CO2 flux over young and snow-covered Arctic pack ice in winter and spring

Rare CO2 flux measurements from Arctic pack ice show that two types of ice contribute to the release of CO2 from the ice to the atmosphere during winter and spring: young, thin ice with a thin layer of snow and older (several weeks), thicker ice with thick snow cover. Young, thin sea ice is characterized by high salinity and high porosity, and snow-covered thick ice remains relatively warm (>−7.5 C) due to the insulating snow cover despite air temperatures as low as −40 C. Therefore, brine volume fractions of these two ice types are high enough to provide favorable conditions for gas exchange between sea ice and the atmosphere even in mid-winter. Although the potential CO2 flux from sea ice decreased due to the presence of the snow, the snow surface is still a CO2 source to the atmosphere for low snow density and thin snow conditions. We found that young sea ice that is formed in leads without snow cover produces CO2 fluxes an order of magnitude higher than those in snow-covered older ice (+1.0± 0.6 mmolCm−2 day−1 for young ice and +0.2± 0.2 mmolC m−2 day−1 for older ice).


Introduction
Arctic sea ice is changing dramatically, with rapid declines in summer sea ice extent and a shift towards younger and thinner first-year ice rather than thick multi-year ice (e.g., Stroeve et al., 2012;Meier et al., 2014;Lindsay and Schweiger, 2015).Although the effects of sea ice formation and melting on biogeochemical cycles in the ocean have previously been discussed (e.g., Vancoppenolle et al., 2013), the effects of sea ice freeze and melt processes on carbon dioxide (CO 2 ) exchange with the atmosphere are still largely unknown (Parmentier et al., 2013).
Recent CO 2 flux measurements on sea ice indicate that sea ice is an active component in gas exchange between ocean and atmosphere (Nomura et al., 2013;Geilfus et al., 2013Geilfus et al., , 2014;;Delille et al., 2014;Brown et al., 2015;Kotovitch et al., 2016).The sea ice CO 2 fluxes depend on (a) the difference in the partial pressure of CO 2 (pCO 2 ) between the sea ice surface and air, (b) brine volume fraction at the ice-snow interface, (c) ice surface condition including the snow deposited on ice, and (d) wind-driven pressure pumping through the Published by Copernicus Publications on behalf of the European Geosciences Union.
snow.For (a), it is known that the air-sea ice CO 2 flux is driven by the differences in pCO 2 between the sea ice surface and atmosphere (e.g., Delille et al., 2014;Geilfus et al., 2014).Brine pCO 2 changes due to processes within the sea ice, such as thermodynamic process (e.g., Delille et al., 2014), biological activity (e.g., Delille et al., 2007;Fransson et al., 2013;Rysgaard et al., 2013), and calcium carbonate (CaCO 3 ; ikaite) formation and dissolution (e.g., Papadimitriou et al., 2012).When pCO 2 in brine is higher than that of air pCO 2 , brine has the potential to release CO 2 to the atmosphere.Brine volume fraction (b) controls the permeability of sea ice (Golden et al., 1998) and thus CO 2 fluxes (Delille et al., 2014;Geilfus et al., 2014).The air-sea ice CO 2 flux is also strongly dependent on the sea ice surface conditions (c) (Nomura et al., 2010a(Nomura et al., , 2013;;Geilfus et al., 2013Geilfus et al., , 2014;;Barber et al., 2014;Brown et al., 2015;Fransson et al., 2015).Nomura et al. (2013) proposed that snow properties (e.g., water equivalent) are important factors affecting gas exchange processes on sea ice.In addition, frost flowers (vapor-deposited ice crystals that wick brine from the sea ice surface) promote CO 2 flux from the ice to the atmosphere (Geilfus et al., 2013;Barber et al., 2014;Fransson et al., 2015).Finally, for (d), it is thought that CO 2 flux is affected by wind pumping through the snow pack (Massman et al., 1995;Takagi et al., 2005) in which the magnitude of CO 2 flux through snow or underlying soil (e.g., Takagi et al., 2005) can increase the transport relative to molecular diffusion by up to 40 % (Bowling and Massman, 2011).These results were mainly found over land-based snow (soil and forest), and thus they are still poorly understood over sea ice (Papakyriakou and Miller, 2011).
