Anthropogenic CO₂ accumulation rates in the North Atlantic Ocean from changes in the ¹³C/¹²C of dissolved inorganic carbon

Abstract

[1] The anthropogenic CO₂ accumulation rate for the North Atlantic Ocean was estimated on the basis of the decrease in the δ¹³C of the dissolved inorganic carbon measured between cruises in 1981 (Transient Tracers in the North Atlantic), 1993 (OACES) and 2003 (Repeat Hydrography). A mean depth-integrated δ¹³C change of −15.0 ± 3.8‰ m yr⁻¹ was estimated by applying a multiple linear regression approach to determine the anthropogenic δ¹³C decrease at 22 stations where δ¹³C depth profiles were compared. The largest and deepest anthropogenic δ¹³C decreases occurred in the subpolar ocean and, in contrast, the smallest and shallowest decreases occurred in the tropical ocean. A mean anthropogenic CO₂ accumulation rate of 0.63 ± 0.16 mol C m⁻² yr⁻¹ (0.32 ± 0.08 Pg C yr⁻¹) in the North Atlantic Ocean over the last 20 years was determined from the mean depth-integrated δ¹³C change and a ratio of anthropogenic δ¹³C to DIC change of −0.024‰ (μmol kg⁻¹)⁻¹. Only half of the accumulated anthropogenic CO₂ in the North Atlantic during the last 20 years was the result of air-sea CO₂ uptake, based on a comparison of the air-sea ¹³CO₂ flux to the DIC¹³ inventory change, with the other half likely a result of northward advective transport.

1. Introduction

[2] The ¹³C/¹²C isotope ratio of dissolved inorganic carbon (DIC) is a useful tracer of oceanic uptake of anthropogenic CO₂. The anthropogenic perturbation signal in the δ¹³C of DIC (hereafter referred to as δ¹³C) is stronger than that for the concentration of DIC itself. For example, in surface waters at the time series station near Bermuda (BATS), Gruber et al. [2002] measured a 20-year-long decrease in δ¹³C of −0.24‰ per decade, about equal to the amplitude of the seasonal cycle (∼0.2‰), and a corresponding increase in surface DIC concentration of ∼6 μmol kg⁻¹ per decade that is ∼20% of the seasonal cycle (30 μmol kg⁻¹). In most regions, decadal changes in δ¹³C resulting from anthropogenic CO₂ accumulation will be easier to separate from changes resulting from natural variations in the ocean (e.g., biological productivity, circulation, mixing, etc.) than will anthropogenic changes in DIC concentration. We make use of this characteristic of δ¹³C to determine the accumulation rate of anthropogenic CO₂ in the North Atlantic Ocean over the last 20 years and the processes contributing to the increase in anthropogenic CO₂ burden in this ocean basin.

[3] To date the utility of δ¹³C as a tracer has been limited by the availability of historic data. In rare cases, like at the subtropical time series stations near Hawaii (HOT) and Bermuda (BATS), decade-long measurements clearly demonstrate the anthropogenic changes in δ¹³C [Gruber et al., 2002; Quay et al., 2003; Keeling et al., 2004]. Alternately, to detect decadal δ¹³C changes one can rely on snapshot comparisons between cruises [Quay et al., 2003]. However, these opportunities are limited by the lack of historic δ¹³C data and subject to natural (nonanthropogenic) δ¹³C variations in space and time. To overcome the latter complication, regression techniques have been applied to separate natural from anthropogenic δ¹³C changes [e.g., Sonnerup et al., 2000; McNeil et al., 2001]. Additionally, anthropogenic δ¹³C change rates have been estimated by determining along isopycnal trends in preformed δ¹³C, i.e., measured δ¹³C corrected for biological effects [Sonnerup et al., 1999; Körtzinger et al., 2003]. The advantage of the preformed method is that historic data sets are not needed. The preformed approach works best in oceanic situations where end-member mixing of water masses does not dominate the spatial δ¹³C trends.

[4] The abundance of high-quality δ¹³C data improved markedly during WOCE with about 20,000 δ¹³C values measured throughout all the ocean basins in the 1990s [Quay et al., 2003]. The CO₂/CLIVAR Repeat Hydrography Program began to repeat carbon measurements along WOCE ocean sections in the North Atlantic in 2003. The Repeat Hydrography (RH) program will provide an excellent opportunity to determine the changes in both the concentration and δ¹³C of DIC that have resulted from anthropogenic CO₂ accumulation since WOCE. In addition to the RH cruises, another significant development has been the increased use of Volunteer Observing Ships (VOS) to measure pCO₂ in the surface ocean. We have taken advantage of these platforms to collect hundreds of surface δ¹³C samples during the last 10 years, which substantially improves our resolution of the surface ocean δ¹³C change.

[5] In this presentation, we determine the δ¹³C change in the Atlantic Ocean between the decades of the 1980s, 1990s and 2000s. We directly compare the depth distributions of δ¹³C measured in the North Atlantic during the 2003 RH cruises A16N, A20N and A22N with δ¹³C measurements from an OACES (NOAA) cruise in 1993 and the Transient Tracer in the North Atlantic (TTNA) cruise in 1981 in the same regions. On a basin-wide scale, we determine the surface δ¹³C values measured during more than two dozen cruises spanning 1981 to 2005. We use the estimated depth-integrated δ¹³C change in the North Atlantic Ocean to estimate the rate of anthropogenic CO₂ accumulation.

[6] We find that the mean basin-wide surface δ¹³C change in the Atlantic Ocean is ∼−0.018 ± 0.002‰ yr⁻¹ which is similar to the rates of change found previously in the Pacific and Indian oceans, i.e., −0.018 and −0.014‰ yr⁻¹, respectively [Quay et al., 2003]. The mean depth-integrated δ¹³C change in the North Atlantic since the 1980s is −15.0 ± 3.8‰ m yr⁻¹, which is twice the mean global ocean rate determined previously [Quay et al., 2003]. We determine that the depth-integrated δ¹³C change in the North Atlantic corresponds to an anthropogenic CO₂ accumulation rate of 0.63 ± 0.16 mol C m⁻² yr⁻¹ (0.32 ± 0.08 Pg C yr⁻¹). By comparing the depth-integrated DI¹³C change to the air-sea ¹³CO₂ flux, we estimate that the air-sea input of anthropogenic CO₂ accounts for only 50% of the DI¹³C inventory change. The other half of the DI¹³C inventory change is likely due to northward advection of surface waters into the North Atlantic that carry anthropogenic CO₂ and ¹³CO₂, as part of the meridional circulation associated with North Atlantic Deep Water formation.

2. Methodology

2.1. Sampling

[7] Samples for δ¹³C depth profiles were collected at 22 stations during the NOAA/OACES cruise in July 1993 on a meridional section along 20°W–30°W between 5°S and 63°N. This same meridional section was reoccupied during the RH-A16N cruise during July 2003 when 12 stations were sampled for δ¹³C. During September 2003, depth profiles for δ¹³C were collected at 11 stations during the RH-A20N and RH-A22N cruises in the region between 15°N–42°N and 50°W–70°W. The δ¹³C values from RH-A20/22N are compared to δ¹³C depth profiles measured by Keeling and Guenther [1994] at 11 stations during the Transient Tracers in the North Atlantic cruises (TTNA) in this same region (15°N–60°N, 42°W–54°W) in 1981. The δ¹³C data from TTNA can be found at http://cdiac.esd.ornl.gov/ftp/oceans/keeling.data/.

[8] Surface samples for δ¹³C measurement (n ∼ 1100) were collected on 28 cruises between 1981 and 2005. These cruises span the North and South Atlantic, with greatest coverage in the tropics and subtropics and poorest coverage in the subpolar latitudes in both hemispheres. The cruise track locations are presented in Figure 1.

Figure 1

The research and VOS cruises on which δ¹³C samples were collected between 1981 and 2005 with their individual years included as part of the cruise name.

