Anthropogenic CO2 Uptake by the Ocean Based on the Global Chlorofluorocarbon Data Set

See allHide authors and affiliations

Science  10 Jan 2003:
Vol. 299, Issue 5604, pp. 235-239
DOI: 10.1126/science.1077429


We estimated the oceanic inventory of anthropogenic carbon dioxide (CO2) from 1980 to 1999 using a technique based on the global chlorofluorocarbon data set. Our analysis suggests that the ocean stored 14.8 petagrams of anthropogenic carbon from mid-1980 to mid-1989 and 17.9 petagrams of carbon from mid-1990 to mid-1999, indicating an oceanwide net uptake of 1.6 and 2.0 ± 0.4 petagrams of carbon per year, respectively. Our results provide an upper limit on the solubility-driven anthropogenic CO2 flux into the ocean, and they suggest that most ocean general circulation models are overestimating oceanic anthropogenic CO2 uptake over the past two decades.

Despite improvements in our understanding of the partitioning of anthropogenic CO2between the atmosphere, ocean, and terrestrial biosphere, substantial uncertainties and insufficient direct observational constraints continue. Recent decadal-scale changes in oxygen concentrations that have been observed in the ocean (1, 2) imply large and uncertain corrections (3–5) to the oceanic and terrestrial sinks for anthropogenic CO2 that have been estimated based on atmospheric O2/N2observations (6, 7), which was the method used in the 2001 report by the Intergovernmental Panel on Climate Change (8). Because the O2/N2technique is based on atmospheric observations, it inherently requires assumptions regarding the partitioning of anthropogenic CO2between the ocean and terrestrial biosphere. Ocean general circulation models (OGCMs) currently simulate oceanic anthropogenic CO2uptake, assuming a steady-state circulation and biological production (9). We present here an observational estimate of the decadal inventory of anthropogenic CO2 in the ocean based on the global chlorofluorocarbon (CFC) data set. Our estimates provide independent observational insights into the contemporary global carbon budget and provide a framework that can be used for direct validation of ocean model predictions.

The most direct way of estimating anthropogenic CO2accumulation in the ocean is to compare dissolved inorganic carbon (DIC) measurements made at one time with those made later in the same region. To isolate the long-term trend from changes due to natural variability, DIC measurements along isopycnal surfaces (10) can be compared, or multiple linear regression (MLR) of DIC against hydrographic properties (11) can be used (Fig. 1C legend). Although these methods provide direct evidence for regional anthropogenic CO2accumulation (10, 12, 13), they currently cannot be used in the global context because of the lack of adequate historical DIC measurements.

Figure 1

(A) Zonal mean CFC-12 concentrations (in pmol/kg) for the Indian Ocean. (B) Estimated CFC-12 ages (years) for the Indian Ocean. (C) Zonal mean estimate of anthropogenic CO2 (in μmol/kg) accumulation in the Indian Ocean from 1978 to 1995 made with the MLR method (12). The MLR method involves regressing DIC against Geochemical Ocean Sections hydrographic data from 1978 and applying the resulting regression to WOCE hydrographic data from 1995 to predict DIC concentrations that would have been present without anthropogenic CO2accumulation. The difference between the measured DIC during WOCE and the DIC predicted from the regression is the change in DIC due to anthropogenic CO2 from 1978 to 1995, corrected for changes in DIC due to natural variability. (D) Zonal mean anthropogenic CO2 accumulation (in μmol/kg) for the same period (1978 to 1995), estimated using the CFC age technique.

