The simulation of atmospheric–land–ocean CO2 exchange for the 1850–2000 period offers the possibility of testing and calibrating the carbon budget in earth system models by comparing the simulated changes in atmospheric CO2 concentration and in land and ocean uptake with observation-based information. In particular, some of the uncertainties associated with the treatment of land use change (LUC) and the role of down regulation in affecting the strength of CO2 fertilization for terrestrial photosynthesis are assessed using the Canadian Centre for Climate Modelling and Analysis Earth System Model (CanESM1). LUC emissions may be specified as an external source of CO2 or calculated interactively based on estimated changes in crop area. The evidence for photosynthetic down regulation is reviewed and an empirically based representation is implemented and tested in the model. Four fully coupled simulations are performed: with and without terrestrial photosynthesis down regulation and with interactively determined or specified LUC emissions. Simulations without terrestrial photosynthesis down regulation yield 15–20 ppm lower atmospheric CO2 by the end of the twentieth century, compared to observations, regardless of the LUC approach used because of higher carbon uptake by land. Implementation of down regulation brings simulated values of atmospheric CO2 and land and ocean carbon uptake closer to observation-based values. The use of specified LUC emissions yields a large source in the tropics during the 1981–2000 period, which is inconsistent with studies suggesting the tropics to be near-neutral or small carbon sinks. The annual cycle of simulated global averaged CO2, dominated by the Northern Hemisphere terrestrial photosynthesis and respiration cycles, is reasonably well reproduced, as is the latitudinal distribution of CO2 and the dependence of interhemispheric CO2 gradient on fossil fuel emissions. The empirical approach used here offers a reasonable method of implementing down regulation in coupled carbon–climate models in the absence of a more explicit biogeochemical representation.
Traditional climate change simulations are performed with specified concentrations of atmospheric CO2 and other greenhouse gases (GHGs), and the physical state of the land surface is specified in terms of the fractional coverage of vegetation, its leaf area index (LAI), roughness length, and rooting depth. These restrictions are removed when terrestrial and oceanic ecosystem components are introduced into coupled general circulation models (CGCMs) that explicitly simulate land– and ocean–atmosphere fluxes of CO2 and other GHGs and their evolution. The GHGs, in this framework, become prognostic climate-dependent quantities as do the structural attributes of vegetation. A variety of coupled climate–carbon cycle models exist including comprehensive models based on CGCMs and earth system models of intermediate complexity (EMICs) (Friedlingstein et al. 2006).
Simulations with the models participating in the Coupled Climate–Carbon Cycle Model Intercomparison Project (C4MIP) indicate that climate warming due to increasing GHGs will decrease the capacity of the ocean and terrestrial biosphere to sequester carbon, leading to higher atmospheric CO2 and stronger warming in a positive climate feedback loop (Friedlingstein et al. 2006). The C4MIP experiment concentrates on the consequences of CO2 emissions (although not those of other GHGs) through 2100. There are only a few published reports that assess model performance against observed atmospheric CO2 concentrations and against inverse model estimates of terrestrial and oceanic carbon uptake for the twentieth century (e.g., Matthews et al. 2005). Here, we analyze the twentieth-century performance of the Canadian Centre for Climate Modelling and Analysis (CCCma) Earth System Model (CanESM1), which is a comprehensive carbon–climate model based on the CGCM3 coupled land–ocean–atmosphere model (Flato et al. 2000) used to generate climate change scenarios for the recent Intergovernmental Panel on Climate Change (IPCC) report (Randall et al. 2007). The focus here is on the biogeochemical behavior of the model: we compare the simulated evolution of the global mean atmospheric CO2 concentration and its seasonal and latitudinal distribution with observation-based estimates. We also compare simulated land– and ocean–atmosphere fluxes of CO2 including latitudinal distributions with estimates from inversion and other studies.
CanESM1 is a fully coupled climate–carbon earth system model. Fully dynamical three-dimensional physical atmosphere and ocean components are coupled with the Canadian Terrestrial Ecosystem Model (CTEM), which provides the terrestrial carbon component, and to the Canadian Model of Ocean Carbon (CMOC), which provides the ocean carbon cycle component. The simulation of climate change over the twentieth century offers the possibility of testing and calibrating the carbon budget in the model against observed changes in atmospheric CO2 concentration and land and ocean uptake. Here, we assess the effect of some of the uncertainties associated with the treatment of land use change (LUC) emissions and the strength of CO2 fertilization of terrestrial photosynthesis. LUC emissions may be specified as an external source of CO2 or calculated interactively based on estimated changes in crop area. The modeled atmospheric CO2 increase over the twentieth century is low compared to observations regardless of the treatment of LUC emissions. This suggests that some process is lacking in the model, such as nutrient limitation, that regulates photosynthesis in nature. Experimental studies in which plants are grown at ambient and elevated CO2 levels are reviewed and provide evidence for down regulation. An empirical representation of down regulation based on these results is implemented and tested in the model, which brings modeled late-twentieth-century atmospheric CO2 much closer to observed.
2. Model description and datasets
a. The physical atmosphere
The atmospheric component of the CanESM1 is the CCCma third generation atmospheric general circulation model (AGCM3) described in detail by McFarlane et al. (2005) and Scinocca et al. (2008, manuscript submitted to Atmos. Chem. Phys.). The version used here employs a horizontal resolution defined by a triangular truncation at total wavenumber 47 (T47) for dynamical terms. The physical terms are calculated on a 96 × 48 (∼3.75°) horizontal linear grid. In the vertical, the model domain extends to 1 hPa atmospheric pressure with the thicknesses of the model’s 32 layers increasing monotonically with height from approximately 100 m at the surface to 3 km in the lower stratosphere.
While many of the parameterized physical processes in the third-generation model are qualitatively similar to AGCM2, key new features include 1) a new parameterization of cumulus convection (Zhang and McFarlane 1995), 2) an improved treatment of solar radiation employing four bands in the visible and near-infrared region, 3) an “optimal” spectral representation of topography (Holzer 1996), 4) a revised representation of turbulent transfer coefficients at the surface (Abdella and McFarlane 1996), 5) a hybrid moisture variable (Boer 1995), 6) the Canadian Land Surface Parameterization Scheme (CLASS) for treatment of the land surface processes (Verseghy 1991; Verseghy et al. 1993), and 7) a new anisotropic orographic gravity wave drag parameterization (Scinocca and McFarlane 2000).