2.2. The δ¹³C Measurements

[9] The δ¹³C measurements made at the Stable Isotope Laboratory at the University of Washington had a reproducibility of ±0.03‰ based on replicate measurements of standards and sample pairs collected from the same Niskin water sampler [Quay et al., 2003]. The analytical method has been described previously [Quay et al., 1992; Quay et al., 2003]. To test for analytical offsets in the δ¹³C data sets, we compared the δ¹³C at depths >2000 m measured at overlapping stations south of 40°N during the OACES (1993) and RH-A16N (2003) cruises. The mean difference was 0.00 ± 0.06‰. A similar deep-water offset test was made between δ¹³C values measured at overlapping stations during RH-A20/22N (2003) and TTNA (1981). The mean difference was +0.01 ± 0.05‰. There was no evidence of significant δ¹³C offset between the data sets collected in 1981, 1993 and 2003.

2.3. Multiple Linear Regression

[10] To separate natural from anthropogenic changes in DIC and δ¹³C, a multiple linear regression (MLR) procedure has often been used [e.g., Wallace, 1995; Sabine et al., 1999; Sonnerup et al., 2000]. The approach uses the correlation between δ¹³C (or DIC) and hydrographic properties of the water measured at an earlier time to predict the δ¹³C (or DIC) at a later time from measured hydrographic properties. Following Sonnerup et al. [2000], we expressed the MLR estimate of δ¹³C as follows:

where β represents the intercept, m1 through m4 represent the regression coefficients for each predictive variable, R represents the residuals between the predicted and observed δ¹³C for each data point, θ is potential temperature, S is salinity, AOU is apparent oxygen utilization and PO₄ is phosphate. The MLR approach ostensibly accounts for temporal and spatial variations in δ¹³C that are a result of changes in the hydrographic and biogeochemical characteristics of the water. The portion of the δ¹³C change that cannot be explained by the MLR approach (i.e., residuals) is assumed to result from the accumulation of anthropogenic CO₂ during the intervening time period. Although this assumption is difficult to prove, the observation that the measured δ¹³C change is very similar to the portion of the change that cannot be explained by the MLR approach (see Figure 4 in section 3.1.2) implies that the anthropogenic δ¹³C change is a robust feature and thus less sensitive to the assumptions of the MLR method.

[11] In a typical application, the MLR approach uses correlations between δ¹³C (or DIC) and hydrographic properties that were determined over the entire depth range and horizontal extent of a particular cruise. Often, the MLR determined this way lead to poor estimates of surface ocean changes (large variability in residuals) and a systematic depth distribution of residuals. Sonnerup et al. [2000] found depth and latitude dependences of the uncertainties of the δ¹³C predictions using an MLR approach to compare δ¹³C measured during GEOSECS (1978) and WOCE (1995) in the Indian Ocean. We modified the MLR approach by determining MLR coefficients for data sorted into 17 isopycnal intervals (range 24 < σ < 27.9), which yielded an average of 22 data points per interval. There are two advantages to this modification. First, it allows the MLR to correlate δ¹³C changes along a more likely circulation pathway for subducting water masses. Second, it can account for spatial variations in the preformed δ¹³C and hydrographic properties in the outcrop regions where water masses are subducted. The isopycnal MLR approach (isoMLR) predicted more accurate estimates of the δ¹³C data (±0.04‰), compared to the predictions determined using a single MLR for the entire density range (±0.09‰), as discussed below. The coefficients for the isoMLR applied to the δ¹³C measured during the OACES cruise in 1993 and the TTNA cruise in 1981 and the residuals of the MLR estimates are presented in the auxiliary material.

[12] We also tested the MLR approach (called eMLR) utilized by Friis et al. [2005] to estimate the anthropogenic DIC change in the subarctic North Atlantic. They determined two MLRs, one each for the recent and earlier DIC data sets, and determined the anthropogenic DIC increase using the difference in the MLR coefficients applied to a single hydrographic data set. Friis et al. [2005] found a substantial reduction in errors using eMLR method, however, our application of the eMLR method yielded a larger range in δ¹³C residuals than the isoMLR approach and systematic depth trends in the residuals.

[13] On the basis of the above observations, we used the isoMLR approach to determine the depth distribution of anthropogenic δ¹³C changes between 1981, 1993 and 2003 in the North Atlantic. We determined the MLR coefficients from two δ¹³C data sets collected on earlier cruises (i.e., OACES 1993 and TTNA 1981) to predict the δ¹³C distributions on two later cruises in 2003 (i.e., RH-A16N and RH-A20/22N, respectively).

3. Results and Discussion

3.1. Depth Trends

3.1.1. Measured δ¹³C Decrease

[14] A direct comparison of the 12 δ¹³C depth profiles collected during RH-A16N (2003) was made with δ¹³C depth profiles collected at the same locations in 1993 during the OACES cruise to the eastern North Atlantic (Figure 2). There is a clear meridional trend in the magnitude and depth of the decadal δ¹³C decrease, which has been observed previously [Sonnerup et al., 1999; Quay et al., 2003]. The shallowest (<400 m) δ¹³C decreases are observed in the tropics (0°–20°N). In the subtropics (20°N–40°N), the δ¹³C decrease typically extends to about ∼1000 m. In the subpolar latitudes (>40°N), the deepest δ¹³C changes were observed often extending into the deep sea (>3000 m). A comparison of the δ¹³C depth profiles collected during OACES and RH-A16N at 18°N 29°W demonstrates a situation where there was a large δ¹³C decrease (up to −0.45‰) between 1993 and 2003, which exceeded the δ¹³C decrease of ∼−0.34‰ in the atmospheric CO₂ over this 10-year interval. At this station the large δ¹³C decrease is associated with an increase in phosphate and apparent oxygen utilization (AOU) [Johnson and Gruber, 2006], indicating that the δ¹³C decrease is primarily not anthropogenic.

Figure 2

Selected δ¹³C depth profiles measured during OACES cruise in 1993 (squares) and RH-A16N cruise in 2003 (triangles) and the MLR predicted profiles for 2003 (line) in the eastern North Atlantic Ocean.

[15] To determine the changes in the δ¹³C depth distribution in the western North Atlantic we compared 7 depth profiles collected during the RH A20N and A22N in 2003 with depth profiles collected at nearby locations during the TTNA cruise in 1981. Because of the 22 year interval between these cruises, the absolute δ¹³C decrease should be about twice that observed in the eastern Atlantic between 1993 and 2003. Generally, we found similar latitudinal trends in the δ¹³C decrease in the western North Atlantic as observed in the eastern portion of the basin. The δ¹³C decrease for the subtropical stations (18°N–36°N) extended to ∼1000 m, whereas at the most northerly station (42°N) in the subpolar region, the δ¹³C decrease extended to 4000 m.

[16] Generally, the observed changes in the δ¹³C depth profiles in the western and eastern basins of the North Atlantic were consistent with previous results from the Pacific and Indian Oceans. Shallow and small δ¹³C decreases occurred in the tropics and larger and deeper decreases occur in the subtropics. However, we had not previously observed δ¹³C decreases that extended into the deep sea as we observed in the subpolar North Atlantic.

3.1.2. Using a Multiple Linear Regression to Predict Anthropogenic δ¹³C Decrease

[17] When the isoMLR method was applied to the δ¹³C data from the OACES (1993) cruise the overall mean residual was ±0.04‰ (±1 SD), which was half of the mean residual (±0.09‰) resulting from application a single MLR determined over the entire OACES δ¹³C data set (Figure 3). In the surface layer the isoMLR had an average residual of ±0.07‰ versus ±0.18‰ for the cruise-wide MLR.

Figure 3

(a) The residuals from a multiple linear regression (MLR) applied on isopycnal intervals to the δ¹³C distribution measured during the OACES cruise in 1993 and the means (squares) binned by depth intervals; (b) the mean residuals from the isopycnal MLR (squares) compared to mean residuals from an MLR applied to all δ¹³C depth profiles measured on the OACES cruise (triangles). Uncertainties of means represent ±1SD.

[18] Notably, the isoMLR-based estimate of the mean anthropogenic δ¹³C decrease between the RH A16N and OACES cruises and the RH A20/22N and TTNA cruises has a depth distribution that closely follows the depth distribution of the difference between the raw δ¹³C values measured on these cruises (Figure 4). This result implies that the anthropogenic δ¹³C change is a robust feature that dominates over the natural δ¹³C variability; that is, the MLR is not extracting an anthropogenic signal out of a noisy background. Despite the robustness of the directly observed δ¹³C change, the MLR approach reduced the overall variability in the estimated mean δ¹³C change between 1993 and 2003 by about half compared to that for the mean measured δ¹³C change, i.e., ±0.08‰ versus ±0.18‰ averaged over 0–1000 m.

