Gas exchange through sea ice is a determining factor in the polar ocean budget of climatically-active gases. We use SF6 and O2 as conservative gas tracers to observe transport between the water, ice and air during conditions of freezing and partial ice cover in artificial seawater. During ice growth, O2 and SF6, as non-polar solutes, were rejected from the ice into the underlying water at a faster rate than that observed for salt. Measurements of the gas exchange rate, k, through partial ice cover exceeded that expected from linear scaling between 100% open water (k100%) and complete ice cover: at 15% open water, k was 25% of k100%. These results indicate that the net flux of gas through the ice pack may not scale linearly with open water area, as circulation processes under the ice affect the gas exchange rate.
 During freezing of seawater dissolved solutes (ions and gases) are excluded from the ice crystal structure [Richardson, 1976], producing a cascade of processes that affect the circulation and the biogeochemistry of the polar oceans [Arrigo and Van Dijken, 2007]. The exclusion of salt produces convection beneath sea ice to such an extent that the mixed layer depth is determined by the amount of sea ice growth during austral winter [Martinson, 1990]. Hypersaline brine carves a porous structure within sea ice that is permeable to both water and gas [Golden et al., 2007; Gosink et al., 1976]. As the sea ice temperature decreases, brine becomes further concentrated and carbonate minerals, such as ikaite (CaCO3 · 6H2O), can precipitate from the solution [Dieckmann et al., 2008; Marion, 2001] decreasing carbonate alkalinity. It has been proposed that the loss of carbonate alkalinity can enhance the partial pressure of CO2 (pCO2) within the sea ice matrix [Papadimitriou et al., 2003]. These processes, as well as diffusion and air-sea gas transfer, may result in a net redistribution of inorganic carbon from the surface ocean to the atmosphere and to the ocean interior [Nomura et al., 2006; Rysgaard et al., 2007].
 This study attempts to isolate the physical processes that can produce mass transport from the biogeochemistry of seawater freezing. The temporal evolution of two aqueous gas tracers, SF6 and O2, are measured in comparison with the salt budget under abiotic, conservative, windless conditions. We examine the influence of unconsolidated ice crystal formation and the growth of columnar sea ice. Results are presented indicating that gas is rejected more efficiently than salt from the freezing interface, possibly by differential diffusion rates. We present measurements of gas exchange that show an enhancement in the gas exchange rate through partial ice cover for both O2 and SF6. Finally, considering the mechanisms of turbulence production beneath sea ice, we attempt to connect these results with the variation in gas exchange produced by variable turbulence.
 A total of six freezing experiments were conducted in a variable temperature cold room at the US Army Corps of Engineers Cold Regions Research and Engineering Laboratory (CRREL) in Hanover, NH. Two gas exchange tanks of different volume were used: a 144 L polypropylene drum (d × h = 49.2 × 75 cm) with a lid and rubber gasket, and a 55 L PVC tank with a lid and O-ring seal (Figure 1). In the 144 L tank, salinity, temperature and pressure were measured using a Seabird SB37SIP CTD probe. Dissolved O2 concentration was measured using an Aandera Instruments 3830 Optode. Three Rule 360 bilge pumps were used to mix the tank and to circulate water through a length of Tygon tubing for discrete SF6 sample collection. A similar configuration was used within the 55 L tank.
 Before each experiment, artificial seawater was prepared by mixing MilliQ water with sea salt for aquariums (Instant Ocean, Spectrum Brands) to obtain salinity between 33 and 35 psu, as verified with a conductivity probe. Prior to introducing the water mixture, the experiment tanks, instruments and tubing were cleaned with 10% bleach solution to minimize the growth of bacteria. To initiate an experiment, the cold room was set to a temperature between −5 and −20°C, and the seawater bath was allowed to cool (see Table S1 of the auxiliary material for more detail on each experiment). As the water temperature decreased below 5°C, the bath was bubbled with air, for O2 evasion, or with N2 for O2 invasion. Prior to achieving 0°C, aeration was suspended to make estimates of k, before the onset of freezing. Subsequently, ca. 10−7 moles of SF6 were added as a concentrated solution dissolved in 5 ml of freshwater, and the solution was allowed to mix entirely. Water samples for SF6 were collected with 50 ml glass syringes and needles. The gases were separated using the headspace method [Wanninkhof et al., 1987] and analyzed by gas chromatography as described by Ho et al. .