In addition to the processes described above, the CO 2 flux over sea ice may also be influenced by the temperature difference between the ice surface and the atmosphere.This has been shown in previous studies in dry snowpacks over land surfaces.These studies show that there is an unstable air density gradient due to heating at the bottom producing a strong temperature difference between the bottom and top of the snowpack (e.g., Powers et al., 1985;Severinghaus et al., 2010).This produces air flow within the snowpack, which is a potentially significant contributor to mixing and transport of gas and heat within the snowpack.We expect that this process would also occur in snow over sea ice, especially during the wintertime when air temperatures are coldest and the temperature difference between sea ice surface (snow bottom) and atmosphere is largest (e.g., Massom et al., 2001).Generally, the sea ice surface under thick snow cover is warm due to the heat conduction from the bottom of sea ice and the insulating effect of the snow cover, and a strong temperature difference between the sea ice surface and atmosphere is observed (e.g., Massom et al., 2001).Such a temperature difference would produce an unstable air density gradient and upward transport of air containing CO 2 degassed at the sea ice surface, thereby enhancing CO 2 exchange between sea ice and atmosphere.
In the ice-covered Arctic Ocean, storm periods which produce high wind speeds and open leads are also important for air-to-sea CO 2 fluxes (Fransson et al., 2017) due to the undersaturation of the surface waters in CO 2 with respect to the atmosphere.In addition, the subsequent ice growth and frost flower formation in open leads promote ice-to-air CO 2 fluxes in winter (e.g., Barber et al., 2014).Given the fact that Arctic sea ice is shrinking and shifting from multi-year ice to first-year ice (e.g., Stroeve et al., 2012;Meier et al., 2014;Lindsay and Schweiger, 2015), the area of open ocean and thinner seasonal ice is increasing.Thus, a potential consequence may be increased contribution of open ocean surface and/or thinner sea ice to the overall CO 2 fluxes of the Arctic Ocean.The dynamics of the thinner ice pack, through formation of leads and new ice, will play an important role in the gas fluxes from the ice pack.However, there is a definite lack of information on sea ice processes during wintertime due to the difficulty in acquiring observations in winter pack ice, as reflected by the fact that most of the previous winter CO 2 flux measurements have been take over landfast ice.
The Norwegian young sea ICE (N-ICE2015) campaign in winter and spring 2015 provided opportunities to examine CO 2 fluxes between sea ice and atmosphere in a variety of snow and ice conditions in pack ice north of Svalbard.Formation of leads and their rapid refreezing allowed us to examine air-sea ice CO 2 fluxes over thin young sea ice, occasionally covered with frost flowers in addition to the snowcovered older ice that covers most of the pack ice area.The objectives of this study were to understand the effects of (i) thin sea ice and frost flower formation on the air-sea ice CO 2 flux in leads, (ii) effect of snow-cover on the air-sea ice CO 2 flux over thin, young ice in the Arctic Ocean during winter and spring seasons, and (iii) of the effect of the temperature difference between sea ice and atmosphere (including snow cover) on the air-sea ice CO 2 flux.

