Operational European Flood Forecasting - Environmental Modelling: Finding Simplicity in Complexity - page 402

Legend: Legend:

Severe
High
Medium
Low

High: 51

Low: 1

0 71 142 213 284 355 km 0 71 142 213 284 355 km

Figure 25.2 Example of an EFAS threshold exceedance maps for Romania based on river discharge predictions from the LISFLOOD
hydrological model driven by (a) single deterministic NWP forecasts, with four critical thresholds defined as those defined in
Table 25.1 and (b) EPS NWP forecasts, with the highlighted pixels referring to the number of EPS above the EFAS high alert defined
in Table 25.1 (right) (Reproduced with permission from Thielen et al ., 2009a Thielen J., Bartholmes J., Ramos M.-H and de Roo A.,
(2009a) The European Flood Alert System - Part 1: Concept and development. Hydrology and Earth System Sciences , 13, 125-40).

and simulated thresholds, allowing the estimation of the
severity of a flood event independently of the absolute
simulated values. This method compensates for any sys-
tematic overprediction or underprediction (bias) of the
model. Another advantage is that the method can be
applied in all areas where measurements are sparse or
nonexistent, like, for example in small basins (that are
often ungauged) and are at risk of flash floods (Alfieri
et al ., 2010) or in bigger basins but where measure-
ments are often not available for a number of different
reasons (Thiemig et al ., 2010). However, although this
approach may produce reasonable results in terms of
threshold exceedance, results can be significantly differ-
ent from the observed hydrograph. To ensure that EFAS
remains credible, such quantitative discrepancies need to
be identified and reduced over time, aided by updating
of river-discharge forecasts and model calibration and
uncertainty analysis (Younis et al ., 2008b; Thielen et al .,
2009a; Bogner and Kalas, 2008).
Although many flood-warning systems have thresh-
olds based on individual points or locations on rivers,
the spatial extent of the flood prediction is also of
key interest (Bates and De Roo 2000; Hunter et al .,
2007), which can be achieved by coupling a distributed
flood-inundation model to the hydrological model (Pap-
penberger et al ., 2005; Bartholmes and Todini, 2005).
However, it increases the computational burden and the
amount of high-resolution data needed dramatically, in
particular for exact topography, locations of buildings
and flood defences as well as detailed data for assessing

model performance (Schumann et al ., 2007). These needs
would be extremely challenging tomeet at the continental
scale, although future improvements in remote sensing
data resolution and supercomputing power may allow
this limitation to be overcome in the future.

25.2.2 Spatial calculationof riverdischarge

In any flood-forecasting system the techniques used to
predict river discharge depend largely on the anticipated
prediction lead time (how far in the future is being
forecast) and the catchment characteristics. Below, three
representative forecasting systems are discussed which
are commonly defined in flood forecasting, with the
later systems building on the techniques contained in the
earlier ones:

System 1 - river-to-river

River-to-river forecasting systems predict discharge using
only observed river discharge. The simplest form of this
method assumes that future discharge is the same as
the currently observed discharge. The method can be
taken further by extrapolating using methods such as
regression or time-series analysis and/or using upstream
river discharge observations to predict some downstream
river discharge (Romanowicz et al ., 2006). Thesemethods
are very efficient for short-term forecasts of a few hours,
for example when a flood wave takes seven hours to travel
from point A (observation point - for example, a trigger
gauging station) to point B (forecast point - such as a