Geology Reference
In-Depth Information
(a)
(b)
8
10
4
ยด
6
7
5
6
5
4
sim min-max
sim 5-95
sim 25-75
sim median
obs
sim min-max
sim 5-95
sim 25-75
sim median
obs
4
3
3
2
2
1
1
0
0
0
100
200 300
Timestep (min)
400
500
0
100
200
300
Timestep (min)
400
500
Fig. 5.4
Model performance for event 1 (17 November 2006) for (a) discharge and (b) sediment concentration.
The simulations comprise those sampled parameter sets that yielded results better than the discharge performance
threshold of
-
||
0.4, with each parameter set weighted by
-
||
to form the GLUE likelihood distributions of
simulations. Note the sedigraph is shown before conditioning on the sediment data, as the latter led us to reject
all parameter sets sampled based on the sediment performance threshold of
MAE
=
=
150.
5.4 Results
performance threshold might in fact appear too
relaxed by conventional expectations of model
performance, but errors in event sediment data
are likely to be high, and so this threshold tries to
reflect the uncertainty in the data with no direct
knowledge regarding the error characteristics.
Importantly within our extended GLUE uncer-
tainty framework, based on the evidence thus far
presented, all models have been rejected and so
further constraining of model simulations, as
described previously, is not possible. However,
for completeness, discussion and assessment of
the variation in model dynamics for the different
events, we present, independently, simulations
from the same set of sampled parameter sets for
the second storm below.
For the second, twin-peaked, event (Fig. 5.5),
the EUROSEM discharge output again does rela-
tively well at following the observed data, although
it drops below the observed interval in between
the consecutive peaks and underestimates the
final rise of the second peak, resulting in a mis-
match of shape between the observations and
Figure 5.4 shows the simulated and observed
hydrograph and sedigraph for event 1, where the
simulations comprise the sampled parameter sets
yielding results better than the discharge perform-
ance threshold of
-
||
0.4 (note, as mentioned
before, that the choice of this performance thresh-
old is arbitrary at this stage but, as will be seen
later, no better performance than
-
||
=
0.3 could
be achieved for event 1 and this puts our initial
threshold into perspective). For the discharge the
model response follows the observed interval
closely, apart from the initial
Q
and the final rise
at approximately 350 minutes. The model response
is also notably more 'spiky' than the observed one,
reacting more quickly to rainfall. The sediment
simulations, on the other hand, are, on average,
one order of magnitude higher than the observed
concentrations. Therefore, none of the parameter
sets were retained further when evaluated against
the
SS
performance threshold of
MAE
=
150, lead-
ing us to reject all the models. Note this
MAE
=
