Interannual Variability in the North Atlantic Ocean Carbon Sink

See allHide authors and affiliations

Science  20 Dec 2002:
Vol. 298, Issue 5602, pp. 2374-2378
DOI: 10.1126/science.1077077


The North Atlantic is believed to represent the largest ocean sink for atmospheric carbon dioxide in the Northern Hemisphere, yet little is known about its temporal variability. We report an 18-year time series of upper-ocean inorganic carbon observations from the northwestern subtropical North Atlantic near Bermuda that indicates substantial variability in this sink. We deduce that the carbon variability at this site is largely driven by variations in winter mixed-layer depths and by sea surface temperature anomalies. Because these variations tend to occur in a basinwide coordinated pattern associated with the North Atlantic Oscillation, it is plausible that the entire North Atlantic Ocean may vary in concert, resulting in a variability of the strength of the North Atlantic carbon sink of about ±0.3 petagrams of carbon per year (1 petagram = 1015grams) or nearly ±50%. This extrapolation is supported by basin-wide estimates from atmospheric carbon dioxide inversions.

The ocean's contribution to the observed interannual variability of atmospheric carbon dioxide (CO2) is poorly established. Estimates based on atmospheric measurements of CO2, oxygen, and stable carbon isotopes indicate that the variability contributed by the oceanic carbon cycle is more than ±1 Pg C year−1(1–4). In contrast, estimates based on direct observations of the partial pressure of CO2(pCO2) in surface waters (5,6) and on modeling studies (7,8) indicate a contribution of less than ±0.5 Pg C year−1, mainly associated with tropical Pacific ocean variability caused by El Niño and La Niña (9). However, many uncertainties are associated with the modeling studies, and the equatorial Pacific is the only region where interannual variability in oceanic pCO2 has been directly observed and documented. Given evidence for substantial extratropical variability in sea surface temperature (SST) (10) and the ocean's state (11), other oceanic regions may contribute substantially to the atmospheric CO2 variability as well. The North Atlantic Ocean is one of the few regions where enough data are available to investigate interannual to decadal variability in the extratropical ocean carbon cycle. Observationally based estimates (12), as well as forward and inverse modeling results (13), indicate that this region constitutes the largest ocean sink for atmospheric CO2 in the Northern Hemisphere, on average taking up about 0.7 ± 0.1 Pg C year−1.

Observations have shown that most of the interannual to decadal climatic variability in the North Atlantic basin occurs in broadly coherent patterns linked to a natural mode of atmospheric pressure variation known as the North Atlantic Oscillation (NAO) (14). The NAO is a large-scale seesaw in atmospheric mass between a subtropical high-pressure system, typically near the Azores, and a subpolar low near Iceland (15). In the Atlantic Ocean, a positive phase of the NAO, i.e., a stronger pressure gradient between these two systems, is expressed by positive SST anomalies in the subtropics and the marginal seas of northern Europe and negative anomalies in the subpolar gyres (16,17). These anomalies are accompanied by changes in oceanic convection, in which regions of warmer SSTs experience less vigorous winter mixing than normal, whereas regions of cold SSTs experience deeper winter mixing (18).

We examined interannual variability in the carbon cycle of the upper layers of the North Atlantic Ocean, and its connection with the NAO, on the basis of an 18-year time series of dissolved inorganic carbon (DIC), total alkalinity (Alk), and the 13C/12C ratio of DIC obtained from near-surface waters at two stations near Bermuda (19) (Fig. 1). The Bermuda time series shows distinct seasonal oscillations for salinity-normalized DIC (sDIC), its 13C/12C ratio [expressed as δ13C (20)], and computed pCO2(21), but it exhibits only occasional short-term changes in salinity-normalized Alk (sAlk) (Fig. 1A). sDIC, δ13C, and pCO2 show distinct long-term trends related to the uptake of isotopically depleted anthropogenic CO2 from the atmosphere (22–24). The observed long-term rate of increase of sDIC over the period from October 1983 to June 2001 is 0.64 ± 0.05 μmol kg−1 year−1(25). The corresponding rate of increase forpCO2 is found to be 1.5 ± 0.1 parts per million (ppm) year−1, very close to the observed atmospheric trend over the same period and consistent with the expectation that subtropical near-surface waters mix too slowly with deeper waters to be substantially displaced from equilibrium with anthropogenically perturbed atmospheric CO2. The long-term decrease in δ13C is –0.024 ± 0.001 per mil year−1, close to the trends observed previously in subtropical regions (23, 26).