Figure 4

(a) The residuals between δ¹³C measured during A16N in 2003 and OACES in 1993 (diamonds) compared to the residuals from the measured A16N minus the isopycnal MLR predicted δ¹³C for A16N (squares). (b) Same for A20/22N in 2003 compared to TTNA in 1981. Error bars represent ±1SD of the mean residual over the depth interval.

3.1.3. Depth-Integrated δ¹³C Change

[19] The depth-integrated anthropogenic δ¹³C change at each station was determined by applying the isoMLR method to 21 stations measured in 2003 during the Repeat Hydrography cruises (A16N, A20N and A22N). At 19 of these stations, where earlier δ¹³C depth profiles were measured during OACES and TTNA cruises, we calculated the depth-integrated anthropogenic δ¹³C change using both the measured δ¹³C change and the isoMLR predicted change (Table 1). The mean depth-integrated δ¹³C change for these 19 stations was −16.5 and −15.7‰ m yr⁻¹ using the measured and isoMLR δ¹³C changes, respectively. Both the measured and isoMLR predicted depth-integrated δ¹³C changes showed clear meridional trends (Table 1). The smallest δ¹³C changes occurred in the tropics (−2.7 to −5.7‰ m yr⁻¹), larger changes were found in the subtropics (−11.7 to −23.0‰ m yr⁻¹) and the largest changes (−19.6 to −36.7‰ m yr⁻¹) occurred in the subpolar North Atlantic where the δ¹³C decrease extended to 3000 m or more (Figure 5). However, there is a limited amount of data, i.e., only 10 pairs of δ¹³C measurements at >2000 m for stations located north of 40°N, thus the magnitude of the subpolar δ¹³C change at depths >2000 m is not well represented. The data available, however, indicate that the anthropogenic δ¹³C signature has penetrated well into the deep sea in the subpolar North Atlantic. This result agrees with the MLR-based estimates of anthropogenic DIC increase for the subpolar North Atlantic (42°N–65°N; 12°W–60°W) between TTNA (1981) and Meteor cruises (1997–1999) by Friis et al. [2005]. They found evidence for an anthropogenic DIC increase of ∼ 10 μmol kg⁻¹ extending to the bottom (∼5000 m).

Figure 5

The depth-integrated δ¹³C change (‰ m decade⁻¹) estimated from applying the MLR to station pairs from OACES (1993) and A16N (2003) and from TTNA (1981) and A20/22N (2003).

Table 1. Depth-Integrated Change in δ¹³C at 19 Sites in the North Atlantic Ocean Where δ¹³C Depth Profiles Were Measured During the Transient Tracer in North Atlantic (TTNA) in 1981, OACES in 1993, and Repeat Hydrography Cruises A16N, A20N, and A22N in 2003^a

Station Pair		Integration Depth MLR, m	Integrated δ^13C Change, ‰ m yr^−1	MLR Integrated δ^13C Change, ‰ m yr^−1
A16N 150 (6°S 25°W)	OACES 02 (5°S 25°W)	500	−4.7	−5.0 ± 1.3
A16N 135 (0°N 25°W)	OACES 12 (0°N 25°W)	500	−5.9	−5.7 ± 1.1
A16N 122 (5°N 26°W)	OACES 22 (5°N 26°W)	550	−2.5	−4.3 ± 1.3
A16N 108 (12°N 29°W)	OACES 29 (12°N 29°W)	575	−6.7	−2.7 ± 1.1
A20N 54 (15°N 52°W)	TTNA 32 (15°N 54°W)	500	−5.9	−5.5 ± 0.6
A16N 96 (18°N 29°W)	OACES 41 (18°N 29°W)	750	−14.8	−4.4 ± 1.6
A20N 45 (21°N 52°W)	TTNA 36 (21°N 54°W)	1000	−11.4	−12.1 ± 1.8
A20N 40 (24°N 52°W)	TTNA 38 (24°N 54°W)	1000	−13.8	−14.0 ± 1.4
A16N 82 (25°N 26°W)	OACES 38 (24°N 27°W)	1250	−19.1	−11.7 ± 2.3
A16N 71 (30°N 23°W)	OACES 35 (30°N 25°W)	1250	−20.7	−17.6 ± 2.7
A20N 29 (32°N 52°W)	TTNA 234 (32°N 51°W)	1350	−16.8	−15.1 ± 2.1
A20N 26 (36°N 52°W)	TTNA 231 (36°N 47°W)	3000	−15.8	−23.0 ± 3.3
A16N 55 (38°N 20°W)	OACES 56 (38°N 20°W)	1250	−16.8	−21.8 ± 2.1
A20N 24 (39°N 52°W)	TTNA 229 (39°N 44°W)	3100	−9.0	−22.3 ± 3.3
A22N 11 (42°N 52°W)	TTNA 228 (42°N 42°W)	4000	−33.2	−19.6 ± 3.8
A16N 43 (44°N 20°W)	OACES 63 (44°N 20°W)	3000	−25.0	−24.6 ± 4.1
A16N 31 (50°N 20°W)	OACES 66 (48°N 20°W)	3600	−32.8	−36.7 ± 4.4
A16N 19 (56°N 20°W)	OACES 75 (57°N 20°W)	1450^b	−25.3	−25.0 ± 1.9
A16N 9 (61°N 20°W)	OACES 80 (61°N 20°W)	2400^b	−23.3	−21.1 ± 3.5

• a The integration depth and depth-integrated δ¹³C change, based on differences between measured δ¹³C and MLR-predicted δ¹³C (see text), are presented.
• b Maximum water depth at station.

[20] The mean (±1SD) depth-integrated δ¹³C change rates were −4.5 ± 1.2, −17.2 ± 4.2 and −25.4 ± 6.7‰ m yr⁻¹ for the tropics (0°–20°N), subtropics (20°N–40°N) and subpolar (40°N–65°N) regions, respectively, based on MLR calculated δ¹³C changes at 21 stations, which yielded an area-weighted basin-wide mean value (±1SD) of −15.0 ± 3.8‰ m yr⁻¹ for the North Atlantic. Alternatively, we estimated a basin-wide depth-integrated δ¹³C change of −12.9 ± 2.4‰ m yr⁻¹ using the correlation between the depth-integrated changes in anthropogenic δ¹³C and bomb ¹⁴C measured during GEOSECS (Figure 6) and a mean bomb ¹⁴C burden of 14.4 × 10⁹ atoms cm⁻² for the North Atlantic estimated by Broecker et al. [1995].

Figure 6

The depth-integrated δ¹³C change (‰ m decade⁻¹) estimated for stations in the Atlantic (Table 1), Pacific [Quay et al., 2003], and Indian [Sonnerup et al., 2000] oceans and the bomb ¹⁴C inventory in 1975 (10⁹ atoms cm⁻²) estimated for nearby GEOSECS stations [Broecker et al., 1995].

3.1.4. Error Analysis

[21] The uncertainty in the depth-integrated δ¹³C change depends primarily on the uncertainty of the MLR used to estimate the δ¹³C depth profiles expected in 2003 if there was no anthropogenic change. The depth-integrated δ¹³C change equals the expected minus the measured δ¹³C depth profile in 2003 at 21 stations (see Table 1). The uncertainty in the MLR estimate of the expected δ¹³C in 2003 was determined on the basis of the residuals (see equation (1)) associated with the application along isopycnal surfaces of the MLR method to the δ¹³C data sets from the TTNA (1981) and OACES (1993) cruises. The δ¹³C residuals averaged ±0.08‰ and ±0.06‰ for the TTNA and OACES data sets, respectively (see auxiliary material). The MLR predicted a δ¹³C value in 2003 at each depth (density) with its associated uncertainty (residual). A Monte Carlo approach was used to randomly select an expected δ¹³C value at each depth (density) based on a normal distribution around a mean (estimated from the MLR) and a standard deviation (SD) equal to the MLR residual on that density surface. The uncertainty in the measured δ¹³C was assumed equal to ±0.03‰ at all depths. The depth-integrated difference between the randomly chosen measured δ¹³C depth profile and MLR predicted δ¹³C profile was determined. This procedure was repeated 300 times and the mean difference and SD of the difference was determined. These are the ±1SD values reported as the errors in the individual depth profiles presented in Table 1. The mean basin-wide average depth-integrated δ¹³C change of −15 ± 3.8‰ m yr⁻¹ was determined by spatially weighting the δ¹³C profile changes at the 21 sites and the uncertainty represents the ±1SD of the 21 profile changes. (The mean error (±1SD) in the individual depth-integrated δ¹³C changes at the 21 sites was slightly smaller at ±2.2‰ m yr⁻¹). The data sparseness could yield a bias in the basin-wide mean value. Although biases are difficult to estimate, the observation that the ¹⁴C normalized δ¹³C change of −12.9 ± 2.4‰ m yr⁻¹ is within error similar to the spatially weighted change of −15.0 ± 3.8‰ m yr⁻¹ suggests that the spatial distribution of stations did not substantially bias the basin-wide average.

