This study evaluates the anthropogenic changes of CO2 (ΔCant) and δ13C (Δδ13Cant) in the Nordic seas, the northern limb of the Atlantic Meridional Overturning Circulation, that took place between 1981 and 2002/2003. The changes have been determined by comparing data obtained during the Transient Tracers in the Ocean, North Atlantic Study (TTO-NAS) with data obtained during the Nordic seas surveys of R/V Knorr in 2002 and R/V G.O. Sars in 2003 using an extended multilinear regression approach. The estimated Δδ13Cant and ΔCant and their relationship to each other and to water mass distribution suggest that the Polar Water entering the Nordic seas from the north is undersaturated with respect to the present atmospheric anthropogenic CO2 levels and promotes a local uptake of Cant within the Nordic seas. In contrast, the Atlantic Water entering from the south appears equilibrated. It carries with it anthropogenic carbon which will be sequestered at depth as the water overturns. This preequilibration leaves no room for further uptake of Cant in the parts of the Nordic seas dominated by Atlantic Water. The upper ocean pCO2 in these regions appears to have increased at a greater rate than the atmospheric pCO2 over the last 2 decades; this is reconcilable with a large lateral advective supply of Cant.
 The increase of the atmospheric CO2 concentration caused by fossil fuel combustion, cement production, and land use change is significantly dampened by oceanic CO2 uptake. Current annual net oceanic CO2 uptake corresponds to roughly one quarter of the fossil fuel and cement manufacturing emissions [Prentice et al., 2001], and the accumulated oceanic uptake since the start of the industrial revolution corresponds to approximately 50% of the emitted fossil CO2 [Sabine et al., 2004]. The size of the oceanic inventory of anthropogenic carbon (Cant) is, however, hard to estimate owing to the absence of oceanic carbon observations from the preindustrial era and estimates such as the quoted Sabine et al.  have to rely on indirect approaches [Gruber et al., 1996] with inherent methodological uncertainties. Our understanding of where and when Cant is taken up by the oceans, how it is transported within them, and where it is stored is therefore uncertain. This situation is now being alleviated by continued observations which enable us to observe and quantify carbon concentration changes in the interior ocean associated with Cant input.
 Decadal changes have been quantified in the Pacific Ocean [Peng et al., 2003], the Indian Ocean [Peng et al., 1998; Sabine et al., 1999], and in the Southern Ocean [McNeil et al., 2001a] by comparing recent carbon data with the data collected during GEOSECS and other expeditions that were carried out during the 1960s and 1970s. In the present paper observations obtained during the Transient Tracers in the Ocean, North Atlantic Study (TTO-NAS) of 1981 are compared with data obtained during two recent surveys of the Greenland, Iceland, and Norwegian seas. This allows for the quantification and increased understanding of the Cant input to the northernmost parts of the Atlantic Ocean over the last 2 decades.
 The Greenland, Iceland, and Norwegian seas, collectively known as the Nordic seas, are considered important for the natural exchange of carbon between the ocean and the atmosphere, representing the preindustrial carbon fluxes. Annually, approximately 6 Sv (Sverdrup, 106 m3 s−1) of warm and salty Atlantic Water (AW) from farther south enter the region [Hansen and Østerhus, 2000]. This water is cooled and transformed to subsurface waters within the Nordic seas themselves, and also in the Arctic Ocean and associated shelf areas. The subsurface waters exit the Nordic seas over the Greenland-Scotland ridge and contribute to a total northern North Atlantic overturning circulation of 15 Sv [Álvarez et al., 2002]. The cooling of the northward flowing AW drives a flux of CO2 from the atmosphere into the ocean so that the waters that overturn in the Nordic seas and contribute to the global oceanic deep water pool are loaded with carbon. As the waters eventually upwell this carbon is released back to the atmosphere, along with carbon that was added to the deep water from decaying organic material.