Study area
This study was performed during N-ICE2015 campaign with R/V Lance in the pack ice north of Svalbard from January to June 2015 (Granskog et al., 2016).Air-sea ice CO 2 flux measurements were carried out from January to May 2015 during the drift of floes 1, 2, and 3 of the N-ICE2015 campaign (Figs. 1 and 2, Table 1).The ice pack was a mixture of young ice, first-year ice, and second-year ice (Granskog et al., 2017), and both the first-and second-year ice had a thick snow cover (Merkouriadi et al., 2017;Rösel et al., 2018).Air-sea ice CO 2 flux measurements were made over young ice (YI stations), first-year ice (FI stations), and old ice (multi-year ice) (OI station).In the N-ICE2015 study region, the modal ice thickness was about 1.3-1.5 m and the modal snow thickness was about 0.5 m (Rösel et al., 2018).Formation of leads and their rapid refreezing provided us the opportunity to examine air-sea ice CO 2 fluxes over thin sea ice, occasionally covered with frost flowers at station YI1 (Fig. 2 and Table 1).Air temperature and wind speed were measured at a 10 m weather mast on the ice floe installed about 400 m away from R/V Lance (Cohen et al., 2017).

CO 2 flux measurements
The air-sea ice CO 2 flux was measured with LI-COR 8100-104 chambers connected to a LI-8100A soil CO 2 flux system (LI-COR Inc., USA) (Fig. 2).This enclosed chamber method has been widely applied over snow and sea ice (e.g., Schindlbacher et al., 2007;Geilfus et al., 2015).Two chambers were connected in a closed loop to the infrared gas analyzer (LI-8100A, LI-COR Inc., USA) to measure CO 2 concentration through the multiplexer (LI-8150, LI-COR Inc., USA) with an air pump rate at 3 L min −1 .Power was supplied by a car battery (8012-254, Optima Batteries Inc., USA).Four CO 2 standards (324-406 ppmv) traceable to the WMO scale (Inoue and Ishii, 2005) were prepared to calibrate the CO 2 gas analyzer prior to the observations.CO 2 flux was measured in the morning or in the afternoon during low-wind conditions (Table 2), to minimize the effect of wind on the flux (Bain et al., 2005).
One chamber was installed over undisturbed snow or frost flowers on the ice surface.The chamber collar was inserted 5 cm into the snow and 1 cm into ice at the frost flower site to avoid air leaks between the inside and outside of the chamber.The second chamber was installed on bulk sea ice af-Table 2. Station, snow density and water equivalent, brine volume fraction and temperature for sea ice (top 20 cm), brine temperature, salinity, dissolved inorganic carbon (DIC), total alkalinity (TA), pCO 2 (pCO 2 b ), and atmospheric temperature, wind speed, pCO 2 (pCO 2 a ) a and pCO 2 b−a .ter removing the snow or frost flowers.Flux measurements were begun immediately in order to minimize the changes of the ice surface condition.In order to evaluate the effect of removing snow on the ice surface temperature, temperature was monitored during CO 2 flux measurements at station FI6.A temperature sensor (RTR 52, T & D Corp., Japan) was installed in the top of the ice (1 cm) surface after snow removal.During the first CO 2 flux measurements (about 30 min), the ice surface temperature was stable at −5.8 • C, suggesting that the effect of removing snow on the variation of sea ice surface temperature was negligible within 30 min.The ice surface temperature decreased from −5.8 to −8.0 • C at 200 min after removal of snow.Therefore, in this paper, the data of the initial 30 min of CO 2 flux measurement after removal of snow or frost flowers were used.The chamber was closed for 20 min in a sequence.The 20 min time period was used because CO 2 fluxes over sea ice are much smaller than over land.The CO 2 concentrations within the chamber were monitored to ensure that they changed linearly throughout the measurement period (example given in Fig. 3).The CO 2 flux (mmol C m −2 day −1 ) (positive value indicates CO 2 being released from ice surface to air) was calculated based on the changes of the CO 2 concentration within the headspace of the chamber with LI-COR software (model: LI8100PC Client v.3.0.1.).The mean coefficient of variation for CO 2 flux measurements was less than 3.0 % for CO 2 flux values larger than ±0.1 mmol C m −2 day −1 .For CO 2 flux values smaller than ±0.1 mmol C m −2 day −1 , the mean coefficient of variation for CO 2 flux measurements was higher than 3.0 %, suggesting that the detection limit of this system is about 0.1 mmol C m −2 day −1 .In this paper, we express the CO 2 flux measured over the snow and frost flowers as F snow , and F ff , respectively.The flux measured directly over the sea ice surface either on snow-free ice or after removal of snow and frost flowers as F ice .F snow and F ff are the natural flux (snow and frost flowers are part of the natural system), and F ice is the potential flux in cases when snow or frost flowers are removed.While removal of snow and frost flowers is an artificial situation, comparisons between F ice and F snow or F ff provide information about the effect of snow and frost flowers on the CO 2 flux.Therefore, in this study, we examine both situations for CO 2 flux.

