Southern Ocean Carbon Export Revealed by Backscatter and Oxygen Measurements From BGC-Argo Floats

Abstract

The Southern Ocean (south of 30°S) contributes significantly to global ocean carbon uptake through the solubility, physical and biological pumps. Many studies have estimated carbon export to the deep ocean, but very few have attempted a basin-scale perspective, or accounted for the sea-ice zone (SIZ). In this study, we use an extensive array of BGC-Argo floats to improve previous estimates of carbon export across basins and frontal zones, specifically including the SIZ. Using a new method involving changes in particulate organic carbon and dissolved oxygen along the mesopelagic layer, we find that the total Southern Ocean carbon export from 2014 to 2022 is 2.69 ± 1.23 PgC y⁻¹. The polar Antarctic zone contributes the most (41%) with 1.09 ± 0.46 PgC y⁻¹. Conversely, the SIZ contributes the least (8%) with 0.21 ± 0.09 PgC y⁻¹ and displays a strong shallow respiration in the upper 200 m. However, the SIZ contribution can increase up to 14% depending on the depth range investigated. We also consider vertical turbulent fluxes, which can be neglected at depth but are important near the surface. Our work provides a complementary approach to previous studies and is relevant for work that focuses on evaluating the biogeochemical impacts of changes in Antarctic sea-ice extent. Refining estimates of carbon export and understanding its drivers ultimately impacts our comprehension of climate variability at the global ocean scale.

Key Points

• We investigate carbon export in the Southern Ocean using a new method based on sinking particulate organic carbon and oxygen drawdown

• We estimate a total basin-integrated carbon export of 2.69 ± 1.23 PgC y⁻¹, where the polar Antarctic zone (PAZ) is the largest contributor (41%)

• The sea-ice zone (SIZ) contributes 8% to Southern Ocean carbon export, and most of the carbon produced in this zone is respired in the upper 200 m

Plain Language Summary

Phytoplankton take up atmospheric carbon dioxide through photosynthesis and help sequester carbon to the deep ocean. Using robotic floats equipped with biogeochemical and physical sensors, we developed a new method to quantify the carbon export from the surface to the deep ocean across the Southern Ocean. We highlight that the Southern Ocean exports 2.69 billion tonnes of carbon per year, which is about a quarter of the mean global ocean carbon export. The sea-ice zone, an area mostly overlooked due to a historical lack of observations, is responsible for 8% of the Southern Ocean carbon export, but this estimate is very sensitive to the depth range considered. Our results suggest that the Southern Ocean carbon export is spatially highly variable and seasonally sea-ice covered regions could play a significant role in the global carbon cycle.

1 Introduction

The Southern Ocean (SO) represents a significant sink for atmospheric CO₂ (Gruber et al., 2009) contributing ∼20%–40% of the total ocean CO₂ uptake (DeVries et al., 2012; Takahashi et al., 2002). The particulate organic carbon (POC) exported to the deep ocean of the SO constitutes about 30% of the ocean global carbon export (Schlitzer, 2002). Sinking POC particles are subject to modification through biogeochemical processes in the water column. Respiration by heterotrophs (zooplankton and bacteria) consumes oxygen to degrade organic matter and return particulate organic matter to inorganic nutrients (remineralization). Respiration is the dominant process consuming most of the POC in the water column, and only 1%–3% of the particles that exit the upper 100 m reach 1,000 m (Cavan et al., 2019; De La Rocha & Passow, 2007; Passow & Carlson, 2012; Steinberg et al., 2008).

The SO biological carbon pump is biogeochemically significant but exhibits strong spatial and temporal variability. Primary production in the sea-ice zone (SIZ) is strongly light-limited due to both seasonal solar patterns and the sea-ice cycle (Hague & Vichi, 2021; Taylor et al., 2013). Nutrient limitation also plays an important role. While macronutrient are abundant, phytoplankton growth is typically limited by iron (Boyd et al., 2000), which is supplied from sea ice (Lannuzel et al., 2016), coastal features (Gerringa et al., 2012; Sherrell et al., 2015; St-Laurent et al., 2019), deep mixing and entrainment of subsurface waters (Carranza & Gille, 2015; Nicholson et al., 2019; Tagliabue et al., 2014), upwelling (Moreau et al., 2023) or dust transported via the atmosphere (Jickells et al., 2005; Tang et al., 2021). Deeper mixed-layers play an increasingly important role in nutrient replenishment in the upper ocean further north, as nutrient-rich coastal sources become less effective (Boyd et al., 2007; Moore et al., 2013). The distributions of grazers, silicate, phosphate and nitrate around the Antarctic Circumpolar Current also directly impact the density and type of phytoplankton in surface waters (Assmy et al., 2013; Smetacek et al., 2012). Considering the biogeochemical and physical differences between zones in the SO (Marinov et al., 2006; Pollard et al., 2002), it is important to understand if and how the processes influencing carbon export vary spatially, as this ultimately impacts POC and associated carbon export. Studies have estimated carbon export or transfer efficiency from several techniques including observations from satellites (Arteaga et al., 2018), moorings (Lourey & Trull, 2001; Manno et al., 2022), model simulations (Karakuş et al., 2021; Lancelot et al., 2000), ship-based observations (Buesseler et al., 2005; DeJong et al., 2017; Ducklow et al., 2008; Pilskaln et al., 2004; Planchon et al., 2013; Puigcorbé et al., 2017; Ratnarajah et al., 2022; Smetacek et al., 2012), and Biogeochemical (BGC) Argo floats (Davies et al., 2019; Hennon et al., 2016; Munro et al., 2015). Some studies also combined these tools (Dall’Olmo et al., 2016; Fan et al., 2020; Nowicki et al., 2022; Schlitzer, 2002; Huang et al., 2023; Huang & Fassbender, 2024).

BGC-Argo floats are autonomous platforms instrumented with sensors to measure physical and biogeochemical variables, including temperature, salinity, dissolved nitrate, downwelling irradiance, pH, dissolved oxygen (DO), chlorophyll-a (chla) fluorescence and particle backscattering at 700 nm (b_bp) (Claustre et al., 2020; Johnson, Plant, Coletti, et al., 2017). BGC-Argo floats allow the study of biogeochemical and physical processes in the upper 2,000 m, and have ice-avoidance capabilities that allow them to gather observations beneath sea-ice (Riser et al., 2018). As part of the Argo fleet, they are programmed to sample on a 10-day cycle frequency. This temporal sampling frequency may not be sufficient to capture events on shorter time scales than 10 days for a single float, such as rapid bloom pulses (Williams et al., 2018) or mixed-layer pump events (Xing et al., 2020), that may lead to bias on long term air-sea exchange (Monteiro et al., 2015). However, they are well suited to capture seasonal carbon export (Lacour et al., 2023) and may also capture fast sinking eddy-driven carbon export (Llort et al., 2018).

