Environmental physical and chemical context

The 26 day experiment encompassed the period of transition from an early spring cold snowpack through to a period of snowmelt and melt-pond formation. Air temperature was highly variable throughout the study (Fig. 2), and the first positive air temperature was recorded on 2 June (DOY 153). the temperature at the ice–snow interface ranged from –9.8^˚C at the onset of the experiment to a maximum value of –3.2^˚C near the end of the experiment. the ice surface warmed rapidly between 10 and 19 May (DOY 130–139) and after 26 May (DOY 148). Storms centered on 11 May (DOY 131), 21 May (DOY 141) and 25–30 May (DOY 145– 150) were associated with high surface winds. Gaps in the time series correspond to periods of particularly harsh weather, including blizzard and/or ice-riming conditions.

Fig. 2.

Time series of temperature and wind speed at the sampling site. (a) Hourly temperatures: air (T _a) in black; snow–ice interface (T _i) in grey. (b) Hourly wind speed (u).

The ice surface and snow-base environments were both saline (Fig. 3a). Upwardly decreasing salinity gradients in snow over seasonal sea ice are common (e.g. Reference Barber, Papakyriakou, LeDrew and ShokrBarber and others, 1995; Reference Langlois, Mundy and BarberLanglois and others, 2007). Salinity at the ice surface and 2 cm into the snowpack varied between 5.9 and

Fig. 3.

Time series of: (a) salinity at the snow base (ice surface) and 2 and 4 cm up into the snowpack; (b) brine salinity for temperatures measured at the snow–ice interface; (c) pCO₂ in the air (open diamond) and in surface sea water (0.5m under the sea ice; solid diamond).

15.3 over our experiment and decreased sharply with distance away from the snow base. At 4 cm from the snow base, the salinity was on average 3.5 less than at the snow– ice interface, while at 6 cm from the snow base the salinity ranged from 0 to 3 (not shown). Snow salinity decreased significantly after 25 May (DOY 145) at 4 cm up from the snow base and 30 May (DOY 150) at the ice surface, presumably the result of snowmelt. Ice salinity profiles showed a weak ‘C-shape’ distribution (not shown), averaging 7.5±0.8 in the upper 15cm of the cores and 5.1±0.6 between 30 and 120 cm. the brine salinity at the ice surface (calculated from measured Otemperatures) decreased over the experiment from 140 to 70 (Fig. 3b) as a result of dilution by melting. Surface sea-water salinity below the ice fluctuated between 32.65 and 32.74, with no obvious trend (not shown). Sea-water pCO₂ (as calculated from TIC and A _T) declined over the period 18 May–4 June (DOY 138–155), yet remained supersaturated relative to atmospheric levels until late in the experiment (Fig. 3c).

Springtime air–surface CO₂ exchange

Figure 4 shows the hourly and daily average air–surface CO₂ exchange. the corrected flux tended to show a low-level CO₂ efflux (averaging 0.36 μmolm^–2 s^–1) prior to 20 May (DOY 140), followed by periods of moderate to strong CO₂ uptake. Maximum hourly efflux exceeded 1.0 μmolm^–2 s^–1, while hourly uptake approached –3.0 μmolm^–2 s^–1 on 30 May (DOY 150).

Fig. 4.

Running mean (5 hour window) of hourly CO₂ flux (FCO₂). Fluxes corrected for sensor heating are shown in black, while uncorrected fluxes appear as grey. A negative flux is directed downwards, toward the snow surface.

