Environmental Engineering Reference
In-Depth Information
Clearly, over a large catchment such as the
Rhine, precipitation inputs in any flood event can
vary significantly across the basin. It is therefore
important to take account of the spatial distribu-
tion of inputs in flood forecasting. There is an
important issue, however, as to whether it is
necessary to use a fully distributed hydrological
model to do so, or whether a simpler model run at
the subcatchment scale might be sufficient. This
is the approach taken in many operational flood
forecasting systems, including the National
Weather Service in the USA (normally based on
subcatchment implementations of the conceptual
Sacramento hydrological model); SMHI in
Sweden, and the FEWS system for the Rhine in
The Netherlands (both based on subcatchment
implementations of the conceptual HBV model);
and theUKEnvironment Agency'sNational Flood
Forecasting System, which uses a variety of dif-
ferent models in different catchments (Whit-
field 2005; Werner and Whitfield 2007).
An important issue here is the utility of real-
time adaptation in flood forecasting. Even with
a well-calibrated model for a catchment or sub-
catchment, it should be expected that the predic-
tions for thenext bigfloodeventwill be (toa greater
or lesser extent) in error, even if only because of
uncertainties associated with the estimates of
catchment rainfalls (and theuncertainties are like-
ly to be worse for events with localized heavy rain,
or rain-on-snowevents). Thus, itmight be possible
to improve predictions of the timing and magni-
tude of a flood peak, by making use of data assim-
ilationinreal time (see, e.g.,Young2002).Thereare
a variety of techniques for real-time data assimila-
tion (see, e.g., Beven 2009) but the most important
data available in flood forecasting will be the cur-
rent water levelmeasurements in the hydrometric
network. These can be compared with model pre-
dicted water levels to see whether the model is
over- or under-predicting and make a correction
with updating every time a new measurement
becomes available at a site.
There is, however, only a limited amount
of information in a water level measurement at
a single site. It would not therefore be sensible to
try to use that information to attempt updating the
hypotheses about how the system functions, since
we expect that even the best process theory avail-
able might not represent fully the complexity of
the real catchment (see also Beven 2001a, 2001b,
2006a, 2010). Rainfall-runoff models, of course,
depend heavily on representations of subsurface
processes that are difficult to study experimental-
ly so it might be hoped that the problems might be
less for distributed hydraulic models. Unfortu-
nately, it seems that this might not be the case.
A number of studies that have attempted to
calibrate hydraulic models against historical
inundation data have found that while it is not
too difficult to reproduce observations of water
levels during a flood event at a gauging site, it
seems to be very difficult to predict patterns of
inundation correctly in the flow domain. In some
cases, this might be because the observations
of inundation are in error (e.g. Pappenberger
et al. 2005a), but even allowing for uncertainty in
such observations it has proven difficult to find
sets of effective roughness coefficients that give
good predictions of the patterns of inundation
everywhere on the flood plain (e.g. Romanowicz
and Beven 1998, 2003; Pappenberger et al. 2007a,
2007b).
Data Assimilation Issues in Using
Distributed Models
Increasingly, spatially distributed input data (from
telemetering raingauges, radar or quantitative pre-
cipitation forecasting) are available for use in flood
forecasting and flood warning (e.g. Pappenberger
et al. 2005b; Collier 2007). The European Flood
Alert System, for example, takes ensemble predic-
tions of precipitation inputs across Europe provid-
ed by the European Centre for Medium-range
Weather Forecasting (ECMWF) up to 10 days
ahead, and feeds these into the LISFLOOD FF
model, run at 5-km grid scale, to provide flood
alerts for all themajor rivers in Europe (e.g. De Roo
et al. 2003; Gouweleeuw et al. 2005; Thielen et al.
2009). The agencies responsible for individual ba-
sins might then run their own flood forecasting
systems at shorter timescales.