 The Nordic seas are also considered an important region for oceanic uptake of Cant. The air-sea flux of anthropogenic CO2 is not driven by biology and meridional water transport as the preindustrial air-sea CO2 flux, but by the increase of the atmospheric CO2 concentration which has perturbed the natural exchanges between the ocean and the atmosphere. Given approximately 1 year the surface layer of the world oceans will equilibrate with the atmosphere [Broecker and Peng, 1974] and further uptake will rely on exposure of older water to the atmosphere. The largest air-to-sea transfer of Cant is thus considered to take place in regions with extensive communication between surface and deep waters such as the tropical and Southern Ocean upwelling regions and the North Atlantic convection areas. A large air-to-sea flux of anthropogenic CO2 and an efficient surface-to-deep ocean carbon transport is expected to occur in the Nordic seas. Indeed, with respect to the latter, Cant estimates do reveal deep penetration in the Nordic seas spreading into the deep Atlantic [Gruber, 1998; Sabine et al., 2004]. As regards the air-sea flux of anthropogenic carbon on the other hand, the situation is somewhat unclear. Three-dimensional ocean carbon cycle models do tend to show a large air-sea Cant flux in the northern North Atlantic and Nordic seas [Orr et al., 2001; Wetzel et al., 2005], but these results have, however, recently been challenged by estimates of oceanic Cant transport which seem to agree that the largest uptake of Cant takes place in the tropical-to-subtropical Atlantic, so that the waters that enter the subpolar regions are already equilibrated with the atmosphere's anthropogenic CO2 concentration, leaving little room for further uptake [Álvarez et al., 2003; Macdonald et al., 2003; Rosón et al., 2003]. In fact, the Rosón et al.  estimate indicates that the subtropical-to-subpolar Atlantic is a source of anthropogenic CO2 to the atmosphere; that is, the uptake of CO2 from the atmosphere has been reduced since preindustrial times. This result is in agreement with output from simple model calculations [Wallace, 2001; Anderson and Olsen, 2002; Bellerby et al., 2005]. Observations from the North Atlantic do indeed seem to verify that the surface ocean pCO2 in this area has increased at a greater rate than the atmospheric pCO2 over the last 20 years [Lefévre et al., 2004], implying a reduced uptake of atmospheric CO2. However, the extent to which this increase is a consequence of anthropogenic CO2 or of biological and/or temperature effects was not resolved by Lefévre et al. . Surface ocean pCO2 appears to have increased in the Barents Sea as well, as presented by Omar et al. . They applied an approach that accounted for changes in biology and hydrography, so their estimated annual mean increase of 1.3 μatm yr−1 between 1967 and 2001 is solely due to anthropogenic carbon. This increase is slightly lower than the atmospheric increase of 1.4 μatm yr−1 in the same time interval. Since no gradients could be observed within the Barents Sea, the authors suggested that the anthropogenic carbon had been advected into the region from the Nordic seas.
 This paper focuses on the region between the areas addressed by Lefévre et al.  and Omar et al. : the Nordic seas. The area was surveyed during the summers of 2002 and 2003 using R/V Knorr and R/V G.O. Sars. Data describing the inorganic carbon system were acquired during both cruises as well as data for temperature, salinity, and nutrient concentrations. These data are compared with the data acquired in the same area during the TTO-NAS of the summer of 1981, aiming to determine the changes of the Cant concentrations over the 2 decades that separate these surveys. To this end an extended version of the multilinear regression (MLR) approach of Wallace  is employed, the eMLR approach as first introduced by Sonnerup et al.  and later formalized by Friis et al. . The eMLR method is also employed on data of the 13C-to-12C ratio, δ13C, acquired during the TTO-NAS and the Sars cruises so that the magnitude of the oceanic 13C Suess effect in the region and its relationship with the anthropogenic CO2 changes can be estimated. The 13C Suess effect is a consequence of the fact that fossil fuel CO2 is relatively enriched with the light carbon isotope 12C so that emissions lead to a decrease of the 13C-to-12C ratio in the atmosphere. Isotopic exchange across the air-sea interface transmits this perturbation to the ocean, giving rise to the oceanic 13C Suess effect. Information on the magnitude of this effect can be used to quantify oceanic uptake of Cant [Quay et al., 1992; Tans et al., 1993; Heimann and Maier-Reimer, 1996; Quay et al., 2003], and the ratio of oceanic Cant and 13C changes provides information on the recent atmospheric exposure histories of water masses [McNeil et al., 2001b; Körtzinger et al., 2003].
2. Data and Methods
 The positions of the data used in the present work are presented in Figure 1. Data selection has been limited to offshore of the 250-m isobath to avoid coastal influence, and at/or north of the Greenland-Scotland ridge.
2.1. TTO-NAS Data
 The TTO-NAS was carried out from April to October of 1981. The TTO-NAS survey consisted of 7 legs and the Nordic seas were surveyed during leg 5, in late July and early August. The data used here were acquired at stations 142 to 159. The data were obtained from the Carbon Dioxide Information Analysis Center (CDIAC), Oak Ridge, Tennessee. The total concentration of inorganic carbon (Ct) in water was determined at sea as well as on shore after the cruise. The at-sea determinations were carried out using potentiometric titration while the onshore determinations were carried out in the late C. D. Keeling's laboratory at Scripps Institution of Oceanography using the gas extraction manometric technique described by Weiss et al. . Subsequent analyses revealed that the Ct values determined at sea were higher than the values determined on shore. This motivated the release of a revised TTO Ct dataset which was calculated from the TTO total alkalinity (TA) and pCO2 data using the carbon system constants of Merbach et al.  [Brewer et al., 1986]. Recently Tanhua and Wallace  have reanalyzed the TTO carbon chemistry data from legs 2, 3, 4, and 7. They find that the TTO TA data are biased high with respect to modern data tied to Certified Reference Material (CRM) and recommend that they should be adjusted down by 3.6 μmol kg−1. They moreover recommend that the TTO Ct values should be recalculated using the adjusted TA data and currently accepted constants and then adjusted upward by 2.4 μmol kg−1 on the basis of comparison with the manometric Ct values obtained during TTO. These recommendations were followed in the present work. Leg 5 Ct values thus computed were on average 0.26 μmol kg−1 higher than the manometric data; this compares favorably with the offset between the manometric and revised Ct data of 2.5 μmol kg−1, the latter being lower.
 The TTO-NAS data of the 13C-to-12C ratio of dissolved inorganic carbon, δ13C, were acquired from CDIAC. The data were measured by the Carbon Dioxide Research Group at Scripps Institution of Oceanography using procedures described by Gruber et al. . All data that were flagged as suspicious in the original data set were discarded, as well as data points where the replicate analyses differed by more than three times the estimated combined error from sampling and analysis, 0.12‰ (3X0.04‰), following Gruber et al. .
