Saturation of the Southern Ocean CO2 Sink Due to Recent Climate Change

See allHide authors and affiliations

Science  22 Jun 2007:
Vol. 316, Issue 5832, pp. 1735-1738
DOI: 10.1126/science.1136188


Based on observed atmospheric carbon dioxide (CO2) concentration and an inverse method, we estimate that the Southern Ocean sink of CO2 has weakened between 1981 and 2004 by 0.08 petagrams of carbon per year per decade relative to the trend expected from the large increase in atmospheric CO2. We attribute this weakening to the observed increase in Southern Ocean winds resulting from human activities, which is projected to continue in the future. Consequences include a reduction of the efficiency of the Southern Ocean sink of CO2 in the short term (about 25 years) and possibly a higher level of stabilization of atmospheric CO2 on a multicentury time scale.

Atmospheric CO2 increases at only half the rate of human-induced CO2 emissions because of the presence of large CO2 sinks in the ocean and on land (1). The sinks are highly variable and sensitive to climate, yet they are poorly constrained by observations. In the ocean, only the large-scale variability and trends in the equatorial and North Pacific have been quantified (2, 3). In other regions, time-series observations and repeated survey analysis exist, but their extrapolation at the scale of a basin is problematic because of the presence of large regional variability (46). Data are particularly sparse in the Southern Ocean, where the magnitude of the CO2 sink is heavily disputed (7, 8), its interannual variability is unknown, and its control on atmospheric CO2 during glaciations is firmly established but still not understood or quantified (9, 10).

We estimated the variability and trend in the CO2 sink of the Southern Ocean during 1981 to 2004 using the spatiotemporal evolution of atmospheric CO2 from up to 11 stations in the Southern Ocean and 40 stations worldwide (Fig. 1). We used an inverse method that estimates the CO2 flux distribution and time variability that best matches the observed atmospheric CO2 concentrations (11). The inversion uses observed atmospheric CO2 concentrations from individual flask pair values and/or hourly values from in situ analyzers, as available (12) (fig. S1). The station set is kept constant throughout the inversion to minimize spurious variability from the inversion setup. We performed an identical inversion over four time periods using (i) 40 atmospheric stations for 1996 to 2004 (9 years), (ii) 25 atmospheric stations for 1991 to 2004 (14 years), (iii) 17 atmospheric stations for 1986 to 2004 time period (19 years), and (iv) 11 atmospheric stations for 1981 to 2004 (24 years). CO2 fluxes and concentrations are linked by the atmospheric transport model TM3, with resolution of ∼4° by 5° and 19 vertical levels, driven by interannual 6-hourly winds from National Centers for Environmental Prediction (NCEP) reanalysis (13). The a priori information does not involve any time-dependent elements. Although we focus on the Southern Ocean (south of 45°S), where the influence of the land is at its minimum, the inversion is global.

Fig. 1.

Footprint of atmospheric CO2 measurement stations. The footprint is defined here as the area where CO2 fluxes of 0.2 mol/m2 year–1 produce a concentration response of at least 1 ppm, on an annual average. The darkest shading shows the region with largest influence on a given station. Stations are Cape Grim (CGO; 40.7°S, 144.7°E); Macquarie Island (MQA; 54°S, 159°E); Baring Head (BHD; 41°S, 175°E); Tierra del Fuego (TDF; 54.9°S, 68.5°W); Palmer Station (PSA; 65.0°S, 64°W); Halley Bay (HBA; 75.7°S, 25.5°W); Cape Point (CPT; 34°S, 19°E); Syowa (SYO; 69°S, 39°E); Mawson (MAA; 68°S, 63°E); Amsterdam Island (AMS; 38°S, 78°E); and South Pole (SPO; 90.0°S). The color coding refers to the length of the station's record used, with light gray stations used since 1981, green stations since 1986, purple stations since 1991, and dark gray stations since 1996.

The variability in integrated sea-air CO2 flux estimated by the inversions is ±0.14 Pg C year–1 (14) over the Southern ocean (Fig. 2). The amplitude of the CO2 variability is about one-third of the amplitude of the flux variability associated with El Niño events in the equatorial Pacific (2) and ∼10% of the variability observed in atmospheric CO2 growth rate (15). The longer inversion reproduces most of the variability of the shorter, better constrained inversions.

Fig. 2.

Sea-air CO2 flux anomalies in the Southern Ocean (Pg C year–1). The contribution of atmospheric CO2 alone (top red curve) is calculated based on observed atmospheric CO2 concentration and a pulse response function that computes the ocean CO2 uptake as a function of time (12, 17). The estimates based on observations use an inverse model of atmospheric CO2. Inversions over four time scales are shown starting in 1981 (thin black, 11 sites), 1986 (green, 17 sites), 1991 (purple, 25 sites), and 1996 (thick black, 40 sites). The gray shading encompasses results from all the sensitivity tests using the 11-site inversion. The lower panel shows results from a process model forced by (full red curve) the 1967 constant winds and fluxes and (blue curve) observed daily winds and fluxes from NCEP reanalysis. Sea-air CO2 fluxes are integrated over 45°S to 90°S. Negative values indicate a flux of CO2 from the atmosphere to the ocean, or a CO2 sink into the ocean. Variability <1 year is removed using a Hanning filter for all time series. The 1995 to 2004 average was removed from all inversions (see table S1 for the spread in the mean). The mean of the atmospheric contribution is normalized to the inverted estimate for the 1981 to 1986 time period.

