Long Term Mean Ocean Uptake

Here I review three largely independent lines of evidence concerning the net uptake of carbon by the global ocean. At the outset I should address some ambiguity about the word "uptake." The word should literally refer to the net storage of carbon in the ocean. However, many of the studies cited here calculate the net flux of carbon across the air-sea interface or deduce the ocean uptake from the terrestrial uptake. Carbon entering the ocean by other routes may not be counted. This becomes particularly confusing in the presence of background circulations of carbon that may predate any anthropogenic perturbation. Such a correction between fluxes and uptake (uptake is often referred to as storage to remove this ambiguity) was cited by Sarmiento and Sundquist (1992) to reconcile the flux estimates of Tans et al (1990) with estimates from, for example, Sarmiento et al (1992). While Sarmiento and Sundquist (1992) quoted a magnitude of 0.4-0.8 Gt C year-1 for the correction I believe this should contribute as much to the uncertainty as to the estimate itself.

2.1 Global and Temporal Perspective

The first strand of evidence for the global ocean uptake comes from models of the ocean carbon cycle. These generally combine descriptions of the relevant carbonate chemistry with some representation of tracer transport in the ocean. As well they need a model (usually highly simplified) of the biological processes in the ocean and finally some parameterization of the surface fluxes of global btogeochemtcal cycles itv the climate system

Text Copyright ■ 200! by Academic 1'ress. All rights of reproduction in any form reserved.

Figures Cop") rightft CSIRO and |>. J. Rayner

CO,. Underlying all these is a model of the physical circulation of the ocean, usually arising from an ocean general circulation model. Three such estimates of the air-sea carbon flux are shown in Figure 1. The estimates are taken from the Ocean Carbon Model Intercomparison (OCMIP) (Orr, 1997).

There is consensus among ocean modelers, with a preliminary estimate for net uptake of anthropogenic C02 in the ocean of 2.0 ± 0.4 Gt C year-1 for 1990. Note that, since most ocean carbon cycle models do not attempt to represent the interannual variability in ocean circulation, these calculations represent an average over an ensemble of possible ocean states for 1990. Given the uncertainties in the underlying physical simulations, this convergence seems surprising. Two factors may help explain the agreement. First, several of the highly uncertain processes controlling uptake do not control the uptake on a decadal time-scale. For example, the surface and ocean are close to equilibrium, so the highly uncertain gas exchange does not force large model-model differences. Similarly, exchange with the deep ocean is not large on decadal scales. The important chemical processes, on the other hand, are well understood and hence consistent among models. Second, there are some integral or global constraints on the global ocean uptake in a model. In particular, most models used for ocean carbon studies have been checked if not tuned against the change in l4C inventory arising from the above-ground nuclear tests in the 1950s and 1960s. While limited data make this check far from perfect (Fleimann and Maier-Reimer, 1996), it does provide a constraint for global uptake estimates.

The range of estimates from OCMIP probably underestimates the total uncertainty in net uptake from ocean models. In Fig. 1 the local agreement is worse than the global agreement. Just as in the global case, the range is one estimate of local uncertainty in air-sea flux. Some of the differences arise from the different positioning of source or sink regions and will vanish when larger-scale integrals are considered. This cancellation will only partly offset the general behavior of uncertainties, which is that they sum quadrati-cally. I believe that large-scale cancellation of differences, e.g., that differences in net uptake in one hemisphere are compensated in the other, is not due to compensating differences inherent in the models. Rather it arises when the models are forced to fit global constraints such as the l4C inventory. I would conclude, then, that the agreement in estimates of net uptake from ocean models does suggest some underlying control, although the estimated uncertainty is larger than this intercomparison would suggest.