2.2. R/V Knorr Data
 The R/V Knorr cruise was carried out in June 2002 and covered most regions of the Nordic seas (Figure 1). The temperature, salinity, and nutrients data acquired during this cruise will be described by R. G. J. Bellerby et al. (manuscript in preparation, 2005). The data for Ct that are employed here were determined by gas extraction of acidified water samples followed by coulometric titration [Johnson et al., 1985, 1987]. The accuracy was set by running CRM supplied by Andrew Dickson of Scripps Institution of Oceanography, and the precision has been determined to better than ±2 μmol kg−1 by comparison of deep samples acquired in the western Greenland basin where the Ct is regarded as being homogenous (R. G. J. Bellerby et al., manuscript in preparation, 2005).
2.3. R/V G.O.Sars Data
 The R/V G.O. Sars cruise was carried out from September to October of 2003. The cruise covered the Barents Sea Opening, the Greenland Sea, the Faeroe-Shetland Gap, the Faeroe Bank Channel, and the Svinøy Section that extends NW off the coast of Norway from 62°N. Data were also acquired at a few stations in the Iceland Sea (Figure 1). Temperature and salinity data were obtained using a SBE 911plus CTD profiler. The initial accuracy of the conductivity and temperature sensors were 0.0003 S/m and 0.002°C, respectively. The CTD profiler carried two temperature and two conductivity sensors and the standard deviation of each pair of sensors averaged over all stations was 0.0023°C for temperature and 0.0012 for salinity. The CTD salinity data were corrected to bottle samples by a polynomial fit. The standard deviation of the difference between the CTD and the bottle values was 0.0047 with a correlation coefficient of 0.9998.
 The nitrate and silicate data were analyzed on the ship using a Chemlab 3-channel autoanalyzer following methods of Grasshoff . Phosphate was analyzed using a modified version of the Murphy and Riley  method. The accuracy and precision of the nutrient analyses have been assessed by running nutrient test samples supplied by the QUASIMEME program [Wells and Cofino, 1997] and laboratory reference material [Aminot and Kérouel, 1998]. The laboratory reference material was run at each station and indicates precisions (1 σ) of ±0.13 μmol kg−1 for nitrate, ±0.02 μmol kg−1 for phosphate, and ±0.27 μmol kg−1 for silicate over the duration of the cruise. Immediately after the cruise the laboratory reference material was analyzed together with test samples from the QUASIMEME program. The test sample results were in very good agreement with the QUASIMEME assigned values.
 Ct was determined using the same method as during the Knorr 2002 cruise. Accuracy was set by running CRM as for the Knorr cruise. The combined sample handling and measurement error of these data were determined to be 0.4 μmol kg−1 by analyzing samples drawn from 10 different Niskin bottles closed at the same depth (1 σ).
 The δ13C samples acquired on this cruise were collected in 100-mL bottles and poisoned with HgCl2 and sealed with rubber stoppers immediately after collection. The samples were stored cold and in the dark. Each sample was analyzed in triplicate using a Finnigan GasBench coupled to a Mat 252 mass spectrometer within 3 months of the cruise. On the basis of 16 samples taken below 2000 m in the Greenland Sea with very similar physical and chemical water mass characteristics, the precision has been determined to be at least ±0.07‰ (1 σ) for these results, just above the long-term precision of the machine based on standard replicates (±0.05‰).
2.4. The eMLR Calculations
 The extended multilinear regression (eMLR) approach builds on the multilinear regression (MLR) approach for determining Cant changes introduced by Wallace  and subsequently employed on several occasions [Sabine et al., 1999; McNeil et al., 2001a; Peng et al., 2003]. The MLR method has also been used to determine the oceanic 13C Suess effect in the Southern Ocean [McNeil et al., 2001b]. The underlying idea of this method is that whereas the carbon system in itself is perturbed as Cant enters the ocean, the natural variations can be accounted for using a number of predictive parameters such as salinity, temperature, apparent oxygen utilization (AOU), nutrients, TA, etc. The approach takes advantage of the fact that changes of the Cant concentration will affect the accuracy of empirically derived functions that estimate Ct, provided the predictive parameters are only influenced by natural processes. The MLR method applies an equation, derived for one region with data obtained at one time, to predict Ct concentrations using data obtained in the same region but at a different time. Any systematic bias revealed through comparing measured with predicted Ct values is ascribed to changes in the concentration of Cant. The eMLR technique was first introduced by Sonnerup et al.  for δ13C data. Later on Friis et al.  further formalized and reviewed the eMLR technique for Ct data. Rather than evaluating the offset between measured and predicted Ct values, the eMLR technique employs two regression equations derived from two data sets collected at different times. The two equations are applied on one set of predictive variables and differences in the predicted Ct values reveal the Cant change. This method minimizes the measurement error associated with the independent variables since they enter the prediction twice, once when they are employed in the modern relationship and once when they are employed in the historical relationship [Friis et al., 2005].
