Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
The CO2 dependence of photosynthesis, plant growth responses to elevated CO2 concentrations and their interaction with soil nutrient status, II. Temperate and boreal forest productivity and the combined effects of increasing CO2 concentrations and increased nitrogen deposition at a global scale
Max-Planck-Institut für Biogeochemie, Postfach 10 01 64, 07701 Jena, GERMANY
1. Appropriate rates of carbon acquisition by temperate and boreal forests are re-evaluated. Based on continental-scale forestry data it is suggested that the productivity of temperate and boreal forests has been overestimated previously.
2. Using these values, a model of the integrated response of ecosystems to carbon dioxide concentration and soil nitrogen availability is presented. The model does not assume constant C/N ratios in plant or soil and considers effects of increases in atmospheric CO2 concentrations and nitrogen deposition separately or together.
3. For temperate-zone forests a co-occurrence of a CO2 increase and nitrogen deposition doubles the increase in net primary productivity and carbon sequestration that would be the case for nitrogen deposition occurring on its own. Considered separately, the effect of the atmospheric CO2 increase is less than even moderate rates of anthropogenic N deposition for temperate or boreal forests. By contrast, for tropical forests, the atmospheric CO2 increase is sufficient to induce large rates of carbon accumulation in plants and soil.
4. Application of the model at the global scale suggests large localized sinks for CO2 in either tropical rain forests or in forested or grassland areas of Europe and North America where appreciable N deposition occurs. Overall, the model suggests a terrestrial sink owing to CO2 fertilization and N deposition of about 0·2 Pmol C per year. About half of this is in the mid-latitudes of the northern hemisphere and about half in the tropics.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Analyses of the historical record of atmospheric CO2 in conjunction with models of oceanic carbon uptake allow some inferences about the historical carbon flux associated with the terrestrial biosphere to be obtained (Oeschger et al. 1975). Depending on the ocean model used, the simulated net release of CO2 from the terrestrial biosphere to the atmosphere is estimated to be between 7 and 13 Pmol C from 1770 to 1984 (Siegenthaler 1993). This is almost certainly less than an estimate of about 12 Pmol C for land-use change from 1765 to 1985 (Houghton & Hackler 1995) suggesting a cumulative flux into the terrestrial biosphere from 1765 to 1985 of 2 ± 3 Pmol C. For the period 1980–1990 a net carbon flux into the terrestrial biosphere of 0·04 and 0·2 Pmol year–1 has been estimated (Schimel et al. 1995). Although there is now some consensus that sequestration of anthropogenically released CO2 by the terrestrial biosphere is probably occurring at the current time (Lloyd 1999), what is still in dispute is the geographical location of that sink and the underlying mechanism(s).
In terms of the former, results from atmospheric inversion studies are ambiguous. Some studies point towards a very strong terrestrial sink in the temperate and/or boreal regions of the northern hemisphere (Tans, Fung & Takahashi 1990; Ciais, Tans, Trolier et al. 1995; Ciais, Tans, White et al. 1995; Fan et al. 1998) with tropical land masses being neutral or even slight sources of carbon to the atmosphere. Moreover, that is after carbon fluxes associated with land-use change have been taken into account. Other atmospheric studies provide a more balanced view on the distribution of the terrestrial carbon sink: the studies of both Keeling, Piper & Heimann (1996) and Rayner et al. (1999) suggesting a significant northern hemisphere sink, but with terrestrial uptake of CO2 in the tropics more or less offsetting the deforestation occurring there. At the other end of the spectrum, the studies of both Keeling, Piper & Heimann (1989) and Enting, Trudinger & Francey (1995) suggested a terrestrial tropical sink larger than that occurring in the northern hemisphere temperate and boreal regions. Despite such conflicting results, there has been little attempt to determine whether the net terrestrial fluxes deduced from these different atmospheric inversion studies are reasonable from a biogeochemical point of view.
This should be possible. The two most important mechanisms leading to long-term net CO2 sequestration by the terrestrial biosphere are increases in plant productivity owing to increasing CO2 concentrations (Gifford 1994) and, in some regions, increased rates of atmospheric nitrogen deposition (Holland et al. 1997). The main purpose of this paper is to provide some new estimates of the magnitude and spatial distribution of a terrestrial sink in response to these atmospheric perturbations; considering the effects of both increases in atmospheric CO2 concentrations and rates of atmospheric N deposition within the one model. Estimates so obtained, are compared to various atmospheric inversion studies and the causes for discrepancies discussed.
Especially in response to comments by some colleagues that previous estimates of temperate and boreal forests productivity (Lloyd & Farquhar 1996) are unusually low, appropriate productivity values for these forests are revisited. Theoretical aspects of appropriate biosphere model structure, especially in terms of the degree of aggregation of component carbon pools with different turnover times are also discussed. Some other aspects of the terrestrial carbon cycle, relating to interactions with phosphorus nutrition, interannual climate variability and a possible carbon sink owing to regrowth of forests in some regions of the northern hemisphere have recently been reviewed elsewhere (Lloyd 1999).
The exchange of CO2 between the atmosphere and the land
Growth rate and net photosynthetic rates of terrestrial ecosystems on the time scales of months to years are referred to as net primary productivity (NP) and gross primary productivity (GP), respectively. Gross primary productivity is equivalent to what plant physiologists normally call CO2 assimilation (photosynthetic) rate, A (Olson 1975); that is, the rate at which plants fix new CO2 into photoassimilate. Net primary productivity is photosynthesis less plant respiration, i.e.
where Rp is the respiration by plants, including that by leaves at night (Olson 1975). Foliar respiration in the light is accounted for in the equations of photosynthesis (Farquhar & von Caemmerer 1982) and is therefore part of GP. For any carbon pool, the rate of change is simply the difference between the fluxes into and out of the pool. That is to say
where MP is the amount of plant carbon present and LP represents the rate of loss of plant carbon through death, abscission and herbivory. Similarly for the soil carbon pool where MS is the amount of soil carbon present and RS is the soil respiration rate:
Thus, the net rate of change in the total amount of carbon in an ecosystem (plant and soil) is equal to NP–RS. In the steady state the flow into each compartment must be exactly equal to the flux out of it. Thus, for example, net primary productivity would have to be balanced exactly by litterfall and death. In formal terms this requires
and for the soil
In this steady state, an important property of each carbon reservoir is the mean residence time (turnover time) τ¯. This is the average time the carbon atoms spend in a particular reservoir. It can be calculated as the amount of carbon in each compartment divided by the flow in (or out) (Nir & Lewis 1975). For terrestrial plant carbon in the steady state
where τ¯P is the mean residence time for plant carbon.
If two of Mp, NP and τ¯P are defined, then the third is as well. This places a constraint on concurrent estimates possible for any ecosystem. Such analyses apply only to mature ecosystems in equilibrium and Fig. 1 shows Mp plotted as a function of NP for various coniferous and broad-leaf forests that were close to maturity at the time of sampling (for original data sources see Lloyd & Farquhar 1996). There is a strong linear relationship for both forest types indicating that τ¯P (the slope of the relationship) is relatively independent of ecosystem size. From maximum likelihood analysis (Kendall & Stuart 1979) average values (± SD) are for τ¯P = 35 ± 7 years for coniferous forests and 27 ± 4 years for broad-leaf forests. From Fig. 1 and equation 3, any concurrent estimate of the net primary productivity and average biomass of cool/cold temperate forests is reasonable if τ¯P is anywhere between about 19 and 49 years (P≤ 0·05). For tropical forests the range is 14–20 years (calculation not shown).
