The severity of the influence of human activity on global ecosystem resources and services is raising concern among scientists and politicians alike (Stern 2006; IPCC 2007). Rapid and long-term changes in environmental conditions, due directly or indirectly to human activities, are occurring simultaneously across a range of spatial and temporal scales and, in some instances, threaten the existence of valued terrestrial ecological resources (Millennium Ecosystem Assessment 2005). A key challenge in the coming years will be to provide a scientific basis for the sustainable use and development of Earth’s biosphere. One area in which ecology has an important role to play is in the development of predictive models that couple the global terrestrial carbon (C) cycle with atmospheric circulation models.
Plant–soil interactions lie at the heart of global biogeochemical and hydrological cycles, with climate change and land use having considerable potential to influence their role in feedbacks of greenhouse gases between the biosphere and atmosphere (Fig. 1). Climate change can affect plant–soil interactions directly as a result of atmospheric warming, alterations in precipitation patterns (IPCC 2001) and lengthening of plant growing seasons (Sherry et al. 2007) and indirectly as a result of longer-term changes in the distribution, productivity and composition of vegetation (Heimann & Reichstein 2008; Sitch et al. 2008). These changes then influence the size, composition and activity of soil biological communities that determine the magnitude of the soil C store and the emission of greenhouse gases at the ecosystem scale (e.g. CO2, N2O and CH4) (Bardgett, Freeman & Ostle 2008). There is also evidence that non-climatic global changes due to land use changes (Smith et al. 2008), nitrogen (N) enrichment (Magnani et al. 2007; Galloway et al. 2008), sulphur deposition (Monteith et al. 2007), CO2 fertilization (Woodward 2002; Norby et al. 2005) and changes in atmospheric ozone concentration (Sitch et al. 2007) are having a significant effect on ecosystem functioning including emissions of greenhouse gases (Meir, Cox & Grace 2006). Improving predictions of the impacts of these multiple stressors on ecosystem biodiversity and biogeochemistry poses a significant challenge in the development of protection, mitigation and adaptation strategies.
Mathematical models offer one means to make predictions and to test hypotheses regarding the effects of changes on responses of ecosystem functioning to global changes. A number of modelling approaches exist spanning a broad range of dynamic and spatial dimensions from global circulation models (GCMs) that operate at the 2° global grid cell scale to soil carbon process models that can be parameterized at the plot, core or microsite scale (Fig. 2). Linking between these are dynamic global vegetation models (DGVMs) of varying degrees of complexity that represent the state-of-the-art for studying the impacts of change on plant–soil interactions and their feedbacks to the climate system, e.g. CLM-CN, IGSM, LPJ, BIOME-BGC, CENTURY, DNDC, HYBRID, SDGVM, TRIFFID, ORCHIDEE (Schimel et al. 1996; Friend et al. 1997; Woodward, Lomas & Betts 1998; Cox 2001; Sitch et al. 2003; Krinner et al. 2005; Thornton & Rosenbloom 2005; Miehle et al. 2006; Thornton et al. 2007, 2009; Sokolov et al. 2008; Xu-Ri & Prentice 2008). These DGVMs can be operated whilst coupled to global atmospheric circulation models (GCMs) to enable explicit simulation of the feedbacks between the biosphere and atmosphere (Friedlingstein et al. 2006). As well as simulating the instantaneous biogeochemical processes of photosynthesis and plant respiration, DGVMs can also simulate longer-term impacts of climate change on vegetation cover and soil carbon storage (Cox et al. 2000; Sitch et al. 2003; Levy, Cannell & Friend 2004; Woodward & Lomas 2004; Sato, Itoh & Kohyama 2007). Most DGVMs typically utilize the concept of ‘plant functional types’ (PFTs; numbering between 3 and 20) to classify global vegetation diversity and its reactivity to climate (Table 1). Each PFT represents a broad class of vegetation type such as deciduous forest or grassland and is parameterized for a core set of physiological processes and ecological phenomena. These physiological processes are represented by different mathematical means in each DGVM and include photosynthesis (i.e. Farquhar, Caemmerer & Berry 1980; Collatz et al. 1991; Collatz, Ribas-Carbo & Berry 1992) C and N allocation using allometric and qualitative relationships (Friedlingstein et al. 1998) stomatal conductance (Jarvis 1976; Ball, Woodrow & Berry 1987;Stewart 1988; Leuning 1995; Haxeltine & Prentice 1996; Cox, Huntingford & Harding 1998), sapwood and/or fine root respiration (Lloyd & Taylor 1994; Gifford 1995) and evapotranspiration (Monteith 1995; Monteith & Unsworth 1990; Ducoudre, Laval & Perrier 1993). Ecological vegetation phenomena that are addressed include the phenological responsiveness of plants, their competition and mortality. Other approaches to the representation of plant functionality propose the use of plant biogeochemical or ecological traits that cover a gradient of overlapping characteristics (Wright et al. 2004). All of these models draw on established ecological understanding and field measurements to develop, parameterize and test their predictive certainty. Table 1 provides information on the range of PFTs simulated by five commonly-used DGVMs: HYLAND, LPJ, ORCHIDEE, SDGVM and TRIFFID (see Sitch et al. 2008 for a detailed breakdown of model compositions).