3.1.5. Comparison to Indian and Pacific Oceans

[22] The depth-integrated anthropogenic δ¹³C change between 1981, 1993 and 2003 in the North Atlantic is much higher than in the Indian and Pacific oceans (Figure 7). In the Indian Ocean, Sonnerup et al. [2000] used an MLR method to calculate depth-integrated δ¹³C decreases at 35 WOCE stations that ranged from −2.0 to −12.5‰ m yr⁻¹ and averaged −6.9‰ m yr⁻¹ from 0° and 53°S between 1978 and 1995. Their smallest δ¹³C decreases occurred in the tropics (10°S–5°N) at −3.0 ± 3.0‰ m yr⁻¹ and largest δ¹³C decrease occurred in the subtropics (20°S–40°S) at −12.0 ± 5.0‰ m yr⁻¹. In comparison, we found a similar δ¹³C decrease for the tropical North Atlantic and significantly larger decreases in the subtropical and subpolar North Atlantic. In the Pacific Ocean, Quay et al. [2003] determined an area-weighted integrated δ¹³C decrease between the 1970s and 1990s that averaged −5.3 ± 4.0‰ m yr⁻¹, however, this estimate is based on the raw δ¹³C change measured at only 10 station pairs. Since the δ¹³C decrease rate in the ocean is increasing with time as the decrease rate of δ¹³C of atmospheric CO₂ accelerates, a more accurate comparison of the results in the North Atlantic, Pacific and Indian oceans accounts for the time offset of the data sets. One estimate of the acceleration of the ocean's δ¹³C decrease rate is obtained from model simulations of the anthropogenic CO₂ perturbation. A GCM simulation of the anthropogenic CO₂ accumulation and resulting δ¹³C changes in the Atlantic Ocean since the 1980s (R. Sonnerup, unpublished data, 2006), indicates that the rate of depth-integrated δ¹³C change in the Atlantic Ocean increased by about −3‰ m yr⁻¹ between the 1980s and 1990s. On the basis of this model simulation, we should add ∼−4.2‰ m yr⁻¹ to the Pacific results and ∼−2.7‰ m yr⁻¹ to the Indian Ocean results to compare with our more recent results from the North Atlantic. Even with this correction, however, the depth-integrated δ¹³C changes in the Pacific and Indian oceans are only two thirds of the −15.0 ± 3.8‰ m yr⁻¹ rate we estimated for the N Atlantic. Furthermore, in the subpolar North Pacific, Quay et al. [2003] detected no significant depth-integrated δ¹³C decrease between the 1970s and 1990s, a situation that contrasts sharply with the large δ¹³C decreases observed for the subpolar North Atlantic (Figure 7). The differences in depth-integrated δ¹³C decreases between the Indian, Pacific and Atlantic oceans mimic the inter basin differences in bomb ¹⁴C burdens (Figure 6); that is, the largest bomb ¹⁴C burdens during GEOSECS in the 1970s were found in the subpolar North Atlantic and the smallest burdens were in the tropical ocean [Broecker et al., 1995].

Figure 7

The depth-integrated δ¹³C change (‰ m decade⁻¹) estimated from: direct station pair comparisons in the Pacific Ocean (diamonds) between 1970 and 1990 [Quay et al., 2003], applying MLR to station pairs in the Indian Ocean (squares) between 1978 and 1995 [Sonnerup et al., 2000] and Atlantic Ocean (triangles) between 1981 and 2003 (Table 1) and from preformed δ¹³C calculations in the Atlantic, Pacific, and Indian oceans (circles) [Sonnerup et al., 1999].

[23] Using a linear regression to approximate the relationship between depth-integrated anthropogenic δ¹³C change and bomb ¹⁴C burden in 1975 (Figure 6) and a global mean bomb ¹⁴C burden of 8.5 × 10⁹ atoms cm⁻² [Broecker et al., 1995] yields a global ocean mean δ¹³C change of −7.2 ± 3.1‰ m yr⁻¹, which is about half the rate estimated for the North Atlantic. This result implies that either the anthropogenic CO₂ accumulation rate or the ratio of anthropogenic δ¹³C to CO₂ change is higher than average in the North Atlantic. The latter explanation is more likely, as discussed below.

3.2. Surface Ocean δ¹³C Changes

3.2.1. North Atlantic Trends

[24] There is a clear δ¹³C decrease for the surface ocean along 20°W–30°W in the eastern North Atlantic during the decade between the 1993 OACES and 2003 RH-A16N cruises (Figure 8). The largest surface ocean δ¹³C decrease at −0.04 to −0.05‰ yr⁻¹ occurred in the subpolar region between 40°N–60°N and exceeded the atmospheric δ¹³C decrease rate of ∼−0.034‰ yr⁻¹ over this time period. This result was unexpected since previously the largest surface ocean δ¹³C changes had been observed in the subtropics [Gruber et al., 1999; Quay et al., 2003]. However, there was a substantial increase in surface ocean phosphate concentrations in the subpolar region between the OACES and RH-A16N cruises (Figure 8), suggesting that changes in water mass properties was partly responsible for the large δ¹³C decrease observed at 40°N–60°N. Using the isoMLR method to estimate the surface ocean δ¹³C changes between the OACES and RH-A16N cruises yielded a different meridional trend. The maximum δ¹³C decrease of ∼−0.025 ± 0.004‰ yr⁻¹ occurred in the subtropics (20°N–40°N) with smaller decreases in the tropics (∼−0.011 ± 0.003‰ yr⁻¹) and subpolar (∼−0.017 ± 0.005‰ yr⁻¹) region (Figure 9). The surface δ¹³C changes between TTNA (1981) and A20/22N (2003) in the western North Atlantic, calculated as the difference between measured δ¹³C in 2003 and isoMLR predicted in 2003, yielded a similar δ¹³C decrease rate for the subtropical region at −0.025 ± 0.002‰ yr⁻¹ (Figure 9). These rates agree well with the δ¹³C decrease rate of −0.024 ± 0.001‰ yr⁻¹ estimated from monthly measurements in the surface layer at BATS (32°N 64°W) between 1984 and 2002 [Gruber et al., 2002]. The surface δ¹³C change in the subtropical North Atlantic estimated using the measured δ¹³C values was −0.022 ± 0.005‰ yr⁻¹ versus the MLR-based δ¹³C change estimate of −0.025 ± 0.004‰ yr⁻¹. The major difference between the measured and MLR-based surface δ¹³C changes occurred in the subpolar latitudes where the MLR predicted change was half the measured change (Figure 9). Thus for tropical and subtropical North Atlantic Ocean, the surface ocean δ¹³C decrease we measured between the 1981, 1993 and 2003 primarily represents the anthropogenic signature of δ¹³C change, whereas in the subpolar North Atlantic the observed δ¹³C decrease was significantly affected by changes in water mass properties.

Figure 8

(a) The δ¹³C and (b) phosphate concentration measured in the surface layer during the OACES cruise in July 1993 (squares) and the Repeat Hydrography A16N cruise in July 2003 (triangles) along 25°W in the eastern North Atlantic Ocean.