1) Greenhouse gases
CO2 is modeled as a three-dimensional tracer whose concentration depends on spatially distributed anthropogenic CO2 emissions, atmospheric transport, and the interactive land– and ocean–atmosphere exchange of CO2. CO2 emissions are based on the Emission Database for Global Atmospheric Research (EDGAR) V1.3 (1890–1960) and EDGAR V3.2 (1970–90 and 1995) datasets (Van Aardenne et al. 2001) (http://www.mnp.nl/edgar/model/). The categories included are fossil fuel consumption, industrial processes (cement production), and biofuel consumption. For 1995, the EDGAR globally integrated emissions for the dominant fossil fuel consumption category are higher than those estimated by Marland et al. (2007), a frequently updated and more widely used dataset that also matches the IPCC [Special Report on Emissions Scenarios (SRES)] values at year 2000. A uniform scaling is applied to the EDGAR spatial distributions for fossil fuel consumption and cement production so that their global totals match the Marland et al. (2007) category totals at each decade (1890–1990, and 1995). The spatial distributions of year 2000 emissions are the same as for 1995 but scaled to give the global total from Marland et al. Biofuel distributions (and global totals) over the 1850–2000 period are taken directly from the EDGAR V3.2 data. The model can be run with land use change emissions specified externally like fossil fuel emissions or they can be calculated interactively by the model from specified land cover change as discussed in section 2d(1). The spatial distribution of specified land use change emissions is based on EDGAR V3.2 data (Van Aardenne et al. 2001), but global totals are adjusted to match values from Houghton and Hackler (2002).
For the decades of interest before 1890 (1850–80), the 1890 spatial distribution of EDGAR emissions for each category are used with the fossil fuel consumption and cement production totals scaled to the Marland et al. (2007) values. For biofuel, it is assumed that the decadal rate of increase of emissions between 1850 and 1880 was the same as that between 1890 and 1900. The globally integrated CO2 emissions, not including specified LUC emissions, are 25.0 Pg CO2 yr−1 (6.82 PgC yr−1) at 2000, slightly below the SRES value of 25.3 Pg CO2 yr−1 (6.9 PgC yr−1). The concentrations of non-CO2 GHGs including CH4, N2O, CFC-11, and CFC-12 are prescribed.
b. The physical ocean
The ocean component (OGCM3.5) of CanESM1 is based on the National Center for Atmospheric Research Community Ocean Model (NCOM1.3) (Gent et al. 1998), which is a primitive equation model with a rigid lid. The model is implemented at a horizontal resolution of 1.86° such that there are four ocean grid boxes underlying each atmosphere grid box. There are 29 levels in the vertical and the vertical resolution increases toward the ocean surface, from 300 m in the deep ocean to 50 m in the top 200 m. The model represents subgrid-scale mixing of tracers via isopycnal diffusion and the Gent and McWilliams (1990) parameterization of eddy stirring effects, with an isopycnal diffusion coefficient of 1 × 103 m2 s−1. There is also a constant background vertical diffusivity for tracers set to 3 × 10−5 m2 s−1. Vertical mixing within the surface mixed layer is parameterized as vertical diffusion of momentum and tracers with large diffusivity, solved implicitly.
The coupled ocean–atmosphere model with AGCM3 and OGCM3.5 is referred to as CGCM3.5. Differences from earlier configurations for CGCM1 described in Flato et al. (2000) and CGCM2 described in Flato and Boer (2001), both based on the Geophysical Fluid Dynamics Laboratory (GFDL) Modular Ocean Model (MOM1), include a modified vertical velocity calculation, use of a special ocean grid cell rather than an island at the North Pole, and exchange of water properties parameterized as instantaneous mixing through the Canadian Archipelago and across Hudson and Gibraltar Straits. Because the Bering Strait is unrealistically wide, only baroclinic transport is allowed across this channel. In contrast to previous model versions including CGCM3.1 employed for the IPCC fourth assessment, no flux adjustments are applied in CGCM3.5.
c. Canadian Model of Ocean Carbon
The Canadian Model of Ocean Carbon incorporates an inorganic chemistry module (solubility pump) and an ecosystem model (organic and carbonate pumps) for simulating the ocean–atmosphere exchange of CO2 (Zahariev et al. 2008). The inorganic chemistry module is based on protocols from the Ocean Carbon Cycle Model Intercomparison Project (OCMIP)–Phase 2 (http://www.ipsl.jussieu.fr/OCMIP/phase2/), with dissolved inorganic carbon (DIC) and total alkalinity as prognostic variables. The piston or gas transfer velocity is proportional to the square of the 10-m wind speed and the intercept is adjusted to conserve the global mean piston velocity (Wanninkhof 1992). There is no gas exchange through sea ice.
The ecosystem component of CMOC (Fig. 1) is based on the NPZD model of Denman and Peña (1999). A single nutrient variable (N) implicitly represents nitrate, ammonium, and urea including inputs from surface dinitrogen fixation. Phytoplankton population (P) growth is limited by light, temperature, nitrogen, and iron. Zooplankton (Z) graze on the phytoplankton population, and their mortality is modeled on the basis of linear and quadratic terms implying predation by unresolved higher tropic levels. The detritus variable (D) implicitly combines dissolved, suspended, and sinking organic matter, with a constant sinking rate. The currency of the model is nitrogen, and the biological effect on dissolved inorganic carbon (DIC) is calculated via a constant Redfield C:N ratio. Chlorophyll (Chl) is a separate prognostic variable based on a varying Chl:N ratio. Phytoplankton growth and remineralization rates are temperature dependent. Iron limitation of photosynthesis is implemented through a surface ocean distribution of an iron limitation factor that is derived from normalized monthly observational estimates of the climatological annual minimum nitrate concentration (Zahariev et al. 2008). This factor is based on the assumption that the lowest nitrate concentration observed during the seasonal cycle is proportional to the degree of iron limitation. A more detailed description of these and other aspects of CMOC is available in Zahariev et al. (2008) and online at http://www.cccma.ec.gc.ca/papers/kdenman/PDF/CMOCreport.pdf.
d. Canadian Terrestrial Ecosystem Model
Land–atmosphere exchange of CO2 in CanESM1 is modeled using the Canadian Terrestrial Ecosystem Model (CTEM) (Arora 2003; Arora and Boer 2003, 2005), which simulates three live vegetation pools (leaves, stem, and root) and two dead carbon pools (litter and soil organic carbon) for nine plant functional types (PFTs) as illustrated in Fig. 2. The CTEM is coupled to the Canadian Land Surface Scheme (CLASS 2.7) (Verseghy 1991; Verseghy et al. 1993) to produce fluxes of energy, water, and CO2 at the land surface. The photosynthesis submodule of the CTEM is based on the biochemical model of Farquhar et al. (1980) and Collatz et al. (1991, 1992). The current version uses a single-leaf photosynthesis approach with coupling between photosynthesis and canopy conductance based on vapor pressure deficit (Leuning 1995). The photosynthesis and autotrophic and heterotrophic respiration submodules of the CTEM, as described in Arora (2003), are used to calculate net primary and net ecosystem productivity. Positive net primary productivity (NPP) is allocated to leaves, stem, and root based on light, root water, and leaf phenological status. The phenology submodule of CTEM uses a carbon-gain approach in which leaf onset is initiated when it is beneficial for the plant, in carbon terms, to produce new leaves. Leaf offset is initiated by unfavorable environmental conditions including shorter day length, cooler temperatures, and low soil moisture (Arora and Boer 2005). Photosynthesis operates at the atmosphere model time step of 20 min; all other submodules of CTEM operate at a daily time step.
