The amount of anthropogenic CO2 (Cant) that entered the Weddell Sea between 1992 and 2008 (Cant1992–2008) was assessed using the extended multiple linear regression (eMLR) method. In the Warm Deep Water (WDW) and the Weddell Sea Bottom Water (WSBW), Cant1992–2008 values were insignificant, whereas values as high as 8 μmol kg−1 were observed over the shelf. Cant1992–2008 concentrations in the surface layer varied with latitude between 2 and 11 μmol kg−1. Weak intrusion of anthropogenic CO2 into Weddell Sea Deep Water (WSDW) was demonstrated (Cant1992–2008 yields 1.5–2 μmol kg−1). That more Cant1992–2008 was found in the WSDW than in the WSBW is surprising, but can be explained by intense ventilation of the WSDW originating from east of the Weddell Gyre. The invasion of Cant1992–2008 provokes a shift in the equilibria of the carbonate system, resulting in acidification and reduction of CO32− concentration. The mean decrease of pH in the upper 200 m layer was 0.016. The largest decrease of calcite and aragonite saturation states was observed at the surface. This implies that surface waters might become undersaturated with respect to aragonite in the future while the underlying WDW is still saturated. Results of this analysis suggest that complete undersaturation of surface waters in the Weddell Sea will be reached after the 21st century.
 Since the 18th century, CO2 emissions have been rising as a result of burning of fossil fuel and land use changes. During 2005 and 2006, global CO2 emissions exceeded the worst case fossil fuel emission scenario A1FI (intensive dependence on fossil fuels) of the Intergovernmental Panel on Climate Change (IPCC) [IPCC, 2001; Raupach et al., 2007; J. Canadell et al., Recent carbon trends and the global carbon budget, http://www.globalcarbonproject.org/carbonbudget/index.htm, 2007]. Not all the emitted CO2 stays in the atmosphere: land and oceans absorb between 25% and 30%, respectively. Thus approximately 45% of the produced greenhouse gas remains in the atmosphere where it constitutes the largest contributor to anthropogenic global warming [Canadell et al., 2007].
 The addition of anthropogenic CO2 increases the H+ concentration and decreases the CO32− concentration, an effect nowadays known as ocean acidification. Global mean surface ocean pH has already fallen from 8.2 to 8.1 because of uptake of anthropogenic CO2, and its further decrease threatens marine organisms and their habitats. Not only acidification as such but also its high rate of change may cause damage to ecosystems as marine organisms are adapted to their environment. Changes in ocean chemistry, especially increasing acidity, have direct (i.e., physiological) and indirect effects. The latter include reduction of calcite and aragonite saturation state, thus influencing calcifying organisms; dissolution of metals in complex form that can be toxic in free dissolved state; changes in availability and composition of nutrients, among others [Kleypas et al., 1999; Raven et al., 2005; Fabry et al., 2008].
 Within the Southern Ocean, major ventilation occurs in the Weddell Sea, a large cyclonical gyre; its water mass structure is relatively stable and different water masses can be defined. From the ocean floor to the surface, the following water masses are found: Weddell Sea Bottom Water (WSBW), Weddell Sea Deep Water (WSDW), Warm Deep Water (WDW) and Antarctic Surface Water (AASW) or Winter Water (WW).
 The Antarctic Circumpolar Current (ACC) transports Circumpolar Deep Water (CDW) that is dragged into the gyre at its eastern edge. This warm, saline, and poorly oxygenated circumpolar water mass is transformed to WDW by mixing with colder, less saline Weddell Gyre water. WDW can be recognized by a temperature and salinity maximum at intermediate depths (θ > 0°C) [Gordon, 1971; Foster and Carmack, 1976]. The WDW is modified on its pathway through the Weddell Gyre; at and near the continental shelf the modified WDW is involved in mixing processes with shelf water which is very dense through the release of heat to the cold atmosphere and rejection of salt during sea-ice formation. The mixing product is able to sink to abyssal depths and depending on its composition is then called WSDW or WSBW [Killworth, 1983; Naveira Garabato et al., 2002; Fahrbach et al., 1995]. Shelf water is a water mass that occurs south of the Antarctic slope front [Jacobs, 1991]. Newly formed WSBW is characterized by minima in temperature and salinity and a maximum in oxygen content (θ < −0.7°C) [Carmack, 1974; Carmack and Foster, 1975]. WSBW is too dense to leave the Weddell-Enderby basin. It mixes with overlying water masses to form WSDW (−0.7°C < θ < 0°C) [Foster and Carmack, 1976] which is able to exit the Weddell Gyre [Orsi et al., 1993]. AASW is extremely cold, but fresher than the underlying water masses (θ > −1.7°C) [Grosfeld et al., 2001]. AASW is uniformly spread from the polar front to the continental margins of Antarctica [Orsi et al., 1995]. It is found in the top hundred meters in the interior of the Weddell Sea, whereas the thickness of the layer increases to more than 600 m off the shelf edge [Fahrbach et al., 1995].
 The objective of this study is to contribute to the understanding of the response of the Southern Ocean to increased CO2 and its further CO2 uptake as exemplified in this particular and important region, namely the Weddell Sea, where major deep and bottom water formation takes place. We apply the extended multiple linear regression method [Friis et al., 2005] on two data sets from the same region.