Previous studies have used BGC-Argo data to report biological activity and carbon export, but these studies are usually regional in scope, over a short time scale, and north of 60°S, thus excluding the SIZ (Bishop et al., 2004; Davies et al., 2019; Le Moigne et al., 2016; Xing et al., 2020). Only a few studies have focused on the SIZ (Briggs et al., 2018; McClish & Bushinsky, 2023; Moreau et al., 2020) despite the susceptibility of this area to extreme environmental changes (Turner et al., 2022), which could alter phytoplankton productivity and subsequent carbon export (Kaufman et al., 2017).

The continuous expansion of BGC-Argo deployments globally, and growing availability of observations allow the study of oceanographic processes over large spatial scales, and with increasingly detailed temporal resolution. These advances have the potential to better understand areas such as the SIZ, where historical data coverage is poor and recent shifts in sea-ice extent are being observed (Gilbert & Holmes, 2024) with unknown impacts on carbon sequestration. In this context, our study makes use of the extensive array of BGC-Argo floats to give a new estimate of carbon export across basins and frontal zones.

2 Material and Methods

2.1 BGC-Argo Floats and Fronts

We selected floats deployed by the Southern Ocean Carbon and Climate Observations and Modeling (SOCCOM), the Southern Ocean and CLIMate (SOCLIM) and the remotely sensed biogeochemical cycles in the Ocean (remOcean) programs south of 30°S equipped with DO and b_bp (Argo, 2024). We used b_bp to derive POC, and we used DO and POC temporal changes to estimate carbon export at depth (see Section 2.3).

Float data were downloaded on 08/06/22 using the OneArgo toolbox for MATLAB (Frenzel et al., 2022) and consist of 25,953 profiles from 212 floats (191 SOCCOM, 8 SOCLIM, 13 remOcean) spanning March 2014 to June 2022, in the open ocean and under sea ice (Figures 1 and S1 in Supporting Information S1). The profiling frequency of SOCCOM floats is usually every 10 days with a vertical sampling resolution decreasing with depth, from 5 m in the first 100 m, to 50 m from 400 m to 2,000 m. SOCLIM and remOcean floats provide higher resolution data, with profiling frequency every 1–7 days at 1–10 m resolution in the first 1,000 m. We also used chla (a proxy for phytoplankton biomass) from bio-optical sensors to investigate biological spatial and temporal patterns, but chla was not used in our carbon export calculations.

Figure 1

Fronts and all BGC-Argo profile locations with co-located backscatter and dissolved oxygen data used in this study. The filled circles with black edges are profiles from the SOCLIM and remOcean fleets, while the filled circles with no black edges are profiles from the SOCCOM fleet. From north to south: subtropical zone (STZ, orange), subantarctic zone (SAZ, red), polar Antarctic zone (PAZ, green) and sea-ice zone (SIZ, blue). The black lines represent the climatology of the fronts (Bushinsky et al., 2017). STZ and SAZ are separated by the subtropical front, SAZ and PAZ are separated by the polar front, and PAZ and SIZ are separated by the seasonal ice front.

Each profile was assigned to a zone defined by fronts or sea-ice extent. Fronts for the polar Antarctic zone (PAZ), the subantarctic zone (SAZ) and subtropical zone (STZ) (Bushinsky et al., 2017) were defined from an updated Argo temperature and salinity 2004–2014 climatology (Roemmich & Gilson, 2009). The STZ and SAZ are separated by the subtropical front, the SAZ and PAZ are separated by the polar front, and the PAZ and SIZ are separated by the seasonal ice front. For the SIZ, we acquired daily EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF SSMIS) 25 km resolution sea-ice concentration data from the Copernicus Climate Data Store website (Copernicus Climate Change Service, 2020) that we used to compute the yearly sea-ice extent from 2014 to 2022. The extent of the SIZ varies widely compared to the STZ, SAZ and PAZ. It is important to accurately quantify the SIZ area, as sea ice influences seasonal stratification and iron fertilization, and the sea-ice extent is prone to rapid and dramatic changes now and in the future (Eayrs et al., 2021; Hobbs et al., 2024). We therefore used the maximum sea-ice extent to define the SIZ front in any given year. This delineation allows us to select floats under the influence of ocean - sea-ice interactions. Other major Southern Ocean fronts that we use to delineate biogeochemical zones (e.g., STZ, SAZ) are strongly linked to local topography (Moore et al., 1999), and less prone to large spatio-temporal variations in their positions. Although some degree of variability in front's position is expected (Freeman et al., 2016, 2018), fixed front positions have been commonly used in recent studies investigating large scale biogeochemistry in the SO (Arteaga et al., 2019, 2020; Huang et al., 2023; Huang & Fassbender, 2024; Johnson, Plant, Dunne, et al., 2017; Llort et al., 2018; Su et al., 2021, 2022).

2.2 BGC-Argo Data Quality Control

We used “adjusted” variables following the initial delayed mode quality control procedures performed on BGC-Argo floats (Bittig et al., 2019; Maurer et al., 2021; Schmechtig et al., 2018). For all variables, we removed all bad or questionable data (QC flag 0, 3, 4, 6, 7, 9). For b_bp, we remove outliers as in Bisson et al. (2019, 2021). All profiles were then interpolated on a common depth grid at 1 m interval. Both b_bp and DO profiles were smoothed using a 7-point moving median for SOCCOM floats (Arteaga et al., 2019) and 9-points moving median for SOCLIM and remOcean (to adjust for their higher resolution compared to SOCCOM floats). This smoothing helps to reduce the backscattering contribution of bubbles and white caps (Stramski et al., 2004), which tend to increase b_bp and potentially lead to an overestimation of variables derived from it. Finally, profiles lacking temperature, salinity (and therefore mixed-layer depth; MLD), time, or location were not considered.

