ECPC’s Mar 2004 Seasonal Forecasts
J. Roads,
M. Kanamitsu, L. De Haan, J. Ritchie
Experimental Climate Prediction CenterScripps Institution of OceanographyUCSD, 0224
La Jolla, CA 92093
1. ECPC’s Experimental
Forecast System
There is a major change underway at the Scripps
Experimental Climate Prediction Center (ECPC). Previously, the ECPC used the
reanalysis I version (Kalnay et al. 1996) of the National Centers for
Environmental Prediction’s (NCEP’s) medium range forecast (MRF) model or global
spectral model (GSM; Roads et al. 2001a) to make routine experimental global
forecasts. These global forecasts (daily out to 7days and weekly out to
16-weeks) start from the NCEP operational 00UTC global analysis and use
persisted SST anomalies (+climatology) as a lower boundary condition.
These GSM forecasts (e.g. Roads et al. 2003a)
are being replaced by an updated version of NCEP's seasonal forecast model
(SFM; Kanamitsu et al. 2002a), which is based on updated physics from the
NCEP/DOE reanalysis II (Kanamitsu et al. 2002b). The SFM has a nominal (a
reduced grid technique is used near the poles) horizontal resolution of T62
(about 2o). There are 28 levels in the vertical sigma coordinate
system. ECPC's SFM is run in a different fashion from the GSM. Starting from
slightly perturbed initial conditions, and forced with observed SST anomalies,
10 simulations are made up to present.
Then, persisted SSTs or forecast SSTs are used to generate a forecast
ensemble. The forecast SSTs come from a
simplified model for the tropical Pacific and are produced by the IRI. This new
SFM is being coupled to an ocean model and sometime in the future we hope to
demonstrate that such a coupled system will be demonstrably better than current
persisted or forecast SSTs as well as our current ocean forecasts, which use
forecast GSM anomalies to drive a Pacific Ocean model (Auad et al. 2003).
A major advantage of the SFM over the GSM is
that the computer code of the SFM was completely rewritten to run on multiple
platforms with single and/or multiple shared memory machines. The code was
improved further to run on massively parallel processor (MPP) machines using
Message Passing Interface (MPI) routines.
The SFM is now running on the COMPAS cluster at the Scripps Institution
of Oceanography. Normally the SFM runs
on 64 processors and takes 2 hours to make a 7-month forecast. Depending upon
the number of ensemble members, a normal 7-month forecast takes between 1-2
days. During the rest of the month background runs are being made to augment
the growing ensemble climatology. In addition, as changes are made in the model
new climatologies have to be developed. In fact, there are a few physical
parameterization differences between the ECPC SFM and the NCEP SFM. The ECPC SFM has an updated set of land
physics state as well as revised formulation of land surface evaporation.
However, it should be noted that the NCEP SFM does start from observed initial
conditions unlike the ECPC SFM, which is starting from previous simulations.
Ignoring the initial conditions is generally thought to be reasonable when
considering long-lead forecasts (greater than a month) although there are
certainly times and places when initial conditions can be important even for
seasonal forecasts (Reichler and Roads 2003a,b,c,d). Another difference between ECPC SFM and NCEP SFM is the initial
condition of the soil moisture. In the
NCEP SFM, climatological soil moisture is used while in the ECPC SFM; the
simulated soil moisture is used. This
may impact the forecast skill particularly in warm seasons.
The ECPC SFM has not yet fully replaced the ECPC
GSM in part because the GSM is currently tightly linked to a number of
additional models and applications. The GSM forces a regional spectral model
(RSM; Juang et al. 1997; Anderson and Roads 2002, Roads et al. 2003b, c, Roads
2003a,b, Chen and Roads 2003) in order to gain increased spatial resolution
(50-25 km resolution) for several selected regions (US, CA, SW, Brazil). The
GSM and RSM are based upon the same physics used in the GSM (and SFM) and can,
in principle, be updated as the GSM (SFM) is updated. We are attempting to
implement these updates and to replace the GSM with the SFM but this process
may take some time due to lack of personnel. Current output products from the
GSM/RSM include a fire weather index and associated variables such as
2m-temperature, relative humidity and 10m-windspeed as well as precipitation
and soil moisture. Additional RSM products are provided to drive US National
Fire Danger Rating System Indices (Roads et al. 2003d) and surface hydrologic
models.