2. Data Sets
 Two data sets 15.5 years apart were used: new data from FS Polarstern cruise ANT XXIV/2 in austral summer 2007/2008 [Bathmann, 2010] was analyzed and compared with the ANT X/4 data set from an FS Polarstern cruise in austral winter 1992 [Lemke, 1994]. Data from the FS Polarstern cruise ANT XIII/4 in austral summer/autumn 1996 was used to estimate nutrient concentrations (see section 3.2) which are lacking for ANT XXIV/2 (Table 1).
Table 1. Overview of Data Sets Available for eMLR Analysis, Measured Variables, and Qualitya
 For ANT XXIV/2 two parameters of the carbonate system, total carbon (CT) and total alkalinity (AT) are available, whereas at ANT X/4 only CT was measured. During ANT X/4 and ANT XIII/4 two transects were sampled: one along the Prime Meridian and one across the Weddell Gyre from Kapp Norvegia to the island of South Orkney and Joinville Island, respectively (Figures 1a and 1b). ANT XXIV/2 provides data from the Prime Meridian and additional sections along 3°E and 3°W. All these three sections were analyzed together as one data set (Figure 1c).
CT was measured by coulometry at both cruises (see Hoppema et al.  for ANT X/4 and Dickson et al.  for ANT XXIV/2). During the ANT XXIV/2 expedition, Certified Reference Material (CRM) from Prof. Andrew Dickson was used for standardization [Dickson et al., 2007]. The precision for CT from cruise ANT XXIV/2 was estimated on the basis of 94 pairs of duplicate samples. The mean absolute difference of the duplicates is 0.8 μmol kg−1. During ANT X/4 CRMs had not yet been made available. However, ANT X/4 carbon data are part of the GLODAP carbon data base [Key et al., 2004] and they were quality controlled and assessed against other carbon data from the Southern Ocean. No offset of ANT X/4 data was found, indicating that the accuracy of the data is good. This result was confirmed by the recent carbon synthesis project CARINA [Hoppema et al., 2009]. The precision as estimated from duplicate analyses is 0.9 μmol kg−1 [Hoppema et al., 1995].
AT of cruise ANT XXIV/2 was measured by potentiometric acid titration [Dickson et al., 2007], while CRMs were used for standardization. The precision was estimated to be 2.0 μmol kg−1 based on duplicate analyses.
 During ANT X/4, oxygen concentrations were analyzed by Winkler titration and automatic endpoint detection. Potassium iodate was used as a standard and the precision was estimated to be 0.7 μmol kg−1. Oxygen concentrations of ANT XXIV/2 were determined by Winkler titration using an automated photometric endpoint detection system built by the Scripps Institute of Oceanography. Standard solutions (potassium iodate, 0.012 N) that were prepared at home in large batch quantity and tested before the voyage were used. The precision for ANT XXIV/2 oxygen data based on duplicate samples is 0.09 μmol kg−1.
 For the extended multiple linear regression (eMLR) analysis (see section 3.1), data should be restricted to a coherent ocean region. The ACC marks the northern boundary of the Weddell Gyre. As only the Prime Meridian transect was sampled on both expeditions, data from the Prime Meridian within the Weddell Gyre, i.e., south of 56°S and between 3.5°W and 3.5°E, were used.
3. Computational Methods
3.1. Extended Multiple Linear Regression
 The eMLR method presented by Friis et al.  is a derivative from the MLR approach [Wallace, 1995] for estimating the temporal increase of anthropogenic carbon. Data sets from different years are used to calculate MLRs that explain the CT content for each year. Subtracting these two equations from each other provides an estimate about anthropogenic carbon content. The idea of the eMLR is the following: assuming that hydrographic properties of the water masses and the underlying natural relationship between the input variables and CT stay the same, physical and biological variations are canceled out. Therefore, the information about anthropogenic carbon is carried only by the regression coefficients and not by the variables. Hence, the result of the eMLR analysis is equal to the increase of carbon due to anthropogenic CO2 increase in the atmosphere. An advantage is that the error of the eMLR is smaller than the error of the MLRs [Friis et al., 2005].
 Input variables are hydrographic and nutrient data. Alkalinity data exist only for cruise ANT XXIV/2, but this data set lacks nutrient data. The variables available for both data sets are temperature, salinity, and oxygen (Table 1). MLRs with all possible combinations of these three variables were tested and the best results were obtained when all variables were included. Potential temperature and salinity are typical characteristics of a water mass, which explain most of the total CO2 concentration, but not all. When in contact with the atmosphere, oceanic CO2 of the surface layer tends to move toward equilibrium with atmospheric CO2. Oxygen is related to CT via primary production, respiration processes and decomposition of organic matter, i.e., processes that are considered to be quantitatively stable with respect to rising CO2 concentrations. Oxygen as a variable in the MLR is assumed to account for modification of CT due to these biological processes. In former studies [e.g., Friis et al., 2005], Apparent Oxygen Utilization (AOU) was used instead of oxygen as one of the independent variables as oxidation is of interest here. However, eMLR does not consider stoichiometry and there are some issues with undersaturation of oxygen in water masses which are ventilated in the Weddell Sea. Variable oxygen concentrations in the source waters would erroneously imply variations in the AOU. It is the small oxygen variations that matter.