The longer inversion further shows a decrease of the CO2 sink in the Southern Ocean between 1981 and 2004 by 0.031 Pg C year–1 decade–1. This decrease is significantly different at the 99.5% level (16) from the trend of –0.051 Pg C year–1 decade–1 in sea-air flux expected in response to the increase in atmospheric CO2 alone (Fig. 2). We estimated the trend caused by increasing atmospheric CO2 alone using two independent methods. First, we used a simple pulse response function, which we integrated in time using the observed atmospheric CO2 growth rate as input (top red curve in Fig. 2) (12, 17). This method takes into account the surface ocean equilibration with atmospheric CO2 and the vertical transport of anthropogenic carbon into the ocean. Second, we used a full Ocean General Circulation Model (OGCM) coupled to a state-of-the-art biogeochemistry model [the Pelagic Interactions Scheme for Carbon and Ecosystem Studies version T (PISCES-T) model; bottom red curve in Fig. 2] (12), which we forced with atmospheric surface conditions from either years 1948, 1967, or 1979 repeatedly for all years (three separate simulations, only the 1967 result is plotted in Fig. 2), and with observed atmospheric CO2 concentration. The pulse response and OGCM estimates have similar variability and a similar trend over the 1981 to 2004 time period (–0.051 and –0.057; –0.046, and –0.072 Pg C year–1 decade–1, respectively) (Figs. 2 and 3).

Fig. 3.

Trend in the sea-air CO2 flux (Pg C year–1 decade–1). Cases A1 to A4 estimate the contribution of atmospheric CO2 alone based on a pulse response model (A1) and an OGCM forced by constant winds and atmospheric fluxes (A2 to A4) from years 1948, 1967, and 1979, respectively. All inverse results are shown in black or gray. Case I is the standard inversion. Cases Is1 to Is5 are sensitivity tests to the model parameters. Cases It1 to It3 are sensitivity tests to the atmospheric transport model. Cases Id1 to Id3 are sensitivity tests to the selection of data. The sensitivity tests hatched dark produce the best match to the station data, whereas those hatched light produce the poorest match (18). Results of the process model using observed atmospheric forcing are shown in blue (M). Error bars for all cases indicate the amplitude of the interannual variability (± 1 SD). Significance of the departure between all the inversion cases and case A1 and between the model M and case A3 is also shown below each case (16).

The significant difference between the observed decrease of the CO2 sink estimated by the inversion (0.03 Pg C year–1 decade–1) and the expected increase due solely to rising atmospheric CO2 (–0.05 Pg C year–1 decade–1) indicates that there has been a relative weakening of the Southern Ocean CO2 sink (0.08 Pg C year–1 per decade–1) as a result of changes in other atmospheric forcing (winds, surface air temperature, and water fluxes). For comparison, the mean Southern Ocean CO2 sink is estimated to be between 0.1 and 0.6 Pg C year–1 (table S1).

Inverse methods are sensitive to errors in the setup and transport model, in the data, and in the selection of the sites. We performed three series of sensitivity tests on the inversion results using the longest inversion. In the first series of tests, we assessed the robustness of the results to the choice of the most sensitive parameters of this inversion set up (11): (Is1 and Is2) We increased and decreased, respectively, the a priori standard deviation of the ocean and land CO2 fluxes by a factor of four; (Is3) we increased the a priori standard deviation over the ocean and decreased that over land by a factor of 2 each; (Is4) we increased the spatial correlation scales by a factor of 2 (in latitude) and 4 (in longitude); and (Is5) we decreased the temporal correlation scale by a factor of 4. In the second series of tests, we assessed the robustness of the results with respect to transport errors by degrading the quality of the transport model: (It1) we reduced the resolution of the transport model by a factor of two; and (It2 and It3) we used the degraded model It1 and applied constant winds for years 1990 and 1995, respectively. In the third series of tests, we used the degraded model It1 and included further available data from (Id1) Baring Head, (Id2) Halley Bay, and (Id3) Cape Grim and Syowa, even though they are not available over all the period.

In all the sensitivity tests, the trend in the CO2 sink in the Southern Ocean is smaller than the trend caused by increasing atmospheric CO2 alone (Fig. 3). Inversions I (standard inversion), Is1, Is3, and Is5 produced the best fit to observations (18). These inversions showed a decrease in CO2 sink of 0.03 to 0.08 Pg C year–1 decade–1, significantly different at the 99% level from the trend caused by atmospheric CO2 alone (16). The inversions with the degraded transport model fit less well to the station data but still show adecrease in the CO2 sink significantly different at the 99% level from the expected trend. The only sensitivity test that produces an increase in the CO2 sink (Is2) also produces the worst fit to the observations (18). However, even this inversion produces a smaller increase in the CO2 sink than that caused by atmospheric CO2 alone, although the significance level is lower (92.5%).