The maximum uptake rates in Figure 4 are in excess of any other values reported over sea ice to date (Reference Semiletov, Makshtas, Akasofu and AndreasSemiletov and others, 2004; Reference Zemmelink, Delille, Tison, Hintsa, Houghton and DaceyZemmelink and others, 2006; Reference MillerMiller and others, in press) and are two to four times larger than hourly fluxes observed over the North Atlantic (Reference McGillis, Edson, Hare and FairallMcGillis and others, 2001). the fluxes observed by Reference Semiletov, Makshtas, Akasofu and AndreasSemiletov and others (2004) approached –0.8 μmolm^–2 s^–1 over first-year sea ice near Barrow, Alaska, while Reference Zemmelink, Delille, Tison, Hintsa, Houghton and DaceyZemmelink and others (2006) reported fluxes up to –0.3 μmolm^–2 s^–1 over ablating Antarctic multi-year ice. In both of these studies, the CO₂ fluxes were measured using eddy covariance systems with open-path CO₂ sensors, consistent with this study, but without the correction of Reference Burba, McDermitt, Grelle, Anderson and XuBurba and others (2008). Wintertime CO₂ fluxes reported by Reference MillerMiller and others (in press) over fast ice in the western Canadian Arctic showed considerable variability and were also during isolated periods quite large. the fluxes were generally directed upward (median flux of 0.14 μmolm^–2 s^–1). In that work, we speculated that the exchange was driven by some combination of sea-ice brine carbonate chemistry and biological activity. By comparison, wintertime hourly efflux in cold terrestrial systems can range anywhere between 0.25 μmol m^–2 s^–1 for wetlands (e.g. Reference Lafleur, Routet, Bubier, Frolking and MooreLafleur and others, 2003) and approximately 0.45 μmolm^–2 s^–1 for upland boreal forest soils in Alaska (Reference Sullivan, Arens and SveinbjSullivan and others, 2008).

Fick’s law for gas diffusion dictates that we would require a C₂concentration at the snow base of approximately 38 mmolm^–3, roughly twice atmospheric levels, to sustain our average early experiment diffusive CO₂ efflux (0.36 μmolm^–2 s^–1) through the snow, assuming a snow depth of 0.3 m and diffusivity of 0.05 cm² s^–1. Diffusivity for gas transport in wind-packed snow can range between 0.02 and 0.08 cm² s^–1 (Reference Albert and ShultzAlbert and Shultze, 2002). an upward flux of 0.36 μmolm^–2 s^–1 is not inconsistent with field observations elsewhere, which indicate that sea-ice pCO² in the spring can exceed atmospheric values (Reference Delille, Jourdain, Borges, Tison and DelilleDelille and others, 2007) and perhaps significantly so (Reference MillerMiller and others, in press). In laboratory experiments, Reference Nomura and ToyotaNomura and others, (2006) observed low-level CO₂ efflux from growing sea ice at a rate (0.004–0.012 μ^molCO2m^–2 s^–1) that increased with increasing brine salinity (decreasing brine temperature). Ice temperature in their experiments ranged between –2.7^˚C and –4.1^˚C, which is warmer than our near-surface sea ice for all but the final days of our experiment (Fig. 2). Consequently, brine salinity in their study was lower by almost a factor of 2, which could partly explain why our observed average efflux was higher.

As the under-ice sea water was supersaturated in pCO₂ relative to atmospheric values for most of our study (Fig. 3c), the prevailing ocean–air pCO₂ gradient could also support an upward flux of CO₂. Through much of the study, the ice in McDougall Sound was likely impermeable to brine because the brine volume constituted <5% (the cut-off for brine permeability in columnar sea ice; Reference Golden, Ackley and LytleGolden and others, 1998) of the sea-ice interior prior to 26 May (DOY 146) (Fig. 5) and hence CO₂ probably could not have been transported by the brine. However, little is known about the comparable temperature cut-off for gas diffusion in sea ice, aside from the laboratory study of Reference Gosink, Pearson and KelleyGosink and others (1976) who observed substantial transport of halogenated gases at temperatures down to –15^˚C. the results from Reference Gosink, Pearson and KelleyGosink and others (1976) imply that CO₂ could have been moving through the ice throughout our study, something that we cannot confirm.

Fig. 5.

The evolution of the 1200 h (local time: North American Central Standard Time) temperature and brine volumetric fraction in the upper meter of sea ice.

