A winter time series of the inorganic carbon system above, within, and beneath the landfast sea ice of the southern Beaufort Sea confirmed that sea ice is an active participant in the carbon cycle of polar waters. Eddy covariance measurements above the ice identified significant vertical CO2 fluxes, mostly upward away from the ice but with short periods of downward fluxes as well. A novel, in situ method revealed extremely high pCO2 values within the ice that are not inconsistent with theory. The total carbon content of the ice increased slightly through the winter season, and increasing variability in the vertical profiles as spring began indicated that the inorganic carbon became mobile as the ice began to melt. During early winter, as the ice formed, inorganic carbon concentrations in the surface waters increased dramatically, along with salinity, partly because of rejection from the ice and partly from advective mixing. Brine drainage was apparently not sufficient to initiate convection, and the excess carbon remained in the surface waters into the summer.
Gosink et al.  were the first to report direct measurements of gas fluxes in sea ice. They found that halogenated gases can move through sea ice at rates as high as 60 cm h−1 at temperatures as low as −15°C, although their experiments with CO2 were limited to conditions with temperatures at −7°C. Recent field studies in the Arctic [Semiletov et al., 2004, 2007; Papakyriakou and Miller, 2011] and in the Antarctic [Delille, 2006] have further confirmed that sea ice is permeable to CO2. All of the studies published to date have been conducted in spring, and most of the observed fluxes have been downward into the ice, although some upward fluxes have also been noted. These observations have prompted questions about what drives the fluxes and what happens to the CO2 absorbed by the ice: is the carbon simply discharged back into the atmosphere when the ice melts [e.g., Miyake and Matsuo, 1963], or is it released into the ocean and ultimately exported into deep waters?
 The biogeochemical processes that ultimately control the fate of that carbon in sea ice are still poorly understood [Thomas et al., 2010]. A number of authors have noted that CaCO3 should be the first salt to precipitate from sea ice brines, at temperatures as high as −2.2°C [Thompson and Nelson, 1956; Assur, 1960], and CaCO3 precipitation in sea ice has recently been confirmed in the field [Delille, 2006; Dieckmann et al., 2008]. Thermal cycling within the ice would cause CaCO3 precipitation and dissolution, which could contribute to the CO2 fluxes observed above the ice. In addition, rejection of the resulting high-pCO2 brines from the ice could provide a mechanism for exporting carbon into deep waters [Lyakhin, 1970; Jones and Coote, 1981; Rysgaard et al., 2007]. Ice algae can also consume substantial quantities of carbon [e.g., Gosselin et al., 1997; Arrigo et al., 2010], but it is not clear whether or under what conditions the resulting organic matter is recycled in surface waters versus exported to depth [e.g., Michel et al., 1996; Rodriguez y Baena et al., 2008]. Bacteria are ubiquitous in sea ice, often occurring at high concentrations, and a vertical column of sea ice can be net heterotrophic (respiratory) due to the distribution of bacteria throughout the ice sheet, even if ice algae are consuming CO2 in the bottom layer [Deming, 2010].
 In an attempt to clarify the role sea ice may play in absorbing and exporting carbon in polar oceans, we collected an integrated data set of atmospheric boundary layer CO2 fluxes along with sea ice and water column carbon biogeochemistry from deep winter into spring at a single location on a shallow Arctic shelf covered by landfast ice. Whereas these data alone do not conclusively answer the questions raised above, they do reveal a surprisingly dynamic system.
2. Field Study and Methods
2.1. Site Description
 The Canadian Arctic Shelf Exchange Study (CASES) included an overwintering in Franklin Bay, on the western Canadian shelf (Figure 1). In December 2003, the NGCC Amundsen was frozen into a smooth pan of landfast ice at 70°02.73 N, 126°18.08 W, 18 km from shore, and remained at that location until 1 June 2004, when we broke out of the ice. We were able to collect data through the coldest phase and into the start of the spring melt period. During our stay, the sea ice grew from an initial thickness of 0.8 m to its maximum of 1.9 m. In general, the top 10 cm of the sea ice cover was composed of granular ice, followed by about a 10 cm thick transition zone to columnar ice. The bulk of the ice column was composed of columnar ice with a porous skeletal structure in the bottommost 1–2 cm, as is typical for landfast sea ice. Significant ice algae production at the bottom of the ice began at the beginning of April, and chlorophyll concentrations increased until June [Riedel et al., 2006].
 Meteorological measurements, as well as ice and surface water samples, were collected from various sites around the ship (Figure 1). The wind direction was variable but mostly WNW or ESE. Water currents at the site were generally toward the west or south, but short-timescale reversals were not uncommon [Kamali Nezhad et al., 2007].
2.2. Atmospheric Measurements
 Vertical CO2 fluxes above the ice were measured between 22 January and 25 May 2004 using the eddy covariance technique. The flux system, including a 3-D ultrasonic anemometer (CSAT3, Campbell Scientific®, Logan Utah) and an open path gas analyzer (LI-7500, LI-COR®, Lincoln Nebraska), was mounted facing 190° at a height of 4.35 m above the snow surface on a open TV antenna-type tower. Information on bulk meteorology, radiation fluxes, and snow and ice temperature was available from sensors mounted on and around this installation. Sensors were powered by a diesel generator that was situated 220 m directly to the north of the instrument tower. The tower was approximately 1400 m east-southeast of the ship (site F, Figure 1) and situated on an ice pan of uniform consolidation approximately 4 km in diameter. Sensors were inspected, and heights and snow depth were recorded, with few exceptions, on a daily basis. Turbulent fluctuations were scanned at 10 Hz using a data logger (model 23X, Campbell Scientific®, Logan Utah), while meteorological sensors, radiometers, and thermocouple-based thermometers were scanned at 5 s intervals by additional data loggers. The raw data were transmitted to the ship by radio frequency telemetry at 10 min intervals and stored on a personal computer for post processing.
 The means, variances, and covariances of the three-dimensional wind velocity, sonic temperature, and scalar concentrations of CO2 and H2O were calculated from the turbulent fluctuations during post processing. Density corrections included both a WPL (Webb, Pearmann, and Leuning) correction [after Massman and Lee, 2002], and a correction to compensate for air density fluctuations in the gas analyzer's optical path associated with heating of the sensor itself [Burba et al., 2008]. Sensor temperature was estimated from air temperature measured at the same height of the gas analyzer using the linear regression model from Burba et al.
 The resulting fluxes were screened on the basis of unfavorable wind direction, and periods not conducive to good flux measurements, including power interruptions and times when the sensors were covered with frost and rime. Data associated with wind directions between 270° and 300° (west-northwest), and between 350° to 10° (i.e., due north) were excluded from the analysis, to minimize possible impacts of the ship, the tower, and generator on computed fluxes. Diagnostic information from the ultrasonic anemometer and gas analyzer provided an objective means of screening the data for periods of frost.
 Uncertainties in eddy covariance estimates of CO2 flux measurements have been discussed by Moncrieff et al. , and accuracy estimates range from ±12% to 25% of daytime values. Burba et al.  found that winter open path CO2 flux measurements are within ±10% of closed path measurements when the effect of instrument heating on the ambient flux is corrected for using their empirical relationships, as we have done in this study. We have no independent measurement of CO2 flux against which to check our fluxes.
