The Effect of Antarctic Sea Ice on Southern Ocean Carbon Outgassing: Capping Versus Light Attenuation

Abstract

We examine the role of sea ice in controlling air-sea carbon fluxes around Antarctica using numerical simulations and idealized theory. Upwelling of carbon and nutrient-rich deep waters in the Southern Ocean promotes outgassing of CO₂ and fuels the biological flux of sinking organic particles. Sea ice inhibits outgassing, by presenting a physical barrier to air-sea exchange (capping), and decreases biological uptake by reducing the flux of photons to the ocean surface (light attenuation). These two compensating effects suggest that changes in sea ice may have a modest impact on the air-sea flux of CO₂ in the region. Numerical simulations support this inference, showing that the net integrated flux remains nearly constant for a large range of sea ice fractions when the ice cover is uniform and time-invariant. Consequently, the outgassing flux is only significantly capped when the ice cover is nearly complete. A simple analytical model shows that the compensation strength can be uniquely characterized by a single parameter that depends on the flow residence time scale under ice, relative to the air-sea equilibration and biological time scales. When the ice is seasonal, compensation between capping and light attenuation is weakened, but still significant. The spring months are particularly important due to the co-occurrence of an extended sea ice cover and the presence of sunlight.

Key Points

• The effect of sea ice on reducing light availability can compensate for its capping effect on the air-sea carbon flux
• This compensation mechanism is strongest during spring and for a largely extended sea ice cover
• The compensation does not occur locally but rather when integrated over the entire sea ice zone

Plain Language Summary

The ocean stores more than 50 times the amount of CO₂ in the atmosphere as dissolved inorganic carbon in the deep ocean. Around Antarctica, these carbon-rich waters come up to the surface, where CO₂ can escape into the atmosphere (outgassing). This dissolved inorganic carbon can also be utilized by living organisms to form organic particles, which ultimately sink into the deep ocean (uptake). The level of atmospheric pCO₂ is strongly dependent on the balance between the competing effects of carbon outgassing to the atmosphere and carbon uptake by biological processes. Moreover, the Antarctic coast is surrounded by sea ice which can affect physical and biological processes that are important for the global carbon cycle. On the one hand, sea ice can act as a lid that reduces exchanges between the ocean and the atmosphere (capping). On the other hand, sea ice can attenuate the amount of light available for organisms living in the ocean and thus reduce the formation of sinking organic particles (light attenuation). We find that the competition between these two effects limits the impact of changes in sea ice cover on the net exchange of carbon between the ocean and the atmosphere.

1 Introduction

The Southern Ocean plays a critical role in shaping the global carbon cycle (Gruber et al., 2009). The unique circulation in these southern high latitudes, fueled by strong westerly winds, drives the upwelling of deep waters to the surface, where they may interact directly with the atmosphere (Marshall & Speer, 2012; Talley, 2013). These deep waters are naturally enriched in dissolved inorganic carbon (DIC) and macronutrients such as nitrate (NO₃) and phosphate (PO₄), due to remineralization of sinking organic matter at depth (see Figure 1). Upwelled deep waters raise surface ocean pCO₂, stimulating outgassing to the atmosphere, while nutrients fuel export of organic carbon from the surface to the deeper ocean, lowering seawater pCO₂ (Lauderdale et al., 2017; Marinov et al., 2006). Diabatic processes associated with high-latitude water-mass transformations then resubduct this surface flow into a northward moving component forming intermediate waters and a southward flow that sinks as bottom waters (Talley, 2013). These dynamics lead to a competition between outgassing of CO₂, biological carbon uptake and export of organic particles, and resubduction of DIC. The outcome of this balance is particularly significant for the maintenance of ocean carbon stores and the regulation of global atmospheric pCO₂ (Frölicher et al., 2015; Ito & Follows, 2013; Toggweiler & Sarmiento, 1985).

Figure 1

Simplified schematic representing the two-cell residual overturning circulation in the Southern Ocean. The thick black lines show these two cells flowing from the deep ocean up to the surface, traveling within the mixed layer (in red), and then resubducting into the deep ocean. The top cell returns northward as intermediate waters, whereas the bottom cell sinks near the shelf to form bottom waters. This overturning circulation carries carbon and nutrient-rich waters up to the surface, which drives carbon outgassing in the south and carbon uptake in the north, as represented by the thin vertical arrows. Sea ice (in blue) can affect both these processes by capping the air-sea flux and reducing light availability.

A significant portion of deep water upwelling occurs underneath the seasonal ice zone (SIZ) (Ferrari et al., 2014). Sea ice can affect the physical and biological dynamics around Antarctica and potentially have global climatic consequences. However, despite the outsized importance of this region, observations in the Southern Ocean have historically been limited and summer biased due to the difficulty of access, particularly in the SIZ. Recently, the deployment of biogeochemical ARGO floats, both in open water and under sea ice, is starting to complement ship-based observations regarding physical and biogeochemical properties in the Southern Ocean. Early results from this ARGO data set tends to confirm the overall seasonal and latitudinal pattern of air-sea carbon flux from past studies (Landschützer et al., 2014; Takahashi et al., 2009) but with the notable exception of more outgassing at the northern margin of the SIZ (Gray et al., 2018). This suggests that some of the mechanisms active in this region and their interactions with sea ice are still poorly understood. Modeling studies also confirm that view, since many state-of-the-art global climate models (GCMs) cannot reasonably capture the magnitude and seasonality of air-sea carbon fluxes in the Southern Ocean (Russell et al., 2018).