Another way to estimate anthropogenic CO2 is to study the distribution of CFCs in the ocean. The release of CFC-11 (CCl3F) and CFC-12 (CCl2F2) to the atmosphere began in the 1930s and accelerated in the 1950s. CFCs are entirely anthropogenic and biologically inert in the ocean. The oceanic CFC distribution thus provides valuable information about the rates and pathways of water mass ventilation processes (14). As part of the World Ocean Circulation Experiment (WOCE) carried out in the 1990s, dissolved CFCs were measured with great accuracy and unprecedented global resolution (Fig. 2). The patterns of oceanic accumulation of CFCs and anthropogenic CO2 are quite different as a result of regional differences in carbonate chemistry, solubilities, and rates of air-sea gas exchange. Concentrations of CFCs (Fig. 1A) are highest in the cold, high-latitude surface ocean and decrease equatorward because of lower solubility at warmer temperatures. Although CO2 solubility also increases with colder temperatures, the observed surface water accumulation of anthropogenic CO2 shows a different pattern (Fig. 1C), with the lowest concentrations in the Southern Ocean and concentrations increasing northward. This pattern is mainly due to regional variations in sea surface alkalinity (15). Because of lower alkalinity in surface waters, the Southern Ocean has less capacity to accumulate anthropogenic CO2 for a given increase in atmospheric CO2 than oceans that are located farther north. Because of these regional differences, the direct use of CFC concentrations to infer anthropogenic CO2 accumulation in the ocean is problematic.

Figure 2

Global distribution of CFC measurements used for this study and taken as part of the WOCE (see

The method used here is based on using observed CFC concentrations to estimate water “ages” (defined as the amount of time since the parcel of water was last at the surface). These age-based methods have been applied on a regional scale (16,17). However, they have yet to be applied and validated on a global scale. The conversion to water ages requires CFC, temperature, and salinity measurements, along with CFC solubility functions (18) and CFC atmospheric histories (19). The ages used in this study were calculated using the oceanic CFC-12 concentration and atmospheric observations (20). The conversion from CFC-12 concentration to CFC age produces a tracer that normalizes the variations in the CFC-12 distribution due to oceanic temperature and salinity distributions (Fig. 1B). Because of mixing, a water parcel is composed of a distribution of transit times since it was last in contact with the atmosphere. The mean of these transit times (or ages) is sometimes referred to as the ideal age (21). Water ages estimated from CFC-12 (22, 23) have been shown to reflect the mean transit time (the ideal age) to within 25% for waters from 0 to 25 years old because of the quasilinear atmospheric history of CFC-12 from 1970 to 1995 (23, 24). For older waters, CFC ages significantly underestimate the ideal ages because of a nonlinear mixing bias resulting from the exponential increase of CFCs earlier than 1970 (23, 24). We focus on water with CFC ages younger than 30 years, based on the good agreement with independent direct anthropogenic CO2 estimates in the Indian Ocean (Fig. 1) and Southern Ocean (25) and the results from our modeling study. The contribution of waters older than 30 years to the decadal anthropogenic CO2 inventory has only a modest impact on our estimates, as discussed below. We now combine our estimates of water mass ages with the atmospheric CO2 history (26) and carbonate chemistry equilibrium equations (27) to calculate the change in DIC in the ocean from one period (t 1) to another (t 2) by using the equationEmbedded Image Embedded Image Embedded Imagewhere Canth(t 2) – Canth(t 1) is the accumulation of anthropogenic CO2 between year (t 1) and year (t 2), DICeq is the equilibrium concentration of seawater with atmospheric CO2concentration (fCO2) when the water parcel was last at the surface (28), T is the temperature, S is the salinity, ALK0 is the preformed alkalinity (29), and τ is the water parcel age in years (30).

From the above equation, we can calculate the change in DIC throughout the 1980s and 1990s. When a comparison is possible, the decadal accumulation of anthropogenic CO2 that is estimated with the CFC age method (Fig. 1D) agrees well with direct observations that have been obtained using the MLR technique (Fig. 1C). Using all available measurements in the Indian Ocean, we use the CFC age method to estimate a basinwide accumulation of 6 Pg of C from 1978 to 1995, in agreement with the direct MLR-based estimate of 6.1 ± 1 Pg of C (12). The accumulation between 20°S and 5°N was 1.5 Pg of C, also in good agreement with the independent estimate from isopycnal analyses of 1.64 Pg of C (10).