Allocation to, and the litter and respiratory losses from, the three vegetation components (leaves, stem, and root) result in time-varying biomasses that are reflected in the structural vegetation attributes used in the energy and water balance calculations of the land surface scheme (Arora and Boer 2005). CTEM does not include N or P cycles, and the effects of nutrient limitation on photosynthesis are not modeled explicitly.
1) Land use change emissions
Land use change emissions can be specified as an external source or can be modeled explicitly in CTEM on the basis of specified changes in land cover. Explicit modeling of LUC emissions ensures that the modeled net land–atmosphere CO2 exchange is consistent with the specified changes in land cover. For instance, specification of LUC emissions with no corresponding land cover changes implies that areas meant to be deforested may act as carbon sinks as CO2 increases. Wang et al. (2006) provide historical (1850–1992) annual time series of the fractional coverage of the CTEM PFTs, including C3 and C4 crops, at 0.5° resolution based on changes in global cropland area (Ramankutty and Foley 1999). Fractional coverage of CTEM PFTs for the 1993–2000 period is also provided by Wang et al. (2006) based on results from the Integrated Model to Assess the Greenhouse Effect (IMAGE) 2 model (Alcamo et al. 1998).
An increase in crop area implies the replacement of natural vegetation by crops (we use the term deforestation). The deforested biomass is divided into three fractions representing the amounts 1) combusted or used for fuel wood with no time delay, 2) left as slash or used for pulp and paper products, and 3) used for durable wood products. The fractions allocated to these three uses depend on the aboveground vegetation biomass density and whether the PFTs are woody or herbaceous (Table 1). The fraction allocated to slash or pulp and paper products is transferred to the litter pool, which has turnover rate of ∼0.5 yr−1. The fraction allocated to wood products is allocated to the model’s soil carbon pool, which has a turnover rate of ∼0.025 yr−1. These maximum turnover rates are reduced to account for the effect of temperature and moisture, so the actual turnover rates are climate dependent. Houghton’s bookkeeping approach (Houghton and Hackler 2002), in comparison, divides the deforested biomass into slash (with turnover time of less than a year) and pools with turnover rates of 1 yr−1 (fuel wood), 0.1 yr−1 (pulp and paper products), 0.01 yr−1 (wood products), and 0.001 yr−1 (elemental carbon) that are not climate dependent.
Over the cropland fraction of a grid cell a simple crop model is used. Crops increase their biomass depending on environmental conditions, and harvesting is initiated when the air temperature remains below 8°C for 5 consecutive days, or when the crop LAI reaches a threshold (3.5 for C3 crops and 4.5 for C4 crops) signifying that the crops have matured (Arora and Boer 2005). Crops are harvested over a period of 15 days and the harvested biomass contributes to the litter pool. This usually leads to one annual crop cycle in midlatitude regions and multiple crop cycles in tropical regions. Harvesting ensures that vegetation biomass does not keep increasing on croplands as CO2 increases, and thus prevents croplands from sequestering aboveground carbon-like forests. Carbon may still be sequestered in the soil carbon pool but the higher soil carbon decomposition rate over croplands (0.031 yr−1) compared to other PFTs (0.013–0.026 yr−1), which accounts for tillage, prevents soil carbon sequestration in croplands. CTEM does not track the age of vegetation explicitly but characterizes the state of the vegetation in terms of the mean values of carbon in the various pools. When croplands are abandoned, the fractional coverage of other PFTs is increased according to the data of Wang et al. (2006). The result is that the vegetation density is reduced, and carbon is sequestered until a new equilibrium is reached, providing a carbon sink associated with regrowth as abandoned croplands revert back to natural vegetation. Biomass lost through LUC, crop growth and harvesting, regrowth of vegetation on abandoned croplands, and soil carbon and litter decomposition on croplands all contribute to LUC-related carbon emissions.
2) Terrestrial photosynthesis response to CO2
Representation of terrestrial primary productivity in the model and its dependence on CO2 concentration is based on a standard biochemical model of photosynthesis (Farquhar et al. 1980; Collatz et al. 1991, 1992). Figures 3a and 3b (thin lines) show the evolution of gross primary productivity (GPP, Gp) and net primary productivity (NPP, Np) simulated by the CTEM in response to historical 1850–1992 CO2 concentrations and driven with 1979–99 observation-based atmospheric reanalysis data (Dirmeyer and Tan 2001), used repeatedly. The increase in GPP with CO2 is determined by the model’s photosynthesis equations that colimit photosynthesis on the basis of Rubisco (Jc) and electron capacity (Je) rates; Jc depends on maximum catalytic Rubisco capacity (Vc,max) and Je on quantum efficiency and incident photosynthetically active radiation (PAR), as shown in the appendix. Luo et al. (1996) define a leaf-level function ℑ that gives the fractional change in Jc and Je per unit change in intercellular CO2 concentration (ci) as
Here ℑ1 and ℑ2 are specified as fractional changes in Jc and Je and are therefore independent of light, nutrient availability, and species characteristics:
Both Θ and Γ are functions of temperature, as shown in the appendix. The ℑ1 and ℑ2 from Luo et al. (1996) and CTEM are compared in Fig. 3c for 25°C, which is a reasonable approximation for globally averaged growing season temperature: CTEM is slightly less sensitive to CO2 than the Luo et al. (1996) model. The differences are mostly due to parameter values. Luo et al. use values of Γ = 45 ppm and K = 414 ppm at 25°C. The corresponding values in CTEM are Γ = 40 ppm and Θ = 239 ppm at 25°C. The decrease in ℑ1 and ℑ2 with increasing CO2 implies that photosynthesis responds less strongly to a change in CO2 concentration when CO2 is higher, that is, CO2 stimulation of photosynthesis saturates. We also calculate for simulated GPP averaged over 1850–1992. This is shown in Fig. 3c for ci equal to 0.7 times atmospheric CO2 concentration (Luo et al. 1996; Haxeltine and Prentice 1996; Rogers and Humphries 2000). The values lies between the ℑ1 and ℑ2 curves from CTEM, indicating that the globally averaged simulated increase in photosynthesis with CO2 in the model is consistent with the expected behavior based on the model formulation.