A large number of studies have discussed the potential impacts of Antarctic sea ice on the carbon cycle. A change in sea ice production affects deep water formation around the continent, which alters the overturning circulation (Ferrari et al., 2014; Jansen, 2017; Sun & Matsumoto, 2010; Watson et al., 2015) and thus modulates the delivery of carbon and nutrients to the surface. Sea ice also acts as a lid that inhibits air-sea carbon exchanges and helps store carbon into the deeper ocean over centennial time scales (Ferreira et al., 2018; Stephens & Keeling, 2000). Both these mechanisms have been implicated in the decrease of atmospheric pCO₂ during the glacial periods of the Pleistocene (Adkins, 2013), which had an expanded sea ice cover in the Southern Ocean (Gersonde et al., 2005). Another effect of sea ice, whose climatic impact has been less widely recognized, is its ability to attenuate the amount of light permeating into the euphotic zone. While there exist organisms that can thrive under the submerged surface of ice floes (Lazzara et al., 2007; Monti et al., 2017; Meiners et al., 2017), light limitation has the potential to cause a reduction in the net biological carbon export to the deeper ocean. Kurahashi-Nakamura et al. (2007) (hereafter KN07) conduct numerical experiments in a GCM where sea ice affects flux capping and light attenuation, either together or in isolation, without changing the ocean's circulation or its gas solubility. Their results show light attenuation dominating over flux capping, leading to an increase in atmospheric pCO₂ under a more compact sea ice cover. Sun and Matsumoto (2010) (hereafter SM10) confirm the view that light attenuation can play a key role in the response but argue that its effect competes with changes in solubility and ocean circulation associated with a colder climate. On the other hand, Ferreira et al. (2018) find that capping can indeed store a significant amount of carbon in the deep ocean, but their model does not include light limitation by sea ice. This motivates us to examine more closely the controls on air-sea flux of CO₂ in the Southern Ocean, in particular, the sensitivity to sea ice cover and the competition between physical and biological influences. We use an idealized channel model of the Southern Ocean and simple theory to investigate how air-sea carbon fluxes around Antarctica may be affected by the competing effects of sea ice on capping and light attenuation.

The paper is structured as follows: In section 2, we describe the frameworks of the channel and analytical models. In section 3, we investigate the drivers that control Southern Ocean carbon fluxes and emphasize the importance of biological uptake versus carbon upwelling in the SIZ. In section 4, we study the competition between the capping and light attenuation effects of sea ice using experiments analogous to the ones performed by KN07. We explore the sensitivity of that competition to sea ice extent, seasonality and patchiness, as well as the background climate state. We also investigate how changes in surface fluxes driven by sea ice can alter the distribution of DIC in the Southern Ocean. In section 5, we discuss and conclude.

2 Modeling Details

2.1 Channel Model

Numerical experiments are conducted with a 2-D channel model of the Southern Ocean with a coupled sea ice/ocean configuration based on the MITgcm dynamical core applied to a beta plane (Marshall, Adcroft, et al., 1997; Marshall, Hill, et al., 1997). The domain spans 3,264 km in the meridional direction with 10.2 km horizontal resolution. In the zonal direction, there is only one grid cell with reentrant boundary conditions (see supporting information [SI] Figure S1). In the vertical, there are 50 levels ranging from the linear free surface to the flat bottom at 4,000 m depth. The top 50 m is resolved at every 10 m, and the intervals between levels increase to 100 m toward the bottom. At the southern boundary, there is a 300 m deep and 80 km long shelf, followed by a continental slope that drops to the bottom within 300 km. The model configuration is analogous to the one used by Doddridge et al. (2019), except that instead of explicitly resolving eddies in 3-D, our 2-D version of the model parameterizes their effects as an advective process (Gent & McWilliams, 1989) and isopycnal diffusion Redi (1982), both with a transfer coefficient of 800 m² s⁻¹. The background vertical diffusion is uniform and set to 10⁻⁵ m² s⁻¹. Ocean convection is represented by enhanced vertical mixing of temperature and salinity via convective adjustment (Klinger et al., 1996). The equilibrated mean states of the 2-D and 3-D models are similar, so we do not expect the lack of explicit eddies to fundamentally change the conclusions of this study. Sea ice thermodynamics are based on the three-layer formulation of Winton (2000), and sea ice dynamics are based on the elastic-viscous-plastic method developed by Hunke and Dukowicz (1997) and described in Losch et al. (2010).

The model also simulates ocean biogeochemistry with an active iron cycle (Dutkiewicz et al., 2006; Parekh et al., 2006). Biological transformations between carbon and nutrients are related using the fixed stoichiometric ratios of (Anderson & Sarmiento, 1994) with a prescribed inorganic to organic rain ratio of 7% (Yamanaka & Tajika, 1997). Biological productivity, B, is limited by light and either macronutrient or iron concentration, whichever is scarcest, as follows:

(1)

where A_b is the maximum rate of biological activity, k_PO4, k_I, and k_Fe are half-saturation constants, and I is the photosynthetically active radiation (PAR).

Air-sea fluxes of CO₂ are calculated prognostically and are dependent on the square of local wind speed (Wanninkhof, 1992). There is no riverine carbon input, sediment carbon burial, nor external CO₂ emissions. The bioavailable iron dust forcing is obtained from Luo et al. (2008) and shown in Figure S3c. It is applied directly at the ocean surface and does not include the effect of sea ice. Sedimentary sources of iron are also included, as per the formulation of Elrod et al. (2004). PAR is taken as 40% of the net incoming solar radiation. Following KN07, we assume that both PAR and the air-sea carbon flux decrease linearly with increasing ice fraction and are equal to 0 when the ice fraction is 1. For simplicity, we also keep the partial pressure of carbon dioxide in the atmosphere (pCO ) constant. In the main body of the paper, it is fixed to a preindustrial value of 278 ppm but the SI presents sensitivity experiments with a postindustrial value of 370 ppm.

