The NCEP Climate Forecast System Status and Plans HuaLu Pan and Suru Saha EMCNCEP

1
The NCEP Climate Forecast System Status and
PlansHua-Lu Pan and Suru SahaEMC/NCEP
2

• Status
• Current CFS system was implemented in 2004
• GFS2003 MOM3
• Direct coupling once a day
• Daily four 9-month forecasts
• One-day delay
• Initial State
• Atmosphere R2
• Ocean GODAS driven by R2
• 14-day lag for final GODAS which drives the
  real-time ocean analysis

3

• Calibrations and skill estimates
• Hindcast using the same system completed prior to
  implementation
• 1981 to 2007 but are continuing due to the slight
  differences in hindcast and real time designs
• CPC blends all forecasts based on the skill
  estimates
• It has become a product heavily used by CPC
• We are gaining converts from climate folks who
  did not believe dynamic models can have any
  skills in seasonal forecasts

4

• Honest assessments
• Skills in predicting the ENSO indices exist
  (however, it seems every model can predict strong
  events six months in advance)
• Skills in US temperature and precipitation
  prediction is low and lower
• It may take a long time to make advances
• We compare this version of the CFS to the
  Barotropic models of the NWP era. Not sure if the
  comparison is adequate yet.
• We believe in taking small but scientifically
  sound steps to move forward using GFS and GDAS as
  base

5

• CFSRR
• We are starting to work towards an implementation
  of the next CFS in 2011
• Reanalysis of the atmosphere, ocean, land and sea
  ice, using a coupled background guess forecast
• Reforecasts using the same model used in the
  reanalysis
• Real time data assimilation and forecast will use
  the same system as the reanalysis and reforecast

6

• CFSRR Plan (I)
• Reanalysis
• Using the CFS as the model to provide the first
  guess to the data assimilation of atmosphere,
  ocean, land, and sea ice
• Using the highest resolution we can afford (we
  did dream big hoping that we will get mist and
  dew when the power 6 machine arrives)
• Using the newest data assimilation and model
  versions tested (drawbacks do exist)

7

• Reanalysis
• T382L64 for atmosphere with GSI, 0.25 - 0.5
  degree 40 layer ocean with stronger SST
  constraint, NOAH land model plus observed
  precipitation for GLDAS, extend the new sea-ice
  analysis back to 1979
• Coupling is on ocean and ice-atmosphere time
  steps
• All satellite instruments calibrated using
  3-month spin up runs.
• Calibrated SSU and MSU data included
• Running on Haze (5-6 streams since June 2008)
• CPC has developed and undertaken close monitoring
  of the CFSR
• We have, within our limit, tried to correct
  mistakes. A few glitches, which would have
  necessitated complete restarts, have been left as
  features.

8

• CFSRR Production Configuration
• Covers 31 years (1979-2009) 25 overlap months
• 6 Simultaneous Streams
• Jan 1979 Dec 1985 7 years 1982/06
• Nov 1985 Feb 1989 3 years 1987/03
• Jan 1989 Feb 1994 5 years 1990/07
• Jan 1994 Dec 1998 5 years 1995/07
• Apr 1998 Dec 2004 6 years 2002/04
• Apr 2004 Dec 2009 5 years 2007/01
• overlap months are for ocean and land spin ups

9

• Reforecast
• All coupled reforecasts will be run at T126L64
  resolution
• 9-month hindcasts will be initiated every 5th day
  and will be run from all 4 cycles of that day,
  beginning Jan 1 of every year, over a 28 year
  period from 1982-2009.
• In addition, there will be a single 35-day
  hindcast run initiated from every cycle between
  these five days, over the entire period.

10
MID JANUARY SEASONAL RELEASE (24 members)
JANUARY MONTHLY (35-DAY) HINDCAST SCHEDULE
11

• Reforecast
• Initially, make complete reforecast for CPCs
  mid-May and mid-November forecast release, to
  assess skill. Once we can demonstrate comparable
  or better skills for ENSO, US temperature and
  precip for winter and summer forecasts, we can
  start the implementation process
• Forecasts generated from high resolution data
  assimilation may allow for a useful monthly
  forecast product for CPC. Month one in the CPC
  parlance is really week3 week 6 forecast

12

• Real Time Configuration for next CFS
• There will be 4 control runs per day, out to 9
  months, from the 0, 6, 12 and 18 UTC cycles of
  the CFS real-time data assimilation system.
• In addition to the control run, there will be 3
  additional runs per cycle, out to 35 days. These
  3 runs per cycle will be initialized as in
  current operations.
• There will be a total of 16 runs every day out to
  35 days (weeks 1-5), four of which will go out to
  9 months (monthly and seasonal)

13
CFSRR Computer Configurations for Retrospective
Forecasts
Each one-year run on power 5 requires 5 nodes for
40 hours
Number of 9 month Hindcasts to be made is 292 per
year X 28 years 8176
Number of extra 35-day Hindcasts to be made is
1196 per year X 28 years 33488 ( 4000 9-month
runs)
Legend of available systems
Haze 75 Compute7 Class 1 100 TB
Zephyr 16 Compute 1 Class 1 40 TB
Mist 140 Compute 8 Class 1 75 TB
Dew 140 Compute 8 Class 1 75 TB
3 Streams
28 Streams
28 Streams
15 Streams
14

