Geology Reference
In-Depth Information
differences of more than one order of magnitude
for all discharge sizes. The scatter for the
daily discharges was much larger than for the
annual discharges. Moreover, in the continuous
simulation a number of discharges were observed
but not simulated, and, for certain rainfall events,
runoff was simulated but not observed. This was
especially the case for smaller events. Boardman
and Favis-Mortlock (1998) concluded amongst
other things that: calibration is almost always
necessary and is most effective if the event(s) to
be estimated lies inside the range of calibration
events. For continuous simulation models, long-
term average results are better simulated than
results for individual time periods. In general,
results are less good for shorter time periods,
although there are exceptions. With these models
it is more difficult to predict days when runoff or
erosion does occur than days when it does not
(when predicted independently).
The difficulty of predicting adequately both
large and small events was remarked upon by
Sadeghi
et al
. (2008) who used the process-based
Hillslope Erosion Model (HEM) to simulate a
series of erosion plots in Iran. They did not suc-
ceed in calibrating HEM using the erodibility fac-
tor alone. The non-linear response of the plots to
more severe events could only be simulated by
applying a non-linear regression between pre-
dicted and observed values. Risse
et al
. (1993)
tested the USLE for over 1700 years of erosion on
208 plots and obtained a reasonable R
2
of 0.58,
with strong overprediction of small events.
Nearing (1998) also remarked that models seemed
to have difficulties in predicting small-scale
events. Nearing
et al
. (1999) and Nearing (2000)
compared the soil loss of pairs of adjacent plots in
the US, selecting plots that can be considered
identical in terms of received rainfall, soils and
surface conditions. One plot from a pair was con-
sidered as a predictor of soil loss of the adjacent
plot. The coefficient of variance (CV) decreases
from 150% for very small events (0.01 t ha
−1
) to
1% for very large events (about 400 t ha
−1
), which
indicates the impossibility of correctly predicting
soil loss of small events with a mathematical
model that is always a simplified representation
3.2.1
Calibration of field/plot scale models
Plot- and field-based research forms the basis of
many erosion studies. The sediment loss from a
plot or field (i.e. a small, relatively homogeneous
spatial unit) is one of the most available experi-
mental data. Experiments at this scale are rela-
tively easy to set up, and a close link exists to
farming units so that direct advice can be given to
farmers for practices that will reduce soil loss.
There are also clear disadvantages: the smaller
the plot, the more rainsplash detachment is
favoured over flow detachment, as larger rills and
gullies will almost never occur on erosion plots.
Also, deposition will not take place on smaller
plots unless there is a break in slope angle or a
change in vegetation. The objective of plot-based
erosion modelling is usually to simulate soil loss
from a single homogeneous unit, whereby a vari-
ety of land uses and soil conservation measures
are often tested, such as mulching, minimum till-
age, and green cover.
There are generally two types of models that
are used: lumped models based on multiple
regression type equations (such as the USLE and
derivatives), and physically-based process mod-
els. The first calculate annual soil loss based on
average rainfall kinetic energy and peak flow (in
the case of RUSLE and EPIC) and use methods
like the SCS Curve Number to derive a runoff
coefficient. The second are models that calculate
runoff from infiltration excess and simulate
splash detachment, flow detachment and sedi-
ment transport (such as WEPP, EROSION 2D,
PSEM2D and HEM). In addition, it should be
noted that most catchment-based models can be
applied as hillslope models.
In the GCTE hillslope comparison exercise
(Favis-Mortlock, 1998), six models were tested
using 73 years of data from seven sites in three
countries. The sites from which the data were
drawn range from plots of 0.01 ha to small catch-
ments of just under 10 ha, with slopes varying
from almost 1% to 18%, a range of soil types, and
land use including both agriculture and natural
vegetation. The results showed a large scatter
between predicted and observed values, with