The surface is forced with a seasonal climatology based on the monthly mean atmospheric data from the Corrected Normal Year Forcing Version 2.0 (CORE v2) product (Large & Yeager, 2009) taken along 30°E. This longitude corresponds to a repeat section (IO6S) between South Africa and Antarctica sampled for both hydrographic and biogeochemical variables, which ensures high-quality observational data are weighted in the optimally interpolated climatologies used as input to the model. Because the channel is 2-D, there are no zonal gradients in forcing. The shortwave and longwave radiative fluxes are applied directly as net heat flux to the surface. Standard bulk formulate are used for the sensible and latent heat fluxes, using the climatological surface temperature, specific humidity, and wind speed (Large & Pond, 1982). Zonal and meridional momentum fluxes are computed based on the prescribed wind velocity at 10 m height. Evaporation and precipitation are also specified from the CORE v2 climatology. The northern boundary has a 100 km sponge layer where physical and biogeochemical properties are gradually relaxed to a fixed vertical profile with 10 day time scale. For these profiles, we use climatology along 30°E for temperature, salinity, DIC, alkalinity, dissolved organic phosphate, O₂, and PO₄ obtained from the World Ocean Atlas version 2 (Locarnini et al., 2013; Zweng et al., 2013) and GLODAP version 2 (Lauvset et al., 2016; Olsen et al., 2016, 2019). The small domain size and presence of the sponge layer at the northern boundary enables the model to reach a well-equilibrated solution in 100 years of integration time. The advantage of such an idealized model over a more comprehensive 3-D model is the ability to conduct a large number of sensitivity simulations while keeping the numerical solutions tractable. The principal drawbacks are that it does not explicitly resolve eddies and that the sponge boundary condition in the north does not allow the quantification of changes in global carbon stores associated with sea ice expansion and retreat.

2.2 Analytical Model

To interpret results from the 2-D channel model, we consider a 1-D analytical model representing the surface flow of carbon under a uniform ice cover. The flow has a carbon concentration C and moves at uniform speed V in a domain of length L and depth H. As carbon travels through the domain, it may outgas to the atmosphere through a term F_e = k_e(1 − α_e)C, where k_e is the piston velocity and α_e is the ice concentration affecting outgassing. The flow may also lose carbon through biological uptake represented by the term F_b = k_b(1 − α_b)C, where k_b is a speed parameter controlling the strength of biological activity, and α_b is an ice fraction affecting light attenuation. The governing equation is as follows:

(2)

In this model, C is analogous to the residual carbon concentration C_dis described in section 3.3. The inlet is fixed to a value C_i, which represents the concentration of regenerated carbon upwelled into the ice zone. To avoid solving a separate equation for nutrients, we assume that their concentration is proportional to C by a constant factor absorbed within k_b. We believe that this simplification is reasonable in upwelling zones, where upwelled and regenerated nutrient and carbon concentrations are linked by biological processes. Solving Equation 2 for the outgassing flux F_e gives

(3)

where

(4)

In section 4.4, we discuss how the nondimensional parameter λ relates to the time scales relevant for understanding the effect of sea ice on the carbon flux.

3 Southern Ocean Carbon Fluxes in a Modern Climate

3.1 Overturning Circulation and Tracer Distribution

The equilibrated channel model state with pCO  278 ppm is shown in Figure 2. The stream function highlights the presence of an upper and a lower overturning cell that bring deep waters in contact with the atmosphere. Upwelling occurs under the SIZ close to the summer sea ice edge, such that at the surface, the meridional velocity is southward south of 65°S and northward north of 65°S. The flow is mostly along isopycnals in the upper cell, but not in the lower cell, due to water-mass transformations. Carbon and nutrient-rich waters are transported up to the surface near the intersection of the two overturning cells and drive outgassing to the atmosphere. The temperature-driven saturated carbon pool is dominant, which causes the warmer part of the domain (north and near the surface) to be less concentrated in DIC than the rest of the model. Deep waters have the highest DIC concentration due to accumulation by the biological carbon pump. Figure S2 shows good agreement between the channel model and observations regarding the latitude-depth distribution of temperature, salt, DIC, and PO₄. Near the surface under the ice zone, the model shows higher salt, DIC, and PO₄ concentrations than observations. These differences may be caused by a number of factors (e.g., biases in surface mixing, sea ice dynamics, and paucity of observations), but we do not believe they change the conclusions of this study.

Figure 2

Equilibrium solution in the 2-D channel model. (top) The residual stream function is shown in the filled contours, where positive is clockwise and negative is anticlockwise. The black contours are isopycnals referenced at the surface. The red arrows indicate the direction of the surface flow, which reverses around 65°S; (bottom) the filled contours show the DIC concentration in the model, and the orange contours indicate potential temperature. The horizontal bars show the summer (dark gray) and winter (light gray) sea ice edges.

The seasonal distribution of surface PO₄ (Figure S3) shows that the model broadly emulates the latitudinal structure of observations, with lower concentrations in the north. Between 50°S and 70°S, observations from the World Ocean Atlas (WOA) suggest somewhat higher nutrient concentrations between June and October, perhaps due to reduced biological activity in these cold and dark months. The simulation also shows this kind of pattern but overestimates nutrient concentrations between 50°S and 70°S by about 35%. This could be a sign of overly weak biological uptake or overly strong supply of nutrients in this region. However, observations from the WOA are quite sparse in the SIZ, and a detailed comparison is difficult. Figure S3c shows the bioavailable iron supply from atmospheric dust deposition applied at the surface of the model. It reveals a significant seasonal cycle, with large dust deposition in summer and fall but weaker in winter and spring, particularly in the southernmost regions. This seasonality is likely a reflection of atmospheric circulation patterns, since the forcing does not include the effect of sea ice.

3.2 Air-Sea Carbon Fluxes: Observed Versus Simulated

Figure 3 shows zonal mean air-sea carbon fluxes and sea ice fractions from observations and the channel model. The modeled sea ice extent shows good agreement over all seasons with observations from Cavalieri et al. (1996). The sea ice fraction is well captured in winter and spring but overestimated near the shelf in summer and fall. Outside the ice zone (north of 55°S), flux observations from Landschützer et al. (2017) show a clear uptake signal over all seasons. Within the ice zone, the observed fluxes are generally weaker but suggest more outgassing in winter and spring and more uptake in summer. The channel model fluxes are shown for two control simulations with different atmospheric carbon concentrations: pCO 278 and 370 ppm. Both these numerical experiments show annual mean carbon uptake in the open ocean south of 55°S and stronger outgassing in winter and spring in the ice zone, as suggested by the observations. The simulation with pCO 370 ppm is generally driven toward more ingassing, which gives a better fit to observations in summer but not as good in fall.