2.3. Ice Sampling
 Ice cores were generally collected weekly from site F, about 400 m north of the meteorological towers (Figure 1). The ice was cored using a 9 cm diameter corer (Mark II, Kovacks Enterprises, Inc.), and the cores were immediately laid out on an area of ice that had been cleared of snow. Using a clean, stainless steel saw, the core was quickly cut into 10 cm sections (starting from the top), which were placed into individual 1 L polycarbonate jars or doubled polyethylene resealable zipper plastic bags (LDPE, “freezer” grade; CO2 permeability estimated at 400–1100 cm3 mm (m2 d atm)−1 at 23°C [Massey, 2003]), and transported back to the ship as soon as possible. Immediately upon return to the ship, within 2 h of coring, the core sections were repackaged to minimize headspace during melting: each core section was tightly wrapped in a new, clean, plastic bag. Starting from 11 April, some bags were also quickly evacuated with a vacuum pump before sealing (the vacuum sealer did not work consistently). On later expeditions, we have used Tedlar bags (Jensen Inert Products; polyvinyl fluoride, CO2 permeability 4.3 cm3 mm (m2 d atm)−1 at 25°C [Massey, 2003]) and a hand pump to remove the headspace, a more satisfactory method which we recommend for future work. The bagged core sections were left to melt at room temperature, which usually took about 24 h and then subsampled immediately (completed within another 3 h). We estimate that the maximum potential CO2 leakage into the LDPE bags, for a very low pCO2 sample collected from a core bottom late in the season, could have contributed no more than 30 μmol kg−1 to the samples. The data became more variable after 11 April, but it is not clear whether that variability is related to changes in how the core sections were handled or due to the onset of spring warming (see section 4.2).
 After melting, some headspace gas (never more than 2% of the total volume, usually much less) was always present in the sample bags, partly from residual headspace left when the cores were wrapped and partly from gas released from the cores as they melted. The melts were mixed gently before subsampling using a 250 mL glass syringe with a long needle inserted through a small opening in the corner of the bag. Both the syringe and subsample bottles were rinsed with a small quantity of the sample. Samples to be analyzed for total inorganic carbon (TIC) and alkalinity (AT) were drawn first, with the syringe needle at the bottom of the bag, and slowly to minimize disturbance and exchange with the air while the sample was both drawn and dispensed. The TIC and AT samples were dispensed into 250 mL glass bottles, leaving a reproducible 1% headspace, poisoned with 100 μL of a saturated HgCl2 solution, capped with ground glass stoppers greased with Apiezon-M, sealed with elastic and plastic clips, and stored cold until analysis (see standard protocols [Department of Energy (DOE), 1994]). Duplicate samples for total organic carbon (TOC) were dispensed into 60 mL glass vials (which had been acid washed and combusted at 550°C for 1 h), without any prefiltration, sealed with screw caps with Teflon-lined silicone septa (acid washed and air-dried), and refrozen at −80°C until analysis. Because the TOC values measured in the ice melts were very low (lower than in seawater), we concluded that the ice samples did not significantly leach organic matter from the plastic bags during the short time required for melting. The full suite of carbon parameters (TIC, AT, and TOC) were sampled from ice cores collected between 29 February and 9 May. Duplicate cores were collected within 4 m of each other on 5 April and 2 May. Measured values from parallel 10 cm sections from the duplicate cores differed by 1 to 17% for TIC and alkalinity and by 6 to 60% for total organic carbon.
 A duplicate core collected on 9 May 2004 was cut into 10 cm sections and stored in zipper plastic bags at −20°C until September–October 2009, when solid CaCO3 was recovered from the ice sections, based on Dieckmann et al. . We essentially followed the Dieckmann et al. method, except that each core section was melted in a Tedlar bag in a 1.5°C water bath. After the crystals collected in the corner of the bag, we drew them out using a pasteur pipette. Because we were melting these samples in a land-based laboratory, when some sections were only partially melted, they were placed over night in a 0.5–2.0°C refrigerator, where surrounded by air, instead of water, the amount of unmelted ice remaining in the samples changed little between evening and morning. The samples from individual core sections were weighed and then combined into a single sample for X-ray diffraction analysis.
 Additional ice cores were collected on a weekly basis from 10 January to 28 March from the same dedicated ice field (Figure 1) and processed to determine particulate matter, bacteria, and extracellular polysaccharide substance (EPS) concentrations, as described in detail by Collins et al. . We converted the bacterial numbers and EPS quantities in the ice reported by Collins et al. to carbon equivalents for comparative budgeting purposes. We calculated bacterial biomass using conversion factors from Bjørnsen and Kuparinen  and carbon attributable to EPS using the conversion factor of 0.7 μg C μg−1 xanthan gum equivalents [Passow, 2002].
 Ice temperatures down to 100 cm were taken from a thermocouple string (sensors at 0.01, 0.05, 0.10, 0.15, 0.20, 0.30, 0.50, 0.60, 0.70, 0.80, and 1.00 m depths) frozen into the ice about 300 m away from the coring site, at the meteorological site (black dots south of site F in Figure 1). Below 1 m, temperatures were measured directly on sampled ice cores, either on the chemistry cores before they were sectioned or on separate cores. As soon as a core came up, narrow boreholes were quickly drilled to the center of the core at 10 cm intervals, starting at 5 cm from the top, and the temperatures were measured by inserting a temperature probe (Hart Scientific 1522) into the hole. At the low air temperatures during deep winter, the ice temperatures measured directly on the cores were imprecise and nearly always falling as the readings were taken. Therefore, we used the measurements from the cores only for depths below 100 cm, the bottom of the thermistor string. Salinities of the core melts (from either the same core used for carbon sampling or a parallel core collected at the same site and time) were measured using a handheld conductivity meter (Hoskin Scientific Cond 330i; accuracy ±0.5% of reading). Subsamples for 18O were taken on 2 May, as well as from ice cores collected further to the west of the ship on 26 May, and stored at room temperature in screw cap bottles (either glass or polyethylene) wrapped in Parafilm. The physical properties (salinity, temperature, density) of the snow and ice surface were monitored at sites F2 and B (Figure 1), as detailed by Langlois et al. [2007, 2008].
 We measured pCO2 within the ice by a new in situ method based on silicone exchange chambers [Owens, 2008], adapted from techniques used in soil science [Holter, 1990; Jacinthe and Dick, 1996; Kammann et al., 2001]. A silicone chamber sampler (“peeper”) is shown in Figure 2a. The chambers were 20 cm lengths of silicone tubing (platinum cured Tygon® 3360 silicone, Cole-Parmer) with an inside diameter of 12.4 mm (for a total volume of 24 cm3) and a wall thickness of 2.0 mm. Each tube was capped at both ends with silicone stoppers, and C-Flex tubing (Podisco, Inc.) led through one end to a surface sampling port consisting of a two-way stopcock valve. Because C-Flex can be slightly permeable to gases, we recommend that stainless steel tubing be used in future deployments. For deployment, sets of three peepers were housed in PVC tubing (Figure 2b), separated and held in place at 30, 70, and 110 cm from the ice-atmosphere interface by foam sealant (Great Stuff™, Dow Chemical).