Figure 1

Time series of properties in the upper-ocean mixed layer of the northwestern Sargasso Sea near Bermuda (Station S at 32°10′N, 64°30′W and the BATS site at 31°50′N, 64°10′W). (A) Observations of inorganic carbon. The reduced isotopic 13C/12C ratio of DIC, δ13C, is shown inverted to make the seasonal and long-term relation with DIC more evident. The concentrations of DIC and Alk (sDIC and sAlk) have been normalized to a constant salinity of 35.pCO2 has been computed from DIC, Alk, SST, and sea-surface salinity and is compared with the partial pressure of atmospheric CO2. The latter has been calculated with the mean atmospheric pressure and humidity observed at St. Davids Head on the island of Bermuda and observations of the atmospheric CO2 mixing ratio [with data from La Jolla, CA (1), for 1982 to 1995 and from Bermuda (46) for 1989 to 2000]. Data labeled CDRG (squares and diamonds) were analyzed at the Scripps Institution of Oceanography; data labeled BBSR (triangles) were measured at Bermuda Biological Station for Research. (B) Seasonally adjusted and long-term linearly detrended time series of SST, δ13C, sDIC, andpCO2. Smoothing splines (cut-off period of 1 year) have been added to emphasize trends. The sDIC anomalies have been inverted to emphasize the interannual correlation with temperature anomalies (δ13C is not inverted).

To highlight interannual variability, we removed the mean seasonal cycle and the long-term trend by subtracting a linear trend and harmonics with 12-, 6-, and 4-month periods from the data. The resulting anomalies revealed significant interannual variability in sDIC and in its isotopic ratio δ13C that are well correlated with anomalies in SST and winter mixed-layer depth (Fig. 1B and table S1). By contrast, the corresponding variability ofpCO2 was relatively low, amounting to only a few ppm. This low variability was a consequence of compensating effects between anticorrelated interannual SST and sDIC anomalies (24). Nevertheless, the forcing ofpCO2 by SST anomalies is stronger than by the sDIC anomalies, as previously noted with regard to the seasonal cycle (27), which leads to a positive correlation betweenpCO2 and SST anomalies (table S1). SST and mixed-layer depth anomalies near Bermuda have been shown to be significantly correlated with the NAO (24, 28), with warmer-than-normal conditions and reduced wintertime mixing during positive phases of the NAO. Our analysis of the 1983 to 2001 data is consistent with these previous analyses. A division of the correlation analysis into three periods (December through March, April through July, and August through November) shows that the strongest correlations are found in winter, when the NAO pattern is most strongly expressed in the atmosphere (14).

We also found a strong correlation of the NAO with sDIC anomalies. In contrast, the correlations for δ13C andpCO2 were weak. Both results agree with the findings of Bates (24), who analyzed the Bermuda Biological Station for Research (BBSR) record for the Bermuda Atlantic Timeseries Study (BATS) from 1988 to 1998.

Our aim was to elucidate quantitatively the processes that influence the surface oceanic carbon cycle (such as air-sea gas exchange, biology, mixing, and transport) and to determine how these processes interact to cause the observed correlations. To this end, we used a simple diagnostic box model of the upper-ocean carbon cycle (29) that exploits the previous finding that the dominant control of seasonal variations in δ13C in near-surface water DIC is net community production (30). We combined an estimate of the isotopic fractionation associated with net community production with estimates of air-sea gas exchange, vertical entrainment, and vertical diffusion, which permitted us to estimate the magnitude and variability of net community production and horizontal transport, the only two processes that cannot be estimated on the basis of simple parameterizations (31).

