Geoscience Reference
In-Depth Information
Czerwinski et al., 2007). The complexity and range of data and parameters required
for these models make them unattractive or unfeasible for most management situa-
tions however (Tan et al., 2006; Pitchford et al., 2007). The relative merits of such
models are reviewed by Tan et al. (2006). Here, examples of predictive potential of
models in two intensively managed ecosystem types (arable agriculture and marine
fi sheries) are given.
Predictability in intensively managed ecosystems:
arable agriculture
Predictability of ecosystem processes can be limited even in ecosystems that have
been greatly simplifi ed or refi ned, and are intensively managed, to ensure that condi-
tions remain relatively fi xed. Agricultural ecosystems are artifi cial, but are designed
and managed to ensure a fi xed return on an ecological resource (e.g., biomass of a
desired crop species) for a continued investment of resources (water, nutrients,
labour). Agricultural ecosystems are simplifi ed systems in which environmental
conditions are made as homogeneous as possible (table 25.3). Despite these attempts
to create ecosystems which are essentially biomass factories, spatial and temporal
variation in crop yields is still common, due to variability in broader environmental
factors which are hard to control or mitigate, e.g., changes in climate and weather
conditions, and to residual variability in factors which are controlled, but cannot
be controlled precisely enough, such as spatio-temporal variations in soil nutrients
(e.g., Basso et al., 2001; Batchelor et al., 2002).
Because of the economic value of agricultural ecosystems, many attempts at yield
prediction using crop models have been made, with varied success. Initial models
were simple, and were based on regression techniques to compare yield to environ-
mental parameters (e.g., Jones and Ritchie, 1996). These failed to adequately predict
yields due to the many non-linear spatial and temporally variable factors which
impact upon crop yield, such as intraspecifi c competition, crop population densities,
weather, pest and pathogen dynamics and water and nutrient dynamics, all of which
lead to complex patterns of plant stress (e.g., Batchelor et al., 2002). Process-
oriented crop models examine the effects of this variability more explicitly and
therefore predict single-crop yield under different environmental and management
scenarios, but are still limited in their application to individual agricultural ecosys-
tems and can display notable error margins (e.g., Basso et al., 2001). Examples
include models such as CROPGRO (Boote et al., 1998) and CERES-Maize (Jones
and Kiniry, 1986), which simulate daily growth of soybean and corn respectively
in relation to differences in carbon, nitrogen, and water inputs. These models
assumed spatial homogeneity and predicted yields based solely on temporal varia-
tion of environmental conditions.
More recently attempts have been made to refi ne these models to account for
spatial variation in environmental parameters. Batchelor et al. (2002) evaluated the
performance of these models using different methods of spatial validation. Environ-
mental (fi eld capacity, rainfall, temperature and solar radiation) and management
(planting date, row spacing, genotypes) characteristics were recorded at a spatial
resolution of 0.2 ha, with 100 points being measured in total (over a 20 ha fi eld).
These inputs were fed into the CERES-Maize model, and simulated yield was cor-
related to measured yield over three years, with an overall coeffi cient of 76 percent.
Although the source of the remaining variability was unknown, measurements of
