The seasonal cycle of ocean‐atmosphere CO2 flux in Ryder Bay, west Antarctic Peninsula

Approximately 15 million km2 of the Southern Ocean is seasonally ice covered, yet the processes affecting carbon cycling and gas exchange in this climatically important region remain inadequately understood. Here, 3 years of dissolved inorganic carbon (DIC) measurements and carbon dioxide (CO2) fluxes from Ryder Bay on the west Antarctic Peninsula (WAP) are presented. During spring and summer, primary production in the surface ocean promotes atmospheric CO2 uptake. In winter, higher DIC, caused by net heterotrophy and vertical mixing with Circumpolar Deep Water, results in outgassing of CO2 from the ocean. Ryder Bay is found to be a net sink of atmospheric CO2 of 0.90–1.39 mol Cm−2 yr−1 (average of 3 years). Seasonal sea ice cover modifies the net annual CO2 uptake, but its effect on gas exchange remains poorly constrained. A reduction in sea ice on the WAP shelf may reduce the strength of the oceanic CO2 sink in this region.


Introduction
The Southern Ocean south of 44 ∘ S is responsible for approximately 25-30% of the global ocean uptake of anthropogenic carbon [Fletcher et al., 2006;Lenton et al., 2013], but accurately quantifying this sink and understanding the processes behind it remains challenging. The net ocean-atmosphere CO 2 flux of the high-latitude, seasonally ice-covered Southern Ocean is especially difficult to quantify due to a scarcity in observational data, particularly during the ice-covered winter months [Bakker et al., 2014]. Not only does ice cover hinder data collection and therefore limit the scope of observational flux estimates, but it also adds complexity to the carbon cycle through its effect on gas exchange, vertical stability, and biological production.
Estimates of CO 2 uptake by the high-latitude Southern Ocean vary between models, inversions, and observations. Most recent estimates characterize the high-latitude Southern Ocean as a small sink or a small source of CO 2 to the atmosphere [Gruber et al., 2009;Takahashi et al., 2009;Lenton et al., 2013]; however, they are largely based on the open ocean and do not account for the variability and importance of the coastal ocean which may be a strong sink of CO 2 [Arrigo et al., 2008]. Furthermore, many ocean biogeochemical models fail to capture the magnitude of the seasonal cycle at high latitudes, the main mode of variability in this region [Lenton et al., 2013]. This inability to reconcile modeled and observed CO 2 fluxes highlights our poor understanding of the high-latitude carbon cycle and limits confidence in our ability to predict future changes to Southern Ocean CO 2 uptake.
Recent work suggests that the Southern Ocean carbon sink may be weakening, and this has been attributed to increased upwelling of dissolved inorganic carbon (DIC) caused by increased Southern Ocean winds associated with a more positive Southern Annular Mode (SAM) [Lenton and Matear, 2007;Le Quéré et al., 2007;Lovenduski et al., 2008]-a climatic change that is known to include an anthropogenic component [Thompson and Solomon, 2002]. However, due to a lack of observational data in this region, our mechanistic understanding of the variability in the ocean-atmosphere flux of CO 2 remains weak. A better understanding of the processes driving seasonal and interannual variablity in carbon flux is required to strengthen our predictions of future carbon uptake by the Southern Ocean. This is true especially for the productive coastal waters, which are the focus of this paper.

Sampling and Analyses
Discrete samples for dissolved inorganic carbon (DIC) and total alkalinity (TA) were collected from 15m depth at the Rothera Time Series site (RaTS), located in Ryder Bay, about 4km offshore on the west Antarctic Peninsula (Figure 1). Sampling was undertaken from a rigid inflatable boat or through a hole in the ice approximately weekly in summer and biweekly in winter, weather and ice permitting. If partial ice cover prevented access to the main RaTS site, then a secondary site was used ( Figure 1). Collection of inorganic carbon samples began in December 2010 and continued until February 2014. Samples were collected in 250 ml or 500 ml borosilicate glass bottles, spiked with mercuric chloride and sealed with greased stoppers. They were transported back to the University of East Anglia, UK, at the end of each field season for analysis. DIC was measured by coulometry [Johnson et al., 1985] following standard operating procedure (SOP) 2 of Dickson et al. [2007], and TA was measured by potentiometric titration [Mintrop et al., 2000] following SOP 3b of Dickson et al. [2007]. DIC and TA were measured using two VINDTAs (versatile instrument for the determination of titration alkalinity, version 3C, Marianda, Germany). Instruments were calibrated using certified reference materials from the Scripps Institution of Oceanography. Sea ice type and fraction of ice cover were visually estimated by marine assistants at Rothera. While these sea ice estimates are inherently somewhat subjective, the Rothera data have been found to agree well with wider-scale satellite estimates [Wallace, 2007]. Temperature, salinity, and pressure were measured during full depth (500m at site 1) conductivity, temperature, depth (CTD) casts [Venables et al., 2013].