Temperate broadleaf evergreen
Temperate needleleaf evergreen
Boreal needleleaf evergreen
|Tropical broadleaf evergreen|
Temperate broadleaf evergreen
Temperate needleleaf evergreen
Boreal needleleaf evergreen
Tropical broadleaf raingreen
Temperate broadleaf summergreen
Boreal broadleaf summergreen
Boreal needleleaf summergreen
Needleleaf deciduous Shrubs
Accurate representation of plant–soil interactions is crucial to improving predictions of climate change on global ecosystems and the feedbacks between them. The overall aim of this paper is to review the integration of plant–soil interactions and their role in C and N cycling into global DGVMs. To achieve this aim, we illustrate four component model processes: (i) soil carbon dynamics, (ii) nitrogen cycling, (iii) drought impacts on vegetation, and (iv) vegetation dynamics. We also discuss the challenges faced by global vegetation modellers and ecologists to reduce uncertainty in their predictions.
Soil carbon dynamics
The balance between photosynthetic assimilation of CO2 and soil organic matter decomposition is critical to the global C balance (Davidson & Janssens 2006), with most terrestrial C being held below-ground in the soil (Lal 2004). Changes in environmental conditions, including land use change, ecosystem degradation and climate change, are having strong impacts on both the ecology and biogeochemistry of soils and vegetation across the globe (Millennium Ecosystem Assessment 2005; IPCC 2007). Plant–soil carbon models have been developed over the past 50 years to maximize soil C turnover for agricultural productivity and to predict land use and global change effects on ecosystem nutrient dynamics. Early soil models simulated soil organic C (SOC) as one homogeneous compartment (Jenny 1941), then two-compartment models were proposed (Beek & Frissel 1973; Jenkinson 1977) and later multi-compartment models were developed (McGill 1996; Molina & Smith 1998). Of the 33 SOC models currently represented within the Global Change and Terrestrial Ecosystems Soil Organic Matter Network (SOMNET) database (Smith, Bradbury & Addiscott 1996; Smith et al. 2001, 2002), 30 are multi-compartment, process-based models. Each SOC pool is characterized by its position in the model’s structure and its decay or ‘turnover’ rate (Fig. 3). Decay rates are usually expressed by first-order kinetics with respect to the C in the pool, i.e. [dC/dt = −kC] where t is time and the rate constant k is related to its half-life [h = (ln 2)/k], time required to halve the C content without inputs, or turnover time (τ = 1/k). Generally, the lower the decay rate constant, the higher the half-life, the turnover time and the stability of the organic pool. Using this approach the complexity and variability of SOC observed in the field is condensed into a mathematical representation of hypothetical pools that are parameterized individually, giving the resultant process based model considerable transferability between soils, ecosystems and land uses. A thorough review of the structure and underlying assumptions of different process-based SOC models is available (Molina & Smith 1998).