Sampling of snow, frost flowers, brine, and sea ice
For salinity measurements, separate samples were taken for snow only, snow and frost flowers, and sea ice surface scrapes.The samples were taken using a plastic shovel, placed into plastic bags and stored in an insulated box for transport to the ship lab for further processing.Samples were melted slowly (2-3 days) in the dark at +4 • C. The temperature of the snow and frost flower samples was measured during CO 2 flux measurements (approximately 60 min after the onset of the CO 2 flux measurement) using a needle-type temperature sensor (Testo 110 NTC, Brandt Instruments, Inc., USA).The accuracy of this sensor is ±0.2 • C. Snow density was obtained using a fixed volume sampler (Climate Engineering, Japan) and weight measurement.The depth of the snow pack and frost flowers was also recorded using a ruler.
Brine was also collected at stations FI3-6 for salinity, dissolved inorganic carbon (DIC), and total alkalinity (TA) measurements.Brine was collected from sack holes as described in Gleitz et al. (1995).The sack holes were drilled using a 9 cm diameter ice corer (Mark II coring system, Kovacs Enterprises, Inc., USA) to a depth of 30 cm.The sack holes were then covered with a lid of 5 cm thick urethane to reduce heat and gas transfer between brine and atmosphere.When brine accumulated at the bottom of the sack holes (approximately 15 min), it was collected with a plastic syringe (AS ONE Corporation, Japan) and kept in 500 mL unbreakable plastic bottles (I-Boy, AS ONE Corporation, Japan) in order to facilitate safe transport to the sampling sites in cold and harsh conditions.The brine bottles were filled without head space and immediately stored in an insulated box to prevent freezing.Immediately after returning to the ship, the brine samples were transferred to 250 mL borosilicate bottles (DURAN Group GmbH, Germany) for DIC and TA measurements using tubing to prevent contact with air.The samples were preserved with saturated mercuric chloride (HgCl 2 , 60 µL for a 250 mL sample) and stored in the dark at +10 • C until analyses were performed at the Institute of Marine Research, Norway.
Sea ice was collected by the same ice corer as described for brine collection and at the same location as snow and frost flowers were collected.Sea ice temperature was measured by the same sensor as described for snow.For the ice cores, the temperature sensor was inserted in small holes drilled into the core.The core was then cut with a stainless steel saw into 10 cm sections and stored in plastic bags for subsequent salinity measurements.The ice core sections were kept at +4 • C and melted in the dark prior to measurement.

Sample analysis
Salinities for melted snow, frost flowers, sea ice, and brine were measured with a conductivity sensor (Cond 315i, WTW GmbH, Germany).For calibration of salinity measurement, a Guildline PORTASAL salinometer model 8410A, standardized by International Association for the Physical Sciences of the Oceans (IAPSO) standard seawater (Ocean Scientific International Ltd, UK), was used.The accuracy of this sensor was ±0.003.
Analytical methods for DIC and TA determination are fully described in Dickson et al. (2007).DIC in brine was determined using gas extraction of acidified sample followed by coulometric titration and photometric detection using a Versatile INstrument for the Determination of Total inorganic carbon and titration Alkalinity (VINDTA 3C, Germany).TA of brine was determined by potentiometric titration of 40 mL sample in open cell with 0.05 N hydrochloric acid using a Titrino system (Metrohm, Switzerland).The average SD for DIC and TA, determined from replicate sample analyses from one sample, was within ±2 µmol kg −1 for both DIC and TA.The accuracy of the DIC and TA measurements were ±2 µmol kg −1 for both DIC and TA, as estimated using Certified Reference Materials (CRM, provided by A. G. ture, salinity, DIC, and TA of brine using the carbonate speciation program CO2SYS (Pierrot et al., 2006).The calculated pCO 2 b values (Table 2) varied within 1.7 % when DIC and TA values were changed within the SD (±2 µmol kg −1 ).We used the carbonate dissociation constants (K 1 and K 2 ) of Mehrbach et al. (1973) as refit by Dickson and Millero (1987) and the KSO 4 determined by Dickson (1990).The conditional stability constants used to derive pCO 2 are only valid for temperatures above 0 • C and salinities between 5 and 50.Studies in spring ice indicated that seawater thermodynamic relationships may be acceptable in warm and lowsalinity sea ice (Delille et al., 2007).In sea ice brine at even moderate brine salinities of 80, Brown et al. (2014) found that measured and calculated values of the CO 2 system parameters can differ by as much as 40 %.However, because the CO 2 system parameters are much more variable in sea ice than in seawater, sea ice measurements demand less precision than those in seawater.Fransson et al. (2015) performed one of the few detailed analyses of the internal consistency using four sets of dissociation constants and found that the deviation between measured and calculated DIC varied between ±6 and ±11 µmol kg −1 , respectively.This error in calculated DIC was considered insignificant in relation to the natural variability in sea ice.
The water equivalent was computed for snow by multiplying snow thickness by snow density (Jonas et al., 2009).Brine volume of sea ice was calculated from the temperature and salinity of sea ice according to Cox and Weeks (1983) and Petrich and Eicken (2010).

Air temperature
Air temperature is shown in Fig. 4.During the study period, the air temperature varied considerably from a low of −41.3 • C (30 January) to a high of +1.7 • C (15 June) (Hudson et al., 2015).Even in wintertime (from January to March), rapid increases of air temperature from less than −30 up to −0.2 • C (e.g., 18 February) were observed.In springtime (from April to June) the air temperature increased continuously, and from 1 June air temperatures were near 0 • C, although rapid increases (and subsequent decreases) of air temperature to near 0 • C were observed on two occasions in mid-May (Cohen et al., 2017).