 Three peeper arrays were deployed at each of sites C and F (Figure 1). The gas samplers were frozen into narrow boreholes drilled into the ice (which was 60–70 cm thick at the time) prior to 19 December 2003. By the time sampling began on 9 January 2004, both the top and middle peepers were encased in the ice. The delay in the start of sampling should have also allowed the peepers to reach equilibrium with their surroundings, even at the low temperatures we encountered [Holter, 1990; Jacinthe and Dick, 1996]. By mid-February, the ice was thick enough to completely enclose the peepers at 110 cm. The peepers at site F were sampled in the morning and those at site C in the evening for purely logistical reasons, to coordinate with other operations and sampling. Gas samples (up to 10 mL) were drawn from each peeper with a syringe fitted with a stopcock valve and immediately injected into evacuated 10 mL glass vials for subsequent gas chromatography (GC) analysis. The evacuated vials would draw the sample out of the syringe as soon as the syringe stopcock was opened; if it was necessary to depress the plunger on the syringe in order to inject the sample, we assumed that the seal on the vial had been compromised, and we would discard the sample. We collected atmospheric samples at the same time, also with syringes, at about shoulder height while facing into the wind. During sampling we were careful not to contaminate the samples with either snow or our breath. The exposed sampling ports were susceptible to damage by wildlife (Arctic foxes), and 6 L plastic buckets were anchored to the ice surface over the sampling ports for protection. After several months of operation, it became more difficult or even impossible to draw gas samples from some of the individual peepers (i.e., we drew only vacuum into the syringes), possibly the result of degradation of the silicone chamber allowing water to enter and freeze within the chamber. Additional details of the peeper design and deployment are presented by Owens .
 Reproducibility between peepers appeared to diminish toward the end of the study, and it is unclear whether this loss was due to degradation of the peepers through the long deployment or increasing variability within the ice as the melt season progressed. Future studies involving short-term peeper deployments in spring need to specifically investigate the spatial heterogeneity of melting sea ice.
2.4. Water Sampling
 While the ship was frozen in the ice, subsurface water (>10 m before mid-February, >20 m after that time) samples were collected with a 24 Niskin bottle rosette fitted with a SeaBird 911+ CTD deployed through the ship's moon pool, generally every 6 days. The overwintering site was also visited twice during November 2003, the month before the ship froze into the ice, and twice more in July and August 2004, when the ship was mobile again, and at those times, the rosette was deployed off the side of the ship, allowing us to collect full profiles to within 2 to 3 m of the surface [Mucci et al., 2010]. Both TIC and total alkalinity (AT) were sampled according to standard protocols [DOE, 1994]. Briefly, the Niskin bottles were tapped into either 250 or 500 mL glass bottles, overflowing at least a full bottle volume to rinse. A reproducible headspace of approximately 1% was introduced, 100 μL of saturated HgCl2 solution was added, and the bottles were sealed with ground glass stoppers greased with Apiezon M, elastic, and plastic clips. Samples were stored cold until analysis. Oxygen-18 samples were tapped directly into screw cap bottles (either glass or polypropylene) and wrapped in Parafilm before storage at room temperature.
 Surface waters (0.5–10 m below freeboard) were collected the day after each moon pool cast through a hole in the ice at 70°02.44 N, 126°17.13 W, about 0.4 km from the ship (site E, Figure 1). A protective shack of plastic sheeting over a wooden frame surrounded the hole. A 2 L vertical acrylic Kemmerer bottle was ordinarily used to draw the water, but under extremely cold conditions a submerged pump with a ceramic magnetic impeller (a Supreme Mag-Drive 5 pond pump) was sometimes used. A flexible rubber tube attached to the outlet of the pump allowed us to control the flow rate and eliminate bubble entrainment while filling the sample bottles. The difference between analytical results from samples collected by bottle versus pump was the same as the instrumental reproducibility throughout the expedition (see section 2.5). Alkalinity and DIC samples were each sampled quickly (to limit freezing) in triplicate into 250 mL glass bottles with small overflow to rinse, and no more than two DIC samples were tapped from a single Kemmerer bottle cast. A reproducible headspace was introduced into each sample bottle, which was closed with a greased stopper and placed immediately into a warm insulated box (a “cooler”) with several snap-activated gel heat packs. After returning to the ship, within 1 h, the samples were poisoned with HgCl2 and sealed for storage according to standard methods [DOE, 1994]. The samples were not poisoned at the sampling site, because the small volumes of saturated HgCl2 solution froze in the pipette tip. The TOC samples were tapped directly into clean 60 mL glass vials, sealed with Teflon-lined silicone septa screw caps, allowed to freeze, stored onboard the ship in a −80°C freezer, and shipped frozen back to the Institute of Ocean Sciences (IOS) for analysis. Duplicate salinity and 18O samples were stored warm for transport back to the ship. Temperature and salinity profiles were also taken through the ice hole using a CTD (SeaBird19, Sea-Bird Electronics, Inc.), but many of those data were lost. Of the data available, the temperatures were never more than 1°C from the freezing point and usually within 0.4°C. Therefore, the freezing point calculated from the bottle salinities [Fofonoff and Millard, 1983] was used in place of the missing temperature data.
2.5. Chemical Analyses
 Total inorganic carbon was measured coulometrically using a SOMMA system [Johnson et al., 1993] fit to a UIC 5011 coulometer, according to standard protocols [DOE, 1994], and calibrated against certified reference materials (CRM Batch 61) provided by Andrew Dickson of Scripps Institute of Oceanography, a standard made at IOS, or a secondary standard made from a large quantity of deep water collected early in the expedition and preserved in the same way as the samples. Both the IOS and deep water standards were regularly calibrated directly against the CRMs. Most analyses were conducted onboard ship, although some were shipped cold back to IOS; all samples were analyzed within a year of collection. Precision, based on the difference between two replicate samples drawn from the same Niskin bottle, varied between expedition legs from 1.6 to 3.9 μmol kg−1.
 Total alkalinity was measured onboard using an automated Radiometer (Titrilab 865) potentiometric titrator with a Red Dot® (pHC2001) pH combination electrode in an open cell in continuous titrant addition mode and with an algorithm specifically designed for shallow endpoint detection. The dilute HCl (about either 0.025 or 0.005 N) titrant was calibrated at the beginning and end of each day using the certified reference materials (CRM Batch 61) and the secondary standards made from deep water. After early April, a modified titration program that had been optimized for low alkalinities and the lower acid titrant concentration was used to analyze the ice melts, but there was no offset between those analyses and earlier ice melt analyses using the method optimized for seawater. While the ship was immobilized in the ice, titration volumes were determined by mass, and while the ship was transiting, the samples were dispensed into the cell with precalibrated, water-jacketed, thermostated (25.0 ± 0.05°C) pipettes (∼80 mL). The average, cumulative (total) relative error, based on the sum of the relative standard deviation on replicate standard and sample analyses, was 0.3%.
