Geoscience Reference
In-Depth Information
tion of individual weather events impossible.
Small errors in the initial conditions used to start
a model simulation invariably grow in magnitude
and spatial scale, and the entire globe will generally
be affected by a small observational error at a
single point before long. Therefore, long-term
weather prediction and climate prediction do not
try to predict individual weather events, since
these would certainly be in error. Instead they
generally try to represent the statistics of the
climate rather than the weather itself and are often
associated with probabilities based on statistical
relationships.
Like numerical forecasting at shorter
timescales, long-range (monthly and seasonal)
outlooks use a combination of dynamical and
statistical approaches in order to assess the
probability of certain weather situations. Long-
range forecasts rely on the idea that some types of
weather, despite being unpredictable in its details,
may, under certain circumstances, be more likely
than in others. One major recent advance in
long-range forecasts is the realization that El
Niño/Southern Oscillation has documented
statistical effects in many parts of the globe. For
any particular El Niño or La Niña it is generally
not realistic to forecast increased/decreased
precipitation at most points in the globe but many
regions show a statistical tendency towards more
or less precipitation or higher/lower temperatures
depending on the phase of ENSO. Long-range
forecasts make use of these statistical relation-
ships.
ENSO has a fairly regular periodicity allowing
for some skill in predicting changes in phase just
from climatology. Several dynamical models also
try to predict the future phase of ENSO, though
these have not been dramatically more sucessful
than a knowledge of the climatology. The phase of
ENSO is the single most important factor going
into long-range forecasts today.
The United States NCEP is again typical of the
methodology used globally. It currently issues
30-day and three-month seasonal forecasts up to
one year into the future. The primary information
used in these outlooks is the phase of ENSO,
recent and extended climate history, the pattern
of soil moisture which can affect temperature and
precipitation far into the future, and an ensemble
of 20 GCM model runs driven with predicted SSTs
from an AOGCM simulation over the period. This
information is used to produce a variety of indices
which predict the probability of three equally
likely categories of temperature (near normal,
above/below normal) and precipitation (near
average, above/below the median) (see
Figures 8.7
and
8.8
), together with tables for many cities.
Figure 8.8A
illustrates the observed height field
corresponding to
Figure 8.7A
for February 2007,
showing that the pattern is well represented on the
forecast chart.
Figure 8.8B
and C show that in this
case, as is usual, the temperature forecasts are
more reliable than those for precipitation.
A statistical techniqe called a canonical
correlation analysis uses all the above information
to produce long-range outlooks. Simulated
700mb heights, global SST patterns, US surface
temperature and precipitation for the past year
are all used to infer possible preferred patterns.
Temperature and precipitation history give
information about persistence and trends over the
year. ENSO is emphasized in this analysis but
other natural modes of variability such as the
North Atlantic Oscillation are also accounted for.
Secondary analyses which use single predictor
variables are also available and become more or
less useful than the correlation analysis under
differing circumstances. The composite analysis
estimates ENSO effects by defining whether a La
Niña, El Niño or neutral conditions are forecast
for the period of interest and then taking into
account whether there is confidence that this
one phase of ENSO will exist. Another index
predicts future tempertaure and precipitation
based on persistence over the past 10-15 years.
This measure emphasizes trends and long-term
regimes. A third secondary index is a constructed
analog forecast from soil moisture patterns.