The rate of increase of GPP per unit increase in CO2 (dGp/dCO2) in Fig. 3a, averaged over the 1850–1992 period, is 0.27 PgC yr−1 ppm−1, which is similar to that of other ecosystem models that do not include nutrient or any other limitation of terrestrial photosynthesis. Law et al. (2006) report that GPP increases from ∼116 PgC yr−1 in 1900 to ∼138 PgC yr−1 in 2000 in the Commonwealth Scientific and Industrial Research Organisation biosphere model yielding a value of ∼0.30 PgC yr−1 ppm−1. For the Lund–Potsdam–Jena (LPJ) dynamic global vegetation model (DGVM), Gerber et al. (2004) report that GPP increases from ∼135 PgC yr−1 at 280 ppm CO2 to ∼159 PgC yr−1 at 370 ppm CO2, which yields a value of ∼0.23 PgC yr−1 ppm−1.
The model’s photosynthetic response to the observed 1850–1992 increase in atmospheric CO2 concentration, C(t), can also be expressed in terms of logarithmic growth factors for GPP and NPP as is done in many simple carbon cycle models (Cao et al. 2001; Alexandrov and Oikawa 2002):
Here Gp(t) and Np(t) are modeled GPP and NPP at time t, Gp0 and Np0 are GPP and NPP at the preindustrial CO2 concentration (C0) of 288 ppm, and γg and γn are obtained from a least squares fit. Figures 3a,b show Eqs. (6) and (7) fitted to simulated GPP and NPP, which yield values of Gp0 = 130 PgC yr−1, γg = 0.90, Np0 = 62 PgC yr−1, and γn = 1.23. The logarithmic growth factor is higher for NPP than for GPP, implying a faster rate of increase for NPP with CO2 because the increase in autotrophic respiration (which depends on vegetation biomass, which in turn depends on turnover time) lags the increase in GPP (Friedlingstein et al. 1995). The value of γn for NPP is higher than is commonly used in simpler terrestrial box carbon cycle models, which typically use values between 0.3 and 0.8 (Alexandrov and Oikawa 2002; Friedlingstein et al. 1995). Using anthropogenic fossil fuel emissions (Marland et al. 2007), land use change emissions (Houghton 2003), observations of atmospheric CO2, and global ocean–atmosphere flux for the 1850–2004 period within a Bayesian calibration framework, Ricciuto et al. (2008) derive an estimate for the logarithmic growth factor γn of 0.72 with a 95% confidence range of 0.55–0.84. Friedlingstein et al. (1995), using a similar approach, obtain a value of γn = 0.68. Norby et al. (2005) estimate the value of γn as 0.60 using data from four Free-Air CO2 Enrichment Experimental (FACE) sites. These estimates of γn suggest that the standard photosynthesis equations yield a greater rate of increase of NPP with CO2 than implied by the observed increase in atmospheric CO2 and the results based on a FACE study. This suggests that the model lacks a mechanism to constrain the response of terrestrial photosynthesis to elevated CO2.
Experimental studies of plants grown in elevated CO2 show that acclimatization to elevated CO2 widely occurs in the form of decline in maximum photosynthetic rate, a process known as photosynthesis down regulation. These studies find that down regulation results from decreases in Rubisco activity (Vc,max), in the maximum rate of electron transport, and in both together (Sage 1994; Rey and Jarvis 1997; Medlyn et al. 1999; Rogers and Humphries 2000; Murray et al. 2000; Ainsworth et al. 2003, 2004; Adam et al. 2004; Kenzo et al. 2006). Down regulation of Rubisco and electron transport rates may result from nutrient limitation (McGuire et al. 1995; Lecain et al. 2003; Luomala et al. 2003; Luo et al. 2004) and/or an excess end-product condition that leads to source–sink imbalances (Midgley et al. 1999; Ainsworth et al. 2004). The latter results from a plant’s inability to form sufficient “sinks” for the additional photosynthate acquired in an elevated CO2 environment, even under nonlimiting nitrogen conditions (Adam et al. 2004). Not all elevated CO2 experiments exhibit down regulation, however. Duke Forest results, for example, suggest that down regulation may not occur as long as soil nitrogen availability keeps pace with photosynthesis and growth processes (Herrick and Thomas 2001), and nitrogen-fixing species do not show any photosynthesis down regulation (Vogel and Curtis 1995). All of these studies are of limited scale, however, and the overall global magnitude of down regulation remains uncertain.
Explicit down regulation of the maximum photosynthetic rate (Vc,max) is not accounted for in the standard photosynthesis model but may be included by coupling terrestrial carbon and nitrogen cycle models. Down regulation of electron transport rates is not currently accounted for in any model, although some experimental studies suggest that electron transport rates could decrease in tandem with Rubsico-dependent rates in an elevated CO2 environment owing to N resource optimization (Midgley et al. 1999).
In the absence of 1) fully coupled C and N cycles in CTEM, 2) a mechanistic approach to limiting maximum electron transport rate, and 3) a representation of source/sink imbalance, we adopt an empirical approach based on experimental plant growth studies. We characterize the down-regulation effect by ξ(C), where the down-regulated GPP is given by Gd = ξ(C)Gp. In the absence of down regulation, γg in Eq. (6) characterizes the strength of the model GPP response to increased CO2, which is not constrained by nutrient limitation or source–sink imbalance, while a similar expression characterizes the down-regulated case (Gd) with a down-regulated photosynthetic growth factor γgd < γg. Then
yields a progressively decreasing ξ(C) as CO2 concentration increases above C0. Equation (8) is used to obtain an estimate of γgd from experimental studies in which plants are grown in both ambient and elevated CO2 environments. Table 2 gives values of γgd estimated from Eq. (8) by substituting experimentally determined value of ξ, γg = 0.90 from CTEM and values of C, and C0. Data sources include the individual studies of Ainsworth et al. (2003, 2004), Adam et al. (2004), Bigras and Bertrand (2006), and two meta-analyses (Medlyn et al. 1999; McGuire et al. 1995). Medlyn et al. summarize results from 15 field-based elevated CO2 experiments and report down regulation of 10%–20%. McGuire et al. summarize results from 77 studies and report a mean decrease of leaf nitrogen concentration of 21% in response to CO2 varying from 350 to 700 ppm. Table 2 shows that estimated values of γgd vary from 0.19 to 0.59 with an average of 0.42. Lower (higher) values of γgd imply more (less) down regulation of photosynthetic rates. We do not give greater weight to γgd from the meta-analyses, because they report results only for forest vegetation while other studies listed in Table 2 report results from grass and crop species.
