Geology Reference
In-Depth Information
2500
3000
Calibration
Validation
2500
2000
2000
1500
1500
1000
1000
500
500
0
0
0
500
1000
1500
2000
2500
0
500
1000
1500
2000
2500
3000
Q obs (m3)
Q obs (m3)
Kineros2
Erosion3D
Medalus
LISEM
AGNPS
ACRU
Kineros2
Erosion3D
Medalus
LISEM
AGNPS
ACRU
Fig. 3.1
Breakdown of the results of the GCTE catchment comparison per model (after Jetten
et al
., 1999).
model and showed that even with optimized val-
ues for saturated hydraulic conductivity (
K
sat
),
prediction was moderate especially for smaller
events. Bathurst and Lukey (1998) compared
observed and predicted soil loss from the SHE
model for the Draix area (France) and showed
results varying from accurate up to a factor 100
difference. Brochot (1998), on the other hand,
obtained good results for the same area using a
simple regression equation between soil loss, pre-
cipitation and infiltration, suggesting that model
complexity and area complexity play a role in
how well a model performs. De Roo (1993) com-
pared the lumped USLE and MMF models with
the spatially distributed ANSWERS and KINEROS
models for three small Dutch catchments, and
concluded that they performed equally well when
the models were tested for the same type of out-
put (annual soil loss).
From the above it is clear that more empirical
models seem to perform equally well compared
with more physically-based models. For exam-
ple, RUSLE variants are used with some success
when coupled to a Sediment Delivery Ratio
(SDR) or runoff estimates using the SCS Curve
Number method, to calculate the fraction of sed-
iment reaching a stream. Arhonditsis
et al
. (2002)
simulated sediment production of a 194 km
2
catchment in Greece, and showed it was possible
to use the USLE at this scale by adding an SDR
term based on the distance (d
k
) between a sedi-
ment source area
k
and the channel: SDR
k
d
k
−0.34
where 0.34 is the calibrated value. Furthermore,
they changed the slope length exponent from 0.6
in the original USLE to 0.72 for their dataset.
Tyagi
et al
. (2008) coupled the USLE, the SCS-CN
method, Horton's infiltration equation and a res-
ervoir routing method to simulate soil loss with
28 calibration and 26 validation events from three
Indian and four US catchments. They showed
good results for total sediment yield and moder-
ate results for peak sediment yield, and concluded
it was important that the calibration dataset
included events that represented different ante-
cedent moisture contents and crop growth stages.
In other words, the wider the range of conditions
in the calibration dataset, the better the validation
results will be. This was also the outcome of the
GCTE plot-scale comparison (Boardman and
Favis-Mortlock, 1998). On the other hand, as noted
earlier, Hessel
et al
. (2003b) showed that it was
not possible to find a single calibration set that
=
