The reinvigoration of the Southern Ocean carbon sink

See allHide authors and affiliations

Science  11 Sep 2015:
Vol. 349, Issue 6253, pp. 1221-1224
DOI: 10.1126/science.aab2620

Uptake uptick

Has global warming slowed the uptake of atmospheric CO2 by the Southern Ocean? Landschützer et al. say no (see the Perspective by Fletcher). Previous work suggested that the strength of the Southern Ocean carbon sink fell during the 1990s. This raised concerns that such a decline would exacerbate the rise of atmospheric CO2 and thereby increase global surface air temperatures and ocean acidity. The newer data show that the Southern Ocean carbon sink strengthened again over the past decade, which illustrates the dynamic nature of the process and alleviates some of the anxiety about its earlier weakening trend.

Science, this issue p. 1221; see also p. 1165


Several studies have suggested that the carbon sink in the Southern Ocean—the ocean’s strongest region for the uptake of anthropogenic CO2 —has weakened in recent decades. We demonstrated, on the basis of multidecadal analyses of surface ocean CO2 observations, that this weakening trend stopped around 2002, and by 2012, the Southern Ocean had regained its expected strength based on the growth of atmospheric CO2. All three Southern Ocean sectors have contributed to this reinvigoration of the carbon sink, yet differences in the processes between sectors exist, related to a tendency toward a zonally more asymmetric atmospheric circulation. The large decadal variations in the Southern Ocean carbon sink suggest a rather dynamic ocean carbon cycle that varies more in time than previously recognized.

Simulations with ocean biogeochemical models have suggested a stagnation or even a reduction of the Southern Ocean carbon sink from the 1980s to the early 2000s (13), a result that has been supported by inversion studies (1) based on atmospheric CO2 data. Such a stagnation has wide-reaching implications for climate, because the Southern Ocean south of 35°S accounts for about 40% of the global oceanic uptake of anthropogenic CO2 (46), thereby removing a disproportionally large share of anthropogenic CO2 from the atmosphere. The trend toward a saturation of the Southern Ocean carbon sink has been attributed mainly to the intensification and poleward shift of the westerly winds associated with a trend toward a more positive state of the Southern Annular Mode (1, 2). The resulting enhanced upwelling of deep waters with high concentrations of dissolved inorganic carbon (DIC) drove an anomalously strong flux of natural CO2 out of the surface ocean, counteracting the increase in the oceanic uptake of anthropogenic CO2 (2).

Several studies based on observations of the surface partial pressure of CO2 (pCO2) (79) corroborated these model-based trends in the Southern Ocean carbon sink, but all of them used the observations without any interpolation. Given the sparsity and spatial heterogeneity of these surface ocean observations (8), the conclusions drawn in these studies regarding the trends turn out to be rather sensitive to the chosen method of trend calculation (9) and the beginning and end year of analysis (10). Nevertheless, these studies tended to support a weakening sink trend up to the mid-2000s. One of these studies (9) also pointed out that the trend may have reversed in recent years, a finding corroborated by the analysis of pCO2 observations along a single meridional transect south of Tasmania (11).

To address the sparse data coverage, we used a neural network technique (12) to interpolate the pCO2 observations in time and space. We then evaluated the results using (i) a complementary pCO2 observation–based product based on the interpolation by a data-driven mixed-layer scheme (13) and (ii) an atmospheric CO2 inverse estimate (14). Both surface ocean–based methods were extended for this study to produce multidecadal distributions of the surface ocean pCO2 field (1517). The air-sea CO2 flux variations were then computed with the use of a standard bulk parameterization (supplementary text 1.4). Although each of these estimates faces limits due to the available information, their combination allows us to gain confidence in the inferred features.