Spring heating raised the ice temperature above –5^˚C by 28 May (DOY 148), with a corresponding increase in sea-ice brine volume above 5% (Fig. 5). Beyond this date, the brine should have been draining freely and mixing with the underlying sea water (Reference Golden, Ackley and LytleGolden and others, 1998; Reference TisonTison and others, 2008). the timing of this transition closely corresponded to the period of maximum CO₂ uptake observed between 28 and 30 May (DOY 148–150; Fig. 4), meaning that atmospheric CO₂ could have offset part of any pCO₂ deficit developing at the ice–water interface as a result of algal production or melt dilution. In regard to the former, sea ice in the Canadian Arctic Archipelago ranks amongst the most productive in the Arctic, with reported primary production rates between 20 and 463 mgCm^–2 d^–1 (Reference Arrigo, Mock, Lizotte, Thomas and DieckmannArrigo and others, 2010). Chl a concentrations within the bottom 5 cm of the ice cover close to our research site (Reference Mundy, Barber and MichelMundy and others, 2005) were between 20 and 45 mgm^–2 by 29 May (DOY 149). Data shown by Reference Lavoie, Denman and MichelLavoie and others (2005) indicate that pronounced ice algal accumulation in 2002 (commensurate with our study) had started at about 10 May (DOY 130) and persisted into the first week of June (approximately DOY 160) in sea ice approximately 70 km southeast of our study site and of comparable thickness and snow depth. Maximum Chl a concentration ranged between 55 and 80 mgm^–2 at their site, with net particulate organic carbon (POC) accumulation rates ranging between 28 and 55 mg Cm^–2 d^–1, depending on snow thickness. Algae are the main contributors of POC at the base of Arctic first-year sea ice (Reference Riedel, Michel, Gosselin and LeBlancRiedel and others, 2008). Accumulation rates were obtained by applying a POC:Chl a ratio of 20 (Reference Riedel, Michel, Gosselin and LeBlancRiedel and others, 2008) to Chl a time rate-of-change estimates obtained graphically from figures provided by Reference Lavoie, Denman and MichelLavoie and others (2005), over the period between 10 May (DOY 130) and 25 May or 10 June (DOY 145 or 161) for thick and thin snow, respectively. Uptake, according to our tower measurements, averaged approximately 23 mgCm^–2 d^–1 between 10 and 30 May (DOY 130–150). This value is close in magnitude to the POC accumulation rates that we derived from Reference Lavoie, Denman and MichelLavoie and others (2005) and on the low end of primary production rates reported for sea ice in the Canadian Arctic Archipelago. Although we expect the ocean mixed layer to be the main supplier of nutrients to the skeletal layer at the ice base (Reference Thomas, Papadimitriou, Michel, Thomas and DieckmannThomas and others, 2010, and references therein), it is not inconceivable for atmospheric CO₂, via brine drainage, to offset short-term CO₂ deficits at the ice base given that: (1) both TIC and CO₂(aq) in sea-ice brine can be depleted under conditions of high primary production and large accumulation of algal biomass (Reference Thomas, Papadimitriou, Michel, Thomas and DieckmannThomas and others, 2010, and references therein); (2) the nutrient supply from the mixed layer to biomass at the ice base can be restricted by the thickness of the molecular sublayer (Reference Lavoie, Denman and MichelLavoie and others, 2005); and (3) downward flushing of nutrients into sea ice associated with snowmelt has been documented and thought able to augment sea-ice biomass accumulation (e.g. Reference Granskog, Kaartokallio and ShirasawaGranskog and others, 2003).

Finally, if CaCO₃ had precipitated in the ice when it formed, the late-spring melting process would have dissolved the salts, lowering pCO₂ within the ice and further encouraging CO₂ drawdown from the atmosphere. Collectively, these observations support the argument that the observed late-season CO₂ uptake was a result of a permeable ice cover and a lowering of pCO₂ in the ice and sea water through photosynthesis, melt and possibly calcium carbonate dissolution.

The magnitude of fluxes reported here raises concerns about possible measurement biases associated with the open-path system. It is evident (Fig. 4) that prior to 20 May (DOY 140) the heating correction constituted a large component of the corrected flux, often changing the flux direction from predominately a regime of low-level uptake to low-level efflux. Accordingly, we are less confident of measurements during this period given the possibility that