Figure 4 shows ξ for γgd = 0.42, γg = 0.90, and C0 = 288 ppm. In the absence of information about the response of individual PFTs, this down-regulation function is applied to all PFTs. These parameter values yield a value of ξ ≈ 0.9 for a CO2 concentration of 370 ppm, resulting in down regulation of photosynthetic rates by ∼10% in the year 2000 compared with 1850. This may be compared with the f (N) function used in the Community Land Model (CLM3) of the Community Climate System Model (CCSM), which is intended to represent the fraction of the potential photosynthesis that is realized in the face of nitrogen limitation. Their PFT-dependent values of f (N) range between 0.60 and 0.84, implying 16%–40% reduction in photosynthesis for the present day compared to the preindustrial era (Oleson et al. 2008; Stöckli et al. 2008).
Implementation of down regulation decreases the rate of increase of GPP and NPP with increasing CO2 and yields lower values of the logarithmic growth factors for GPP (γg = 0.51) and NPP (γn = 0.69) in the fully coupled model. The value of γn = 0.69, after down regulation, compares reasonably well with those from Ricciuto et al. (2008) (γn = 0.72), Friedlingstein et al. (1995) (γn = 0.68), and Norby et al. (2005) (γn = 0.60), although we used an entirely different approach from these authors.
We perform simulations for 1850–2000 with different treatments of land use change emissions and photosynthesis to attempt to understand the behavior of the carbon budget and compare results with available observations. The simulations were initialized from a 2000-yr preindustrial control run with no anthropogenic emissions and fixed 1850 land cover. Land and ocean carbon pools are near equilibrium with a mean surface atmospheric CO2 concentration of ∼286 ppm. Details of this control simulation are presented in Christian et al. (2009). Simulations were then performed for 1850–2000 with and without down regulation of terrestrial photosynthesis and with LUC emissions treated in two different ways: 1) interactively, as discussed in section 2d(1), based on historical changes in crop area and 2) by injecting Houghton and Hackler (2002) LUC emissions into the atmosphere without any changes to land cover (similar to C4MIP simulations). These four simulations—A, B, C, and D—are summarized in Table 3. Emissions of CO2 and concentrations of other GHGs are prescribed as described under the heading “Greenhouse gases” in section 2a.
a. Terrestrial and oceanic biostates in 1850
Figure 5a displays the global pattern of simulated terrestrial net primary productivity (tNPP) for the preindustrial (year 1850) control simulation. Figure 5b shows simulated tNPP when CTEM is driven offline with 21 years (1979–99) of repeated reanalysis data (Dirmeyer and Tan 2001) and 1850 land cover with atmospheric CO2 set to 286 ppm. In the absence of spatially distributed observation-based estimates of tNPP, the simulated spatial pattern may be compared to that of average tNPP simulated by 17 terrestrial ecosystem models (Fig. 5c) in the Cramer et al. (1999) study, with the caveat that participating models in this study use CO2 concentrations between 340 and 350 ppm. Modeled tNPP exhibits the expected spatial structure, with higher values in the warm and humid tropics and lower values at higher latitudes and in arid and semiarid regions. Simulated global tNPP is 60.8 PgC yr−1 for the fully coupled simulation and 62.0 PgC yr−1 when CTEM is driven offline with reanalysis data for preindustrial CO2 concentration. These estimates are compared with observation-based and other model values in Table 4, which also shows the simulated values for the 1980–2000 period from simulation C (see Table 3). Observation-based estimates in Table 4 reflect the late-twentieth-century values, while model values are from a range of simulations, so the corresponding CO2 concentration or the time period is also noted. Simulated terrestrial NPP and GPP values lie within the model estimates, although our values for the 1980–2000 period are slightly higher than observation-based estimates. Compared to the simulation driven with reanalysis data and the Cramer et al. (1999) values, tNPP in the fully coupled model is lower over Amazonia and higher over Central Africa. This is due to lower and higher than observed model precipitation over Amazonia and Central Africa, respectively. The globally averaged annual value of modeled precipitation (2.73 mm day−1) for the 1981–2000 period compares well with the observation-based estimate from Xie and Arkin (1997) (2.70 mm day−1), but differences remain in the spatial pattern. Simulated global total vegetation biomass and litter and soil carbon pools compare reasonably well with observation-based and other model estimates (Table 4).
Figure 6 compares the simulated ocean net primary productivity (oNPP) during the 1850s with that estimated from satellite-derived surface chlorophyll using the Vertically Generalized Production Model (VGPM) of Behrenfeld and Falkowski (1997). For comparison, simulated oNPP during the late twentieth century is also shown from simulation C. The spatial pattern of simulated oNPP does not change significantly from the 1850s to the late twentieth century. Simulated primary production shows greater contrast between more and less productive regions than the estimates from the VGPM. Possible reasons for this behavior include the 50-m thickness of the uppermost model layer, inadequate vertical mixing and lack of a dynamic mixed layer, and an ecosystem model with a single phytoplankton species and a fixed Redfield ratio. The modeled global total annual primary production is 34 PgC yr−1 and is at the lower end of estimates shown in Table 5. Our values are close to those for the open ocean (Fung et al. 2000). A recent model intercomparison study (Carr et al. 2006) reports mean values of 51 PgC yr−1 for satellite-based methods (mean of 24 models) and 55 PgC yr−1 for ocean circulation/ecosystem models (7 models). Our values are comparable with the lower estimates of the latter group: Dutkiewicz et al. (2005) report a value of 35 PgC yr−1, and Aumont and Bopp (2006) report a value of 41 PgC yr−1. Part of the reason may be that about 17% of the observed total occurs on the continental shelves, which are not well resolved in our model. CMOC-simulated values of organic and inorganic carbon export, and organic flux to sediments, compare reasonably well with observation-based estimates (Table 5). Simulated nitrogen fixation is somewhat lower than observation-based estimates (Table 5). For 1850 the global mean air–sea CO2 flux is close to zero (Christian et al. 2009). CMOC validation is discussed extensively in Zahariev et al. (2008).
b. Atmospheric CO2 and land and ocean CO2 fluxes
1) Global trends
Figure 7a compares observation-based surface CO2 concentrations for 1850–2000 (dotted line) with results from the four simulations listed in Table 3. Simulated CO2 concentrations are those of the lowest atmospheric model level. Figures 7b and 7c show simulated land–atmosphere and ocean–atmosphere CO2 fluxes, with observation-based estimates for the decades of the 1980s and 1990s together with their uncertainty ranges. Positive values indicate flux from the atmosphere to the land/ocean. The observation-based estimates for land and ocean carbon uptake during the 1980s (0.3 ± 0.9 and 1.8 ± 0.8 PgC yr−1) and 1990s (1.0 ± 0.6 and 2.2 ± 0.4 PgC yr−1) are from the IPCC Fourth Assessment Report (AR4) (Denman et al. 2007).
