The NCEP operational Climate Forecast System : configuration, products, and plan for the future

1
The NCEP operational Climate Forecast System
configuration, products, and plan for the future
Hua-Lu Pan Environmental Modeling Center NCEP
2
The NCEP coupled Climate Forecast System (CFS)
model

• 1. Atmospheric component

• Global Forecast System 2003
• T62 in horizontal 64 layers in vertical
• Recent upgrades in model physics
• Solar radiation (Hou, 1996)
• cumulus convection (Hong and Pan, 1998)
• gravity wave drag (Kim and Arakawa, 1995)
• cloud water/ice (Zhao and Carr,1997)

2. Oceanic component

• GFDL MOM3 (Pacanowski and Griffies, 1998)
• 1/31 in tropics 11 in extratropics 40
  layers
• Quasi-global domain (74S to 64N)
• Free surface

3. Coupled model

• Once-a-day coupling
• Sea ice extent taken as observed climatology

3
The other components of CFS

• Atmospheric initial condition NCEP reanalysis-2
• The reanalysis-2 is now an operational product
  and part of the CFS with a 4-day lag
• The pentad global precipitation analysis will
  become operational in the near future
• Ocean initial condition NCEP GODAS driven by
  the NCEP reanalysis-2 fluxes
• GODAS runs in two modes, a 14-day lag final cycle
  and a 7-day lag CFS analysis

4
Ensemble strategy

• One 10-month run per day
• This strategy makes the most sense for the
  operational center computer usage
• CPC can obtain a 20-30 member ensemble at any
  time
• 7-day lag GODAS analysis
• Daily update from the 14-day lag GODAS analysis
• 14-day lag GODAS is cycled but the 7-day lag one
  is not cycled
• PLEASE USE ONLY THE ENSEMBLE AND NOT ANY SINGLE
  RUN!!!!!

5
CFS Products

• Monthly mean fields
• Atmospheric fields in GRIB form 2.5 degree
  global grid of height, temperature, winds (U and
  V), relative humidity, etc at 17 levels
• Ocean fields in GRIB form (2degx1deg)
  temperature, wind, etc at 40 levels
• Single level fields in GRIB form global
  Gaussian grid of precipitation, 2-meter
  temperature, 10-meter winds, surface fluxes of
  heat and momentum, etc

6
CFS product (II)

• ftp//tgftp.nws.noaa.gov/SL.us008001/SL.opnl/MT.cf
  s_MR.fcst for real data
• ftp//tgftp.nws.noaa.gov/SL.us008001/SL.opnl/MT.cf
  s_MR.clim for corresponding climatology
• All data will stay on the site for 7 days to
  allow time to download (rotating 7-day archive)

7

• Seasonal retrospective forecasts by CFS to
  provide calibration and a priori skill assessment
• 15-member ensemble over 23 years from 1981-2003
• Runs are complete
• 10 month runs
• Initial atmospheric states Five 00Z analyses
  centered on the 1st, the 11th and the 21th day of
  each month from the Reanalysis-2 archive
• Initial ocean states the 1st, the 11th and the
  21th of each month from NCEP GODAS (Global Ocean
  Data Assimilation System)

8
Retrospective forecast productsmonthly and
seasonal mean

• CFS uncorrected forecast climatology
• Reanalysis-2 and GODAS climatology
• CFS forecast standard deviations
• Reanalysis-2 and GODAS standard deviations
• CFS uncorrected forecast root-mean-square error

9
(No Transcript)
10
3-month lead forecast
11
Three-month mean Nino34 SST correlation skill
Initial month 0-month lead 0-month lead 3-month lead 3-month lead 6-month lead 6-month lead
Initial month CLIPER CFS03 CLIPER CFS03 CLIPER CFS03
Apr 0.85 0.94 0.57 0.89 0.68 0.87
Jul 0.82 0.94 0.87 0.93 0.84 0.89
Oct 0.96 0.98 0.86 0.94 0.69 0.68
Jan 0.95 0.96 0.69 0.67 0.60 0.49
Three-month mean Nino34 SST RMSE (K)
Initial month 0-month lead 0-month lead 3-month lead 3-month lead 6-month lead 6-month lead
Initial month CLIPER CFS03 CLIPER CFS03 CLIPER CFS03
Apr 0.34 0.23 0.66 0.47 0.84 0.57
Jul 0.46 0.36 0.62 0.49 0.53 0.55
Oct 0.34 0.20 0.49 0.63 0.47 0.72
Jan 0.29 0.33 0.47 0.68 0.63 0.90
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
Conclusions

• The CFS system has predictive skill in the the
  equatorial tropical SST comparable to the
  statistical methods.
• With a complete retrospective forecast for 23
  years, the prediction skill for North America
  temperature and precipitation can be assessed.
• The CFS is run without flux correction and can be
  used as a 1-tier system.

21
Caveats

• North America seasonal prediction skills are
  still low but the skill masked regions seem
  complementary to the statistical tool results.
• We performed the kind of rigorous tests van den
  Dool and Livezey have been advocating for years.
  This is the kind of evaluation that CPC
  forecasters have long demanded.
• Statistical tools used by CPC are the benchmark
  for models. CFS is beginning to be competitive.
  Longer term goal is to beat statistical tools
  consistently just like NWP in the early days
  against human forecasters.

22
Why?

• Weather model for climate?
• Model tropics improved a lot over the past 15
  years
• 28 layer versus 64 layer?
• Stratosphere influence
• Surface stress and model drift?
• AMIP and CFS surface stress

23
Lessons learned

• Retrospective forecast requires huge human and
  computer resources and must be planned.
• Consistent re-analysis of the atmosphere and the
  ocean should be done using the same model and
  analysis system for the real time forecasts.
• Consistent retrospective forecast configuration
  and real-time configuration should be used.

24
User participation

• Some of the results so far seem encouraging and
  we need more people to look at the products
• We will provide all re-forecasts on a NOMADS
  server as soon as we can
• We encourage MOS development for seasonal
  predictions

25
Plan for the future

• Twice daily run in 2005
• T126L64 system in 2007 with re-analysis and
  re-forecast with goal for 2008 implementation
• Ocean model may be MOM4 or HYCOM (resolution?)
• Physics in a well-tested GFS will be used

26
Other topics

• Regional Climate Models
• Disk space requirement for CFS re-forecast
• Fair re-forecast evaluation
• Added value versus cost
• Multi-model ensemble
• Added value versus cost
• Maintenance issues