the flux has been overcorrected (Reference JärviJärvi and others, 2009) and hence is artificially high. With this uncertainty a validation of the open-path eddy covariance system, and associated corrections, is required over sea ice to confirm that low-level fluxes of this sort are in fact above system noise. However, the heating correction to the CO₂ flux represents only a small component of the strong midday uptake that occurred on 21, 28 and 30 May and 3 June (DOY 141, 148, 150 and 154). We are also reasonably confident that the measured uptake at these times is not the result of an obstruction of the open-path lens (Serrano-Orti and others, 2008) or cross-sensitivity to the water vapour flux (Pyrtherch and others, 2010). In regard to the former, we do observe a decrease in atmospheric CO₂ concentration of 4–5% corresponding to CO₂ uptake on 28 and 30 May (DOY 148 and 150) (Fig. 6a), which if related to lens obstruction would artificially lower our flux measurement by as much as 30% through a reduction in the WPL correction (Serrano-Orti and others 2008). the sensor lens was checked for cleanliness daily, and diurnal fluctuations in ambient CO₂ of >10% are not uncommon over ecosystems with strong 24 hour cycles of net photosynthesis (Reference Anthoni, Law and UnsworthAnthoni and others, 1999; Reference Yang, Black, Neumann, Novak and BlankenYang and others, 1999), hence we expect our measurements over these periods to be within the sensor’s specifications. High-frequency measurements of CO₂ and H₂O concentration are not available from this study to explore possible sensor cross-sensitivity, but data from subsequent experiments over winter- or springtime sea ice show no coherent relationship between high-frequency measurements of CO₂ mixing ratio and relative humidity. Further, our measured latent heat flux (±7Wm^–2) between 26 and 31 May (DOY 146–151; Fig. 6b) is close to a factor of 10 smaller than reported in Pyrtherch and others (2010) over open water, indicating that if there was cross-sensitivity in the gas concentration measurements, it was not strongly correlated to vertical velocity or, hence, the CO₂ flux.

Fig. 6.

The time series of average (a) atmospheric pCO₂ and (b) latent heat flux (R _E) between 26 and 31 May (DOY 146–151). the flux of CO₂ (FCO₂) is provided in both plots for reference.

Diurnal variation in air–surface CO₂ exchange

Prominent shifts from efflux to uptake were observed several times, including on 10 May (DOY 130), 21–22 May (DOY 141–142) and then 26, 28 and 30 May and 2 June (DOY 146, 148, 150 and 155) (Fig. 4). Two periods with large CO₂ flux (FCO₂) ranges are shown in Figure 7a, showing the measured flux between 20 and 24 May (DOY 140–142) and between 26 and 30 May (DOY 146–150). Also shown are the corresponding variations in surface net radiation (R _n, Fig. 7b), ice surface temperature (T _i, Fig. 7c), horizontal wind speed (u, Fig. 7d) and air temperature (T _a, Fig. 7e). the interface temperature reflects the thermodynamic state of the surface. Net radiation provides a measure of heating and cooling of the surface via radiation, while horizontal wind speed is related to surface shear and forced convection.

Fig. 7.

Hourly variation in (a) CO₂ flux (FCO₂), (b) surface net radiation (R _n), (c) snow–ice interface temperature (T _i), (d) wind speed (u) and (e) air temperature (T _a) for the periods 20–23 May (DOY 140–143) and 26–30 May (DOY 146–150). No data are available from 29 May (DOY 149).

Uptake on 21 May (DOY 141) started during the night of 20 May (DOY 140) and increased over the course of the morning and afternoon of 21 May (Fig. 7a). the uptake occurred during the first high wind event following the intense heating of the snowpack on 19 May (DOY 139) (Fig. 2). the start of the CO₂ uptake during the evening of 20 May (DOY 140) coincided with the rise in wind speed (Fig. 7d), and uptake stopped when the wind speed dropped again below 6 ms^–1 in the early morning of 22 May (DOY 142). the Gaussian-shaped timeline of hourly R _n centered midday on 21 May (DOY 141; Fig. 7b) indicates clear-sky conditions, which allowed the surface to cool intensely over much of the night. Peak uptake on 21 May (DOY 141) also corresponded in time to the 24 hour maximum in R _n and a rise in T _i. the efflux observed at about midnight at the start of 22 May (DOY 142) corresponded to the 24 hour minimum in R _n.

Strong CO₂ uptake was also observed on 26, 28 and 30 May (DOY 146, 148 and 150), with efflux occurring just about midnight on 27 and 28 May (DOY 147 and 148). the periods of efflux corresponded to night-time minima in R _n, air temperature and cooling at the snow base, again associated with clear-sky conditions. Over this period, the hourly pattern in FCO₂ mirrored wind speed. High winds (generally >6ms^–1) gave rise to large CO₂ uptake rates, and low winds corresponded to periods of little exchange or efflux. the largest uptake rate observed over the entire experiment occurred on 30 May (DOY 150) under conditions of extremely high wind speed (>9ms^–1) and high interface (greater than –5.5^˚C) and air (approximately –5^˚C) temperatures.

