Prediction of floods in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 213

community, and are at the base of the derived flood fre-
quency framework discussed in Section 9.2.1 and of the
similarity theory of flood frequency discussed in Section
9.2.2 . Both design storm methods and

assumed to be uniform across the catchment, although for
larger catchments the design rainfall intensity estimated
from the IDF curve may be reduced using a so-called areal
reduction factor or ARF (e.g., Australian Rainfall and
Runoff, 1987 ). Stochastic rainfall models are now avail-
able to generate realistic spatial patterns that can mimic the
scaling of average rainfall intensity with increasing catch-
ment area (Menabde and Sivapalan, 2001 ; Burton et al.,
2008 ). However, to fully benefit from these, the rainfall

methods
that estimate the entire population of flood events are
essentially event-based methods with a major distinction:
in the first case the events are not actual observed ones, but
are designed in such a way that the mapping between storm
and flood return periods is known; in the second case the
method involves real (or close to real) events and the
mapping to the flood frequency curve is derived, either
analytically or numerically (in the frequency domain).

' scientific '

-
runoff models must be spatially distributed as well (see
Chapter 10 ).

Model structure The structure of the event-based

rainfall runoff model is often kept simple, particularly in

ungauged basins. The runoff generation in rainfall

-

Design storm methods
Design storm methods simulate floods for a discrete
'

runoff
models used for flood estimation in ungauged catchments
is often based on the concept of effective rainfall repre-
sented by empirical methods such as Phi-index and
W-index methods and SCS curve numbers (Australian
Rainfall and Runoff, 1987 ; SCS, 1985 ). The most common
approach adopted for runoff routing involves the use of the
unit hydrograph (Dooge, 1959 ). Although the unit hydro-
graph is based on strong assumptions, which are only
rough approximations of the flood runoff processes, the
use of unit hydrographs has proved to be a very useful
approximation, especially when large flood events are used
for calibration (Lamb, 2005 ). Other runoff routing methods
used in practice include storage routing models, based on a
set of linear or non-linear reservoirs or the kinematic wave
approach (Pilgrim and Cordery, 1993 ). Again, these
models often perform well when their parameters can be
calibrated against observed data.

-

storm and have a long tradition in engineering
hydrology (Pilgrim and Cordery, 1993 ; Viglione et al.,
2009b ). They usually consist of three parts: (i) a design
storm that is usually expressed in the form of a joint
probability of rainfall intensity and duration; (ii) a deter-
ministic rainfall

design '

runoff model that transforms the rainfall
input (i.e., design storm) into the flood peak - this model
usually contains a runoff generation sub-model or a com-
ponent to estimate effective rainfall, and a flow routing
component to convert the volume of runoff generated (or
rainfall excess) to the peak runoff or flood peak; and (iii) a
methodology or framework in which the two above com-
ponents are combined within a probabilistic framework,
and conditioned on regional flood data, to enable predic-
tion of the flood of a specified return period for an
ungauged basin. Each of these components is discussed
in turn.
Rainfall inputs to drive the runoff model can usually be
obtained from two types of sources: (i) a design rainfall
and (ii) a stochastic rainfall model. Design rainfall values
are published for many regions in the world, e.g., Australia
(Australian Rainfall and Runoff, 1987 ), Germany (DWA,
2012 ), USA (Chow et al., 1988 ; Bonnin et al., 2004 ), UK
(Houghton-Carr, 1999 ; Kjeldsen, 2007 ). They are usually
presented in the form of IDF curves (Svensson and Jones,
2010 ). Within the chosen storm duration, the rainfall inten-
sity can be assumed to be uniform in time, or to follow a
predefined (design) within-storm intensity pattern. Alter-
natively to assuming standard within-storm patterns, one
could generate a population of equally likely within-storm
patterns using stochastic rainfall models while maintaining
the storm duration and mean rainfall intensity (e.g., Acre-
man, 1990 ; Robinson and Sivapalan, 1997b ; Onof et al.,
2000 ; Haberlandt et al., 2008 ). This way, instead of produ-
cing a single design flood, one could generate an ensemble
of design storms, and hence flood peaks, using the rainfall

-

Estimating model parameters in ungauged basins
Regardless of the combination of rainfall runoff models

used, the biggest challenge and concern for their applica-
tion to ungauged catchments is the choice of appropriate
parameters: a parameter related to the runoff generation,
and a parameter related to runoff routing. There are numer-
ous recommendations regarding how to obtain model par-
ameters in ungauged catchments, but they all fall into one
of two categories: (i) transpositions from similar, gauged
catchments; and (ii) the use of empirical formulas based on
some form of regional analysis (e.g., Snyder, 1938 ;
Mockus, 1957 ; SCS, 1985 ; Akan, 1993 ; Pilgrim and Cor-
dery, 1993 ; USACE, 1994 ; ASCE, 1996 ; Houghton-Carr,
1999 ; Merz et al., 2006 ). Australian Rainfall and Runoff
( 1987 ) presents design values for infiltration capacity and
initial loss for several regions of the country as a function
of return period. One of the most widely used methods
around the world to estimate runoff generation in
ungauged catchments is the US SCS curve number method
(SCS, 1985 ; USACE, 1994 ). The method calculates

-

-
runoff models. In space, the rainfall intensity is usually