Table 1 shows discrepancies between the usually quoted estimates of MP and NP and those associated with the 1990 FAO Forest Resources Assessment as used by Lloyd & Farquhar (1996). For example, of the nine MP estimated listed in Table 1 only the tropical-forest plant carbon density of Olson (1983) compares closely to that given by FAO (1995). Similarly, NP estimates of Whittaker & Likens (1975), Atjay et al. (1979) and Olson (1983) are also substantially greater than those calculated from the FAO data.
The average values of MP cited by ecologists usually come from a small number of plots (typically less than 20 per biome) but FAO estimates come from national forest inventories and therefore provide a better estimate of regional forest carbon stocks. Similar differences between the two approaches have been noted before for tropical forests (Brown & Lugo 1984). The usual explanation for the high values used by ecologists is that they have a tendency to study sites which epitomize ‘high quality’ instead of average ecosystem development (Olson 1983). From Fig. 1 it can also be concluded that these sites probably have a higher than average NP and thus studies which simply scale up these biassed NP measurements to a global scale must overestimate the true average value. Many complex ecosystem models calibrate their NP against these few high-productivity sites failing to take into account that they are almost certainly unrepresentative.
It is also possible to use FAO growth increment data for temperate zone forests to provide an independent estimate of NP using Net Annual Increment. This is the average increment volume less natural losses over a given period (UNECE/FAO 1992), dV/dt where V is the volume of live wood in the forest. The relationship between dV/dt and NP is
where D is the mean wood specific gravity (g DW m–3), φ is the fraction of carbon in dry matter, ΦB is the proportion of NP going into boles and Υ is the proportion of bole net primary productivity measured as new growth rather than as mortality. The extent and importance of variation in Υ was pointed out by Schulze (1982) and can be gleaned from forester’s yield tables as is shown in Fig. 2. In Fig. 2a and 2Fig. 2b the total volume of live stems, the cumulative volume of dead stems and their sum (the total volume of bole produced over the life of the stand) are plotted for a Pinus sylvestris stand in Finland (Ilvessalo 1920) and a Betula pubscens stand in Germany (Schober 1975). Differentiating with respect to time, the rate of increase in live plant bole volume, dV/dt, the rate of bole loss owing to tree mortality, LB and their sum, the gross increment in stem volume per year are shown in 2Fig. 2c,d. This shows the bole net primary productivity to peak at about 80 years, after which some decline is observed for both stands. The cause of this decline is unclear but it may be because of reduced canopy assimilation rates in older trees (Yoder et al. 1994).
2Figure 2c,d shows Υ to change markedly during stand development. For both stands it is close to 1 early on (almost all bole net primary productivity is measured as new growth). Around the time of maximum bole net primary productivity it is about 0·7 for both stands and at 130 years of age it is much lower, 0·22 for the P. sylvestris stand and 0·32 for Fagus sylvatica.
Based on plots such as Fig. 2, stand scale data in Cannell (1982) and UNECE/FAO (1992) we take (± SD) Υ = 0·7 ± 0·2 for coniferous forests and Υ = 0·75 ± 0·2 for broad-leaf forests. If a forest were to be fully mature, dV/dt would be 0, with Υ = 0, i.e. all the new volume growth would be observed as litterfall. However, the age structure of commercial forests in Europe (Kuusela 1994) and the USA (Turner et al. 1995) shows relatively young forests with a high Υ making up most of the exploitable growing stock surveyed by foresters for wood increment analyses.
Based on Cannell (1984) we take (± SD) D = 410 ± 40 kg m–3 for coniferous forests and 610 ± 90 kg m–3 for broad-leaf forests and using data summarized in Atjay et al. (1981) we take φ = 0·0400 ± 0·00167 mol C g–1 DW.
Table 2 shows estimated dV/dt and NP for coniferous and broad-leaf stands in the boreal and temperate zones and areas of each forest type in the Northern hemisphere (UNECE/FAO 1992). Also shown is the 95% confidence interval for the NP estimates, based on the means and variances given above and an assumption that for dV/dt the standard deviations are 10% of the average values given by FAO. This gives estimates of NP higher in the temperate than the boreal zone, with boreal coniferous forest NP lower than boreal deciduous forest NP. Also shown in Table 2 are the NP estimates of Lloyd & Farquhar (1996). With the exception of temperate coniferous forests agreement between the two methods is good. Both estimates are markedly lower than the usually quoted numbers, the average of which (using the areas of Table 2) is 0·75 Pmol year–1, That is three to four times larger than the values calculated here.
Table 2. . Forest areas, bole volume increment, dV/dt and associated net primary productivity, NP for coniferous and broad-leaf forests in the temperate and boreal zones as calculated using equation 4. Values from Lloyd & Farquhar (1996) are also shown
One reason that the Lloyd & Farquhar (1996) estimate of NP for temperate coniferous-zone forests is lower than the above analysis, only just within in the 95% confidence range (Table 2), might be that a relatively high proportion of forests in the temperate zone are actively regrowing at the current time (Kauppi, Mielikäinen & Kuusela 1992; Sedjo 1992; Turner et al. 1995) and the steady state assumptions of equation 2 are not therefore correct. In such a case, rather than MP/NP reflecting a mean residence time, it is more applicable think of it as a biomass accumulation ratio, B. (Whittaker & Marks 1975). Variations in B with stand development can be seen by plotting MP as a function of NP, with stand age being the source of variation. This is shown for the P. sylvestris and F. sylvatica stands in Fig. 3. Bole volumes have been converted to carbon densities based on the average values of D and φ for coniferous forests (given above) and allometric relationships in Lloyd & Farquhar (1996). Similarly, NP has been estimated from annual bole increments and equation 4. Also shown in Fig. 3 are lines corresponding to B of 10, 20, 30 and 40 years. This shows large changes in B during stand development, increasing from less than 5 years early on and for P. sylvestris levelling out at a little over 40 years at a stand age of about 120 years, after which we would expect B to be approximately equal to τ¯P. For the still growing F. sylvatica stand, B is 20 years at 130 years of age, but would be expected to increase further as the stand matured.
If a large proportion of the coniferous forest resource in the temperate zone was to be currently regrowing after logging or some other disturbance, then the steady-state assumption does not apply, leading to an underestimate of NP using this method. There is evidence that this is the case as a considerable proportion of the commercially utilized forest resource is of a reasonably young age (Kuusela 1994; Turner et al. 1995). This explanation is supported by the observation that for boreal-coniferous forests, both ways of estimating NP from the FAO statistics give similar answers. The majority of the world’s coniferous forest is in Russia, but as much of this forest has not been commercially exploited in the past, it may be that Υ is significantly less than for equivalent forests in Europe. In that respect, the fact that the numbers used for the calculations in Table 2 are predominantly North American and European forest characteristics, is reiterated.
Despite these uncertainties, it is still hard to reconcile the disparity between the usual high NP estimates of ecologists and the FAO carbon densities, especially in the boreal zone. This requires average B≤ 10 years or an average value of Υ of less than 0·2, both of which are unlikely.