 Carbon dioxide mole fractions in the gas samples from the peepers were analyzed onboard ship by gas chromatography using a portable SRI 8610B gas chromatograph (GC) equipped with a thermal conductivity detector (TCD) and a Poracil C column (mesh: 80/100; length: 60 cm; diameter: 0.3 cm). The TCD was set at 100°C, the GC oven temperature was set at 40°C, and the carrier gas was He. Samples were injected into the gas sampling valve manually. Starting from the middle of February, a 15 cm (reduced to 2 cm in May) water trap (Mg(ClO4)2) was mounted before the valve to protect the TCD from moisture. Concentrations were determined from peak area using PeakSimple 3D10 software with a typical CO2 retention time of 2.2 min. The gas chromatograph was regularly calibrated (usually daily) against a 412 (±2%) ppmv standard gas. Analyses of the standard using three different sized sample loops confirmed that the detector response was linear to at least 8000 ppmv. The estimated error for the gas analysis was ±40 ppmv CO2, based on standard analyses performed throughout the project. Additional details are presented by Owens . Because in situ pressures within the ice were not available, we converted the mole fraction values to partial pressures (μatm) using the barometric pressure at the time of analysis. We note that precision and accuracy of the GC analyses are not as good as with IR detection [e.g., Delille, 2006], which we recommend for future applications with sea ice peepers.
 The X-ray powder diffraction analysis of the solid material recovered from the stored ice core was performed on a Bruker D8 Focus Bragg-Brentano diffractometer, and quantitative phase analysis was done with Rietveld program Topas 3 (Bruker AXS). The sample was found to be composed of at least 97.4% CaCO3 minerals (93.1% calcite and 4.3% vaterite; we note that ikaite could not be identified in the sample, because it had been allowed to warm up to room temperature before analysis). As the specimen was sufficiently thin to transmit X-rays, the results should be considered approximate.
 Total organic carbon analyses were conducted at IOS by high-temperature combustion with a Tekmar-Dohrmann Apollo 9000 TOC analyzer, calibrated with potassium phthalate solutions [Spyres et al., 2000]. Analytical precision, based on analysis of replicate samples, was less than 7 μmol L−1, and accuracy was assured by additional calibration against seawater TOC certified reference and blank materials supplied by D. Hansell and W. Chen, Rosenstiel School of Marine and Atmospheric Science, University of Miami, USA.
 Oxygen isotope (δ18O) measurements were made at the University of Ottawa (G.G. Hatch Isotope Laboratories) using a gasbench flushed with a gas mixture of 2% CO2 in helium and a DeltaPlus XP isotope ratio mass spectrometer (ThermoFinnigan, Germany). Isotopic composition is expressed as δ18O referenced to the V-SMOW standard according to the formula
Analytical precision estimated from pooled duplicate samples sent to the laboratory over the period of analysis has been evaluated as sp = ±0.05‰ (ν = 50).
3. Atmosphere-Ice Fluxes
 The CO2 flux between the surface (snow/ice) and the atmosphere was highly variable throughout the study and quite large at times (Figure 3), with a maximum daily average flux of 0.86 μmol (m2 s)−1 upward, away from the ice, on 4 May. The fluxes were mainly upward, but substantial downward fluxes (as much as 0.70 μmol (m2 s)−1 toward the ice surface, on 11 February) were not uncommon.
 The largest fluxes in both directions often occurred during periods of high wind speed or when the temperature was at a relative maximum or increasing (Figure 3c). High wind speed accelerates vertical exchange of CO2 and ventilates the snow cover [Albert and Shultz, 2002; Takagi et al., 2005], and the relationship to temperature is consistent with the ice becoming more permeable as it warms and brine volume increases [Kelley and Gosink, 1979; Petrich and Eicken, 2010]. At brine fractions below about 5%, columnar sea ice becomes impermeable to fluids [Golden et al., 1998], but the analogous cutoff for gases is unknown. Our data imply that air-ice CO2 exchange is possible at very low temperatures, which may also indicate transport within the sea ice, consistent with the results of Gosink et al.  and Kelley and Gosink , who found significant gas fluxes in sea ice at temperatures as low as −25°C.
 Even in deep winter, we observed significant CO2 fluxes, ranging up to 1 μmol (m2 s)−1, values at the higher end of what is seen over open water [e.g., McGillis et al., 2001]. It is difficult to imagine that sea ice at these temperatures is sufficiently permeable to allow movement of large quantities of CO2 without extensive cracking and structural failure of the ice sheet, and we suspect that many of the fluxes we observed at the lowest temperatures were exchanges with brine in the snow cover on top of the ice. For example, the particularly high CO2 uptake observed between 9 and 12 February coincided with a surface warming as the warm sector of a low-pressure system arrived at the site [Langlois et al., 2008]. We suspect that at this time, the CO2 may have been drawn down by CaCO3 dissolution in the brine at the snow-ice interface. This hypothesis is plausible, given that the ice surface and adjacent snow were quite salty: from 30 January to 15 February, the salinity at the ice surface was substantially higher than in the snow just centimeters above the ice surface [Langlois et al., 2008].
 Between 30 March and 9 April, the near-surface ice temperature rose rapidly (Figure 3c), corresponding to intense storms and an increase in snow thickness to more than 20 cm. Fluctuations in near surface ice temperature were dampened under the thick snow, but both the amplitude and frequency of the CO2 exchange variations increased during this period, roughly in step with diurnal fluctuations in air temperature and wind speed. Although high rates of CO2 uptake were observed under the thick snow regime, most of the flux as spring approached was toward the atmosphere, consistent with observed inorganic carbon distributions within the ice (see sections 4.1 and 4.2). That is, because pCO2 values in the sea ice were higher than in the atmosphere (Figure 4), when the sea ice warmed and brine channels reopened, large upward CO2 fluxes became possible (Figure 3). The fluxes reported here do raise interesting questions about the CO2 source-sink capacity of the snow and sea ice. Additional flux work using multiple methodologies is required to better understand the mechanics of the air-surface exchange and by extension, the limitations of the various techniques.
 Gaps in our data set limit confident extrapolation of our flux measurements to calculate a seasonal total, but a first-order estimate indicates that between the end of January and the end of May 2004, the sea ice in Franklin Bay released a net of almost 3 mol CO2 m−2 to the atmosphere. During the open water seasons immediately prior to and after this study, the waters of Franklin Bay were a mild sink for atmospheric CO2 [Mucci et al., 2010], with fluxes between 0 and −10 mmol (m2 d)−1 (or up to −0.1 μmol (m2 s)−1). Comparison between these low fluxes and the larger fluxes we observed over the ice during the winter challenges the conclusion derived from summertime studies that seasonally ice-covered seas are net annual CO2 sinks [e.g., Yager et al., 1995; Miller et al., 2002; Sweeney, 2003; Bates et al., 2006; Bates and Mathis, 2009]. Whether Franklin Bay and similar seasonally ice-covered coastal waters are net annual sinks or sources of CO2 will depend on the fluxes while the ice forms in autumn. Laboratory studies [Nomura et al., 2006] have indicated that as sea ice freezes, it releases CO2 to the atmosphere, but that process has not been confirmed in the field. Extrapolation of the results from Nomura et al. indicate that as much as 1 mol CO2 m−2 could have been released from the newly forming ice by the time our eddy covariance measurements began, when the ice was 1 m thick (assuming 60 days since ice formation began). In another laboratory experiment, Nagurnyi  also found that sea ice formation provides a net atmospheric CO2 source. On the other hand, Anderson et al.  have proposed that turbulent boundary processes may enhance CO2 drawdown into the very cold water as frazil ice forms, possibly limiting the source implied by the work of Nagurnyi  and Nomura et al. .