In this study, we consider the floats to be quasi-Lagrangian. They can in principle cross frontal boundaries when they are drifting at depth, potentially introducing biases in the calculation for each zone defined by a fixed area. Water masses may also differ between profiles (10-day sample period on average) as floats are drifting. To improve biases due to the sampling nature of profiling floats, we checked that pairs of consecutive profiles were located in the same area and month. If not, the said pair of profiles was discarded from the analysis. We then investigated the physical properties of the water column between every pair of valid profiles. Following Johnson, Plant, Dunne, et al. (2017), we removed pairs of profiles that observed either: (a) a change of salinity by more than 0.05 psu at 500 m, (b) a change of latitude by more than 5°, or (c) a change of longitude by more than 8.8°. However, we strengthened these criteria by also discarding pairs of profiles showing changes in salinity of more than 0.05 psu at 300 m. By doing so, we ensure that consecutive profiles used to derive carbon export come from the same water mass, strongly decreasing biases that could be due to floats crossing lateral gradients. After carefully investigating our database, 4,815 pairs of profiles breaching at least one of these criteria were removed from the analysis. We also discarded pairs of profiles if either POC or DO were missing. This brought our final pool of profiles to 13,549.

2.3 Selecting Export and Horizon Depths

Most primary productivity occurs in the upper layer of the ocean. The vertical extent of the productive layer is debatable (Buessler et al., 2020; Marra et al., 2023) and the choice of export depth (E_d) can largely influence estimates of carbon export (Palevsky & Doney, 2018, 2021). Previous studies reporting carbon export estimates have either used a fixed depth or a changing productivity layer depth (Z_p), which is defined as the deepest of either the MLD or the euphotic depth (Z_eu), as their export depth (Dall’Olmo & Mork, 2014; Hennon et al., 2016; Moreau et al., 2020; Su et al., 2022). We defined the MLD as the depth where the density exceeds the density at 10 m by 0.03 kg m⁻³ (de Boyer Montégut et al., 2004) and estimated the Z_eu from the BGC-Argo chla fluorescence data following Morel and Maritorena (2001) for each profile.

During the productive season, MLD and Z_eu can be relatively shallow. Oxygen from air-sea exchange, and POC and oxygen produced from the phytoplankton bloom may impinge below E_d, if it is defined as E_d = max(MLD, Z_eu). Therefore, we defined the export depth as E_d = max(MLD, Z_eu, 200), where 200 m is roughly the deepest climatological winter MLD observed, following Moreau et al. (2020) (Figure S2 in Supporting Information S1). 200 m is also the depth where we observe a strong attenuation in chla and POC (Figure S3 in Supporting Information S1) likely implying that the signal we observe below this depth is from export, and not production. We chose 1,000 m as the lower limit (horizon depth, H_d) to encompass potential deep remineralization, following Su et al. (2022).

2.4 Carbon Export and Respiration Calculation

First, POC is estimated from particle backscattering from an empirical relationship following Haëntjens et al. (2017):

$$$\text{POC}=3.12\,\cdot \,{10}^{4}\mathit{\cdot }{b}_{\text{bp}}+3.0\,\left(\text{mg}\,{\mathrm{m}}^{-3}\right)$$$ (1)

Common methods to derive carbon export from BGC-Argo floats use DO, POC, or nitrate. Some studies looked at nutrient drawdown in the upper layer (Arteaga et al., 2019; Johnson, Plant, Dunne, et al., 2017; Weis et al., 2024), oxygen depletion (Hennon et al., 2016; Su et al., 2022), a combination of both (Plant et al., 2016) or changes in POC only in the upper mixed layer (Lacour et al., 2023; Xing et al., 2020). In this study, we develop a new dual-tracer approach for estimating in situ POC export from the temporal evolution of POC (abbreviated to just C in the below equations) and DO at depths between E_d and 1,000 m. The balance equation of POC and DO can be written as

$$$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}+\overrightarrow{u}\cdot \nabla C={K}_{H}{\nabla }_{H}^{2}C+{K}_{z}\frac{{\mathit{\partial }}^{2}C}{{\mathit{\partial }z}^{2}}+\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{sinking}}+\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{prod}}-\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$$ (2)

$$$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}+\overrightarrow{u}\cdot \nabla DO={K}_{H}{\nabla }_{H}^{2}DO+{K}_{z}\frac{{\mathit{\partial }}^{2}DO}{{\mathit{\partial }z}^{2}}-\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$$ (3)

$$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ are the observed change in POC and DO along the float trajectories. The second term on the left-hand side represents horizontal advection. The first term in the right-hand side represents horizontal turbulent fluxes.

$${K}_{z}\frac{{\mathit{\partial }}^{2}C}{{\mathit{\partial }z}^{2}}$$ is POC resulting from vertical turbulence, thereafter called $$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{turb}}$$, and $${K}_{z}\frac{{\mathit{\partial }}^{2}DO}{{\mathit{\partial }z}^{2}}$$ is DO resulting from vertical turbulence, thereafter called $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{turb}}$$. $$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{sinking}}$$ is the sinking POC exported to the deep ocean, which is what we aim to quantify, and $$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$ are the POC and DO changes due to respiration. Since changes in POC and DO are influenced by both biotic and abiotic effects, we make the following assumptions:

1. Primary production is negligible below the export depth. We consider 200 m as the shallowest export depth to remove any productivity or air-sea exchange happening in the epipelagic zone (0–200 m).

2. Turbulent vertical exchange can influence POC and DO at the upper (E_d) and lower (H_d) boundaries of the mesopelagic zone (200–1,000 m) through vertical diffusivity.

3. The horizontal water mass exchange is minimal between profiles. When deriving changes in POC $$\left(\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}\right)$$ and DO $$\left(\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}\right)$$ between pairs of profiles along the float trajectories, we assume that floats have an equal likelihood of passing through water masses with both positive and negative oxygen changes, which offset each other when using a large array of floats. We applied a criterion to reduce horizontal effects (as detailed in Section 2.2) and demonstrated that the spatial and temporal scales considered support this approximation (see Section 4.3). Consequently, we attribute respiration to heterotrophic activity. We acknowledge that despite these conservative criteria, gradual change in POC and DO over each pair of profiles may still occur and result in an accumulation of bias while deriving $$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}$$, and thus the final interpretation of the results. We advise that this approach is only reliable when using a large array of BGC-Argo floats with broad spatial and temporal coverage. Negative respiration values were removed before computing the total carbon export; see our explanation at the end of this section.

4. The temporal change in oxygen at depth can be a direct proxy for oxygen consumption by heterotrophic activity and carbon degradation, once turbulent vertical exchange is accounted for.