Ocean-Atmosphere CO 2 Flux
The ocean-atmosphere flux of CO 2 was calculated for each day during the study period using equation (1). A negative value signifies an uptake from the atmosphere by the ocean.
where s is the solubility of CO 2 at in situ temperature and salinity, calculated using the equation of Weiss [1974], Δf CO 2 is the difference in the fugacity of CO 2 between seawater and the air, and k is the gas transfer velocity. Water f CO 2 was calculated from DIC, TA, silicate, and phosphate measurements using the CO2SYS program [Van Heuven et al., 2011]. Total alkalinity data are presented in the supporting information. Seawater temperature and salinity, seawater f CO 2 , and open water fraction were linearly interpolated into daily vectors. Atmospheric f CO 2 was calculated from the dry mole fraction of CO 2 (xCO 2 ) using the coefficients of Weiss [1974], and water vapor pressure was calculated following Weiss and Price [1980]. Atmospheric xCO 2 data are from weekly flask samples from Palmer Station, approximately 400km northeast of Rothera [Dlugokencky et al., 2014]. These discrete flask data were interpolated and smoothed using a LEGGE ET AL.
weighted linear least squares method. Daily averaged wind speed and sea level pressure were measured at Rothera. Wind speed has been corrected from the measurement height of 42 m to 10 m above sea level following Fairall et al. [2011].

Ice Cover and Gas Transfer Velocity
Gas transfer velocity (k) and its relationship with ice cover were calculated using two separate methods. First, k was calculated from wind speed using the parameterization of Wanninkhof et al. [2013]. As in previous studies, the gas transfer velocity was scaled linearly with sea ice cover by multiplying k by the fraction of open water. Even at 100% ice cover, there is likely to be some gas exchange with the atmosphere due to leads, fractures, and brine channels [Semiletov et al., 2004;Loose and Schlosser, 2011], so, consistent with other studies [Bates et al., 2006;Mucci et al., 2010;Roden et al., 2013], we have set the minimum open water fraction to 0.01. A scenario with no ice is also considered in which k is not modified by ice cover and also a scenario in which brash and grease ice do not affect k. The second method for calculating gas transfer velocity is a recently published parameter model for gas exchange in the sea ice zone [Loose et al., 2014]. This model relates k to wind speed through mean squared wave slope and also includes shear-driven and convection-driven turbulence. The k derived from both methods was corrected to the Schmidt number following Wanninkhof [1992]. Full details of gas flux calculations are given in the supporting information.

Uncertainties
Uncertainties were calculated from the measurement uncertainty of DIC, TA, silicate, phosphate, and atmospheric CO 2 ; the uncertainty of the carbonate system dissociation constants [Millero, 2007]; and the uncertainty of the parameterization of the gas transfer velocity [Wanninkhof , 2014]. The propagation of uncertainties using a Monte Carlo approach is described in Appendix A.