4. Evolution of Ice Biogeochemistry
4.1. The pCO2 in the Ice
 The CO2 partial pressures (pCO2) we measured within the ice were extremely high and variable (Figure 4). In the case of the highest values, often only one of the three chambers sampled at a given time registered an extremely high value (i.e., greater than 10,000 μatm). However, that chamber would consistently give higher values than the other chambers, and the extremely high values were limited to the upper portions of the ice in the peepers sampled in the evening, indicating that the variability may be due to spatial and temporal heterogeneities within the ice. Therefore, we have chosen to leave the extremely high values in the data set, despite the lack of corroboration across all the sampling chambers.
 Notably, the samples collected at site C in the evening were generally higher and more variable than those collected at site F in the morning. Possible natural explanations could include photochemical or temperature-controlled respiration variations. However, we expect that both effects would have been subtle, particularly early in the season, when there was little light or diurnal temperature variation (the sun rose on 18 January and by the middle of February was up for only 7 h each day). While it is conceivable that the PVC (polyvinyl chloride) casing around the peepers (Figure 2b) could have been conducting and amplifying subtle temperature variations, the thermal conductivity of PVC (0.2 W mK−1) is actually an order of magnitude lower than that of ice (greater than 2 W mK−1 [Pringle et al., 2007]). Therefore, the PVC casings would have been dampening any thermal variations, not enhancing them. A more likely explanation for the differences between the evening and morning samples is that they were collected at different sites. Not only was site C quite close to the ship, and therefore more subject to atmospheric (if not thermal) contamination from the ship, but we would also expect normal spatial variability in the ice to give different results at the two sites.
 Notwithstanding the extremely high pCO2 values measured toward the top of the ice, the generally high pCO2 levels observed in the ice overall are startling. A very significant potential artifact could have been introduced into the measurements because the sampling procedure drew vacuum on the peeper chambers. This could have disrupted the carbonate equilibria within the ice, artificially drawing CO2 into the gas phase. In addition, silicone is nearly 7 times more permeable to CO2 than N2 [Massey, 2003], and therefore, CO2 in the evacuated peepers would have equilibrated with the surrounding ice much faster than N2, giving erroneously high CO2 partial pressures inside the peepers if they were sampled before the N2 had also equilibrated. The measured pCO2 values did rise with sampling frequency, implying that reequilibration times between samplings may have been inadequate to relieve the sampling-induced vacuum. Although we see no relationship between pCO2 and brine volume fraction (Figure 5a), implying that brine transport was not a primary factor controlling the measured pCO2 values, the brine volume fraction was below 5%, the cutoff for brine mobility [Golden et al., 1998] throughout our study.
 On the other hand, the observed values are not inconsistent with the limited number of previous measurements: Matsuo and Miyake  found CO2 concentrations between 0.4 and 20% (presumably by volume) in sea ice samples collected in Antarctica and northern Japan; and Tison et al.  found even higher CO2 concentrations (up to nearly 50%, by volume) in sea ice grown under controlled laboratory conditions. Tison et al. attributed their high pCO2 observations to bacterial contamination, which does not compromise the comparison with our natural samples, which also contained bacteria (see section 4.3). Comparing our measured pCO2 values (up to 12,000 μatm) with those calculated from measured temperature, salinity, TIC, and AT in the ice melts (up to 45,000 μatm) indicates that the measured values could even be minima (Figure 5b), despite their surprisingly high magnitudes. Furthermore, the calculated values would be even higher, if we had considered CaCO3 precipitation. The lack of agreement between the measured and calculated values is not entirely surprising, since the calculations are based upon stability constants for the carbonate system which are only known to be valid under seawater conditions (temperatures down to 0°C, salinities up to 40 practical salinity units (psu)), not at the low temperatures and high salinities of sea ice brines. Nevertheless, the theoretical values are so high that CO2 loss during sampling and analysis would be expected, and we encourage other investigators to avoid collecting discrete samples for GC analysis in favor of measurements conducted in the field using nondispersive infrared detectors [e.g., Delille, 2006].
 Mechanisms that could result in gas enrichment within sea ice are: release from freezing seawater, release from brine with further freezing, and release by biological activity. Biologically, CO2 could be released into the ice if respiration exceeds photosynthesis, a scenario which may be common in sea ice [e.g., Grossmann and Dieckmann, 1994; Deming, 2010]. The primary abiotic pathway by which freezing seawater and brines would likely release CO2 is through precipitation of calcium carbonate [Jones and Coote, 1981; Tison et al., 2002; Papadimitriou et al., 2004; Dieckmann et al., 2008], with additional salting out contributions as the brine salinity increases.
 Despite these theoretical rationales for high pCO2 values within sea ice, our peeper design and sampling protocols were experimental. Substantial additional studies are required to confirm under what conditions the peepers accurately measure pCO2 in the ice and to perfect the methodology. For now, in the absence of other wintertime data on pCO2 in sea ice, we present these data to await confirmation or contradiction from future studies.
4.2. Ice Carbon Contents
 The inventory of TIC in the ice increased throughout the 2 month time series (Figure 6a), although the increase in total carbon was not as definitive because of variability in the total organic carbon trend. Between the end of February and the beginning of May, the total carbon content (organic plus inorganic) in the ice increased by 0.13 (±0.04) mol m−2 (the increase in TIC alone constituted 0.12 ± 0.03 mol m−2). This increase in the total carbon inventory is almost certainly due to entrainment from seawater into the growing ice sheet. On the other hand, the average total carbon concentration within the ice decreased throughout the winter (Figure 6b), emphasizing that similar to salinity, carbon is not quantitatively entrained in sea ice as it forms.
 Total inorganic carbon and alkalinity measured in the ice melts appear to be tightly coupled (slope = 1.06, r2 = 0.96), implying that their concentrations are controlled by concentration and dilution with ice freezing and melting. However, if concentration and dilution were the only processes affecting TIC and alkalinity, normalizing the data to a constant salinity would remove all variability in each parameter, whereas normalizing our data gives a relationship that still has a slope close to one (Figure 7). Circumstantial evidence from past studies indicates that as ice ages, bicarbonate is retained in the ice while bulk salinity is lost [Tsurikov, 1974], consistent with the hypothesis that calcium carbonate is precipitating and staying in the ice as brines drain. If CaCO3 precipitation alone were controlling TIC and AT in the ice, we would expect the data to fall along a line upward from the fall seawater end-member with a slope of two as brine was lost, leaving CaCO3 behind to be measured in the ice melts. Because the slope is closer to one, there must have also been some input of CO2 gas, most likely from respiration, although absorption from the atmosphere could also have been contributing at some times (Figure 3). Note that the effect of CaCO3 precipitation shown in the rosette diagram on Figure 7 is reversed from the typical presentation [e.g., Broecker and Peng, 1989], because we are examining the impact of CaCO3 left behind in the ice when the carbon-depleted brine has drained away. Interestingly, the two samples that appear to follow the photosynthesis trend, in the middle of Figure 7, were collected on 2 May, well into spring.