• Resources
• Human resources
• Funding
• Computer

15
Human resources

• Suru Saha is the leader of the group who maintain
  the continuous running of the reanalysis and the
  reforecast
• Xingren Wu, Jiande Wang, Sudhir Nadiga and
  Patrick Tripp do the actual monitoring of jobs
  and runs
• S. Moorthi built the reanalysis and reforecast
  scripts from end to end
• Suru and Patrick Tripp built the scripts for HPSS
  archives
• Jack Woollen provides most of the observation
  support
• Bob Kistler provides overall support of the
  reanalysis
• CPC formed four teams of about 20 people to
  provide monitoring of the reanalysis
• Huiya and Wesley provided continued support for
  the NCEP post and the GRIB2 conversions
• Xingren worked on the ice model and assimilation,
  Jiande worked on the checkout of MOM4 for our
  configuration
• Jun Wang worked with Xingren to produce the
  MOM4-GFS coupler
• Lidia helped with the assimilation and evaluation
  of CHAMP and COSMIC data

16
Human resources (II)

• Haixia Liu from the data assimilation group
  provided help with the satellite instrument bias
  estimates
• Glenn White has provided diagnostics of the
  stratus
• Diane Stokes provided SST data support as well as
  general data support
• Dave Behringer provided GODAS
• Jesse, Ken and the NOAH land team provided GLDAS
• Bob Grumbine provided sea ice analysis
• NESDIS provided recalibrated SSU and MSU data
  (Goldbergs team)
• Yutai Hou provided CO2 changes as well as codes
  to incorporate the CO2 changes
• Paul van Delst updated the CRTM for all the
  instruments
• Several people from the assimilation team (Daryl
  and Russ in particular) helped with the initial
  set up and running of the GSI
• Pingping Xie and others from CPC provided the
  gauge only global precip data for GLDAS
• George Gayno provided snow analysis with guidance
  from Ken
• Computing support from Doris, Carolyn and George
  are crucial

17
Funding

• Funding is provided by the Climate Program Office
  for reanalysis, ocean data assimilation and
  climate forecasting
• Funding for Reanalysis should continue. As there
  are other reanalysis proposals within NOAA, our
  shares of the funding may change. We hope to fund
  three positions for data collection, data
  assimilation, and transfer of corporate
  experience

18
Computer

• CFS upgrade can only be done when there is a
  computer upgrade
• Only way to do the reanalysis and reforecast is
  to leverage the usage of old computers after EMC
  and NCO migrated to the new ones
• RD computer is helpful but not sufficient
• Delivery of the milestone depends entirely on the
  availability of the computers
• Reanalysis is run entirely on Haze

19
Whats new

• New data
• SST, SSU, AMMA, CO2
• New assimilation system
• GSI with FOTO GODAS with strong SST constraint
  and asymmetric data window, GLDAS with observed
  precip simple sea-ice initialization using
  observed ice fraction
• New model
• The package that was in parallel in 2007 for GFS
• Key portion is RRTM short-wave which allowed for
  easy CO2 updates
• MOM4 with a sea-ice model
• NOAH for land

20
Anticipated improvements

• R1 had a very quiet tropics and R2 had a very
  noisy tropics
• The GFS system since 2000 has shown a much
  improved tropical analysis
• Tropical ocean is crucial to the seasonal
  prediction so better forcing from the atmosphere
  should lead to better ocean analysis
• This benefit can start in week two and go to
  monthly and seasonal
• We plan to interact with CPC to evaluate the
  monthly hindcasts and the skill estimate issues

21
Problems encountered

• New data
• SST and sea ice mask problems
• Sea ice over land and sea
• AMMA data issue
• New model
• Stratus
• New assimilation system
• QBO, SAO and background error estimates
• New compiler
• A bug was introduced when converting to new
  compiler. Took a long time to realize and correct
• Human resource issue
• 24/7 monitoring takes the toll. The length of the
  project becomes an issue if we suffer delays in
  the production
• Computer resource issue
• Delays of the computer upgrade continues to cause
  delays in our planned implementation

22
Samples of the monitoring of the reanalysis
23
Web site for monitors
24
Monitoring of 5-day forecast skill (500 hPa
height anomaly correlation) for 2005 Green line
is operational GFS at high resolution, black line
is CFSRR with forecast made at T126, Red line is
with CFS system but with both data assimilation
and forecasts done at T62 with a yearly cold
start)
25
Monitoring of 5-day forecast skill (500 hPa
height anomaly correlation) for 1980 Black line
is CFSRR with forecast made at T126, Red line is
with CFS system but with both data assimilation
and forecasts done at T62 with a yearly cold
start)
26
Sample of daily monitor of the reanalysis using
short-range forecast skill to check
27
Comparing analysis with other reanalysis results.
Here the comparison is with R2
28
Comparing analysis with other reanalysis results.
Here the comparison is with era40
29
Monitoring MJO in the data assimilation system
against R1, R2, and era40 against R1 climatology
30
(No Transcript)
31
CFSRR
GFS operational
Monitoring of fits of analysis and first guess to
surface observations showed improved fit of the
CFSRR over the then operational system (12-48
hour forecasts for CFSRR were made with lower
resolution so gives slightly worse fits to obs)
32
Monitoring land surface temperature against
independent observations
33
Monitoring land surface energy balance
34
Monitoring sea ice distribution against data used
in operation
35
Monitoring SST against OISST
36
Monitoring depth of 20C against offline analysis