The mean seasonal fluxes for the 1983 to 2001 period derived from this diagnostic approach are very similar to those found for the 1991 to 1994 period (29), which indicates relatively stable seasonal behavior. These results demonstrate that the seasonal dynamics of mixed-layer DIC near Bermuda is dominated by the net removal of DIC in the spring and summer as a result of net community production. This biologically induced drawdown of DIC is aided in summer by the outgassing of CO2 due to the winter-to-summer warming of the surface ocean. The seasonal cycle is closed by vertical entrainment and uptake of CO2 from the atmosphere in fall and winter. Vertical diffusion plays a minor role, whereas our calculations indicate a small addition of DIC throughout the year as a result of horizontal transport processes. The magnitude and sign of this inferred horizontal transport are consistent with estimates computed from the horizontal DIC gradient and from the observed horizontal velocities (32). Summarized over the annual cycle, the carbon balance in the mixed layer is established between the net removal of carbon by net community production (3.4 ± 0.8 mol m−2 year−1) (33), the net addition of carbon by gas exchange (1.9 ± 0.2 mol m−2year−1), entrainment (1.6 ± 0.4 mol m−2year−1), diffusion (0.6 ± 0.3 mol m−2year−1), and horizontal transport (1.3 ± 0.4 mol m−2 year−1). The flux budget is closed by a “rectifier” flux term that arises because of the strong covariance of mixed-layer depth and DIC (31).

Interannual variability strongly modifies these annual mean fluxes with variations up to ±50% from their mean values (Fig. 2). These variations strongly correlate with mixed-layer depth variability and SST anomalies (Fig. 2 and table S1) and suggest that these latter two parameters represent key processes controlling interannual variability near Bermuda. Because DIC increases with depth, the flux of inorganic carbon into the mixed layer is increased during winters with deeper convection. As a consequence, a positive DIC anomaly is generated in conjunction with anomalously cool SSTs, which are caused by increased heat loss at the surface and by cooling induced by mixing with deeper waters (16). Our model calculations indicate that these years of cool water tend to have up to 0.8 mol m−2 year−1higher-than-average uptake of CO2 from the atmosphere, a joint result of negative pCO2 anomalies caused by the cooler-than-normal SSTs and of the stronger-than-normal winds. The contribution from horizontal transport also tends to be enhanced but to a significantly smaller degree. Finally, annual net community production is up to 1 mol m−2year−1 higher than average, a finding corroborated by concurrently observed increases in net primary productivity (24). These results are consistent with the argument that, in nutrient-limited systems such as that found near Bermuda, the increase in the entrainment of nutrients should lead to enhanced biological productivity (34).

Figure 2

Twelve-month running mean fluxes of CO2 gas exchange, entrainment, horizontal transport, and net community production as computed by our diagnostic model. All processes are significantly correlated with mixed-layer anomalies shown in the top panel (correlation coefficients, r, are shown on the right ), with years of deeper mixed layers tending to have higher uptake of CO2 from the atmosphere, higher entrainment fluxes, increased horizontal transport, and elevated net community production. The shading indicates a 1-σ uncertainty interval determined from Monte Carlo simulations. DJFM, December through March; MLD, mixed-layer depth; trsp, transport.

Figure 3 summarizes the findings of our analysis of interannual variability of the carbon cycle near Bermuda by contrasting a year with shallow winter mixing and warm SST anomalies with a year with opposite characteristics. The positive DIC anomalies and negative δ13C anomalies, accompanying a cooler-than-average year, are primarily caused by the increased vertical supply of waters from the thermocline, which is high in DIC and low in δ13C, and increased uptake of isotopically light CO2 from the atmosphere. The magnitudes of these anomalies are reduced by an increase in net community production that removes isotopically light carbon from the near-surface ocean water. However, this compensation is incomplete, leading to a positive DIC anomaly, which persists for the entire year. During warmer-than-normal years, the opposite is the case, with net community production compensating less than normal. This interaction between mixed-layer depth variability, air-sea gas exchange, and biological activity not only leads to the generation of DIC and δ13C anomalies but also affects the magnitude of the seasonal cycle. Years of cool water and deep winter mixing tend to show large seasonal cycles, because all processes are acting vigorously; years of warm water and shallow mixing typically show a reduced seasonal cycle.