Figure 9

The δ¹³C change rate (‰ per decade) in the surface waters of the North Atlantic Ocean estimated from: differences between δ¹³C measured during RH-A16N (2003) and OACES (1993) (diamonds), differences between δ¹³C measured during RH-A16N (2003) and δ¹³C predicted for this cruise by isoMLR (triangles), and differences between δ¹³C measured during RH-A20/22N (2003) and δ¹³C predicted by isoMLR for this cruise in 2003 (squares). Mean δ¹³C changes ±1SD over 5° latitude bands are presented.

3.2.2. Basin-Wide Trends

[25] To spatially broaden the estimated surface δ¹³C change in the Atlantic Ocean, we utilized the transit legs during several research and VOS cruises for underway δ¹³C sample collection (Figure 1). The advantage of this approach is that it yielded an additional 800 surface δ¹³C measurements on 12 cruises since the 1990s and, as a result, provided reasonable coverage for most of the Atlantic Ocean, except south of ∼50°S. One disadvantage of this approach is that temperature and salinity were typically the only ancillary measurements available for underway samples. Thus we can calculate the surface δ¹³C change using the raw data but not the MLR method. On the basis of the good agreement between measured and MLR-based estimates of δ¹³C change between OACES and RH-A16N and between TTNA and RH-A20/22N cruises, discussed above, we expect that the measured δ¹³C change in the tropics and subtropics likely reflects mainly anthropogenic δ¹³C change (Figure 9). However, for higher latitudes, where seasonal and spatial δ¹³C variability is greater, a significant portion of the observed decadal δ¹³C change may reflect natural variability.

[26] We extended the calculation of the surface δ¹³C change rate in the Atlantic Ocean back to the 1980s utilizing the δ¹³C measurements of Keeling and Guenther [1994] during several cruises in the 1980s (e.g., TTNA, TTTA, SAVE, AJAX, ANTIPODES) and of Mackensen et al. [1996] during Polarstern cruises in the South Atlantic Ocean. As mentioned above, a deep water comparison of our recent δ¹³C measurements during A20/22N to those of Keeling and Guenther during the TTNA cruise yielded no significant offset (0.01 ± 0.05‰). We do not have measured δ¹³C depth profiles south of 50°S in the South Atlantic to make a similar deep water comparison to the δ¹³C data of Mackensen et al. [1996].

[27] This compilation of surface ocean δ¹³C data (1050 measurements collected on 28 cruises) shows meridional trends (Figure 10) similar to those observed previously in the Pacific and Indian oceans [Gruber et al., 1999; Sonnerup et al., 2000; Quay et al., 2003]. The highest δ¹³C values are found at ∼40S°–50°S and the lowest values are found in the Southern Ocean (south of 60°S). The δ¹³C measured in the tropics is higher than values measured in the adjacent subtropical gyres. Generally, these meridional trends in δ¹³C are found in the separate data sets from the 1980s, 1990s and 2000s (Figure 10). The meridional trend in surface ocean δ¹³C results from the interaction of three processes, i.e., air-sea CO₂ exchange, export of organic carbon and advection/mixing. The export of organic carbon increases the δ¹³C of DIC because the organic carbon typically has a δ¹³C of about −20‰ [Goericke and Fry, 1994]. The effect on δ¹³C of air-sea CO₂ exchange depends on temperature. In cold waters (<∼10°C), air-sea CO₂ exchange increases the δ¹³C, whereas at warmer temperatures it decreases the δ¹³C because of the temperature dependence of the equilibrium isotope fractionation effect between the bicarbonate and carbonate ions and dissolved CO₂ gas [Zhang et al., 1995]. The impact of air-sea CO₂ exchange depends on the residence time of the water in the surface layer because of the ∼10 year isotopic air-sea equilibration time for the δ¹³C of DIC. Upwelling and vertical mixing brings δ¹³C depleted subsurface waters into the mixed layer and, thus, decreases the δ¹³C. The low δ¹³C values in the Southern Ocean indicate that upwelling of δ¹³C depleted subsurface waters dominates over the δ¹³C enriching effects of export of organic carbon and air-sea CO₂ gas exchange. However, by about 50°S the δ¹³C enriching effects of organic carbon export and air-sea CO₂ exchange overcome the δ¹³C depletion caused by upwelling and vertical mixing to yield high δ¹³C values. The low δ¹³C values in the subtropical gyres are primarily the result of long surface water residence times and air-sea CO₂ exchange at warm temperatures. Despite equatorial upwelling, the δ¹³C near the equator is higher than in the adjacent subtropical gyres because the short surface residence time of the upwelled water limits the impact of air-sea CO₂ exchange and organic carbon export rates are high. The northward increase in δ¹³C into the subpolar N Atlantic implies that the δ¹³C enrichments caused by air-sea CO₂ exchange (at cold temperatures) and organic carbon export overcome the δ¹³C depleting effects of deep winter mixing.

Figure 10

(a) The δ¹³C (‰) in surface waters of the Atlantic Ocean measured since 1981 using 22 research and VOS cruises (see Figure 1 for locations). Generally, cruises from the 1980s are in red, from the 1990s in green, and from 2000s in blue. (b) The mean δ¹³C (±1SE) over 5° latitude bands in the mid-1980s (diamonds), 1990s (squares), and 2000s (triangles).

[28] In general, the observed δ¹³C decrease in the Atlantic Ocean agrees well with previous estimates of the δ¹³C decrease rate in the surface waters of the Atlantic, Pacific and Indian oceans (Figure 11) [Gruber et al., 1999; Sonnerup et al., 2000; McNeil et al., 2001; Körtzinger et al., 2003; Quay et al., 2003]. In the subtropical Atlantic (20°N–40°N and S), the δ¹³C has decreased at a mean rate of −0.023 ± 0.003‰ yr⁻¹, which is similar to the −0.025‰ yr⁻¹ resulting from application of the MLR-based method to the δ¹³C changes between A16N versus OACES and A20/22N versus TTNA in the North Atlantic, as discussed above. In the tropical Atlantic (20°S–20°N), the δ¹³C decrease rate was similar at −0.020 ± 0.003‰ yr⁻¹. In the subpolar (40°N–65°N) North Atlantic, the estimated δ¹³C decrease rate at −0.015 ± 0.006‰ yr⁻¹ is slightly lower but with greater uncertainty due to greater spatial and seasonal δ¹³C variability. In this region, a better estimate of the δ¹³C decrease rate may result from the time rate of change in preformed δ¹³C along isopycnals that yield a mean rate of −0.022 ± 0.003‰ yr⁻¹ since these rates integrated over longer time and larger space scales [Sonnerup et al., 1999; Körtzinger et al., 2003]. Between 40° and 50°S in the Atlantic, there is a transition from subtropical δ¹³C decrease rates of ∼−0.025‰ yr⁻¹ to a negligible δ¹³C change of ∼ −0.001 ± 0.010‰ yr⁻¹. In the South Pacific, both Sonnerup et al. [1999] and McNeil et al. [2001] observed a similar sharp decline in the δ¹³C decrease rate from −0.015 to −0.005‰ yr⁻¹ across the subantarctic front (SAF) (Figure 11). We expect that the transition zone for surface δ¹³C change associated with the SAF would occur further north in the South Atlantic versus South Pacific because of the more equatorward location of the SAF (at ∼ 45°S) in the South Atlantic [Orsi et al., 1995]. South of 50°S in the Atlantic there appears to be δ¹³C increase (Figure 10), however the δ¹³C data is sparse (e.g., only 17 samples collected prior to 2000 and none collected in the 1990s) and collected from different regions in the 1980s versus 2000s (e.g., eastern versus western portion of the South Atlantic basin). With the available data, we cannot determine a representative δ¹³C change in the surface ocean south of 50°S.

Figure 11

The surface ocean δ¹³C change rate (‰ decade⁻¹) in the Atlantic, Pacific, and Indian Oceans estimated from: MLR-based estimates [Sonnerup et al., 2000; McNeil et al., 2001] (also this work), direct cruise δ¹³C comparisons [Gruber et al., 1999; Quay et al., 2003], δ¹³C measurements at three time series stations ALOHA [Quay et al., 2003], BATS [Gruber et al., 2002] and KNOT [Tanaka et al., 2003] and calculations of preformed δ¹³C time trends along isopycnals [Sonnerup et al., 1999; Körtzinger et al., 2003].