We assessed the influence of the choice of stations further by comparing the trends in the long inversion with that of the 1986 to 2004 inversion, which uses 17 atmospheric stations instead of 11 (3 additional Southern Ocean stations). The trend in sea-air CO2 flux in the two inversions for the overlapping period is similar, with 0.047 and 0.035 Pg C year–1 decade–1 for the 11-station and 17-station inversion, respectively, showing that the trend is correctly captured in the longer inversion.

The CO2 flux variability from the longest inversion correlates with the Southern Annular Mode (SAM), an index of the dominant mode of atmospheric variability in the Southern Ocean. We use the SAM definition of Marshall (2003) (19), based on the difference in mean sea level pressure between 40°S and 65°S, which is entirely based on observations and fully independent of our inversion. The correlation of the monthly mean anomalies is small (r = +0.22) but significant at the 99% level (16, 18). The positive correlation indicates that the ocean outgasses CO2 compared with its mean state when the SAM is positive, i.e., when the winds are intensified south of 45°S (20), and suggests that wind-driven upwelling and associated ventilation of the subsurface waters rich in carbon dominates the variability in CO2 flux (18).

To examine whether the results of the inversion can be traced back to physical processes, we estimated the variability and trend in CO2 fluxes using the OGCM-PISCES-T model (12), now forced by daily wind stress and heat and water fluxes from the NCEP reanalyzed data for 1948 to 2004 (13), similar to (21). This process model reproduces similar patterns of variability in CO2 flux as estimated by the inversion, with a smaller CO2 sink (more positive sea-air CO2 flux) during 1993 to 2003 compared with 1983 to 1993 and 2003 to 2004 (Fig. 2). The process model also produces a decrease in the CO2 sink between 1981 and 2004 of 0.018 Pg C year–1 decade–1 (Fig. 3). The difference in sea-air CO2 trend of +0.064 Pg C year–1 decade–1 between this simulation using observed atmospheric forcing and the simulation using constant forcing (–0.046 Pg C year–1 decade–1 using 1967 forcing) is entirely attributable to changes in ocean biogeochemistry caused by changes in surface atmospheric forcing. Thus, the process model attributes the decrease in CO2 sink to an increase in outgassing of natural carbon (sea-air flux of +0.064 Pg C year–1 decade–1) overcompensating the increase in the uptake of anthropogenic CO2 (sea-air flux of –0.046 Pg C year–1 decade–1), in agreement with results of the inversion based on observations.

We performed two additional simulations. First, the winds alone were kept constant at year 1967, but the heat and water fluxes were allowed to vary interannually. Results from this simulation show a trend in sea-air CO2 flux that is close to the simulation where both winds and fluxes are kept constant (–0.034 compared with –0.046 Pg Cyear–1 decade–1). Second, the winds were kept constant in the formulation of the gas exchange only but were allowed to vary in the physical model. The difference in trend with the variable gas exchange was very small (<3%). The results of the process model suggest that the changes in the CO2 sink are dominated by the impact of changes in physical mixing and upwelling driven by changes in the winds on the natural carbon cycle in the ocean (18) (fig. S5), as suggested by the positive correlation between the inversion and the SAM. The process model also shows that the acidification of the surface ocean is accelerated by this process (18) (fig. S5).

On a multicentury time scale, results of simple models based on well-known carbon chemistry show that the ocean should take up 70 to 80% of all the anthropogenic CO2 emittedtothe atmosphere (22). This estimate takes into account changes in carbon chemistry but not the physical response of the natural carbon cycle to changes in atmospheric forcing. In the past, the marine carbon cycle has responded to circulation changes and cooling during glaciations by taking up enough carbon to lower atmospheric CO2 by 80 to 100 parts per million (ppm) (9). Changes in Southern Ocean circulation resulting from changes in Southern Ocean winds (23) or buoyancy fluxes (24) have been identified as the dominant cause of atmospheric CO2 changes (9, 10, 25). We showed that the Southern Ocean is responding to changes in winds over a much shorter time scale, thus suggesting that the long-term equilibration of atmospheric CO2 in the future could occur at a level that is tens of ppm higher than that predicted when considering carbon chemistry alone.

Observations suggest that the trend in the Southern Ocean winds may be a consequence of the depletion of stratospheric ozone (26). Models suggest that part of the trend may also be caused by changes in surface temperature gradients resulting from global warming (27, 28). Climate models project a continued intensification in the Southern Ocean winds throughout the 21st century if atmospheric CO2 continues to increase (28). The ocean CO2 sink will persist as long as atmospheric CO2 increases, but (i) the fraction of the CO2 emissions that the ocean is able to absorb may decrease if the observed intensification of the Southern Ocean winds continues in the future and (ii) the level at which atmospheric CO2 will stabilize on a multicentury time scale may be higher if natural CO2 is outgassed from the Southern Ocean.

Supporting Online Material


Figs. S1 to S8

Tables S1


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article