 Residuals for the MLR of the 2008 data with these variables showed a bias to negative values with increasing depth. This is explained by the fact that deep data points are underrepresented in this data set. In order to assure that all depth levels are equally weighted within the fit, pressure was added as a variable resulting in evenly distributed residuals (Figures 2a and 2b). Pressure is usually not seen as a typical variable to be used in an eMLR, but there are several reasons to justify it. First, hydrography in the Weddell Sea is relatively stable, as described in section 1. Variability of the thickness of the surface layer over the shelf is less problematic because the use of pressure as variable only affects deep layers as the regression coefficients for pressure are on the order of 10−4 to 10−5 (Table 2). Second, the main objective of eMLR is to obtain the best correlation for each data set and pressure significantly improves the fit. This can be done in particular because the method does not claim to infer stoichiometric ratios from the coefficients [Friis et al., 2005] and including pressure only emphasizes this.
Table 2. Derived Multiple Linear Regression Coefficients and Statistics for a CT Prediction Using Prime Meridian ANT X/4 and ANT XXIV/2 Data, According to Equation (2)a
ANT X/4 1992
ANT XXIV/2 2008
R2 is the coefficient of determination, σ is the standard deviation, and n is the number of data points that were used for the fit.
b0 (μmol kg−1)
b1 (μmol kg−1°C−1)
b2 (μmol kg−1)
b4 (μmol kg−1 dbar−1)
4.80 × 10−5
6.76 × 10−4
σ (μmol kg−1)
 Statistics (Table 2 and Figure 2) for the MLRs show that the variables we use are suitable for predicting CT: the standard deviations (Table 2) of the predicted values are low (σ = 3.8 μmol kg−1 for the historic data set and σ = 4.8 μmol kg−1 for the modern data set, i.e., equal or lower than those of Friis et al. , who yield σ = 4.7 μmol kg−1 for the historic and σ = 6.6 μmol kg−1 for the modern data set). Residuals for both cruises are normally distributed, this may well be the most important criterion for valid implementation of eMLR.
 The multiple linear regression is of the form
where the units of CT and O2 are μmol kg−1, θ is in °C and p is in dbar.
 Although in many studies [e.g. Sabine et al., 1999; Wallace, 1995] using the MLR method to predict CT, the surface layer was excluded, inclusion of surface data is not unusual according to some recent studies [Friis et al., 2005; Feely et al., 2003]. Friis et al.  justify the input of surface data by the assumption that temporal changes in preformed values are either insignificant or associated with changes in preformed CT that are accounted for by the regression coefficients. This is especially valid for the Weddell Sea being an upwelling region; hence surface data are included in the analysis. This argument persists when it comes to seasonal variations as seasonal changes in CT are accompanied by changes in temperature, salinity, and oxygen. The equation describing the eMLR for the difference between 1992 and 2008 is (see also Table 2):
 Note that Cant1992–2008 is the change in CT observed between 1992 and 2008. Variables θ, S, O2, and p can be used either from the ANT X/4 data set (“forward case”) or from the ANT XXIV/2 data set (“backward case”). The data set with the highest quality should be used to calculate Cant1992–2008 from equation (3) [Tanhua et al., 2007]. Precisions are very similar for both data sets. As CRMs were used in 2008, but data coverage was better in 1992, especially in WSDW and WSBW, results from forward and backward calculations will be shown.
 As stated above, stoichiometric ratios cannot be inferred from the regression coefficients [Friis et al., 2005]. Regression coefficients for the same variable vary significantly between two years dependent on the data set (Table 2) and the combination of other variables used for the fit. This is because the multiple input variables are not independent from each other [Friis et al., 2005], θ and S are both anticorrelated with O2 (Table 3) and also θ and S and S and p are not independent from each other.
Table 3. Correlation Coefficients for All Combinations of θ, S, O2, and p From the ANT X/4 and ANT XXIV/2 Data Sets
3.1.1. Error Estimation
 The standard deviation of the eMLR must be lower than the standard deviation of the MLRs because measurement errors, in fact, appear twice in the calculation and with different sign:
that is, the measurement errors for the ANT XXIV/2 data go into the prediction twice: once in combination with regression coefficients for 1992 and once in combination with the regression coefficients for 2008. These offsets are then subtracted from each other by conducting the eMLR and partly cancel out [Friis et al., 2005].
 However, there is still no mechanistic understanding of how errors propagate in an eMLR analysis [Friis et al., 2005; Tanhua et al., 2007]. Tanhua et al.  suggest a Monte Carlo simulation to estimate accuracy and precision of the eMLR method. They randomly perturbed an idealized data set in order to make a general statement about errors in eMLR analyses. We performed a Monte Carlo simulation similar to Tanhua et al.  where we randomly disturbed the ANT X/4 and ANT XXIV/2 data sets with a noise on the order of twice the precision of each measurement (δθ = 0.002°C, δS = 0.004, δO21992 = 1.4 μmol kg−1, δO22008 = 0.174 μmol kg−1, δCT1992 = 1.8 μmol kg−1, δCT2008 = 1.6 μmol kg−1). We used these perturbed data sets to calculate a perturbed Cant1992–2008 (Cant-err1992–2008) 10,000 times and measure precision of our eMLR-based Cant1992–2008 estimates as the standard deviation of all Cant-err1992–2008 values and accuracy as the difference between the mean Cant-err1992–2008 and unperturbed Cant1992–2008.