With these assumptions, the changes in POC and DO can be written as

$$$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}=\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{sinking}}+\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{turb}}-\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$$ (4)

$$$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}=\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{turb}}-\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$$ (5)

$$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$ can also be re-written as changes in DO such as:

$$$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{rsp}}=\theta \,\mathit{\cdot }\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{rsp}}$$$ (6)

where

$$$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{rsp}}=\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{turb}}-\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}$$$ (7)

By substituting changes in POC by changes in DO, Equation 4 can be re-written as

$$$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}=\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{sinking}}+\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{turb}}+\theta \hspace*{.5em}\mathit{\cdot }\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}-\theta \cdot \frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{turb}}$$$ (8)

or

$$$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{sinking}}=\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}-\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{turb}}-\theta \,\mathit{\cdot }\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}+\theta \,\mathit{\cdot }\,\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{turb}}$$$ (9)

Specifically, over any given time period between each pair of profiles (profile_t1–profile_t0, Figure 2), our calculation sums the newly introduced particles in the layer due to sinking and vertical turbulent fluxes as well as the particles lost via respiration and vertical turbulent fluxes. Therefore, our method emphasizes POC sinking as the only contributor to carbon export. When integrated over depth, the final equation is:

$$${\text{POC}}_{\mathrm{exp}}=\int \nolimits_{{E}_{d}}^{{H}_{d}}\frac{\mathit{\partial }c}{\mathit{\partial }t}{\vert }_{\text{obs}}dz-\int \nolimits_{{E}_{d}}^{{H}_{d}}\frac{\mathit{\partial }c}{\mathit{\partial }t}{\vert }_{\text{turb}}dz-\theta \cdot \int \nolimits_{{E}_{d}}^{{H}_{d}}\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}dz+\theta \cdot \int \nolimits_{{E}_{d}}^{{H}_{d}}\frac{DO}{\mathit{\partial }t}{\vert }_{turb}dz$$$ (10)

Where POC_exp (for exported POC) is the total carbon rate of change (in mmolC m⁻² d⁻¹) between E_d = max(MLD, Z_eu, 200) and H_d (1,000 m) for a given pair of profiles. θ is the stoichiometric C:O₂ quotient for organic matter respiration (0.69; Anderson and Sarmiento (1994)) and the depth-integration of the vertical divergence of the turbulent fluxes is:

$$$\underset{{E}_{d}{}}{\overset{{H}_{d}}{\int }}\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{turb}}\,dz={\mathrm{K}}_{\mathrm{Z}}\mathit{\cdot }\frac{\mathit{\partial }POC}{\mathit{\partial }z}{\vert }_{{E}_{d}}-{\mathrm{K}}_{\mathrm{Z}}\mathit{\cdot }\frac{\mathit{\partial }POC}{\mathit{\partial }z}{\vert }_{Hd}$$$ (11)

$$$\underset{{E}_{d}{}}{\overset{{H}_{d}}{\int }}\frac{DO}{\mathit{\partial }t}{\vert }_{\text{turb}}\,dz={\mathrm{K}}_{\mathrm{Z}}\mathit{\cdot }\frac{\mathit{\partial }DO}{\mathit{\partial }z}{\vert }_{{E}_{d}}-{\mathrm{K}}_{\mathrm{Z}}\mathit{\cdot }\frac{\mathit{\partial }DO}{\mathit{\partial }z}{\vert }_{Hd}$$$ (12)

Figure 2

Example of particulate organic carbon (POC) and dissolved oxygen (DO) co-located profiles to perform the calculation (from float 5905376, red line on map in panel (b)). Profile t₁ is on average 10 days after profile t₀. The two black dashed lines represent the export depth (E_d) and horizon depth (H_d). The dashed red lines represent the boundaries used to calculate the vertical turbulent diffusive flux of POC and DO at both E_d and H_d. Integrated POC and DO were calculated from the export depth to the horizon depth for both profiles.

K_z represents the turbulent diffusivity coefficient (m⁻² s⁻¹). To estimate depth-integrated $$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{turb}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{turb}}$$, we calculated turbulent fluxes at the depths E_d and H_d. For each profile, we calculated the vertical gradient of POC and DO over a range of ±25 m around E_d and H_d. For instance, if E_d and H_d were at 200 m and 1,000 m, we calculated $$\frac{\mathit{\partial }POC}{\partial \mathrm{z}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }z}$$ between 175–225 m and 975–1,025 m respectively. The vertical gradient was then multiplied by an average K_z within those depth ranges. K_z was derived from direct microstructure measurements collected during two Southern Ocean cruises: SOFine (Naveira Garabato, 2009) and the Diapycnal and Isopycnal Mixing Experiment (DIMES; Gille et al., 2012; Ledwell et al., 2011; Sheen et al., 2013). We used the average profile from both to estimate the turbulence diffusivity at each boundary, averaging over a 50 m depth range similarly to $$\frac{\mathit{\partial }POC}{\partial \mathrm{z}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }z}$$ (Figure S4 in Supporting Information S1), acknowledging that vertical eddy diffusivity can vary significantly across the SO (Whalen et al., 2012; Wu et al., 2011).

We performed this calculation for each pair of profiles for each float. Then, we averaged all pairs of profiles values obtained per month and per zone from the entire fleet, giving a monthly mean climatological value of POC_exp for each zone (mmolC m⁻² d⁻¹). We then integrated in time, giving a climatological total yearly accumulated carbon (molC m⁻² y⁻¹) hereafter called carbon export (note that this is also called annual net community production, or ANCP, in the literature). We finally integrated in space to a full SO estimate for each zone (PgC y⁻¹). Note that before calculating carbon export, negative respiration values were discarded similar to the way respiration data were treated by Su et al. (2022) and Arteaga et al. (2019), who only focus on periods of DO decrease (e.g., positive respiration). This approach fits our focus on the respiration activity. The performance of our method is compared to previous studies in the results and discussion sections.

3 Results

3.1 Spatial and Temporal Distributions of POC, Chla and Dissolved Oxygen

Figure 1 shows the spatial distribution of float profiles for all zones. From a visual inspection, the SIZ, PAZ and SAZ have a relatively even circumpolar spread of profiles. In the STZ, there are more profiles in the Indian and Atlantic sectors compared to the Pacific. Most of the profiles are in the SIZ (5,125; 38%) and PAZ 4,805 (35%), while the SAZ and STZ account for 1,855 (14%) and 1,764 (13%) profiles, respectively. Profiles from the SOCLIM and remOcean fleet are mainly located in the polar Antarctic zone (PAZ) between the sea ice extent and polar front (Figures 1 and S1b in Supporting Information S1).