 The predictive equations for Ct and δ13C were derived using salinity, potential temperature (θ), nitrate, and silicate as independent variables. These parameters were found to be sufficient and necessary. Adding the parameters AOU and TA did not improve the fit. Nitrate data were preferred to phosphate as comparison of the phosphate data collected at the Knorr and Sars cruises revealed significant differences in deep waters, ∼0.2 μmol kg−1, indicating issues with the phosphate analyses on one of the cruises. There were no significant deep water differences between these cruises with respect to the other variables used in this study. A cutoff value of 0.5 μmol kg−1 was applied to the nitrate data to avoid the potential impact of carbon overconsumption at very low nutrient levels [Falck and Anderson, 2005]. The regression analyses were carried out using the software package R [R Development Core Team, 2004]. The coefficients and statistics of each relationship are listed in Table 1. The residuals are plotted as a function of fitted values in Figures 2a and 2b, and versus depth in Figures 3a and 3b. In general the relationships predict Ct and δ13C to within ±6 μmol kg−1 and ±0.07‰ as evaluated from the root mean square (rms) values (Table 1). The accuracy of the Ct relationships appears uniform over the whole range of Ct values as evaluated from Figure 2. The predictions are biased slightly high deeper than 2000 m, around 1 to 2 μmol kg−1, with the TTO results being higher (Figure 3). Precision of the Ct predictions appears stable with depth to ±5 μmol kg−1; this is taken as a measure of the uncertainty of the Cant estimates, except in the upper 100 m where precision decreases to ±10 μmol kg−1. The precision of the δ13C predictions appears to be within an acceptable ±0.1‰ at all depths, but it was not possible to evaluate deep TTO precisions due to the limited number of samples.
Table 1. Regression Coefficients and Statistics for the Predictive Equations Which Were of the Form Ct = a × S + b × θ +c × NO3 + d × Si + k and δ13C = a × S + b × θ +c × NO3 + d × Si + k
 The anthropogenic changes of Cant (ΔCant) and δ13C (Δδ13Cant) were estimated as the difference between Knorr/Sars equations and the TTO equations when both were applied to the TTO data. The equations were employed on the TTO data because these data define the region where all of the regression equations are known to be valid. Moreover, ΔCant was only determined for samples within the salinity, θ, nitrate, and silicate ranges for which both the Knorr/Sars and the TTO relationships were valid, i.e., the common validity range. The same was the case for the Δδ13Cant estimates. These ranges are provided in Table 2.
Table 2. Common Application Ranges of the Predictive Equations for Ct and for δ13C
 A schematic of the Nordic seas upper ocean circulation is presented in Figure 4. It is dominated by the northward flow of warm and saline AW in the east and the southward flow of fresher and colder Polar Water (PW) from the Arctic Ocean in the west. These water types meet and mix in the area between and form Arctic Surface Water, and as this water is cooled during winter several types of Arctic Intermediate Water (ArIW) are formed. The deep waters are not only formed locally such as Greenland Sea Deep Water (GSDW), but are also advected into the region from the Arctic Ocean. Canadian Basin and Eurasian Basin Deep Waters enter the Nordic seas through the Fram Strait, meet and mix with GSDW and moves south into the deep Lofoten and Norwegian basins as Norwegian Sea Deep Water (NSDW).
 The distributions of temperature and salinity along a south-to-north section based on the TTO data are displayed in Figure 5. The positions of the stations have been specified in Figure 4. The upper 600 m of the southernmost station (Sta. 142) in the Faeroe Bank Channel are dominated by the relatively warm and salty (S > 35) AW entering the region, and below this the cold overflow water exiting to the North Atlantic prevails. Station 143 in the southern Norwegian Basin is to the west of the Atlantic inflow which turns sharply east after crossing over the Iceland-Faeroe Ridge, thus salinities do not exceed 35 at this station. The upper part of this station has salinities in excess of 34.95 and captures the Recirculated Faeroe Current. This current carries AW that has recirculated and been mixed with ArIW in the Norwegian Basin. The lens of water with S < 34.9 between 400 and 600 meter is Norwegian Sea Arctic Intermediate Water (NSArIW), and NSDW dominates below this. The AW is reencountered at the next stations, 144 and 145, and it reaches its deepest extent at the latter of these, in the Lofoten Basin. NSArIW is present at both of these stations as well, as a salinity minimum at 700 and 1000 m, respectively, and below this NSDW prevails. The sharp gradient in salinity over the Mohns Ridge at stations 146 and 147 is the Arctic Front, north of which the increased influence of PW is manifest as a substantial freshening above 1000 m. The Greenland Sea appears otherwise quite homogenous with respect to both temperature and salinity, but then again we stress that the resolutions applied for salinity and temperature data are inadequate to resolve the finescale structures. As the section moves northward through station 149 over the Boreas Basin and then turns east onto the West Spitsbergen shelf, the increasing Atlantic influence at intermediate depths becomes evident with salinities ranging from 34.9 up to 35. Surface waters, though, are quite fresh throughout. In the Boreas Basin, this is due to PW and particularly when close to the shelf it is due to the West Spitsbergen Coastal Current, a continuation of the East Spitsbergen Current that carries water from the northern Barents Sea.
