Calibration of Erosion Models - Erosion Modelling - page 35

of reality. Although the low CV for large events is
encouraging, small events are of much greater
interest in recent studies of eroded agricultural
lands as a source of polluted sediment: only small
quantities of sediment are sufficient to pollute
surface water if the clay particles are saturated
with agrochemicals.

grazing (often overgrazing), small field sizes and
torrential rainfall. On the other end of the spec-
trum, land reallocation in western Europe has
caused the creation of uninterrupted large fields
that produce runoff and sediment even at low
slope angles. From the point of view of this multi-
purpose modelling, proper calibration at this scale
is perhaps the most difficult to achieve.
The GCTE catchment model comparison test
was reported by Jetten et al . (1999). Six models
were tested using five calibration and five vali-
dation events in a 40 ha catchment in The
Netherlands. Calibration was mostly done using
parameters that influence infiltration (such as ini-
tial moisture and saturated hydraulic conductiv-
ity). It appeared that the overall performance of
the models was moderate for the calibration data-
sets, and less good for the validation datasets (see
Fig. 3.1). Total discharge predictions were reason-
able, and generally better than peak discharge,
while both were better predicted than sediment
discharge. It should be said that some models
were not meant to be used at this scale, and there-
fore may not have performed well. Also, contribu-
tors were asked to find average calibration factors
for the five training events, which were quite dif-
ferent in magnitude. Fig. 3.1 shows that some
models have difficulty with predicting larger
events when calibrated for small events. This may
also have influenced the results. Whilst individ-
ual calibration of each rainstorm would probably
have yielded better calibration results, it would
still have been interesting to see how this would
have affected predictions for the validation events.
For example, Hessel et al . (2003b) analysed the
performance of LISEM for a small catchment in
northern China and concluded that it was impos-
sible to find a single calibration set that would fit
all measured rainfall events. From the discussions
during the GCTE meetings it was clear that addi-
tional 'soft' information, in particular when mod-
ellers use knowledge of changes in soil structure
as a result of agricultural activities and/or climate
in their calibration strategy, can improve the qual-
ity of input data and model results.
Similar results were obtained in other model
comparisons. Zhang et al . (1996) tested the WEPP

3.2.2 Calibration of catchment-scale models

The term 'catchment scale' is used rather loosely
here to indicate a range of scales from catchments
of roughly 1 km 2 to several hundreds of km 2 . The
difference between the 'plot' and 'catchment'
scale is the detailed spatial characterization of a
catchment, either using grid-based GIS systems
or dividing the catchment into functional ele-
ments such as slope segments, channel segments
and ponds.
Generally two types of models are used for
catchment-based simulations: models that use
breakpoint rainfall data to model individual rain-
fall events, or series of rainstorms, and models
lumped in time that simulate annual or multi-
annual erosion (Jetten and Favis-Mortlock, 2006).
Most models at this scale are physically-based
and attempt to simulate the individual hydrologi-
cal and sediment transport processes.
There are multiple objectives at this scale: (i) to
simulate soil loss from the catchment, e.g. to
quantify downstream floods and sediment prob-
lems (Boardman et al ., 1994); (ii) to gain insight
into the spatial distribution of sources and sinks
in a catchment, e.g. to estimate the effect of con-
servation measures and agricultural policies
(Jetten and De Roo, 2001; Lundekvam et al ., 2003;
Hessel et al ., 2003a; Hessel and Tenge, 2008); (iii)
to study connectivity between sources and sinks,
e.g. to emphasize the role of tillage and wheel-
tracks (Takken et al ., 2001) and crusting (Le
Bissonnais et al ., 2005); and (iv) to predict the
impacts of future changes in, for example, climate
and land use (Nearing et al ., 2005). The results
can be very different for different climates and
land uses. Semi-arid areas are known for their
limited connectivity between hillslopes and
stream beds, combined with mixed land use of