Climatologically integrated POC from the surface to E_d (monthly mean of POC stock, thereafter called iPOC) shows clear seasonality (Figure 3), with a latitudinal gradient in maximum values. iPOC peaks in the productivity layer in late winter (September) in the STZ, and spring (November) in the SAZ. Further south, iPOC peaks in the PAZ and SIZ in summer (January) and late summer (March) respectively. The highest iPOC is observed in the PAZ during austral summer (8.29 gC m⁻²). The climatological depth-time plots (gridded average monthly profiles in each zone, Figure 4) highlight the vertical and seasonal changes in POC and chla. The exported POC is visible at depth for all zones (Figures 4a–4d). Seasonal variability in POC and chla is greater in the PAZ and SIZ (Figures 4c, 4d, 4g, and 4h) compared to the SAZ and STZ. Focusing on the SIZ (Figure 5), the rate of change in DO and POC through time and depth highlights that during the productivity season (October to April), most of the POC produced in the first 100 m that is exported at depth is respired between Z_p and 200 m (where 200 m is almost always the export depth used for our calculation in the SIZ). Very little further respiration occurs below 200 m in the SIZ (Figure 5).

Figure 3

Monthly time series of depth-integrated POC in the upper layer defined as the surface to the export depth, E_d = max(MLD, Z_eu, 200), such as $$iPOC=\underset{0{}}{\overset{Ed}{\int }}POC\,dz$$, averaged over the four Southern Ocean zones (see legend). Shading represents the standard error.

Figure 4

Climatological depth-time plots of particulate organic carbon (POC; left, a–d) and chlorophyll-a (Chla; right, e–h) for the four zones. The blue and red lines represent the climatological euphotic depth (Z_eu) and mixed-layer depth (MLD) respectively. The white filled circles represent the climatological productivity layer depth (Z_p, the deepest of the Z_eu or MLD). The black line is the 200 m threshold. The climatological export depth E_d is always 200 m, except for the SAZ in August.

Figure 5

Rate of change of (a) dissolved oxygen (DO; $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}$$) and (b) particulate organic carbon (POC; $$\frac{\mathit{\partial }POC}{\mathit{\partial }t}$$) for the sea-ice zone. The blue and red lines represent the climatological euphotic depth (Z_eu) and mixed-layer depth (MLD) respectively. The white filled circles represent the climatological productivity layer depth (Z_p, the deepest of the Z_eu or MLD). The black line is the 200 m threshold.

3.2 Carbon Export Estimates by Frontal Zones

The annual climatological POC_exp across the frontal zones (Figure 6) shows that the PAZ is responsible for larger export in the SO with 11.40 ± 13.72 mmolC m⁻² d⁻¹, followed by the SAZ and STZ with 6.13 ± 9.85 mmolC m⁻² d⁻¹ and 5.48 ± 7.64 mmolC m⁻² d⁻¹ respectively. The SIZ has the smallest observed POC_exp with 2.74 ± 2.92 mmolC m⁻² d⁻¹. The SIZ and the PAZ also stand at opposite extremes when integrated annually, with 0.84 ± 0.24 molC m⁻² y⁻¹ and 3.28 ± 0.99 molC m⁻² y⁻¹ respectively (Figure 7a). The SAZ and STZ have lower carbon export than the PAZ but higher than the SIZ (2.17 ± 0.71 and 1.78 ± 0.64 molC m⁻² y⁻¹ respectively, Figure 7a). Finally, when integrated spatially over the entire area of each zone, the SIZ remains the smallest contributor to the circumpolar carbon export with 0.21 ± 0.09 PgC y⁻¹, followed by the SAZ with 0.54 ± 0.25 PgC y⁻¹. The contribution slightly increases in the STZ with 0.84 ± 0.43 PgC y⁻¹, and the PAZ makes the largest contribution with 1.09 ± 0.46 PgC y⁻¹. The overall carbon export for the entire SO is 2.69 ± 1.23 PgC y⁻¹ (Figure 7b).

Figure 6

Yearly climatological boxplot of POC_exp for the four frontal zones. The red bars indicate the median POC_exp for each zone, while the blue diamonds indicate the mean POC_exp for each zone (monthly median and mean of POC_exp, n = 12). The blue top and bottom edges of the boxes indicate the 75th and 25th percentiles, respectively. The whiskers extend to the extreme data points not considered outliers. Red crosses are outliers.

Figure 7

(a) Annual carbon export per unit area and (b) total area-integrated carbon export for the four frontal zones. Estimates from previous studies are overlaid (see legend), and classified by zones to facilitate the comparison and include: Arteaga et al. (2018, 2019), Weis et al. (2024), Johnson, Plant, Dunne, et al. (2017), MacCready and Quay (2001), Munro et al. (2015), Lourey and Trull (2001), Hennon et al. (2016), McNeil and Tilbrook (2009), Shadwick et al. (2015), Bender and Jönsson (2016), Riser and Johnson (2008), Martz et al. (2008), Schlitzer (2002), Su et al. (2022), Louanchi and Najjar. (2000), Lee (2001), Moore and Abbott (2000), Li et al. (2021), Chang et al. (2014), Pan et al. (2023), and Huang and Fassbender (H&F) (2024). Black bars represent the standard deviation for each estimate. F_poc = POC sinking flux; EP_poc = export potential of POC; EP_doc = export potential of dissolved organic carbon (doc); EP_toc = export potential of total organic carbon (TOC = POC + DOC). Studies listed used BGC-Argo floats, satellite, model and ship-based estimations.

