All values are below the expected content of instant freezing seawater (IFSW) (Cox and Weeks, 1983) and ranged between 4 and 21 mL STP kg-1 ice (Fig. 2f). A peak in gas content was observed between 25 and 35 cm, which is where both the bulk ice salinity, and the brine volume fraction were at their lowest and where the ice was depleted in δ18O.

The concentration of dissolved CH4 in the seawater ranged from 5.7 to 18.4 nmol L-1 and the maximum concentrations were measured at the ice–water interface (Fig. 4). The bulk ice methane concentration, [CH4]bulk ice, ranged from 1.8 to 12.1 nmol L-1 (Fig. 3). The methane, as part of total gas volume (i.e., mixing ratio : [CH4]bulk ice divided by the total gas content of the ice), ranged from 3.2 to 28.7 ppmv during the study, with an average value of 11.8 ppmv. The dissolved methane measured in the brine liquids, [CH4]br, ranged from 12.0 to 17.0 nmol L-1.

The pCO2 of the underlying seawater (below 0.5 m) was slightly undersaturated compared to the atmosphere (390 ppmv) and very undersaturated (77–130 ppmv) compared to seawater at the ice–water interface. For the ice column, the average bulk ice pCO2 was 194 ppmv. Except on 16 March, the bulk ice pCO2 increased from a minimum of lower than 185 ppmv at the bottom of the ice to a maximum, exceeding 330 ppmv, in the top layers of the ice (Fig. 5.).

A thin surface layer with low salinities (Ssurf < 20) and strong isotope depletion (δ18Osurf < -8 ‰) characterized the upper part of the water column. The low salinities and heavy isotope depletion could result from a dilution process, which implies a mixing of the fjord water with a water mass either from melting ice or from the Kapisillit river, or both. Assuming a conservative mixing between the three end members, the fraction of river water and the melting ice in the fjord water surface can be deduced from the following equations: 1=friv+fice+ffjord,Ssurf=Sriv⋅friv+Si⋅fice+ffjord⋅Sfjordδ18Osurf=friv⋅δ18Oriv+fice,⋅δ18Oice+ffjord⋅δ18Ofjord, where friv, fice and ffjord are the fraction of river water, melting ice and fjord water present at the surface of the fjord. Ssurf, Sfjord, Sriv and Sice are the salinity of the surface water (Ssurf < 20), the fjord water at 9 m depth (Sfjord= 32.9), the river water (which is assumed to equal 0) and the bottom bulk ice salinity (4.6 < Sice < 8.8), where δ18Osurf, δ18Oriv, δ18Oice and δ18Ofjord are the δ18O content of the fjord surface layer (-6.44 ‰ < δ18Osurf < -7.82 ‰), the river water (δ18Oriv = -14.84 ‰), the bottom of the ice (δ18Oice= -4.55 ‰) and the fjord water at 1.09 m depth (δ18Ofjord= -0.957 ‰). Based on the results from these equations, the fraction of freshwater during the sampling period varied from 33 to 50 % and water from the melting ice was less 2 %. This is in agreement with the study of Long et al. (2012) from the same area at the same period, which revealed low rates of ice melt at a maximum of 0.80 mm day-1. As the river was unfrozen below its surface, we suggest that the stratification results from an input of freshwater from the Kapisillit river.

In the previous section we have shown that freshwater from the Kapisillit river contributed to the low salinities close to the ice–water interface. During earlier ice growth, a freshwater input may have produced similar changes of salinity, gas content and isotope content into the water from which the sea ice had formed. If no change had occurred in the parent water, we would expect that the isotopic composition of sea ice from the under-ice water would monotonically shift towards higher δ18O values due to isotopic fractionation favoring the incorporation of heavy isotopes into the solid phase (ice) as compared to the liquid phase (Eicken et al., 1998; Souchez et al., 1988). Similarly, due to salt rejection during the freezing process, the salinity of sea ice would shift towards lower salinities than the parent seawater. The salt rejection rate as well as the isotopic fractionation depends mostly on the sea-ice growth rate (Eicken, 2003). Once sea ice is formed, desalination occurs. The combination of salt segregation, gravity drainage and brine expulsion explains the typical “C-shape” of the salinity profiles, which evolve into the “I-shape” during the melting period (Eicken, 2003). The δ18O profiles should, however, still monotonically increase towards the bottom of the ice since most of the isotopic signal is recorded into the ice crystals themselves.