 Gas exchange was calculated using the control volume approach. Before each experiment the water was conditioned to enhance the partial pressure gradient between water and air. For SF6, the gas evasion method was used [Ho et al., 1997]. For O2, both evasion and invasion were used during successive experiments. For a cylindrical tank of depth (h), the gas exchange rate (k) can be estimated from the initial and final concentration (Ci and Cf) as, k = h(Δt)−1ln[(Ci − CEQ)/(Cf − CEQ)]. CEQ = ßX, where ß is the Bunsen solubility coefficient (volume of gas/volume of water) and X is the volume of pure gas per volume of air.
3. Results and Discussion
3.1. Onset of Freezing
 The onset of freezing is evident in both the temperature and O2 concentration. In the 55 L tank (Figure 2, Expt. 2 and 6) the temperature decreased below the expected freezing point (dashed line) by 0.08 to 0.14 °C, before exhibiting an increase in temperature, followed by an increase in O2 concentration, which coincided with the rapid appearance of unconsolidated ice crystals in the upper 2–3 cm of the tank. As freezing progressed during Expt. 2, the aqueous O2 concentration increased, reflecting gas and solute rejection. The slower cooling rate during Expt. 6 caused partial ice cover to persist, following the first onset of freezing, so that the aqueous O2 concentration continued to decrease beyond 0.1 days.
3.2. Columnar Ice Growth and Solute Rejection
 The response of salinity and O2 to freezing can be compared by normalizing the change of each solute measured in the water, during an equivalent time interval: d2 = dO2/O2 and d = dS/S. During the growth of consolidated ice (Figure 2b, t ≤ 1 days), d2 and d converge and become nearly identical as ice growth tapers off, near t = 1.1 days. Throughout the period of ice growth, the magnitude of d2 exceeded d. The same holds true for the period of relatively constant ice thickness (t = 3 – 5 days). Whereas in the absence of ice growth or melt, d2 and d should be almost zero, they are slightly positive indicating residual brine drainage. To compare d2 and d during successive experiments, we fit a linear trend line to the normalized salinity and O2 time series, during a period of constant ice growth. The magnitude of the slope reflects the normalized rate of change of each parameter. The observations indicate that d2 exceeded d by 20 – 80%. The same effect was observed when comparing the normalized increase in SF6 during rapid ice growth. The values of d2, d, and d, and the R2 coefficients of each fit are listed in Table 1.
Table 1. Comparison of the Normalized Change in Salinity (d) With the Normalized Changes in O2 (d2) and SF6 (d6), Measured Beneath the Ice During Conditions of Constant, Rapid Ice Growtha
d, d2 and d6 reflect the change during a period of one day. The ratios of d2:d and d6:d greater than 1 indicate that the dissolved gases increased in concentration more rapidly than salinity.