Characteristics of snow, sea ice, and frost flowers
The snow and ice thickness at the observation sites ranged between 0.0 and 60.0 cm and between 15.0 and > 200 cm, respectively (Table 1).The thin snow and ice represent newly formed ice in leads at station YI1.The thickness of the frost flowers ranged from 1.0 to 2.5 cm.
Figure 5 shows vertical profiles of snow and ice temperature and salinity in the top 20 cm of ice.Temperatures within the snowpack depended on the air temperature at the time of observation.However, the bottom of the snow and the surface of the sea ice were relatively warm (T > −7.5 • C), except for the frost flower station YI1 and the multi-year ice station OI1 (Fig. 5a and Table 2).High salinities (S > 18.6) characterized the bottom of the snow and the surface of the sea ice, except for the multi-year ice station OI1 (Fig. 5b).At the multi-year ice station OI1, salinity was zero through the snow and top of sea ice.Salinity of frost flowers was up to 92.8 for the thin ice station YI1 (Fig. 5b).Snow density and water equivalent ranged from 268 to 400 kg m −3 and 11 to 180 kg m −2 , respectively (Table 2).

Physical and chemical properties of brine
The brine volume fraction, temperature, salinity, DIC, TA, and calculated pCO 2 are summarized in Table 2. Brine volume fraction in the top 20 cm of ice was between 9 to 17 %, except for the value of 0 % at the multi-year ice station OI1 (Table 2).Brine temperatures and salinity ranged from −5.3 to −3.3 • C and 51.8 to 86.6, respectively.DIC and TA of brine ranged from 3261 to 4841 µmol kg −1 and 3518 to 5539 µmol kg −1 , respectively.The pCO 2 of brine (pCO 2 b ) (334-693 µatm) was generally higher than that of atmosphere (pCO 2 a ) (401 ± 7 µatm), except for station FI4.

CO 2 flux
Table 3 summarizes the CO 2 flux measurements for each surface condition.For undisturbed natural surface conditions, i.e., measurements directly on the snow surface (F snow ) or the frost flowers (F ff ) on young ice, the mean CO 2 flux was +0.2 ± 0.2 mmol C m −2 day −1 for F snow and +1.0 ± 0.6 mmol C m −2 day −1 for F ff .The potential flux in cases when snow or frost flowers had been removed (F ice ) was +2.5 ± 4.3 mmol C m −2 day −1 .The air-sea ice CO 2 fluxes measured over the ice surface (F ice ) increased with increasing differences in pCO 2 between brine and atmosphere ( pCO 2 b−a ) with significant correlation (R 2 = 0.9, p < 0.02), but this was not the case for F snow (R 2 = 0.0, p < 0.96) (Fig. 6).

Effect of snow cover on the physical properties of sea ice surface
In this study, we examined CO 2 fluxes between the sea ice and atmosphere in a wide range of air temperatures and diverse snow and ice conditions (Table 2).The bottom of the snow pack and the surface of the sea ice remained relatively warm (> −7.5 • C) (Fig. 5a, Table 2), except for stations OI1 and YI1, even though air temperature was sometimes below −40 • C (Fig. 4).Relatively warm ice temperatures were likely due to the upward heat transport from the bottom of the ice and in some cases the thick insulating snow cover, except for stations OI1 and YI1 (Table 2).Therefore, snow acted as thermal insulator over sea ice, and in general the snow depths observed during N-ICE2015 point towards this being representative for first-year and second-year or older ice in the study region in winter 2015 (Rösel et al., 2018).
The young and first-year ice surfaces were characterized by high salinities (Fig. 5b).During sea ice formation, upward brine transport to the snow pack occurs (e.g., Toyota et al., 2011).In addition, brine within the sea ice was not completely drained as compared to that of multi-year ice.Furthermore, formation of frost flowers and subsequent wicking up of surface brine into the frost flowers also provides high salinity at the surface of sea ice (Kaleschke et al., 2004 a Data of first measurement after removal of snow or frost flower.b "−" indicates no data.c Number of measurements in bracket.d Data of station OI1 were not included.fus et al., 2013;Barber et al., 2014;Fransson et al., 2015) as observed in this study (S > 92) (Fig. 5b).Snowfall over the frost flowers would have preserved the high salinity at the bottom of snow pack and top of sea ice for young and firstyear ice.
As a result of the combination of the relatively high temperature and high salinity at the top of sea ice, brine volume fractions in the upper parts the sea ice were high, up to 17 % (Table 2).It has been shown that ice permeability increases by an order of magnitude when brine volume fraction is greater than 5 % as compared to when the brine volume fraction is less than 5 % (Golden et al., 1998;Pringle et al., 2009;Zhou et al., 2013).A brine volume fraction of 5 % would correspond to a temperature of −5 • C for a bulk ice salinity of 5 -the so-called "law of fives" (Golden et al., 1998).Because sea ice temperatures are low, thereby reducing the permeability in winter season, air-sea ice CO 2 flux is generally at its minimum in the winter (e.g., Delille et al., 2014).However, in our study, the brine volume fractions were generally > 9 %, except for station OI1 with fresh ice at the surface, providing conditions for active gas exchange within sea ice and between sea ice and atmosphere.This situation was likely made possible due to the thick snow cover and relatively thin and young sea ice.

