35.8 at the bottom (Figure 2a). All waters outside the fjord have salinities well below this, with the Atlantic waters having the highest salinity, up to 35.1. All the waters with S > 35 collected at stations 1 to 6 within Storfjorden were at, or very close to, the freezing point (Figure 3). Consequently, the high salinity observed in the deep Storfjorden can only result from sea ice production with subsequent brine release to the underlying waters as also been observed in the past by others [e.g., Haarpaintner et al., 2001a, 2001b].
 Inside the fjord, only small variations with depth in oxygen and nutrient concentrations were observed. There was, however, a significant decrease in pH with depth that was accompanied by an increase in fCO2 and salinity normalized CT (CT-35), but not by a significant change in normalized AT (AT-35) (Figures 2g–2j). (Note that since CT and AT are nearly proportional to salinity (S), they are “normalized” to a constant salinity, i.e., CT-35 = 35CT/S and AT-35 = 35AT/S, in order to emphasize effects not related to variability in salinity.) The increase in fCO2 and CT-35, and the nearly constant nutrient and oxygen concentrations with increasing depth indicate that other than biological processes are affecting the inorganic carbon system as will be discussed below.
 At station 7 the top few meters, which had temperatures below 0°C and lower salinity and nutrient values than neighboring stations farther into the fjord, also had oxygen concentrations close to 100% saturation, typical of a primary production signal in this stratified surface water (Figures 2a–2f). This primary production signal is also supported by the observed minimum in fCO2 (Figure 2h).
 The average depth profiles of the five most northern stations inside Storfjorden (Figure 1) are shown in Figure 4, where the error bars reflect the variability among these stations (Figure 2). As mentioned above, the only source of the high-salinity deep water of the fjord is brine from sea ice production. Other studies [Haarpaintner et al., 2001a] have shown that a major source region for these waters is the eastern shallow area and that the brine drains down along the bottom to the deep central parts. Assuming that the top 30 meters are homogeneous within the fjord, we can treat the observed surface water as the source water from which the brine is produced. The small trend in the nutrient profiles (Figure 4c) indicates that little in the way of biochemical transformations has occurred. The increase in concentration starting at 40 m depth is seen in all nutrient profiles, but nitrate shows a more pronounced decrease deeper than 100 m. This signal is very small considering the variability, but it could come about by some denitrification occurring in low oxygen environments at the sediment interface.
 The nearly constant AT-35 profile of Figure 4d (2315 ± 2 μmol kg−1) also indicates minimal dissolution of metal carbonates from biologically produced shells. Furthermore, calcium carbonate can in principle precipitate in the brine channels of sea ice [Assur, 1958], significantly affecting total alkalinity, but no signal of this is seen. The observed variability in other carbon cycle parameters, increasing CT-35, decreasing pH and increasing fCO2 with depth, is therefore likely a result of abiotic processes.
 The observed nutrient and oxygen profiles are in contrast to those from a study made in Storfjorden in late July of 1986, which showed nutrient concentrations 20% to 50% higher in the salinity enriched bottom water relative to that 50 m above the bottom [Anderson et al., 1988]. This trend was accompanied by a decrease in oxygen concentrations that corresponded closely to what is expected when organic matter decays [Redfield et al., 1963]. The difference between this study and the one in1986 is a 3-month period during which primary productivity occurs with subsequent sedimentation and decay of organic matter.
 In order to compensate for the small biological contribution to the CT-35 signal as observed in the nutrient profiles (Figure 4c), we use the change in phosphate and multiply this by 106 as determined by the C:P relationship in organic matter [Redfield et al., 1963]. Phosphate is used instead of nitrate as it has the largest relative increase with depth, therefore making the largest compensation. This corrected CT-35 (CT-35corr) was computed for all data of the five most northerly stations using the equation
where the measured properties are noted by superscript “meas” and the average concentration in the top 30 m by superscript “surf.” It should be noted that this correction includes uncertainties as the C:P relationship of 106 is an approximation that is strictly valid only over very long time and large space scales, when variations in biological communities is average out. However, the correction we apply is very small, and the intention is only to show that the observed increase of CT with depth is not supported by decay of organic matter with the subsequent increase of nutrients.
 The resulting average concentration profile still shows a clear increase with depth (Figure 5) of about 10 μmol kg−1 between the top 30 m and the bottom. What could be the cause of this observed trend? One indication is given by the CFC profiles, which show increasing concentrations toward the bottom (Figure 4f). CFC-11 increases by about 8% from the surface to the bottom, where it is close to saturation, while CFC-12 increases by about 5% from just under saturation at the surface to slightly over saturation at the bottom. The increase in CFC concentrations versus depth is well above the precision of the measurements, but the concentrations relative to saturation levels should be considered with care. There are uncertainties both in the accuracy of the CFC data (even though calibrated with NOAA standards) and in the solubility equations especially at these low temperatures. The topic of the accuracy of the solubility equations is outside the scope of this work, but the calibration and the accuracy of the solubility equations would have little effect on the changes in concentrations with depth. Thus the significant increase of CFC-11 and CFC-12 concentrations with depth supports an air-sea uptake by the high salinity waters.
 Oxygen on the other hand does not show any increase in saturation with depth (Figure 4b). However, as oxygen is consumed during decay we make a correction using the same approach as for CT-35 (equation (1)), a difference being that we multiply the change in phosphate by −135, the O:P ratio during mineralization of organic matter [Redfield et al., 1963]. The resulting profile (Figure 6) shows an increase in oxygen saturation from about 94% at the surface to about 96% at the bottom. This increase toward equilibrium with the atmosphere is of the same relative magnitude as that observed for fCO2 (Figure 4e), about a third of the way (from ∼286 μatm to ∼320 μatm, relative to the atmospheric pCO2 ∼ 376 μatm being the mean of April 2002 at the Zeppelin Mountain outside Ny-Ålesund on Svalbard (http://www.misu.su.se/∼kim/pict/trend.jpg)).