 Accuracy (Figure 3a) is rather uniformly distributed within the section. The bias due to uncertainties in the measurements ranges from −0.4 to 1.6 μmol kg−1 with the highest values at the surface. The range of the standard deviation of Cant-err1992–2008 (Figure 3b) is of similar magnitude (0.2–2.0 μmol kg−1). The lowest standard deviation (highest precision) is revealed between about 1000 and 2000 m depth and increases from here, both to the surface and to the bottom. The highest standard deviation (lowest precision) is found at the surface at the northernmost station, which is located in the ACC. It is known that MLRs are only valid in coherent ocean regions and cannot be extrapolated to the entire ocean. The slightly higher standard deviation hints that this is the boundary between the Weddell Gyre and the ACC.
 It is crucial to include phosphate (PO4) and silicate (SiO4) concentrations in the definition of alkalinity, especially in the Southern Ocean where nutrient concentrations are high. In the Weddell Sea, seasonal variations of nutrients are small. Because no nutrient data exist for cruise ANT XXIV/2, PO4 and SiO4 were estimated from ANT XIII/4 data, applying a linear curve fit as a function of potential temperature and salinity:
where R2 is the coefficient of determination, σ is the standard deviation, and n is the number of data points that were used for the fit. It has to be noted that the nutrients estimated based on temperature and salinity do not affect Cant1992–2008 as they are not used for the eMLR. They are used for calculating the change in other parameters of the carbonate system from Cant1992–2008 and AT2008. They do not affect changes in pH, [CO32−], Ω, and pCO2 at all, because they go into the calculation twice and with different sign and any possible error cancels out. Estimated nutrients are also used for the calculation of parameters of the carbonate system in 2008, but the error introduced by this fit is very small (a change in SiO4 concentration of 10 μmol kg−1 leads to a change of approximately 0.0004 in pH or 0.4 μatm in pCO2, a change in PO4 concentration of 0.05 μmol kg−1 leads to a change of approximately 0.0002 in pH or 0.2 μatm in pCO2).
3.2.1. Calculation of the Change in pH, [CO32−], ΩC, and ΩA
 The difference of the acidification parameters (pH, [CO32−], ΩC, ΩA) is calculated using the ANT XXIV/2 data and the Cant1992–2008 from the eMLR results. First, the parameters are calculated for 2008 with the ANT XXIV/2 data. Then, they are calculated for 1992 using CT1992 = CT2008 − Cant1992–2008, together with parameters from 2008 (T, S, AT) and PO4 and SiO4 concentrations as input data. By subtracting the 1992 carbonate parameters from the 2008 carbonate parameters, the change (Δ) in pH, [CO32−], [HCO3−], ΩC and ΩA is determined. The terms ΩC and ΩA denote the calcite and aragonite saturation states that are defined as
where K*sp is the stoichiometric solubility product [Zeebe and Wolf-Gladrow, 2001; Mucci, 1983]. The calcium concentration as a function of salinity for the calculation of ΩC and ΩA has been taken from Dickson et al.  and the solubility products of calcite and aragonite, K*sp C and K*sp A from Mucci . Alkalinity may change because of dissolution of carbonate sediments on timescales of several thousand years [Gehlen et al., 2007]. On a short timescale it is safe to assume that alkalinity distribution has not changed significantly. Seasonal variation of alkalinity may occur, but it is not well known in this region; therefore, we assume a constant total alkalinity distribution. The changes in the carbonate system can be seen as a summer estimate, as they are consistently calculated with alkalinity, temperature, and salinity from summer 2008. Changes in wintertime values may be different.
3.2.2. Calculation of the Change in Surface ΔpCO2
 The difference between oceanic surface layer pCO2 and atmospheric (atm) pCO2 is denoted by ΔpCO2, and its change is denoted by ΔΔpCO2. The latter is computed similar to ΔpH: seawater pCO21992 is calculated using CT1992 = CT2008 − Cant1992–2008 together with all other necessary parameters from 2008 (T, S, AT) and PO4 and SiO4 concentrations as input data. Seawater (sw) pCO22008 is computed with measured T, S, AT, CT, and PO4 and SiO4 data. Then, ΔpCO2 is determined by
 The change in ΔpCO2 is
 Negative values for ΔpCO2 indicate undersaturation. A negative ΔΔpCO2 indicates a trend toward stronger undersaturation, positive results would show a trend to saturation. Atmospheric CO2 mixing ratios at the South Pole were taken from Keeling et al.  and converted to partial pressure (Table 4) with the mean atmospheric sea level pressure from ANT XXIV/2 (south of 56°S) [König-Langlo, 2008]. The mean atmospheric pressure is 986.2 hPa which is very near to the annual mean of 985.8 hPa in 1992 which was measured at Neumayer station [König-Langlo and Herber, 1996].
