Geology Reference
In-Depth Information
For some changes it is difficult to assess whether
the improvement to the model also results in
an improvement in simulation. If one starts
with a calibrated model and then implements a
theoretical improvement, it can be expected that
the adapted model gives less good predictions. To
evaluate whether the implemented change is an
improvement in terms of simulation accuracy,
one has to recalibrate the model. If this results in
either a better fit with observations or the use of more
realistic calibrated parameter values, the change
can be considered an improvement. This method
most easily applies to changing Manning's
n
and
to slope correction because these changes affect
the hydrograph. Therefore, the hydrographs can
be compared with the predicted hydrographs of
the original version. The other changes affect only
sediment transport and are more difficult to test
since the measurements of sediment concentration
are less frequent and probably less reliable, so that
there is also uncertainty about the accuracy of the
measurements.
(1)
Saturated conductivity (
K
sat
).
K
sat
is the model
parameter with the largest influence on discharge.
(2)
Initial suction. Initial suction determines the
unsaturated conductivity (and thus infiltration)
during the start of a rainfall event. Initial suction
was only used for calibration if calibrating on
saturated conductivity proved insufficient.
(3)
Manning's
n
. This influences the velocity of
runoff and therefore affects the shape and timing
of the hydrograph.
(4)
Channel length. In LISEM, pixels can be
defined that contain a channel characterized by a
separate Manning's
n
. The width of these chan-
nels can be defined by the user, but must be
smaller than the pixel size. Flow velocity in the
channel will generally be higher as a result of dif-
ferent hydraulic radius. Changing the channel
length therefore influences timing and shape of
the simulated hydrograph.
All these parameters were changed within rea-
sonable boundaries; that is, within boundaries
that could be argued to be realistic given the
available amount of data and its uncertainty.
Because of the limited number of storms, a split
in calibration and validation events was not pos-
sible. Instead, each event was calibrated separately.
This resulted in five different calibration sets.
Each calibration set was validated by applying it
to the other four events.
Peak discharge calibrations are most suited
to evaluate the performance of LISEM because
they use time to peak, peak discharge and the
shape of the hydrograph. The use of a goodness-
of-fit coefficient (e.g. Nash & Sutcliffe, 1970)
was less appropriate because these coefficients
are very sensitive to a time shift in runoff.
Therefore the fitting was done by eye, and the
Nash-Sutcliffe coefficient was only calculated
to compare the final calibrated versions of
LISEM 163 and LISEM LP.
Total runoff volume calibrations were done
because sediment loss is calculated as the prod-
uct of runoff volume and concentration. The total
runoff volume calibrations used the peak dis-
charge calibration as a starting point. For the total
runoff calibrations, only saturated conductivity
was changed. Where the fit between predicted
12.5.3
Calibration and validation
In this study, the prediction of both catchment soil
loss and spatial erosion patterns was evaluated. To
do this, a two-step approach was used. Firstly, the
LISEM model was calibrated on runoff and sedi-
ment yield measured at the catchment outlet.
Then, the simulation results of the calibrated
model were evaluated in a spatial way using field
observations on erosion patterns. Both the original
version of LISEM (LISEM 163) and the adapted ver-
sion (LISEM LP) were calibrated and the results of
both versions were compared with each other.
Model calibration had several objectives: to
simulate correctly the peak discharge, the total
discharge and the total soil loss. The LISEM model
was calibrated first on peak discharge (including
time to peak and hydrograph shape) to obtain the
correct shape of the hydrograph, and after that an
adjustment was made to obtain the correct total
discharge. Once the discharge prediction could
not be improved any more the model was cali-
brated on sediment yield. Several parameters were
used to calibrate on peak discharge:
