Prediction of floods in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 198

small catchment. Due to the relationship between catch-
ment area and response time, increasing catchment area
also leads to a higher attenuation of the flood peaks. Rates
of runoff generation could also change with increasing
catchment area, due to changes in the dominant runoff
processes. For example, specific mean annual floods (i.e.,
flood peaks scaled by the catchment area) tend to decrease
with increasing catchment area (Eaton et al., 2002 )asitis
less likely that large areas will be fully covered by rain-
storms and also fully saturated (Viglione et al., 2010a , b ).
For all these reasons, catchment size is the main reason
why the specific runoff peaks of the large Ohio River
catchment (526 000 km 2 )( Figure 9.1 ) are so much lower
than those of the small Trisanna catchment (385 km 2 ).
The slopes of the two flood frequency curves in
Figure 9.1 are very different. The small Trisanna catch-
ment has a steeper flood frequency curve than the Ohio. As
discussed above, many authors have investigated the
nature of the scaling of the coefficient of variation of
floods with area (Smith, 1992 ; Gupta and Dawdy, 1995 ;
Blöschl and Sivapalan, 1997 ; Robinson and Sivapalan,
1997a ; Iacobellis et al., 2002 ) and have come out with
different interpretations. Catchment size is again the main
reason for the steeper slope of the Trisanna

and increased potential for soil development, which in turn
tends to dampen flood response. On the other hand, in the
Buwe region floods are produced by convective storms
with partial coverage of catchments and the soils are shal-
low, thus resulting in flash floods. Because of this, there is
high erosion, which reduces soil depth and increases the
efficiency of the drainage network, which in turn tends to
increase the flashiness of flood response. Figure 2.3 ( Chap-
ter 2 ) illustrates the processes involved. This example of
comparative hydrology illustrates that the patterns apparent
in the landscape are largely the result of the co-evolution of
climate, vegetation, soils and landform, and may give
predictive power to similarity indices such as drainage
density that goes beyond hydraulic relationships.

Event similarity
While the similarity measures discussed above relate to
hydrological characteristics of the catchment system as a
whole, in order to understand why two catchments are
similar in terms of their flood frequencies it may often be
very useful to break down the similarity into individual
events. These can then be used for regionalisation in
ungauged basins. One way of framing event similarity in
terms of the component process is by the derived flood
frequency framework, proposed by Eagleson ( 1972 ), later
generalised by Wood ( 1976 ), continued by Fiorentino and
Iacobellis ( 2001 ), and extended by Sivapalan et al.( 2005 ).
Each independent flood peak in the complete data series is
caused by an independent precipitation event. The relation-
ship between precipitation (characterised by intensity and
duration) and the resulting flood peak involves two kinds
of transformations, i.e., runoff generation and runoff
routing, and these are impacted by antecedent soil wetness.
Probability enters the picture by the fact that event charac-
teristics of precipitation (e.g., rainfall intensity and dur-
ation) are random, and so is antecedent soil wetness.
Derived distribution analysis can enable us to obtain ana-
lytically (in simple cases), or otherwise numerically, the
cumulative distribution function (cdf) of the population of
flood peaks that occur in a catchment. Extreme value
theory, on the basis of the number of flood events within
one year (or in the general case, the uneven distribution of
events within the year if they are seasonally dependent)
allows us to derive or estimate the extreme value distribu-
tion and thus the (annual) flood frequency curve (see
Sivapalan et al., 2005 ). The power of the derived flood
frequency approach is that, in a mechanistic way, it helps
to isolate the various contributions to the flood frequency
curve, and hence has considerable explanatory power to
analyse similarity of flood frequency behaviour between
catchments. There have been several efforts at developing
such a similarity framework, focusing separately on each
of the component processes: climate, catchment runoff and

s flood fre-
quency curve compared to that of the Ohio River. This is
because, in any particular year, the probability of an
extreme rainfall event hitting the small Trisanna catchment
is lower, resulting in high variability and a steep flood
frequency curve. In contrast, it is more likely that during
any particular year a part of the large Ohio catchment is
affected by an intense rainfall event.
Other catchment characteristics that contribute to simi-
larity and differences in flood frequency curves include
soil characteristics and land use. An important catchment-
related similarity measure is the fraction of the catchment
area that is urbanised. This is an indicator of surface flow
processes, which will increase the flood peaks. Geomor-
phological characteristics include drainage density and
topographic elevation. Figure 9.8 shows photographs and
flood hydrographs from the Gurk and Buwe regions in
Austria. The Gurk catchment has a dampened flood
response while the Buwe catchment is very flashy. The
catchment sizes do not differ much (432 and 184 km²) and
annual precipitation and elevation are also similar. How-
ever, they are different in their landforms. The landscape of
the Gurk is mountainous with flat valley bottoms and well-
rounded hills, while the landscape of the Buwe region
shows deeply incised channels. They are also different in
the flood-producing storm types and the geology. The
storm events in the Gurk are mainly synoptic, there is large
subsurface storage (highly permeable rock) and the surface
flow paths are tortuous ( Figure 9.8 ), thus resulting in
dampened floods. Because of this, there is little erosion

'