Geology Reference
In-Depth Information
would enable the model LISEM to predict accu-
rately all events in a small catchment in China
(see also Chapter 12).
An alternative approach is to use decision
support-based models that apply a rating to
erosion factors. One example is STREAM, a
physically-based distributed model that gives
simulations lumped for single events. It uses
observed crusting classes according to a system
developed in northern France for silty loam (loess)
soils (Cerdan
et al
., 2002). These classes are
related to infiltration rates and surface roughness
values in a built-in dataset in the model. The
strength of the model lies in the fact that an
accurate image of runoff-contributing areas in a
catchment is made by observing crusting classes,
which allows good predictions in spite of the fact
that a classified system is used.
A classified approach is also used by De Vente
et al
. (2008) who employ a Spatially Distributed
Scoring Model (SPADS) to predict the annual
sediment yield of 61 basins in Spain. This model
is based on a rating with scores of 1-3 for vegeta-
tion cover, topography, lithology, annual rainfall,
gully density and inverse distance from a river
stream. This yields a catchment index which is
then related to the area specific sediment yield
(SSY) by a regression analysis. De Vente
et al
.
compared this approach to predictions of SSY
from the models WATEM/SEDEM and PESERA.
After splitting the dataset into basins with high
(>5%) and low (<5%) sediment delivery ratios,
good results were obtained, where SPADS
explained 67% of the variation and WATEM/
SEDEM 48%, while PESERA did not perform
well. The poor performance was attributed to the
lack of gully erosion, channel erosion and chan-
nel sediment delivery processes in PESERA.
However, the explained variance is not the most
important result of this analysis; instead it is the
ease with which specific data (such as gully infor-
mation) can be included in a factorial analysis. It
allows the easy testing of the relative importance
of certain processes in a given area, which may
shed some light on the performance of other ero-
sion models. Also clear from this analysis is the
fact that Sediment Delivery Ratios are very area-
specific and cannot be used in different areas
without a thorough analysis. This is significant
for modelling at larger scales.
3.2.3
Calibration of large-scale models
There have been many attempts to calculate the
soil loss from large areas, such as river basins,
countries or even continents. The difference
between 'catchment scale' and 'large scale' is
assumed here to be a generalization of the spatial
characteristics of an area. Where the catchment
models explicitly characterize areas of detach-
ment and accumulation, in large-scale models
many processes happen inside a spatial element,
such as grid cells of >1 km
2
, first-order catch-
ments or even administrative units. Two types of
models seem to be used: a direct use of RUSLE
with minor or major adaptations, and more
sophisticated models that attempt to retain proc-
ess descriptions for runoff, detachment, transport
and deposition. There are several purposes for the
model results at this scale, from showing average
annual soil loss (often called erosion risk) to use
by policy-makers (see Van der Knijff
et al
., 2000)
to predict the effects of soil conservation policies
on the sediment loads of rivers and the siltation
of dams and reservoirs. Consequently calibration
at this scale varies from non-calibrated, through a
non-quantitative verification of spatial erosion
patterns, to a quantitative comparison of pre-
dicted and observed sediment levels in rivers or
sedimentation behind dams.
It is clear that the USLE, originally meant to
predict annual soil loss from a slope or field, can-
not be used directly to predict soil loss on a large
scale, unless adaptations are made. The original
USLE is based on soil loss by rainfall energy modi-
fied with slope angle and slope length, which are
used as a proxy for the flow detachment processes.
The RUSLE and MUSLE improved this by adding
a runoff factor to the driving force. There is usu-
ally no provision for deposition in these models.
The popularity of the model on a variety of scales
probably stems from the ease of use in a GIS, in
particular the derivation of the slope length L and
slope angle S from a DEM that are combined in