Currently, the most widely applied soil C models employ multiple SOC pools to simulate soil organic matter dynamics (e.g. DNDC; Li, Frolking & Harriss 1994; CENTURY Del Grosso et al. 2006; ROTHC Coleman & Jenkinson 1996; SUNDIAL Bradbury et al. 1993). They generally include key soil biogeochemical processes and have been used to determine soil C and N turnover in a range of ecosystems from arable croplands to grasslands, forests and peatlands (McGill 1996; Smith et al. 1998; Smith 2001, 2002; Peltoniemi et al. 2007). Most soil models have adopted the representation of soil C in multiple pools of organic matter decomposing using first-order rate kinetics (Paustian 1994), but others have attempted to include more detailed understanding of soil ecology in the form of food webs that predict the energy and nutrient flows between organisms (Hunt et al. 1984; Hunt, Coleman & Ingham 1987; Hunt, Trlica & Redente 1991; de Ruiter et al. 1993; de Ruiter & Van Faassen 1994; de Ruiter, Neutel & Moore 1995). All current DGVMs use process based descriptions of soil C dynamics (Smith et al. 2009a), some with a single C soil pool (e.g. Cox et al. 2000), others with multiple SOC pools similar to those found in soil models (e.g. Parton, Stewart & Cole 1988; Comins & McMurtrie 1993; Jones et al. 2005a,b) .
Major uncertainties and areas for research and development of representations of soil C in DGVMs relate to their inability to represent potentially important ecological phenomena including priming of soil C decomposition at depth (Fontaine et al. 2007), the ‘gadkil’ effect and non-equilibrium dynamics (Schimel & Weintraub 2003; Fontaine & Barot 2005; Neill & Gignoux 2006; Wutzler & Reichstein 2007) and for assumptions about the temperature sensitivity of different soil carbon pools (Davidson & Janssens 2006). Other challenges include understanding the nature of impacts of land management change on the stability of different pools of soil carbon (e.g. the physical protection of organic matter; Denef et al. 2004, 2007) and the inclusion of microbial feedbacks of decomposition that are not captured by first-order assumptions (e.g. soil methanogenesis and methanotrophy).
Nitrogen is crucial for plant growth and productivity in all terrestrial ecosystems (Vitousek & Howarth 1991; LeBauer & Treseder 2008). The availability of N is determined by the balance between inputs from biological N fixation, anthropogenic inputs and atmospheric N deposition, the recycling of plant residues and losses in gaseous (N2O and N2), inorganic (NO3− and NH4+) and dissolved organic matter forms (Vitousek et al. 2002). Ecosystem N limitation is common and the rate of plant N demand often exceeds the rate of soil N supply (Luo et al. 2004). The N cycle is, therefore, tightly coupled with soil–plant C dynamics so that N supply and limitation also affect the structure and function of ecosystems with important feedbacks to the global C cycle (Lloyd 1999). Land surface models and DGVMs used in the IPCC assessments have, historically, not included the plant–soil N cycle, and thus predict C sequestration rates far greater than that which expected N limitation would allow (Cramer et al. 2001; Hungate et al. 2003; Sitch et al. 2008). Here we review recent developments in DGVM modelling to introduce the N cycle to constrain C cycling (Zaehle & Friend 2006; Thornton et al. 2007, 2009; Sokolov et al. 2008; Xu-Ri & Prentice 2008).
Plants can acquire N from the soil via four pathways: advection (passive uptake), retranslocation (resorption), active uptake and biological N fixation (Ourry, Gordon & Macduff 1997; Hopmans & Bristow 2002; Wright & Westoby 2003). Well-established general ecosystem and DGVM models initially included plant N uptake simply as proceeding at the rate of demand until depleted from the soil (e.g. BIOME-BGC, CENTURY, DNDC, HYBRID, SDGVM) (Schimel et al. 1996; Friend et al. 1997; Woodward, Lomas & Betts 1998; Thornton & Rosenbloom 2005; Miehle et al. 2006). More recently energetics-based models have been developed that couple soil–plant C and N cycles by prescribing a vegetation C cost for N acquisition (Vitousek & Field 1999; Rastetter et al. 2001; Dickinson et al. 2002; Vitousek et al. 2002; Wang, Houlton & Field 2007). This latter set of models is particularly important for modelling plant C dynamics because it represents a second pathway for C expenditure other than growth and respiration making the link between C and N cycles explicit (Chapman et al. 2006).