2.
Forecast Skill Evaluations
Five years worth of forecasts (260 forecasts)
were previously used to develop GSM/RSM forecast climatologies, which are
dependent upon season as well as forecast lead-time. Both means and standard
deviations were derived in order to provide normalized (by their respective
standard deviation) anomalies. As discussed by Roads et al. (2001a,b), Roads
and Brenner (2002), Roads et al. (2003a,b); Roads (2003), Chen et al. (2001),
Chen and Roads (2003), the GSM/RSM provides skillful forecasts of temperature,
precipitation, soil moisture and fire danger indices at long forecast ranges.
Although the greatest skill occurs initially and then rapidly decays, monthly
and seasonal averages can still demonstrate significant skill (Reichler and
Roads 2003a,b,c,d), which may be comparable to empirical long-range forecast
methodologies.
ECPC SFM forecast skill evaluations are still
underway in collaboration with the IRI and will be reported upon later. Suffice
it to say that the ECPC SFM has skill comparable to other forecast models used
by IRI, namely MPIM's ECHAM, NCAR’s CCM and COLA’s GCM. ECPC SFM forecast skill
apparently exceeds that of others in some areas and in some seasons, and thus
contributes to making a better multi-model ensemble forecast for the IRI and
NCEP.
As a preliminary evaluation of the SFM
forecasts, we have been comparing the GSM anomaly forecasts with the
corresponding SFM forecasts. It should be noted here that the GSM is
initialized at the beginning of the month, whereas the SFM ensemble members are
initialized one month earlier from continuous simulations using observed SSTs.
The GSM and SFM also have different base climatologies (5 years for the GSM and
52 years for the SFM). Nonetheless, there are some remarkable agreements as
well as some major disagreements. We first show the comparison in which the SFM
is forced by persisted SSTs. There are many similarities and some differences.
We then show SFM forecasts forced by the IRI forecast tropical SSTs.
Fig. 1 shows that during MAM,
both models show forecast temperature anomalies being high over Alaska, Canada,
Europe, Russia and China and Africa. Anomalies are low over Southeast Asia and
the Southwest US. There are a few differences. GSM has Europe below normal
while the SFM has it above normal.
Precipitation anomaly forecasts show greater
similarities. Fig. 2 shows that
during MAM both models are forecasting above normal precipitation over the
Indian Ocean and in a tropical strip north of the equator. Above normal
precipitation is also being forecast for the US and Europe. Slightly below
normal precipitation is being forecast over Australia and Africa. There are
certainly some differences. For example, the SFM tends to be dry over Central
America and the Indonesian Archipelago, whereas the GSM has above normal
rainfall there.
3. Global
seasonal GSM forecasts and US monthly RSM forecasts
Figs. 3,4 show the SFM seasonal
forecast anomalies for additional months. Note there are similarities and
differences between these forecasts and those described above using persisted
SSTs.
Persistent below normal seasonal temperatures (Fig. 3) continue to be forecast for
most of the NH land regions, except for western Russia. Above normal
temperatures tend to be forecast for the Arctic. Above normal temperatures are
also being forecast for most SH land areas, except for Antarctica.
Precipitation (Fig. 4) also shows this remarkably persistent character in the
tropics, which may be related to the persistent temperature forecasts. Above
normal precipitation is being forecast by the SFM over the southern Indian
Ocean, the equatorial Pacific, the US Northwest, and below normal precipitation
is being forecast over South Africa, India, and the Indonesian Archipelago.
References
Anderson,
B.T., and J. O. Roads, 2002: Regional Simulation of Summertime Precipitation
over the Southwestern United States. Journal
of Climate, 15, 3321-3342.
Auad, G., A. Miller, J. Roads 2003: Ocean Forecasts. J. Marine
Res. (submitted)
Chen, S-C. J. O. Roads, and M. Wu, 2001: ECPC’s Asia forecasts. Journal
of Terrestrial-Atmosphere-Oceanography, 12, 377-400.