The second line of evidence comes from the time trends of some atmospheric species, mainly CO, and oxygen. Developing the oxygen budget of the atmosphere, which contains some problematic terms, is beyond the scope of this chapter. Briefly, the global, long-term budgets for 0,/N, and CO, can be written schematically as a„(co2) d,

-100

- Takahashi

---Princeton

GS GL MS ML

-120

Latitude

FIGURE 1 Several estimates of the zonal mean flux (Mt C year-1 per degree latitude) from the ocean into the atmosphere. The three blue lines are taken from OCMIP (Orr, 1997), the red line from a regionally aggregated estimate of Takahashi et al. (1997) and the four green lines from four atmospheric inversions as described in the text.

where q refers to atmospheric concentration,/,,, f0, and/B refer to fluxes to the atmosphere from fossil fuel, the ocean, and the terrestrial biosphere, respectively, and the S factors are the stoichiometric ratios of O, to CO, associated with each of these fluxes. Briefly, this assumes that the ocean makes no contribution to the atmospheric 0,/N2 budget, which means the equation is invalid at seasonal (and probably interannual) time-scales. There are also difficulties associated with treating budgets like this over finite time intervals, particularly regarding estimates of the trend from noisy series. Readers are referred to Enting (1999) for a fuller treatment of these problems. The oxygen budget also assumes the marine biomass is in steady state, an assumption questioned by Galloway et al. (1995). Potential unaccounted fluxes like this should be regarded as contributing to the uncertainty in derived flux estimates. Given reasonable knowledge of (Marland and Boden, 1997) Eq. la reduces to two algebraic equations in the unknowns fn and fH. All this relies on measurements of the C02 and 02/N2 trends, which were pioneered respectively by C. D. Keeling (Keeling, 1960) and R. F. Keeling (Keeling, 1988). For CO,, the long-term trend is well characterized by the global sampling programmes in place since the 1960s, e.g., Conway et al. (1994) and Keeling et al. (1995). For 02/N2, two estimates of the trend have been made on decadal time-scales. One study (Battle et al., 1996)

uses the record of air trapped in firn at the South Pole and some more recent atmospheric measurements. The firn measurements constrain the trend for the period 1978-1985. The other study (Langenfelds et al., 1999) uses air sampled at Cape Grim, Tasmania, and archived. This record determines the trend for 1978-1997. Both records have their difficulties and are hard to compare because of the different periods they cover. Their overall trend is similar. I will use the value of — 3.5 ± 0.2 ppmv year-1 for 1978-1997 from Langenfelds ct al. Table 1 lists data and derived values for these terms. The contributions to the uncertainty are as follows:

1. Quoted uncertainty in both the CO, and 0,/N, trends.

2. Fossil-fuel source. Note that this has a significant impact on the terrestrial uncertainty because of its high oxygen-carbon stoichiometric ratio.

3. Uncertainties in stoichiometric ratios.

4. The potential for the occan oxygen budget to be out of balance on any given time-scale.

Most of these terms are difficult to assign and I have used some judgment for some of them. More serious, however, is the question of just what budget is established by this simple calculation. The derived so-called terrestrial flux is in fact the net amount of carbon being reduced with carbon/oxygen ratios characteristic of photosynthetic material. According to Galloway et al. (1995) some of this carbon could be reduced in the ocean, resulting in changes of marine biomass, although this is probably not a major contribution on the lime-scales considered here. The calculation also says nothing about the ultimate fate of the organic material formed by photosynthesis. Some is washed into the ocean by rivers to be outgassed again through the ocean surface or buried in sediments. Such a circulation probably existed in the preindus-trial carbon cycle and the impact of perturbations in this budget is highly uncertain. However, Sarmiento and Sundquist [1992] estimate a maximum contribution of 0.3 Gt C year-1 for this term.

TABLE 1 Parameters Used in the Derivation of the Long-Term Mean Global Land and Ocean Fluxes from Trends in Atmospheric C02 and 0,/N2.

Parameter

fr

1.4 ± 0.05 ppmv year 1 5.73 ± 0.3 Gt C year-1

>„(o,} >1

— 3.5 0.2 ppmv year-1

Sr.

1.38 ± 0.02

So

0

S„

-1.05 ± 0.05

fn h

-2.2 ± 0.4 GtC year-1 -0.6 * 0.5 Gt C year-1

See Eq. la for the definition of the parameters. Uncertainties refer to one standard deviation. CO: growth rates are taken from the output of Rayner el al. (1999).