It is suggested below that low values for MP and NP from the FAO Forest Resources Assessment mitigate against the dominant component of the terrestrial sink being located in the northern hemisphere boreal-forest zone.
Although these analyses are not straightforward (Tans et al. 1993), there is a general agreement between these studies and those based on models of oceanic CO2 exchange, all of which suggest that sequestration of CO2 by the terrestrial biosphere was somewhere between 0·04 and 0·2 Pmol year–1 during the 1980s (Schimel et al. 1995). Although it is now generally accepted that the terrestrial biosphere is a major sink for anthropogenically released CO2 there is still considerable disagreement both to the nature and location of that sink (Lloyd 1999).
For example, one synthesis inversion of the latitudinal gradient in concentration and carbon isotopic composition (δ13C) of atmospheric CO2 (Enting et al. 1995) concluded that most of the terrestrial sink was probably located in the tropics. Similarly, Keeling, Piper & Heimann (1989) estimated a terrestrial sink of 0·13 Pmol year–1 for the equatorial regions and only 0·05 Pmol year–1 north of 16 °N. However, from a two-dimensional inverse deconvolution of atmospheric CO2 and δ13C observations, Ciais, Tans, White et al. (1995) and Ciais, Tans, Trolier et al. (1995) came to the opposite conclusion, i.e. most of the terrestrial net carbon uptake is in the temperate/boreal zone of the Northern Hemisphere, with an estimated net uptake (30 ° N to 90 ° N) of 0·246 Pmol in 1993 (Ciais, Tans, Trolier et al. 1995; Ciais, Tans, White et al. 1995). The magnitude of this discrepancy is shown in Fig. 4 where the latitudinal gradient in modelled net uptake of CO2 by the terrestrial biosphere is shown on a land-area basis.
There are many uncertainties in such analyses, the main one being that the determination of surface sources and sinks from observations is a poorly determined inverse problem. That is, direct estimates of sources are subject to arbitrarily large errors arising from errors in any or all of the observations, the transport model or the inversion technique employed (Enting et al. 1995; Bousquet et al. 1996). A second major uncertainty is that for both oceanic and terrestrial exchange there should be a difference between the carbon isotopic composition of CO2 entering and leaving the land and ocean reservoirs (Tans et al. 1993). This is because the isotopic composition of atmospheric CO2 has become increasingly depleted in δ13C with time as fossil fuels have a (depleted) carbon isotopic signature similar to C3 plant matter (Keeling, Bacastow et al. 1989). As the mean residence times for both terrestrial and oceanic carbon are appreciable, this difference can be substantial and it significantly hampers the estimation of sources and sinks for anthropogenic carbon where the rate of change or latitudinal gradient in the δ13C of atmospheric CO2 is used to separate the oceans from the land (Tans et al. 1993; Heimann & Maier-Reimer 1996; Rayner et al. 1999). This ‘disequilibrium effect’ is particularly a problem for oceanic exchange as the local isotopic composition of air in equilibrium with surface water is a strong function of temperature and depends also on the local isotopic composition of the ocean surface dissolved inorganic carbon (Tans et al. 1993; Heimann & Maier-Reimer 1996). Despite the large effects of such uncertainties on the calculated spatial distribution of sources and sinks for CO2, there has been little attempt to determine whether the magnitudes and locations of these deduced terrestrial sources and sinks are reasonable in terms of plant physiological, ecological and biogeochemical considerations. These provide an additional constraint on the location of terrestrial sinks for anthropogenically released CO2. For example, take the typical values for northern hemisphere forest NP in Table 2 (excluding fine-root turnover), on an area weighted basis about 20 mol C m–2 year–1. The values of net terrestrial uptake supposed by (Ciais, Tans, White et al. 1995) in Fig. 4, peak at about the same value near 45 ° N. A long-term rate of net terrestrial uptake of this magnitude would require soil respiration rates to be close to zero. The same argument can be applied to a recent paper by Fan et al. (1998). They propose a massive sink of carbon in North America, 0·14 Pmol C year–1, and with most of it below 51 ° N (i.e. almost all of it in the conterminous United States of America). Model NP estimates for the conterminous USA are only about twice that amount: 0·26–0·31 Pmol C year–1 (Schimel, VEMAP participants & Braswell 1997). This means that the Fan et al. (1998) estimate requires that, averaged over the entire USA, soil respiration rates are at most only 70% of NP. I show below that from a terrestrial viewpoint, such large imbalances are hard to imagine, at least as a consequence of ecosystem response in to increasing CO2 concentrations or enhanced rates of atmospheric nitrogen deposition. And this is even as a short-term anomaly.
The response of the terrestrial biosphere to increasing CO2 concentrations
There have been many studies examining the relationship between atmospheric CO2 concentrations and plant growth. In almost all cases, increased growth in response to increases in CO2 concentrations have been observed for C3 plants: a result expected from the CO2 dependence of C3 photosynthesis for CO2 partial pressures less than about 100 Pa (Farquhar & von Caemmerer 1982).
When reasonable representations of the CO2 dependence of plant growth are combined with models describing the flow of carbon through ecosystems, all models predict that the terrestrial biosphere is accumulating carbon at the present time (Kohlmaier et al. 1987; Goudriann 1992; Taylor & Lloyd 1992; Gifford 1994; Friedlingstein et al. 1995). This is because increases in GP and NP in response to increased [CO2] are almost instantaneous but the increase in ecosystem respiration in response to these increases in productivity is not. From analytical considerations one can obtain (Taylor & Lloyd 1992; Thompson et al. 1997)
The broad principle illustrated by equation 5 is that the ability of any biome to increase is total mass (i.e. to sequester atmospheric carbon) depends on the turnover time of carbon in that ecosystem, as well as the rate of increase in NP (Taylor & Lloyd 1992). This is also clear from empirical modelling results (Friedlingstein et al. 1995).
Although equation 5 shows some important principles it only applies for a linear increase in NP and even then only when both litter fall and soil respiration are also linearly increasing. For example, a linear increase in NP with time gives
where N0 is the initial NP and a is a constant that describes the linear increase in NP expressed as a fraction of N0 per year and M0 is MP at t = 0. As t becomes large relative to τ¯E then the subtractant becomes insignificant and equation 6 becomes the simple integral form of equation 5 with dNP/dt = aNP (Thompson et al. 1997). Thus, in this simple formulation dME/dt is simply proportional to a and τ¯EN0, the latter being equivalent to the ecosystem carbon density, M0.
However, τ¯E is typically of order 10–100 years and this is about the same time-scale over which [CO2] and NP have been increasing. Thus, many ecosystems will not be accumulating carbon to the degree that equation 5 suggests. This is especially the case for tundra and boreal forest with a high τ¯E and is illustrated in Fig. 5 where dME/dt is plotted against M0 for the different ecosystems modelled by Taylor & Lloyd (1992), Polglase & Wang (1992), and Lloyd & Farquhar (1996), who looked at forest ecosystems only. As predicted, there is a positive relationship between dME/dt and M0, but both studies showed tundra to fall below the general relationship. This was also true for boreal forests modelled by Polglase & Wang (1992) and Lloyd & Farquhar (1996) and it arises because these ecosystems are characterized by long τ¯E. A further complication is that different ecosystem carbon pools have different turnover times and this affects overall behaviour.