Figure 3

Seasonal climatology of air-sea carbon fluxes and sea ice fractions from zonal mean observations and the channel model simulation. The observed fluxes (dashed lines) are taken from Landschützer et al. (2017) in the period 1998 to 2011, and error bars indicate one standard deviation temporal variability. The observed sea ice fractions are obtained from Cavalieri et al. (1996) and averaged between 1978 and 2018. Channel model fluxes are taken from simulations with pCO  278 ppm (solid lines) and 370 ppm (dotted lines). The individual panels show (a) annual mean, (b) summer (December–February), (c) fall (March–May), (d) winter (June–August), and (e) spring (September–November) conditions. Outgassing is defined as positive.

In both simulations, the peak ice zone outgassing is significantly stronger than in the observations for winter, spring, and the annual mean. The temporal variability represented by the error bars does not entirely account for this discrepancy. Some differences can be expected due to the idealized geometry of the model and the simplified biological and physical schemes. In the Southern Ocean, iron and light colimit biological productivity, such that modeled carbon fluxes could be sensitive to their representation. Furthermore, observations in the Antarctic sea ice zone have been historically limited and summer biased due to the difficulty of access, particularly in winter and spring. Therefore, the error bars in Figure 3 may not capture the full range of uncertainty in the observed fluxes. The recent deployment of bio-ARGO floats in the Southern Ocean tend to confirm the latitudinal and seasonal structure of the fluxes reported by Landschützer et al. (2017) but suggest larger outgassing in the open water region close to Antarctica, possibly due to a better representation of wintertime fluxes (Gray et al., 2018). The availability of more observational data in the upcoming years will help narrow the range of uncertainty in these quantities.

As will be discussed in section 3.3, the air-sea carbon flux in the Southern Ocean is a small resultant of several large canceling components. Therefore, capturing its spatiotemporal distribution is a challenging task, even in state-of-the-art GCMs (Russell et al., 2018; Verdy & Mazloff, 2017). To mitigate against some of this discrepancy, we perform sensitivity experiments using two different base states (pCO  278 and 370 ppm). The simulations with pCO  370 ppm have less outgassing in the ice zone and are perhaps more representative of the modern postindustrial climate sampled by the observations. On the other hand, simulations with pCO  278 have stronger outgassing in the ice zone and could be more characteristic of a preindustrial climate. In the main body of the paper, we focus on simulations with pCO  278 ppm but include results for pCO  370 ppm in the SI and discuss the differences throughout the text. In section 4.3, we discuss how the effect of sea ice on the carbon cycle may change with different base states.

3.3 Flux Decomposition

Net carbon fluxes (Figure 3) result from the competition between various physical and biogeochemical drivers. Understanding the contribution of these individual mechanisms is a useful first step in investigating the role of sea ice on the net flux. This is challenging to do over the seasonal cycle, but it is possible to achieve a meaningful flux decomposition for the steady annual mean state of the model using the method outlined in Lauderdale et al. (2016) (hereafter LAU16). Figure 4 shows the annual mean carbon flux budget in the channel model in terms of the following drivers: heat (F_heat), freshwater (F_fresh), biology (F_bio), and disequilibrium (F_dis). These individual components and the total reconstructed flux ( ) can be estimated using the following set of equations, which use the quantities described in Table 1:

(5)

(6)

(7)

(8)

(9)

Figure 4

Decomposition of the annual mean air-sea carbon flux in the channel model using the method of LAU16 (positive is outgassing). The thin black line is the model-diagnosed flux, and the thick black line is the total reconstructed flux . The red line shows the heat-driven carbon saturation flux F_heat. The orange line is the fresh water-driven carbon flux F_fresh, which includes the effects of direct dilution of DIC, salt, and alkalinity. The dotted blue line is the biological flux F_bio, which includes the soft tissue and carbonate pumps. The blue dashed line is the disequilibrium flux component F_dis, and the solid blue line is the sum F_bio + F_dis. All the colored solid lines add up to the reconstructed flux (thick black line). The horizontal bars show the summer (dark gray) and winter (light gray) sea ice edges. The background state has pCO  278 ppm.

Table 1. Quantities Relevant to the Flux Decomposition

Quantity	Definition	Value	Units
θ	Temperature	—	°C
S	Salinity	—	psu
AT	Alkalinity concentration	—	g eq m^−3
DIC	Dissolved organic carbon concentration	—	g C m^−3
Csat	Saturated carbon concentration	—	g C m^−3
Cdis = DIC − Csat	Disequilibrium carbon concentration	—	g C m^−3
P	Phosphate concentration	—	g P m^−3
Qθ	Net heat flux at the surface	—	W m^−2
Qfw	Net fresh water flux at the surface	—	kg m^−2 s^−1
h	Mixed layer depth	—	m
	3-D velocity field	—	m s^−1
κ	Eddy diffusivity tensor	—	m^2 s^−2
	Advection/diffusion operator for a scalar quantity X	—	s^−1
ρ	Mean ocean density	1,030	kg m^−3
ρfw	Fresh water density	1,000	kg m^−3
cp	Ocean specific heat capacity	3,190	J kg^−1 K^−1
RC:P	Carbon to phosphorus ratio	117	—
RCa	Carbonate rain ratio	0.07	—
γθ	Carbon solubility coefficient for θ	−105	g C m^−3 K^−1
γS	Carbon solubility coefficient for S	−53.4	g C m^−3 psu^−1
γA	Carbon solubility coefficient for AT	10.4	g C (g eq)^−1