CO 2 fluxes over different sea ice surface types
The CO 2 flux measurements over different surface conditions indicate that the snow cover over sea ice affects the magnitude of air-sea ice CO 2 flux, especially for stations FI5 and FI6 (Table 3).For undisturbed natural surface ditions, flux measured directly over snow-covered first-year ice and young ice with frost flowers (F snow and F ff ) was lower in magnitude than that for potential flux obtained directly over the ice surface after removing snow (F ice ) for stations FI5, FI6, and YI1.
F ff indicates that the frost flower surface on young thin ice is a CO 2 source to the atmosphere and F ff was higher than F snow , except for station FI1.This finding was consistent with the previous studies (Geilfus et al., 2013;Barber et al., 2014;Fransson et al., 2015).At multi-year ice station OI1, neither snow or ice surface acted as a CO 2 source/sink.The surface of multi-year ice did not contain any brine (Fig. 5b and Table 2), and the top of the ice was clear, colorless, and very hard, suggesting superimposed formation at the top of sea ice.This situation would be similar as for freshwater ice and superimposed ice as these non-porous media block gas exchange effectively at the sea ice surface (Delille et al., 2014).Snow ice and superimposed ice were frequently found in second-year ice cores during N-ICE2015 (Granskog et al., 2017), so the "blocking" of gas exchange in second-year and multi-year ice may be a widespread process in the Arctic.
The magnitude of positive F snow is less than F ice for stations FI5 and FI6 (Table 3), indicating that the potential CO 2 flux from sea ice decreased due to the presence of snow.Previous studies have shown that snow accumulation over sea ice effectively impedes CO 2 exchange (Nomura et al., 2013;Brown et al., 2015).Nomura et al. (2013) reported that 50-90 % of the potential CO 2 flux was reduced due to the presence of snow/superimposed ice at the water equivalent of 57-400 kg m −2 , indicating that the snow properties are an important factor that controls the CO 2 exchange through a snowpack.Comparisons between stations FI5 and FI6 for F snow /F ice ratio (0.23 for FI5 and 0.02 for FI6) and water equivalent (11 kg m −2 for FI5 and 127 kg m −2 for FI6) indicate that the potential CO 2 flux is reduced (80 % for FI5 and 98 % for FI6 of the potential CO 2 flux) with increasing water equivalent.Although the magnitude of the potential CO 2 flux through the sea ice surface decreased by the presence of snow for stations FI5 and FI6 (Table 3), the snow surface still presents a CO 2 source to the atmosphere for low snow density and shallow depth conditions (e.g., +0.6 mmol C m −2 day −1 for FI5).
For F ice , there were negative CO 2 fluxes at stations FI3 and FI4 (−0.6 mmol C m −2 day −1 for FI3 and −0.8 mmol C m −2 day −1 for FI4) (Table 3).These fluxes corresponded to low or negative pCO 2 b−a (Table 2 and Fig. 6).Negative CO 2 fluxes should correspond to negative pCO 2 b−a .Therefore, the uncertainty for the calculation of carbonate chemistry may be one reason for the discrepancy in pCO 2 calculation at station FI3 (Brown et al., 2014).