[29] The estimated area weighted basin-wide mean δ¹³C decrease rate in the Atlantic Ocean was −0.018 ± 0.002‰ yr⁻¹ between 50°S and 65°N over the last two decades. In comparison, basin-wide surface ocean δ¹³C decrease rates of −0.018‰ yr⁻¹ for the Pacific Ocean between 1970 and 1993 [Quay et al., 2003] and −0.014‰ decade⁻¹ for the Indian Ocean between 1978 and 1995 [Sonnerup et al., 2000]) have been determined previously.

3.3. Air-Sea δ¹³C Disequilibrium

[30] The surface ocean δ¹³C currently is not in equilibrium with the δ¹³C of atmospheric CO₂. The 10-year air-sea equilibration time for the δ¹³C of DIC causes the mean δ¹³C change in the surface ocean to lag the atmospheric δ¹³C change and is slow enough to allow other processes, like mixing and biological production of organic carbon, to keep the observed surface δ¹³C from fully equilibrating with atmospheric CO₂ on regional spatial scales. The difference between the measured δ¹³C in the surface ocean and that expected at equilibrium with the atmosphere is referred to as the air-sea δ¹³C disequilibrium [Tans et al., 1993] and equals the δ¹³C of the dissolved CO₂ gas minus the δ¹³C of the dissolved CO₂ gas were it isotopically equilibrated with atmospheric CO₂. The δ¹³C of dissolved CO₂ gas is calculated from the measured δ¹³C of the DIC, mean annual mean sea surface temperature (SST) [Levitus and Boyer, 1994] and the empirically determined temperature dependence of the equilibrium isotopic fractionation between the DIC and CO₂ gas (α_DIC-g) [Zhang et al., 1995].

[31] In all basins, the measured δ¹³C of warm (>10°C) surface waters exceeds the atmospheric equilibrated δ¹³C value and of cold surface waters is lower than the atmospheric equilibrated δ¹³C value because of the substantial temperature dependence of α_DIC-g [Quay et al., 2003]. The area-weighted mean air-sea δ¹³C disequilibrium for the North Atlantic (0° to 60°N) was 1.09, 1.11 and 1.25‰ for the 1980s (n = 94), 1990s (n = 267) and 2000s (n = 285), respectively, based on the surface δ¹³C data shown in Figure 10. The area-weighted and air-sea CO₂ flux weighted mean δ¹³C disequilibrium values were 0.62, 0.62 and 0.83‰ in the 1980s, 1990s and 2000s, respectively, where the air-sea CO₂ flux was determined using the relationship between wind speed and gas transfer rates by Wanninkhof [1992] and climatological NCEP winds and CO₂ solubility [Weiss, 1974] based on climatological mean SST. The error in the basin-wide δ¹³C disequilibrium and has been estimated at ±0.08‰, as discussed below, and depends on the uncertainties in the surface δ¹³C, α_DIC-g, SST, and atmospheric δ¹³C. The accuracy of the estimate is better for the 1990s and 2000s because of the larger surface ocean δ¹³C data sets. The magnitude of the ocean-wide air-sea δ¹³C disequilibrium is growing since the δ¹³C of atmospheric CO₂ is decreasing at a faster rate (∼−0.027‰ yr⁻¹) than the δ¹³C in the surface ocean (∼−0.016‰ yr⁻¹). Thus the global air-sea disequilibrium should be increasing by about 0.1‰ per decade, assuming wind speeds, SST, organic carbon export and circulation have remain constant.

4. Anthropogenic CO₂ Accumulation Rates

[32] To estimate the anthropogenic CO₂ accumulation rate in the North Atlantic, we have to convert the basin-wide depth-integrated δ¹³C change of −15.0 ± 3.8‰ m yr⁻¹ to an integrated DIC change. This conversion is accomplished by dividing the depth-integrated δ¹³C change by RC, which has been defined as the ratio of the anthropogenic δ¹³C change to the anthropogenic CO₂ change [McNeil et al., 2001]. The range of RC in the ocean is significant, that is, from −0.005‰ (μmol kg⁻¹)⁻¹ in the Southern Ocean [McNeil et al., 2001] to −0.032‰ (μmol kg⁻¹)⁻¹ if the ocean fully equilibrated with the atmospheric δ¹³C and CO₂ rates of change [Körtzinger et al., 2003]. The magnitude of RC depends on the air-sea CO₂ gas exchange rate and exposure time of the surface waters before subduction. Because the air-sea equilibration time for δ¹³C, at ∼10 years, is 10 times longer than for CO₂, a water parcel that remains at the surface for a longer time will allow for greater δ¹³C equilibration and have a higher RC value. For the North Atlantic, RC has been determined previously by Körtzinger et al. [2003] on the basis of the ratio of preformed DIC and δ¹³C changes along isopycnals. They found a very consistent value of RC = −0.024 ± 0.003‰ (μmol kg⁻¹)⁻¹ for isopycnals between 26.6 and 27.4 that extend to a depth of ∼1000 m and represent ventilation times of 20 to 50 years.

[33] An anthropogenic CO₂ accumulation rate of 0.63 ± 0.16 mol C m⁻² yr⁻¹ for the North Atlantic Ocean over the last 20 years is calculated by dividing our estimate of the depth-integrated δ¹³C change by RC, i.e., −15.0 ± 3.8‰ m yr⁻¹/−0.024 ± 0.003‰ (μmol kg⁻¹)⁻¹. Basin-wide this corresponds to 0.32 ± 0.08 Pg C yr⁻¹ of anthropogenic CO₂ accumulation in the North Atlantic Ocean during the last 20 years.

4.1. Anthropogenic CO₂ Sources: Air-Sea Exchange Versus Meridional Circulation

[34] To estimate the importance of air-sea ¹³CO₂ exchange as the source of the anthropogenic δ¹³C changes observed in the North Atlantic, we compare the rate of depth-integrated δ¹³C change to the air-sea ¹³CO₂ flux. The flux of ¹³CO₂ (F₁₃) across the air-sea interface can be described as follows:

where k is the gas transfer (or piston) velocity, K_H is the solubility of CO₂ in seawater [Weiss, 1974], and pCO₂^g and pCO₂^sea are the partial pressures of CO₂ in air and seawater, respectively. The carbon isotope fractionation factors are as follows: α_k is the kinetic fractionation factor for CO₂ diffusion through the boundary layer, α_aq−g is the equilibrium fractionation factor for CO₂ gas dissolution, and α_DIC−g is the equilibrium fractionation factor between seawater DIC and gaseous CO₂ [Zhang et al., 1995]. and R_DIC represent the measured ¹³C/¹²C ratios of CO₂ in air and seawater DIC, respectively, where R = (δ¹³C/1000 + 1)*R_PDB. We assume ¹²CO₂ adequately represents (98.9%) bulk CO₂.

[35] The impact of air-sea ¹³CO₂ flux on the ¹³C/¹²C of the DIC in the ocean can be separated into two components. First, there is the effect of gross air-sea CO₂ flux times the air-sea δ¹³C disequilibrium and, second, there is the net addition of CO₂ with the δ¹³C of the dissolved CO₂ gas. These effects can be quantified by rewriting equation (2) as follows:

where R_dis = and ΔpCO₂ = pCO₂^g − pCO₂^sea. The net CO₂ gas flux (F₁₂) is:

In a simple 1-D vertical ocean, the changes in DIC and DI¹³C inventories equal the air-sea CO₂ and ¹³CO₂ fluxes, respectively, and can be expressed as follows:

where ΔDIC = (DIC_t − DIC_o) and ΔDI¹³C = DIC_t (R_t − R_o) + R_o (DIC_t − DIC_o). We can solve equations (3) and (4) for the ΔpCO₂ using the estimate of ΔDIC/Δt = 0.63 mol m⁻² yr⁻¹ and ΔDI¹³C/Δt = 0.006679 mol m⁻² yr⁻¹ (calculated from our estimates of Δδ¹³C/Δt = −15.0‰ m yr⁻¹, ΔDIC/Δt = 0.63 mol m⁻² yr⁻¹, DIC_t = 2100 μmol kg⁻¹, and δ¹³C_o = 1.5‰), to substitute for F₁₂ and F₁₃, respectively, and an air-sea disequilibrium of 0.71‰ (as discussed above), as follows:

[36] We determined an average value for ΔpCO₂ = 5.4 μatm for the North Atlantic from equation (7) that, notably, does not depend on the gas transfer rate. A value of the gas transfer rate (k) of 9.7 m d⁻¹ would be needed, using equation (4), to yield a ΔDIC/Δt = 0.63 mol m⁻² yr⁻¹ with a value of ΔpCO₂ = 5.4 μatm. This gas transfer rate is double the value of 4.6 m d⁻¹ predicted using climatological NCEP winds and relationship between wind speed and k determined by Wanninkhof [1992]. In other words, a rate of gross air-sea CO₂ exchange of 50 mol m⁻² yr⁻¹, which is more than twice the expected rate of 20 mol m⁻² yr⁻¹, is required to obtain the observed inventory changes in the δ¹³C of the DIC of −15.0‰ m yr⁻¹ given an air-sea δ¹³C disequilibrium of 0.7‰.