Figure 3

Schematic depiction of the seasonal carbon cycle near Bermuda, modified in response to interannual variations in mixed-layer depth and SST. In warmer-than-normal years that have shallow winter mixed layers (a typically positive NAO index), little DIC and few nutrients are brought into the upper ocean, which promotes less net community production than normal. In addition, the uptake of CO2 from the atmosphere is reduced because of positive SST anomalies and reduced wind speeds. As a result, the seasonal cycle of sDIC is suppressed. In contrast, colder years exhibit deeper mixed layers with larger entrainment, enhanced net community productivity, and higher CO2 uptake from the atmosphere, which produces an enhanced seasonal cycle of sDIC. SSTA, SST anomalies.

Anomalies in mixing and SST near Bermuda during winter tend to extend over the entire western subtropical gyre (16, 35) and are linked to the NAO (17). Given the importance of mixing and temperature variability in driving interannual variability in the carbon cycle near Bermuda, we would expect that this cycle varies in a concerted manner with Bermuda over the entire subtropical gyre. This might most likely be the case for air-sea gas exchange, which showed a high and significant correlation with the NAO near Bermuda (table S1). If the amplitude of the air-sea CO2 flux anomaly observed near Bermuda (±0.8 mol m−2 year−1) is representative of the entire gyre, the subtropical North Atlantic carbon sink (15°N to 40°N) has varied by up to ±0.2 Pg C year−1 since 1983, enough to contribute measurably to Northern Hemisphere–scale atmospheric CO2 anomalies. We unfortunately lack long-term time-series observations of pCO2 at other places in the North Atlantic subtropical gyre with which to assess our extrapolation. Some supporting evidence comes from deeper ocean DIC observations made at BATS from 1988 to the present (36), which revealed substantial DIC variability in subtropical mode waters. Bateset al. (36) show that these variations most likely reflect NAO-related variations in the net uptake of CO2 from the atmosphere in the regions where these subtropical mode waters are formed, thus representing a record of past variations in air-sea CO2 fluxes in the regions several hundred kilometers north of Bermuda.

Further support for our extrapolation is provided by an inverse modeling analysis of atmospheric CO2 variations (37), which showed a pattern of anomalous CO2 flux for the entire North Atlantic consistent in phasing and magnitude with our Bermuda results (Fig. 4). This suggests that possibly the entire North Atlantic might vary in concert with Bermuda. If we scale the air-sea CO2 flux anomalies from the Bermuda site to the entire North Atlantic, this oceanic sink for atmospheric CO2 may vary on the order of ±0.3 Pg C year−1, or nearly ±50% of its mean strength.

Figure 4

Comparison of the air-sea CO2 flux anomalies estimated for Bermuda from direct observations (solid line) with the air-sea CO2 flux anomalies estimated with an inverse model applied to the entire North Atlantic region (broken line) (37). The model estimate of flux (mol m−2 year−1) is derived from an inversion of atmospheric CO2 data yielding an integrated flux, which then was divided by the area of the North Atlantic between 20°N and 80°N (30 × 1012m2). The broken line represents the mean result of eight atmospheric inversions; gray shading indicates the range of results. The air-sea flux anomalies at Bermuda have been smoothed with a 1-year running average filter.

Although the air-sea CO2 flux anomalies estimated from our long-term record near Bermuda are consistent with atmospheric CO2 inversions, the lack of direct oceanic observations, and the uncertainties associated with atmospheric inversion analysis, make these extrapolations tentative. Time-series observations of the subpolar gyre of the North Atlantic are required to confirm our estimates and to clarify the response of the extratropical carbon cycle to interannual variations in meteorological forcing (8). Understanding the climate sensitivity of the oceanic carbon cycle on these shorter time scales is imperative if we are to predict how the oceanic carbon cycle will respond to future climatic change.

Supporting Online Material

Materials and Methods

Table S1

References and Notes

  • * To whom correspondence should be addressed. E-mail: ngruber{at}


View Abstract

Navigate This Article