According to our data, the average bulk ice salinity (Si) of 3.2 was lower than typical values of first-year ice by a factor of 2 (Cox and Weeks, 1988; Eicken, 2003) and close to the bulk ice salinity (2 < Si < 4.7) of sea ice grown under the influence of brackish water in the Laptev region of the Arctic Ocean (Eicken et al., 2005). The low Si induced a low brine volume fraction (Vb) that exceeded 10 % only in the bottom horizons of the ice and fell below the permeability threshold of 5 % (Golden et al., 1998, 2007) in the first 5 cm, between 20 and 30 cm and at 45 cm depth (Fig. 5; dotted area). Thus, the brine network was stratified and not fully interconnected during the sampling period. Our measurements suggest that no fluid transport by convective processes could have taken place during the sampling period since the Rayleigh numbers (Ra) were well below the threshold of 10 (Notz and Worster, 2009) or 5 (Vancoppenolle et al., 2010).

There was a positive correlation between the bulk ice salinity and the isotope content between the 20–30 cm interval (Fig. 6). The isotopic composition measured on bulk sea-ice results from a contribution of both brine and pure ice. Since the brine accounts for less than 10 % of the ice volume, the ice isotopic composition measured originates mainly from the pure ice. As no fractionation occurs in the solid phase during melting (Jouzel and Souchez, 1982), the δ18O content in pure ice should not change over time and should depend only on processes occurring during freezing and on the composition of the underlying seawater (Eicken et al., 1998). On the other hand, the bulk ice salinity originates from the brine medium. Like the ice isotopic content, the bulk ice salinity depends on the ice growth velocities and on the composition of the underlying seawater. However, once the ice is formed, desalination occurs. Sea ice looses salt primarily through brine drainage, convective transport, or flushing (Untersteiner, 1968). Since the bulk ice salinity is still correlated with the ice isotopic signal (Fig. 6.; R2= 0.64, p < 0.001), it appears that no major desalinization process has been active within the ice cover. Therefore, this ice should hold the original characteristics inherited from the parent water during the freezing process.

S-shaped profiles in the ice isotopic distribution and bulk ice salinity are uncommon and cannot be explained by the freezing or by desalinization processes. Since the ice core samples were obtained before the onset of surface melt, and as there are no traces of either convective processes, meltwater infiltration or superimposed ice in any of the samples suggests that both the anomalies in the ice salinity profiles and the heavy isotopic depletion in the ice are mainly due to changes of water mass. We also note that the repeatedly oriented columnar ice crystals observed between 24 and 32 cm (Fig. 3) suggest the presence of a current at the ice–water interface.

The bulk ice salinity and the ice isotope distribution can be used as a proxy to track the salinity and the isotopic composition of the parent water, and hence indirectly quantify the fraction of river water. Thus, we may derive the fraction of river water present in a volume of sea ice from isotope mass balance using the equation developed by Eicken et al. (2005): friv=δ18O-ε-δ18Ofjordδ18Oriv-δ18Ofjord, where δ18Oice,δ18Ofjord,δ18Oriv are the isotopic composition of the ice, fjord water at 1.09 m depth and river water, respectively. ε is the sea-ice fractionation coefficient ranging between 1.5 and 2.7 ‰ (Eicken et al., 1998). Since δ18Ofjord and δ18Oriv are constant over winter (November to April) (Fitzner et al., 2014), we can determine the temporal changes in surface water composition, namely the fraction of river water friv and the salinity of the surface parent water mass Sp. According to Eicken et al. (2005), Sp is given as Sp=1-friv⋅Sfjord. The maximum friv deduced from the ice isotopic mass balance occurs when the sea ice reached a thickness of between 20 and 30 cm. The computed salinity shows that at this time Sp had to drop by a factor of 1.7 and could not exceed 18. This shift probably induced the drop in bulk ice salinity in the horizon at 25 cm below the surface.