The total reconstructed flux (thick black line) matches closely with the model-diagnosed flux (thin black line), indicating that the decomposition works reasonably well. F_heat (red line) shows that the atmosphere cools the surface ocean south of 58°S, driving carbon uptake, whereas it warms the ocean north of 58°S, which drives outgassing. F_fresh (orange line) encapsulates the effects of fresh water on the dilution of DIC, salinity, and alkalinity. As noted by LAU16, F_fresh does not contribute much to the net CO₂ flux, since the salt-driven component is small, and the alkalinity-based change in solubility compensates for the direct dilution of DIC. F_bio (blue chained line) describes the production or remineralization of organic soft tissue and the production of calcium carbonate within the surface layer of the ocean. F_bio drives a net uptake of carbon everywhere and is dominated by the soft tissue term because the carbonate rain ratio is small. Biological activity is relatively constant north of 65°S, and tails off southward due to reduced light availability. F_dis (blue dashed line) represents the advection/diffusion of the disequilibrium carbon reservoir C_dis, defined in the surface layer as C_dis = DIC − C_sat (Ito & Follows, 2013). F_dis may be driven by (i) the upwelling of subsurface waters rich in regenerated carbon of biological origin and (ii) the incomplete realization of the C_sat potential due to the limited amount of time available for surface equilibration. In this model, F_dis drives outgassing everywhere and has a large peak around 65°S, which is likely caused by the upwelling of regenerated carbon from the deep ocean up to the surface. As the flow travels away from the upwelling zone (both northward and southward from 65°S), the surface carbon concentration equilibrates with the atmosphere, and F_dis decreases.

In summary, the annual mean pattern of uptake/outgassing is largely controlled by the sum of F_bio and F_dis (shown in the solid blue line), with some mediation in its magnitude by F_heat. The outgassing of upwelled regenerated carbon dominates south of 58°S, whereas nutrient-driven uptake dominates north of 58°S. F_heat opposes F_bio + F_dis everywhere, but not strongly enough to change the sign of the net flux. The fact that F_bio and F_dis dominate the net flux is consistent with the latitudinal and seasonal structure seen in the observations, although the details of the flux decomposition may vary regionally (Bates et al., 1998; Inoue & Sugimura, 1988; Takahashi et al., 1997, LAU16). Figure S4 presents the flux decomposition applied to the simulation that has pCO  370 ppm. F_heat, F_fresh, and F_bio are unchanged from the results shown in Figure 4, but F_dis is reduced over much of the domain and becomes negative north of 46°S. The increase in pCO relative to the carbon concentration of upwelled waters decreases the disequilibrium between them and the amount of outgassing. This situation is analogous to the transient postindustrial state of the climate, where deep waters have not yet fully equilibrated with increased pCO . Consequently, in this model configuration, a large fraction of the ice zone is dominated by F_bio, rather than F_dis.

4 The Effect of Sea Ice on Southern Ocean Carbon Outgassing

4.1 Compensation Between Capping and Light Attenuation

The importance of biological uptake in the Southern Ocean, as highlighted by the flux decomposition in section 3.3, suggests that reduced light availability due to sea ice could have a notable effect on the surface carbon flux in the SIZ. Figure 5 shows air-sea carbon fluxes in simulations where the seasonal ice cover affects the following:

1. capping only: α_b = 0
2. light availability only: α_e = 0
3. capping and light attenuation together: α_b = α_e = α

Figure 5

Air-sea carbon flux in the channel model for cases where the seasonal ice cover affects capping only (orange), light attenuation only (green), or both together (solid black). The “no-ice” case is also shown in the dotted black line. The horizontal bars indicate the model's sea ice fraction with latitude over the seasons, as well as the PAR undisturbed by ice.

For simplicity, the physical dynamics are kept fixed to the modern climate described in section 3, and the atmospheric pCO₂ is kept constant to 278 ppm. When ice affects capping only, the outgassing flux in the SIZ is reduced significantly compared to the case with no ice. On the other hand, when ice affects light availability only, the outgassing flux is higher than the case with no ice. When both effects are active (as in Figure 3), the flux magnitude is somewhere in between Cases (i) and (ii).

To investigate this behavior, we perform idealized experiments in which we artificially impose a uniform and time-invarying ice fraction over the net outgassing zone of the model (between 70°S and 55°S). As before, we conduct three sets of simulations with fixed physics and atmospheric pCO₂, where sea ice affects (i) capping only, (ii) light attenuation only, and (iii) both together. We consider the surface carbon flux integrated over the imposed ice zone (Figure 6a). When only capping is active (orange dots), the ice simply acts as a lid and prevents air-sea carbon exchange. The integrated flux decreases with increasing ice fraction and reaches 0 when the ice fraction is 100%. When only light attenuation is affected by ice (green dots), biological productivity decreases, less organic carbon is exported into the subsurface ocean, and therefore more carbon can outgas from the mixed layer. Consequently, the integrated flux increases uniformly with ice fraction. When ice affects both capping and light availability (black dots), the two effects tend to compensate. The integrated flux remains nearly constant for a large range of ice fractions, before sharply dropping to 0 when the ice fraction reaches 100%, in which case no air-sea exchange is allowed by definition. We subsequently refer to this property as the “compensation” mechanism. A similar compensation behavior occurs when integrating over the whole domain (Figure 6b), but in this case the flux is not constrained to reach 0 when α_e= 1. The integrated flux for α= 0 is close to 0, which simply indicates that uptake almost balances outgassing in this scenario when integrating over the whole domain.

Figure 6

Air-sea carbon flux integrated over (a) the imposed ice zone and (b) the whole domain for three sets of experiments where ice affects capping only (orange), light attenuation only (green), and both together (black). The imposed sea ice is uniform and constant throughout the year between 70°S and 55°S. The dots show channel model results, and the solid lines are corresponding fits from the analytical model. The background state has pCO  278 ppm.