The importance of incorporating both the plant–soil C and N cycles in these coupled DGVMs and climate models was demonstrated by Thornton et al. (2007) when the biophysical framework of the Community Land Model (CLM 3.0; Bonan & Levis 2006; Dickinson et al. 2006; Oleson et al. 2004) was coupled to the terrestrial biogeochemistry model (Biome-BGC 4.1.2.; Thornton et al. 2002; Thornton & Rosenbloom 2005) resulting in a new model, the Community Land Model (CLM-CN; Thornton et al. 2007). The simulated change in global terrestrial C uptake in response to increasing atmospheric CO2 concentration was reduced in CLM-CN by 74%, relative to the C-only counterpart model when N limitation was included. The importance of including N was further demonstrated when Sokolov et al. (2008) compared terrestrial C uptake in response to increased surface temperatures as simulated by the MIT Integrated Global Systems Model (IGSM2;Sokolov et al. 2005) coupled to the Terrestrial Ecosystem Model (TEM; Melillo et al. 1993; Felzer et al. 2004) either with or without C–N interactions. TEM includes two carbon pools (vegetation and soil detritus), heterotrophic respiration being dependent on C:N ratio, temperature and moisture. A change in terrestrial C uptake with increased surface temperatures was observed when N was included, resulting in a net sequestration of C in the soil–plant system and a reduced CO2 feedback to the climate system. Similarly, Xu-Ri & Prentice (2008) have coupled DyN, a comprehensive process-based model of the C and N cycling in terrestrial ecosystems, into the Lund–Potsdam–Jena Dynamic Global Vegetation Model (LPJ). The DyN model represents the uptake, allocation and turnover of N in plants, and soil N transformations including mineralization, N2 fixation, nitrification and denitrification, NH3 volatilization, N leaching, and N2, N2O and NO emissions.
Although the C–N cycle models within CLM-CN, IGSM and LPJ (Thornton et al. 2007, 2009; Sokolov et al. 2008; Xu-Ri & Prentice 2008) all represent state-of-the art N cycling models, none of them have applied the energetics-based plant N uptake theory developed by Vitousek, Rastetter and colleagues (Vitousek & Field 1999; Rastetter et al. 2001; Dickinson et al. 2002; Vitousek et al. 2002; Wang, Houlton & Field 2007). Currently, these models allow for plant N uptake to proceed at the rate of demand, and unmet demand results in a proportional downscaling of vegetation productivity. As part of the UK-based model development project, QUEST, a new energetics-based plant N uptake model (Fixation and Uptake of Nitrogen – FUN), has been developed (Fisher et al., unpublished data) within the JULES land surface model (JULES 2009) and that is coupled to the soil C and N model ECOSSE (Smith et al. 2007; Smith et al. 2009a). ECOSSE uses a five SOC pool approach (two plant debris and three soil organic matter pools) derived from the ROTHC (Coleman & Jenkinson 1996) and SUNDIAL (Bradbury et al. 1993) models (Fig. 4a), and with the incorporation of the FUN model allows plant N acquisition from all four major pathways: passive uptake through transpiration, re-translocation of N from senescing leaves, active uptake from the soil, and biological N fixation (Fig. 4b). All pathways except for passive uptake require C expenditure for N uptake based on the energetics theory (Vitousek & Field 1999). The model selects the least energetically costly N uptake routes, C is then allocated to optimize growth until N demand is met. Under N limitation, increasing C is expended on a decreasing soil N supply, resulting in less photosynthate C being allocated to growth.