4 Discussion

4.1 Contrast in Biological Seasonality Between Zones

The seasonality of iPOC in the productivity layer is highlighted in Figure 3. The iPOC peak occurs later southward from September in the STZ to March in the SIZ. The magnitude of the iPOC peak is the smallest further north in the STZ, and increases south, except at high latitudes in the SIZ, where iPOC in surface waters is lower than in the PAZ (Figure 3). The low iPOC in the STZ most likely reflects nitrate limitation in these oligotrophic waters (Levitus et al., 1993). On the other hand, the SAZ and PAZ are characterized by deep mixing layers and high macronutrient (nitrate, phosphate, silicate) concentrations (Moore et al., 2013). The strong upwelling and mixing at the subtropical and polar fronts (Carranza & Gille, 2015; Sarmiento et al., 2004) replenishes surface waters with nutrients, potentially leading to higher iPOC from phytoplankton blooms in the SAZ and PAZ regions. In addition in the PAZ, the sea-ice retreat at the SIZ-PAZ boundary can trigger ice edge blooms (Lancelot et al., 1993; Smith & Nelson, 1985). Ship-based studies have reported enhanced levels of chla just north of the ice edge in early December, where large diatoms are abundant in the southern PAZ, while small pennate diatoms and Phaeocystis dominate in the SIZ (Kauko et al., 2022; Landry et al., 2002; Lenss et al., 2024). This may explain why higher POC is also observed in the PAZ, as greater POC might be representative of larger silicifying diatoms. This hypothesis concurs with Arteaga et al. (2019), who found a strong silicate drawdown in the region, increasing the POC ballasting effect ANCP. During spring/summer, the SIZ displays similar characteristics to the PAZ as sea ice reaches its maximum retreat, resulting in fully open water. This may account for the comparable magnitude (to a certain extent) and seasonal trends in terms of iPOC in the SIZ and PAZ, particularly during periods of sea-ice retreat.

In the SIZ, the phytoplankton bloom is strongly influenced by the sea-ice seasonal cycle (Stammerjohn et al., 2012). When sea ice retreats in spring, surface ocean stratification is enhanced and phytoplankton is relieved from light limitation (Taylor et al., 2013; Vaillancourt et al., 2003), which triggers an increase in primary productivity until nutrients become depleted, grazing overcomes primary productivity or sea ice forms again in autumn (Figure S5; Text S1 in Supporting Information S1). Despite the good circumpolar distribution of BGC-Argo profiles (Figure 1), data gaps remain near the Antarctic continent and over continental shelves, where the highest phytoplankton production is usually found, particularly in coastal polynyas (Arrigo & van Dijken, 2003; Liniger et al., 2020; Moreau et al., 2019). This is a known limitation of Argo floats, but sampling under ice has dramatically improved the breadth of data available in recent years. Recent successful BGC-Argo deployments in the Ross Sea show the possibility for more coastal sampling in the future (Cai et al., 2024; Porter et al., 2019).

4.2 Perspective From Previous Studies

Our estimates of carbon export are within the range of previously reported ANCP and carbon export estimates (Figure 7). Many approaches have been used to estimate carbon export and ANCP. Some studies have looked at nutrient drawdown in the upper layer (Arteaga et al., 2019; Johnson, Plant, Dunne, et al., 2017; Munro et al., 2015), but this technique does not account for nutrient replenishment from below (via vertical mixing or advection), heterotrophic activity within the mixed layer, or productivity outside of the defined seasonal production period. Others quantified export below their defined productivity layer from sediment traps (Lourey & Trull, 2001; Pilskaln et al., 2004), a technique which does not account for respiration. Both methods are likely to be prone to underestimation, as respiration can represent up to 90% of the export production in the mesopelagic zone (Jacquet et al., 2011). Thorium isotopes have also been used (Le Moigne et al., 2016; Planchon et al., 2013; Puigcorbé et al., 2017; Smetacek et al., 2012), accounting for all the POC exported from the surface waters. Many of the studies compared in Figure 7b did not extend the Southern Ocean boundary as far north as 30°S, excluding estimates from the STZ. Some studies omitted the SIZ or relied on export depths shallower than 200 m to estimate ANCP from DO depletion (Hennon et al., 2016; Su et al., 2022), including DO changes that would not be solely due to respiration or photosynthesis, therefore introducing bias in their ANCP calculation. Aside from the definition of regions, we argue that a basin scale calculation using all available floats allows for a better estimation of carbon export compared to studies relying on extrapolation from very localized observations. In particular, our work provides a new complementary but distinctive approach that accounts for both changes in POC and DO in the mesopelagic layer and extends previous work to include a circumpolar SIZ estimate, an area which has historically been largely unstudied because of the paucity of observational data under sea ice.

When performing our calculations (based solely on changes in DO, excluding turbulence and POC) over the 100–500 m depth range for comparison with the results of Arteaga et al. (2019), we obtain very similar results (Figure 8, blue bars). Both approaches rely on oxygen drawdown but applied differently. We analyze changes between consecutive profiles, while Arteaga et al. (2019) used full seasonal cycles of 123 floats to estimate respiration terms. Our estimates are 1.11, 2.01, 1.40 and 1.14 molC m⁻² y⁻¹ for SIZ, PAZ, SAZ and STZ, respectively, compared to 0.9, 1.80, 1.80 and 1.25 molC m⁻² y⁻¹ in Arteaga et al. (2019). When the DO turbulence term is included for the same depth range (Figure 8, orange bars), the estimate for the SIZ increases substantially, reaching 5.93 molC m⁻² y⁻¹. The estimate for the PAZ also rises slightly (2.49 molC m⁻² y⁻¹), while the SAZ and STZ estimates remain comparable (1.62 and 1.59 molC m⁻² y⁻¹). Applying the same method (changes in DO + turbulence) over the depth E_d and 1,000 m yields a much lower estimate for the SIZ (0.28 molC m⁻² y⁻¹), while the values for the PAZ, SAZ and STZ slightly increase (Figure 8, yellow bars). These findings highlight two key points. First, accounting for turbulence in shallow productive regions is critical, as it likely captures DO from phytoplankton blooms at the subsurface, and DO changes from air-sea exchange, which are not included by Arteaga et al. (2019) method. Second, the choice of depth range impacts the estimates, particularly in regions where productivity and respiration are concentrated in the surface layer. For instance, in the SIZ, estimates vary dramatically, from 5.93 molC m⁻² y⁻¹ in the upper layer (100–500 m) to 0.28 molC m⁻² y⁻¹ in the mesopelagic layer (E_d–1,000 m).

Figure 8

Carbon export (or ANCP) estimates from this study were obtained using three different methods: changes in dissolved oxygen (DO) only between 100 and 500 m (blue bars), changes in DO and turbulence between 100 and 500 m (orange bars), and changes in DO and turbulence between the export depth E_d and 1,000 m (yellow bars) for the four frontal zones of interest (i.e., SIZ, PAZ, SAZ, and STZ). Estimates from Arteaga et al. (2019) were averaged per zone (cyan dots) to facilitate the comparison. Black bars represent the standard deviation for each estimate.

