Geology Reference
In-Depth Information
when the erosion took place. At present, perhaps
too many model users are obsessed with predicting
numbers when, for many practical purposes, it is
enough to know where and how frequently erosion
occurs. Users should be aware, however, that when
using a model to predict the location of erosion, its
sensitivity may differ from when it is used to pre-
dict the amount of erosion. LISEM, for example, is
very sensitive to the size of the grid cells when
used to predict the spatial patterns of erosion
observed in the landscape (Jetten
et al
., 2003).
Model users should also consider whether it is
sufficient to be able to predict the direction and
magnitude of changes in erosion: relative values
rather than absolute numbers. If rainforest is
cleared to make way for rubber or oil palm plan-
tations or a tourist complex, environmental man-
agers and planners will need to know by how
much erosion and sediment delivery to adjacent
watercourses will increase. Some models may
well predict the percentage change in erosion
closely enough for practical purposes, even
though the absolute predictions of erosion rates
are an order of magnitude adrift. Similarly, the
ability to predict order-of-magnitude changes in
erosion as a result of changes in rainfall amounts
and intensity will be valuable in assessing the
effects of climatic change.
oriented and more complex, it is a more difficult
exercise. Many models now require input data for
20 or more parameters and it is virtually impos-
sible to calibrate a model on more than about four
parameters. As already noted, these models can
produce identical predictions of erosion through
different routes. Different parameters may need
to be calibrated depending on whether erosion is
limited by the rate of sediment detachment or the
sediment transport capacity. Where the same
parameter affects both sets of processes, it may
take different calibrated values for each set. A fur-
ther complication is that calibrated values usu-
ally pertain to a very specific set of conditions, so
that those for soil properties when the soil is bare
may not be the same as those when the soil has a
crop or vegetation cover. Values for bare soil may
also differ between winter and summer, or
between the start and the end of a rainy season.
In other words, as is the case with measured
parameter values, calibrated values are likely to
vary in time and space. There is also the issue that
many of the process-based models give several
outputs. EUROSEM, for example, provides for
each storm a prediction of total runoff, total soil
loss, peak runoff, peak sediment concentration, a
storm hydrograph and a storm sediment graph. It
is possible to calibrate the model to produce good
predictions of total runoff and total soil loss with-
out being able to reproduce acceptable hydrographs
and sediment graphs (Quinton & Morgan, 1998).
Overall, the logistics of obtaining appropriate sets
of calibrated parameter values are such that many
model users now take the view that calibration is
an impossible exercise and has no meaning or
validity for complex process-based models.
Others, however, believe that calibration is vital
and should be attempted by all model users (see
Chapter 3 and sections 7.3.3 and 14.4).
2.8.3 Calibration
One approach to assigning values to data inputs
which are hard to obtain or difficult to measure
accurately is to determine what values are required
to allow the model predictions of erosion to match
observed values. This is the process of calibration.
In the 1960s and 1970s calibration was common
practice and widely used with hydrological models
to determine parameters like Manning's
n
, rainfall
interception and saturated hydraulic conductivity.
Before adopting a calibrated value for a parameter,
a check should be made to ensure that its value
lies within the range of those observed in the field
or obtained experimentally in the laboratory.
More recently, calibration has been viewed by
many model users with some cynicism, largely
because, as models have become more process-
2.8.4 Validation
Before erosion models can be used to develop pol-
icies or design management systems, the user
must have confidence that the model outputs are
realistic. This means that the chosen model must
be validated. Validation is usually achieved by