The total gas content was always lower than the 23.75 mL STP kg-1 value expected in IFSW (Cox and Weeks, 1983). This is in agreement with previous reports that gases in seawater are preferentially expelled from the growing ice, as with other impurities (Cox and Weeks, 1983, 1988; Killawee et al., 1998; Loose et al., 2009, 2011; Tison et al. 2002). The range of total gas content values for all the samples was 4 to 21 mL STP kg-1, which is comparable to the range obtained by Matsuo and Miyake (1966) for natural sea ice (2.2 to 21.2 mL STP kg-1) and the range of 3 to 18 mL STP kg-1 obtained by Tison et al. (2002) for artificial sea ice. In this study there was a peak of gas content between 25 and 35 cm associated with the lowest salinities. As demonstrated above, the change of salinity was caused by a change in the parent water due to an input of freshwater. By computing friv and Sp during the ice growing period, we can further assess the freezing temperature of the parent ice water following the equation of UNESCO (1978). Based on the salinities and temperatures of the surface waters, we can compute the gas content of the surface parent water at each stage of ice growth using the solubility law given by Garcia and Gordon (1992) for O2, and Hamme and Emerson (2004) for N2 and Ar. When the ice reached a thickness of 25 cm, the salinity of the parent water decreased by a factor of 1.7, increasing the gas solubility by a factor 1.05, although the sea-ice gas content had been increased by a factor of 2.1. The increasing gas content in the parent water, dictated by the increasing solubility, could not explain the gas peak in the middle horizons of the ice. Tison et al. (2002) showed that the initial gas content of sea ice is affected by a current at the ice–water interface that was 3 times higher than the gas content of sea ice grown with a stagnant water–ice interface. According to Tison et al. (2002), the current zone is characterized by a thinner boundary layer in which both salts and gases are controlled by diffusion. Due to this thin boundary layer, the current zone reaches the critical value necessary for early bubble nucleation. Once started, bubble nucleation ensures that that the gasses that would otherwise have diffused as a solute towards the water will be entrapped as bubbles in the sea ice. Moreover, the total gas content depends on ice growth velocity; while ice from freshwater grows faster than ice from salty water, more gases were trapped in the sea-ice cover. Thus, we propose that the peak in sea-ice gas content is a result of the increasing gas content in the parent water and the presence of a current under the sea-ice–ocean interface, both caused by a flux of fresh water.

The methane concentration was measured in the water column [CH4]sw, in the ice cover [CH4]bulk ice and in the brine medium [CH4]br. First, we compare [CH4]sw and [CH4]bulk ice with the methane concentration at saturation, [CH4]sat sw and [CH4]sat bulk ice. The concentration at saturation is the concentration of methane dissolved in a liquid (i.e., brine or seawater) in equilibrium with the atmospheric partial pressure of methane. The saturation is determined by the solubility, which depends on the temperature and salinity. CH4 saturation with respect to the atmosphere at the sea surface was calculated assuming a 2010 atmospheric mixing ratio of 1.9 ppmv, using the CH4 solubility equation of Wiesenburg and Guinasso (1979) corrected for the in situ temperature and salinity. We assumed that the relationships from the reference are valid for the range of temperature and salinity found in the ice. The bulk ice saturation [CH4]sat bulk ice was obtained by multiplying the calculated solubility in brine (nmol L-1 brine) by the relative brine volume (L brine L-1 bulk ice). The ratio of the observed [CH4]bulk ice to the [CH4]sat bulk ice determines the supersaturation factor.

