Forecast Descriptions
This page describes the methods behind the production of the dynamical seasonal
climate forecasts. Additional information pages describe how the output from
the dynamical model is manipulated for more refined prediction information, including
a description of the Rebuilt probabilistic model predictions, and the bias corrected
ensemble probability model predictions. An additional tutorial webpage is also available
that gives a more general explanation of the "why"s and "how"s of seasonal climate
prediction.
Updated 13 April, 2001
Dynamical Seasonal Climate Forecasts -
The dynamical climate predictions are currently made with 4 different
atmospheric general circulation models: CCM3.2, ECHAM3.6, NCEP-MRF9, and
NSIPP.
Each month, each of these models is run 10 times (forming an ensemble) with
one or both of
two possible scenarios for the global sea surface temperature (SST), one
to three seasons (three to nine months) into the future.
In each set of ensemble runs for a given SST scenario, the only
difference between the individual runs is the initial atmospheric state
(the current weather) at the beginning of the run.
On the web pages for the various dynamical prediction maps/tools a
table is presented in which each entry represents a particular model
forced by a particular SST scenario. For example, "ECHAM(psst)" refers
to the ECHAM model (described below) forced with the PSST (persisted SST
described below) scenario.
Predicted climate anomalies, such as anomalies of temperature or precipitation,
are calculated as the departure of the model predictions from average conditions
for that season. The average conditions
are computed for the climatological base period of 1969-1998 (Note: The climatological
base period was updated in early 2001. This is close to the 1971-2000 climatology
period used by IRI's Monitoring Group and many National Meteorological Services,
but still incorporating the higher quality, higher resolution observed data
available from the Climate Research Unit of the University of East Anglia). The
climate anomalies for each model are taken with respect to each model's
average climatology.
The IRI climate prediction system is part of an ongoing experimental
research effort. The atmospheric models and the methods for producing
SST scenarios (SST forecasts) have changed during the years since IRI began producing
seasonal climate predictions in 1994 (predictions since 1997 are available
on our web site) and will continue to evolve.
The current approach is described below. There is also a link
to a page that describes previous approaches to forecasting SSTs, as these older
forecasts of SST, and of the climate anomalies resulting from them, are available
on our web site.
Atmospheric General Circulation Models (AGCMs) -
CCM3.2 -
This model was developed at the National Centers for Atmospheric
Research (NCAR) in the United States. It is freely available to the community.
The resolution of the model is T42 (approximately 2.8x2.8 degrees
horizontal resolution) with 18 vertical layers.
Initial atmospheric conditions are supplied by
restart files from an integration in which CCM3 has been forced with
observed SSTA for many years up through the forecast start date.
At the beginning of the forecast for the first ensemble member,
nine sets of restart files are generated, each for a successive
model day, to yield nine additional forecast initial conditions.
ECHAM3.6 -
The European Community - HAMburg (ECHAM) model was developed at
the Max Plank Institut fur Meteorolgie in Germany.
The resolution of the model is T42 with 19 vertical layers.
(Barnett et al, 1994, Tellus, 46A, 381-397;
Bengtsson et al, 1993, Science, 261,1026-29).
Initial atmospheric conditions for each forecast ensemble member are supplied by
restart files from separate ensemble members of a simulation in which ECHAM has
been forced with observed SSTA up through the forecast start date.
NCEP-MRF9 -
The NCEP-MRF9 atmospheric climate model was developed at the National
Centers for Environmental Prediction (NCEP) in the United States,
based on a version of the
medium range forecast model used by the National Weather Service.
The resolution of the model is T40 with 18 vertical layers.
Initial atmospheric conditions are derived as with CCM3 from
restart files from an integration in which NCEP has been forced with
observed SSTA for many years up through the forecast start date.
At the beginning of the forecast for the first ensemble member,
nine sets of restart files are generated, each for a successive
model day, to yield nine additional forecast initial atmospheric conditions.
NSIPP -
This model was developed at the NASA's Seasonal to Interannual
Prediction Project (NSIPP) at Goddard Space Flight Center.
The resolution of the model is 2.5 degrees longitude by 2.5 degrees
latitude with 34 vertical layers. Each forecast consists of a 9 member
ensemble.
Initial atmospheric conditions for the nine ensemble members are supplied
by restart files from nine
integrations in which NSIPP-1 has been forced with observed SSTs for many
years up to the start date.
More details on the model and its performance can be obtained at:
http://nsipp.gsfc.nasa.gov/atmos/atmosdescrip.html
Sea Surface Temperature Scenarios (SST Forecasts) -
The global SST anomalies are predicted as unchanging from
the latest observed field.
The prescribed SST field is thus persisted (held fixed).
The observed SST anomalies are taken from the month previous to when
the forecast was made and added to the climatological monthly
average SSTs to obtain the total SST values that are used as
boundary conditions to force the AGCM. For example, for the forcasts made in
August, the observed SST anomalies are taken from July, and the
forecast season is Sep-Oct-Nov. Dynamical predictions using
persisted SST forcing are run only 1 season into the future.