 Although we did not measure CaCO3 distributions in the ice during our study, we have confirmed that the salt could have precipitated in the ice under wintertime conditions. A duplicate core that was collected on 9 May 2004 (when the temperature of the ice varied between −8°C at the top and −3°C at the bottom) and then stored for more than 5 years at −20°C contained substantial quantities of CaCO3 (Figure 8). The amount of CaCO3 shown in Figure 8 may be greater than what was present as actual solid salt precipitate in the core when it was collected, because the core had been stored at a much lower temperature than its in situ temperatures at the time of collection. However, the upper portions of that ice sheet did experience temperatures below −20°C during the months before we extracted the core (Figure 9), and therefore, it is likely that substantial CaCO3 precipitation had occurred in the ice during our study. With precipitation of CaCO3 and other salts, at some point, we would expect that the total alkalinity of ice brines would no longer be dominated by the carbonate system. However, in the absence of measurements of the other major ions in the ice brines, we will continue to assume that at the temperatures we encountered, carbonate alkalinity largely followed total alkalinity.
 The ice was generally enriched in both TIC and AT, relative to salinity. That is, starting from the surface waters observed in Franklin Bay during the preceding autumn, more salt appears to have been lost from the ice than carbon. However, the top 10 cm of the ice was usually depleted in both TIC and AT (Figure 7; the one exception was from the very first core we collected in February). This depletion in the surface ice could result from a combination of two processes: CO2 loss to the atmosphere and upward migration of CaCO3-depleted brines. While most of the brine excluded from young ice migrates downward, upward permeability in the ice can be greater than downward permeability [Ono and Kasai, 1985], and at least some brine appears to be forced upward [Martin, 1979; Perovich and Richter-Menge, 1994; Wettlaufer and Worster, 1995]. If CaCO3 had formed as the ice froze, upward migrating brine would be depleted in both TIC and AT, explaining the low salinity-normalized values at the top of the ice. However, having left behind solid CaCO3, the alkalinity in the upward migrating brine would be depleted by twice as much as TIC, contrary to what we observed. The most likely explanation for this discrepancy would appear to be CO2 loss to the atmosphere, which would reduce TIC in the ice but not AT. Although very cold ice is expected to be impermeable, a direct connection between its surface and the atmosphere is conceivable; the top of the ice in our study was generally composed of granular ice, which would have had more brine channel branching and connections between pore spaces than the columnar ice lower down [Weissenberger et al., 1992], possibly allowing more efficient gas transport at lower temperatures. Because our shallowest peepers were installed at 20 to 40 cm from the top of the ice, below the level of TIC and AT depletion, we are unable to establish whether pCO2 at the surface of the ice was also high (as would be expected for low-alkalinity brines) or closer to equilibrium with the atmosphere (if there was open exchange between the two reservoirs).
 During the coldest period, through March and early April, salinity-normalized TIC and AT were both relatively constant with depth below the top of the ice (Figure 10a). The data became more variable after 11 April, which could be related to changes in how the core sections were handled (see section 2.3.1), but that was also when the spring warming began, and the observed changes are consistent with the environmental conditions. As temperatures rose and the ice began to melt, normalized TIC and AT increased, primarily because salinity decreased. Figure 10b shows that the measured (unnormalized) TIC and AT concentrations remained relatively constant. Therefore, we conclude that as the brine became mobile again, transporting salinity downward, it left some inorganic carbon behind, possibly as solid CaCO3 salt. Interestingly, the unnormalized TIC and AT distributions for the core collected on 9 May (Figure 10b) are remarkably similar to the CaCO3 distribution in the duplicate core collected on that same day (Figure 8).
 Our TIC results from the ice melts were slightly higher, and total alkalinity was slightly lower, than those obtained by another group also working in Franklin Bay during April 2004 (Table 1) [Rysgaard et al., 2007]. The values given in Table 1 are averages of the measurements from each core; Rysgaard et al. apparently did not include the top 20 cm of their cores, contributing to their lower within-core variability. When the within- and between-core variability in each study is taken into consideration the differences are not actually significant, which is remarkable, considering the different locations and methods used. The small difference in TIC could be partly explained by the fact that Rysgaard et al. melted their sections in syringes filled with water, while we melted our core sections in plastic bags, from which it was not possible to eliminate all headspace; at 0°C, pCO2 in the ice melts would have been quite low, potentially contributing to CO2 absorption from the headspace during melting of our samples. Complete equilibration with laboratory air at 400 μatm would have added only 30 μmol kg−1 TIC to the melts, significant, but within the variability in the measurements. The differences in total alkalinity are harder to explain. Brine loss from our samples during transport back to the ship could have accounted for no more than 3 μmol kg−1 in AT. Potentially, incomplete CaCO3 dissolution in our melted samples could have had a significant impact, but even taking the average amount of CaCO3 we recovered from our stored core as a maximum (since at the time of sampling, the ice was warmer in situ than our storage temperature of −20°C), residual solid CaCO3 could have accounted for a maximum of only 30 μmol kg−1 of alkalinity in our samples, again well within the variability between our samples. An additional explanation for at least part of the difference between our results and those of Rysgaard et al. is simply spatial variability, since the sampling sites were more than 1 km apart (Rysgaard et al. sampled quite close to the ship, 150 m NW off the stern, Figure 1). As the duplicate core sections collected on 2 May show (Figure 11), spatial variability can be quite high. Curiously, our AT/TIC ratios fall between those Rysgaard et al. found in Franklin Bay and Young Sound. Regardless of the source of the differences and despite the circumstantial evidence presented above, neither the quantities of CaCO3 we recovered from our stored core nor the AT/TIC ratios we measured in our fresh cores can conclusively confirm a preferential retention of alkalinity over TIC in the ice as it formed and rejected brine.
Table 1. Comparison of Ice Carbon Concentrations Between Investigators, Franklin Bay, Spring 2004a
TIC (μmol kg−1)
AT (μmol kg−1)
Values are averages and standard deviations of ice melt measurements taken from each core.
 In February, ice pCO2, as measured by the peepers, increased dramatically (Figure 4) as the temperature decreased (Figure 3c). The increase occurred in both evening and morning samples, as well as at the different depths in the ice. We suggest that the overall pCO2 increase in the ice was due, at least partly, to CaCO3 precipitating from the brines as the ice cooled [Dieckmann et al., 2008], but this hypothesis is based on theoretical considerations and cannot be confirmed with our data set. In addition, CaCO3 precipitation cannot explain many of our observations, and other processes must have also contributed to the trends in the sea ice carbon system.
 In seawater, pCO2 roughly doubles with every 16°C increase in temperature [Takahashi et al., 1993], all else kept constant (i.e., solubility decreases with increasing temperature), and such a relationship is consistent with our observations in the ice, particularly in spring (Figures 3c and 4). Direct temperature control of pCO2 may also help explain the pCO2 decreases we observed during March, the coldest period of the expedition. On the other hand, temperature was clearly not the dominant factor setting sea ice pCO2 during February. Even in spring, the highest pCO2 values were often observed at the top of the ice (Figure 4b), which was still colder than deeper in the ice (Figure 9), further indicating that temperature was not the only factor controlling pCO2.
 Without exception, organic carbon was enriched, relative to salinity, in all the samples we analyzed from the sea ice (Figure 12). Such organic carbon enrichment in sea ice has been observed by others [Thomas et al., 1995, 2001; Giannelli et al., 2001] and has been attributed mainly to concentration during initial ice formation and consolidation [Garrison et al., 1983; Ackley and Sullivan, 1994], as well as production within the ice [Thomas et al., 2010]. Collins et al.  tracked EPS in ice cores collected in parallel to ours (within 4 m) and detected a significant increase in EPS throughout the upper ice from January to March. This increase, attributed to in situ bacterial production of EPS as a cryoprotectant [Krembs et al., 2002; Collins et al., 2008], could account for at least part of the TOC increase that we observed at the start of our study, at the beginning of March (Figure 6b). At the same time we collected our cores, Riedel et al.  found concentrations of EPS in the bottom of the sea ice in Franklin Bay (approximately 0.5 km to the east of our coring site) that were in the same range as our TOC measurements (Figure 12). Both data sets indicate that EPS constituted a significant fraction of the TOC we measured. After an initial increase, TOC in the ice decreased through the remainder of the winter, even as the ice continued to grow (Figure 6).
