Geology Reference
In-Depth Information
interaction between vegetation growth (using a
continuous model) and soil erosion processes at
different spatial scales (using an event-scale
model).
As for the selection of a specific model to
implement the study, some recommendations by
Jetten
et al
. (1999) can be taken into account:
●
input data quality (both quantitative measure-
ments and qualitative knowledge of the study
area), calibration procedures, and the knowledge
of modellers can be more important than model
structure for successful erosion simulations (see
Chapter 3 for further discussion);
●
models usually perform better for the processes
and at the spatial and temporal scale they were
designed to operate in, so evaluating different
model structures and applications can provide an
insight on the most appropriate model for a given
case study;
●
when modelling for changing conditions, proc-
ess-based models can accommodate processes
that do not currently occur, while empirical
approaches are constrained by currently operat-
ing processes.
Finally, when designing a modelling study,
some attention should also be given to calibration
and validation strategies for climate change sce-
narios. Calibration and validation is a complex
process, especially for models requiring large
amounts of input data. Typical problems include
the lack of measured data at the appropriate
scale used by the model; parameter equifinality,
where different sets of calibrated parameters pro-
vide equally good results; and over-calibration,
where model parameterization is optimized using
an excessively small sample of observations (e.g.
Quinton, 1997; Beven, 2000; Boardman, 2006).
Since models usually perform best for the range
of conditions for which they were calibrated
(Favis-Mortlock
et al
., 2001), calibrating and vali-
dating a model for future conditions presents a
number of additional problems. For example, cal-
ibrated model parameters can have limited trans-
ferability in time, particularly in the face of
significant changes to climate parameters or
watershed conditions (Apaydin
et al
., 2006).
Parameter equifinality can present a similar
challenge, since parameter sets performing
equally well for current conditions can lead to
significant differences in climate change predic-
tions (Wilby 2005). Toy
et al
. (2002) defined a
robust model as one able to perform reasonably
well with similar parameter values, including
highly dynamic ones, for the widest possible
range of conditions; calibration and validation
problems for uncertain future conditions call into
question the robustness of runoff and erosion
models for climate change analysis (Beven, 2000;
Morgan & Quinton, 2001).
To address this problem, models used for cli-
mate change studies should demonstrate an
increased degree of robustness considering both
current conditions and those as close to possible
changes as achievable. One important strategy to
increase robustness is to demonstrate that the
model can simulate alterations to hydrological
and erosion processes caused by changes in cli-
mate; this can be achieved using a 'space-for-time'
approach, where the consequences of future cli-
mate change are studied using a comparative
analysis between one study area and another with
climatic characteristics resembling GCM predic-
tions (Imeson & Lavee, 1998). In practice, this
strategy can be implemented by calibrating and
validating a model for different study areas with
different climates and hydrological and erosion
processes operating, or by using periods with dif-
ferent climate conditions in the calibration and
validation process, especially if these conditions
represent in some way the expected climate
change scenarios (e.g. Bronstert, 2004; Xu &
Singh, 2004). This approach can be further
detailed by reproducing climate change in the
calibration and validation process; for example,
Xu and Singh (2004) proposed that, if the goal is
to simulate a drier climate scenario, a model
should be calibrated for a wet year and validated
for a dry year, thus demonstrating its ability to
simulate a wet/dry transition.
Another approach to increase model robust-
ness is multi-process validation, i.e. to calibrate
and validate a model for the highest possible
number of variables representing different proc-
esses occurring at different scales, such as splash,