The predicted SST field contains forecast SST anomalies for
the tropical oceans and damped-persisted observed SST anomalies
for the mid-latitude oceans. At this time, the SST predictions
are made separately for each of the tropical ocean basins.
In the tropical Pacific Ocean (15S-15N) the forecast SSTs are produced
at NCEP using their coupled climate model, CMP12 (Pacific basin ocean/global
atmosphere), where the ocean has been initialized
with assimilated observed ocean temperature data as measured by the TAO buoy array.
For the tropical Atlantic Ocean (18S-30N) SST anomalies are
forecast at CPTEC (Brazil) using the statistical method of CCA (canonical correlation analysis).
The predictors for the tropical Atlantic CCA prediction of SST is
the recent observed SST anomalies in the tropical Pacific and Atlantic
Oceans.
The Indian Ocean SST anomalies (15S-15N) are also forecast using a CCA
technique, which is run at the IRI. The predictors for the tropical
Indian Ocean SST are the recently observed SST anomalies in the
tropical Pacific and Indian Oceans and also the NCEP forecasts for
the tropical Pacific.
The tropical oceans where forecast SST anomalies
are specified are smoothed into the mid-latitude observed SST
anomalies over approximately 8 degress of latitude.
In the mid-latitudes, the SST field is damped from observed initial SST
anomalies to climatology (1969-1998 reconstructed SST), with an
e-folding time of approximately 90 days.
Previous approaches to global SST prediction :
BSST = blended SST anomalies :
The prescribed SST field is a blend of forecast SST anomalies for the
tropical Pacific anomalies based on the tropical Pacific SST forecasts
from the NCEP coupled GCM and the Scripps Hybrid Coupled Model (HCM).
The observed SSTs for the initial time of forecast are also used in
forming the blended SST product for the first season.
Weights assigned to the observations and various forecasts change spatially,
and with season and forecast lead-time (weights for the observations are
zero after the first season). The tropical Indian Ocean SSTs
are statistically predicted based on the final forecast for the tropical
Pacific.
NSST :
The nsst forecast uses a unique blend of SST anomalies forcing the
specified atmospheric GCM. Over the Pacific from 25S to 30N, monthly
forecast SSTAs from NCEP are used. The NCEP SSTA forecast is then used to
make a statistical estimate of SSTAs in the Indian Ocean as far south as
25S. In the other global oceans, the latest observed SSTAs are persisted.
As these data sets are joined together month by month, the SSTAs are
smoothed where the different elements come together in order to avoid
abrupt SSTA changes.
HCM = Hybrid Coupled Model Predicted SST anomalies :
The prescribed SST field is predicted from the integration of
a hybrid coupled ocean-atmosphere model -- the atmospheric
component is statistical, and the ocean component is a GCM
covering the tropical Pacific.
(See Barnett et al., J. Clim., 6 (1993), p. 1545-1566).
Over the rest of the ocean basins time weighted values of the
current observed SSTA are used. Time weights are 100% for the
first four months of a forecast and then decrease to 90% in the
fifth month and then linearly to 0% in the ninth month.
The initial condition for the integration is the monthly
mean fields of the previous month taken from the ocean model
driven with statistically-derived (i.e. based on historical
SST-surface wind stress relationships) winds.
The forecasts presented are the ensemble averages of 10 runs.
FSST = forecast SST anomalies :
(These only exist prior to March 1997)
The prescribed SST field is forecast for the tropical Pacific
ocean (15S-15N) using the NCEP coupled climate model, CMP12 (Pacific
basin ocean/global atmosphere), where the ocean has been
initialized with assimilated observational sea temperature data.
Elsewhere, the SST field is damped from observed initial SST
anomalies to climatology (using 1961-1990 reconstructed SST data);
tropical oceans are damped with an e-folding time of 90 days,
all extra-tropical oceans are damped to climatology with an
e-folding time of 30 days.
MSST = mixed SST anomalies :
The prescribed SST field is a combination of forecast and persisted
observed SST in the tropical Pacific.
Elsewhere, the SST field is damped from observed initial SST
anomalies to climatology (using 1961-1990 reconstructed SST data);
tropical oceans are damped with an e-folding time of 90 days;
all extra-tropical oceans are damped to climatology with an
e-folding time of 30 days.
For example, a forecast made in March, through the upcoming Apr-May-Jun-Jul
would use the following weights for the SST fields to obtain the
prescribed SST during the integration. Anomalies are added to observed
climatology.
tropical Pacific sst (15S-15N)
osst fsst
FEB-MAR 1.0 0.
APR 0.75 0.25
MAY 0.5 0.5
JUN 0.25 0.75
Jul- 0. 1.00
where, osst: observed sst anomalies
fsst: forecast sst anomalies
extratropical sst (poleward of 25N/S):
observed sst damped to climatology with e-folding time 30days
Feb15 1
Mar15 1/e ~ 0.37
Apr15 1/(e*e)~ 0.14