See Eq. la for the definition of the parameters. Uncertainties refer to one standard deviation. CO: growth rates are taken from the output of Rayner el al. (1999).

Global constraints like the 02/N2 trend form part of the data used by Rayner et al. (1999) in their synthesis inversion. They solve for the spatial distribution of surface sources and use the long-term constraint provided by the oxygen budget as a global constraint. With their large excess of degrees of freedom, they deduce a budget of — 2.1 Gt C year-1 for ocean flux and — 0.7 Gt C year-1 for land. The slight mismatch between this calculation and the purely global one detailed in Table 1 comes from the extra information from spatial CO, gradients and the use of prior estimates. Note, too, that the estimated net flux for the terrestrial biosphere includes a large positive contribution from land-use change.

2.1.1 Ice-Core Records and S13C

02/N2 is a useful quantity for partitioning land-ocean uptake since it is hoped to be a tracer of terrestrial and not ocean processes. The ratio of 13C to 12C in the atmosphere can play a similar role since photosynthesis discriminates strongly against l3C while dissolution in the ocean does not. We could, in principle, use S13C in the same way as 02/N2 but there are several complicating factors. The carbon in fossil fuel, being a product of photosynthesis itself, is isotopically different from the carbon in the current atmosphere so that its input changes <513C in the atmosphere (the Suess effect). There are also large so-called gross fluxes between the atmosphere and underlying reservoirs. These gross fluxes influence the 13C budget so we need to take them into account when using this species. The gross flux can be thought of as the number of molecules crossing the interface between two reservoirs (in one direction). In the ocean the gross flux is driven by the continual exchange between surface waters and overlying air. For the terrestrial biosphere, the important reservoir is the photosyn-thesized material. The amount of carbon entering this reservoir each year is known as the gross primary productivity. We can neglect the larger amount of carbon that passes in and out of the stomata of leaves without being assimilated into plants. Both these gross fluxes can be large even when there is no driver for a net flux, i.e., in steady state.

Gross fluxes can impact the isotopic composition of the atmosphere in the absence of net fluxes. If there are, for example, higher concentrations of l3CO, molecules in the ocean than the atmosphere then, on average, more of these are likely to leave than enter the ocean. The resulting isotopic flux is often called the isoflux as a convenient shorthand. The isoflux acts as a restoring term to bring the ocean and atmosphere back to equilibrium. The same holds for the terrestrial biosphere. The strength of the restoring term depends on the size of the gross flux (number of molecules crossing the interface) and the difference in the concentration of 13C02 molecules on each side of the interface, i.e, the magnitude of the isotopic disequilibrium. The disequilibrium, in turn, depends on the rate of change in the atmosphere and the adjustment times of underlying reservoirs. A combination of large reservoirs (hence slow adjustment) and relatively rapid change in the atmosphere has led to a large disequilibrium at present between the atmosphere and both the main reservoirs. The upshot is that the <5I3C budget of the atmosphere is sensitive to the magnitudes of the gross fluxes. Unfortunately, these gross fluxes are very hard to quantify from direct measurement. It is even more difficult when one considers that the disequilibrium is spatially variable so that the global isoflux must be properly flux-weighted for the effect of different regions. A side benefit of the calculations based on 02/N, has been an indirect estimate of the global isoflux. To do this one first calculates the ocean and terrestrial fluxes as above, then calculates the isoflux as the residual from these fluxes, the fossil fuel input, and the atmospheric trend. Both Langenfelds et al. (1999) and Rayner et al. (1999) present such a calculation.