Table 4. Monthly Atmospheric CO2 Mixing Ratios at the South Pole and Derived Partial Pressure Used for the Determination of ΔpCO2a
Cant1992–2008 calculated by eMLR with the variables θ, S, O2, and p is exhibited for the “backward case” (eMLR with 2008 data, Figure 4a) and the “forward case” (eMLR with 1992 data, Figure 4b). There are two regions with high Cant1992–2008: at the shelf, penetrating down to about 1000 m, and between 56.5°S and 60°S. The backward and forward cases differ only by the maximum surface Cant1992–2008 concentrations. In the surface layer Cant1992–2008 concentrations vary between 2 and 11 μmol kg−1 for the backward case and between 5 and 8 μmol kg−1 for the forward case. At the shelf, between 5 and 8 μmol kg−1Cant1992–2008 are dissolved in the ocean. The vertical distribution shows a distinct Cant1992–2008 minimum between 58°S and 63.5°S in the WDW around 500 m. This represents WDW that has been circulating within the Weddell Gyre and is characterized by lower temperature and salinity due to mixing with waters above and below. WDW that has more recently entered the Weddell Gyre in the east is found in the southern part of the gyre (64–69°S) at a depth of 300–750 m and carries Cant1992–2008 of about 2 μmol kg−1. The source water of this recent WDW in the ACC can be distinguished at the northern end of the section between about 300 and 1000 m.
 Only very little Cant1992–2008 was found to be stored in the deep and bottom layers of the Weddell Sea (Figures 4a and 4b). In the central Weddell Sea, Cant1992–2008 concentrations ranged between 1 and 2 μmol kg−1 between 1500 and 4000 m with values below 1 μmol kg−1 near the bottom. These values are so low that they are close to the uncertainty of the method. Low values in deep Weddell waters have been reported earlier by Poisson and Chen , Hoppema et al. [2001b], and Sabine et al. , which support our results. Despite the low level of Cant1992–2008, a well-structured distribution of it can be discerned even in the intermediate and deep water (Figures 4a and 4b) which complies with the current knowledge of the hydrography. This is an additional indication that the level of Cant1992–2008 and its distribution, though very low, is significant.
 It is somewhat surprising that the Cant1992–2008 concentration in the WSDW is higher than that in the bottom water. Traditionally, the main ventilation pathway of deep Weddell water has been thought to be the formation of WSBW and its subsequent mixing up into the deep water body. In that case, a Cant1992–2008 maximum would be expected in the WSBW. The fact that Cant1992–2008 in the WSDW is higher suggests that the WSDW is also ventilated by a mechanism not involving the WSBW. Hoppema et al. [2001a] have reported significant ventilation of the deep Weddell Gyre by a core of water with a CFC maximum at 3000–4000 m, entering the gyre from east along the continental slope. Further to the west in the Weddell Sea, traces of this water mass were found in most parts of the basin. Via this mechanism Cant1992–2008 is also transferred into the deep gyre. At about 69°S, a Cant1992–2008 maximum is found, hugging the continental slope at about 3000 m depth, the same location where the CFC maximum is usually observed [Klatt et al., 2002].
 At about 59°S, there is a core of WSBW which has been relatively recently ventilated in the western Weddell Sea [Klatt et al., 2002], which does have a Cant1992–2008 maximum of about 1–2 μmol kg−1 (Figure 4b), that is, higher than Cant1992–2008 of the WSBW found in the central and southern basins. The explanation for this difference is as follows. Significant Cant uptake takes place by recently formed WSBW. WSBW, which has been circulating in the bottom layer of the gyre for longer times already, as found in the central and southern basins, did not yet absorb that much Cant because of the lower pCO2 in the atmosphere at the time of formation.
 Between June 1992 and December 2007 the partial pressure of CO2 in the atmosphere at the South Pole rose by 26.8 μatm (Table 4) [Keeling et al., 2001]. Using typical AT, S, T, PO4, and SiO4 values from the uppermost layer of ANT XXIV/2 data to convert pCO2 to CT reveals that this atmospheric pCO2 increase should result in a CT increase of 10.3–10.9 μmol kg−1 assuming complete CO2 equilibration and no mixing. This is consistent with the Cant1992–2008 surface values, which are about 11 μmol kg−1. However, these high values occur only between 56.5°S and 60°S. This southern boundary exactly reflects the extent of the sea ice in the Weddell Sea, which reached as far as 60°S during ANT XXIV/2, curtailing equilibration with the atmosphere. High Cant1992–2008 in shelf water is transported to the Prime Meridian as part of the coastal current. The occurrence of polynyas along the coast allow for a better air-sea equilibration [Hoppema and Anderson, 2007], which leads to a higher Cant1992–2008 indeed (Figures 4a and 4b).
 On average, 0.06 μmol kg−1 of Cant1992–2008 have been accumulated per year in WSBW (below 4000 m) between 1992 and 2008. This agrees fairly well with Hoppema et al. [2001b] who found 0.02–0.04 μmol kg−1 a−1 for the period 1973–1998. As regards to the latter estimate it should be realized that during the 1970s and 1980s the Cant uptake has been less than that in more recent years because of the lower level of Cant in the atmosphere.