4.1. Anthropogenic CO2 Increase
 The estimated change in anthropogenic CO2 concentration along the TTO section between 1981 and 2002/03 is displayed in Figure 6. The largest variations of ΔCant are evident in the upper 100 m. This approximately represents the seasonal mixed layer and is the zone with the largest methodological uncertainty (±10 μmol kg−1). These estimates should be interpreted with care and have been shaded gray in the figure. Below this ΔCant values range from less than 4 μmol kg−1 to more than 20 μmol kg−1 with systematic spatial variations that are fairly coherent with the water mass distribution. The largest values are associated with the Atlantic domain of the Nordic seas. In the core of the Atlantic inflow, above 600 m in the FBC, ΔCant lies between 18 and 20 μmol kg−1. This corresponds to annual changes of around 0.9 μmol kg−1 yr−1. The ΔCant decreases slightly moving into the Norwegian Sea but values are clearly higher here than in the rest of the section; in the AW at stations 144 and 145 as well as in the recirculated AW of station 143 values are between 14 and 18 μmol kg−1, corresponding to annual changes of 0.7 to 0.8 μmol kg−1 yr−1. The values decrease with depth and in the NSDW ΔCant is less than 6 μmol kg−1. The isolines in the Norwegian Sea slope northward and are deepest in the Lofoten Basin where also the Atlantic layer is deepest. The overflow water in the Faeroe Bank Channel has ΔCant values in the same range as the upper NSDW, between 8 and 10 μmol kg−1. Similar to salinity, ΔCant decreases sharply across the Arctic Front over the Mohns Ridge and the surface to deep ocean ΔCant gradient is smallest in the Greenland Sea as a result of deep convective mixing in this region. Approaching the West Spitsbergen Shelf where the influence of AW increase, the ΔCant lies between 12 and 14 μmol kg−1.
4.2. Anthropogenic δ13C Changes
 The estimated changes of δ13C attributable to the oceanic 13C Suess effect, Δδ13Cant, are displayed in Figure 7. The deep parts of the Norwegian Basin and most of the Faeroe Bank Channel are excluded as they were outside the common validity range of the δ13C equations. There are less surface samples, so the contours are not drawn all the way to the surface.
 The 1981 to 2003 oceanic 13C Suess effect exhibits substantial regional variation in the Nordic seas, and appears to be even more closely tied to water mass distribution than anthropogenic CO2. In the core of the AW in the northern Norwegian Basin and the Lofoten Basin the δ13C decrease ranges between −0.4‰ and −0.6‰, and corresponds to annual decreases between −0.018‰ yr−1 and −0.024‰ yr−1. The changes are smaller in the Greenland Sea, only −0.1‰ which corresponds to −0.004‰ yr −1. As the AW is reencountered west of Spitsbergen larger changes are evident: between −0.3‰ and −0.4‰.
 The results on the changes of anthropogenic CO2 concentration and the 13C Suess effect over the last 20 years, the magnitude of these changes, and their spatial variation and relation to water mass distribution provides a framework that enables us to specify how anthropogenic CO2 is brought into the Nordic seas.
 The atmospheric δ13C has decreased at a fairly steady rate of −0.027‰ yr−1 since at least 1962 up to 2003 (calculated from the Southern Hemisphere record of Francey et al. , extended with the record from the Cape Grim Observatory (B. H. Vaughn and J. W. C. White, personal communication, 2005). The δ13C decreased in the AW by up to −0.024‰ yr−1, so this water mass is close to equilibrium with the atmospheric 13C Suess effect. However, this equilibrium cannot have been established solely within the Nordic seas because it takes 10 to 12 years to establish isotopic equilibrium between ocean and atmosphere [Broecker and Peng, 1974; Lynch-Stieglitz et al., 1995] so the transit time of about 1 year from the Iceland-Scotland gap up to the Lofoten Basin [Furevik, 2000, 2001] is insufficient. Thus the changes must have been advected laterally from farther south. The AW has been given time to equilibrate with the atmosphere as it travels toward the Nordic seas via the Gulf Stream–North Atlantic Current–North Atlantic Drift system. Although a large CO2 uptake within the Nordic seas would preferentially introduce isotopically light carbon to the ocean due to kinetic fractionation [Lynch-Stieglitz et al., 1995], the Ct change of 15 to 20 μmol kg−1 means that this kinetic effect can only account for a total decrease of around −0.06‰ (assuming starting conditions of Ct of 2160 μmol kg−1 at δ13C of −1‰, and a one-way addition of 15 μmol kg−1 CO2 at −10‰). The present estimates for the AW 13C Suess effect are similar to what has been seen upstream of the Nordic seas. Körtzinger et al.  found a mean 13C Suess effect of −0.026‰ yr−1 between 35°N and 60°N using isopycnal analysis, and annual changes at Hydrostation S close to Bermuda have been estimated to −0.025‰ yr−1 [Bacastow et al., 1996; Gruber et al., 1999].