The two surface ocean data–based air-sea CO2 flux products confirm that the Southern Ocean carbon sink (south of 35°S) weakened through much of the 1990s, in agreement with the model-based studies and the atmospheric inversions (1, 2), but they reveal that it has strengthened substantially since about 2002, increasing by more than ~0.6 Pg of C year–1 (Fig. 1) to a vigorous uptake of ~1.2 Pg of C year–1 in 2011. This increase has returned the Southern Ocean sink to levels expected from the increase in atmospheric CO2 (5), computed from an ocean biogeochemistry model forced with just the increase in atmospheric CO2 (18). The increase in the Southern Ocean carbon uptake since 2002 is responsible for roughly half of the global trend in the ocean carbon sink over this period (15), highlighting the importance of the Southern Ocean in moderating the growth of atmospheric CO2.

Fig. 1 Evolution of the Southern Ocean carbon sink anomaly south of 35°S.

The lines show the integrated air-sea CO2 flux derived from two complementary surface ocean pCO2 interpolation methods [a two-step neural network technique (15) and a mixed-layer scheme (17)] as well as the integrated flux from an atmospheric inversion based on measurements of atmospheric CO2 (14). These estimates are compared with the expected uptake based on the growth of atmospheric CO2 alone, based on a simulation with the ocean component of the Community Climate System Model (CCSM3) (18). All data are plotted as anomalies by subtracting the 1980s mean flux from each method.Negative values indicate anomalous uptake by the ocean.

Both surface ocean observation–based methods rely on the to-date largest sea surface pCO2 observation database [Surface Ocean CO2 Atlas (SOCAT) version 2] (19), which contains more than 2.6 million observations in the Southern Ocean south of 35°S. The neural network technique (12, 15) interpolates these observations to a 1° × 1° global grid at a monthly resolution for the period from 1982 through 2011, resulting in a multidecadal reconstruction of the global ocean carbon sink. The method relies on nonlinear but robust relations between the limited pCO2 observations and the properties that are measured more frequently, such as sea surface temperature, sea surface salinity, satellite measurements of chlorophyll a, and mixed-layer depth (supplementary text 1.2). The mixed-layer scheme (13, 17) (version oc_v1.2), in contrast, does not regress pCO2 variations to physical, chemical, or biological driver data but directly assimilates the available pCO2 observations into a mass budget of the mixed layer at a resolution of 4° × 5° in space and daily in time. This method also uses several ancillary observations to parameterize the air-sea CO2 exchange, solubility, and carbon chemistry but does not use them to interpolate the pCO2 to regions without observations. Instead, it interpolates the pCO2 data directly.

Extensive validation of the neural network–based estimate using independent observations reveals that the method is able to map the sparse pCO2 data with little bias (with mean differences between SOCAT observations and neural network estimates of generally less than 2 μatm; table S1) in space and time. Both methods agree well regarding the sign and the magnitude of the decadal trends within the two decades from 1992 through 2001 and 2002 through 2011 (Fig. 1 and table S2), when the majority of surface ocean pCO2 observations were made (fig. S3).

However, given the methodological differences in the data treatment in data-sparse regions (interpolation versus regression), there is less agreement regarding higher-frequency variability such as the year-to-year variations in the sink strength. This lower agreement is a result of the weaker signal-to-noise ratio of the pCO2 data in the interannual frequency band. Under such conditions, the direct interpolation scheme of the mixed-layer method tends to extrapolate high-frequency noise present in the observations to the data-sparse regions, probably generating overly strong variations there. In contrast, the neural network scheme suppresses the high-frequency noise by being constrained by the ancillary observations, resulting in a possible underestimation of the year-to-year variability in the data-sparse regions. In contrast, the relatively strong pCO2 signals that underlie the decadal changes in the Southern Ocean are distinctly captured by the two methods, resulting in very similar decadal trends.