There are a number of uncertainties in our understanding and modelling of soil N and plant N uptake that require further research. First, the impact of physical protection on soil organic matter decomposition is poorly understood and so introduces uncertainty into simulations of N supply especially following land use change (Oades & Waters 1991; Beare, Hendrix & Coleman 1994; Six, Elliott & Paustian 1999, 2000; Six et al. 2002; Del Galdo et al. 2003; Denef et al. 2004, 2007; Jiao, Whalen & Hendershot 2006). Second, the various N uptake processes operate across a range of time scales whereas they are confined to prescribed time steps in mathematical models. Third, existing PFT categorizations are based on C function, growth and competition that may not necessarily reflect their N cycling characteristics (i.e. assumptions with respect to the C:N ratios for different plant compartments may propagate uncertainty due to PFT definitions). Fourth, it is unclear under what circumstances plants favour NH4+ versus NO3− versus organic N uptake (Marschner, Haussling & George 1991; Falkengrengrerup 1995; Nordin, Hogberg & Nasholm 2001; Jones et al. 2005a,b; Weigelt, Bol & Bardgett 2005). Fifth, there is considerable evidence that other limiting nutrients including phosphorous (P) have an important regulatory function in terrestrial ecosystems (Lambers et al. 2008; Wardle et al. 2009). Finally, there is no unifying mathematical framework with which to accurately model and/or predict the amount of N supplied to the plant through mycorrhizal symbioses in exchange for plant C, despite the evident importance of this interaction (Allen et al. 2003; Smith et al. 2009b).
Drought impacts on vegetation
As climate change progresses, rainfall is expected to reduce in several regions of the world with extreme scenarios, such as that predicted by some climate models, leading to a large-scale loss of forest cover across the Tropical biomes (Cox et al. 2000; Salazar, Nobre & Oyama 2007; Sitch et al. 2008). DGVMs vary widely in their simulation of the impacts of reduced rainfall on forest productivity and carbon storage (Galbraith et al., personal communication). This is due both to differences in their parameters and in process descriptions between models. Here, we review some current approaches to the simulation of drought impacts on forest productivity, using Amazonia as a regional case study.
The sensitivity of ecosystems to drought depends on factors that determine both the onset of drought stress and factors that modulate the rate at which declines in productivity proceed once hydraulic stress has become established. Plant available water (PAW) is dependent on rainfall, soil physical properties and on the availability of soil water to plant roots. Key soil parameters that influence PAW include soil water holding capacity, soil hydraulic conductivity and soil water retention, a measure of the relationship between soil moisture content and soil water potential. The characteristics of the root system, including rooting depth and the distribution of roots at different depths, are also important determinants of PAW. The simulation of PAW in DGVMs is generally simple with soil divided into discrete layers or horizons and water percolating between them (e.g. Prentice et al. 2007). Often entire model grid cells (3.75 º longitude by 2.5 º latitude) are assumed to have a uniform soil texture, uniform soil water holding capacity and the same soil hydraulic characteristics. Values for key parameters determining PAW are presented in Table 2 for several DGVMs. Differences in parameterization can lead to large differences in the amount of water accessible to the vegetation in different models. For example, Harris et al. (2004) found that calibration of the soil hydraulics parameters in MOSES (precursor of JULES) with data from an Amazonian forest site resulted in a considerable reduction in PAW and an increased sensitivity to soil moisture stress in the model. The uncertainty in the representation of PAW will inevitably result in poor predictions of climate change effects on plant and ecosystem productivity and CO2 feedbacks to the atmosphere.