 The atmosphere is the source of CFCs and the observed increase with depth must hence be a result of air-sea flux. Such a flux requires a mechanism that is associated with the process responsible of the formation of these bottom waters, for instance sea ice formation. Air-sea gas exchange is very dependent on the surface film properties and mixing of the near-surface water. The effect by wind, increasing the turbulence in the very top surface water, together with the difference in partial pressure between the atmosphere and the surface ocean, has traditionally been used to parameterize the air-sea gas exchange [e.g., Wanninkhof, 1992]. Sea ice formation likely also increases the turbulence at the air-sea interface when ice crystals form together with brine at the very surface. We hypothesize that a highly efficient gas exchange takes place in conjunction with the heat exchange that occurs when ice is produced in open water with a partial pressure different from that of the overlying atmosphere. Very low temperatures in the salt-enriched water that surrounds the ice crystals during formation could amplify this gas exchange. This water in the very top layer that takes up the gas brings the concentration closer to equilibrium with the atmosphere. The addition of salt to the surface water during ice formation increases its density, causing it to sink and to be replaced by near-subsurface water having the “original” surface water partial pressure. Consequently, sea ice formation also amplifies mixing of surface water by the continuous removal of the water that has taken up the gas and hence promotes air-sea exchange. The mixing of the surface water is also supported by the strong easterly winds pushing the sea ice away and maintaining open water along the eastern edge of the fjord. The strong winds are crucial as they maintain open water, which both enhance sea ice production and provide efficient air-sea gas exchange.
 Thus we suggest that the increase in CT-35corr with increasing salinity (and depth) is a result of uptake of CO2 from the atmosphere. The difference in CO2 partial pressure between the surface water and the overlying atmosphere is about −90 μatm, 286 μatm relative to 376 ppm. This strong undersaturation in the surface water results in a gradient that will drive a flux from the atmosphere into the sea, if the conditions are favorable, such as during sea ice production. A similar process functions for CFCs and oxygen, but with less effect as their surface water concentrations are closer to being in equilibrium with atmospheric concentrations.
 The total CO2 uptake, computed by integrating the excess CT-35corr down to 150 m relative to the average CT-35corr concentration in top 50 m of the profile (Figure 5), is about 0.7 mol m−2 or close to 9 g C m−2. If this entire signal is attributed to the winter season of 2001/2002, the CO2 uptake is comparable to the flux associated with the biological pump in the fjord in 1986 [Anderson et al., 1988]. Compared to the surrounding region, this areal air-sea flux is about 25% of that estimated into the Barents Sea [Fransson et al., 2001]. The latter flux is driven by both cooling and biological primary production.
5. Concluding Remarks
 We have shown that the waters below 50 m in the Storfjorden had an excess of dissolved inorganic carbon in the order of 9 g C m−2 after correcting for the effect of microbial decay of organic matter. We suggest that sea ice formation causes this excess by enhancing air-sea exchange of CO2. This enhancement is hypothesized to be a result of an efficient exchange across the surface film during the ice crystal formation, increased solubility in the low-temperature brine, and the rapid transport of the brine-enriched high-salinity water to deeper waters.
 The oceanic uptake of atmospheric CO2 has not previously been considered in high latitudes during the winter season. Our suggested mechanism will be functioning and efficient in taking up atmospheric CO2 in regions of polynyas and leads if the surface water is undersaturated with respect to CO2 in this period. In a larger context, we suggest that the processes associated with deep water formation and transport of CO2 and other gases in Storfjorden to deeper waters are similar to those operating in regions of the Arctic Ocean with extensive sea ice production [e.g., Jones et al., 1995; Schauer et al., 1997; Anderson et al., 1999a]. These processes can thus provide a sink for atmospheric CO2 even in winter and are thus relevant to the understanding of gas transport in a much larger Arctic region. A rough estimate shows that the brine added to deep Storfjorden requires sea ice formation corresponding to near 3 m if two thirds of the salt is expelled into the brine. This is based on an average “excess” salt of 0.5 over 120 m depth (Figure 3a) and a loss of 2/3 × 34.8 salinity to the brine, or sea ice production [m] = 120[m] × 0.5/(2/3 × 35). With the observed fCO2, these 3 m give a flux of 9 gC m−2 or 3 gC m−2 per meter ice produced. One meter of ice is typically formed each year in the Arctic Ocean. If 10% of this ice production occurs in polynyas and open leads, which are kept open by wind throughout the productive season, there is a potential of an air-sea flux of 3 × 1012 gC yr−1 if all of the central Arctic Ocean and the major shelf seas areas are considered. The total air-sea flux in the Arctic Ocean is estimated to be 41 ± 18 × 1012 gC yr−1 [Anderson et al., 1998], and if our computation is correct 5% of this is attributed sea ice production. Furthermore, the CO2 taken up through this process is directly included in the cold, high-salinity brine-enriched water that sinks and can, under the right conditions, reach deep into the ocean.
 We are grateful to the physical oceanography group on board for running the CTD rosette package and for letting us have access to the CTD data. This research has been supported by grants from the European Union's 5th Framework Programme project TRACTOR (contract EVK2-CT-2000-00080), by the Swedish Research Council, by the Swedish Polar Research Secretariat, and by the Canadian Panel on Energy Research and Development.