 As for anthropogenic CO2 it appears as if the AW goes beyond maintaining equilibrium with the atmosphere because the pCO2 growth rate in this water actually exceeds the atmospheric growth rate based on the equivalent pCO2 increase displayed in Figure 8. Data from Mauna Loa shows that from 1981 to 2002 the atmospheric CO2 concentrations increased from 339.9 ppm to 373.1 ppm and further to 375.6 ppm in 2003 [Keeling and Whorf, 2005]. This gives an annual mean atmospheric CO2 increase of 1.6 ppm yr−1 over the time span covered by the present study (1981 to 2002/03, 21.5 years). A similar trend was observed in data obtained within this time interval at Ocean Weather Station Mike (66°N, 2°E) in the Norwegian Sea [Tans and Conway, 2005]. The equivalent pCO2 increase in the regions dominated by AW in the Nordic Seas appears larger than this. In particular, in the Faeroe Bank Channel the estimated growth rates are above 2 μatm yr−1, and in the most saline layers of the northern Norwegian Basin and the Lofoten Basin they range between 1.6 and 1.8 μatm yr−1. These excessive growth rates appear below the upper 100 m, the seasonal mixed layer. At these depths the error of the ΔCant estimates was ±5 μmol kg−1 which corresponds to an uncertainty of ±0.37 μatm yr−1 in the estimated growth rates (see caption of Figure 8 for details). This gives a range of probable growth rates in the inflow region of between 1.6 and 2.4 μatm yr−1, and from 1.2 to 2.2 μatm yr−1 in the AW farther north and so it is not crystal clear that pCO2 growth rates in the AW are exceeding those of the atmosphere. However, an inspection of the residuals of the MLR equations revealed no systematic bias with region or water mass properties which could ultimately have resulted in a systematic overestimation of the AW pCO2 growth rates. Moreover, the ±0.37 μatm yr−1 is a measure of the random error of each individual estimate and so the error in the gridded data presented in Figure 8 is probably smaller since independent random errors tends to cancel each other during averaging. This, and the fact that excessive growth rates have also been observed in the source region for the AW entering the Nordic Seas [Lefévre et al., 2004; Friis et al., 2005; Omar and Olsen, 2006] lead us to conclude that it is highly likely that the pCO2 growth rates in the AW feeding the Nordic Seas are exceeding those of the atmosphere. Local uptake of anthropogenic CO2 cannot sustain these growth rates. We suggest that the excessive increase is the manifestation of a negative feedback on the North Atlantic CO2 solubility pump brought about by the advection and cooling of water loaded with anthropogenic CO2. This phenomenon can be conceptualized in a number of ways [Wallace, 2001; Anderson and Olsen, 2002; Völker et al., 2002; Bellerby et al., 2005]. We have chosen the following approach. The AW that enters the Nordic seas originates farther south. As the water moves northward it is cooled and the increased solubility induces a transfer of CO2 from the atmosphere to the ocean. This cooling-induced uptake is part of the natural preindustrial exchange of CO2 between the ocean and the atmosphere; the CO2 solubility pump. As a result of the uptake of Cant, the Ct content of the source water from the Gulf Stream system has increased. Assume that the pCO2 of the warm source waters increases at the same rate as the atmosphere; this implies a given anthropogenic change in Ct. The water moves northward and carbon concentrations increase as a consequence of cooling-induced uptake. When Ct increases, the pCO2 response for a given increase in Ct becomes larger; that is, the buffer capacity decreases. Therefore, as the water moves northward, the effect on pCO2 of the given anthropogenic change in Ct increases. Thus the increase in the south, tracking the atmospheric increase, may bring about changes greater than the atmospheric increase farther north.
 The anthropogenic changes of both δ13C and Ct that have been estimated for the AW in the Nordic seas thus indicate that this water mass enters the region preequilibrated with the atmosphere. Within the Nordic seas it meets with PW and these two mix. The PW has recently exited from under the Arctic ice and has not been in contact with the contemporary atmosphere for long, so both ΔCant and Δδ13Cant are lower for this water mass than for the AW. The mixing with PW dilutes the anthropogenic carbon system signature of the AW, yielding the ties with water mass distribution seen for both ΔCant and Δδ13C in the Nordic seas. The relationship to water mass is stronger for Δδ13Cant than for ΔCant (Figures 6 and 7). Regression of ΔCant versus salinity in the upper 1000 to 100 m yielded an r2 of 0.40, which is much less than the r2 of 0.80 for Δδ13Cant versus salinity (not shown). We believe the different strengths of the relationships are the consequence of the uptake of anthropogenic CO2 from the atmosphere which has increased the initially low ΔCant of the PW after it has entered the Nordic seas, while the low Δδ13Cant signature has been preserved to a greater extent. This effect is due to the atmosphere-ocean CO2 equilibration rate being ten times greater than the 13C equilibration rate; only 1 year is required to equilibrate the surface ocean with respect to atmospheric CO2 changes whereas 10 years are required to equilibrate with respect to 13C changes [Broecker and Peng, 1974; Lynch-Stieglitz et al., 1995]. Thus it appears as if the influx of PW gives rise to a local sink for anthropogenic CO2 in the Nordic seas.