There is an alternative method for estimating the current isoflux. Air trapped in ice cores can provide long-term histories of C02 and ¿>I3C at one point over centuries or millennia. With such records, and knowledge of anthropogenic inputs, the combined C02 and <5I3C budgets can be inverted to solve for the ocean and terrestrial net fluxes. Such a calculation is presented by Joos et al. (1999). A similar calculation using only the CO, record, a model of ocean uptake and the ¿>I3C record as cross validation is presented by Trudinger et al. (1999). Both calculations use the ice-core data of Etheridge et al. (1996) and Francey et al. (1999) in simple box models of the atmosphere, ocean, and terrestrial biosphere. Such calculations also use simple models to track the SI3C of the reservoirs and hence the disequilibrium with the atmosphere. Thus, assuming the gross fluxes are known, and subject to the accuracy of these models, this calculation will estimate the isoflux over time. Such calculations are stabilized by the role of the gross flux as a restoring term. A large gross flux will generate a small disequilibrium while a small gross flux will generate a large disequilibrium. Their product, the isoflux, may be less sensitive than either. In a set of sensitivity calculations, Trudinger et al. (1999) showed that the calculated isoflux for the period 1980-1990 was not very sensitive to the specified gross fluxes although a more detailed test of model parameters is still to be performed. Thus, used in this long-term role, SI3C forms a valuable constraint on the current partition of uptake.

Rather than comparing estimates of net fluxes (which vary rapidly on interannual periods) Figure 2 compares the isoflux predicted from the 02/N2-based calculations of Rayner et al. (1999) and the calculations of Trudinger et al. (1999). There is close agreement on both magnitude and slope. Note that, when computed by an inversion, the slope of the isoflux is derived from curvature in the §I3C record, which is a very subtle feature. (Indeed, the magnitude of the slope is not supported by the findings of Gruber et al. (1999) using the trends in ocean disequilibrium. They show no trend in disequilibrium between the mid-1970s and mid-1990s.) Their mean value is, however, consistent with estimates presented here. While the uncertainties in both methods are substantial, corresponding to a net flux of ±0.5 Gt C year-', these two somewhat independent calculations agree that net ocean uptake in the period 1980-1990 was 2.0 ± 0.5 Gt C year-1.

0 60

1 40

20 0

1900 1920 1940 1960 1980 2000 Year

FIGURE 2 Isoflux (Gt C year-l%o) from the box diffusion model of Trudinger et al. (1999) and the synthesis inversion of Rayner et al. (1999). Uncertainties for the synthesis inversion are calculated by considering the joint uncertainty in the mean value and slope.

2.2 Spatial Perspectives

In summary, then, it would appear that ocean models, the trend in 02/N2 and the long-term histories of C02 and <5I3C agree on an ocean uptake for the 1980s around 2 Gt C year-1. The apparent consensus breaks down when spatial gradients of C02 and S13C in the atmosphere are considered. While an exhaustive overview of these studies is beyond us here, the general methods and results have been similar. The major forcing for the meridional structure of C02 in the atmosphere is the north-south gradient in fossil-fuel combustion. When this forcing is used as a flux-boundary condition for an atmospheric transport model, this produces a north-south gradient substantially larger than observed. Based on this mismatch in gradients, several authors, e.g., Enting and Mansbridge (1989), Tans et al. (1990) and Keeling et al. (1989), suggested that there should be a sink in the Northern Hemisphere to produce a countervailing negative gradient.

How this gradient was interpreted depended on ancillary information. Keeling et al. (1989) had already set the global ocean uptake from a calculation like that of Trudinger et al. (1999). They were left to decide an apportionment of the uptake among ocean basins. They posited an extra source in the southern ocean balancing an extra northern oceanic sink. This structure is consistent with the description by Broecker and Peng (1992) of the transport of carbon from formation of North Atlantic Deep Water into the southern ocean. Tans et al. (1990), using SpC02 measurements argued that the northern ocean could not account for the required sink. They therefore preferred a scenario of a substantial northern land sink. In their preferred scenario, the global ocean sink was less than 1 Gt C year-1. An ocean sink in keeping with SpC02 measurements (with a relatively large Southern Hemisphere uptake) would exacerbate the gradient mismatch.

1900 1920 1940 1960 1980 2000 Year

FIGURE 2 Isoflux (Gt C year-l%o) from the box diffusion model of Trudinger et al. (1999) and the synthesis inversion of Rayner et al. (1999). Uncertainties for the synthesis inversion are calculated by considering the joint uncertainty in the mean value and slope.

