Geoscience Reference
In-Depth Information
intervals. The model equations are then integrated
into the future from this starting point.
The GFS currently runs out 17 simulations
which are identical except for very small differ-
ences in initial conditions four times a day. The
repetition of numerical forecasts incorporating
minor differences in the initial conditions allows
the effects of uncertainties in the observations,
inaccuracies in the model formulations, and 'the
chaotic' nature of atmospheric behavior to be
accounted for in terms of probabilities. Errors in
numerical forecasts arise from several sources.
One of the most serious is the limited accuracy
of the initial analyses due to data deficiencies.
Coverage over the oceans is sparse, and only a
quarter of the possible ship reports may be
received within 12 hours; even over land more
than one-third of the synoptic reports may be
delayed beyond six hours. However, satellite-
derived information and instrumentation on
commercial aircraft fill gaps in the upper-air
observations. Another limitation is imposed by
the horizontal and vertical resolution of the
models and the need to parameterize subgrid
processes such as cumulus convection. The small-
scale nature of the turbulent motion of the
atmosphere means that some weather phenomena
are basically unpredictable, for example, the
specific locations of shower cells in an unstable air
mass. Greater precision than the 'showers and
bright periods' or 'scattered showers' is impossible
for next-day forecasts. The procedure for
preparing a forecast is becoming much less
subjective, although in complex weather situa-
tions the skill of the experienced forecaster still
makes the technique almost as much an art as a
science. Detailed regional or local predictions can
only be made within the framework of the general
forecast situation for the country and demand
thorough knowledge of possible topographic or
other local effects by the forecaster. The average
of these ensembles is used for the short-term
forecast. The primary analysis products issued
every six hours are MSL pressure, temperature
and relative humidity at 850mb and 700mb,
respectively, wind velocity at 300mb, 1000-500mb
thickness, and 500mb vorticity.
NCEP also computes medium-range ensemble
forecasts from the 17 ensemble runs performed
at each interval. For example, the probability that
the 24-hour precipitation amount some days in
the future will exceed a certain threshold can be
computed by counting the number of model runs
where the value is exceeded in a certain grid box.
This is a rough estimate of the probability because
17 simulations cannot span all possible weather
scenarios, given the uncertainty in initial condi-
tions and model formulation. Current forecasts
are given as a 6-10-day outlook and 8-14-day
outlook of the departure of temperature and
precipitation from normal.
In order to calculate forecasts with more
regional detail, NCEP utilizes a limited area 'eta'
model which makes up to 84-hour forecasts over
North America only. Like all operational weather
models the eta is in a continual cycle of im-
provement and redesign. At present, however,
the eta model has a 12km grid spacing and 60
vertical layers. A specialized vertical coordinate is
employed in order to handle the sharp changes
in topography which a high-resolution model
encounters. Eta has a similar suite of output
variables as the GFS.
Because a typical weather forecast, even in the
highest resolution regional models, is meant to
depict an average over a large grid box, the actual
conditions at any single point within that grid
box will not generally be accurately predicted.
Forecasters have always subjectively applied
model information to forecasts at a single point
using their own experience as to how accurate
model information has been in the past under
certain circumstances (i.e., a subjective assessment
of model bias). An effort to make such localized
use of information more objective is called model
output statistics (MOS) and actual weather
conditions at specific weather stations are now
commonly predicted using this technique.