 One might ask to what extent it is not mixing with deep, rather than Polar Waters, that dilutes the anthropogenic signature of the AW. This is in accordance with the classical view of the northern high latitude anthropogenic CO2 sink [Takahashi et al., 2002], and modeling results [Orr et al., 2001; Wetzel et al., 2005]. However, the impact of deeper waters appears to be limited. This becomes clear through examining the ratio of the anthropogenic isotopic and carbon changes, ΔRc = Δδ13Cant/ΔCant. The atmospheric perturbation history of δ13C and anthropogenic CO2 has been similar and so one might assume similar oceanic penetration as well. This would imply that the ratio of the changes, i.e., ΔRc, should be constant. This theory was invoked by Heimann and Maier-Reimer , who, when introducing the “dynamic constraint” method for estimating oceanic uptake of anthropogenic CO2, determined the global ΔRc to be −0.016‰ (μmol kg −1)−1. However, because of the different equilibration times of 13C and CO2, significant spatial variations in ΔRc may arise. In particular, for situations where the surface residence time of a water mass has been inadequate for establishment of isotopic equilibrium the ratio will be less negative than for situations were isotopic equilibrium has been established. In the Southern Ocean, ΔRc was shown to vary between −0.007‰ (μmol kg −1)−1 for water masses with short surface residence time and down to −0.015‰ (μmol kg −1)−1 for waters with longer surface residence time [McNeil et al., 2001b]. Nordic seas ΔRc estimates are plotted as a function of salinity in Figure 9. They take on a wide range of values and, noticeably, vary as a strong function of salinity above 1000 m. This variation reflects the differences and mixing between the PW and AW as discussed above. The AW, which is the most saline water mass, is characterized by the most negative ΔRc which seems to stabilize at −0.032‰ (μmol kg −1)−1 at the highest salinity values. The fact that this is similar to the values observed by Körtzinger et al.  in the AW upstream of the Nordic seas and close to the value expected for a water mass whose CO2 and δ13C changes matches those of the atmosphere corroborates that the AW is equilibrated with the atmosphere when entering the Nordic seas. ΔRc approaches zero as salinity decrease, i.e., as the influence of PW increase. This confirms that the PW has not been exposed to the present atmosphere for very long and so is closer to atmospheric equilibrium with respect to CO2 than with respect to 13C.
 The deeper waters, defined here simply as waters below 1000 m, takes on a wide range of values independent of salinity. We identify three reasons why this is so. First, Nordic seas deep waters are, despite of the small variation of salinity, formed from a variety of water masses at a variety of places. The deep water originates in the Arctic Ocean as well as in the Nordic seas so variations in the magnitude of the ΔRc are to be expected. Second, limited and variable atmospheric exposure times for deep waters during convective events will introduce ΔRc variability. Finally, the relative uncertainty of the ΔCant and Δδ13Cant estimates at these depths is greater. The impact of the deep waters on the variability of ΔRc in the upper 1000 m seems to be limited, as most of the points here lie on the straight line connecting the PW and AW. Only at salinities around 34.9 does this slope seem perturbed by deep mixing with many points higher than the expected value. This is apparently a result of mixing with deeper waters, as these typically have higher ΔRc than what is expected from the PW-AW mixing line. As can be seen from Figure 5, it is only in the homogenous Greenland Sea that these waters reach the surface and support an anthropogenic carbon uptake.
6. Conclusions and Implications
 The increase of the anthropogenic carbon concentrations and the δ13C decrease in the Nordic seas have been quantified using an extended version of the multilinear regression approach of Wallace . The distribution of the changes and in particular their relationship to each other showed a high dependency on water masses and circulation. This dependency has illuminated the processes that bring anthropogenic CO2 into the Nordic seas. First, anthropogenic CO2 is advected into the region with the AW. The isotopic changes were too large to have taken place solely within the Nordic seas where the residence time is only a few years. Advection and cooling of water from farther south in the Atlantic provides a reasonable explanation for the excessive pCO2 increases evident in the core of the AW [Wallace, 2001; Anderson and Olsen, 2002]. The large negative ΔRc values of the AW corroborate this view; they reflect that the water has been exposed to the present atmosphere for a considerable time and are similar to the ΔRc values observed by Körtzinger et al.  south of the Scotland-Greenland ridge. Second, and in contrast to the AW, in the PW the anthropogenic CO2 changes are relatively large compared to the isotopic changes, perhaps best reflected in the trend of increasing ΔRc with decreasing salinity. This observation and the fact that the PW exits from under the Arctic ice imply that the influx of PW promotes an uptake of anthropogenic CO2 from the atmosphere within the Nordic seas. Finally, convection in the Greenland Sea does seem to lead to surfacing of deep water which takes up anthropogenic CO2 from the atmosphere.
 Given this, we expect that the air-sea flux of anthropogenic CO2 in the Nordic seas exhibits a distinct spatial pattern that is intimately linked to surface water mass distribution. In the eastern part, the Atlantic region, the uptake of Cant from the atmosphere is probably limited, if not even negative; that is, it may be a source of anthropogenic CO2 to the atmosphere. This latter phenomenon would be apparent as a decreased undersaturation compared to preindustrial times. Our findings indicate that the undersaturation has decreased at least over the last two decades. The area toward the west, on the other hand, across the Arctic Front, is most likely a sink of anthropogenic CO2 from the atmosphere as the surface waters of this region are appreciably influenced by PW and deep mixing. This feature, i.e., a quasimeridional gradient approximately following the position of the Arctic Front is not evident in estimates of the air-sea flux of anthropogenic CO2 from coarse-resolution ocean carbon model runs. There are discrepancies between different models, but most of them seem to agree that there is a basin-wide uptake of anthropogenic CO2 in the Nordic seas [Orr et al., 2001; Wetzel et al., 2005]. This flux is brought about by deep water formation in the models [Wetzel et al., 2005]. The difference between our inferred spatial pattern and model output supports the need for instance for better spatial resolution and improved representation of convective processes in order for models to properly reproduce oceanic uptake of anthropogenic CO2.