The carbon cycle community has sought hard the solutions to this ocean-atmosphere paradox. At global scale, Sarmiento and Sundquist (1992) adduced the skin-temperature correction from Robertson and Watson (1992) as well as the river-flux correction already mentioned. The mismatch in gradients was also reduced when the atmospheric transport of CO was considered as in Ent-ing and Mansbridge (1991). Next, since the initial results were derived from two related models it was thought (perhaps hoped) that their transport was aberrant. If modeled transport between hemispheres was much slower than observed then the mismatch between modeled and observed meridional gradients would be purely a model artifact. A comparison of 12 atmospheric transport models was reported by Law et al. (1996). in the TransCom (transport comparison) study. An estimate of fossil-fuel source and seasonal biospheric exchange (annually balanced) were used as flux-boundary conditions for the contributing models. Although the response to the fossil-fuel source (embodied in the mean interhemispheric difference at the surface) varied by a factor of 2, the models from the earlier studies were at the more rapid end. Thus these models would produce a smaller north-south gradient than most, meaning the mismatch was perhaps even worse than was first thought.

While this study was under way the evidence for large northern land sinks seemed to strengthen. Two studies by Ciais ct al. (1995a, b) included the spatial gradients in S13C as well as C02 in an atmospheric inversion. The results suggested northern land sinks in the range 2.5-3.5 Gt C year-1 for the period 1992-1994. Studies that calculated the year-to-year changes in this sink, such as Conway et al. (1994) and Rayner et al. (1999), suggest this was a period of anomalously large northern sink.

A further complication was added by Denning et al. (1995). Using the Colorado State University (CSU) general circulation model (GCM), they simulated a strong annual mean response to the annually balanced biospheric source. The annual mean response arose from the covariance between the seasonality of transport and the seasonality of the source. The effect had been noted earlier in Keeling et al. (1989) and dubbed the atmospheric rectifier by analogy with the production of a dc signal from an ac source in electrical circuits. The signal was positive in the Northern Hemisphere, strongest over land but also carried to the Northern Hemisphere observation sites. If the effect occurs in nature it would further strengthen the north-south gradient, requiring a yet larger sink in the Northern Hemisphere. The Denning et al. (1995) study provoked such strong interest because of the size of the effect, generating large-scale gradients from the annually averaged biosphere source roughly half those from fossil-fuel sources. Law ct al. (1996) noted the same effect in several contributing models. In a recent sensitivity study, Law and Rayner (1999) found that the impact of the rectifier effect on atmospheric inversions was governed not only by the strength of the signal over source regions but how well it was advected to observation sites. The CSU model used by Denning et al. (1995) produced stronger signals at observation sites than did the NCAR-MATCII model used by Law and Rayner (1999) and hence a greater impact. The third phase of the TransCom study will investigate these model-model differences in inversions systematically.

Finally, the terrestrial carbon cycle community has had little trouble identifying candidates for this enlarged role for land biota. Inventory studies such as Kauppi et al. (1992) or flux studies such as Grace et al. (1995) have suggested large sinks in northern and tropical forests, respectively. The possible mechanisms include the direct stimulation of net uptake by increased C02 concentration (so-called C02 fertilization), regrowth of forests on abandoned agricultural land, increased nutrient supply from other anthropogenic sources, and impacts of climate change. Many of these will be discussed elsewhere in this book.

So, 10 years after the paradox was first raised, the global carbon budget stands in a curious position. In general, the scientific community has adopted the combined evidence of the 0,/N2 temporal trend and ocean carbon cycle models. Thus I would propose a value for inorganic fixation of 2.1 ± 0.5 Gt C year-1 for the decade averaged around 1990. This is probably close to the rate of change of ocean carbon inventory although the uncertainty for this may be larger. This choice reflects a belief in the simplicity particularly of the 0,/N2 record and suggests a lack of confidence in the ability to interpret spatial gradients in the atmosphere. Given that various proposals are in place to interpret these gradients at smaller scales (almost certainly more difficult), it is important to understand the inversion of the large-scale gradient.

