Operational European Flood Forecasting - Environmental Modelling: Finding Simplicity in Complexity - page 413

Date of this report : 2011010512
EUE : Number of members out of 51; COS : Number of members out of 16;

5 6 7 8 9 10 11 12 13 14

Forecast Day

DWD ↑↑↑ * ↓↓

*

↑↑↑ ↓ ↓↓↓↓

ECMWF

33 27 1

EUE > HAL

EUE > SAL

1

COS > HAL

Figure 25.8 EFAS threshold box diagram for
the forecast of 5 January 2011, 12:00 (From
EFAS).

COS > SAL

obtain the best forecast information (Nobert et al ., 2010;
Demeritt et al ., 2010). It is an illusion that one can
make reliable and accurate flood forecasts by relying only
on deterministic model predictions without considering
uncertainty (Pappenberger and Beven, 2006). Even with
increasing technology and knowledge, model uncertain-
ties will certainly remain, and ignoring them does not
result in their disappearance!
Making decisions from probabilistic flood forecasts
is not very simple. A single forecast provides an easy
yes-or-no answer whereas, probabilistic forecasts by their
nature shift responsibility towards the end user for the
interpretationof results for decisionmaking. For example,
what is the minimum probability value when it makes
sense to issue a warning for a severe flood event? Are these
probabilities thresholds the same for medium and severe
events? For end-users that are used to having forecasts
that predict an exact amount of flooding at a particular
point in time, how can they begin to use probabilistic
information instead?
Increased communication between the developers of
probabilistic systems and the end users, and more tar-
geted end-user training can help in identifying the correct
answer to these questions. End-users need to become
familiar with probabilistic forecast products. In partic-
ular, they need to understand exactly what probabilistic
forecasts are (and what they are not), and in what ways
they aremore useful than single, yes-or-no forecasts (such
as their better potential for early warning and capturing
uncertainty). Commonly used training approaches range
from lectures and games in an artificial setting to training
in realistic case studies and in situ training. For example,
using case studies of real floods for training purposes is
an effective tool and thus allows a realistic participatory
learning approach. In such case studies, participants have

25.4 Lessons and implications

The EFAS is currently one of the few (pre)operational
flood-warning systems worldwide making full use of
HEPS driven by EPS to increase the predictability of
floods. It has probably contributed to the acceleration in
adoption of a HEPS approach in national and regional
flood forecasting systems in Europe (Cloke et al ., 2009,
Table 25.3). The national flood-forecasting centres of
Sweden, Finland and theNetherlands, have already imple-
mented HEPS in their operational forecasting chain and,
for example, in France, Germany, the Czech Republic
and Hungary, hybrids or experimental chains have been
installed (Cloke et al ., 2009). The applicability of HEPS
for smaller river basins was tested in MAP D-Phase, an
acronym for 'Demonstration of Probabilistic Hydrolog-
ical and Atmospheric Simulation of flood Events in the
Alpine region' which was launched in 2005 as a Fore-
cast Demonstration Project of World Weather Research
Programme of WMO, and entered a preoperational and
(still active) testing phase in 2007 (Zappa et al ., 2008).
Examples outside Europe include the system of Hopson
andWebster (2008, 2010) who develop and run an opera-
tional ensemble flood forecasting system for Bangladesh.
Perhaps one reason for the slow transition from deter-
ministic to probabilistic modelling systems is the radically
different way of thinking, communication and decision
making required. Moving from deterministic to proba-
bilistic forecasting systems requires training of staff to
understand the sources of uncertainty and how they
propagate through a highly nonlinear system, ways of
visualizing multiple forecasts without losing focus, and
guidance on how to communicate this information to
different end-users and decision-makers in particular to