Environmental Engineering Reference
In-Depth Information
results in improvement tomodel fields (Lopez and
Bauer 2007; (Xiao and Sun 2007); (Rihan et
al. 2008); (Pu, Li and Sun 2009); (Xiao et al. 2009).
Alternatively, moisture observations can be
treated without VAR to generate pseudo-observa-
tions. The UK Met Office runs the Moisture
Observation Pre-processing Syste (MOPS) opera-
tionally, which treats moisture observations be-
fore ingestion. The system combines IR satellite,
rain radar and surface observations with the
background model state to generate a distribution
of humidity pseudo-observations (MacPherson
et al. 1996) that can then be assimilated more
readily against model humidity parameters. Radar
observations are included in operational model
runs via the Latent Heat Nudging (LHN) method
(Bell 2009) (Dixon et al. 2009) described by Jones
and MacPherson (1996). LHN involves modifying
the latent heat in the model by the ratio of model
to radar rain rate.
Regardless of the schemeused to assimilate rain
and cloud observations into a model as initial
conditions, the predictive power of this system
will need to be tested. Comparison runs of the
model with and without cloud and rain assimila-
tion need to be performed. In this way the impact
of changes to the initial conditions on the forecasts
can thus be assessed by comparing the model
outputs to observations. Further tests are then
undertaken to characterize the sensitivity of the
model to perturbations in the new initial condi-
tions. This could be achieved by performing runs
with subtly different rain/cloud schemes or by
perturbing other model fields (Hohenegger and
Schar 2007).
Preliminary results suggest that if the initiali-
zation is attempted by simply increasing the hu-
midity field the extra moisture 'rains out'
relatively quickly, leading to only short-term ef-
fects. If, however, the rainfall and cloud moisture
fields are introduced by increasing the updrafts in
the model, resulting in increased surface conver-
gence and divergence aloft, then the model can
move to a higher energy and greater rainfall state.
There has been some success reported with the
addition of water vapour information to the initial
conditions. Benedetti et al.
Some meteorological observations can be com-
pared almost directly to prognostic model vari-
ables. Wind, for example, is treated explicitly in
NWP models, so a comparison between model
wind and anemometer readings requires only a
simple observational operator to account for
interpolation from model space to a point
measurement.
The treatment of precipitation-related mea-
surements is somewhat more troublesome, as
more complicated models are required to compare
observations and prognostic model variables. Spe-
cial caremust be takenwith the conditional on/off
thresholds common to precipitation microphysi-
cal models, which can be difficult to represent as
linear approximations.
Assimilation of large fields of observational
data in the presence of clouds and precipitation
data via VAR is a complex process. Inferring de-
tails of clouds and precipitation from radiance
measurements is complicated because the exact
propagation of radiation is highly dependent on
the spatial distribution and micro-physical prop-
erties of the hydrometeors. The large number of
parameters contributing to the measurement re-
sults in a poorly constrained problem, with errors
that are difficult to determine. Often rain or cloud-
affected satellite radiance data are excluded from
operational assimilation schemes for this reason
(Errico, Bauer and Mahfouf 2007), although in the
long term this is not a tenable position to take
given the high percentage of the globe that is
covered by clouds at any time. There is much
research currently underway in this area.
Rain radar data are also problematic to assim-
ilate due to the inherent discontinuity in forward
modelling from precipitation (or more correctly
reflectivity) to humidity and temperature. Again
significant difficulties are associated with the
division between precipitation/no precipitation
situations. This discontinuity can result in bimod-
al probability distributions (Errico et al. 2000),
which are difficult to deal with in minimization
schemes.
Nonetheless, case studies at various forecast
centres have found that variational assimilation
of radar reflectivity and/or radial velocity data
(2005) assimilated