The solid lines in Figure 6a are obtained by fitting the analytical model to the channel model results. We impose: L = 1,650 km, V = 0.02 m/s, and H = 100 m, which represent the length of the imposed sea ice zone, a characteristic meridional current speed and the diagnosed annual mean mixed layer depth, respectively. We then fit the analytical model to the scatter points for Cases (i) and (ii) using a gradient descent method. This fit gives values for C_i, k_e, and k_b, from which we infer a characteristic exchange time scale τ_e = H/k_e = 0.9 years, a biological time scale τ_b = H/k_b = 0.5 years, and a residence time scale under ice τ_res = L/V = 2.6 years. Using these values, we plot the analytical curve corresponding to case (iii). The analytical model fits the channel results well in all three cases, which suggests that it can provide useful insights into the compensation mechanism.

4.2 Latitudinal Dependence

The latitudinal structure of the air-sea carbon fluxes give useful insights into the compensation mechanism (see Figure 7). In the channel model, upwelling occurs around 65°S, and the surface flow splits into a part traveling northward and the other moving southward, both underneath the imposed ice zone that extends between 70°S and 55°S. By analogy, the inlet of the analytical model can be regarded as the upwelling point, and the flow moving from left to right.

Figure 7

Meridional structure of the air-sea carbon flux in the channel model (left) and the analytical model (right) for a range of sea ice fractions. Each row shows experiments in which sea ice affects capping only (top), light attenuation only (middle), and both together (bottom). Outgassing is defined as positive. The extent of the uniform and time-invariant sea ice is highlighted in blue. The black arrows show the flow direction in the imposed ice zone, and the black dots indicate the location of either the upwelling point (in the channel model) or the inlet (in the analytical model). The background state has pCO  278 ppm.

When sea ice affects capping only (Figure 7, top row), the outgassing flux is weaker for simulations with a larger ice fraction. As in the analytical model, this difference is large at the upwelling point, but the curves subsequently converge, as the air-sea disequilibrium reduces along the flow direction.

When sea ice affects light attenuation only (middle row), the outgassing flux in the imposed ice zone is stronger for simulations with a larger sea ice fraction. At the upwelling point, all curves have the same flux, but they subsequently diverge, as the difference in biological uptake accumulates along the flow direction. In the analytical model, the curves also diverge from the inlet, but then asymptote to 0 as eventually all the input carbon is removed from the layer by either uptake or outgassing. In the channel, this asymptotic behavior does not occur, likely because the delivery of carbon and nutrients is more continuous along y. Greater light availability in the north may also enhance the light attenuation effect with increasing latitude. Moreover, channel model simulations with large sea ice fractions and enhanced outgassing in the ice zone show correspondingly larger uptake in the open ocean (55°S to 40°S). This enhanced uptake is evidence of a redistribution of biological activity into the open ocean, as light attenuation prevents it in the ice zone. Nevertheless, when integrated over the whole domain, the net outgassing flux increases with sea ice fraction, which shows that the redistribution of biological activity is only partial (green dots in Figure 6b).

When sea ice affects capping and light attenuation together (bottom row), the flux decreases with ice fraction in the southern part of the ice zone (67°S to 62°S) but increases with ice fraction in its northern extent (62°S to 55°S). This suggests that the compensation does not hold at any given location but only when integrated over a capping-dominated region in the south and a region dominated by light attenuation in the northern part of the ice zone. This property is also captured by the analytical model. In the open ocean region of the channel, there is again evidence of a redistribution of biological uptake discussed in the previous paragraph, which tends to reduce the overall effect of light attenuation. However, as shown in Figure 6b, integrating the fluxes over the whole domain does not significantly alter the compensation mechanism. Therefore, in the rest of this study, we focus on the ice zone.

4.3 Sensitivity to the Background State

The sensitivity experiments explored thus far were all conducted in a background state with pCO  278 ppm. This state has strong outgassing in the ice zone, driven by the disequilibrium between carbon-rich circumpolar waters upwelling in the ice zone and the atmosphere. This disequilibrium depends on processes occurring upstream of the Southern Ocean and is poorly constrained for the preindustrial climate. In this section, we thus explore the compensation mechanism in an alternative background state dominated by ingassing rather than outgassing under the ice. As was described in section 3.3, this state is obtained by increasing pCO from 278 to 370 ppm, which reduces the disequilibrium of upwelling waters with respect to the surface. Consequently, capping is weakened and light attenuation dominates the response. Increasing the sea ice fraction drives more outgassing (or less uptake) when both effects of ice are included (Figure 8). The latitudinal structure of the flux in these experiments shows that while light attenuation acts similarly in both background states, capping behaves differently in net ingassing regions (Figure S5). Firstly, capping decreases uptake, thus enhancing the effect of light attenuation. Secondly, the ingassing flux is not as strongly affected by capping than in outgassing regions, consistent with it being dominated by F_bio rather than F_dis. Both these asymmetries cause light attenuation to dominate the response to sea ice in net ingassing regions.

Figure 8

As in Figure 6 but in the background state where pCO  370 ppm.

4.4 Sensitivity to the Sea Ice Extent

We study the effect of sea ice extent by halving the length of the imposed ice zone, compared to the simulations presented in Figure 6. For ease of comparison, fluxes are normalized with respect to their corresponding value at α = 0 (Figure 9). In both the channel and analytical models, the integrated flux falls more smoothly to 0 in the case where the sea ice extent is halved. This suggests that the effect of light attenuation strengthens for a wider sea ice extent, consistent with the results of KN07. This result is linked to the fact that capping dominates the response in the southern extent of the ice zone, whereas light attenuation dominates in its northern part (Figure 7). Shortening the ice zone proportionally reduces light attenuation, causing capping to become more dominant. We use the analytical model to interpret this result in terms of the exchange (τ_e), biological (τ_b), and residence time scales under ice (τ_res) introduced in section 4.1. We consider the integrated flux F_i given by

(10)

where

(11)

Figure 9

Air-sea carbon fluxes integrated over the imposed ice zone and normalized with respect to the integrated flux when no ice is present. Black dots show the channel model results in the reference case where sea ice extends between 70°S and 55°S. Blue dots are channel model results obtained by halving the length of the imposed ice zone (70°S and 63°S) and integrating over that reduced domain. Solid lines are corresponding results from the analytical model using the full ice zone length L (in black) and half its length L/2 (in blue). In these simulations, sea ice affects both capping and light attenuation together (α_e = α_b = α). The background state has pCO  278 ppm.

