Large-scale estimates of the Southern Ocean CO2 sink do not adequately resolve the fluxes associated with Antarctic continental shelves. Using a mechanistic three-dimensional biogeochemical model of the Ross Sea, we show that Antarctic shelf waters are a strong sink for CO2 due to high biological productivity, intense winds, high ventilation rates, and extensive winter sea ice cover. Net primary production (NPP) in these waters is ∼0.055 Pg C yr−1. Some of this carbon sinks to depth, driving an influx of CO2 of 20–50 g C m−2 yr−1. Although currently unaccounted for, the total atmospheric CO2 sink on the Ross Sea continental shelf of 0.013 Pg C yr−1 is equivalent to 27% of the most recent estimate of the CO2 sink for the entire Southern Ocean. Given these results, these and other highly productive waters around the Antarctic continent need to be included in future budgets of anthropogenic CO2.
 Of the 7–9 Pg C yr−1 released as CO2 into the atmosphere by humans, 25–35% is taken up by the ocean through physical and biological processes. A variety of methods have been employed to assess the role of the ocean as a sink for atmospheric CO2, including atmospheric and oceanic inversions, analysis of transient tracers, measured (or inferred) surface ocean partial pressure of CO2 (pCO2), and global climate models. These techniques generally agree that the largest CO2 sinks are found in the North Atlantic and the Southern Ocean, but disagree as to the magnitude of these sinks, particularly in the Southern Ocean, where historical estimates range widely from 0.10 to 0.56 Pg C yr−1 [Takahashi et al., 2002; Gloor et al., 2003; Roy et al., 2003; Gurney et al., 2004; Sabine et al., 2004; Wetzel et al., 2005; Mikaloff Fletcher et al., 2007; McNeil et al., 2007]. However, the most recent analysis of the largest available observational pCO2 data set supports estimates at the lower end of this range, suggesting that the Southern Ocean removes only 0.05 Pg C yr−1 from the atmosphere (T. Takahashi et al., Climatological mean and decadal change in surface ocean pCO2 and net sea-air CO2 flux over the global oceans, submitted to Global Biogeochemical Cycles, 2008). Unfortunately, all of these estimates are of low spatial resolution, making it impossible for them to fully resolve many of the most productive regions of the Southern Ocean, including much of the continental shelf (waters of <1000 m depth) and its associated coastal polynyas (areas of open water surrounded by sea ice), where the air-sea flux of CO2 (FCO2) is expected to be relatively high.
 The narrow continental shelves around Antarctica are important ventilation sites because they are regions of active Antarctic Bottom Water (AABW) formation [Jacobs et al., 1970, 1985; Gordon et al., 2004]. Beginning in austral autumn, cold and intense offshore winds initiate the growth of sea ice within the coastal polynyas (Figure 1a) that ring the Antarctic continent [Arrigo and Van Dijken, 2003]. As the associated rejection of salt increases surface salinity, these dense waters sink, driving deep convection and bringing waters high in both nutrients and CO2 to the surface. Once the sea ice melts in spring, some CO2 escapes to the atmosphere, but because surface nutrient and trace metal concentrations on the shelf are high [Fitzwater et al., 2000; Coale et al., 2005], intense phytoplankton blooms develop, rapidly reducing pCO2 to very low levels [Sweeney, 2003]. Some of this newly fixed organic carbon sinks below the surface layer, driving an influx of atmospheric CO2. How much total carbon is exported off the shelf, including the anthropogenic component (Cant), is currently not known.
 To quantify the importance of the Antarctic continental shelf and its associated coastal polynyas as sinks for anthropogenic CO2, we implemented a coupled physical/biogeochemical model of the southwestern Ross Sea [Arrigo et al., 2003] with the spatial resolution required to accurately simulate phytoplankton dynamics and CO2 fixation, carbon export, and air-sea CO2 exchange on the shelf in response to physical forcing (for model validation see auxiliary material). The Coupled Ice, Atmosphere, and Ocean (CIAO) model was run over an annual cycle with identical biology under both pre-industrial (280 μatm) and contemporary (380 μatm) atmospheric pCO2 conditions; anthropogenic CO2 fluxes and inventories were calculated as the difference between these two runs. Model physics were spun up under climatological forcing for 80 years, at which time the biogeochemical components of the model were switched on and the model was run until the carbon system on the continental shelf had reached steady-state (10 years, about double the residence time of water on the shelf [Jacobs et al., 1970]).