 The results of this study do not agree with Cant results by Lo Monaco et al. [2005a, 2005b]. As Lo Monaco et al. [2005b] focus on Cant at the Indian-Atlantic boundary of the Southern Ocean, AABW outflow of the Weddell Sea is taken into account. On the basis of a back-calculation technique [Körtzinger et al., 1998], they provided two estimates: one assuming that oxygen in surface waters is equilibrated with the atmosphere and the other one presuming that due to sea ice, surface waters are mostly undersaturated with respect to atmospheric oxygen. Although the former option yields Cant values of 8–10 μmol kg−1 in the deep water of the Weddell Sea, the latter results in Cant accumulation of 22–23 μmol kg−1, prompting questions about the role of the oxygen disequilibrium. High Cant values for WSDW were also found by the “Tracer Combining Oxygen, Inorganic Carbon and Total Alkalinity” (TrOCA) method [Lo Monaco et al., 2005a]. These values appear to be too high in comparison with the low Cant1992–2008 concentrations we found in the deep Weddell Sea. As our method gives a more direct estimate of Cant using data, opposite to back-calculating techniques, we think our results are more reliable. This holds true also for the high estimation of carbon inventories [Vázquez-Rodríguez et al., 2009] on the same transect as Lo Monaco et al. [2005a]. Rather high inventories might be forced by back-calculation techniques, which introduce errors by using a poorly known air-sea CO2 equilibrium [Matsumoto and Gruber, 2005].
 The amount of anthropogenic CO2 stored per kg of seawater is, in any case, less than that in the North Atlantic. Friis et al.  reported an increase in Cant of 1–20 μmol kg−1 below 100 m in the subpolar North Atlantic between 1981 and 1997. Even below 2000 m they estimated excess CO2 of up to 15 μmol kg−1. This is anticipated as extensive deep and bottom water formation processes are known to occur in the North Atlantic in ice-free regions. About 17 Sv of deep water are formed in the North Atlantic, a similar amount as in the entire Southern Ocean (14 Sv) of which only a part is ventilated in the Weddell Sea [Orsi et al., 2002]. Additionally, the large extent of sea ice in the Weddell Sea hampers air-sea CO2 exchange, surface water CO2 concentrations get diluted by mixing with intermediate and deep waters with little anthropogenic CO2, and the residence time for nascent bottom water is too short for atmospheric CO2 to penetrate into the water as extensively as in the North Atlantic [Poisson and Chen, 1987; Hoppema et al., 2001b; Lee et al., 2003].
4.2. Acidification of the Weddell Sea
 Acidification is expressed by oceanic pH decrease. Alternatively, this is done by the decreasing CO32− concentration and saturation states of calcite (ΩC) and aragonite (ΩA). The change in pH (ΔpH) between 1992 and 2008 based on the eMLR Cant1992–2008 calculation is shown in Figure 5a. The same patterns are seen as in Cant1992–2008, reflecting water mass structures and the remote source from the east. In the core of the WDW and in WSBW, minimal pH change with a decrease of only 0.002 units is observed. In WSDW, a decrease of up to 0.005 pH units is found. Near the shelf, ΔpH exceeds −0.02 units, even down to 500 m, and in the uppermost layer north of 60°S a drop of 0.03 pH units occurs.
 The theoretical Weddell Sea surface layer pH change due to atmospheric pCO2 increase is about −0.03 in the period considered, again assuming complete CO2 equilibration and no mixing. Values in this range are only found north of 60°S. The theoretical values were not reached in areas where sea ice inhibits gas exchange.
4.2.1. Change in Carbonate Ion Concentration
 If the CO2 concentration changes, the carbonate ion concentration changes as well as a result of shifts in the equilibrium (see equation (1)). There is a decrease in [CO32−] between 3 and 4.5 μmol kg−1 at the shelf, up to 7 μmol kg−1 at the surface north of 60°S and below 1 μmol kg−1 in the deep sea (Figure 5b). The average decrease of carbonate ions is 2.1 μmol kg−1, representing a reduction of 2.5% compared to 1992 in the entire water column.
 In the Southern Ocean, surface layer [CO32−] concentrations are naturally low compared with global concentrations. Current average [CO32−] varies between 105 μmol kg−1 in the Southern Ocean and 240 μmol kg−1 in tropical regions [Orr et al., 2005]. The ANT XXIV/2 data from 2008, however, show even lower values with an average of 95.9 μmol kg−1 in the upper 100 m (range: 80.6–126.0 μmol kg−1). On top of that, Southern Ocean [CO32−] decreases seasonally by about 15 μmol kg−1 in winter as a result of lower temperatures [Orr et al., 2005]. We find a decrease at the surface of 3.9 μmol kg−1 or 4.0% due to anthropogenic CO2 invasion. Seasonal variations may occur in addition to our estimate. The annual rate of [CO32−] reduction in the surface layer is 0.25 μmol kg−1. By comparison, according to Orr et al.  modern surface [CO32−] in the Southern Ocean has decreased by 18 μmol kg−1 from preindustrial levels, which is converted to 0.07 μmol kg−1 between 1750 and 1994. As expected, because of nonlinear CO2 increase during this period, this demonstrates that the rate of [CO32−] decline has significantly accelerated.
