Operational European Flood Forecasting - Environmental Modelling: Finding Simplicity in Complexity - page 404

estimate of the uncertainty of the flood forecast (Pappen-
berger and Beven, 2006). There has been an increasing
awareness and acceptance that uncertainty is a fundamen-
tal issue of flood forecasting and needs to be dealt with at
the different spatial and temporal scales as well as at differ-
ent stages of the flood generating processes (Cloke et al .,
2009). Hydro-meteorological forecasting systems have
several components of which the meteorological forcing
is often seen to be the most uncertain one at lead times
beyond two to three days (a comprehensive review of this
topic is given by Cloke and Pappenberger, 2009). Addi-
tional sources of uncertainty include: the errors associated
with collection, transmission, processing and storing of
real-time data, corrections and downscaling procedure
of the meteorological predictions; antecedent conditions
of the system; methods of data assimilation for example
for soil moisture; geometry of the system (including
flood-defence structures); the possibility of infrastructure
failure (dyke failure, or blockage and subsequent backing
up of drains); characteristics of the model system (in
the form of model parameters); and limitations of the
hydrological model in fully representing processes (for
example surface- and subsurface-flow processes in flood
generation and routing). The importance of the individ-
ual components will vary in time and space, depending
on the dominant flow regime and the uniqueness of each
catchment (Beven 2002).

information derived from satellite data over the oceans
has represented a major step in improving the skill of
meteorological forecasts (e.g. Vidard et al ., 2009). This
approach is based on the incorporation of observed data
such as pressure and temperature in the generation of the
initial conditions for the weather forecast. One particular
development that has been a true milestone has been
the introduction of ensemble prediction systems (EPS),
which are multiple numerical weather forecasts issued
for the same forecast period (Buizza et al ., 1999; Buizza,
2002), as opposed to single 'deterministic' forecasts.
Ensemble prediction-system forecasts take account of
the nonlinear and chaotic behaviour of the atmosphere
as well as the uncertainties introduced by limits in spatial
and temporal resolution of meteorological observations.
Ensemble prediction systems are now well established in
operational weather forecasting (Park et al ., 2008).
This representation of forecast uncertainty makes EPS
forecasts an attractive product for flood forecasting sys-
tems which will always be sensitive to these uncertainties
(Cloke and Pappenberger, 2009; Cloke et al ., 2009). Sev-
eral operational weather centres around the world issue
such EPS meteorological forecasts, which are often based
on different methods to represent these uncertainties (for
a review see Park et al ., 2008). Not all NWP systems
have the same performance and some models are more
adapted to some climatic regions or particular weather
systems. The choice of NWP and EPS must therefore be
carefully examined for the application. For example, for
Europe, forecasts of the European Centre for Medium
Range Weather Forecasts (ECMWF), which were first
issued in 1993 (for a review see Palmer et al ., 2007) tend
to outperform the other systems (Park et al ., 2008).
Spatial patterns of meteorological forcing are difficult
to forecast: rainfall forecasts are still heavily dependent
on subgrid scale parameterizations of the rainfall form-
ing processes that cannot take complete account of our
understanding of the behaviour of rainfall cells in dif-
ferent meteorological conditions (Hewitson and Crane,
1996). Convective cells, in particular, cannot be resolved
at the current mesoscale grid sizes of 10-40 km used in
predictions for 5 to 10 days ahead. Thus rainfall fore-
casts are still limited by the resolution of the simulated
atmospheric dynamics and the sensitivity of the solu-
tions to the pattern of initial conditions and subgrid
parameterizations (Buizza et al ., 1999; Downtown and
Bell, 1988; Harrison et al ., 1999). Applying coarse res-
olution numerical weather predictions to regional- and
local-scale hydrological models necessitates some form of
'downscaling' (Xu 1999; Wilby et al . 2004; Fowler et al .

25.2.3 Spatial predictionofmeteorological
forcing

Numerical weather-prediction forecasts have been
used as meteorological forcing for the hydrological
models in order to extend forecast lead times in
atmospheric-to-river flood forecasting systems for a
long time (for example, Kitanidis and Bras, 1980).
Forecasts use coupled atmospheric, ocean and land
surface models to extrapolate current weather conditions
in order to predict the future state of the atmosphere
(Coiffier, 2011; see also Chapter 9). This is an extremely
computationally expensive modelling activity and
NWP forecasts are typically generated with the use of
supercomputers and parallel processing techniques (for
instance, www.ecmwf.int/services/computing/overview/
ibm_cluster.html). Key improvements to such systems
in recent decades have included developments in
physical parameterization of the coupled models as
well as improvements in techniques for data analysis
and data assimilation methods (Rabier, 2005; Lynch,
2006). In particular, the possibility of assimilating