To demonstrate this point it is mathematically easiest to work with the reciprocals of turnover times, these being equivalent to rate constants describing the relationship between the amount of carbon in a particular pool and the associated rate of carbon acquisition or loss:
where Nl, Nb, Nbo and Nr are the amount of net primary productivity going into the leaves, branches, boles and roots, respectively. Denoting the fractions of the total plant carbon in each of these as ϕl, ϕb, ϕbo and ϕr and the actual amount of total carbon in each of the organs as Ml, Mb, Mbo and Mr, then because MP= Ml/ϕl and 1/ϕl = Nll Ml–1 (and likewise for branches, boles and roots) then
8 shows that the overall turnover time of plant carbon is a function of the turnover times of the various pools, weighted according to the proportion of the total plant carbon density that they represent. This means that a plant growth model having plant ‘compartments’ with different turnover times will only give a value of dMP/dt different to a single pool where the proportions of the total plant carbon density in the various organs changes. However, this actually happens in most simple biosphere CO2 response models (e.g. Goudriann 1992; Polglase & Wang 1992). This is because it is the fraction of NP entering each pools that is assumed to remain constant. Although it has been sometimes assumed that this results in the fraction of MP in each pool also staying constant (e.g. Siegenthaler & Oeschger 1987) but this is not true.
Let us re-express equation 6 specifically for leaves only, looking at the relative increase:
where, as above, the subscript ‘l’ refers to the leaf compartment. As Nl(0)/Ml(0) = 1/τ¯l
The same applies for branches,
and so on for boles and roots.
9a–c show that, for a linear increase in net primary productivity, the relative rate of increase in a plant compartmental pool is not totally independent of the mean residence time of carbon in that compartment. The same also applies to exponential or other non-linear rates of increase in NP. Where a model has a set fraction of NP allocated to different organs, the carbon density of long residence times organs (such as boles) will initially increase at a slower rate (expressed in proportion to the initial carbon density) than fast turnover tissues such as leaves. This means that for such an allocation scheme, ϕ values will change and, from equation 8, so will τ¯P. Although the change in τ¯P is small, it significantly affects estimations of dMP/dt.
To illustrate this, I have run a simulation for a tropical rain forest, using the atmospheric CO2 record with the allocation parameters of Polglase & Wang (1992) viz τ¯l. = 1 years, τ¯br = 10 years, τ¯bo = 50 years and τ¯r = 10 years with the fractions of total annual NP going into these four pools being 0·3, 0·2, 0·3 and 0·2, respectively. Initial NP was taken to be 50 mol C m–2 years–1 and a β-value of 0·3 assumed (Polglase & Wang 1992). Results are shown in Fig. 6.
6Figure 6a shows the modelled NP and rates of litter fall for the aggregated model (four plant compartments) and a disaggregated one (one compartment). For the aggregated model, rates of litterfall are substantially less than for the disaggregated model, despite the same NP input of carbon. The difference between NP and litterfall (i.e. dMP/dt) is about 30% greater for the aggregated model, than for the disaggregated one. The changes in ϕl, ϕb, ϕbo and ϕr responsible for this are small. For example, ϕl increases from 0·0155 in 1730–0·0161 in 1980 (an increase of 4%), ϕb and ϕr both increase from 0·1036 to 0·1061 and ϕbo decreases from 0·7772 to 0·7717. Calculation of τ¯P also shows only a small change, decreasing from 19·30 to 18·95 years (Fig. 6b). Although this is a decrease of just less than 2%, it is enough to account for the 30% difference in 6Fig. 6a. This is because dMP/dt in the above simulation is only about 2·5% of NP. From equation 2a dMP/dt is the difference between much two larger terms, and so small changes in either of these cause a much greater proportional change in dMP/dt. The reason that the decreased τ¯P in 6Fig. 6b does not totally wipe out the CO2 fertilization response is that MP is itself smaller for the disaggregated example (1012 vs 1025 mol C m–2). In this simple model it is MP/τ¯P that determines the rate of litterfall.
Thus, allocation schemes can give rise to subtle changes in plant allometric relationships which have large effects on modelled rates of CO2 sequestration. However, for plant allocation schemes where allometric relations are kept constant (as opposed to the fractions of NP allocated being the conserved quantity) there is no effect. This was shown empirically by Lloyd & Farquhar (1996) and can also be seen from equation 8.
Although early models of soil carbon dynamics assumed a single carbon pool (e.g. Jenny 1949), soil carbon has a wide spectrum of turnover times (Post 1993). To deal with this, modern phenomenological models of soil carbon dynamics have at least three pools: short turnover time carbon in living microbes (microbial carbon), an intermediate turnover time compartment (‘slow organic carbon’) and a ‘passive (or recalcitrant) soil organic carbon’ fraction which is formed from turnover of microbial and slow soil organic carbon. The separation of soil carbon into these three pools is common to both the Century (Parton et al. 1987) and Rothamsted (Jenkinson & Raynor 1977) models. Both models also separate litter carbon into a decomposable fraction and a resistant fraction.
The principles of aggregation and disaggregation previously discussed for plants applies equally well to soil carbon. Indeed, the passive carbon pool responds on a time scale of thousands of years but leaf litter pools may turn over in less than a year. This means that effects of aggregation on modelled soil carbon response to changes in litter inputs can be large. For example, running an ecosystem carbon dynamics model for various forest types using the atmospheric CO2 record, Lloyd & Farquhar (1996) showed that a disaggregated carbon pool responded much more slowly to increased litter inputs than an aggregated one, with rates of soil carbon increase varying twofold with aggregation scheme.
Thus, in scenarios where carbon dioxide concentrations are supposed to stabilize such as those presented by the Intergovernmental Panel of Climate Change (IPCC), with models considering carbon uptake in response to CO2 fertilization (Enting, Wigley & Heimann 1994) the modelled temporal pattern of biosphere response to increased CO2 concentrations (and hence deduced permissible emissions where this is the purpose of the exercise) will be strongly dependent on the degree of aggregation in the biosphere model structure. Uncertainties in this sort of biosphere modelling are so great that it is not reasonable to suppose that one can deduce the nature of a terrestrial sink (CO2vs N fertilization, for example) solely on the basis of a modelled temporal pattern vs that inferred by the atmospheric CO2 record (Friedlingstein et al. 1995).
Global estimates of CO2 fertilization
There have now been several studies attempting to determine the magnitude of terrestrial carbon sequestration in response to increases in atmospheric [CO2] (Kohlmaier et al. 1987; Goudriann 1992; Taylor & Lloyd 1992; Gifford 1994; Friedlingstein et al. 1995). Although variations in the modelled magnitude are considerable (from less than 0·04 to more than 0·3 Pmol C year–1), all models agree that some uptake should be occurring. This is not surprising. Any increase in NP in response to increased [CO2] combined with a residence time greater than a few years will give rise to some modelled sequestration. Where models differ is in the assumed residence times and ecosystem net primary productivities as well as the sensitivity of NP to changes in CO2 concentration and how it may be modified by ‘nutrient limitations’ and the degree to which carbon pools are aggregated. Some models (e.g. Taylor & Lloyd 1992; Gifford 1994; Friedlingstein et al. 1995) assumed a single carbon pool, and from the discussion above it seems likely that these models resulted in unduly high current rates of carbon accumulation in the soil. On the other hand, models which simply apportion a set fraction of NP to various plant organs (e.g. Polglase & Wang 1992) probably underestimate the rate of increase in plant carbon.