Simulations A and B (without down regulation) yield atmospheric CO2 concentrations lower than observed and also lower than those from simulations C and D (with down regulation). CO2 concentrations in year 2000 for simulations A and B are ∼20 and ∼15 ppm, respectively, below the observed value of ∼370 ppm and their land–atmosphere (ocean–atmosphere) fluxes are higher (lower) than observation-based estimates (Figs. 7b,c). Low atmospheric CO2 is due to high carbon uptake by the land in simulations A and B, and low ocean uptake is at least partially a consequence of low atmospheric CO2 concentration.
Differing treatments of LUC emissions modify these results only modestly and introduce a difference of 5–7 ppm CO2 in 2000 (Fig. 7a). Simulation A with interactive treatment of LUC emissions gives a lower CO2 concentration than simulation B with specified LUC emissions because the interactive land use change is based on changes in cropland area only. Estimates of LUC emissions from pasture lands range from 13% (Houghton 1999) to nearly 50% (Strassmann et al. 2008) of total emissions, and overall LUC emissions are uncertain by up to ±50% (Ramankutty et al. 2007). Shevliakova et al. (2009) give contribution from pastures that range from a sink of 0.15 to a source of 0.37 PgC yr−1 for the 1990s. CanESM1 simulations (not shown here) with land cover reconstructions based on increase in crop area only and increase in both crop and pasture area show that the LUC emissions caused by taking into account the increase in pasture area are to a large extent compensated by increase in soil carbon over pasture lands so that the overall simulated twentieth-century carbon budget does not change substantially (V. K. Arora and G. J. Boer 2009, unpublished manuscript). Our implied LUC emissions for the 1850–2000 period, inferred as the difference in net land–atmosphere exchange from simulations with and without LUC (not shown), is around 70 PgC compared to about 155 PgC based on Houghton and Hackler (2002). This 70 PgC of LUC emissions for the 1850–2000 period due to increase in cropland is comparable to 84 PgC obtained by Strassmann et al. (2008) for the 1700–1999 period.
Simulations A and B indicate a mismatch between simulated and observation-based carbon budgets at the end of the twentieth century that is not reconcilable by the differing treatments of LUC emissions considered here. Excessive carbon uptake by the land indicates that the photosynthetic response to increasing CO2 levels is too strong in the model despite the tests discussed in section 2d(2) and Fig. 3c, showing that accepted photosynthesis mechanisms are properly implemented in the model. This kind of behavior is, however, consistent with the down regulation of photosynthesis observed in plant growth experiments as discussed in section 2d(2). In simulations C and D (with down regulation), land uptake decreases (Fig. 7b) and atmospheric CO2 concentrations and land–atmosphere and ocean–atmosphere fluxes are more consistent with observations (Fig. 7). The reduced land uptake of carbon with down regulation is consistent with results of an EMIC (Sokolov et al. 2008) and a comprehensive earth system model (Thornton et al. 2009) that explicitly model the interactions between the terrestrial carbon and nitrogen cycles.
Figure 7 shows that simulation C, with down regulation and interactive LUC emissions, agrees best with observed CO2 concentrations, while simulation D, with specified LUC emissions, results in CO2 concentrations higher than observed. Both C and D (with down regulation) yield land uptake during the 1990s (∼0.7 and ∼0.3 PgC yr−1) that compares reasonably well with observation-based estimates (1.0 ± 0.6 PgC yr−1) from IPCC AR4, whereas A and B do not. Ocean uptake also increases in C and D because of higher atmospheric CO2 concentration. CO2 fluxes for simulations C and D lie within the uncertainty bounds of observation-based estimates (Figs. 7b,c). Simulated CO2 concentration in 2000 for the four simulations in Fig. 7a varies from ∼350 to 379 ppm, compared with a range of ∼346–400 ppm for the C4MIP models (Friedlingstein et al. 2006).
2) Cumulative results
Table 6 compares the cumulative increase in the atmospheric carbon burden and the land and ocean carbon uptake for 1850–2000 with observation-based estimates. The atmospheric carbon burden is known from observed atmospheric CO2 concentration. Observation-based estimates of cumulative carbon uptake by the global ocean are derived from Sabine et al. (2004) and Denman et al. (2007). Estimated emissions are from Marland et al. (2007), as discussed in section 2a. Cumulative uptake by land is the residual. Table 6 shows that cumulative ocean uptake in the model is low compared with Sabine et al., even when comparing with an estimate (115 Pg) excluding the extrapolation poleward of 65° (includes the Arctic Ocean and marginal basins adjacent to Antarctica). However, Matsumoto and Gruber (2005) suggest that the Sabine et al. estimate may be high by ∼7%. Cumulative ocean uptake may be biased low because of too warm ocean temperature, weak vertical exchange, and changes in export production or calcification. North Atlantic Deep Water formation is slightly weaker in the model than observation-based estimates, and sea surface temperature has a warm bias in the southern midlatitudes (Zahariev et al. 2008; Christian et al. 2009). Vertical exchange overall is possibly limited by the coarse vertical resolution and mixing parameterizations used. Cumulative land uptake is much higher than observation-based estimates in simulations with no down regulation (Table 6). Simulation C (with down regulation and interactive LUC emissions) yields the best comparison with observation-based estimates although the partitioning of carbon uptake by land and the ocean is biased owing to low carbon uptake by the ocean. Simulation D can be compared with Ricciuto et al. (2008) since the rate of NPP increase with CO2 is similar and both use the Houghton and Hackler (2002) LUC emissions. Ricciuto et al. use ocean uptake from Sabine et al. (2004), while we use a full ocean with a lower cumulative carbon uptake (Table 6).