The changes in the Southern Ocean carbon sink are almost entirely driven by changes in the air-sea difference of pCO2; i.e., Embedded Image (Embedded Image = atmospheric pCO2), because the direct effect of changes in the wind and temperature on the gas transfer coefficient is small (fig. S8). The spatial pattern of the trends in ΔpCO2 from the neural network method reveals for both decades a very uniform trend pattern across the entire Southern Ocean, with the strongest ΔpCO2 trends at high latitudes (Fig. 2, A and B). The spatial trends for the mixed-layer scheme are similar, although at coarser resolution and with somewhat more zonal variations, part of which may be spurious because of missing data constraints there, reflecting the more variance-producing nature of this method in data-sparse regions (fig. S6). From 1992 through 2001, the trend in ΔpCO2 was strongly positive, driven by the surface ocean pCO2 increasing at nearly twice the rate of Embedded Imagearound Antarctica. In contrast, from 2002 onward, the growth of surface ocean pCO2 nearly stalled, strongly increasing the degree of surface ocean undersaturation, which ultimately drove the increasing uptake of atmospheric CO2.

Fig. 2 Trends in ΔpCO2 based on the neural network output and its two components for the two analysis decades: 1992 through 2001 and 2002 through 2011.

(A) Linear trend in ΔpCO2 for the 1990s. (B) As (A) but for the 2000s. Linear trend in (C) thermal pCO2 and (D) nonthermal ΔpCO2 for the 1990s. (E) and (F), as (C) and (D) but for the 2000s. Positive (red) ΔpCO2 trends indicate a faster increase of pCO2 in the surface ocean than in the atmosphere (i.e., a decreasing sink) and vice versa for positive (blue) trends. Hatched areas indicate where the linear trends are outside the 5% significance level (P ≥ 0.05). (G) and (H) illustrate decadal trends of sea level pressure (shading) and 10-m wind (vectors) from 1992 through 2001 (G) and 2002 through 2011 (H) based on data from the ERA-Interim reanalysis (20).

We tested the robustness of this result on the basis that such strong decadal changes in the CO2 uptake across most of the Southern Ocean should leave an imprint on atmospheric CO2, taking advantage of the lack of land regions with substantial CO2 fluxes south of 35°S. Specifically, we used an atmospheric inversion method (14) to infer the air-sea CO2 fluxes that are optimally consistent with the atmospheric CO2 data, while taking into consideration atmospheric transport and mixing. The setup of the employed version s85_v3.7 uses the atmospheric winds from the ERA-Interim reanalysis (20). The evolution of the Southern Ocean carbon sink from this inversion of atmospheric CO2 data also supports our postulated larger–than-expected increase in the strength of the Southern Ocean carbon sink within the past decade (Fig. 1), even though it shows much less of a weakening during the 1990s. Thus, two complementary pCO2 database estimates, as well as an atmospheric CO2 inversion, confirm that the Southern Ocean carbon sink has experienced a significant strengthening since the early 2000s.

This reinvigoration after the early 2000s cannot be a simple reversal of the Southern Annular Mode–driven wind trend that has been suggested to cause the weakening of the Southern Ocean carbon sink over the past decades (1, 2), because the ERA-Interim reanalysis winds (20) do not show such a signal (Fig. 2, G and H). Instead, the atmospheric circulation became more zonally asymmetric with a wavenumber two pattern, reminiscent of the lower-frequency pattern of variability of the Antarctic Circumpolar Wave (21). But how can this zonally asymmetric forcing result in a relatively zonally uniform response of the surface ocean pCO2?

Insight into the drivers is gained by separating the ΔpCO2 trend pattern into a component driven by changes in sea surface temperature [the thermal trend (Fig. 2, C and E)] and one driven by changes in the DIC and/or alkalinity [the nonthermal trend (Fig. 2, D and F)] (22). For both analysis periods, the trends in the thermal and nonthermal components are generally opposed for any given location, which is in line with previous studies (2224). The thermal component shows a sink increase in both decades in the Pacific sector, where the advection of cold air from Antarctica and sea-ice changes led to a persistent surface cooling trend (25). In the lower latitudes of the Atlantic and Indian sectors, we find a reduced thermally driven uptake in the 2000s due to surface ocean warming, which is probably related to the more asymmetric atmospheric circulation that caused a reduced northward Ekman transport (fig. S11) of cold polar waters in these regions.