Seasonal and Interannual Variability
The 3 years of DIC data presented here ( Figure 2) display a clear, highly assymetric, seasonal cycle with an amplitude of ∼200 μmol kg −1 . DIC decreases sharply during December/January due to the rapid increase of biological production, and summer months are characterized by low and variable DIC values. There is a gradual increase from around April to maximum DIC concentrations of ∼2200 μmol kg −1 in September. This increase during autumn and winter is caused by net heterotrophy and mixing with relatively old, carbon-rich Circumpolar Deep Water (CDW) as the mixed layer deepens. The rapid spring DIC drawdown is preceded by a more gradual decrease corresponding to reduced sea ice cover ( Figure 2), an increase in photosynthetically active radiation, and a shoaling and warming mixed layer. The gradual decrease prior to the main spring bloom is therefore likely to be caused by the onset of algal production in the water column and ice but may also be influenced by ikaite dissolution during ice melt (see section 3.2).
Due to the pronounced seasonal cycle in DIC, the Δf CO 2 changes sign between winter and summer ( Figure 2), making the water column a potential source of CO 2 to the atmosphere during winter and a potential sink during summer. This gradient determines the direction of the CO 2 flux, but its magnitude is strongly influenced by wind speed and ice cover. The daily ocean-atmosphere CO 2 flux, as calculated with a linear scaling between k and ice cover (Figure 3), ranges from approximately −15 mol Cm −2 yr −1 in summer (into the ocean) to approximately 10 mol Cm −2 yr −1 in winter (out of the ocean). In all 3 years sampled, the summer uptake of CO 2 by the ocean is greater than the winter outgassing, making the water column in Ryder Bay a net sink of atmospheric CO 2 over an annual cycle. Summing the daily fluxes gives the cumulative ocean-atmosphere CO 2 flux (Figure 3), clearly showing a net ocean sink from all estimates. The mean net annual CO 2 uptake by the ocean from 1 January 2011 to 31 December 2013 was 0.90 mol Cm −2 yr −1 as calculated using the linear scaling of k with ice cover (solid blue line, Figure 3) with net annual uptake ranging from 0.56 mol Cm −2 yr −1 in 2013 to 1.34 mol Cm −2 yr −1 in 2011. When calculated using the sea ice gas exchange model (solid red line, Figure 3), the mean net annual ocean uptake was 1.39 mol Cm −2 yr −1 ranging from 1.00 mol Cm −2 yr −1 in 2013 to 1.94 mol Cm −2 yr −1 in 2011. Our estimates of net annual ocean carbon uptake are consistent with the estimates of Roden et al. [2013], who calculated a net ocean uptake in Prydz Bay, East Antarctica of 0.54 ± 0.11 mol Cm −2 yr −1 and Arrigo et al. [2008] who calculated a net annual uptake by the Ross Sea of 1.7-4.2 mol Cm −2 yr −1 using a biogeochemical model. There is significant interannual variability in the ocean-atmosphere flux of CO 2 , especially in the amount of winter outgassing (Figure 3). Atmospheric CO 2 increases by roughly 2.3 ppm/yr from ∼387 ppm in January 2011 to ∼394 ppm in January 2014, with a seasonal amplitude of ∼3 ppm. Despite this increase in atmospheric CO 2 during the study period, the ocean to atmosphere CO 2 flux increases through each successive winter in the three observed years, and we hypothesize that this interannual variability is caused by two main drivers. First, higher water f CO 2 in winter increases the concentration gradient of CO 2 between the ocean and the atmosphere (Figure 2), making the ocean a stronger potential source of CO 2 to the atmosphere. We suggest that the increase in f CO 2 and DIC in surface water in winter is caused by increased mixing of CDW into the surface layer, and this theory is supported by a slight increase in winter salinity from 2011 to 2013. Reduced ice cover in Ryder Bay during winter allows enhanced mixing due to the exposed water surface, and this can result in reduced stratification in the following spring [Venables and Meredith, 2014]. A decrease in ice cover could therefore affect ocean-atmosphere CO 2 fluxes by increasing the mixing of CDW into the surface and increasing surface water f CO 2 and DIC.
Second, a reduction in ice cover during winter, when Δf CO 2 is positive, allows more gas exchange across the air-water interface. The three consecutive winters shown here exhibit increasing Δf CO 2 and a longer period during which Δf CO 2 is positive, while ice cover is below 50%. These factors combined result in increased outgassing of CO 2 to the atmosphere and a reduced net CO 2 sink over the seasonal cycle. When the effect of ice cover on k is ignored, the water column is still found to be a net sink of atmospheric CO 2 (dashed lines, Figure 3). The role of winter sea ice cover in reducing CO 2 flux from the ocean to the atmosphere is consistent with previous observations [Gibson and Trull, 1999;Sweeney, 2003;Bakker et al., 2008] and supports the hypothesis that increased Antarctic sea ice cover during glacial periods could have lowered atmospheric CO 2 by reducing deep water ventilation [Stephens and Keeling, 2000].
LEGGE ET AL.