According to Weisenburg and Guinasso (1979), the methane concentration in equilibrium with the atmospheric pressure for seawater salinities > 33.9 and at -1.9 ∘C is 3 to 4 nmol L-1. Thus, the dissolved methane in the surface water was 4.5 times the concentration at saturation (i.e., 450 % supersaturated). Such high concentrations, up to 15 nmol L-1, suggest that the top layer of water column was somehow affected by nearby sources (i.e., sediment degassing, riverine input). These measurements are similar to the values (5–55 nmol L-1 water) reported by Damm et al. (2007) in Storfjorden (Svalbard archipelago) and in the fjords of Spitsbergen (Damm et al., 2005). The observed concentrations are, however, lower than the concentration in the East Siberian Arctic Shelf (ESAS) measured by Shakova et al. (2010). Due to the thawing sub-sea permafrost, CH4-rich bubbles released from the sea floor rise up through the water column. As far as we are aware, no sub-sea thawing permafrost has been reported for Godthåbsfjord. Damm et al. (2008) suggested that methane plumes in the water originate from sediments during winter and from in situ production during phytoplankton blooms during the summer. As our studies took place before the algal bloom, one of the main methane sources should be the sediment. The methane released from sediments rises through the water column by convective processes and accumulates under the ice. However, we have to consider the Kapisillit river that runs through a fen (Jensen and Racsh, 2011) as potential source of dissolved methane at the surface of the water column. The fen in the fjord area is a source of CH4 due to the permanently wet conditions that promote anaerobic decomposition, by which CH4 is an end product (Jensen and Racsh, 2011). Like CH4 maxima in the overlying sea ice also coincide with salinity minima, the Kapisillit river is most likely the major source of CH4 into the fjord system.

The range of CH4 concentrations in the bulk ice (1.8–12.1 nmol L-1 ice) is in agreement with measurements (15 to 25 nmol L-1 ice) reported by Lorenson and Kvenolden (1993) and Zhou et al. (2014). The deduced mixing ratio ranged between 3 and 28 ppmv indicating that the methane concentration within the sea ice was consistently higher than that in the atmosphere (1.9 ppmv). The sea-ice CH4 concentrations exceeded atmospheric saturation with saturation state ranging from 1200 to 70 000 % and an average saturation level of 2400 %. Thus, sea ice could plausibly act as a methane source to the atmosphere.

First, we compare [CH4]bulk ice with [Ar]bulk ice (bulk ice Ar, O2, N2 were measured by gas chromatography and the profiles are discussed in Crabeck et al. (2014)). As the diffusion process controls the gas motion within the sea ice Crabeck et al., (2014), Ar can be used as a tracer of physical processes. In marine biology, the O2 : Ar ratio is commonly used to remove the physical contribution to oxygen supersaturation for determining the biological oxygen production. Our CH4 samples were randomly distributed related to the Ar (R2≤ 0.21, p < 0.01; Fig. 7.). The CH4 : Ar ratio was systematically higher than the atmospheric and seawater ratios. Hence, CH4 was preferentially accumulated within the sea-ice cover relative to Ar. This could be related to in situ biological production, since methane is produced by methanogen bacteria in anoxic conditions. DNA (deoxyribonucleic acid) from methanogen archaea as well as anaerobic pathways have been observed in Arctic sea ice (Collins et al., 2010; Rysgaard and Glud, 2004). However, high-resolution O2 measurements in the Kapisillit sea ice showed there were no anoxic conditions during the sampling period (Crabeck et al., 2014). Denitrification, which is an anaerobic process producing more energy per organic carbon unit than methanogenesis, was not active within the sea ice due to presence of oxygen. In addition, the Ar : N2 ratio showed no significant deviations from the seawater ratio (Fig. 7). Therefore we conclude that the methane accumulation was not related to biological in situ production. We suggest that the accumulation of methane was due to (1) the initial CH4 concentration in the seawater and (2) the bubble formation that can act as a methane trap. Assuming a steady rate of incorporation for each gas, the accumulation and supersaturation (up to 450 %) of CH4 in the surface water leads to a larger incorporation of CH4 in the ice, and consequently greater CH4 accumulation relative to Ar. The bulk ice methane concentration (CH4]bulk ice represents the total methane content within the sea ice as the sum of the dissolved methane in the brine medium (CH4br) and the methane trapped in bubbles ([CH4]bubbles). We can deduce the fraction of the methane trapped in bubbles f[CH4]bubbles, using the following relationships: [CH4]bulk ice=[CH4]bubbles+CH4br, f[CH4]bubbles=[CH4]bulk ice-CH4br[CH4]bulk ice⋅100. Based on these relationships, the trapped methane bubbles contributed up to 70 % of the sea-ice methane content. According to our estimates, methane bubbles were preferentially accumulated in the upper layers of the sea ice. We pinpoint two sources of methane bubbles; (1) the methane bubbles trapped during the freezing process and (2) the in situ formation of methane bubbles within the brine medium. During seawater freezing, dissolved solutes (ions and gases) are segregated from the ice matrix (Cox and Weeks, 1983; Killawee et al., 1998; Untersteiner, 1968) and become concentrated in the liquid near the freezing interface. As the partial pressures of gases increase beneath growing ice, bubbles nucleate and are included along ice crystal boundaries (Bari and Hallett, 1974; Cox and Weeks, 1983; Killawee et al., 1998; Tison et al., 2002). Taking into account that the surface water was initially supersaturated with CH4, the value to trigger the nucleation process will be reached earlier than for the other gases. Thus, the growing ice was enriched in methane by bubble formation that would otherwise have diffused as a solute towards the water reservoir (Tison et al., 2002).