3. Results and Discussion
 The spatial pattern of modeled annual net primary production (NPP) by phytoplankton on the Ross Sea continental shelf (Figure 1b), which is the same in the pre-industrial and contemporary model runs, agrees well with estimates made both in situ [Arrigo et al., 2000] and using satellite ocean color data [Arrigo and Van Dijken, 2004, 2007; Arrigo et al., 2008]. Annual NPP is highest (and pCO2 lowest) in the relatively ice-free central Ross Sea polynya region (Figure 1a), ranging from ∼140 to >200 g C m−2 yr−1 (Figure 1b). NPP is lower along the western side of the continental shelf (west of 170°E) where ice cover is more persistent (Figure 1a) and strong winds mix the water column more deeply. The off-shelf waters in the northeast sector of our study region exhibit extremely low annual NPP because of persistent sea ice cover and low trace metal concentrations [Fitzwater et al., 2000; Coale et al., 2005]. Total NPP on the Ross Sea continental shelf is approximately 54.7 Tg C yr−1.
 In winter, surface waters of the Ross Sea are supersaturated in pCO2 (∼425 μatm, Figure 2a) due to upwelling of deep, high CO2 water onto the shelf and convective mixing, initiated by cooling and brine rejection during sea ice formation, that mixes this high CO2 water to the surface. Once sea ice melts in the austral spring and light becomes sufficient to support net phytoplankton growth, rates of NPP increase and surface pCO2 in the polynya is rapidly reduced to below atmospheric levels in both the pre-industrial and contemporary runs (Figure 2a). Surface pCO2 on the shelf drops to as low as 150 μatm at the peak of the phytoplankton bloom and persists at low levels from December through March in both model runs, in excellent agreement with in situ observations (Figure 2a). The calculated summer surface pCO2 deficits are virtually identical for the two runs because both NPP and the CO2 content of the waters that upwell onto the continental shelf (and eventually into surface waters) are assumed to be the same. Although intermediate waters of the Southern Ocean have 10–30 μmol kg−1 more total CO2 (TCO2) today than they had in pre-industrial times [Sabine et al., 2004], the assumption in the model that subsurface TCO2 has remained unchanged impacts estimates of FCO2 by only <3% (see sensitivity analysis in auxiliary material). Increased winds in late austral summer and autumn enhance FCO2, with maximum fluxes of −0.55 and −0.85 g C m−2 d−1 (Figure 2b) in the pre-industrial and contemporary runs, respectively (negative sign denotes flux from atmosphere to ocean), causing surface pCO2 to begin to rise in February (Figure 2a). Despite similar surface pCO2 in the two runs, FCO2 is higher in the contemporary run because its higher atmospheric pCO2 results in a larger ΔpCO2. Remineralization of organic matter and enhanced convective mixing of deep CO2-rich waters into the surface layer also increase surface pCO2, thereby reducing FCO2 throughout March (Figure 2b). By April, newly-formed sea ice inhibits further CO2 exchange even as surface water pCO2 rises to >400 μatm (Figure 2a), in excess of atmospheric levels.
 In the contemporary run, annual FCO2 ranges from −20 to −50 g C m−2 yr−1 in waters associated with the Ross Sea polynya (Figure 1d). The spatial distribution of the annual FCO2 in both model runs is directly related to spatial variability in annual NPP (Figure 1b); high NPP leads to larger drawdown of CO2, generating regions of low surface pCO2 and a large ΔpCO2. FCO2 was also mostly negative in the pre-industrial run (Figure 1e), but because ΔpCO2 was smaller than in the contemporary run (due to lower atmospheric pCO2), so was FCO2. FCO2 depends not only on ΔpCO2 but also on wind speed, and thus FCO2, calculated as a fraction of NPP (Figure 1c), is greatest near the coast along the western margin of the continental shelf where winds are most intense. In these coastal regions, annual FCO2 is equivalent to 40–70% of annual NPP.
