Environmental Engineering Reference
In-Depth Information
range quantitative precipitation forecast ensem-
bles, but bring their own set of uncertainties. The
source andmagnitude of the uncertainty inNWPs
is often poorly understood. Forecasts from NWP
models are usually pre-processed to remove bias
and often increased to match raingauge observa-
tions. NWPmodels are discussed inmore detail in
later sections.
There is a role for both nowcasting and NWP
models in real-time flood forecasting, but it is
important to understand inwhich situations com-
bining the two is useful and improves the accuracy
and confidence in the hydrological forecasts pro-
duced. Figure 10.1 shows the errors in quantitative
rainfall forecast using various techniques and a
combined system. The combined system consis-
tently produced improved results. Where a hydro-
logical application such as a small urban
catchment requires a short lead time up to about
90 minutes, NWPmodels provide little value, and
radar and real-time raingauge data are often used
exclusively in these situations. By contrast, for
larger catchments, intense but short-lived rainfall
activity usually poses less concern than moder-
ately heavy but longer duration rainfall over the
entire region. This is particularly true where there
are concerns regarding the reliability of levees or
dam emergency spillways.
It makes sense to determine if a combined QPF
is going to improve the accuracy and usefulness of
a flood forecasting system before adopting this
logistically complex and computationally inten-
sive approach. Examining the response times of
the catchment(s) is a logical first step. Particularly
in mixed urban-rural settings, or where a large
river flows through an urban setting and flooding
may result from either overbank flow or insuffi-
cient capacity in the city's storm water drainage
system, there may be a real advantage in merging
several precipitation forecasting methods in a
combined QPF. This is true even if separate hy-
drological models are being used.
Flood forecasters must attempt to minimize
losses due to flooding while ensuring they do not
lose the public trust through unnecessary warn-
ings; knowledge of the uncertainty in the flood
forecast
uncertainty in the forecasts can be simulated
by perturbing the initial conditions of the NWP
models being used by amounts that represent the
uncertainty in the input data and running the
forecast many times. The hope is that the distri-
bution of the ensemble of modelled outputs gives
a measure of the uncertainty in the forecast. For
the purposes of this chapter, this approach will be
termed 'ensemble methods'.
Ensembles of rainfall input for rainfall-runoff
models generated by combining ensembles of rain-
fall extrapolated from remotely sensed data with
ensembles of output fromNWPmodels present an
interesting and useful tool to flood forecasters and
emergency responders. Instead of a deterministic
process for developing the 'most likely' flood peak
or emergency situation, the user now possesses a
probability distribution of future hydrological
conditions, which is much more useful for deci-
sion-making.
Hydrological Considerations
Despite the fact that lumped models often outper-
form distributed models, the latter are more often
used for real-time flood forecasting because of
their ability to deal more directly with spatially
heterogeneous catchment characteristics and in-
puts, common in extreme events. However, recent
studies have shown that careful parameterization
and calibration can improve distributed model
performance beyond what can be achieved used
a lumped approach. The size of the catchment, the
spatial resolution of the data available to describe
the catchment, and the resolution of rainfall data
used as input may dictate which modelling ap-
proach is most appropriate.
Quantitative Precipitation Forecasting from
projections of radar and/or satellite data (nowcast-
ing) has been used to estimate floods with greater
lead time, but the accuracy of the forecasts di-
minishes dramatically for forecasts beyond 90
minutes. Mesoscale NWP models (often a high-
resolution model nested within the grid of a lower
resolution model and centred over the catchment
of interest) have been used to provide better long-
is very important. This uncertainty