Comparison to earlier studies on sea ice to air CO 2 flux
The CO 2 fluxes measured over the undisturbed natural surconditions (F snow and F in this study ranged from +0.1 to +1.6 mmol C m −2 day −1 (Table 3), which are at the lower end of the reported range based on the chamber method and eddy covariance method for natural and artificial sea ice (−259.2 to +74.3 mmol C m −2 day −1 ) (Zemmelink et al., 2006;Nomura et al., 2006Nomura et al., , 2010aNomura et al., , b, 2013;;Miller et al., 2011;Papakyriakou and Miller, 2011;Geilfus et al., 2012Geilfus et al., , 2013Geilfus et al., , 2014;;Barber et al., 2014;Delille et al., 2014;Sørensen et al., 2014;Brown et al., 2015;Kotovitch et al., 2016).Direct comparison to these previous studies is complicated because CO 2 flux measurements with both chamber and eddy covariance techniques were used during different conditions and ice surface characteristics.In addition, discrepancies between chamber and eddy covariance measurements of airice CO 2 fluxes have been repeatedly observed.The footprint size of CO 2 exchange measured with the two approaches (Zemmelink et al., 2006(Zemmelink et al., , 2008;;Burba et al., 2008;Amiro, 2010;Miller et al., 2011Miller et al., , 2015;;Papakyriakou and Miller, 2011;Sørensen et al., 2014) may be one reason for the large difference.The eddy covariance method reflects a flux integrated over a large area that can contain several different surface types.Therefore, eddy covariance appears to be more useful for understanding fluxes at large spatial and temporal scales.In contrast, the chamber method reflects the area where chamber was covered, and it is useful for understanding the relationship between fluxes and ice surface conditions on smaller scales.The different spatial scales of the two methods may therefore be one reason for the discrepancy in CO 2 flux measurements.
Comparison of the natural CO 2 flux range (+0.1 to +1.6 mmol C m −2 day −1 for F snow and F ff ) (Table 3) with previous estimates derived from the chamber method (−5.2 to +6.7 mmol C m −2 day −1 ) (Nomura et al., 2006(Nomura et al., , 2010a(Nomura et al., , b, 2013;;Geilfus et al., 2012Geilfus et al., , 2013Geilfus et al., , 2014;;Barber et al., 2014;Delille et al., 2014;Brown et al., 2015;Kotovitch et al., 2016) (these studies include both natural and potential fluxes) shows that CO 2 fluxes during the N-ICE2015 experiment are at the lower end of positive values.However, our potential CO 2 flux (F ice ) was a larger CO 2 source (up to +11.8 mmol C m −2 day −1 ) than reported in previous studies (+6.7 mmol C m −2 day −1 ).In our study, the maximum potential flux (+11.8 mmol C m −2 day −1 ) was obtained for F ice at station FI6 (Table 3).In this situation, pCO 2 b−a (293 µatm) was the highest (Table 2 and Fig. 6), and it is reasonable to consider this as the highest magnitude of positive CO 2 flux within our study.However, a previous study by closed chamber method showed that even for a similar pCO 2 b−a (297 µatm) and brine volume fraction (10-15 %), the CO 2 flux was +0.7 mmol C m −2 day −1 for artificial sea ice with no snow in the tank experiment (Nomura et al., 2006).
The CO 2 flux between the sea ice and overlying air can be expressed by the following equation: where r b is the ratio of surface of the brine channel to sea ice surface, and we assume that the value of r b is equal to brine volume fraction, k is the gas transfer velocity, α is the solubility of CO 2 (Weiss, 1974), and pCO 2 b−a is the difference in pCO 2 between brine and atmosphere.The equation is based on the fact that CO 2 transfer between seawater and air is controlled by processes in the near-surface water (Liss, 1973).The gas transfer velocity (k) calculated from F , r b , α, and pCO 2 b−a was 5.12 m day −1 for F ice at station FI6 and 0.29 m day −1 for the tank experiment examined in Nomura et al. (2006).This result clearly indicates that the gas transfer velocity for F ice at station FI6 is higher than that of the tank experiment examined in Nomura et al. (2006) even with very similar pCO 2 b−a and brine volume fraction.
Here, we surmise that the gas transfer velocity, and thereby CO 2 flux, is greatly enhanced by the temperature difference between sea ice surface and atmosphere.Previous studies indicate that there is an unstable air density gradient in a dry snowpack due to basal heating and the strong temperature difference develops between bottom and top of snow (e.g., Powers et al., 1985;Severinghaus et al., 2010), which enhances the flow of air through the snowpack.We propose that the mixing and transport of gas within the snowpack could also occur over sea ice.Because temperatures at the bottom of snow and the top of sea ice were relatively warm due to a thick insulating snow over sea ice, there was a strong temperature difference between sea ice surface and atmosphere when air temperature was low (Fig. 5a and Table 2).For station FI6, the temperature difference between the sea ice surface and atmosphere was 20.2 • C after snow removal.In contrast, in the tank experiment by Nomura et al. (2006), the temperature difference between sea ice surface (top 1.5 cm) and air in the headspace was only 4.5 • C.
Figure 6 shows the relationship between mean air-sea ice CO 2 fluxes and temperature difference between ice and atwww.biogeosciences.net/15/3331/2018/Biogeosciences, 15, 3331-3343, 2018 mosphere.The strong dependence of CO 2 flux with temperature difference (T ice -T a ) was observed, especially for F ff and F ice (R 2 > 0.7, p < 0.01, linear fitting) (Fig. 6).Due to the high brine volume fractions (Table 2), the sea ice surface had enough permeability for gas exchange.In addition, ice temperatures were similar for young and first-year ice (Table 2), indicating that pCO 2 at the top of the sea ice and CO 2 flux would be of similar order of magnitude if thermodynamic processes dominated.Therefore, our results suggest that the CO 2 fluxes, even over the frost flowers as a natural condition, would be enhanced by the upward transport of air containing high CO 2 from the surface of sea ice to the atmosphere due to the strong temperature difference between sea ice surface and atmosphere.Although the presence of snow on sea ice has potential to produce a larger temperature difference between sea ice surface and atmosphere and promote the upward transport, the magnitude of the CO 2 flux decreased due to the presence of snow.However, for young sea ice with frost flowers (e.g., station YI1), ice surface temperature was warm (Table 2), suggesting that CO 2 flux would be enhanced by the large temperature difference between sea ice surface and atmosphere.