Unresolved Complexity
Quantification of the gas transfer velocity in ice-covered or partially ice-covered waters is challenging, particularly as different ice types are likely to affect gas fluxes in different ways, and this heterogeneity is not well understood. To investigate the sensitivity of our flux estimates to different ice types, ocean-atmosphere CO 2 flux was also calculated with brash ice ignored (green line, Figure 3). This scenario results in a stronger annual ocean uptake of CO 2 than when k is affected by all ice types (solid blue line) because brash ice predominates in summer, when the Δf CO 2 is negative, so its removal allows more CO 2 flux into the water. The parameter model of Loose et al. [2014] (red line, Figure 3) gives higher k values and therefore a larger net annual sink than the scenario using a linear scaling of k with ice cover. This is partly caused by the way in which these two methods calculate k from wind speed. The contribution of this factor can be seen as the difference between the red and blue dashed lines which do not include ice effects. Another reason for the discrepancy between the linear scaling and the parameter model is the inclusion in the model of shear-driven and convection-driven turbulence which can increase the gas transfer velocity by 40% [Loose et al., 2014]. It is important to note that the uncertainty caused by the handling of ice cover and ice type in gas flux estimates dwarfs that caused by uncertainty in the carbonate parameters and the gas transfer LEGGE ET AL.
velocity (shading on blue and green lines, Figure 3). More observational data, from both laboratory ice tank experiments and field studies, are needed to increase our confidence in calculating ocean-atmosphere gas flux in seasonally ice-covered waters.
As carbon samples were collected from 15 m depth, the seawater f CO 2 used for flux calculations may not accurately represent surface concentrations when the mixed layer shoals to above this depth in summer. Due to biological activity in the surface mixed layer, it is likely that our water f CO 2 is overestimated between approximately December and March, giving an underestimate of the difference between the concentration of CO 2 in the atmosphere and the sea water. The net annual ocean CO 2 uptake presented here is therefore likely to be an underestimate. Mixed layer depth data are presented in the supporting information.
Sea ice does not simply act as a cap to ocean-atmosphere gas exchange; sea ice-atmosphere CO 2 exchange has been observed in both the Arctic and Antarctic [Gosink et al., 1976;Semiletov et al., 2004;Delille, 2006;Zemmelink et al., 2006;Miller et al., 2011;Nomura et al., 2013]. The estimates of CO 2 flux presented here do not account for these ice-atmosphere interactions and so may overestimate or underestimate the net annual sink of the water column. Although sea ice has been demonstrated to act as both a source and a sink of CO 2 over the seasonal cycle [Delille et al., 2014], we have applied the maximum flux of CO 2 from Antarctic sea ice to the atmosphere of 1.9 mmol m −2 d −1 observed by Delille et al. [2014] in order to obtain a first-order estimate of the maximum possible reduction in CO 2 sink strength caused by ice cover. Multiplying the percentage cover of fast or pack ice in Ryder Bay during the study period by this value results in a weakening of the net annual sink of atmospheric CO 2 of approximately 0.2 mol m −2 yr −1 .
As well as modulating gas exchange via physical processes, sea ice also affects CO 2 flux through its effect on carbonate chemistry. The precipitation of the hydrated calcium carbonate mineral ikaite (CaCO 3 ⋅ 6H 2 O) in sea ice increases CO 2 in brine, and its dissolution on ice melt reduces CO 2 of the seawater [Rysgaard et al., 2011]. Although no carbon measurements in sea ice were made here, ikaite has been observed in Antarctic sea ice [Dieckmann et al., 2008] and has been found to significantly increase brine pCO 2 during precipitation and decrease brine pCO 2 on dissolution [Papadimitriou et al., 2004;Delille et al., 2007;Geilfus et al., 2012]. Rysgaard et al. [2012] found the dissolution of ikaite during sea ice melt to correspond to an air-sea CO 2 uptake of ∼3.9 mol Cm −2 yr −1 , given a reduction in ice thickness of 0.2 m/week. This process is of a significant magnitude compared to the ocean-atmosphere CO 2 fluxes presented here (Figure 3). It is likely that some of the increase in water pCO 2 observed during winter in Ryder Bay is caused by the release of high CO 2 brine from sea ice and, similarly, that some of the f CO 2 decrease observed in spring is caused by ikaite dissolution. However, the quantification of the impact of ice formation/melt on carbonate chemistry in Ryder Bay is complicated by the fact that an observed change in ice cover is not directly related to ice formation/melt as much of the brash and pack ice may be blown in/out of the bay rather than forming/melting in situ [Meredith et al., 2010]. Biological production by ice algae was also not quantified quantified here and it almost certainly contributes to reducing DIC and f CO 2 in the surface layer, thereby affecting the ocean-atmosphere CO 2 flux in spring.