 Heterotrophic remineralization by bacteria [Deming, 2010] or protists [Caron and Gast, 2010] may also have contributed to the high pCO2 values we observed in the ice. Some studies have found evidence that significant bacterial activity can occur in sea ice, even in winter [Junge et al., 2004; Wells and Deming, 2006], notwithstanding relatively low biomass. Although bacterial biomass in our cores (Table 2) was about 3 orders of magnitude lower than TOC, the total inventory we measured in the full ice column decreased between 27 January and 28 March, despite increasing ice volume, indicating that the bacteria were participating in the carbon cycle to at least some extent. Respiration, like bacterial growth, is a temperature-dependent process [e.g., Rivkin and Legendre, 2001; Deming, 2010; Caron and Gast, 2010], and therefore in addition to the direct temperature effects discussed above, any increase in temperature would have contributed indirectly to our observed sea ice pCO2 increases by increasing respiration, both as spring progressed and through the course of the day (Figure 4).
Table 2. Bacterial Biomass Inventories in the Sea Icea
Ice Thickness (cm)
BB (mmol C m−2)
BB, bacterial biomass inventories. Based on a conversion factor of 40 fg C cell−1, assuming 0.1 μm3 cell−1 [Bjørnsen and Kuparinen, 1991]. Estimated uncertainties are based on the observed gradients within the cores.
0.187 ± 0.016
0.157 ± 0.011
0.096 ± 0.025
 Working about 1.5 km from our site during April and early May 2004, Rysgaard et al.  also found evidence for significant heterotrophic activity within the ice, based on in situ oxygen distributions and dynamics. They measured a net oxygen consumption, particularly in the lower parts of the ice, indicating that heterotrophy exceeded primary production, which they estimated at about 7 μmol C (m2 d)−1 (assuming production of 0.3 μg C (L d)−1 over the bottom 30 cm). Rysgaard et al. speculated that heterotrophic O2 consumption had begun before the bloom, consistent with evidence that bacteria were actively producing EPS [Collins et al., 2008] and in some cases even growing (albeit slowly [Wells and Deming, 2006]) in this ice field during the winter. Translating the oxygen results from Rysgaard et al. into a net carbon release is problematic, since O:C ratios of metabolic processes are poorly defined even in bulk seawater. The Redfield N:C ratio is more robust, and making a first-order assumption that it applies to ice communities, the results of Cota et al.  on the rates at which heterotrophic ice organisms remineralized nitrogen in the Canadian Archipelago can be used to estimate a maximum carbon remineralization in the ice approaching 1 mmol (m2 d)−1. This value agrees with our observed TOC decrease of 60 mmol m−2 between early March and mid-May (Figure 6a) to within an order of magnitude, indicating that although we cannot confidently quantify it, heterotrophic respiration could have played a significant role not only in daytime pCO2 increases within the ice as spring progressed (Figure 4b), but also in the TOC decreases during winter (Figure 6).
 Circumstantial evidence also points to abiotic photochemical remineralization as a possible additional source of inorganic carbon and high pCO2 within sea ice [e.g., Belzile et al., 2000; Norman et al., 2011]. At the end of our time series, Xie and Gosselin  found high concentrations of carbon monoxide at both the top and the bottom of the sea ice in Franklin Bay. Based on laboratory experiments, they proposed that the CO could have been produced photochemically, which would also imply photochemical CO2 production [Miller and Zepp, 1995]. Photochemistry could also explain why pCO2 levels were higher in the ice in the evening than in the morning (Figure 4) [Owens, 2008].
 Despite the likelihood that significant amounts of TOC were remineralized within the ice, either biotically or abiotically, the decrease in the TOC inventory was only about 30% of the increase in TIC (Figure 6a). Therefore, while remineralization may have been contributing to the changes we observed in the inorganic carbon pool, it was not the dominant factor.
 In conclusion, the carbon dynamics of sea ice must be controlled by a mixture of biotic and abiotic processes interacting in complex ways that change as the seasons progress. Direct quantification of the relative contributions from temperature, CaCO3 precipitation, heterotrophic remineralization, primary production, and photochemistry will require that future expeditions comprehensively examine those controlling factors throughout the ice column at the same narrow spatial and temporal resolution, involving further development of nondestructive analytical methods. In addition, the inorganic carbon system needs to be better defined in ice and brines, clarifying the thermodynamic and kinetic relationships between solid CaCO3, pCO2, alkalinity, and dissolved inorganic carbon.
5. Ice-Water Fluxes
 In our data set, carbon was preferentially retained in the ice as brine was rejected (Figure 7), indicating that the brines were depleted in TIC and AT, relative to salinity. However, those brines would still have had quite high TIC concentrations, compared to seawater. Indeed, while the total inorganic carbon inventory of the ice increased throughout our study (because of the increasing ice thickness), the average concentration in the ice decreased (Figure 6b). Taking a final average TIC concentration in the sea ice of about 360 μmol kg−1 and assuming that the initial seawater TIC concentration was about 1950 μmol kg−1 (Figure 7) gives a total carbon loss from the sea ice (final thickness about 190 cm with an average density of 910 kg m−3) of nearly 3 mol m−2. While some of that carbon may have been lost to the atmosphere (see section 3), the same calculation for alkalinity (average sea ice concentration of 370 μmol kg−1 and an initial seawater concentration of 2040 μmol kg−1) gives a similar loss of nearly 3 mol m−2, and alkalinity could not be lost to the atmosphere.
 The carbon content and alkalinity of the surface waters beneath the ice did increase substantially over our sampling period (Figure 13). Over an average surface mixed layer of 14 m throughout our study, both TIC and AT increased by about 3 mol m−2, again consistent with what was lost from the ice. The salinity increase in the surface waters between November 2003 and February 2004 is also consistent with brine rejection during ice formation (resulting in an ice sheet 1.4 m thick with an average bulk salinity of 3.7). Given the concomitant increases in surface water TIC, alkalinity, and salinity, it is reasonable to assume that much of the inorganic carbon increase was also derived from the freezing ice. While brine rejection influenced the composition of the surface mixed layer, the density increase was apparently insufficient to initiate even shallow convection, because the mixed layer did not deepen throughout the winter, and there was no significant increase in carbon concentrations in deeper waters.