Chen, S. and J. Roads,
2003: Regional Spectral Model Simulations for South America. J. Hydrometeor.
(submitted)
Juang, H.
-M. H., S. -Y. Hong and M. Kanamitsu, 1997: The NCEP regional spectral model:
an update. Bulletin Amer. Meteor. Soc., 78, 2125-2143.
Kalnay,
E. et al., 1996: The NMC/NCAR reanalysis project, Bull. Am. Meteor. Soc., 77, 437- 471.
Kanamitsu, M., Cheng-Hsuan Lu, Jae Schemm and W. Ebisuzaki,
2003a: The predictability of soil
moisture and near surface temperature in hindcasts of NCEP Seasonal Forecast
Model. J. Climate, 16, 510-521.
Kanamitsu, M. and Kingtse, Mo, 2003b: Dynamical Effect of Land
Surface Processes on Summer precipitation over the Southwestern United States. J.
Climate, 16, 496-509.
Reichler,
T. J. and J. O. Roads, 2003: The Role of Boundary and Initial Conditions for
Dynamical Seasonal Predictability. Nonlinear
Processes in Geophysics, 10 (3) May/June 2003, 1-22.
Reichler,
T. and J. O. Roads , 2003: Time-space distribution of long-range atmospheric
predictability. J. Atmos. Sci.,
(submitted).
Reichler,
T. and J. O. Roads, 2003: Long-range predictability in the tropics. Part I:
monthly averages. J. Climate, (submitted).
Reichler,
T. and J. O. Roads, 2003: Long-range predictability in the tropics. Part II:
30-60 days variability. J. Climate,
(submitted).
Roads, J.O., S-C. Chen
and F. Fujioka, 2001a: ECPC’s Weekly to
Seasonal Global Forecasts. Bull. Amer.
Meteor. Soc., 82, 639-658.
Roads, J., B. Rockel, E.
Raschke, 2001b: Evaluation of ECPC’s Seasonal Forecasts Over the BALTEX Region
and Europe. Meteorologische Zeitschrift Vol.
10 (4) p. 283-294.
Roads, J. and S.
Brenner, 2002: Global Model Seasonal Forecasts for the Mediterranean Region. Israel Journal of Earth Sciences. 51 (1),
1-16.
Roads, J. 2003:
Experimental Weekly to Seasonal, Global to Regional US Precipitation Forecasts J.
Hydrology (in press)
Roads, J., S.-C. Chen,
M. Kanamitsu, 2003a: US Regional Climate Simulations and Seasonal Forecasts. Journal
of Geophysical Research-Atmospheres (in press).
Roads, J., S. Chen, S. Cocke, L. Druyan, M. Fulakeza, T. LaRow, P. Lonergan, J.-H. Qian, S. Zebiak, 2003b: The IRI/ARCs Regional Model Intercomparison Over S. America. J. Geophys. Res. (in press).
Roads,
J., F. Fujioka, S. Chen, R. Burgan, 2003c: Seasonal Fire Danger Forecasts.
Inter. J. Wild. Fire (submitted).
Fig.
1 Seasonal GSM (upper) and SFM (lower) temperature (2 m) forecasts
(K). Note the different temperature scales for each panel. The GSM was
initialized on 02/28 and the SFM was initialized at the beginning of 02/04. The
SFM also shows the ensemble mean of 10 forecasts.
Fig.
2 Seasonal GSM (upper) and SFM (lower) precipitation forecasts
(mm/day). Note the different scales for each panel. The GSM was initialized on
02/28 and the SFM was initialized at the beginning of 02/04. The SFM also shows
the ensemble mean of 10 forecasts.
Fig.
3 Seasonal SFM temperature (2 m) forecast anomalies (K). The SFM
ensemble was initialized at the beginning of 02/04 and forecasts were made for
the next 7 months. 3 month running mean forecasts are shown in the 4 panels.
Fig.
4 Seasonal SFM precipitation forecast anomalies (mm/day). The SFM
ensemble was initialized at the beginning of 02/04 and forecasts were made for
the next 7 months. 3 month running mean forecasts are shown in the 4 panels.