4.3 Uncertainties and Caveats of Carbon Estimations

The primary goal of our study is to quantify the contribution of each frontal zone to the overall circumpolar SO carbon export. However, sea ice prevents the floats from surfacing and transmitting their data in real time (Hague & Vichi, 2021; Riser et al., 2018). When floats are under sea ice, the location of each profile is linearly interpolated between ice-free GPS fixes. Therefore, errors in distances between consecutive under ice profiles (Chamberlain et al., 2018) may bias our SIZ carbon export estimates compared to PAZ, SAZ and STZ. In the SIZ, 99% of consecutive pairs of profiles under sea-ice are within 50 km from each other (Figure S6a in Supporting Information S1), while the range increases to 150 km in SIZ open water (Figure S6b in Supporting Information S1), suggesting that floats travel less under ice compared to open water. The pairs of profiles used to derive our metrics of interest under sea ice are therefore likely representative of processes in the same water masses. Distances between pairs of profiles for the entire SO are also shown for comparison (Figure S6c in Supporting Information S1). The decorrelation scale, defined as the distance and time after which the correlation between subsequent profiles becomes non-significant at the 0.01 probability level, was also investigated. The average decorrelation scale in space and time is greater than the mean space and time interval between profiles (Figure S7 in Supporting Information S1). These additional analyses suggest that all consecutive pairs of profiles used to derive $$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ likely capture the same water masses.

The sinking rate of Southern Ocean POC partly depends on iron availability (Obernosterer et al., 2008). The SOIREE experiment reported sinking rates reaching 1.6 m d⁻¹, 2.5 m d⁻¹ and up to 4 m d⁻¹ in the SIZ following iron additions (Boyd et al., 2000; Maldonado et al., 2001; Waite & Nodder, 2001). South of Tasmania, Cassar et al. (2015) described a sinking rate of 6 m d⁻¹ on average, reaching a maximum of 19.5 m d⁻¹. Fecal pellet sinking rates were also shown to be high in the SO, ranging from 82 m d⁻¹ to 437 m d⁻¹ north of the Antarctic Peninsula in the top 400 m (Liszka et al., 2019). This wide range of sinking rates observed during ship-based campaigns suggests that the fastest sinking particles could be overlooked by the typical 10 days sampling frequency of BGC-Argo floats, such as the SOCCOM floats. To address this, we conducted a detailed analysis comparing export estimates from SOCCOM, and SOCLIM and remOcean floats only, which indicates that the sampling frequency can significantly influence export calculation (Figure S1; Table S1 and Text S2 in Supporting Information S1). While these limitations should be considered, the limited high frequency sampling data in this study and the restrictive location of SOCLIM and remOcean floats compared to SOCCOM floats prevent a definitive conclusion that such events were consistently missed. This highlights the need for broader circumpolar analysis using high profiling frequency floats.

Another avenue to address the sampling frequency deficiency would be to merge profiles from all floats for a given province and derive parameters from the closest profiles in time, regardless of which float made the profiles. Opting for this method certainly allows for a greater sampling frequency compared to a usual 10-day interval, from the same day to 2 days in most cases (Figures S8a–S8d in Supporting Information S1). However, this method implies that every calculation made from the closest time t₀ to time t₁ can be performed on profiles from different floats that are in different areas, and possibly different water masses within the same zone. With this approach, about 90% of all consecutive profiles used to derive $$\frac{\mathit{\partial }C}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ are from different floats (Figure S8e in Supporting Information S1), with large spatial gaps (Figure S9 in Supporting Information S1). Therefore, we believe that comparing consecutive profiles from the same quasi-Lagrangian float, and then averaging within months and zones, introduces less bias and is the most suitable method for this study.

If large, fast-sinking particles disaggregate into smaller particles or dissolved organic matter (DOM) that cannot be detected by the bio-optical sensor, this would lead to underestimates of POC_exp. If some of these particles or DOM are respired within the layer of interest, the underestimation of POC_exp could be minimized. Huang et al. (2023) found enhanced dissolved organic carbon (DOC) in the SIZ and STZ, so our export estimates could therefore be particularly underestimated in these zones. This argument also applies to very small particles produced in the surface layer and sinking into the mesopelagic but still missed by the backscatter sensor.

Considering the floats as quasi-Lagrangian, looking at large spatial and temporal scales (monthly variability) and averaging from an extended fleet could mean that smaller scale positive and negative changes in oxygen balance each other (Hennon et al., 2016; Martz et al., 2008; Najjar & Keeling, 1997). If we do not include the turbulence terms, our basin-integrated estimate is 2.52 PgC y⁻¹, representing a small 6% difference, as the calculated turbulence terms are on average about two orders of magnitude smaller than $$\frac{\mathit{\partial }POC}{\mathit{\partial }t}$$ and $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}$$. This implies that below 200 m, such terms could be reasonably neglected as they have little impact considering the large size sample, especially at depth (Bilheimer et al., 2021). However, turbulence may be relevant above 200 m where vertical gradients can be large and higher productivity occurs, explaining our higher estimates in the SIZ (Figure 8, orange bar) than those from Arteaga et al. (2019) who ignored vertical mixing.

As anticipated from previous studies (Palevsky & Doney, 2018), our results and conclusions are sensitive to the choice of export depth, particularly if we select E_d as E_d = max(MLD, Z_eu), and do not consider 200 m as a potential export depth. As mentioned in the method section, E_d calculated this way can be very shallow during the spring/summer season and include high DO and POC values over the upper 200 m. The SIZ is the most sensitive to these nuances, because of high productivity and shallow respiration (Figure 5a) (Briggs et al., 2018; Usbeck et al., 2002). Using E_d = max(MLD, Z_eu) increases the carbon export estimates in the SIZ to 0.58 compared to 0.21 PgC y⁻¹. There are smaller differences for the PAZ, SAZ and STZ (1.55, 0.81 and 1.09 PgC y⁻¹ respectively, compared to 1.09, 0.55, 0.85 PgC y⁻¹ when using E_d = max(MLD, Z_eu, 200)). Using E_d = max(MLD, Z_eu) increases the total SO estimates to 4.03 PgC y⁻¹, with the SIZ contributing up to 14%.