Although the CFC age technique reproduces the observed decadal accumulation of anthropogenic CO2, we must carefully evaluate the main assumptions in the CFC age technique: (i) that CFC ages give a reasonable estimate of ideal water mass ages, (ii) that CFCs and anthropogenic CO2 maintain similar saturation states at the surface ocean, and (iii) that the air-sea CO2disequilibrium does not change over the time interval in question (31). To explore these combined uncertainties, we used results from a simulation with an OGCM forced with the observed atmospheric histories of CO2 and CFCs (32), following the protocol of the Ocean Carbon Model Intercomparison Project (9). By comparing the simulated global anthropogenic CO2 inventory from between 1980 and 1999 to those computed using the CFC age method in the model, we evaluated the potential biases in the CFC age method. We found that the CFC age method reproduces the simulated decadal anthropogenic CO2inventory to within 10% (Fig. 3A). The Southern Ocean (south of 40°S) is the region of greatest error, because the rapid exchange of surface waters with deep waters results in a violation of all three of the assumptions of the CFC age approach (Fig. 3B). However, the bias in the Southern Ocean may be overestimated in the model, given the good agreement between the observational CFC age estimates and the direct estimates south of 40°S (Fig. 1) (25). In any case, because the total anthropogenic CO2 inventory in the Southern Ocean is small, the biases in this region have little effect on our global inventory estimates. An important factor in the relatively small bias in estimating the decadal anthropogenic CO2 inventory using CFC ages is the relatively linear atmospheric histories between CFCs and CO2 from 1970 to 1990 (19,26). Any mixing effects that result in biases in the CFC age estimation for young waters (23,24) will have a similar impact for anthropogenic CO2 and will result in a relatively small bias when the CFC age method is used to infer decadal anthropogenic CO2 in the ocean.

Figure 3

(A) The change in global inventory of anthropogenic CO2 from 1980 to 1999 simulated in the CSIRO model (black line) compared with the model inventory in waters with an age younger than 30 years (red line) and the inventory change calculated using the CFC age technique for waters younger than 30 years (data points). From 1 June 1980 to 1 June 1999, the total simulated anthropogenic CO2 inventory change is 32.8 Pg of C. In waters younger than 30 years, the inventory change is 27.8 Pg of C. The CFC age technique gives an inventory change of 30.1 Pg of C for waters younger than 30 years, overestimating the inventory change in waters younger than 30 years by 2.3 Pg of C because of age biases and air-sea exchange assumptions (data points versus red line), but missing the 5 Pg of C inventory in waters older than 30 years. These partially cancel to give a reasonably consistent estimate of the decadal anthropogenic CO2 inventory (∼10%) in the model. (B) Fractional error in the water column inventory of anthropogenic CO2 from 1980 to 1999 calculated from the CFC ages as compared to the simulated inventory of anthropogenic CO2 in the model. Negative values indicate that the CFC age estimate is too high. This plot highlights the regions with most bias in using the CFC age technique from our model analysis. In the Southern Ocean, analysis of observations south of 40°S (Fig. 1) (25) shows that the model is most likely overestimating the extent of the bias based on the CFC age approach.

The decadal accumulation of anthropogenic CO2 shows a similar pattern in each ocean (Fig. 4). The lowest values are found in the Southern Ocean (south of 40°S) and the equatorial regions (10°S to 10°N), associated with the upwelling of old water to the surface (12, 13). The largest inventories are found in the southern subtropical gyre (20° to 40°S) and in the North Atlantic, where anthropogenic CO2 penetrates throughout the water column, consistent with penetration of North Atlantic Deep Water.

Figure 4

Basinwide estimates of the accumulation of anthropogenic CO2 (in μmol/kg) for the period from 1980 to 1999, using the CFC age method for the (A) Pacific and (B) Atlantic Oceans.

