Geology Reference
In-Depth Information
cover in inter-rill areas, and
BASI
is the produc-
tion of the fraction of basal surface cover in inter-
rill areas and total basal surface cover.
These procedures have been developed using a
calibration and validation dataset and are there-
fore subject to the same limitations as statistical
erosion models. As process-based models are very
sensitive to input data and/or parameter values,
one should therefore only apply them for these
conditions where the statistical input value esti-
mation procedures were properly validated. If
input values need to be estimated for conditions
that are different from those for which input
value estimation procedures were developed,
erroneous results may be expected.
Even if the model is used in conditions similar
to those for which input value estimation proce-
dures were developed, there will be considerable
uncertainty associated with the parameter values
obtained from the estimation procedures, and the
resulting uncertainty on the final model result
may well be bigger than that for a statistical
model. Quinton (1997) convincingly showed that,
in a simple application of the EUROSEM model,
the impact of input value uncertainty on the final
model result was such that a meaningful model
evaluation became impossible. The range of pos-
sible output simulations was such that the meas-
ured outcome always fell within the lower and
upper possible limits, and it was therefore impos-
sible to evaluate the model properly; given the
uncertainty in parameter estimates, it was possi-
ble that the model properly simulated the meas-
ured events but poor simulations were in fact
equally probable (see also Chapter 5).
The uncertainties in input values are not only
due to the statistical nature of parameter estima-
tion techniques. There is also a considerable error
associated with the actual measurement or esti-
mation of the data necessary for input value esti-
mation. We often have to rely on very few and
very small samples or a limited number of point
measurements to estimate soil bulk density, mois-
ture content, roughness and vegetation cover. The
loss of prediction accuracy due to uncertainty in
both input data and in input value estimation may
outweigh possible gains in accuracy due to a better
process description, thereby leading to a decrease
in model performance with increasing model com-
plexity (Van Rompaey & Govers, 2002).
This does not imply that we should not apply
process-based, spatially distributed models to
ungauged fields and catchments, but we should
not expect their predictions to be quantitatively
superior to those obtained with simpler, statisti-
cal or black-box model structures.
A priori
cali-
bration/evaluation is as necessary for spatially
distributed process-based models as for statistical
models; neglecting this prerequisite may lead to
erroneous evaluations which may have impor-
tant practical implications (Nyssen
et al
., 2006;
see also Chapter 3).
7.3.4 Misconception: Erosion models
now simulate what is needed by science
and society: further model development
and testing is therefore unnecessary
Given the fact that a wide range of soil erosion
models exists, one might indeed expect there to be
an erosion model for every potential application
either from a scientific or a societal viewpoint. We
would argue that this is not (yet) the case. Below
we focus on two 'missing links' which are rela-
tively poorly represented in most erosion models:
the modelling of sediment size-selectivity and the
modelling of erosion-soil interactions.
Current needs with respect to soil erosion mod-
elling are no longer limited to the simulation of
the total amount of soil redistribution within a
landscape. Recent literature devotes a conside-
rable amount of attention to the movement of
sediment-associated nutrients, pollutants and
organic matter. As the latter are mainly bound to
the fine fraction, understanding how the fine soil
fraction is mobilized and transported through the
landscape becomes crucial. Many current models
have the ability to deal with different size frac-
tions and are capable of simulating the redistribu-
tion of sediment-associated soil constituents.
Various approaches have been proposed, ranging
from the use of simple enrichment ratios in
association with statistical models (Foster
et al
.,
2003), to process-based approaches describing the