Rather than estimating the response of plant growth to changes in [CO2] by using a simple β factor, Lloyd & Farquhar (1996) expressed NP in relation to Gross Primary Productivity (ecosystem photosynthesis) as
where GP is the Gross Primary Productivity, φo is the proportion of carbon that is consumed in the conversion of recently fixed carbon to structural biomass and m is a whole-plant maintenance respiration coefficient. By simulating the response of GP to increasing atmospheric [CO2], this allows the dependency of NP on [CO2] and MP to be simulated in a realistic manner. As was noted at the time, the estimates of NP used did not include fine-root turnover. A recent study has estimated that, on a global basis this accounts for about 33% of NP (Jackson, Mooney & Schulze 1997). This underestimation of NP is of little consequence for estimation of dME/dt as fine roots are characterized by a very fast turnover time (< 1 years) and thus higher estimates of NP when fine-root production is allowed for is almost completely balanced by a reduction in τ¯P.
Nevertheless, in what follows, calculations of Lloyd & Farquhar (1996) are repeated, allowing for an extra-fast turnover (1 year) fine-root pool, with commensurate higher values of NP. The new estimates of NP for coniferous forests are also used from the re-analysis in Table 2. The analysis is also extended to other ecosystems, also considering interactions between the nitrogen and carbon cycles.
Interactions of CO2 fertilization with ecosystem nitrogen status
It has long been recognized that soil nitrogen can exert a profound effect on plant growth, and it has been widely assumed that when plants are growing under conditions of suboptimal nitrogen nutrition, the ability of these plants to grow faster in response to elevated carbon dioxide concentration is reduced (e.g. Bazzaz 1990). Nevertheless, as has been pointed out by Idso & Idso (1994) and Lloyd & Farquhar (1996) evidence for this from glasshouse experiments is weak and when expressed as a proportional stimulation in response to a doubling of CO2 concentration, low-nitrogen plants are observed to have higher growth stimulations (compared to well-fertilized plants) almost as often as they have lower growth enhancements (Lloyd & Farquhar 1996; Poorter 1998). Much of the ‘evidence’ for a reduced CO2 growth response at low-nitrogen nutrition comes from modelling experiments which assume conserved C/N ratios in plant and soil tissue (Comins & McMurtrie 1993; Melillo et al. 1993). However, plant C/N ratios (or those of individual plant organs) are not constant, increasing with plant size (Lloyd & Farquhar 1996).
Many models also assume that soil humus and litter C/N ratios are also conserved quantities. However, examination of available data for temperate forest and grasslands shows that this not the case (Fig. 7). Ecosystems with a high soil C density have higher C/N ratios than the same ecosystem at low C densities.
First, the dependence of photosynthetic rate (per unit leaf carbon) is considered a saturating function of leaf nitrogen content (also expressed per unit leaf carbon), using the relationship defined by Cromer, Kriedemann et al. (1993) viz.
where GP(s) is the gross primary productivity at saturating nitrogen concentration, k is a constant, [Nl] is the average leaf nitrogen concentration and [No] is the foliar leaf nitrogen concentration below which no photosynthesis is observed. Across a range of leaf types, [No] seems quite constant at around 0·010 mol N mol–1 C (Field & Mooney 1986; Cromer, Kriedemann et al. 1993; Medina & Francisco 1994; Reich et al. 1994). Whether there is a generic value for k is less clear, but lacking evidence for the contrary, the value of Cromer, Kriedemann et al. (1993) of – 53·2 mol–1 N is used. As in Lloyd & Farquhar (1996), changes in GP(s) with changes in leaf area are accounted for. Available data on the relationship between total canopy leaf nitrogen and total canopy leaf carbon in Lloyd & Farquhar (1996) (their Fig. 16) is taken as a standard case. That is to say, for any canopy nitrogen concentration:
where GP* is the gross primary productivity modelled to occur without consideration of leaf nitrogen concentrations, eqn 11 of Lloyd & Farquhar (1996), and [Nl*] is the associated ‘normal’ leaf nitrogen concentration at the modelled canopy carbon density, Fig. 16 of Lloyd & Farquhar (1996).
Plant nitrogen uptake (U) is modelled as a simple Michaelis–Menton dependence
where [NS] is the plant available soil nitrogen concentration, Mfr is the density of fine roots (soil area basis), γmax is the maximum rate of N uptake per unit fine root (mol N mol–1 C) and KN is a Michaelis–Menton constant. Equation
12 of Lloyd & Farquhar (1996) is modified to allow for changes in the nitrogen content of the soil carbon pools and separating inputs into that from plant litter and directly from the atmosphere, i.e.
where I is the rate of input into the available soil N pool from the atmosphere (wet and dry deposition), [NL] is the nitrogen concentration in litterfall, LP is the rate of litterfall, LN is the rate of loss of N from the system as a result of leaching, [Nlit] is the nitrogen concentration in plant litter (mol N mol–1 C), Mlit is the amount of carbon present in plant litter (soil area basis), [Nhum] is the nitrogen concentration in soil humus, Mlit is the amount of carbon present in soil humus, [Nmic] is the nitrogen concentration in the soil microbe pool and Mlit is the amount of carbon present in the soil microbe pool. The use of the last three terms in equation 14 follows from the models of Townsend et al. (1996) and Holland et al. (1997).
Two cases are considered. First, as in Townsend et al. (1996) and Holland et al. (1997) [Nlit], [Nhum] and [Nlit] are taken as constants, using values from Table 1 of Townsend et al. (1996). In the second case [Nhum] is allowed to change according to the relationships of Fig. 7. In this case, an increase in Mfr in response to reductions in leaf nitrogen status (induced by CO2 stimulated growth) is also allowed for. This is achieved by adjusting Mfr (initially taken as 20% of the total root density) in proportion to [Nl]/[Nl*]. There is then the question of how to distribute nitrogen throughout the plant. This is achieved by assuming that the relatively small concentrations of nitrogen in boles, shoots and roots represent a fixed requirement for metabolic function that must be fulfilled and that only the foliar nitrogen pool is allowed to vary its concentration on the basis of anything other than the total carbon pool size. From some fertilizer trials there is some evidence that this is a reasonable approach (e.g. Birk & Turner 1992) but other studies are not always consistent in this respect (e.g. Heilman & Gessel 1963). The results are compared with a third case where plant growth in response to changes in past CO2 concentrations is independent of any changes in plant or soil nitrogen status.
These simulations have been carried out with a low pre-industrial rate of atmospheric nitrogen input into these ecosystems (2 kg N ha–1 year–1 = 0·014 mmol N m–2 years–1). The model is initialized assuming equilibrium conditions in 1730. This is also carried out for the nitrogen pools, requiring that atmospheric inputs are exactly balanced by leaching out of the system. Thus having prescribed I, and writing LN = kL[Ns], the initialization solves for kL as a time-invariant constant.