The global inventory (33) is estimated to be 14.8 Pg of C from mid-1980 to mid-1989 and 17.9 Pg of C from mid-1990 to mid-1999, with the Pacific (∼46%) and Atlantic (∼35%) contributing the most (Table 1). These estimates imply an average uptake rate of 1.6 ± 0.4 and 2.0 ± 0.4 Pg of C per year for the 1980s and 1990s (34).

Table 1

Estimates of the inventory of anthropogenic CO2 (in Pg of C) from 1980 to 1999 (33). Uncertainties are 20% and taken to be an upper limit (34).

View this table:

The CFC age method assumes a steady-state ocean, whereas both models and recent observations (1, 2) suggest that the ocean may be changing as a result of global warming. CFCs were added to the model described above (1, 32) to evaluate climate change impacts on the CFC age–derived estimates of anthropogenic CO2 accumulation (35). We compared the change in anthropogenic accumulation for the period from 1980 to 1999 that was estimated from the CFC simulation with and without climate change. As the ocean warms and becomes more stratified because of global warming, CFC transport into the ocean slows, which results in older water masses remaining in the ocean interior. From the model simulations, climate change reduces the oceanic uptake during the 1990s by less than 1% relative to a simulation that neglects this process. Hence, the potential error associated with climate change is small enough to neglect.

The CFC age technique used here explicitly addresses the abiotic solubility-driven anthropogenic CO2 flux into the ocean and is independent of other observational estimates of anthropogenic CO2 uptake. The assumptions used for our observational estimate here are exactly the same as those used to estimate anthropogenic CO2 from models (36). The potential biases in the CFC age technique result in an overestimation of the anthropogenic storage in the ocean and can therefore provide a firm upper limit on anthropogenic CO2 uptake in the ocean. However, our estimates only include waters younger than 30 years and will therefore miss the anthropogenic CO2 inventory in older waters. To determine the likely contribution of waters older than 30 years, we used the results from the OGCM and also from the direct observations of Sabineet al. (12). Based on the model (Fig. 3), the anthropogenic CO2 inventory for waters older than 30 years is about 18% of the total from 1980 to 1999 and only 5% using the direct observations reported by Sabine et al.(12). If we take the model results as the upper bound and add the contribution to our estimates, then the absolute maximum net oceanic solubility-driven anthropogenic CO2 flux from 1980 to 1999 is about 39 Pg of C, corresponding to 1.9 and 2.3 Pg of C per year for the past two decades. In a recent intercomparison project of 12 international models that simulate anthropogenic CO2uptake (37), three models are close to this observational upper limit [Princeton University in Princeton, New Jersey (PRINCE), the Lawrence Livermore National Laboratory in California (LLNL), and the Commonwealth Scientific and Industrial Research Organisation in Hobart, Australia (CSIRO)], whereas the majority of models overestimated the uptake of anthropogenic CO2 during the past two decades (Table 2). The reason for this is not clear and requires closer regional examination in the models.

Table 2

Annual mean oceanic anthropogenic CO2uptake (in Pg/year) from the Ocean Carbon Model Intercomparison Project (9, 37) for the past two decades. MIT, Massachussetts Institute of Technology, Cambridge, Massachusetts; NCAR, National Center for Atmospheric Research, Boulder, Colorado; IPSL, Institut Pierre Simon Laplace, Paris, France; MPIM, Max Planck Institut for Meteorologie, Hamburg, Germany; SOC, Southhampton Oceanography Centre/Hadley Centre, UK Meteorological Office, England; IGCR, Institute for Global Change Research, Tokyo, Japan; AWI, AlfreWegener Institute for Polar and Marine Research, Bremerhaven, Germany; PIUB, Physics Institute, University of Bern, Switzerland; NERSC, Nansen Environmental and Remote Sensing Centre, Bergen, Norway.

View this table:


View Abstract

Navigate This Article