 The annual rate of decline that we calculated is at the lower end of Orr et al.'s  predictions for the next decades. In their prediction with the IS92a scenario (“business-as-usual,” atmospheric CO2 reaches 550 ppm in 2050), the Southern Ocean surface [CO32−] becomes about 70 μmol kg−1 in 2050, i.e., a mean reduction of approximately 0.5 μmol kg−1 per year between 2000 and 2050. This is twice the annual reduction observed in our study. If we extrapolate the annual rate of 0.25 μmol kg−1 a−1 to the year 2050 linearly, a carbonate concentration of 85.4 μmol kg−1 would be reached. This is higher than all predictions in all scenarios used by Orr et al. , even the moderate scenario. The linear extrapolation may be problematic because it is not constrained, but it is worthwhile noting that also in the paper of Orr et al.  all predictions (except for scenario A1FI) between 2000 and 2050 are approximately linear. Also, Zeebe and Wolf-Gladrow  (their Figure 1.6.27) demonstrate that a decline in [CO32−] based on the IS92a scenario is almost linear. Therefore, the calculations made above are not significantly biased by the assumption of a linear reduction of [CO32−] based on IS92a. It should be noted that Raupach et al.  and Canadell et al. (http://www.globalcarbonproject.org/carbonbudget/index.htm, 2007) report that CO2 emissions are already higher than the worst-case scenario A1FI. A steepening of the [CO32−] reduction is hence likely to happen around the year 2030 [Orr et al., 2005], but the estimates of the [CO32−] decrease of IS92a and A1FI do not start to deviate significantly from each other before 2050.
 The reduction of [CO32−] in Weddell Sea surface waters is lower than that in the averaged Southern Ocean as estimated from global biogeochemical circulation models [Orr et al., 2005]. The Weddell Sea seems to react differently to increasing atmospheric CO2 than the mean Southern Ocean surface waters. This discrepancy may be explained by averaging over the entire Southern Ocean, by insufficient spatial resolution of global models in the Southern Ocean, by underestimating the large effect of sea-ice in the Weddell Sea or by neglecting biological feedback mechanisms in the models. We underline the importance of regional studies in the Southern Ocean to determine spatial deviations of the mean response of the Southern Ocean to increasing atmospheric CO2.
4.2.2. Changes in the Saturation States of Calcite and Aragonite
 The spatial distribution of changes in ΩC (Figure 6a) and ΩA (Figure 6b) corresponds closely to that of the parameters discussed in sections 4.1 and 4.2.1. The decrease in ΩA amounts to about two thirds of the decrease in ΩC. Note that ΩC dropped by up to 0.18 at the surface north of 60°S, 0.1 at the shelf and 0.01 in WSDW. Near the seafloor and in the WDW the reduction was weaker, between 0 and 0.01. Likewise, ΩA was reduced by 0.09 at the surface north of 60°S and 0.06 at the shelf, whereas reduction was below 0.01 in WSDW and below 0.005 in WDW and WSBW. On average, ΔΩC was −0.05 and ΔΩA −0.03.
 The future projection of ΩA was estimated by an extrapolation, which was conducted as follows: ANT XXIV/2 data were used as starting point and to each data point the product of the annual ΔΩA (as calculated at this latitude, longitude, and depth) and the time difference in years to 2008 was added. This leads to
 The number of data points of ΩA < 1 in the upper 20 m divided by the total number of data points in the upper 20 m gives the percentage of surface area undersaturated with respect to aragonite. This extrapolation assumes a constant linear decrease of [CO32−] (discussed above), and constant temperature and salinity distributions. This is a qualitative estimate about the change of ΩA as a consequence of the accumulation of Cant; seasonal variations will occur on top of that. It is therefore only a rough estimate, but it shows that average Weddell Sea surface waters will not be completely undersaturated (90%) with aragonite at the end of the 21st century. Although a few single spots of permanent aragonite undersaturation in the surface layer may occur earlier, only a small part of the surface will be constantly undersaturated by 2100 (Figure 7). Temporary undersaturation in winter may occur earlier [McNeil and Matear, 2008].
 Nowadays ΩA is largest at the surface and decreases with depth (Figure 8a). An interesting feature is that the reduction of ΩA is strongest at the surface. Hence, ΩA could fall below 1 in the surface layer whereas the WDW below is still saturated with aragonite (Figure 8b). This is different from other regions of the world oceans.
4.3. The CO2 Saturation State of the Weddell Sea
 ΔpCO2 was calculated for 2008 from CT and alkalinity, while for 1992 it necessarily had to be obtained from the hydrographic parameters for 2008 and CT1992 = CT2008 − Cant1992–2008. Thus, changes in ΔpCO2, i.e., ΔΔpCO2, are the result only of Cant1992–2008 and changes in atmospheric pCO2. In 1992, the Weddell Gyre was undersaturated with CO2 in almost all areas, except in the 3°W transect south of 67.5°S (Figure 9a). It should be pointed out again that ΔpCO21992 shows how the situation would have been in 1992 if CT changed as calculated based on the eMLR analysis. ΔpCO22008 varied between −103 and +9 μatm, with a mean value of −42.5 μatm (Figure 9b). Thus, there was undersaturation of CO2 in almost the entire Weddell Sea, also reported for the summertime Weddell Sea by Hoppema et al. . Saturation was observed in the western transect near the shelf. As discussed in section 3.1, despite the generally robust eMLR estimates, errors tend to be largest at the surface. Furthermore, variations in pCO2 can be high indeed at the surface and even higher at the shelf. Both, undersaturation and local supersaturation have been reported previously for the Weddell Gyre [Hoppema et al., 1995, 2000; Bellerby et al., 2004]. Gibson and Trull  mentioned whole year undersaturation for another coastal region, Prydz Bay in East Antarctica.