Conclusions
We measured CO 2 fluxes along with sea ice and snow physical and chemical properties over first-year and young sea ice north of Svalbard in the Arctic pack ice.Our results suggest that young thin snow-free ice, with or without frost flowers, is a source of atmospheric CO 2 due to the high pCO 2 and salinity and relatively high sea ice temperature.Although the potential CO 2 flux from the sea ice surface decreased due to the presence of snow, the snow surface still presents a modest CO 2 source to the atmosphere for low snow density and shallow depth situations.The highest ice-to-air fluxes were observed over thin young sea ice formed in leads.During N-ICE2015 the ice pack was dynamic, and formation of open water was associated with storms, where new ice was formed.The subsequent ice growth in these leads is especially important for the ice-to-air CO 2 fluxes in winter since the flux from young ice is an order of magnitude larger than from snow-covered first-year and older ice.

Figure 1 .
Figure 1.Location map of the sampling area north of Svalbard during N-ICE2015.Image of the sea ice concentrations (a) and station map (b) were derived from Special Sensor Microwave Imager (SSM/I) satellite data for mean of February 2015 and from Sentinel-1 (Synthetic Aperture Radar Sensor) satellite data, respectively.
Temperature Wind speed pCO 2 a pCO 2 b−a (kg m −3 ) pCO 2 a (µatm) was calculated from CO 2 concentration (ppmv) at Ny-Ålesund, Svalbard (http://www.esrl.noaa.gov/gmd/dv/iadv/),taking into account the saturated water vapor and atmospheric pressures at sampling day.b Mean values for column.c "−" indicates no data.Due to logistical constraints, data of snow, sea ice, and brine were not obtained.

Figure 2 .
Figure 2. Photographs of the CO 2 flux chamber system at station YI1 north of Svalbard on 13 March 2015.CO 2 flux chamber was installed over the frost flowers on the new thin ice in the refreezing lead.

Figure 3 .
Figure3.Example of the temporal variation in CO 2 concentration ( CO 2 ) in the chambers installed at station YI1 that is use to calculate the CO 2 flux.CO 2 indicates the change in CO 2 concentration inside the chamber since the chamber was closed.

Figure 4 .
Figure 4. Time series of air temperature measured at the weather mast over the ice floe (10 m height) (Hudson et al., 2015).Blank period indicates no data.Colored symbols indicate the date for the chamber flux measurements.The horizontal dashed line indicates air temperature of 0 • C.

Figure 5 .Figure 6 .
Figure 5. Vertical profiles of temperature (a) and salinity (b) in snow and sea ice (top 20 cm).The horizontal line indicates snow-ice interface.area indicates sea ice.The triangle in panel (a) indicates the air temperature for each station.For stations FI7 and YI2 and 3, we have no salinity data.

Table 1 .
Station, date for CO 2 flux measurement, position, floe number, surface condition, ice type and thickness of snow, frost flowers, and sea ice.

Table 3 .
CO 2 flux measured over the snow (F snow ), frost flowers (F ff ) and ice surface (F ice ).Natural flux was emphasized as bold.