Geoscience Reference
In-Depth Information
Figure 10.2
Simulation results for the Harrowdown Old Mill catchment (194 km
2
) treated as if ungauged with
observed and ensemble prediction streamflows (the grey range is the unconstrained ensemble; the white range
is the ensemble constrained by predicted hydrograph indices) (after Zhang et al., 2008, with kind permission
of the American Geophysical Union).
parameter sets that would satisfy multiple constraints. An example of their resulting ensemble of model
predictions for a catchment treated as ungauged is shown in Figure 10.2.
The advantages of this approach are that it does not depend on a particular model structure, nor on
the particular calibration methodology of the gauged catchments. It also focuses some attention on the
uncertainty in the hydrograph indices when they are used in constraining the model predictions. This
study might, however, have been too ambitious in its assumptions that single model structures for both
the regression equations and for the hydrological model could be applied everywhere. It might be more
interesting to see whether differences within a group of similar catchments, analogous to the pooling
group concept, could be distinguished in this way. Bulygina
et al.
(2009) have, in fact, used an analogous
approach to predict how afforestation and soil degradation might impact on hydrological indices and
model parameters for a gauged catchment.
It is worth noting that there have also been attempts to regionalise complete flow duration curves
for ungauged sites using different methods (Yu and Yang, 2000; Holmes
et al.
, 2002; Yu
et al.
, 2002;
Castellerin
et al.
, 2007; Li
et al.
, 2010). These might also be used as a constraint on model parame-
terisations, though the uncertainty in the discharge estimates should be taken into account in doing so
(e.g. Blazkova and Beven, 2009; Westerberg
et al.
, 2010b).
10.9 Comparing Regionalisation Methods for Model Parameters
With so many different methods available for the estimation of rainfall-runoff model parameters on
ungauged catchments, it is not surprising that there have been a number of studies comparing different
methods. These reveal, as noted above, that it is difficult to beat themethod of scaling themeasurements for
an upstreamor downstreamgauge, for those special cases where such observations are available (Merz and
Bl oschl, 2004). More generally, however, the results of such comparisons have been equivocal (Parajka
et al.
, 2005; Kay
et al.
, 2006b; Zhang and Chiew, 2009), with some suggestion that a combination of
methods might perform better than any individual method (e.g. Viviroli
et al.
, 2009). Oudin
et al.
(2008)
note that none of the methods compared on a sample of 913 catchments in France performed well relative
to full model calibration. Gotzinger and Bardossy (2007) report that the four methods in their study
using a distributed form of the HBV model failed to perform well in subcatchments with karst geology