 While the local rivers of the southern Beaufort Sea did not contribute significantly to the surface mixed layer in Franklin Bay during the winter of 2003–2004, H218O (δ18O) data from our study indicate that mixing with other, apparently oceanic, water masses did contribute to the changes we observed in the surface waters. To estimate water mass changes in the mixed layer over winter, we analyzed the salinity and δ18O relative to meteoric water (MW; runoff), sea ice melt (SIM), and Polar Mixed Layer (PML) end-member values (Table 3). Depending on the application, end-members usually must be chosen which are appropriate to the specific site [Macdonald et al., 1995], and here, we used a salinity of 0 and δ18O of −18 ‰ for MW, which complies with many large runoff sources to the Arctic Ocean. For sea ice, we averaged the values measured on six ice cores collected in Franklin Bay during our study, excluding the top 35 cm to avoid snow contamination. The properties we chose for the saline end-member (PML) correspond to the bottom of the mixed layer in the salinity-δ18O plot. These values are close to those used by Macdonald et al.  for Beaufort Shelf waters. The contribution from MW and SIM in the surface waters varied systematically through the winter (Figure 14). Most significantly, while the ice melt fraction decreased and became negative (a result of net ice brine injection, relative to the average PML end-member), the fraction of meteoric water also decreased, indicating a subtle shift in the Franklin Bay waters during our study away from riverine and toward marine influences.
Table 3. End-Members Used to Identify Water Mass Variations
 Although we confirmed that the sea ice in Franklin Bay exported carbon into the underlying waters, that process does not necessarily contribute to a net CO2 sequestration. That is, simply depositing carbon into the surface waters can leave it available for release back into the atmosphere within a year, and export to deeper waters is required to isolate carbon from the atmosphere for longer periods of decades or more. Although dense shelf waters are sometimes exported from the shallow shelves of the southern Beaufort Sea [Melling, 1993], such export apparently was not occurring in Franklin Bay during our study, and the TIC increase was limited to the surface waters. On the other hand, a number of investigators have hypothesized that even without convection, brine rejection from sea ice can effectively isolate gases from the atmosphere, because the ice forms a freshwater lens at the surface as it melts [Jones and Coote, 1981; Hood, 1998]. The resulting strong density gradient would prevent the gas-rich, relatively high-salinity surface waters that had formed under the ice from interacting with the atmosphere, and if the freshwater layer is not degraded or removed (by wind) before it freezes again in the fall, the gases exported with brine during the winter before could remain in the ocean throughout the summer. Indeed, when we revisited the overwintering station during the summer of 2004, the subsurface waters were isolated from the atmosphere by a surface freshwater lens (Figure 15). Interestingly, the surface waters in Franklin Bay apparently never became supersaturated in CO2 under the ice, an unusual although not unheard of phenomenon [Semiletov et al., 2007], and thus, the question of outgassing when the ice cleared was moot. Therefore, it appears that the carbon exported from the ice and into the underlying waters may have remained there for much of the summer and possibly into the fall [see also Mucci et al., 2010].
 Our observations in Franklin Bay during the winter of 2004 indicate that the inorganic carbon cycle associated with first-year sea ice is dynamic and complex. Most notably: (1) large quantities of carbon dioxide move between the atmosphere and the ice, with a net release from the ice during winter; (2) the total carbon content of the ice increased throughout the winter season, primarily because of the growing ice volume, while the average carbon concentration decreased, due to drainage with brines and outgassing to the atmosphere; and (3) inorganic carbon is exported from the ice into the underlying waters with draining ice brines, although during our study that carbon did not appear to be sequestered into deeper waters.
 Very high CO2 partial pressures observed in the ice, as well as temporal and spatial variations in the TIC, alkalinity, and TOC profiles within the ice, indicate that ice carbon dynamics are controlled by a combination of processes, including temperature-mediated brine drainage and expulsion, CaCO3 precipitation and dissolution, and biological community progression. In Figure 16, we propose the following sequence of events. Initial ice formation between November and January coincided with the most rapid increase in salinity and inorganic carbon in the surface waters (Figure 13) partly because of brine drainage, which slowed during the winter (Figure 14) as the ice grew thicker. In February, when temperatures were dropping (Figure 3c) and the ice was still relatively thin, CaCO3 precipitation likely raised pCO2 within the ice (Figure 4) and contributed to a small upward CO2 flux from the ice (Figure 3), to the extent allowed by the rapidly closing brine channels. During the coldest period in early March, pCO2 in the ice dropped (Figure 4), probably due to increasing gas solubility. Rising temperatures at the end of March would have contributed to the increases in upward CO2 fluxes (Figure 3), while signals within the ice became more variable (Figures 4 and 10). The slow increase in the “daytime” pCO2 measurements in the ice through the spring (Figure 4b) are consistent with increasing temperatures and possibly associated with bacterial activity (Table 2) before primary production was able to dominate [Rysgaard et al., 2008]. Even at their lowest, the sea ice pCO2 values were still much higher than the atmosphere, so that as the ice warmed and brine channels reopened, large upward fluxes were observed by the eddy covariance system (Figure 3). Based on the late spring work of Delille et al.  in Antarctic landfast ice, as spring progresses primary production and possibly CaCO3 dissolution eventually draw down pCO2 within the ice, presumably drawing CO2 down from the atmosphere, as well. Once the ice has cleared, conditions revert to those ‘typical’ of summertime polar waters, with CO2 drawdowns into biologically productive waters [Mucci et al., 2010].
 The very rough budget we were able to assemble for total carbon in the sea ice indicates that the ice lost about 3 mmol C m−2 to the atmosphere (section 3) and lost about 3 mmol C m−2 to the underlying water (section 5), while gaining about 0.1 mmol C m−2 (section 4.2). The majority of these changes were in the inorganic component (TIC), with TOC cycling making only minor contributions (section 4.3). While it appears there may be another, yet unidentified, source of carbon in Franklin Bay, all the fluxes and inventories we have presented are associated with very large uncertainties. With improved methodologies, future studies should be able to more confidently identify whether the budgetary gap implied here is real.
 Most importantly, our combined data sets demonstrate that rather than serving as an impermeable barrier, preventing CO2 from moving between the atmosphere and the ocean, the sea ice is an active participant in the carbon cycle of polar waters. We cannot yet assess the net effect of sea ice CO2 fluxes on the annual integrated source versus sink balance of ice-covered waters, both because of uncertainties in our measurements and because we lack data from the fall freezeup period. Competing hypotheses predict that freezing seawater could either release CO2 to the atmosphere [Nomura et al., 2006; Nagurnyi, 2008] or absorb atmospheric CO2 [Anderson et al., 2004]. Resolution of this question awaits results from detailed autumn studies.
 This work could not have been completed without the assistance of many dedicated people. Mike Arychuk, Pascale Collin, Marty Davelaar, and Constance Guignard provided invaluable assistance with logistics and sample analyses. Help with sampling on and under the ice from Eric Braekevelt, Carrie Breneman, Shelly Carpenter, Mark Gordon, Dan Leitch, C. J. Mundy, Takahashi Ota, Carlos Pedrós-Alio, Sebastien Roy, and Shinya Yamamoto is gratefully acknowledged. Thanks go to J.-L. Tison, B. Delille, and S. Papadimitriou for very useful conversations that helped to crystallize our discussion. We also thank Sylvain Blondeau and all the CASES chief scientists and CTD rosette operators as well as the captains and crew of the NGCC Amundsen for all their help and support. Finally, thanks go to two anonymous reviewers, whose encouraging reviews identified yet a few more possible explanations and helped us clarify a number of issues. Funding for this work was provided by the Fisheries and Oceans Canada Strategic Science Fund, an NSERC network grant for the Canadian Arctic Shelf Exchange Study, NSERC Discovery grants, and NSF awards OPP0327244 and DGE-980713.