The normalized integrated flux F_n is then given by

(12)

Equation 12 shows that the shape of F_n(α) uniquely depends on the nondimensional parameter λ₀, which is equal to the ratio between τ_res and the sum of τ_e and τ_b. Halving the extent of the ice zone reduces the amount of time spent by the flow underneath sea ice, relative to the exchange and biological time scales. This favors capping over light attenuation, since the former effect decreases along the flow, while the latter accumulates along the flow (section 4.2). Equation 12 also more generally reveals how the impact of sea ice on the carbon cycle depends on the balance between these three time scales. This framework may help interpret the effect of sea ice under different climatic conditions, where the strength of the biological carbon pump and the physical circulation may differ.

4.5 Sensitivity to Ice Leads and Seasonality

To test the effect of seasonality on the compensation mechanism, we conduct experiments with sea ice active during winter and spring only (June to November). The physical dynamics and atmospheric pCO₂ are again kept fixed, and sea ice is imposed uniformly between 70°S and 55°S. We consider the air-sea carbon flux integrated over the full year in the imposed ice zone (Figure 10a). When ice affects capping only, the integrated flux decreases faster than when ice affects light attenuation and capping together (i.e., the compensation mechanism). The strength of this mechanism is weaker than when ice is present year-round (Figure 6a) because in winter, there is naturally less light at these high latitudes, so ice does not affect light attenuation as much as it does during the rest of the year. In spring, however, there is enough light for sea ice to reduce biological productivity and thereby activate the compensation. This is likely the reason we see evidence of the compensation working in the case of more realistic sea ice fractions shown in Figure 5. Similar results are obtained when integrating over the whole domain (Figure 10b).

Figure 10

As in Figure 6, except that the imposed sea ice is only applied during the winter and spring months (June to November). The rest of the year is completely ice-free.

The effect of sea ice on surface carbon fluxes may also be influenced by the presence of large scale leads or polynyas, which could allow the release of large amounts of carbon accumulated under ice and potentially negate its trapping effect. To investigate this possibility, we compare the net flux in a simulation with 75% uniform ice fraction with one where sea ice is imposed as a pattern alternating between 15 km of full ice coverage and 5 km of no ice, such that the mean ice cover is also 75% (Figure 11). The net integrated flux over the ice zone (between 75°S and 55°S) is 8.6 · 10⁶ g C m⁻¹ yr⁻¹ for the uniform ice cover case and 7.6 · 10⁶ g C m⁻¹ yr⁻¹ when ice is applied intermittently. Both these values are significantly lower than the reference case with no ice, where the integrated flux is 19 · 10⁶ g C m⁻¹ yr⁻¹. This suggests that sea ice can still have a net trapping effect, even in the presence of kilometer-scale leads.

Figure 11

Annualmean air-sea carbon flux in the simulations with no imposed ice (solid black line), in a simulation with 75% uniform ice cover (solid green line) and in a simulation with intermittent ice cover averaging 75% (dashed green line). The ice extends between 70°S and 55°S and affects capping only. The net integrated flux over the ice zone is 19 · 10⁶ g C m⁻¹ yr⁻¹ for the no-ice case, 8.6 · 10⁶ g C m⁻¹ yr⁻¹ for the uniform ice cover case, and 7.6 · 10⁶ g C m⁻¹ yr⁻¹ when ice is applied intermittently. Outgassing is defined as positive.

4.6 Impact on DIC

In Figure 12, we plot the DIC anomaly in the domain after 100 years for simulations with 100% sea ice fraction, compared to an experiment with no sea ice. Capping prevents carbon-rich waters upwelling in the ice zone from outgassing to the atmosphere and causes a resubduction of DIC by the lower cell of the circulation. This results in a positive carbon anomaly permeating into the deep ocean (Panel a). On the other hand, light attenuation prevents biological activity from occurring under the ice, which reduces organic carbon export and leads to a negative DIC anomaly in the deep ocean (Panel b). The weakened biological activity under ice initially leads to anomalous carbon and nutrients in the surface layer. A fraction of this anomalous carbon escapes to the atmosphere, while another is utilized by enhanced biololgical activity north of the ice zone, promoted by extra nutrients carried there by the upper cell. This explains the positive DIC anomaly in the upper circumpolar waters seen in Panel (b) and is consistent with the redistribution of biological activity discussed in section 4.2. When both capping and light attenuation are active (Panel c), the pattern in DIC distribution is a combination of the two individual responses. Antarctic Bottom Waters (AABWs) near the shelf are enriched in carbon, while deep circumpolar waters are depleted in DIC, consistent with capping dominating in the south and light attenuation dominating in the north of the ice zone. Moreover, the upper cell in Panel (c) displays a larger DIC anomaly than in Panel (b) due to the effect of capping on the anomalous surface carbon originating from reduced biology under ice.

Figure 12

DIC anomaly after 100 years in simulations where the sea ice fraction is 100% between 70°S and 55°S and applied year-round. Ice affects (a) capping only, (b) light attenuation only, and (c) both together. The anomaly is calculated relative to a simulation with no imposed sea ice. The background state has pCO  278 ppm.