3) Meridional structures
Figure 8 shows the zonally averaged distribution of land– and ocean–atmosphere CO2 fluxes averaged over the 1981–2000 period for each of the simulations. In Fig. 8a, all simulations show a flux of CO2 into the land, and hence a sink of atmospheric CO2, at mid to high northern latitudes. The simulations with down regulation exhibit weaker sinks than those without down regulation, as expected. For the 1981–2000 period, the specified LUC emissions in simulations B and D are concentrated in the tropics, resulting in a large flux into the atmosphere there. Tropical South America and Asia are known LUC hotspots during the 1981–2000 period (Lepers et al. 2005), consistent with the Houghton (2003) LUC emissions estimates (2.06 ± 0.6 PgC yr−1 for the tropics out of a global total of 2.09 ± 0.8 PgC yr−1). Simulations A and C, with interactive LUC emissions based on changes in crop area, do not show this large tropical CO2 source but rather are neutral or small carbon sinks. The Southern Hemisphere south of 30°S is neutral or a small carbon sink in all simulations. A terrestrial CO2 sink at mid to high northern latitudes and absent or small carbon sinks in the tropics and the Southern Hemisphere are consistent with the results reported in IPCC AR4 (Denman et al. 2007) and other studies (Myneni et al. 2001; Goodale et al. 2002; Kaufmann and Stock 2003; Stephens et al. 2007). The large tropical terrestrial source of atmospheric CO2 associated with specified LUC emissions in simulations B and D is not consistent with any published studies that we are aware of and suggests that either the Houghton and Hackler (2002) LUC emissions may be excessive in this region (Denman et al. 2007; Stephens et al. 2007) or that specifying LUC emissions without any corresponding changes at the land surface leads to inconsistencies.
The zonal structure of the ocean–atmosphere fluxes in Fig. 8b is similar in all the simulations, including offline simulations with the ocean carbon model (Zahariev et al. 2008). There is net ocean uptake in the Northern Hemisphere extratropics and the subantarctic and outgassing in the tropics and the high-latitude Southern Ocean. Southern Ocean CO2 uptake and export production are reasonably consistent with observation-based estimates, although the magnitude of high-latitude CO2 outgassing is uncertain (Zahariev et al. 2008; Christian et al. 2009). The high-latitude Southern Ocean is believed to have changed from a net source to a net sink as a result of anthropogenic emissions (Gloor et al. 2003), and some studies have calculated that the Southern Ocean as a whole was a CO2 source in preindustrial times (e.g., Jacobson et al. 2007). In CanESM1 the net preindustrial Southern Ocean CO2 flux is small, with high-latitude outgassing largely balanced by midlatitude uptake (Christian et al. 2009). However, the high-latitude outgassing may be too strong in our model (Fig. 8b). The region south of 60°S does not become a sink until around 2020 depending on the emission scenario used (not shown), whereas Gloor et al. (2003) estimate that this had already occurred by 1990. The observation-based estimates of Takahashi et al. (2009) show the same latitudinal pattern (Fig. 8b) but both the source and the sink in the Southern Ocean are much weaker than in the model. However, this may result in part from incomplete temporal and spatial coverage in the ship-based observations, particularly in austral spring and fall when outgassing is particularly strong (Christian et al. 2009).
For all criteria combined, including the temporal evolution of atmospheric CO2, globally averaged terrestrial and oceanic carbon sinks, and latitudinal distribution of land– and ocean–atmosphere fluxes, simulation C—with terrestrial photosynthesis down regulation and interactive LUC emissions—yields the best comparison with observation-based estimates.
c. Comparison of contemporary results with observations
Figures 9a and 9b show the zonally averaged distribution of observed and simulated surface CO2 concentration from simulation C for 1991–2000. The observations are from the NOAA Earth System Research Laboratory (ESRL) (Masarie and Tans 1995). These data are based on more than 80 sites worldwide including shipboard stations, with about 30% of the stations in the Southern Hemisphere. Observed and simulated surface CO2 concentrations show similar patterns, with higher CO2 concentration and a larger annual cycle in the Northern Hemisphere. The latter is the result of the larger Northern Hemisphere landmass and the associated large terrestrial uptake of CO2 during boreal summer and release during winter.
Figure 9c compares the globally averaged observed and simulated monthly CO2 concentrations for the 1991–2000 period. Simulated CO2 concentration is about 2 ppm higher than observed in 2000. The annual cycles of globally averaged observed and simulated CO2, after removal of the annual mean and the increasing trend, are compared in Fig. 9d. Thin lines correspond to individual years and the thick line is the mean. The average amplitude of the modeled annual cycle is 4.2 ppm, slightly larger than the observed value of 3.8 ppm. The decrease in atmospheric CO2 during Northern Hemisphere spring and summer due to vegetation growth and the increase in fall and winter is captured well by the model, although there are some differences in timing and amplitude (Fig. 9d). The amplitude of the annual cycle for simulation A, without terrestrial photosynthesis down regulation, is 5.1 ppm (not shown).
Figure 9e shows zonal mean CO2 anomalies for 1991–2000, computed as the difference of CO2 concentration from the South Pole (after removing the global mean and the increasing trend), which allows interpretation of the interhemispheric gradient in a straightforward manner as the difference between South and North Poles values. Simulated values from simulation C show a slightly larger interhemispheric gradient (3.46 ppm CO2) than the observations (3.37 ppm). The corresponding value from simulation A, without photosynthesis down regulation, is 2.56 ppm (not shown). Model values show outgassing of CO2 from the Southern Ocean around 60°S that is not apparent in the observations. This may reflect both model biases and limited spatial coverage in the observations, as is the case for ocean–atmosphere CO2 fluxes (Fig. 8). Values from 6 of the 11 models that participated in the Atmospheric Tracer Transport Model Intercomparison Project (TransCom) (Law et al. 1996) are also shown for comparison (gray dotted lines). Maximum modeled concentrations occur at around 50°N for CanESM1 and the TransCom models, a feature that is absent in the ESRL observed fields but is present in early CO2 retrievals from the Atmospheric Infrared Sounder (AIRS) satellite (Tiwari et al. 2006).
The north–south interhemispheric gradient of CO2 in our model is due entirely to higher emissions in the Northern Hemisphere; in the absence of fossil fuel emissions the gradient reverses sign (Christian et al. 2009). Figure 10 shows the relationship between CO2 concentration difference between Mauna Loa and the South Pole, a proxy for the interhemispheric gradient, against anthropogenic emissions. Observed and modeled gradients compare well, with higher Southern Hemisphere concentrations corresponding to weak emissions early in the 1850–2000 period and higher Northern Hemisphere concentrations for higher emissions.