 When collapsing pCO2 data collected over a number of years to a 1995 climatology Takahashi et al.  assumed no temporal trend of surface ocean pCO2 north of 45°N in the Atlantic. They argued that the anthropogenic signal in surface ocean pCO2 in the region should be completely diluted by deep mixing. On the other hand Olsen et al.  assumed that the whole region followed the atmospheric pCO2 increase. Our results show that neither of these assumptions are totally correct. Olsen et al.  appear closer to the truth than Takahashi et al.  but the temporal evolution of surface ocean pCO2 differs from region to region. The limited spatial coverage of the TTO data and increased random uncertainty of the ΔCant estimates in the upper 100 m prohibit a comprehensive analysis, but as a first-order estimate we suggest that the annual pCO2 increase varies as function of salinity according to: ΔpCO2 = 2.074 × S − 71.13, r2 = 0.6, n = 76, RMS = 0.2. This equation was derived through the linear regression of the annual pCO2 increase as estimated for Figure 8 and salinity in the upper 500 to 100 m. 100 m was used as the upper cut off because of the increased random uncertainty of the ΔCant estimates shallower than this.
 The ratio of the anthropogenic isotopic and CO2 changes, ΔRc has been used as a conversion factor to get from estimates of the oceanic 13C Suess effect to anthropogenic CO2 estimates. Several authors have multiplied their estimates of δ13C changes by −0.016‰ (μmol kg −1)−1, the ΔRc estimate of Heimann and Maier-Reimer , to get to Cant estimates [Bauch et al., 2000; Ortiz et al., 2000]. This procedure cannot be recommended since different surface residence times of water masses produce substantial variations in ΔRc at regional scales, as was also shown by McNeil et al. [2001b] for the Southern Ocean. The ΔRc values of the AW are exceptional, and much more negative than the values determined by Tanaka et al.  at 44°N in the Pacific. This reflects the special circulation of the Atlantic. Tanaka et al.  suggested that their value was applicable in general in subpolar and polar regions; this is not correct.
 The present findings support the results of anthropogenic carbon transport calculations in that the AW brings with it a large amount of anthropogenic CO2 to the northern North Atlantic [Álvarez et al., 2003; Macdonald et al., 2003; Rosón et al., 2003]. In particular Álvarez et al.  estimated that 56% of the anthropogenic carbon sequestered north of 24.5°N was brought in by oceanic transport whereas 44% was taken up from the atmosphere within the region. Just how large the ratio is in the Nordic seas has yet to be determined, but the advective flux must be significant. Given our findings it is reasonable to assume that the AW is saturated with anthropogenic CO2. Temperature and TA of the AW entering the Nordic seas is around 8°C and 2325 μmol kg−1, respectively; this gives a total anthropogenic CO2 concentration of ∼53 μmol kg−1 (atmospheric pCO2 280 μatm vs. 380 μatm) which is reasonable compared to Körtzinger et al. . About 6 Sv overturn in the Nordic seas [Hansen and Østerhus, 2000]. Combining these numbers we estimate that overturning leads to an annual sequestration of roughly 0.12 Gt Cant in the Nordic seas. This is 6% of the annual oceanic accumulation of 1.8 Gt Cant [Prentice et al., 2001]. Future studies should elaborate our initial, conceptual appraisal of how mixing with PW dilutes the anthropogenic carbon system signature of the AW within the Nordic seas and quantify the relative importance of air-sea fluxes and oceanic transport for importing anthropogenic carbon into the region.
 What we would finally like to emphasize here is the fact that the Cant transport can bring around the excessive pCO2 increase seen in the core of the AW [Anderson and Olsen, 2002; Wallace, 2001]. Similar large increases have been seen farther south by Lefévre et al. , Friis et al. , and Omar and Olsen . This gives us reason to believe that the AW feature is the northernmost manifestation of a trend that occurs in all regions fed with water from the Gulf Stream; that is, surface pCO2 levels in the whole North Atlantic Drift–Norwegian Atlantic Current system are currently approaching equilibrium with the atmosphere.
 Financial support from the EU through IP CARBOOCEAN (511176-2) and TRACTOR (EVK2-2000-00080) is greatly appreciated. The Cape Grim Observatory δ13C record from 1991 to 2003 was kindly provided by Bruce H. Vaughn and James W.C. White at the Institute of Arctic and Alpine Research, University of Colorado. The authors would like to thank the officers and crews of R/V Knorr and R/V G.O. Sars and the chief scientists of the Knorr cruise, Jim Swift and Bill Smethie, for their cooperation and organization. Yoshie Kasajima kindly provided the Sars temperature and salinity data, and Jim Swift kindly provided the Knorr temperature, salinity, and nutrients data. James Orr and Rolf Sonnerup provided thorough, intelligent and helpful comments on the manuscript. This is contribution A 131 from the Bjerknes Centre for Climate Research.