Investigation of this paradox between spatial and temporal information requires inversions with both types of information present. The inversion of Rayner et al. (1999) was one such, using both large-scale spatial gradients and the long-term 02/N2 constraint. Four different versions of this study, using two different observing networks and two atmospheric transport models, are shown as the four green lines in Fig. 1. The "GS" line is similar to the calculation of Rayner et al. (1999) except that it uses only the long-term mean sources, not the full year-to-year variations as in that calculation. Prior uncertainties are halved compared to Rayner et al. (1999) to reflect this lack of interannual variability. However, the prior uncertainty on the long-term mean in Rayner et al. (1999) is given by cr/n"2 where ct is the uncertainty for one year and n is the number of years of data in their experiment. This expression arises from the form for the standard error of the mean. Accounting for this, the uncertainties here are twice those in Rayner et al. (1999). The calculation uses the Goddard Institute for Space Studies (GISS) transport model and a relatively small observing network of 12 C02-observing sites. Unlike Rayner et al.

(1999) it does not use <513C data. The setup of their calculation meant that S13C data had no impact on the long-term mean sources so this is not an important difference. Data are taken from GL0BALVIEW-C02 (1999) and cover the period 1980-1995. The calculation uses data from only those months where actual observations exist. Data uncertainties are taken from Peylin et al.

(2000) and reflect the ability of a monthly mean concentration to fit the actual flask data. They range from 0.3 ppnrv at clean sites such as South Pole to over 3 ppmv at difficult observing sites near large terrestrial and industrial sources. The "GL" curve uses the same transport model but a larger C02-observing network, which, at its maximum, contains 65 CO,-observing sites. The "MS" and "ML" lines are the same as the "GS" and "GL" lines except that they use the MATCII transport model as used by Law and Rayner (1999). The prior sources for all these experiments are the same balanced biospheric source from Fung et al. (1987) as used in Rayner et al. (1999) but for the ocean estimate we use the flux compilation of Takahashi et al. (1997). This flux is shown as the red line in Fig. 1.

The clearest feature from the various inversion calculations is the large scatter among them, and between them and the prior estimate. First, there is a substantial difference between the global ocean uptake as reported in Takahashi et al. (1997) and that required to match the oxygen constraint. The inversion is required to substantially increase ocean sinks to match this global constraint, how much and where being determined by the spatial gradients in data and transport characteristics. The inversions usually increase net uptake in both the Northern Hemisphere and Southern Hemisphere oceans except for GS, which decreases net uptake in the Southern Hemisphere (compensated by a very large increase in Northern Hemisphere ocean net uptake) and ML, which decreases Northern Hemisphere ocean net uptake (compensated by a very large Northern Hemisphere land net uptake). Tropical sources are increased by all the inversions, much more so for the GISS than MATCH transport models.

The general sensitivity of source estimates to both transport model and data network is of some concern. There are many potential causes for this (probably undue) sensitivity. One is purely statistical. The same consideration of errors in long-term means that holds for sources also holds for data. When using long atmospheric records as in these calculations, we may imply a tight constraint on the long-term mean spatial gradient. While this might be a fair reflection of observational uncertainties, it does not reflect the ability to model these observations. In the inversion formalism used here, the limitations of the model, embodied in the so-called model error, are included as part of the data uncertainty. The model error contribution does not approach zero with increasing record length. The solution to this problem is a more careful and complex treatment of data uncertainty, which takes explicit account of the model-error contribution.

Another problem is the difference in sampling between the model and real atmosphere. Generally flask sampling is timed to reduce contamination by heterogeneous local sources, which often means a bias toward marine sampling where possible. As treated here, the model includes no such selection. The effect is a stronger observation of land sources in the model than in the real world. The effect grows with an increasing network since many of the extra stations are coastal. The presence of the rectifier effect in the MATCH model further enhances the bias since the model may sample a larger gradient from the rectifier effect than is really observed. This problem can be treated with more judicious sampling in the model atmosphere, closer to the protocols in use by observers.

It should be noted, in passing, that all these inversions produce a lower uptake for North America than in the study of Fan et al. (1998), and that all produce larger uptakes for Eurasia than North America. This is not a direct contradiction of their work, however, since the calculations here have not fixed the ocean uptake as they did.

0 0

Post a comment