Environmental Engineering Reference
In-Depth Information
sembles such as the UK Met Office MOGREPS
(Bowler et al. 2008). The resulting ensemble of
outputs gives distributions for the predicted QPF,
which can then be used to run many hydrological
simulations, thus ultimately yielding a distribu-
tion of possible flow rates or river stages.
However, whilst this conceptually attractive
procedure sometimes gives reasonable results for
the likely distributions of the primary variables,
much of the variability tends to be confined to the
wind fields, and hence the QPF range is often
smaller than actually observed, and in particular
may not include the sort of extreme events likely
to cause floods. To address this issue there have
been attempts to consider a range of rainfall-pro-
ducing processes in the NWP models in an
attempt to capture the range of possible dominant
cloud andmicrophysical processes that are param-
eterized in the NWP model. This approach
is sometimes described as a 'physics ensemble'
technique and can be accomplished by either
using a suite of different NWPs or applying sto-
chastic perturbations to parameters within one
model's microphysics.
An alternative strategy is the so-called
'perturbation breeding' approach (Toth andKalnay
1997) where the input data are perturbed every-
where and the NWP model is run for some equiv-
alent time. Regions where large increases in the
rainfall occur are then identified and the error field
renormalized. The process is repeated to maxi-
mize the magnitude of the extreme event again.
The idea is to determine the most exteme event
that could actually happen from the modelled
physics of the system. It is possible that this idea
will make a useful contribution to extreme pre-
cipitation and flood frequency analysis.
radar reflectivity to adjust the humidity and tem-
perature fields and observed an improvement in
model skill, with the increases in moisture prop-
agating to change the system dynamics. Water
vapour has also been integrated based on micro-
wave radiometer observations, and resulted in
improved model predictions. A great deal of re-
search is currently underway in this area and it is
probably reasonable to expect that it will result in
significant improvements in QPF in the next
decade.
At the mesoscale, attempts to initialize NWP
models with nowcast rainfall and cloud patterns
are underway.
Ensembles
Since both the atmosphere and the hydrological
response of the catchment contain highly non-
linear processes it is not plausible that a single
deterministic representation of either QPF or hy-
drological output is likely to yield accurate results
on all occasions. Thus there has been a growing
interest in attempting to predict the distribution
of QPF rainfall amounts as well as the distribution
of possible hydrological responses. There are a
variety of ways of trying to achieve this, which
have only relatively recently become widely avail-
able because of the large demand such procedures
usually place on computer resources. As discussed
earlier, the initialization data for the NWPmodels
are usually temperature, pressure, wind and hu-
midity, historically derived primarily from radio-
sonde balloon ascents and more recently
supplemented by satellite data, which are assim-
ilated into the current cycle of theNWP process. A
frequently used strategy is to randomly perturb the
new input data by amounts comparable with the
likely measurement errors and to then re-run
the model. This process, known as Ensemble
Forecasting, can then be repeated as many times
as computer resources allow. Traditionally, oper-
ational ensembles have focused on synoptic scale
baroclinic instability, which involves models not
suitable for making predictions of rainfall on
catchment scales. More recently, increasing com-
puting power has seen the advent of regional en-
Meteorological scale and process problems
For the purpose of considering the role and type of
strategy required for effective flood warning, it is
useful to dividemeteorological situations into two
distinct types of events: those that are forced from
the large scale (e.g. fronts and cyclones) and those
triggered or forced from the small scale (e.g. air
mass thunderstorms and small-scale orographic