Temp 5 cm above ice (°C)
Growth rate (cm s−1)
1.53 × 10−5
2.56 × 10−5
6.47 × 10−5
dS, S (psu)
dO2, O2 (mmol L−1)
dSF6, SF6 (pmol L−1)
 To quantify the partitioning of gases between the water and ice, we compared the mass of solute rejected during an interval of ice growth (MR) to the total mass of solute that was initially in the volume of water that has been frozen (MICE). This can be calculated by differencing the solute concentration over a short period (SI, and SF), and by estimating the decrease in water volume. We define the term rejection ratio (R) as MR/MICE = VT(ρFSF − ρISI)/(VICEρISI). ρF and ρI are the initial and final density of the water, VT is the volume of the tank after freezing has taken place, and VICE is the water equivalent of the frozen ice. To minimize the effects of gas diffusion through the ice we restrict the calculation to a period less than two days. For constant growth of 4.5 cm of ice during Expt. 3 the salt rejected (MR) is 143 g. This value is within 12% of the estimated salt rejection using the empirical equations of Cox and Weeks  (126 g), who use air temperature and ice growth rate to calculate solute rejection and brine volume. This produces a rejection ratio for salt (RS) of 45%. The same relationship, applied to dissolved O2 results in a rejection ratio (RO2) of 61%. A similar calculation for relatively constant growth conditions during Expt. 5 produces values for RS of 40% and RO2 of 45%. The agreement between the measured mass of rejected salt (MR) and the result using the method by Cox and Weeks  was within 22%. The rate of ice growth is a strong determining factor for the rate of salt rejection, [Cox and Weeks, 1983] - the faster the growth rate the larger the amount of salt that remains included in the ice. During Expt. 3 the ice growth rate was 2.56 × 10−5 cm s−1, ∼40 % of the growth rate during Expt. 5 (6.47 × 10−5 cm s−1). These results indicate that 39 – 55 % of the O2 within the unfrozen water remained included within the sea ice, dissolved within the brine or nucleated into gas bubbles [Craig et al., 1992].
3.3. Gas Exchange Rate During Partial Ice Cover
 During Expts. 1, 3, 4 and 5 in the 144 L tank, ice was removed from the surface by drilling holes to measure the rate of gas exchange during partial ice cover. Once a constant fraction of open water was established, the rate of gas exchange was measured for a period of 8 – 20 hours, using the control volume approach as described in the methods section [Ho et al., 1997]. During these periods the ice thickness remained constant, and residual brine drainage was minimal. Because the O2 concentration in brine was not measured, a correction for brine drainage was not attempted. However the increase in salinity was less than 0.5% during each measurement period. If the O2 in brine is proportional to salinity, then brine expulsion would produce a source of O2 that is less than 5% of the change in O2 observed during each estimate period, resulting in a 7% difference in k. This 7% bias has been added to the standard error in k, which ranges between 10 and 80%, as indicated by the error bars in Figure 3. Similarly, the change in water volume produced by brine drainage accounts for ∼0.1 – 0.3% bias in k.
 To evaluate the tendency of gas exchange rate with an increasing fraction of open water, the measurements of kO2 and kSF6, normalized by the gas exchange rate at 100% open water (k100%(O2) = 1.02 and k100%(SF6) = 0.83 cm h−1), are plotted versus the fraction of open water surface in Figure 3 (see Figure S1 of the auxiliary material for more detail on k100%(O2) and gas exchange prior to freezing). The expectation is for k/k100% to range proportionally between k/k100% = 0 (or the rate of gas diffusion through ice) when there is no open water, and k/k100% = 1 at 100% open water. Without exception, the measurements are clustered above the 1:1 line, exhibiting a tendency that is in excess of the proportional scale. At 15% ice cover, k is 25% of k100%.
 The existence of a lead, or opening in the ice pack, can establish several modes of circulation in the water column underneath the ice. In the limit of zero horizontal current motion between the sea ice and the water column, the production of dense water at the ice opening induces free convection that draws water from beneath the adjacent ice in a cellular circulation pattern [Morison et al., 1992]. At the other extreme, motion of the water relative to the ice produces boundary layer turbulence that entrains dense water from the ice opening into the lateral turbulent flow. Morison et al.  define the turbulent lead number, or ratio of the buoyancy production to shear production beneath the lead to characterize this spectrum of flow conditions. A lead number less than one describes a condition dominated by boundary layer turbulence, while a lead number greater than one indicates free convection. In both extremes, the existence of lateral flow towards the opening can affect a relatively large heat and mass transfer within the water column relative to the size of the lead opening.
