Environmental Engineering Reference
In-Depth Information
with the conservation of atmospheric mass, ther-
modynamic energy and momentum. In some
models, vertical momentum considerations are
simplified by making use of the hydrostatic ap-
proximation. More advanced non-hydrostatic
models solve this explicitly. The direct treatment
of vertical momentum turns out to be important
on small scales where convection and updrafts
need to be resolved by the model. At large scales
these processes can be included in so-called sub-
grid parameterizations. Since these processes are
the ones associated with clouds and precipitation
their treatment can be particularly important for
hydrological applications.
The high-resolution cloud and rain information
discussed in the previous section needs to be
assimilated into the NWP model, either as initial
data or by 'nudging' a continuously functioning
operationalmodel.Whilst the estimation of initial
precipitation and cloud fields from the NWPmod-
el is a relatively straightforward process, although
not a particularly accurate one, the introduction of
the cloud and rainfall information intoNWPmod-
els is not. Themodels carrywater as fields of cloud
liquid, frozen water and water vapour, and for
practical purposes treat rainfall as a diagnostic
rather than a prognostic parameter. Analysis
methods have yet to be fully established to adjust
the moisture parameters and convergence fields
such that the observed rain patterns are generated
by the model at the initialization time. This is a
non-trivial task.
Nowadays, 3D-VAR (three dimensional varia-
tional) (e.g. Lorenc et al. 2000) and 4D-VAR (e.g.
Rawlins et al. 2007) methods are considered to be
state of the art for the assimilation of observed data
into the initial conditions of NWPs. In the nD-
VAR assimilation systems, differences between
observations, the model and background fields
from a previous model run, and a new analysis
are minimized in a least squares sense, weighted
by estimates of observational and model error.
Observations might include radiosonde sound-
ings, measurements from ground stations, satel-
lite images or radar data. Such variationalmethods
have now largely superseded older methods such
as nudging and diabatic or physical initialization.
active research that will inevitably lead to im-
proved resolution and accuracy of global mapping
of cloud properties. Assimilation of this additional
information into NWP models, provided that it is
of sufficient quality, will improve predictions of
the NWP models, particularly for QPF.
Nowcasting
As indicated earlier, at the convective scale of a
kilometre or so the only practicalway to obtain the
precipitation pattern is by weather radar. The
extrapolation of radar or satellite remote-sensed
rainfall and cloud patterns to make QPF predic-
tions a few hours ahead - nowcasting (Browning
1982) - has been used operationally for years,
particularly for thunderstorm hazards at airports
but also for urban and flash flooding. Various
attempts have been made to combine nowcasting
with NWP to take advantage of the superior pre-
dictive capability of nowcasting for the first couple
of hours while retaining the guidance afforded by
the NWP on the longer timescale. The most suc-
cessful early attempt was the UKMetO Frontiers
System (Browning 1982). More recently the Aus-
tralian Bureau of Meteorology System, STEPS, has
demonstrated some success (Bowler et al. 2006). In
the latter system the spatial resolution of the
output is reduced the further ahead is the forecast
time, recognizing the impossibility of predicting
small-scale features far into the future. Research-
ers are actively involved in looking for ways to
statistically downscale the low-resolution fore-
casts to give an ensemble of QPF outcomes for
hydrological models.
Numerical weather prediction (NWP)
Numerical Weather Prediction (NWP) can be con-
sidered a boundary and initial value problem - in
order to make forecasts the starting point needs to
be specified. Various NWPs can then be used to
propagate the initial conditions (analysis) forward
in time to generate a forecast.
The models are all similar in that they solve
formulations of dynamic equations associated
Search WWH ::




Custom Search