Modelled changes in net primary productivity and soil respiration for the three simulations for a temperate zone deciduous forest are shown in Fig. 8. This shows a large effect of nitrogen limitation assumptions on modelled changes in net primary productivity, soil respiration and hence their difference (the net ecosystem productivity, NEP, equal to the net rate of carbon accumulation in plant and soil). Where no carbon/nitrogen interactions are considered the increase in NP is 18%, but where constant carbon/nitrogen ratios in soil and no change in root/shoot ratio are assumed it is less than 3%. For the intermediate case (i.e. changing C/N ratios in soil organic matter and increased fine-root production as canopy nitrogen status declines) the modelled increase is about 7%.
There are some small changes in the overall mean residence time for ecosystem carbon between the different models, but in all cases soil respiration lags net primary productivity by about 60 years. The proportional differences in modelled carbon sequestration (NP–RS) are therefore similar to the proportional differences in modelled stimulation of NP in response to increased atmospheric CO2 concentrations: 3·1 mol C m–2 year–1 for the ‘no-nitrogen limitation’ scenario, 0·5 mol C m–2 year–1 for the constant C/N ratio scenario and 1·3 mol C m–2 year–1 for the intermediate (changing soil C/N ratio and root/shoot ratio) scenario.
Figure 9 shows changes in the modelled canopy nitrogen concentration relative to 1760. Where constant soil C/N ratios and root/shoot ratios are assumed, leaf N concentrations in 1985 are modelled to be only 88% of those in 1760 whereas for the intermediate (changing soil C/N ratio and root/shoot ratio) scenario this reduction is only to about 92%. Similarly, if one looks at modelled losses of N through leaching, there is little change where constant soil C/N ratios and root/shoot ratios are assumed, but a significant reduction is observed for the case of the changing soil C/N ratio and root/shoot ratio scenario (not shown). This is mostly a consequence of enhanced fine-root production resulting in appreciably more N uptake by the vegetation.
It is hard to know which scenario is the most realistic, but the truth probably lies somewhere between the no-nitrogen limited case and the intermediate one. This is because the latter does not allow for an enhancement of ecosystem nitrogen fixation as atmospheric CO2 increases. This is expected on theoretical grounds (Lloyd & Farquhar 1996) and is consistent with changes in herbarium specimen and tree-ring δ15N content (Penuelas & Estiarte 1997).
Although the simulations of Figs 8 and 9 indicate that it is possible for nitrogen availability to constrain ecosystem CO2 responses, it is not a corollary that ecosystems with lower levels of nitrogen availability should have a lower relative sensitivity to increases in atmospheric carbon dioxide concentrations, even though this is sometimes assumed (e.g. Friedlingstein et al. 1995). This is because there are other complicating effects. For example, low-nitrogen plants tend to have higher respiratory costs relative to their rate of carbon assimilation and this increases their sensitivity to increased atmospheric CO2. Also, they tend to have lower leaf areas and thus, reinvestment of any CO2 stimulated growth into extra leaf area with commensurate increases in light interception tends to amplify their proportional growth responses as compared to plants with higher leaf areas. However, a greater sensitivity to lower foliar N concentrations, such as is observed in 9Fig. 9a, tends to act in the opposite direction and thus the interaction can go either way. A more complete discussion of these issues is given in Lloyd & Farquhar (1996), Poorter (1998) and Lloyd & Farquhar (1999).
Combining effects of nitrogen deposition and CO2 increase
The changing soil C/N ratio and plant root/shoot ratio developed and examined above is consistent with observations of fine-root growth responses and consequent changes in N uptake ability with changes in CO2 concentration (BassiriRad et al. 1996; Tingey et al. 1997) as well as successfully predicting a small decrease in foliar N concentrations over the last century (Penuelas & Estiarte 1997). In what follows the same model is, therefore, used to evaluate the effects of deposition of anthropogenically produced nitrogen on carbon sequestration by vegetation.
Using such a model has advantages because it does not assume that plant C/N ratios are constant (cf. Schindler & Bayley 1993; Townsend et al. 1996; Holland et al. 1997). Generally speaking these decline when nitrogen deposition occurs. For example, using British herbarium specimens and current-day samples of bryophytes Pitcairn, Fowler & Grace (1995) showed that over about 40 years, tissue nitrogen concentrations had increased by 38–62% for sites with high rates of anthropogenic N deposition, but that no change was observed for a remote site. They also showed that foliar nitrogen concentrations for Calluna vulgaris were more than 100% higher in areas characterized by high N deposition rates (> 15 kg N ha–1 years–1) than for remote sites. Cape et al. (1990) and Lumme, Arkipov & Kettunen (1995) showed that for Picea abies and P. sylvestris foliar N concentrations increase with increasing N deposition.
A model of ecosystem and plant growth response to nitrogen inputs should therefore simulate more foliage with increased N deposition, but with that foliage also having a higher N concentration than when lower rates of N deposition occur. In response to high rates of nitrogen deposition it should also predict higher fractions of nitrogen inputs leaving ecosystems as leachates (Tietema et al. 1997).
The model is tested by observing the simulated growth stimulation for a mature deciduous forest in response to nitrogen deposition rates of 10, 30 or 60 kg N ha–1 year–1 in 1990, with or without additional CO2 fertilization. As was carried out by Townsend et al. (1996) and Holland et al. (1997) it is assumed that the history of N deposition follows the same pattern as fossil-fuel CO2 emissions. Results for this simulation are shown in Fig. 10. This shows that for even moderate rates of anthropogenic nitrogen deposition (10 kg N ha–1 years–1 in 1990) modelled growth stimulations are substantially greater than for CO2 alone, an increase of 25%vs 7% being observed. Allowing for the simultaneous CO2 concentration increase as well increases this growth stimulation to about 40%. Thus effects of increased nitrogen deposition and atmospheric [CO2] increase are not simply additive. This is also the case for higher rates of nitrogen deposition (30 and 60 kg N ha–1 years–1), but here there is no difference between the two higher rates in the simulated net primary productivity for 1990. There is, however, a different temporal pattern between these two treatments, with the 60 kg N ha–1 years–1 increasing initially at a much faster rate. Thus at this very high rate, the model predicts that some ‘saturation’ of the response of net primary productivity to nitrogen deposition is already occurring.
This is even more marked when the rate of ecosystem carbon accumulation is examined (Fig. 11). This shows that, although initially greater at 60 kg N ha–1 year–1, rates of carbon sequestration are simulated to be less than at 30 kg N ha–1 year–1 after about 1970. Thus, the high carbon sequestering potential of forests with very high rates of N deposition is predicted to be rapidly disappearing. Compared to CO2 fertilization alone (1·3 mol C m–2 year–1), the additive effect of moderate rates of nitrogen deposition on the modelled carbon sequestration is substantial, ranging between 8·0 mol C m–2 year–1–9·8 mol C m–2 year–1 for a deposition rate of 30 kg N ha–1 year–1.
