Operational European Flood Forecasting - Environmental Modelling: Finding Simplicity in Complexity - page 406

the severe threshold, thus the frequency of the event
occurring is 2/51 or roughly 4%. This frequency is often
termed the 'probability' of the flood event, although this
expression is not correct from a mathematical point of
view because such ensemble forecasts are often neither
highly reliable nor precise. For example, in a perfect
model an event forecast with a 25% probability should
be observed one out of four times, whereas in three cases
flooding should not take place. However, such reliability
is reduced by bias in the model (a tendency to overpredict
or underpredict discharge values because of the model
equations used), and precision reduced by uncertainties
in input data, forecasting model parameters and model
uncertainty. It is therefore not recommended to refer to
this as 'probability' unless the forecasts have been properly
bias corrected. Past forecasts (hindcasting) can be used
to increase reliability and precision, through calibration
of parameters, improvements in models and estimation
of systematic errors in the output (see Bogner and Kalas,
2008). This approach tends to improve flood forecasts
significantly.
The value of hydrological forecasts (discharge, water
stage, soil moisture etc.) based on ensemble predic-
tions can be evaluated with various metrics (Jolliffe
and Stephenson, 2003). However, long time series or
a large number of events are usually required to apply
these metrics in a meaningful way, which makes the
evaluation of the quality of flood forecasts difficult as
flood events are rare and thus it is difficult to estab-
lish reliable statistics. Even if there were enough data
from different flood events at different locations, it does
not account for non-stationarity, which means that one
catchment is often very different from the next one
(Beven, 2000) and for example, the form of a river bed
often changes dramatically after flood events thus mak-
ing consecutive evaluations incomparable. Thus, there
may be no other option than to analyze the performance
of EPS-driven flood forecasts on a case-by-case basis
(Pappenberger et al ., 2008). The analysis of such case
studies (see Cloke and Pappenberger, 2009) shows that,
in general, literature agreement is that EPS flood fore-
casting is a useful activity and has the potential to inform
early flood warning.
Table 25.2 provides a summary of the substantial setup,
activities, data sets and information needed for an oper-
ational flood forecasting system at the European scale. In
order to start up with correct initial conditions and in
order to performmeaningful discharge post-processing, a
considerable amount of observational data need to be col-
lected in real-time. These data need to bemade available at

a high frequency and be available not only in real time but
also for historic time series. Any operational forecasting
system also has substantial technical requirements.

25.3 The European Flood Alert System
(EFAS)

In response to the particularly severe floods of 2002, the
European Commission commissioned the development
of the European Flood Alert System (EFAS) in order
to reduce the impact of transnational floods through
early warning (European Commission, 2002). The EFAS
has two complementary objectives: first, complement-
ing member states' activities on flood preparedness, for
example by providing national hydrological services with
early flood information in addition to their own local
and, mostly often, shorter range forecasting information
and, second, providing the European Commission with
an overview of ongoing and expected floods in Europe,
and thus an early-warning flood forecasting system that
can be useful for crisis management in the case of large
transnational flood events that might need intervention
on an international level (Thielen et al ., 2009a). The
EFAS project followed on from the successful research
project, the European Flood Forecasting System (EFFS,
1999-2003) (Gouweleeuw et al ., 2005). The EFAS has
been running in a preoperational mode since 2007 and
currently produces medium-range flood alerts for the
whole of Europe on a 5 km grid for 27 national hydro-
logical partner organizations and the Monitoring and
Information Centre (MIC) of the European Commis-
sion, one of the instruments of the EU Civil Protection
Mechanism.

25.3.1 TheLISFLOODhydrologicalmodel: spatial
modellingoffloods
The hydrological model that is at the core of the Euro-
pean Flood Alert system is called LISFLOOD. LISFLOOD
is a hybrid between a conceptual and a physically based
rainfall-runoff-routing model, which is capable of sim-
ulating the hydrological processes that occur in large
and transnational river catchments. Using the PCRas-
ter Dynamic Modelling Language the model works in
a GIS environment, which has the great advantage of
a flexible and easy adaptable model structure (Thielen
et al ., 2009a; van der Knijff et al ., 2010). LISFLOOD was
designed to make the best possible use of spatial databases
that contain pan-European information on topography,