The patterns of DIC changes shown in Figure 12a are analogous to those reported by Ferreira et al. (2018), whose simulated ice cover was completely insulating to carbon fluxes and did not include the effect of light attenuation. As discussed in section 4.3, the effectiveness of the capping mechanism is controlled by the disequilibrium of upwelling circumpolar waters with respect to the surface. In simulations with pCO  370 ppm instead of 278 ppm, the disequilibrium is reduced, and capping stores comparatively less carbon into the deep ocean (Figure S6a). The DIC anomaly is instead dominated by light attenuation (Figure S6c), and its pattern is consistent with the one reported by KN07. The balance between capping and light attenuation thus depends on the strength of the biological activity occurring in the ice zone and the disequilibrium of upwelling waters. Our results suggest that the DIC anomalies induced by changes in Antarctic sea ice may propagate beyond the Southern Ocean and affect the global carbon cycle. However, the open boundary condition at the northern edge of our model precludes a full carbon budget analysis. These issues should therefore be investigated more quantitatively in a global model with closed boundaries.

5 Discussion and Conclusions

This study uses numerical simulations to investigate the role of sea ice on the carbon cycle around Antarctica. The presence of sea ice can affect the carbon flux by presenting a barrier to air-sea exchange at the surface (capping) and by reducing the amount of light available for local biological productivity (light attenuation). KN07 and SM10 use global models to argue that light attenuation may dominate over capping and that an expansion of the ice cover drives an increase in pCO , rather than the decrease predicted by Stephens and Keeling (2000). Our study frames this competition in terms of air-sea fluxes and uses idealized models and theory to better understand the factors controlling the net effect of capping and light attenuation.

The numerical framework for our simulations is a 2-D channel model of the Southern Ocean, which successfully reproduces the broad latitudinal structure of the surface carbon flux seen in observations, namely, outgassing around the SIZ and uptake in the open ocean. A flux decomposition reveals that this structure is primarily controlled by the balance between biological uptake and the upwelling of carbon from the deep ocean. Biological uptake is relatively constant with latitude, except for a progressive drop south of 65°S, due to reduced light availability. The upwelling-driven outgassing peaks at the outcrop location (near the summer sea ice edge) and decreases away from those latitudes. The temperature-driven solubility effect is significant but does not affect the sign of the net flux in the annual mean. These results are generally consistent with observations and with the decomposition presented by LAU16 in a global ocean model.

Simulations with fixed physical dynamics and pCO show that in regions of net outgassing, capping reduces the carbon flux out of the ocean, whereas light attenuation increases it. When both effects are active, the net flux is modestly affected by sea ice and is only significantly capped when the ice cover is nearly complete. We attribute the strong compensatory effect of light attenuation to the dominant role played by biological uptake in this region, as evidenced by the flux decomposition. An analytical model reveals that the compensatory effect is enhanced as the residence time scale under ice increases with respect to the local biological and exchange time scales. Consequently, simulations with reduced sea ice extent are more strongly dominated by capping, rather than by light attenuation. Furthermore, a seasonal ice cover weakens the compensation effect, due to the asynchrony between months of extended sea ice and months during which light reaches high latitudes. Light attenuation may nevertheless have a strong effect in spring because these months have enough sunlight and sea ice is at its maximum extent.

An expansion of Antarctic sea ice has been implicated in the rearrangement of the carbon cycle during the glacial periods of the Pleistocene. Ferreira et al. (2018) find that flux capping contributes to significant oceanic carbon drawdown in the glacial state of their GCM. On the other hand, KN07 and SM10 report that capping only modestly accumulates carbon in the deep ocean and that light attenuation dominates instead. The experiments we conduct in different background states of the channel model may help reconcile these results. In a state dominated by outgassing under ice, capping and light attenuation contribute similarly to increasing and decreasing carbon storage in the deep ocean, respectively. However, in a state where ingassing dominates (achieved here by increasing pCO ), the influence of capping is weak, and light attenuation controls the response. The effectiveness of the capping mechanism thus depends on the disequilibrium between incoming circumpolar waters and the Southern Ocean surface, which may vary between different climatic states and models. Furthermore, while light attenuation decreases carbon accumulation in the deep ocean, our simulations show an increase in DIC concentration in the upper cell of the model, due to a partial redistribution of biological activity toward the north. This effect, along with a full budget of carbon changes, should be further investigated in a global model.

The compensation mechanism explored in this study is contingent upon sea ice reducing biological productivity in the underlying water column. While a number of studies have investigated the optical properties of sea ice and their influence on resident ice algal communities (Campbell et al., 2014; Nicolaus et al., 2010; Perovich et al., 1998), few have examined the effect of sea ice on the entire euphotic zone. Laney et al. (2017) find that the vertical distribution of PAR under Arctic sea ice frequently deviates from a canonical exponential decay, due to the spatial heterogeneity of sea ice and its scattering effects. An accurate representation of light availability under sea ice may thus require a very different parameterization than in the open ocean. Moreover, colimitation of primary productivity by iron and light is an essential aspect of the Southern Ocean carbon cycle. In our experiments, we did not couple the atmospheric dust forcing with sea ice, but this effect could be important for the biological response (Moore et al., 2000). Finally, changes in Antarctic sea ice can significantly affect the global ocean circulation, local thermal insulation and mixing under ice (McPhee, 1992; SM10, Ferrari et al., 2014; Jansen, 2017). These effects were not taken into account in this study, but they likely modulate the light attenuation and capping mechanisms in ways to be further explored.

Acknowledgments

M. G. acknowledges support from the FESD Ozone Hole grant. M. J. F. and J. M. L. acknowledge funding from NSF PIRE Project 1545859. We are grateful to Jean-Michel Campin and Hajoon Song for their help with the model setup. We also thank two anonymous reviewers and the Editor for their valuable comments on the manuscript.

Open Research

Data Availability Statement

The relevant files used to configure the MITgcm, the scripts, and data employed for producing the figures are available online (at https://doi.org/10.5281/zenodo.3786199). The flux decomposition routines are available online (at http://doi.org/10.5281/zenodo.1304267).