As the sea-ice temperature decreases, the brine becomes further concentrated and supersaturated, and gases can nucleate from the solution (Killawee et al., 1998; Tison et al., 2002). As CH4 has the lowest solubility (Weisemburg and Guinasso, 1979), bubble formation had to be triggered. Once formed, the bubbles can only migrate upward due to their buoyancy until they are blocked under the impermeable layer or released into the atmosphere. In contrast, the dissolved component in the brines can diffuse to the underlying seawater. Hence, bubble formation leads to an accumulation of CH4 inside the ice cover. The upward migration of the bubbles could partially explain the preferential bubble accumulation in the top 20 cm of the ice column, a process recently discussed for the Ar in Zhou et al. (2013) and modeled by Moreau et al. (2014).

The pCO2 of the water column was slightly undersaturated (354–384 ppmv) compared to the atmosphere (390 ppmv), whereas at the ice–water interface the pCO2 was largely undersaturated (77–130 ppmv). As noted above, the thin surface layer was under the influence of mixing processes, which could have produced a reduction of the water pCO2 at the ice–water interface. Moreover, the annual monitoring measurements provided by the Nuuk Ecological Research Operations observed year-round pCO2 undersaturated surface water in the Godthåbsfjord system (Rysgaard et al., 2012).

The range of 77–330 ppmv and average of 194 ppmv for the bulk ice pCO2 is in the same range as the profiles of bulk sea-ice pCO2 measured on natural and experimental sea ice (Geilfus et al., 2012a). According to Geilfus et al., (2012a, b) and Delille et al. (2007) the bulk ice pCO2 is undersaturated once ice temperature is above -4 ∘C. One of the main factors controlling the inorganic carbon dynamics within sea ice appears to be temperature. As temperature increases, the subsequent decrease of the salinity promotes the brine dilution and a decrease of the brine pCO2. According to our data, the bulk ice pCO2 was inversely correlated to the ice temperature. Indeed, the high values of bulk ice pCO2 (up 330 ppmv) were associated with the coldest temperature (-3 to -4 ∘C), and as the temperature increased with the ice thickness, the bulk ice pCO2 decreased to ≈ 100 -190 ppmv and reached the same range of pCO2 concentration as the underlying seawater, ≈ 76–130 ppmv.

Other processes affect the pCO2 concentrations within sea ice such as the precipitation and dissolution of calcium carbonate (Dieckmann et al., 2008; Geilfus et al., 2012b, 2013a,b; Papadimitriou et al. 2007, 2012; Rysgaard et al., 2007, 2011, 2013). During the sea-ice melt, the carbonate dissolution promotes lower pCO2 conditions (Rysgaard et al., 2011). Søgaard et al. (2013) suggested that the main factor controlling the total alkalinity and the dissolved inorganic carbon of the sea ice of Kapisillit was the dissolution of calcium carbonate crystals. Hence, dissolution of calcium carbonates, associated with the high temperature and the subsequent decrease of salinity was likely the reason for the observed low bulk ice pCO2 during our study also.

Finally, primary production as well bacterial respiration can affect the inorganic carbon dynamics within sea ice (Delille et al., 2007; Kaartokalio et al., 2013; Papadimitriou et al. 2012; Rysgaard et al., 2007, 2009; Søgaard et al., 2013). Ice melting through spring and summer will produce both a continuous reduction of the bulk ice pCO2 and an increase of the ice permeable features leading to pCO2 sub-saturated sea ice relative to the atmosphere and hereby enhance the air–sea flux of CO2.