Future Implications
The mechanistic understanding of observed variability in the ocean CO 2 uptake that we have developed has significant implications for how we can expect the carbon sink in this region to behave in the future. The west Antarctic Peninsula has been one of the most rapidly warming regions in the world over the last 50 years [Vaughan et al., 2003;Meredith and King, 2005] with an associated rapid decrease in sea ice cover [Stammerjohn et al., 2008;Li et al., 2014]. On a larger scale, and contemporaneous with this, there has been a strengthening of zonal winds south of 45 ∘ S, which is projected to increase in future [Le Quéré et al., 2007], and modeling studies suggest that this will increase the upwelling of carbon-rich deep water in the Southern Ocean, slowing the increase in the carbon sink expected with rising atmospheric CO 2 levels [Lenton and Matear, 2007;Lovenduski et al., 2008]. Decadal changes in the properties and production of dense water around Antarctica have also been observed Purkey and Johnson, 2012;Lavergne et al., 2014], emphasizing further the potential sensitivity of the Southern Ocean carbon sink to climatic change.
The observations presented here support the theory that reduced winter ice cover and increased surface DIC concentrations in future will reduce the rate at which the region takes up atmospheric CO 2 . Furthermore, a reduction in the spring phytoplankton bloom in Ryder Bay following winters with reduced sea ice LEGGE ET AL.
©2015. The Authors. cover has been found [Venables et al., 2013], and it has been suggested that a continued reduction in WAP sea ice cover is likely to lead to reduced phytoplankton blooms in the region. A reduction in winter sea ice would therefore not only increase winter outgassing but would also reduce summer uptake of CO 2 by the ocean, further reducing the net annual CO 2 sink.

Appendix A
Uncertainties were calculated using a Monte Carlo approach to ensure that all nonlinearities in the carbon system calculations were accounted for in error propagation. This is hard to achieve using formal error propagation. The propagation of uncertainties was divided into two steps. First, the uncertainty on water f CO 2 for each measurement of DIC and TA was calculated as follows: 10 5 parameter value sets were created including the parameters DIC, TA, silicate, phosphate, pK1, and pK2. For each parameter, values were determined by random sampling from a distribution based on the measured value (or calculated value in the case of carbonate system constants) as the population mean with a standard deviation based on the measurement or parameter uncertainty (see Table A1). The CO2SYS program was run on each set of parameter values to determine a probability density function (pdf ) for f CO 2 , twice the standard deviation of which was taken as a measure of that data point's uncertainty. The errors on measured values and carbonate system constants used in the uncertainty analysis (see Table A1) were assumed to be normally distributed. One standard deviation was used to define the characteristics of the distribution of each of the input parameters in the uncertainty analysis. Two standard deviations (approximately 95% confidence) is quoted when referring to calculated uncertainties.
The second step was the calculation of uncertainty on the daily calculated ocean-atmosphere flux. Following interpolation of all input parameters on a daily basis, the uncertainty on each daily calculated flux was calculated as follows: 10 5 parameter value sets were created for water f CO 2 , air f CO 2 , and k as described above.
The uncertainty on water f CO 2 was taken from the previous step, the uncertainty on air f CO 2 is described in Table A1, and the uncertainty on k was taken as 20% [Wanninkhof , 2014] and given a uniform distribution. The ocean-atmosphere flux was calculated for each set of values as above, and twice the standard deviation of the resulting pdf was taken as a measure of the uncertainty of that day's flux. The uncertainty on the cumulative flux was calculated as the square root of the sum of the squares of the uncertainties up to and including each point which assumes randomly distributed rather than systematic uncertainty. Uncertainties were only quantified for the fluxes calculated following Wanninkhof et al. [2013] with a linear scaling of ice cover and k. The uncertainty on the fluxes calculated following Loose et al. [2014] was not quantified as the uncertainties in this new model are not well understood.
LEGGE ET AL.

Erratum
In the originally published version of this article, the calculation of the fugacity of carbon dioxide in air used the incorrect units for the gas constant, R. The following have since been corrected, and this version may be considered the authoritative version of record.
In section 3.1, the end of the second paragraph, beginning with "The mean net annual...," has been corrected; and in the final paragraph of this section, the phrase "although the strength... Figure 3)" has been deleted. Figure 3 has been remade to include corrected flux and cumulative flux data.
In the supporting information, the units of the gas constant, R, have been corrected.
LEGGE ET AL.