In the nonthermal component, we find more distinct differences between the two periods. Between 1992 and 2001, the nonthermal component increased the oceanic pCO2 over most of the Southern Ocean (Fig. 2D), in particular in the high latitudes and in the Pacific sector. The estimated changes in Ekman pumping velocity [estimated from ERA-Interim winds (20), fig. S11] support the hypothesis that wind changes led to an increased surface divergence and an associated upwelling of DIC-rich waters into all sectors of the high-latitude Southern Ocean (2) in the first period. During the subsequent period, the nonthermal component primarily reduced the oceanic pCO2 in the Atlantic and Indian sectors and over the Antarctic shelf (Fig. 2F). In contrast, this component continued to increase pCO2 in most of the Pacific sector, although at a much weaker rate than in the 1990s. This much weaker DIC- and/or alkalinity-induced increase in pCO2 in the Pacific sector could no longer compensate for the thermal trend, so that the negative trend in the total pCO2 in this region for the period after 2001 is dominated by the thermal trend. In contrast, in the Atlantic and Indian sectors and over the Antarctic shelf, the negative nonthermal trend dominates the thermal changes. Thus, overall, the temperature-dominated pCO2 trend in the Pacific sector and the DIC/alkalinity–driven trend in the other regions have worked in tandem to prevent pCO2 from increasing across the entire Southern Ocean since the early 2000s. Over the same period, atmospheric CO2 continued to rise strongly, resulting in a substantial increase in the undersaturation of the surface ocean with regard to atmospheric CO2, hence driving a strong increase in the flux of CO2 into the entire Southern Ocean.

We interpret this zonal asymmetry of thermally and DIC/alkalinity–driven changes to be primarily the result of an increased asymmetry in the Southern Hemisphere atmospheric circulation in the years since 2001 (Fig. 3). Specifically, the conditions became more cyclonically dominant in the Pacific sector, and more anti-cyclonically dominant in the Atlantic and parts of the Indian sector (Fig. 2H). As a result of the associated increase in the meridional wind components, more cold air was advected from the Antarctic continent over the Pacific sector, and more warm air was advected from subtropical latitudes over the Atlantic and part of the Indian sectors. Together with the changes in northward Ekman transport (fig. S11), this provides an explanation for the strengthened asymmetry in the sea surface temperature trends, which underlie the thermal trends in Fig. 2E.

Fig. 3 Schematic of the processes governing the changes in the ΔpCO2 trends in the Southern Ocean since 2001.

The trend toward a zonally more asymmetric distribution of the atmospheric pressure systems in the past decade led to stronger meridional winds bringing either colder air (Pacific sector) or warmer air (Atlantic sector) to the open Southern Ocean, causing strong cooling of the sea surface in the Pacific sector and warming in the Atlantic sector. The changes in wind also affected the oceanic circulation pattern, with the net effect being an increase in the DIC/alkalinity–driven pCO2 component in the Pacific sector and a decrease of this component in the Atlantic sector (i.e., opposing the effect of sea surface temperature on pCO2). In the Pacific sector, the effect of the cooling trend on pCO2 prevails, whereas in the Atlantic sector, the effect of circulation/mixing on DIC/alkalinity prevails, also causing a lowering trend in pCO2. Thus, owing to the interaction between temperature and circulation changes, the zonally asymmetric forcing caused a zonally relatively symmetric response of the ocean carbon sink.

