Operational European Flood Forecasting - Environmental Modelling: Finding Simplicity in Complexity - page 410

Previous flood
forecasts

EFAS - schematic view

Previous flood
forecasts

Previous flood
forecasts

Real-time Weather Forecasts:
ECMWF DET (1) & EPS (51)
COSMO-LM (1) & GM (1) & LEPS (16,
1x day)

2

LISFLOOD 1-6-24 h
(EPS)

Flood
forecasts 5

Real-time Observed Meteo/Hydro Data:
EU-FLOOD-GIS/ETN-R station data
(1300/500 stations across Europe)

Initial conditions

Post-
processing

1

LISFLOOD
1-6-24 h Persistent

4

Static European Datasets:
-topography
-landuse
-river channel dimensions
-geology

yes

3

Q>Thresh

Q-Thresholds LISFLOOD daily

Historic observed Meteo data
Ca 2000 station data from 1990 onwards)

Real-time processing, 2x a day
Real-time processing, after decision
Offline processing
External alerts

Figure 25.5 Schematic view of the EFAS system in 2011 (From EFAS).

the time of last observation and start up time of the flood
forecast is filled with the most recent weather forecast
data. In order to obtain a measure on the uncertainty in
the initial conditions, the different deterministic weather
forecasts are used to calculate the initial conditions at
the start up of the forecasts. The flood forecast itself is
calculated in a real-time mode running twice a day with
the latest weather forecast issued at 12:00 1 and 00:00.
Once all flood forecasts have been calculated, the
analysis module processes all results and compares the
simulated discharges against the critical thresholds. These
are calculated with LISFLOOD in an offline process, using
long-term historic meteorological station data. This cal-
culation is repeated every time the calibrationof themodel
changes in order to keep model consistent setup and
thresholds. At those points where both historic and real-
time data are available, post-processing routines (Bogner
and Kalas, 2008) are applied and visualized. This method
has the advantage that the modelled EFAS outputs are fit-
ted to observed discharges and thus can be incorporated
directly by the data providers into their own visualiza-
tion tools. Should the forecasts exceed a threshold and
meet certain criteria (for example, more than five clus-
tered river pixels exceed the thresholds, or the upstream
area is larger than 4000 km 2 ) then these are visualized
graphically as threshold exceedance maps or time series

of threshold exceedances. All EFAS results are presented
as up-to-date information online in a web interface, a
password-secured website that allows remote access for
the EFAS partners (such as the national hydrological ser-
vices) at any time. If the flood forecasts show a newly
developed flood signal, the forecast team is put on alert.
If the next forecasts also show the signal as persistent then
EFAS enters into an active alert. In this case, if a dedicated
EFAS partner institution exists for the river basin of con-
cern, EFAS forecasters send alert emails to the partners,
which then monitor the event on the EFAS-IS webpage
(Thielen et al ., 2009a).
It is of particular importance that all EFAS results are
clear, concise and unambiguous. This requirement does
not leave a lot of room for textual information, because
English, the language adopted in EFAS reports, is not
necessarily the native language of the majority of national
forecasters. Hence, EFAS visualization and decision-
support products have to be understood intuitively, in
order to best help EFAS users in making decisions on
flood warning (Ramos et al ., 2007, Thielen et al ., 2009a).
So EFAS uses two fundamental ways to display proba-
bilistic flood forecasts (Ramos et al ., 2007; Cloke et al .,
2009), which have been carefully tested with the end-users
of the forecasts: (i) spatial overviews in the form of maps
and (ii) time-series evolution at points including tables
of warnings. Visualizing probabilistic results effectively
demands a strategy involving combinations of colours,

1 COSMO-LEPS is available only once a day for the 12:00 forecasts.