Including 200 m as a potential export depth, such as E_d = max(MLD or 200); E_d = max(Z_eu or 200), or E_d = max(MLD or Z_eu or 200), very little variability in carbon export estimates is observed regardless of the chosen export depth (Figures S10 and S11 in Supporting Information S1). The small difference is unsurprising as most of the export depth across the entire float array is 200 m (Figure S12 in Supporting Information S1). However, the horizon depths seem more important in the PAZ, STZ and SAZ, with a greater increase in carbon export with depth. This implies that respiration occurs more evenly through the water column (i.e., the upper mesopelagic layer and down to 1,000 m depth) in these regions compared to the SIZ, potentially because of more abundant mesopelagic zooplankton communities (Smetacek et al., 2004) via vertical export of zooplankton fecal pellets (Cavan et al., 2015; Le Moigne, 2019). Zooplankton usually migrate to the surface and feed at night to avoid predation, and defecate at deeper depths during the day (Steinberg & Landry, 2017), resulting in carbon transfer deeper in the water column and therefore deeper respiration. This hypothesis is also supported by our results in Figure 8, where we find higher carbon export estimates derived from $$\frac{\mathit{\partial }DO}{\mathit{\partial }t}{\vert }_{\text{obs}}$$ only, at depth (E_d–1,000 m), in the PAZ, SAZ and STZ compared to the SIZ (Figure 8, yellow bars). This transport of organic carbon by mesozooplankton may represent less than 40% of the total POC flux (Turner, 2015). In addition, fecal pellets can be highly resistant to degradation (Riou et al., 2018; Tamburini et al., 2006), leading to high carbon flux and transfer in the water column.

Another factor to consider is the calcium carbonate signal in the Great Calcite Belt, extending from 30° to 60°S. Balch et al. (2016) showed that b_bp is high in this area due to high particulate inorganic carbon (PIC) from calcification. Using our method, it is not possible to distinguish PIC from POC, so the presence of PIC likely causes some overestimation of POC. Methods have been developed to detect coccolithophore blooms using BGC-Argo floats (Terrats et al., 2020) based on b_bp and chla. However, we argue this would likely have little effect on our carbon export estimation as (a) PIC was shown to have very little contribution to annual net community production compared to POC (Haskell et al., 2020), (b) high calcium carbonate production does not increase POC export (Balch et al., 2016) and (c), most of the calcium carbonate is remineralized in the photic zone, therefore having little effect on export (Ziveri et al., 2023). Recent budgets of distinctive carbon pools also highlighted the lower contribution of PIC to the SO carbon export potential (Huang et al., 2023; Huang & Fassbender, 2024) compared to POC and DOC (70% for POC: 2.15 PgC y⁻¹; 20% DOC: 0.61 PgC y⁻¹ and 10% PIC: 0.31 PgC y⁻¹). This however underscores that while POC is the dominant contributor and a reliable proxy to estimate carbon export, other components (especially DOC) must also be included in the overall budget (Nevison et al., 2018). This is particularly crucial since considering all carbon tracers in the total carbon budget shifts the SO from a CO₂ source to a sink (Huang et al., 2023), drastically altering our understanding of the SO dynamics. Huang and Fassbender (2024) also revealed comparable zonal contributions (Figure 7a), with the SIZ and STZ contributing the least to POC fluxes, while the PAZ and SAZ contributed the most.

5 Conclusions

In this study, we developed a novel methodology to calculate circumpolar SO carbon export in the mesopelagic zone. We found that the PAZ contributes the highest (41%) to the total circumpolar carbon export, while the SIZ contributes only 8%, and the total SO estimate based on BGC-Argo float observations from 2014 to 2022 is 2.69 ± 1.23 PgC y⁻¹. We highlight the strong shallow respiration that takes place in the upper 200 m in the SIZ, showing how export and respiration appear to be closely linked in this region, pointing to a very active marine ecosystem prone to high variability in environmental conditions that can influence productivity, such as the recent record lows in winter sea-ice extent. We also demonstrate that the turbulence term may be negligible at depth (below 200 m) and that the choice of methodology (i.e., the choice of investigated layer) is critical for deriving carbon export. Several studies emphasize the need for monitoring the SIZ to improve the accuracy of SO carbon sink estimates, as this could significantly affect our understanding of global climate variability. However, our findings indicate that, at present, the SIZ might contribute only a small portion of SO carbon export. We do not yet have a complete understanding of how large fluctuations in sea-ice may influence CO₂ sequestration, or whole-of-ecosystem processes, or potential feedbacks.

Acknowledgments

Guillaume Liniger, Peter Strutton and Delphine Lannuzel thank the University of Tasmania, the Australian Research Council Centre of Excellence for Climate Extremes (CE170100023), and the Australian Centre for Excellence in Antarctic Sciences (ACEAS; SR200100008) for their support. Sebastien Moreau received funding from the Research Council of Norway (RCN) for the project “I-CRYME: Impact of CRYosphere Melting on Southern Ocean Ecosystems and biogeochemical cycles” (Grant 335512), the Norwegian Centre of Excellence “iC3: Center for ice, Cryosphere, Carbon and Climate” (Grant 332635) and the EU for the project “WOBEC: Weddell Sea Observatory and Biodiversity and Ecosystem Change” (Grant 350906). Delphine Lannuzel is funded by the Australian Research Council through a Future Fellowship (FT190100688). This work was sponsored by NSF's Southern Ocean Carbon and Climate Observations and Modeling (SOCCOM) Project under the NSF Awards PLR-1425989 and OPP-1936222 and 2332379, with additional support from NOAA and NASA. Logistical support for this project in the Antarctic was provided by the U.S. National Science Foundation through the U.S. Antarctic Program. We would also like to extend our thanks to the BGC-Argo program and the several Southern Ocean projects that make the data freely available. These include SOCLIM, remOcean and SOCCOM. Data were collected and made freely available by the Southern Ocean Carbon and Climate Observations and Modeling (SOCCOM) Project funded by the National Science Foundation, Division of Polar Programs (NSF PLR -1425989 and OPP-1936222 and 2332379) supplemented by NASA, and by the International Argo Program and the NOAA programs that contribute to it (http://www.argo.ucsd.edu, https://www.ocean-ops.org/board). The Argo Program is part of the Global Ocean Observing System. The remOcean project is funded by the European Research Council (GA 246777) and the SOCLIM project is funded by BNP Paribas, ENS, UPMC and CNRS. We thank all the people involved in the conception, deployment, and quality control of the BGC-Argo float. We finally extend our gratitude to the reviewers for their help in improving the quality of the manuscript. Open access publishing facilitated by University of Tasmania, as part of the Wiley - University of Tasmania agreement via the Council of Australian University Librarians.

Conflict of Interest

The authors declare no conflicts of interest relevant to this study.

Open Research

Data Availability Statement

All BGC-Argo float data (Argo, 2024) and sea-ice concentration (Copernicus Climate Change Service, 2020) used in this study are publicly available. All analysis were conducted in MATLAB.