A strengthening of the carbon sink in the Pacific sector, combined with the further intensification of the winds (Fig. 2H) during the 2000s, provide a paradox at first sight, because the increased upwelling in the Pacific sector should have increased the surface DIC content further. A possible explanation is that the recent stabilization of the surface waters (26) counteracted the wind-induced upwelling. In the Pacific sector and in coastal regions, strong surface freshening (26, 27) might have caused most of this stabilization, whereas in the lower latitudes of the Atlantic and Indian sectors, warming stabilized the surface waters. The reduction in northward Ekman transport to the lower-latitude Atlantic and Indian sectors during the 2000s (fig. S11), which is probably the result of the zonally more asymmetric atmospheric circulation, also reduced the northward advection of high-latitude waters, lowering the DIC content and/or increasing the alkalinity at the surface.

The trend toward a zonally more asymmetric atmospheric circulation may be related to long-term variations of the tropical sea surface temperature; i.e., to the more prevalent La Niña conditions in the Pacific since the early 2000s (28) and the more positive phase of the Atlantic Multidecadal Oscillation over recent decades (29). Alternatively, it may be driven by a zonally asymmetric response of the Southern Hemisphere near-surface circulation to the anthropogenic forcing (25).

Our results indicate that Earth’s most important sink for anthropogenic CO2 (5, 6) is more variable than previously suggested and that it responds quite sensitively to physical climate variability. This also suggests that should current climate trends reverse in the near future, the Southern Ocean might lose its recently regained uptake strength, leading to a faster accumulation of CO2 in the atmosphere and consequently an acceleration of the rate of global warming.

Supplementary Materials

Supplementary Text

Figs. S1 to S12

Tables S1 and S2



  1. ACKNOWLEDGMENTS: This work was supported by European Union (EU) grant 264879 (CARBOCHANGE) (P.L., N.G., D.C.E.B., M.H., S.v.H., and N.M.) and EU grant 283080 (GEO-CARBON) (P.L. and N.G.), both of which received funding from the European Commission’s Seventh Framework Programme. F.A.H. was supportetd by ETH research grant CH2-01 11-1. T.T., R.W., and C.S. acknowledge funding for the pCO2 from ship projects from the Climate Observation Division of NOAA. T.T. and the Ship of Opportunity Observation Program were supported by a grant (NA10OAR4320143) from NOAA. C.S.’s contribution to this research was made possible by support from the U.S. National Science Foundation’s Office of Polar Programs (grants AOAS 0944761 and AOAS 0636975). B.T. was funded through the Antarctic Climate and Ecosystems CRC, the Australian Climate Change Science Program, and the Integrated Marine Observing System. N.M. is grateful for support from the Institut National des Sciences de l'Univers/Centre National de la Recherche Scientifique and the Institut Polaire Française for the Océan Indien Service d’Observation cruises.. C.R. thanks the providers of atmospheric CO2 measurements and the Deutsches Klimarechenzentrum computing center for their support. SOCAT is an international effort, supported by the International Ocean Carbon Coordination Project, the Surface Ocean Lower Atmosphere Study, and the Integrated Marine Biogeochemistry and Ecosystem Research program, to deliver a uniformly quality-controlled surface ocean CO2 database. The many researchers and funding agencies responsible for the collection of data and quality control are thanked for their contributions to SOCAT. We also thank A. Hogg for fruitful discussions. The surface ocean CO2 observations are available from the SOCAT website ( The sea surface pCO2 and air-sea CO2 flux data leading conclusions of this manuscript are available to the public via the Carbon Dioxide Information Analysis Center ( The mixed-layer scheme and inversion data supporting the main findings can be obtained from and P.L. and N.G. designed the study and wrote the paper together with F.A.H. P.L. developed the neural network estimation and performed the majority of the analyses, assisted by F.A.H. C.R. developed the mixed-layer scheme and the atmospheric inversion. S.v.H., M.H., N.M., C.S., T.T., B.T., and R.W. were responsible for the collection of the majority of the surface ocean CO2 data in the Southern Ocean. D.C.E.B. led the SOCAT synthesis effort that underlies this work. All authors discussed the results and implications and commented on the manuscript at all stages.
View Abstract

Stay Connected to Science

Navigate This Article