Figure 11 compares simulated and observed monthly CO2 anomalies at six sites located from Barrow, Alaska, at 71°N to the South Pole at 90°S. Observed data are from the NOAA ESRL air-sampling network (Tans and Conway 2005). Simulated CO2 anomalies compare reasonably well with observed estimates at the sites that exhibit a clear seasonal cycle (Barrow, Niwot Ridge, Mauna Loa), although the simulated minimum values occur about a month late. Simulated and observed values at the Southern Hemisphere sites (Ascension Island, Cape Grim, the South Pole) show a weak seasonal cycle; the timing of the weak maxima and minima is generally not well reproduced.
4. Discussion and conclusions
Aspects of the physical behavior and climate of the CCCma climate model are described in earlier studies (e.g., Flato et al. 2000; Arora and Boer 2002; Scinocca and McFarlane 2004). Here, we have assessed the biogeochemical behavior of the CCCma earth system model (CanESM1) against observations for 1850–2000. We compared simulated atmospheric CO2 concentration, its latitudinal and seasonal distribution, and land–atmosphere and ocean–atmosphere CO2 fluxes with available observations and observation-based estimates. Four simulations were performed: with and without down regulation of terrestrial photosynthesis and using two treatments of LUC emissions. In the absence of explicit nutrient or other limitations on modeled terrestrial photosynthesis, experimental results on photosynthetic down regulation in plants grown in elevated CO2 were applied to represent this effect in the model. The rate of increase of NPP with CO2 after down regulation, expressed in terms of the logarithmic growth factor, is similar to other studies (Ricciuto et al. 2008; Friedlingstein et al. 1995; Norby et al. 2005), and the resulting down regulation is conservative compared to that used in models that explicitly include a down-regulation or nitrogen limitation term (Oleson et al. 2008; Stöckli et al. 2008). Simulations with terrestrial photosynthesis down regulation, and thus with a reduced CO2 fertilization effect, compare better with observed atmospheric CO2 and land and ocean carbon uptake estimates for the 1980s and 1990s than do simulations lacking this process. Implementation of down regulation also yields better agreement in terms of the interhemispheric CO2 gradient. In the absence of down regulation, the interhemispheric gradient is weaker than estimates from observations. Overall, simulated ocean carbon uptake is low, but is much higher with down regulation (Table 6).
For 1981–2000, use of specified LUC emissions, which are concentrated in the tropics (Lepers et al. 2005; Houghton 2003), results in a large tropical terrestrial source of atmospheric CO2 that is inconsistent with other studies that suggest the tropics are near neutral or small carbon sinks (Denman et al. 2007; Stephens et al. 2007).
The effect of down regulation can also be inferred indirectly by diagnosing LUC emissions in a simulation with specified CO2 concentration (since fossil fuel emissions are known). We performed two additional simulations with specified CO2 concentration (dotted line in Fig. 7a), with and without down regulation (not shown). Diagnosed LUC emissions from the simulation with down regulation (141 PgC) compare reasonably well with observation-based estimates (e.g., 155 PgC in Houghton and Hackler 2002); the value for the simulation without down regulation is more than twice as large (287 PgC). Cumulative ocean uptake in these simulations is around 92 PgC (see Table 6).
The annual cycle of globally averaged CO2 concentration, dominated by Northern Hemisphere terrestrial photosynthesis and respiration, is simulated reasonably well although its amplitude is slightly larger than observed. The simulated latitudinal distribution of CO2 compares reasonably well with observation-based estimates. The simulated interhemispheric CO2 gradient is slightly larger than observed but low compared with TransCom models (Law et al. 1996). In the absence of anthropogenic emissions, the interhemispheric CO2 gradient reverses sign, consistent with observation-based estimates (Denman et al. 2007).
The down-regulation approach used here gradually reduces terrestrial photosynthesis rates based on an empirically determined function of CO2 concentration. This is consistent with the progressive nitrogen limitation (PNL) framework of Luo et al. (2004) in which, without new nitrogen inputs or decreased nitrogen losses, the availability of mineral nitrogen declines over time at elevated CO2 levels. This simple approach is effective but limited. In the absence of information about the response of individual PFTs, we apply the same down regulation to all PFTs. The approach used does not take into account increased nitrogen deposition, which is believed to have increased terrestrial photosynthesis (Magnani et al. 2007), except as implicitly included in the down-regulation function. Using a coupled carbon–nitrogen model driven offline with increasing CO2 concentrations, Thornton et al. (2007) estimate that the nitrogen deposition effect on terrestrial photosynthesis during 1976–2000 is ∼3% of the CO2 fertilization effect. The availability of increased nitrogen due to higher decomposition rates associated with higher temperatures is not captured, although other studies that include terrestrial carbon and nitrogen cycle interactions suggest that the warming-enhanced nitrogen availability is a secondary effect compared to limitation of carbon uptake by nitrogen availability (Sokolov et al. 2008; Thornton et al. 2009). Despite its caveats, the empirical approach used here offers a reasonable, straightforward, and experimentally consistent method of down regulating photosynthesis rates for coupled carbon–climate simulations, while more explicit representations of the processes are being developed.
We thank Mike Eby and the reviewers for their useful comments. Observation-based NOAA CO2 data provided by Pieter Tans and air–sea CO2 flux data provided by Takahashi et al. (2009) are also acknowledged. We would also like to thank the journal editor, Professor Andy Pitman, for his help.
Photosynthesis Dependence on CO2
In CTEM, the photosynthetic rate for C3 plants is colimited by assimilation rates based on Rubisco (Jc) and light (Je) limited rates, and depends on atmospheric CO2. Photosynthesis for C4 plants is insensitive to CO2. The gross photosynthetic rate limited by the photosynthetic enzyme Rubisco is given by
where Vm (mol CO2 m−2 s−1) is the temperature-adjusted maximum catalytic capacity of Rubisco (Vc,max), ci is the intercellular CO2 concentration, Oa (Pa) is the partial pressure of atmospheric oxygen, Γ is the CO2 compensation point, and Kc and Ko (Pa) are the Michaelis–Menten constants for CO2 and O2, respectively; Γ, Kc, and Ko are all temperature dependent:
in which σ is the Rubisco specificity for CO2 relative to O2 and is estimated as σ = 2600 f25(0.57), where f25 is the standard Q10 temperature function and T is the canopy temperature. The Michelis–Menten constants for CO2 and O2, Kc and Ko (Pa), respectively, are estimated as
The gross photosynthsis rate limited by the amount of available light is given by
which are differentiated with respect to ci to obtain Luo et al. (1996) leaf-level functions
* Additional affiliation: Fisheries and Oceans Canada, Institute of Ocean Sciences, Sidney, British Columbia, Canada.
Corresponding author address: Vivek Arora, Canadian Centre for Climate Modelling and Analysis, Environment Canada, University of Victoria, Victoria, BC V8W 2Y2, Canada. Email: email@example.com