Geoscience Reference
In-Depth Information
making a compromise between the model-calculated fields defined by the GCM at
the time of the initiation (which are, of course, consistent with the equations used
in the GCM), and any observations available at that time. The values of the weather
variables specified in the initial state of the GCM thus become a weighted average
of that predicted by the GCM based on an earlier initiation and those currently
observed. Plausible values are assumed for errors in the observational data and
model calculated values, and these are used to define the weighted average initial
fields. If the observed values are substantially different to model calculated values
they are rejected as implausible and not used in the initiation.
The GCM is then run forward in time for some days ahead and in this way
numerical predictions of actual weather made for the future. In practice, there are
always shortcomings in the description of the atmosphere represented in the GCM
not least because the representation is made at grid scale using approximate
equations that
parameterize
processes that occur at much smaller scale.
Consequently, the accuracy of weather predictions degrades with time ahead.
Currently GCMs used for weather prediction can make reasonable predictions for
a few days ahead, and some weather centers attempt forward look predictions for
as long as 8-10 days ahead. Thus,
when used for NWP, GCMs extrapolate observa-
tions forward in time using a physically realistic representation of atmospheric
circulation to predict the actual future weather for periods of several days ahead.
Some time ago it was realized that the process of NWP could provide an
important byproduct. The six hourly initiation process using data assimilation
calculates values of the meteorological variables used in the GCM everywhere in
the atmosphere. These values are in significant part based on model predictions,
but they are at least in part influenced by observations. Globally available fields of
weather variables cannot be routinely obtained at regular six hourly intervals in
any other way. But they are needed, and for this reason model-calculated data
based on the assimilation of observations into GCMs have now become widely
used as a source of data in their own right. Thus, the initiation fields of GCMs used
for NWPs are often stored and can be made available as a data source which has
the advantage of being broadly consistent with the laws of atmospheric physics (to
the extent these are represented in the GCM), while also reflecting relevant
observations available at the time of initiation to the extent allowed by the data
assimilation process.
One feature of such model-calculated data when provided as time series is that
the amount and nature of observational data that influence the data product
through data assimilation can change with time. Consequently, the relative
influence of observations versus model in the data also changes with time. Over
the years weather forecast centers have sought to use more and more observations
to better define the initial states used for their forecasts, and the amount of remote
sensing data so used has also greatly increased. As a result more recent
model-calculated data are arguably a better reflection of reality. An important
shortcoming of model-calculated data arises because weather forecast centers are
always striving to improve the realism with which atmospheric processes are
represented in their model. Consequently, improved versions of the GCMs used