[37] The proportion of anthropogenic DI¹³C accumulation in the North Atlantic explained by air-sea ¹³CO₂ exchange can be estimated from the ratio of the air-sea ¹³CO₂ flux to the anthropogenic DIC¹³ inventory change. The air-sea ¹³CO₂ flux approximately equals the gross air-sea CO₂ flux (20 ± 6 mol m⁻² yr⁻¹) times the air-sea δ¹³C disequilibrium (0.71 ± 0.08‰). The DI¹³C inventory change equals the depth-integrated δ¹³C change (−15 ± 3.8‰ m yr⁻¹) times the DIC concentration (2.1 ± 0.25 moles m⁻³). Thus air-sea CO₂ exchange can account for ∼45 ± 19% of the anthropogenic DI¹³C accumulation in the North Atlantic. Applying this proportion to anthropogenic CO₂ implies that the air-sea CO₂ uptake rate is only 0.28 ± 0.07 mol m⁻² yr⁻¹ in the North Atlantic, which is only ∼60% of the global ocean average of ∼0.45 mol m⁻² yr⁻¹ (corresponding to 2 Pg C yr⁻¹). If air-sea CO₂ exchange can explain only half of the observed anthropogenic CO₂ inventory change in the North Atlantic, what other process comes into play?

[38] Another potentially important source of anthropogenic CO₂ and ¹³CO₂ input into the North Atlantic results from the advective input of surface waters from the South Atlantic. The formation of North Atlantic Deep Water (NADW) results in northward transport of surface or near surface currents that contain anthropogenic CO₂ and ¹³CO₂, whereas the NADW exported southward at 2000–3000 m at the equator contains little or no anthropogenic CO₂ [e.g., Gruber, 1998]. There have been several estimates of the magnitude of net northward input of anthropogenic CO₂ to the North Atlantic. MacDonald et al. [2003] estimated a flux of 0.20 ± 0.08 Pg C yr⁻¹ across 24°N in 1998 (corresponding to 0.70 ± 0.28 mol m⁻² yr⁻¹), based on an inverse model of geostrophic and Ekman transport of anthropogenic DIC, which implies virtually no air-sea anthropogenic CO₂ uptake north of 24°N in the North Atlantic. Recently, Mikaloff Fletcher et al. [2006] used multiple GCM inversions to estimate an anthropogenic CO₂ budget for the North Atlantic that had an accumulation rate of 0.92 mols m⁻² yr⁻¹ (0.48 Pg C yr⁻¹), a northward DIC transport rate across the equator of 0.33 mols m⁻² yr⁻¹ (0.17 Pg C yr⁻¹) and an air-sea CO₂ uptake rate of 0.59 mols m⁻² yr⁻¹ (0.31 Pg C yr⁻¹). Our δ¹³C-based estimate that air-sea CO₂ uptake accounts for 0.28 ± 0.07 mols m⁻² yr⁻¹ (0.14 ± 0.06 Pg C yr⁻¹) of the 0.63 ± 0.16 mols m⁻² yr⁻¹ of accumulating anthropogenic CO₂ (0.32 ± 0.08 Pg C yr⁻¹) implies a northward advection of 0.35 ± 0.17 mols m⁻² yr⁻¹ (0.18 ± 0.10 Pg C yr⁻¹), which agrees reasonably well with previous estimates. This northward anthropogenic CO₂ would carry with it an anthropogenic δ¹³C input of −8.4‰ m yr⁻¹, assuming these waters have an RC = −0.024‰ (umol kg⁻¹)⁻¹, which would be a significant portion of the observed depth-integrated δ¹³C change of −15.1 ± 3.8‰ m yr⁻¹. Thus the meridional overturning circulation is likely an important source of the buildup of anthropogenic CO₂ and ¹³CO₂ in the North Atlantic and comparable to the uptake of CO₂ by air-sea exchange. The advective input of surface waters carrying substantial amounts of anthropogenic CO₂, with a high RC value due to long air-sea equilibration times, helps explain why the rate of depth-integrated δ¹³C change we observed in the North Atlantic since the 1980s is about twice the rate we observe for the rest of the ocean.

4.2. Error Analysis

[39] The uncertainty in the calculated rate of anthropogenic CO₂ accumulation in the North Atlantic depends on the error associated with two terms, that is, the depth-integrated δ¹³C change and the ratio of anthropogenic δ¹³C to DIC change (RC). The uncertainty in the basin wide depth-integrated δ¹³C change of −15 ± 3.8‰ m yr⁻¹ was discussed above. The value and uncertainty for RC at −0.024 ± 0.003‰ (μmole kg⁻¹)⁻¹ was taken from Körtzinger et al. [2003]. Although acknowledging that RC varies substantially between ocean basins, from −0.005‰ (μmole kg⁻¹)⁻¹ in the Southern Ocean [McNeil et al., 2001] to −0.024‰ (μmole kg⁻¹)⁻¹ in the North Atlantic [Körtzinger et al., 2003], the Körtzinger et al. estimate of RC is most appropriate for our study site since it was determined in the same geographic region of the North Atlantic, over the same depth interval (0–1000 m) as our δ¹³C measurements and over the same time period of the last 20 to 40 years. As a result, the basin-wide anthropogenic CO₂ accumulation rate and uncertainty is determined to be 0.63 ± 0.16 moles m⁻² yr⁻¹ based on propagating the uncertainties in the depth-integrated δ¹³C change and RC.

[40] The δ¹³C-derived anthropogenic CO₂ accumulation was separated into air-sea CO₂ uptake and anthropogenic CO₂ transport based on a comparison of the air-sea ¹³CO₂ flux to the DI¹³C inventory change. The uncertainty in the air-sea CO₂ uptake calculation depends on the uncertainties in the air-sea ¹³CO₂ flux and DI¹³C inventory change. Since the air-sea ¹³CO₂ flux is approximately equal to the gross air-sea CO₂ flux times the air-sea δ¹³C disequilibrium (see equation (3)), the uncertainties in these two terms determine the error in the air-sea ¹³CO₂ flux. We estimated a mean basin-wide gross air-sea CO₂ flux rate of 20 moles m⁻² yr⁻¹ using the spatially weighted (for 5° latitude bands) wind speeds (NCEP), the relationship between gas transfer rate and wind speed of Wanninkhof [1992] and assuming the surface ocean was in equilibrium with atmospheric pCO₂ at the mean sea surface temperature (with a skin cooling effect of 0.3°C). The uncertainty in the air-sea gross CO₂ flux rate was assumed to be ±30%. We estimated the uncertainty in the air-sea δ¹³C disequilibrium (over 5° latitude bands) following the procedure of Quay et al. [2003] by propagating the ±1SD errors in the surface δ¹³C (±0.13‰), atmospheric δ¹³C (±0.03‰) and air-sea δ¹³C solubility and diffusion fractionation effects from both empirical uncertainties (±0.05‰) from Zhang et al. [1995] and as a result of variability in sea surface temperature (±1.5°C) from Levitus and Boyer [1994]. These errors combine to yield an estimated uncertainty of ±0.2‰ for the δ¹³C disequilibrium. We determined an uncertainty of ±0.08‰ for the basin-wide average δ¹³C disequilibrium using a Monte Carlo approach, as described by Quay et al. [2003], by randomly selecting δ¹³C surface ocean values, δ¹³C atmospheric values, fractionation effects, SSTs and air-sea CO₂ transfer rates (for weighting purposes) on the basis of their mean and ±1SD values for 5° latitude bands. The δ¹³C disequilibrium calculation was repeated 300 times using these randomly selected parameters sets and then the mean and SD of the 300 disequilibrium values were calculated. Although we did not attempt to determine any bias in the calculated air-sea δ¹³C disequilibrium, the similarity of the mean values of 0.62‰, 0.62‰ and 0.82‰ calculated using δ¹³C data from the 1980s, 1990s and 2000s respectively, suggests that spatial biases are likely not in excess of the uncertainty. Thus an uncertainty of ±4.5 mols ‰ m⁻² yr⁻¹ for the air-sea ¹³CO₂ flux results from propagating the errors in the air-sea CO₂ flux (20 ± 6 mols m⁻² yr⁻¹) and the δ¹³C disequilibrium (0.71 ± 0.08‰).