We conclude that, in general, CO₂ uptake and efflux mirrored wind speed, with winds in excess of approximately 5–6ms^–1 supporting uptake, while lighter winds were associated with efflux. the relationship between wind speed and FCO₂ is nonlinear in nature, as evident in Figure 8 for the period 26–30 May (DOY 146–150). High winds ventilate the snow (Reference ColbeckColbeck, 1989; Reference Albert and ShultzAlbert and Shultz, 2002) and above a variable threshold (4.4–11.0ms^–1) give rise to blowing snow (Reference Li and PomeroyLi and Pomeroy, 1997). Snow ventilation allows exchange between the atmosphere and saline snow base/ice surface. In addition, blowing snow has a high surface area:mass ratio and is subject to intense atmospheric turbulence, hence the potential for turbulent transfer of trace gases (in addition to heat) to and from the snow grains (Reference Pomeroy, Davies, Jones, Marsh, Peters and TranterPomeroy and others, 1999) or, in our case, brine-wetted snow grains. the efficiency of gas adsorption/desorption on ice grains will be proportional to the changing snow surface area (Reference Legagneux, Cabanes and DominéLegagneux and others, 2002). Hence, blowing snow during conditions conducive to down-directed sensible heating (i.e. a warming atmosphere) allows for rapid heating of the brine-wetted snow base and ice surface by convection in the snow. on the other hand, in the absence of high winds, CO₂ exchange with the snow base and ice surface is controlled by diffusion, and the upper snow may moderate the exchange by storing air (and CO₂).

Fig. 8.

Relationships between the measured CO₂ flux and wind speed between 26 and 30 May (DOY 146–150). the trend line fit to the data is a second-order polynomial.

Uptake also corresponded to local peaks in both net radiation and temperature, while efflux generally occurred in association with local minima in net radiation and temperature, which occurred largely at night. Previous studies have shown that the dielectric properties of brine-wetted snow and sea ice are quite sensitive to incident radiation (long- and shortwave) through changing brine volume at the snow base (Reference Barber, Papakyriakou, LeDrew and ShokrBarber and others, 1995; Reference Barber and ThomasBarber and Thomas, 1998; Reference Langlois, Fisico, Barber and PapakyriakouLanglois and others, 2008). Reduced radiative cooling (increased net radiation) allows the snow/ ice system to retain a greater proportion of energy available from both oceanic and atmospheric sensible heat, thereby supporting warming. the opposite occurs during periods of pronounced radiation loss.

Our data indicate that the diurnal fluctuations in meteorological conditions can largely explain the efflux and uptake sequences we observed. Nocturnal cooling and freezing of the brine at the snow base and ice surface degasses CO₂ through a reduced solubility in the brine (Reference Killawee, Fairchild, Tison, Janssens and LorrainKillawee and others, 1998; Reference MockMock and others, 2002; Reference Tison, Haas, Gowing, Sleewaegen and BernardTison and others, 2002) and possibly precipitation of calcium carbonate (Reference Papadimitriou, Kennedy, Dieckmann and ThomasPapadimitriou and others, 2003). It follows that the midday uptake often observed with warming was associated with any combination of: increasing gas solubility in the brines in the upper sea ice and snow base; strong air-to-brine pCO₂ gradients associated with nocturnal degassing and CO₂ depletion in the brine (Reference Delille, Jourdain, Borges, Tison and DelilleDelille and others, 2007; Reference Rysgaard, Glud, Sejr, Bendtsen and ChristensenRysgaard and others, 2007); and possibly dissolution of calcium carbonate (Reference Papadimitriou, Kennedy, Dieckmann and ThomasPapadimitriou and others, 2003). the pattern of daytime uptake and nocturnal efflux observed here is indicative of a biologically controlled system of diurnal photosynthesis and respiration. Although we are unable to explore relationships to biology with this dataset, it is intuitive that net photosynthesis could contribute to our observed daytime CO₂ uptake, given the dependency of photosynthesis on photosynthetically active radiation.