Calibration of Erosion Models - Erosion Modelling - page 43

(see Cerdan et al ., 2002; Le Bissonais et al ., 2005).
It allows one to pinpoint the appearance of fields
that are likely sources and sinks during a growing
season. Jetten et al . (1996) also showed that while
different patterns of runoff contributing areas can
give the same discharge and sediment loss, these
crusting classes help in choosing the likely source
fields when there are limited point observations.
(ii) Field erosion measurement . Takken et al .
(1999) attempted a spatial validation of LISEM
using the measured erosion pattern in a 2.9 km 2
catchment in Belgium resulting from an extreme
event on 8 June 1996. By measuring rills and
other erosion and deposition features, they esti-
mated sediment loss aggregated over 12 land use
types. These were then compared with LISEM
simulation results which were aggregated in the
same way. The results show that the observed
variation in erosion rates for different crop types
was not well predicted, with an overprediction of
erosion rates on fields with a high vegetation
cover. Correlation between predicted and meas-
ured erosion rates per crop type at locations of
the measured rill transects was weak. Predicted
and observed patterns of deposition were gener-
ally similar, although important deposition
against vegetation barriers and roads was not
simulated, which resulted in overall discrepan-
cies (Takken et al ., 1999). Hessel et al . (2003a)
attempted a similar exercise but on a pixel-
by-pixel basis for several events in a small catch-
ment on the Chinese loess plateau (see
Chapter 12). They showed the difficulties of
comparing predicted and observed patterns, as
there were both areas of good prediction but also
large discrepancies. They attributed this to both
simulation errors (e.g. LISEM has the tendency to
overestimate deposition in certain grid cells) and
inherent uncertainties in the dataset. For
instance, if the DEM is not accurate enough to
simulate a converging flow path in the right posi-
tion compared with reality, a model can never
produce correct erosion estimates on that loca-
tion. Also the model simulates all detachment of
soil particles, while in reality erosion is only vis-
ible once rills are being formed. Thus if a spatial
comparison is attempted, an aggregation to fields

or slope segments seems to be preferable over a
pixel-based comparison.
(iii) Farmers' knowledge . Vigiak et al . (2007)
compared the predicted erosion maps of five mod-
els to the field observations done with the ACED
method (see above). One of the models was a deci-
sion tree based on farmers' knowledge in the area.
This knowledge consists both of observed erosion
features, and also of related factors such as poor
crop development, soil colour changes and stoni-
ness/rockiness. The decision tree appeared to be
very capable of predicting the observed eroded
area, indicating the strength of this kind of data
for use in model comparisons and calibration.
(iv) 137-Caesium patterns and chemical proxies .
Several models have also been tested with 137 Cs
data (De Roo & Walling, 1994; Walling et al .,
2003; Van Oost et al ., 2003; Owens & Walling,
1998; Porto et al ., 2003; see also Section 14.7.2).
For example, Walling et al . (2003) compared the
simulated erosion and deposition patterns of the
models ANSWERS and AGNPS with the observed
distribution derived from 137 Cs observations in
two small catchments of 4.6 and 0.52 km 2 in the
UK. Although both models produced catchment
total runoff, peak runoff and sediment losses in
agreement with the measured data of seven
events, the predicted spatial patterns of soil redis-
tribution and the sediment delivery ratios were
very different for the two models. The AGNPS
model predicted deposition zones at the bottom
of the slopes, with highest erosion at mid-slope,
in accordance with the 137 Cs-derived distribution,
while ANSWERS showed a continuous increase
of detachment from top to bottom of the slopes.
Also Ritchie et al . (2005) showed that it is pos-
sible to derive quantitative erosion/deposition
patterns from 137 Cs fallout in a semi-arid water-
shed in Arizona, which could then be used for
model verification. Van Oost et al . (2003) included
tillage displacement using the WATEM/SEDEM
model to further explain 137 Cs patterns found in a
Belgian catchment. Also other chemical elements
may enable model verification. Van der Perk and
Jetten (2006) showed that a pattern of Cu concen-
trations could be established in a French vineyard
due to long-term use of copper as a fungicide.