|Model (Ref).||Rooting depth (m)||Root distribution||Leaf:Root allocation|
|CLM-DGVM (Levy, Cannell & Friend 2004)||10.0||Fixed fraction in each of 10 layers.||Allocation to roots increases under drought|
|CTEM (Arora & Boer 2003)||Varies according to root biomass, up to 2.0||Varies with root biomass||Allocation to roots increases under drought|
|HYLAND (Levy, Cannell & Friend 2004)||Not explicitly considered||Not explicitly considered||Fixed|
|IBIS (Foley et al. 1996)||4.0||Exponential decline, following Jackson et al. 1996||Fixed|
|JULES (Cox 2001)||3.0||Exponentially declining||Fixed|
|LPJ (Sitch et al. 2003)||1.5||Fixed fraction in each of two soil layers||Allocation to roots increases under drought|
|MC1 (Bachelet et al. 2001, Daly et al. 2000)||1.5||Fixed fraction in each of 10 soil layers||Fixed|
|ORCHIDEE (Krinner et al. 2005)||1.25||Exponentially declining||Allocation to roots increases under drought|
|SDGVM (Woodward & Lomas 2004)||1.0||Not explicitly simulated||Allocation to roots determined by transpiration|
|SEIB-DGVM (Sato, Itoh & Kohyama 2007)||Not explicitly considered||Not explicitly considered||Fixed|
|SIB3 (Baker et al. 2008)||3.5||Exponential decline, following Jackson et al. 1996||Not explicitly considered|
The below-ground treatment of roots in DGVMs has been given considerably less attention than the above-ground vegetation (Woodward & Osborne 2000). At their most basic representation, such as in the HYLAND DGVM, roots are bypassed completely when simulating the effects of drought on vegetation. The amount of available water from a particular soil layer is usually weighted by the fraction of roots in that layer to determine total available water (e.g. as in the LPJ and JULES models). Rooting depth in most DGVMs is usually a PFT-specific parameter and tends to be not greater than 3 m (Poulter, Heyder & Cramer 2009), whereas in reality much deeper rooting depths have been observed in Amazonia (Nepstad et al. 1994; Bruno et al. 2006). The root distribution in most models is assumed to be constant, often declining exponentially with depth (e.g. JULES model). A number of root adaptation mechanisms have been proposed under drought conditions, including increased root production and root:shoot ratios (Joslin, Wolfe & Hanson 2000), vertical plasticity of the rooting profile (Schymanksi et al. 2008) and increases in the surface-area-to-mass ratios of fine roots (Metcalfe et al. 2008). However, due to limited data availability, these are generally not addressed in DGVMs. The phenomenon of hydraulic redistribution, or vertical movement of soil moisture by plant roots from deeper layers to more superficial layers, may also confer additional advantage to plants under drought conditions (Horton & Hart 1998; Oliveira et al. 2005) and has recently been incorporated into some ecosystem models (Baker et al. 2008; Lee et al. 2006).
A common strategy for simulating the effect of soil moisture stress on plant productivity in DGVMs is to use a dimensionless soil moisture stress factor, (β), which directly scales photosynthesis or stomatal conductance. For example, in JULES, the β scalar is defined mathematically as:
where β is equal to 1 when the actual soil moisture content (θ) (m3 H2O m−3 soil) is greater than the critical soil moisture (θc), equal to zero at soil moistures below the wilting soil moisture (θw) and decreases linearly between θc and θw. JULES assumes a direct effect of β on net leaf photosynthesis, A (Cox, Huntingford & Harding 1998):
where Ap is the potential rate of net photosynthesis (μmol CO2 m−2 s−1), without water stress. The parameters θc and θw are soil texture-specific and correspond to the soil moisture content at suction pressures of −0.033 and −1.5 MPa, respectively. More detailed formulations of the effect of drought on productivity exist, including optimality approaches (Cowan & Farquhar 1977a,b;Williams et al. 1996) and more detailed mechanistic models of stomatal behaviour (Buckley, Mott & Farquhar 2003). The former approach has been used successfully to simulate changes in vegetation water fluxes in a throughfall exclusion experiment in an Amazonian rainforest (Fisher et al. 2007). Such detailed formulations, however, have generally been absent from DGVMs, although Hickler et al. (2004) recently implemented a more detailed description of plant hydraulic architecture in the LPJ DGVM.
