Prediction of flow duration curves in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 156

some cases (e.g., low flows for sub-basin 82) is biased
relative to observed low flows. There also appears to be
a problem with over-prediction of runoff peaks, which
may be related to inadequate storage, infrequent exces-
sive surface runoff generation, or inadequate attention to
attenuation of flood peaks due to floodplain storage. The
use of FDCs has allowed these model deficiencies to be
identified, and has prompted further study towards
model improvement.
Even though the use of continuous models allows for a
more detailed description of the driving processes, it has
the consequence of significantly increasing model com-
plexity, possibly increased accuracy, but also lengthening
simulation times and increasing the number of parameters.
As a result, there is a danger that the individual role of
each climatic factor and hydrological process contributing
to the shape of the FDC cannot be isolated easily, leading
to limited transferability to other catchments/climates.
A range of approaches are therefore needed for introducing
process understanding into predictive models, ranging
from simple, targeted models to complex, distributed
models. In future, process-based methods should be
designed in such a way to benefit from and complement
the experiences arising from well-established statistical
methods, and in this way to help improve the performance
of statistical regionalisations as well (see e.g., Di Prinzio
et al., 2011 ).

7.5.1 Level 1 assessment
Table A7.1 lists 13 studies that deal with the estimation of
FDCs in ungauged basins. Some of the studies reported
performance measures that were not compatible with the
other studies and/or performed goodness of fit analysis
instead of cross-validation. The remaining 10 studies per-
formed leave-one-out cross-validation and the performance
measures were broadly similar. These were used in the
Level 1 assessment (indicated in Table A7.1 ). The number
of catchments evaluated in each study ranges from 8 to
1080, with a median of 49. There are several studies that
compare different hydrological models and/or regionalisa-
tion approaches, which give a total of 27 results for pre-
dictive performance. The regionalisation methods used are
index methods, regression approaches and estimation from
short records. The studies are quite heterogeneous in terms
of performance measures and the way they are applied.
Typically applied performance measures are the absolute
normalised error in the centre of the FDC; the proportion of
sites with Nash Sutcliffe (NSE) calculated over quantiles

lower than 0.75; the proportion of sites with absolute
normalised error (ANE) lower than 1; and the mean rela-
tive root mean square error. Even though these perform-
ance measures are not strictly speaking comparable, values
close to 0 imply good performances, and large values
imply a lower performance. Note that these all represent
errors (rather than skill), so they have been plotted down-
wards on the vertical axis to make them consistent with the
performance measures in the other chapters, i.e., higher up
in the plot is better. Different performance measures were
indicated by different symbols in the plots. For comparison
with the other runoff signatures in Chapter 12 , the NSE of
the quantiles of the FDC were back-calculated from the
percentage of sites with the NSE lower than 0.75 by
applying an empirical relationship from Level 2. The
25% and 75% quantiles of these NSE are 0.60 and 0.90,
respectively.
Figure 7.18 and Table A7.1 indicate that the studies
were performed in Europe, Asia, Australia and North
America. Most available studies were for humid and trop-
ical climates. Three main science questions are addressed
below.

-

7.5 Comparative assessment
The aim of the comparative assessment of FDC predic-
tions in ungauged basins is to learn from the similarities
and differences between catchments in different places,
and to interpret the differences in performance in terms of
the underlying climate

landscape controls. Understanding
these controls sheds light on the nature of catchments as
complex systems and provides guidance on what methods
to choose in a particular environment. The assessment is
performed at two levels (see Section 2.4.3 ). The Level 1
asses sment is a meta-analysis of studies reported in the
literature. The Level 2 assessment involves a more
focused and detailed analysis of individual basins from
selected studies of Level 1, in terms of how the perform-
ance depends on climate and catchment characteristics as
well as on the method chosen. In both Level 1 and Level
2 assessments, the performance was evaluated by leave-
one-out cross-validation, where each catchment was
treated as ungauged and the runoff predictions were then
compared to the observed runoff. The performances
obtained by the comparative assessment are estimates of
the total uncertainty of

-

How good are the predictions in different climates?
In Figure 7.19 it is important to compare similar perform-
ance measures. The absolute normalised errors (full
circles) in humid regions are smaller than those in the arid
regions. The proportion of sites with NSE lower than 0.75
(pluses) in humid regions show some scatter and the
majority of them are larger than that for the cold regions.
This means that from this limited comparison there is a
tendency for the regionalisation methods in humid regions

runoff predictions in these

ungauged basins.