Concentrating on the 30 kg N ha–1 year–1 simulation the CO2/nitrogen interaction is probed further in 12Fig. 12a and b. 12Figure 12a shows the modelled change in leaf nitrogen relative to 1760 in the presence and absence of the additional CO2. Independent of CO2, leaf nitrogen concentrations are modelled to currently be 25% higher than before the industrial revolution and differences of this magnitude are consistent with transect studies (Cape et al. 1990; Lumme et al. 1995). Interestingly, the rate of increase in average canopy N concentration is initially simulated to be higher in the absence of CO2, illustrating a realistic representation of the CO2 effect of canopy N concentrations (discussed above). The temporal pattern of the simulated nitrogen deposition is shown in 12Fig. 12b, along with the associated calculated rates of leaching. This shows that in the absence of a concurrent atmospheric CO2 increase, rates of ecosystem nitrogen loss would be larger than for the CO2 increase occurring in conjunction with N deposition. This is a consequence of the greater tree (and hence fine root) biomass for the latter. The calculated proportions of atmospheric N input retained by the ecosystem is (with CO2 included) is 0·65, 0·40 and 0·32 for 10, 30 and 60 kg N ha–1 years–1, respectively, illustrating that less retention at high rates of N deposition (Tietema et al. 1997) are successfully reproduced by the model. This is even before ecosystem dysfunction at damaging rates of N deposition is taken into account (Durka, Schulze & Voekelius 1994).
The modelled stimulation of net primary production by the combined CO2 concentration and atmospheric nitrogen deposition increase in Fig. 10 is high (an increase of around 50% compared to pre-industrial revolution values), but compared to some forest nutrition experiments this stimulation is relatively small (e.g. Cromer, Cameron et al. 1993) and its magnitude is broadly consistent with recent changes in net annual increment and site index for many European forest research plots (Spiecker et al. 1996), perhaps even being a little bit on the low side (Pretzsch 1996).
The model gives similar results for coniferous forests and is easily applied to other ecosystems. To do this, typical estimates of NP have been taken from the literature, for herbaceous biomes assuming mean residence times of 1 year for foliage, 2 years for structural roots and 1 year for fine roots. The same root and shoot carbon/nitrogen relationship as is applied above for temperate deciduous forests is used. For deserts and shrublands it is assumed the same mean residence times and C/N relationships as for tropical-forest trees (although these trees are, of course smaller) and for savannahs I have simply assumed that 60% photosynthesis is by C4 grasses and 40% by C3 trees, using the model of Collatz, Ribas-Carbo & Berry (1992) to simulate the photosynthetic dependence of C4 grasses on atmospheric CO2. The latter turns out to be essentially zero in this simulation. This is probably an underestimate as effects of increases in [CO2] on C4 grass water-use efficiency are ignored. As was carried out by Melillo et al. (1993) I have also made one important assumption: that the high nitrogen availability in tropical-forest soils (Vitousek & Sandford 1986) means that no significant C/N interactions occur. Consistent with high relative abundances of nitrogen-fixing trees and shrubs in savannahs and other non-temperate woody ecosystems (Högberg 1986; Sprent et al. 1996; Zitzer et al. 1996; Schulze et al. 1998) this is also considered to be the case for savannahs, shrublands and deserts.
Simulations have been carried out for all biomes for the case of the effects the historical increase in atmospheric [CO2] on dME/dt as well as the combined effects of nitrogen deposition and atmospheric [CO2] increase using a rate of 30 kg N ha–1 year–1 for 1990 for mid-latitude temperate biomes where this is considered likely to occur. No effect of N deposition of dME/dt is simulated for cultivated land as it is usually considered that these receive large inputs of nitrogen through fertilization (Holland et al. 1997). Estimates of NP (CO2 fertilization only) and dME/dt (with additional nitrogen deposition calculations where appropriate) are shown in Table 3.
Table 3. . Modelled average net primary productivities (NP) and rates of ecosystem carbon accumulation (dME/dt) for various biomes
This shows that in the absence of appreciable nitrogen deposition tropical forests are likely to be the major sink for CO2 on ground area basis. As discussed by Lloyd & Farquhar (1996), this is attributable to higher temperatures during CO2 assimilation leading to a greater sensitivity of photosynthesis to atmospheric CO2 concentrations, high respiratory losses in tropical ecosystems giving rise to high growth enhancements as a consequence of these photosynthetic increases, and overall, high productivities combined with reasonably long carbon residence times. The modelled rate of sequestration is consistent with the direct measurements of Grace et al. (1995). About 70% of this sequestration is modelled to be above ground (Lloyd & Farquhar 1996) and thus the estimate for above-ground carbon accumulation of Amazon forests from extensive inventory data of about 5 mol C m–2 years–1 for the period 1970–1990 (Phillips et al. 1998) is also consistent with the model presented here.
Even in the absence of assumed nitrogen limitations, dME/dt for temperate and boreal forests is modelled to be much less than in the tropics (Lloyd & Farquhar 1996), but this effect is even more amplified in the simulation here where nitrogen effects are included. Indeed, it is interesting to note that cultivated land has a modelled value of dME/dt greater than for boreal forests. Of course, for the former, almost all this accumulation is modelled to be in the soil. It is sometimes assumed that agricultural systems cannot sequester carbon over the long term, but there is no reason why this should be the case. Although it is true that cultivation usually reduces soil carbon stocks, this usually occurs over decadal time scales (Oades 1988; Lefroy, Blair & Caboche 1993), and much of the world’s cultivated area, especially in the temperate zone, was converted to agriculture before the turn of this century. As would be expected, dME/dt for temperate grasslands is modelled to be small, and almost all of this is below ground.
The ranking of biomes is very different when N deposition is considered to occur as well. In this case, temperate forests are modelled to have values for dME/dt that are comparable with tropical forests. This rate of accumulation is lower for the boreal forest types and, importantly, moderate rates of carbon sequestration (2·1 mol C m–2 year–1) are modelled to occur even for temperate zone grasslands. Again, almost all of this is below ground.
In order to estimate the global total carbon sequestration as a result of CO2 fertilization and nitrogen deposition, the results of Table 3 have been combined with NP estimated from a model of plant productivity with global coverage (J. Lloyd & F. Kelliher, unpublished data). For each 1°× 1° grid square in this model, the calculated NP has been obtained and dME/dt for each grid square scaled according to the ratio of the modelled NP of that grid square to the modelled NP for the biome in that grid square as listed in Table 3. This approach effectively assumes that the mean residence time of carbon for any particular biome is independent of the NP or carbon density.
A second issue is the amount of anthropogenic nitrogen deposition. Although there are now numerous models of the atmospheric nitrogen cycle, there is also considerable disagreement amongst models in terms of rates of N deposition onto terrestrial vegetation (Holland et al. 1997). Furthermore, with the exception of Dentener & Crutzen (1994), ammonia deposition is not considered in these models. The model resolution of Dentener & Crutzen (1994) is also rather coarse, 5°× 5°. In view of these large differences and uncertainties in model output, it has been simply assumed that some areas are exposed to anthropogenic N deposition and some are not. The distribution of areas exposed to anthropogenic N deposition was done of the basis of visual inspection of figures in Holland et al. (1997), with countries and/or their constituent states (where available) assigned as either being exposed to N deposition or not, using the global data base of Lerner, Matthews & Fung (1988). For grid squares where N deposition is assumed to occur the values of dME/dt appropriate to a deposition rate of 30 kg N ha–1 year–1 have been used.