 If no change in the oceanic uptake rate for CO2 would occur, ΔΔpCO21992–2008 would be zero. Note that ΔΔpCO2 shows mostly negative values, i.e., a trend to stronger undersaturation from 1992 to 2008 (Figure 9c). Highest negative values are found on the Prime Meridian near the shelf. Some positive values are observed in the central Weddell Gyre. However, these are isolated data points which confirm the pCO2 variability. The mean ΔΔpCO2 is −5.4 μatm.
 The overall trend to negative values of ΔΔpCO2 indicates that between 1992 and 2008 oceanic pCO2 did not rise as fast as atmospheric pCO2. This is confirmed by the mean surface layer (100 m) pCO2 rise of 16.2 μatm as calculated using eMLR in comparison to the 26.8 μatm increase of atmospheric pCO2. Hence, the Weddell Sea still has a large uptake capacity and absorbs CO2 without a proportional increase of seawater pCO2. This leads to a decreasing buffer capacity and increasing Revelle factor. In the long term, the reduced buffer capacity will cause surface pCO2 to rise proportionally faster than CT, thereby diminishing ΔpCO2, the driving force of CO2 exchange between ocean and atmosphere [Völker et al., 2002; Sabine et al., 2004; Thomas et al., 2007]. The reduction of buffer capacity due to the CO2 uptake is, though slightly, noticeable: the Revelle factor in surface waters (upper 100 m) has increased from an average of 15.1 in 1992 to 15.5 in 2008.
 Recently, the possibility of a declining capacity of the Southern Ocean as CO2 sink has been discussed in the literature [e.g., Le Quéré et al., 2007; Zickfeld et al., 2008; Law et al., 2008; Le Quéré et al., 2008; Canadell et al., 2007]. Le Quéré et al.  argue that due to human activities Southern Ocean winds have increased, leading to intensified upwelling, thereby bringing water with high natural CO2 concentrations to the surface; consequently enhanced outgasing would reduce the oceanic CO2 sink capacity. If this mechanism were active, an increase of ΔpCO2 (from negative values toward zero) would be expected. Our results point to opposite changes in ΔpCO2.
 From a model run for the time period 1981–2004, Le Quéré et al.  conclude that the Southern Ocean sink was not further increasing as expected, but rather remained constant. We conducted the same eMLR analysis with the model output as described by Le Quéré et al.  for May/June 1992 and December 2007/January 2008. The model was subsampled for the ANT XXIV/2 stations. This resulted in a similar increase of undersaturation as found with the data based eMLR (see above), i.e., strengthening of the oceanic sink in this region between 1992 and 2008 (Figure 9d). The mean ΔΔpCO2 as obtained by data from Le Quéré et al.  appeared to be −3.8 μatm. Although it does not reproduce the strong increase of the sink near the shelf as does our data, the overall findings match very well. This emphasizes that regionality is high in the Southern Ocean. Although the Southern Ocean CO2 sink as a whole could be decreasing, the Weddell Sea could be an important and increasing CO2 sink. The apparent contrast between Le Quéré's model results for 1980–2004 (entire Southern Ocean) and 1992–2008 (Weddell Sea) also indicates that because of an accelerating atmospheric CO2 increase in recent years, the Southern Ocean oceanic sink is reinforced.
 It was demonstrated that eMLR is an appropriate method to determine the amount of anthropogenic carbon accumulated at a decadal timescale. Between 1992 and 2008, a significant amount of Cant accumulated in Antarctic Surface Water (2–8 μmol kg−1 per decade) and shelf water (3–5 μmol kg−1 per decade). Low or insignificant amounts of Cant1992–2008 were found in Warm Deep Water and Weddell Sea Bottom Water and low concentrations in Weddell Sea Deep Water (1 μmol kg−1 per decade). Cant1992–2008 found in AASW closely follows the increase of atmospheric pCO2, although there is spatial variation as well.
 The storage of Cant1992–2008 leads to ocean acidification. A decrease of pH was observed in AASW (−0.003 to −0.026 units per decade) and shelf water (−0.007 to −0.013 units per decade). This was accompanied by a decrease in [CO32−] of −0.7 to −4.8 μmol kg−1 per decade. The reduction of carbonate ions is, however, less than that estimated for the mean Southern Ocean surface waters by studies based on global biogeochemical circulation models [Orr et al., 2005]. Calcite and aragonite saturation states, which vary with [CO32−], also decreased significantly near the surface (by 0.016 to 0.116 units per decade and 0.016 to 0.071 units per decade, respectively). Tentative results of our analysis suggest that complete undersaturation of surface waters in the Weddell Sea will be reached after the 21st century.
 The Weddell Sea surface water is mostly undersaturated with respect to CO2. A weakening of the CO2 sink of the Southern Ocean as currently discussed in the literature could not be confirmed. In contrast, undersaturation was more pronounced in 2008 than in 1992. Because of the uptake of Cant1992–2008, the Revelle factor increased by 0.2 per decade which will only lead to a weakening of the sink in the far future.
 The support of captain and crew of FS Polarstern is greatly appreciated. We are indebted to Craig Neill for strong technical support and performing parts of the CT and AT analyses and to Kelly Brown for the oxygen measurements. We are grateful to Volker Strass for providing hydrographic data. This research was supported by two Norwegian Research Council Projects: Bipolar Atlantic Thermohaline Circulation (BIAC, IPY cluster 23) and Southern Ocean Biogeochemistry and Education Project 180328/S50. Additional support came from the EU project CARBOOCEAN (511176; GOCE).