Modelling Plant Ecology - Environmental Modelling: Finding Simplicity in Complexity - page 203

a forest canopy does not necessarily match that of light,
with more N at the bottom and less at the top than
expected from the rate of exponential decay. Therefore,
it is likely that at least one additional axis of optimization
is employed by plants in their distribution of nitrogen.
Possible causes include water limitation at the top of the
canopy, and the enhanced assimilation of diffuse light at
the bottom, compared to the Beer's Law approximation.
Again, to enable the optimization approach to work
effectively, it is necessary to represent the exact target(s)
of the optimization process.
Another set of simplifications employed by some
DGVMs is the linkage between different plant traits along
'tradeoff' gradients. Wright et al . (2004) demonstrate the
existence of a global trade-off surface for leaf properties
over a wide range of climate and life forms. The observed
relationships between leaf mass per unit area, leaf lifes-
pan, nitrogen content and assimilation rate for a sample
of 2548 species in their database all fall along a single axis
of variance. Therefore, given one plant trait, for example
leaf lifespan, all the other properties should be read-
ily predictable. The existence of plant tradeoff surfaces
derives from the competitive evolution of plants - those
with suboptimal allocations of resources to leaf-defence
compounds and photosynthetic structures in a given
environment are less likely to persist in the gene pool;
therefore a tradeoff surface is formed consisting of all the
successful allocations strategies. This property is used in a
basic form in some DGVMs (Moorcroft et al ., 2001; Sitch
et al ., 2003) but increased use of observed correlations
between plant properties is likely to be a feature of future
modelling efforts.

area of land simulated, larger scale heterogeneity resulting
from disturbance processes cannot be captured, and the
output is still stochastic, owing to the random nature of
tree mortality in a small plot. To resolve this issue, Moor-
croft et al . (2001) proposed a 'size and age structured
approximation' of a gap model, called the 'Ecosystem
Demography' approach (ED). The ED model overcomes
these issues by dividing the land surface into a series
of 'patches', each corresponding to a different stage of
vegetation succession. Within these patches, the individ-
ual trees represented by a gap model are grouped into
'cohorts' of similar size and plant functional type. Each
cohort is represented in the model as a single average
tree, so the resulting array of tens of cohorts per site thus
constitutes a compromise between traditional DGVMs,
which typically track the properties of one average tree
per plant type, and gap models, which track hundreds or
thousands of trees per geographical unit.
The hope of second-generation models is that, by
simulating ecosystems at the scale at which ecological
processes occur and at which most observations are made
(the tree and stand scale) it should be possible to: (a) rep-
resent timescales appropriately in climate predictions,
(b) constrain the parameters in vegetation models using
a greater range of ecological observations and (c) better
represent biodiversity and the co-existence of different
vegetation types within ecosystems (Purves and Pacala,
2008). To expand on this latter point, because of the
spatially homogenous growing conditions and simple
representations of plant diversity represented, DGVMs
typically struggle to represent the co-existence of even
two or three vegetation types. This low functional diver-
sity means that, when climatic conditions change such
that the one or a few plant types occurring in a given
location can no longer persist, rapid 'dieback' is likely
to occur. The expectation is that, in an analogous real-
world situation, the co-occurrence of other, slightly better
adapted vegetation types will likely reduce the impact of
the changing conditions on ecosystem properties rele-
vant to climatic feedbacks (leaf-area index, total carbon
storage, albedo, evaporative flux).
Ideally, future vegetation models should be able to
represent the scope and co-existence of more numerous
vegetation types. However, in classical ecology, explain-
ing the reasons for co-existence of species has in fact
proved elusive. Clark et al . (2007) refer to this lack of
explanation as the 'biodiversity paradox'. They propose
that a paradox exists between the theories of plant species
co-existence and the observed evidence. The dominant
theory espoused in ecological literature (Pacala et al .,

12.3 The research frontier

Recently, a new 'second generation' of vegetation models
has been developed that combines the global prediction
capacity of DGVM models with the ecological process rep-
resentation of gap models. Two contrasting approaches
are used to achieve this. First, the LPJ-GUESS model
(Smith et al ., 2003; Hickler et al ., 2008) and the SEIB
(spatially explicit individual based) DGVM (Sato et al .,
2007) both employ a gap-model approach at a global
scale by simulating only a small area ( 30 m)

of land. This approach has many benefits over a tra-
ditional DGVM - light competition, canopy structure,
vegetation coexistence along light competition gradients
and realistic timescales of responses to change can all
be simulated more effectively. However, given the small

∼ 30 m ×