Results of the simulation are shown in Fig. 13. This shows the planet to consist of several ‘hot spots’ of carbon accumulation, corresponding to either areas of characterized by high N deposition, mostly for forests or grasslands in Europe, China and North America or to areas dominated by tropical rain forest in Africa, Asia and South America. However, in this simulation much of the rest of the world’s surface is characterized by small rates of carbon accumulation: less than 2 mol C m–2 years–1. Indeed, about 15% of the earth’s surface accounts for over 80% of the modelled sequestration. However, the rates of carbon sequestration in high N deposition areas are significantly less than either Townsend et al. (1996) or Holland et al. (1997), even though the additional CO2 fertilization effect has been taken into account. The total global sink modelled here is 0·194 Pmol C year–1 with about half of this in the tropics and southern hemisphere and about half in the northern hemisphere.
The latitudinal distribution of the modelled terrestrial sink is shown in Fig. 14. This shows a moderate sink throughout much of the northern hemisphere, but a much stronger localized sink in the tropics, especially south of the equator. Shown as well is a recent estimate of the net terrestrial carbon sink (including deforestation) based on a Bayesian synthesis inversion of atmospheric CO2,13CO2 and O2/N2 ratios (Rayner et al. 1999) and an estimate of the latitudinal distribution of the deforestation carbon flux from Enting et al. (1995). Also included is the terrestrial carbon sink as inferred by (Ciais, Tans, White et al. 1995). Removing the deforestation flux reveals the net uptake of CO2 by the terrestrial biosphere (‘biotic flux’) as inferred by Rayner et al. (1999) and Ciais, Tans, White et al. (1995). Comparison of the estimate of Rayner et al. (1999) with that of the model presented here shows remarkable agreement especially when the crudeness of the N deposition fields and other uncertainties is considered. Also in broad agreement with the study here and Rayner et al. (1999) is the work of Keeling et al. (1996). Investigating the latitudinal gradient in O2/N2 ratios, they deduced an northern hemisphere sink of 0·16 ± 0·08 Pmol C year–1 for the period 1991–1994, also concluding that the tropics were neither a net source nor a net sink for CO2. That is to say, that any releases of CO2 from tropical deforestation must have been offset by CO2 uptake occurring elsewhere in the tropics: in broad agreement with Fig. 14 as well as in agreement with other work (Grace et al. 1995; Phillips et al. 1998).
There is no such agreement between these results and those of (Ciais, Tans, White et al. 1995). This is probably attributable, at least in part to their only performing a two-dimensional (as opposed to a three-dimensional) deconvolution of their atmospheric CO2 concentration and δ13C observations. Subsequent work by the same group and comparing the two model types has indicated that the studies of Ciais, Tans, White et al. (1995) and Ciais, Tans, Trolier et al. (1995) probably overestimated the northern hemisphere sink in 1992/1993 by around 30% (Bousquet et al. 1996). This was a consequence of less efficient transport of anthropogenic CO2 out of the mid-northern latitudes in the two-dimensional model requiring a larger sink for CO2 in this region. It is not clear, however, whether this overestimation should have applied equally to both terrestrial and oceanic sinks. If anything, it is the oceanic sink which would have been overestimated the most, as a net carbon flux into the ocean has an isotopic signature most similar to atmospheric transport. A second consideration is that the period studied by Ciais, Tans, White et al. (1995) and Ciais, Tans, Trolier et al. (1995) seems to have been an unusual one in the recent history of the carbon cycle, with the globally averaged rate of increase and magnitude of the north-to-south gradient in concentrations both being unusually low (Conway et al. 1994). At least in part, this could be as a result of lower ecosystem respiration rates owing to the lower northern hemisphere temperatures observed after the Mount Pinatubo eruption. For example, from the model of Lloyd & Taylor (1994) the 0·7 °C decrease in northern hemisphere temperatures following the Mount Pinatubo eruption (Dutton & Christy 1992) would have decreased respiration rates by about 5%. Assuming that this applied to both autotrophic and heterotrophic respiration, then for an ecosystem respiration rate of 4 Pmol C year–1 for ecosystem respiration north of 20°N (suggested by the model used here) then the associated decrease in ecosystem respiration would have been about 0·2 Pmol year–1, enough to explain some of the differences between Ciais, Tans, Trolier et al. (1995) and the study here. On the other hand, Ciais, Tans, White et al. (1995) also inferred a rather large northern hemisphere terrestrial sinks in 1990 (0·40 Pmol between 30°N and 65°N), before the Mount Pinatubo eruption, suggesting a more fundamental difference. Perhaps then the problem relates to the use of carbon isotopes to partition fluxes between land and ocean. As was discussed above, and in more detail by Tans et al. (1993) and Heimann & Maier-Reimer (1996), uncertainties in the disequilibrium terms, especially for the oceans, means that gradients in the carbon isotopic composition of atmospheric CO2 cannot provide an unambiguous constraint on terrestrial vs oceanic uptake.
The study of Fan et al. (1998) is also in contrast to the one here. However, when one compares their estimated sink for North America of 0·14 Pmol with current estimates of NP for the USA of around 0·3 Pmol year–1 (Schimel et al. 1997), the magnitude proposed by Fan et al. (1998) has a value equal to at least 30% of net primary productivity (even after allowing for some uptake in Canada). However, from the study here (Table 3), even with appreciable rates of nitrogen deposition a ratio of terrestrial sink to net primary productivity of more than about 20% is unlikely and averaged over all of North America a value of about 10% is more credible. Thus, as is the case for the study of Ciais, Tans, White et al. (1995) and Ciais, Tans, Trolier et al. (1995), the analysis presented here does not provide much support for the highly localized terrestrial CO2 uptake in North America as proposed by Fan et al. (1998).
Finally, note that the temperate and tropical sinks are likely to respond in different ways the future. In both cases, of course, some saturation must eventually occur. Given that the temperate sink is a consequence of both increased nitrogen deposition and increased CO2, it seems likely that it might be the most persistent. On the other hand, the analysis here suggests that some saturation of the nitrogen response might already be occurring in areas subject to high rates of deposition, and eventual ecosystem dysfunction as a consequence of continual high rates of nitrogen inputs is likely (Durka et al. 1994). Simulations of future carbon sequestration would also have to take into account changes in the terrestrial carbon balance owing to the increased temperatures considered likely to occur over the next century. The exact nature of the temperature sensitivities of plants and the various soil carbon pools is still a matter of debate. A further consideration for forest ecosystems in the northern hemisphere is that many of these are managed for wood production. In that context, the analysis here is somewhat callow as it ignores the harvesting and regrowth of these forests. Consideration of this may not change the overall picture substantially, but in the post-Kyoto world, the development of integrated forest production models that also account for changes in CO2 and nitrogen inputs and their biogeochemical effects on both above- and below-ground carbon stocks would be a valuable step forward. This would provide good guidance as to appropriate ways to manage forests and plantations as to maximize their abilities for carbon sequestration: a factor that, over the next century, has the potential to also have an impact on the atmospheric carbon balance.
Thanks to Simon Barry for advice on the statistical analyses used in Table 3 and Frank Berninger for useful discussions on forest productivity and for help with obtaining the forest yield data used in Figs 2 and 3. Ian Enting, Philippe Ciais and Peter Rayner generously provided their inversion study model outputs. Frank Berninger, Martin Heimann, Frank Kelliher and Colin Prentice all provided valuable comments on an earlier draft of the manuscript.