 Because of its unique physical and biological characteristics, including high rates of ventilation and primary production, annual FCO2 on the Ross Sea continental shelf, including less productive offshore waters with lower wind speeds, is remarkably high, amounting to −13.3 Tg C yr−1 (24% of annual NPP for these waters). Furthermore, approximately 70% of contemporary FCO2 is anthropogenic, with Cant fluxes ranging from −40 to −70 g C m−2 yr−1 along the western continental shelf (Figure 1f) and approximately −20 to −30 g C m−2 yr−1 further offshore. Surprisingly, although the Ross Sea continental shelf comprises only 0.36% of the open water area in the Southern Ocean south of 50°S, its annual rate of FCO2 is equivalent to 27% of the most recent estimate of the total CO2 sink for the entire Southern Ocean (which does not include the high FCO2 we report for the Ross Sea continental shelf) (Takahashi et al., submitted manuscript, 2008). Total FCO2 for all Antarctic continental shelves is likely to be considerably higher, since the Ross Sea accounts for only 20% of total shelf area around Antarctica (see auxiliary material). Our results suggest that Antarctic continental shelves, which are largely unaccounted for in global estimates of the air-sea flux of Cant, may represent the most efficient long-term sinks of Cant in the global ocean.
 Our physical model indicates that much of the CO2 entrained in surface waters is ultimately moved off the shelf in the deep waters that form there. Because of active mixing and deep convection on the continental shelf, both the pre-industrial (Figure 3b) and anthropogenic (Figure 3c) components of contemporary TCO2 (Figure 3a) penetrate to the bottom. In these shelf waters, concentrations of Cant are approximately 12–25 μmol kg−1 throughout the water column, in good agreement with previous estimates of Cant from the Ross Sea made using the MIX (Cant = 10–25 μmol kg−1) and TrOCA (Cant = 12–30 μmol kg−1) methods [Sandrini et al., 2007]. Eventually, this deep Cant is entrained in bottom waters that move downslope in a narrow northward-flowing plume along the western margin of the continental shelf (Figure 4). CIAO estimates that these waters flow off the shelf at a rate of 4.2–5.3 Sv (106 m3 s−1), consistent with previous measurements of off-shelf bottom water transport that range from 3.9–5.0 Sv [Gordon, 1974; Häkkinen, 1995; Orsi et al., 1999]. Energetic plumes of dense water of the same temperature (−1.7°C to −0.8°C) and salinity (34.66 to 34.80) as those in the model were observed in this location during the AnSlope program [Gordon et al., 2004], reaching velocities of 1 m s−1. The saltiest waters observed on the Ross Sea continental shelf [Jacobs et al., 1985; Orsi et al., 1999] form in the western shelf region where FCO2 is highest, providing an efficient mechanism for transporting Cant into the deep ocean. After flowing down and off the shelf, these dense waters move northward and spread into the deep ocean, carrying Cant with them (Figure 4).
 New approaches have hinted at the importance of Antarctic continental shelf processes in the sequestration of anthropogenic CO2 [Lo Monaco et al., 2005] but our study is the first to quantify the magnitude and distribution of this flux and to relate it to meteorological and oceanographic processes that are unique to the shelf environment. Recent evidence has suggested that the ability of much of the Southern Ocean to take up anthropogenic CO2 may have decreased in the last two decades due to a poleward shift and intensification of westerly winds enhancing ventilation of deep CO2-rich water [Le Quere et al., 2007]. Additionally, it has been suggested that as more anthropogenic CO2 enters the surface ocean, particularly in regions with strong stratification, the ability of the ocean to act as a CO2 sink diminishes [Le Quere et al., 2007]. While these processes may impact continental shelf systems, their effect will be minimized in regions with both high rates of primary production and vigorous deep-water ventilation, such as on the Antarctic continental shelf, where CO2 can be efficiently transported from the surface to the deep ocean. Thus, the relative importance of anthropogenic CO2 sinks in regions like the Antarctic continental shelf is likely to increase in years to come.
 This research was supported by the Ocean Carbon Sequestration Research Program, Biological and Environmental Research (BER), U.S. Department of Energy (grant DE-FG03-01ER63176) and the NASA Oceanography Program (grant NAG5-11264).