DGVM simulations of drought impacts on ecosystems could be improved in several ways. First, data on root and soil properties influencing PAW are still scarce for many regions of the world, especially the tropics. For example, only a small number of studies have provided data on soil hydraulic properties (Hodnett & Tomasella 1997; Belk et al. 2007; Fisher et al. 2008) and on root properties (Trumbore et al. 2006; Metcalfe et al. 2008) in Amazonia. Second, root water uptake is still poorly understood in many natural ecosystems. For example, the relationship between amount of roots available and the uptake of water from different soil layers is unclear, as soil depths with low root density have been shown to contribute significantly to evapotranspiration (Bruno et al. 2006). This misunderstanding is partly due to poor measurement of root activity, or the fraction of roots that are actively taking up moisture from the soil, although recent methodological advances are beginning to address this (Čermák et al. 2006). Third, work is needed to translate plant species differences in sensitivity to drought (e.g. Engelbrecht et al. 2007) into an optimal set of PFTs that best captures this variability. Current DGVM approaches to simulate competition for water between plant functional types (e.g. differences in rooting distribution) lack strong empirical support. Recently, differences in stomatal control strategies have been implicated as the cause of differential species mortality under prolonged drought in the United States (McDowell et al. 2008). Finally, more research into the relative importance of below-ground and above-ground processes and interactions in determining species responses to drought is ultimately required if uncertainty in C and N cycle predictions is to be reduced.
Typically, the distribution of PFTs and vegetation in DGVMs is controlled to a large extent by ‘climate envelopes’ defined as the range of temperature and rainfall conditions under which a particular vegetation type can establish with changes in vegetation occurring in response to shifts in these conditions (e.g. Smith et al. 2001; Bonan et al. 2003; Sitch et al. 2003; Woodward & Lomas 2004; Krinner et al. 2005; Sato, Itoh & Kohyama 2007). The use of climate envelopes limits the extent to which other factors, including soil and nutrients such as N, and disturbance can potentially affect vegetation. Most DGVMs are ‘area-based’ models in which grid cell fractions occupied by homogeneous populations of PFTs exist without any real age or size structure and do not mechanistically simulate the process of vegetation succession or competition for light resources between PFTs. In recent years, however, several advances have been made towards addressing these deficiencies. For example, the LPJ-GUESS (Smith et al. 2001) and SEIB-DGVM (Sato, Itoh & Kohyama 2007) DGVMs are based on individual-based forest gap models. These models can simulate vertical competition for light among individual trees in small patches within each modelled grid cell. Similarly, the Ecosystem Demography (ED) model (Moorcroft, Hurtt & Pacala 2001), recently coupled to JULES (Huntingford et al. 2008), simulates vegetation succession following disturbance on patches of land characterized by ‘time since last disturbance’. In this model, different cohorts of PFTs, each corresponding to a particular height and plant functional type, vertically compete for light by occupying different canopy layers. These approaches have allowed for vegetation dynamics to be represented in a much more ecologically realistic manner and will contribute to resolving long-standing DGVM weaknesses, such as inadequate simulation of tropical savannas. For example, Scheiter & Higgins (unpublished data) recently developed a DGVM to simulate the effect of fire on individual trees, as a function of height, resulting in improved simulations of current African vegetation. On the other hand, current PFTs do not account for functional groups such as bryophytes which are clearly important in C cycling in high latitude and temperate systems with globally significant C stocks below-ground (Beringer et al. 2001; Cornelissen et al. 2007a,b).
The opportunities and challenges that inevitably come from the development of new, more complex DGVM models are many. The absence of a model of vegetation disturbance and recovery impedes our ability to: (i) represent the time scales associated with vegetation change and succession; (ii) represent vegetation co-existence caused by vertical and spatial ecosystem heterogeneity; (iii) represent the mechanisms by which fire disturbance affects vegetation and vice versa (Grigulis et al. 2005); (iv) have an interactive soil N model; and (v) operate at the scale of ecological observations (i.e. individual trees) (Purves & Pacala 2008). However, the more complex the models become the more data and information they require, making them difficult to parameterize and leading to co-dependence of parameters on input variables. This can, in turn, lead to multiple stable equilibria in soil and vegetation status, as small perturbations of the initial equilibrium lock the system into a high- or low-nutrient status. Other multiple stable equilibria are likely to occur in fire-dominated ecosystems (Bond, Woodward & Midgley 2004), and in systems with strong biosphere–atmosphere feedbacks (Betts et al. 2004). This problem illustrates that the challenge of representing the natural world as a single general mathematical description is not necessarily made easier by the development of such complex models, but only by bringing our models closer to the real world, to real observations and experimental findings will we improve our chances of correctly predicting the role of the biosphere in future climate change.