[41] The proportion of the observed DI¹³C inventory change that resulted from air-sea ¹³CO₂ exchange equals the ratio of the air-sea ¹³CO₂ flux divided by the DI¹³C inventory change. The uncertainty in this proportion is determined by propagating the errors in each term. The DI¹³C inventory change equals the depth-integrated δ¹³C change (−15 ± 3.8‰ m yr⁻¹) times the mean DIC concentration over 1000 m (2.1 ± 0.25 moles m⁻³), which yields a value of −31.5 ± 8.8 mol ‰ m⁻² yr⁻¹. Thus the proportion of the inventory change that is a result of air-sea CO₂ exchange is 0.45 ± 0.19 (i.e., −14.2 ± 4.5 mols ‰ m⁻² yr⁻¹/−31.5 ± 8.8 mol ‰ m⁻² yr⁻¹).

4.3. North Atlantic Versus North Pacific Oceans

[42] Application of the MLR method to DIC concentration changes measured on cruises in the North Pacific Ocean during GEOSECS (1970s), NOAA (1990s) and WOCE (1990s) programs yield estimated anthropogenic CO₂ accumulation rates of 1.1 to 1.3 mol C m⁻² yr⁻¹ [Feely et al., 2003; Peng et al., 2003; Sabine et al., 2004], which is twice the rate of 0.63 mol C m⁻² yr⁻¹ we estimate for the North Atlantic. However, CFC-based estimates of anthropogenic CO₂ accumulation in the North Pacific are significantly lower at 0.4 to 0.5 mol C m⁻² yr⁻¹ [Watanabe et al., 2000; McNeil et al., 2003] and more in line with our estimates for the North Atlantic. It is difficult to reconcile an anthropogenic CO₂ accumulation rate for the North Pacific that is twice the rate in the North Atlantic, given the apparent importance of meridional circulation as a significant source of anthropogenic CO₂ build up in the North Atlantic and the absence of this circulation in the North Pacific. Our results imply that only about half of the anthropogenic CO₂ inventory change in the North Atlantic resulted from air-sea CO₂ exchange (i.e., ∼0.3 mol C m⁻² yr⁻¹), with the remainder resulting from northward advection of anthropogenic CO₂ laden surface waters needed to supply sinking NADW. Since there is no equivalent meridional circulation cell in the North Pacific, an anthropogenic CO₂ accumulation rate of 1.2 mol C m⁻² yr⁻¹ resulting from only CO₂ gas exchange would imply an air-sea CO₂ flux that is ∼4× greater in the North Pacific than the North Atlantic.

[43] One issue to consider carefully during these comparisons is how well the data represent the geographic region. For example, data sets weighted toward the subtropics will yield substantially higher CO₂ accumulation rates than data sets weighted toward the subpolar regions in the North Pacific [Watanabe et al., 2000]. This would not be the case in the North Atlantic, where deep penetration of the anthropogenic CO₂ into the deep sea in the subpolar regions [Friis et al., 2005] can yield high inventory changes (as we observed for δ¹³C). Thus the data coverage needs to be fairly uniform throughout the basin to be comparable. Estimates of the DIC and δ¹³C changes based on a comparison of distributions measured during WOCE (1990s) with those currently being measured during the CLIVAR Repeat Hydrography program should help resolve the apparent differences between CO₂ accumulation rates in the North Pacific and North Atlantic. At this time, the apparent 2x difference between MLR-derived anthropogenic CO₂ accumulation rates in the North Pacific and North Atlantic should be viewed with caution. If true, the mechanisms yielding such high rates of CO₂ accumulation in the North Pacific and/or low rates in the North Atlantic, for example, like the slow down in subpolar North Atlantic air-sea CO₂ uptake rates proposed by Lefèvre et al. [2004], must be identified.

5. Summary

[44] The δ¹³C of dissolved inorganic carbon in the North Atlantic has decreased significantly over the last two decades as a result of the accumulation of anthropogenic CO₂. On average, the surface ocean δ¹³C decrease was −0.018‰ yr⁻¹ and the depth-integrated δ¹³C decrease was −15.0 ± 3.8‰ m yr⁻¹ in the North Atlantic. The anthropogenic δ¹³C perturbation extends into the deep sea in the subpolar latitudes, to ∼1000 m in the subtropics and <400 m in tropics. Although a multiple linear regression was used to determine these anthropogenic δ¹³C decrease patterns, a comparison of the measured δ¹³C changes between cruises in 1981, 1993 and 2003 shows similar results. The anthropogenic δ¹³C decrease is a robust, not subtle, signal easily measured over a decade.

[45] The rate of anthropogenic CO₂ accumulation in the North Atlantic was estimated at 0.63 ± 0.16 mol m⁻² yr⁻¹ based on the depth-integrated change of −15.0 ± 3.8‰ m yr⁻¹ and the ratio of anthropogenic δ¹³C to CO₂ change of −0.024 ± 0.003‰ (umol/kg⁻¹)⁻¹ previously determined by Körtzinger et al. [2003]. Integrated over the North Atlantic this yields a net CO₂ accumulation rate of 0.32 ± 0.08 Pg C yr⁻¹ that represents ∼15% of the global ocean CO₂ accumulation rate. Comparing the depth-integrated δ¹³C change with the air-sea δ¹³C disequilibrium of 0.7‰ implies that air-sea CO₂ exchange explains only half of the depth-integrated δ¹³C change in the North Atlantic since the 1980s. Another significant source of anthropogenic CO₂ and δ¹³C input into the North Atlantic Ocean is via the meridional circulation associated with the formation of North Atlantic Deep Water which essentially exports anthropogenic CO₂ free water at depth while replacing it with anthropogenic CO₂ laden surface waters. A northward transport of anthropogenic CO₂ of 0.35 ± 0.17 mols m⁻² yr⁻¹ (0.18 ± 0.10 Pg C yr⁻¹) would account for the difference between the rate anthropogenic DI¹³C accumulation and air-sea ¹³CO₂ flux in the North Atlantic and is in good agreement with previous estimates. This northward advection of anthropogenic CO₂ laden surface waters makes a major contribution to the observed anthropogenic δ¹³C changes in the basin and likely explains why the δ¹³C inventory changes in the North Atlantic are twice the global ocean average. Thus these δ¹³C results indicate that the North Atlantic is a basin with a higher than average anthropogenic CO₂ accumulation rate but lower than average CO₂ uptake rate.

Acknowledgments

[46] We thank: Niki Gruber, John Bullister, Chris Sabine, and Ruth Curry for collecting seawater samples for us during the Repeat Hydrography cruises, Patricia Frickers and Nick Pope for sample collection during the AMT cruises, and Aida Rios and Colm Sweeney for sample collection during the Ficaram and Gould cruises, respectively. We especially appreciated the efforts of the many UW undergraduate students who have helped with both sample collection and sample preparation in our lab. This study was supported in part by the UK Natural Environment Research Council through the Atlantic Meridional Transect consortium (NER/O/S/2001/00680) and core strategic research funding to Plymouth Marine Laboratory. This is contribution 145 of the AMT program. In particular we want to acknowledge the financial support that NOAA's Office of Global Programs has provided, most recently under the Joint Institute for the Study of Atmosphere and Oceans (JISAO) under NOAA Cooperative Agreement NA67RJ0155.