 During these measurements of gas exchange under partial ice cover, turbulence was produced via mixing from the submersible pumps, as well as brine-driven buoyant convection (Figure 2) and heat-loss at the opening. To measure the turbulent mixing during partial ice cover, an Acoustic Doppler Velocimeter (ADV) was submerged 15 cm beneath the ice surface, during one experiment and turbulence intensity was varied by decreasing the submersible pump voltage during conditions of 26% open water fraction. The ADV was set to sample at 25 Hz [Zappa et al., 2003] to compute the scalar quantities, q, proportional to the turbulent kinetic energy and u*, proportional to the Reynolds stress as q = (〈u′2〉 + 〈v′2〉 + 〈w′2〉)1/2 and u* = (〈u′w′〉2 + 〈v′w′〉2)1/4 [McPhee, 1992]. Decreasing the pump voltage from 12 to 6 Volts produced a 10% decrease in q, from 4.18 to 3.75 cm s−1, and a ca. 14% decrease in k/k100% of O2, from 0.61 to 0.53 (Figure 3). Using the range of u* measured with the ADV (0.4 – 0.8 cm s−1), the turbulent lead number ranged from 0.5 to 0.15, i.e., significantly less than one, indicating that shear production from the submersible pumps dominated buoyant convection. There are few field measurements of turbulence beneath winter sea ice in the presence of leads, but McPhee  reported measurements of q from 2.63 to 2.84 cm s−1 and u* from 0.9 to 1.26 cm s−1 under drifting pack ice in the Arctic. These values are similar to the turbulence measured during the partial cover conditions of this experiment - the TKE in the tank was ca. 30% greater, and the estimate of friction velocity was smaller by 80% than the field measurements of McPhee .
 Several of the complex physical processes that take place during the freezing of saltwater have been investigated using O2 and SF6 as conservative dissolved gas tracers of mass transport. Measurements of the gas exchange rate, k, during conditions of fractional ice cover indicate that gas flux through the ice pack may not scale linearly with the open water area. In these experiments turbulence beneath the ice was produced using submersible pumps, and a decrease in mixing by reduced pump power produced a decrease in TKE and a correspondent decrease in k, as expected. While the estimates of TKE in the tank were of similar magnitude to the measurements of TKE in the field beneath drifting ice by McPhee , we don't expect that these experiments represent the range of turbulent conditions and processes beneath sea ice nor the magnitude of gas transfer through the ice pack. Instead, they indicate that an estimate of the effective gas flux through sea ice should account for turbulence and circulation processes under the ice. The limiting conditions of strong boundary layer flow and free convection that Morison et al.  use to describe circulation beneath leads depict a renewal of water exposed to the opening in the ice, so that heat and mass transport through a lead may produce a larger “footprint” than is indicated by the surface area of open water.
 From the perspective of the dissolved gases in the surface layer of the ocean, the comparison of O2 and SF6 with salinity indicates that gas solutes are rejected from the ice more effectively than salt. The greater rejection of gases may be the result of faster gas diffusion away from the freezing interface into the region of water convection. Indeed, Killawee et al.  observed a diffusive boundary layer for ionic solutes beneath growing ice. Comparing the rejection of solutes in this study with a similar laboratory study involving CO2 during freezing, high rejection efficiency is also observed for CO2 and for total inorganic carbon (DIC) [Nomura et al., 2006]. Those authors report that 66.9% of DIC in seawater that became sea ice was rejected to the underlying water volume, while comparatively little CO2 was transferred from the ice to the air (0.8% of the initial CO2 in frozen seawater). The combined effect of these processes, as well as the biological processes that take place within the ice [Delille et al., 2007], determine the seasonal redistribution of DIC from the surface ocean, that occurs as a result of sea ice advance and retreat. These experiments offer some insight into the physical processes acting on dissolved gases during ice formation and deformation; the magnitude of these effects is an open question.
 We would like to thank Jackie Richter-Menge, Bruce Elder, Chris Polashenski, David Cole and David Ringelberg for their generous support of this work at CRREL. Bill Smethie, David Ho, and Eugene Gorman contributed to the design of the SF6 sampling and analysis procedures. Chris Zappa supported the ADV component of the study. Support for this work was provided by the Climate Center at the Lamont-Doherty Earth Observatory, an NSF IGERT Fellowship to BL, NSF grants OPP 01-25523/ANT 04-40825 (PS) and OCE 5-25297 (WM), and NOAA 5-62121 (WM). LDEO contribution 7234.