Geoscience Reference
In-Depth Information
their methodology is similar and owing to
increasing computing power they are becoming
less distinguishable as separate types of forecast.
winds and hence no convergence or divergence.
The movement of systems could be predicted,
but not changes in intensity. Despite the great
simplifications involved in the barotropic model,
it has been used for forecasting 500mb contour
patterns. The latest techniques employ multi-level
baroclinic models and include frictional and
other effects; hence the basic mechanisms of
cyclogenesis are provided for. It is noteworthy
that
fields
of continuous variables, such as
pressure, wind and temperature, are handled and
that fronts are regarded as secondary, derived
features. The vast increase in the number of calcu-
lations that these models perform necessitated a
new generation of supercomputers to allow the
preparation of forecast maps to keep sufficiently
ahead of the weather changes!
Forecast practices in the major national
weather prediction centers around the globe are
basically similar. As an example of the operational
use of weather forcasting models we discuss
the methods and procedures of the National
Centers for Environmental Prediction (NCEP) in
Washington, DC, established in 1995. NCEP
currently runs a global spectral model opera-
tionally. The Global Forecast System (GFS) model
(formerly known as the AVN/MRF for aviation/
medium-range forecast) model has a spectral
truncation of T170 (approximately 0.7
1 Short- and medium-range
forecasting
During the first half of the twentieth century,
short-range forecasts were based on synoptic
principles, empirical rules and extrapolation of
pressure changes. The Bjerknes model of cyclone
development for mid-latitudes and simple
concepts of tropical weather (see Chapter 9)
served as the basic tools of the forecaster. The
relationship between the development of surface
lows and highs and the upper-air circulation was
worked out during the 1940s and 1950s by C-G.
Rossby, R. C. Sutcliffe and others, providing the
theoretical basis of synoptic forecasting. In this
way, the position and intensities of low and high
pressure cells and frontal systems were predicted.
Since 1955 in the United States - and 1965 in
the United Kingdom - routine forecasts have been
based on numerical models. These predict the
evolution of physical processes in the atmosphere
by determinations of the conservation of mass,
energy and momentum. The basic principle is that
the rise or fall of surface pressure is related to mass
convergence or divergence, respectively, in the
overlying air column. This prediction method was
first proposed by L. F. Richardson, who, in 1922,
made a laborious test calculation that gave very
unsatisfactory results. The major reason for this
lack of success was that the net convergence or
divergence in an air column is a small residual
term compared with the large values of conver-
gence and divergence at different levels in the
atmosphere (see
Figure 6.7
). Small errors arising
from observational limitations may therefore
have a considerable effect on the correctness of
the analysis.
Numerical weather prediction (NWP)
methods developed in the 1950s use a less direct
approach. The first developments assumed a one-
level barotropic atmosphere with geostrophic
0.7-
degree grid), 42 unequally spaced vertical levels
out to seven days. The truncation is increased to
T62 with 28 levels out to 15 days. It should be
noted that typically the computer time required
decreases several-fold when the grid spacing is
doubled. In order to produce a forecast, an
analysis of currently observed weather conditions
must first be generated as an initial condition for
the model. Very sophisticated data assimilation
algorithms take a large amount of observational
data from a variety of platforms (surface stations,
rawinsondes, ship, aircraft, satellite) which are
often measured at irregular intervals in both
space and time and merge it into a single coherent
picture of current atmospheric conditions on
standard pressure levels and at regular grid
×
