Update of NCEP ShortRange

1

Update of NCEP Short-Range Ensemble Forecast
(SREF) System inclusion of WRF members
Jun Du, Jeff McQueen Dusan Jovic, Xiaoxue
Wang, Binbin Zhou, Geoff Dimego, Matt Pyle,
Hui-Ya Chuang, Eric Rogers, George Gayno and
Mike Baker http//wwwt.emc.ncep.noaa.gov/mmb/SR
EF/SREF.html
2
WRF Member Setting
Log Started Aug. 1, 2005 Implemented Dec. 6,
2005 Took EMC and NCO 4 months to accomplish
3
SREF System

• IC/LBC Uncertainty (1) multi-analysis (gdas and
  ndas) and (2) bred modes
• Model Physics Uncertainty (1) multi-models (Eta,
  RSM, NMM, ARW) and (2) 6 convective schemes
• Post-Processing currently missing but will
  implement bias correction within 2006
• 21-member, 87hr, 09z/21z, North America/
  Alaska/Eastern Pacific (Hawaii), both grid and
  bufr, individual/mean/spread/prob, conventional
  and aviation products, web, ftp

4
RMSE (SREF mean fcst, 8/25/05-9/17/05)
SLP
500H
850U
850T
250U
850RH
5
SREF mean fcst vs 12km NAM (rmse)
SLP
500H
850U
850T
850RH
250U
6
Ensemble mean RMSE with height at F48hr (Nov.
2005)
RMSE
BIAS
7
Ensemble Mean RMSE (Nov. 2005)
10m Wind
2 m Temp
RMSE
BIAS
8
Ensemble spread (15mem vs 21mem, 8/25/05-9/17/05)
SLP
500H
850T
850U
250U
850RH
9
Ensemble Spread vs Forecast Error
21-member SREF
15-member SREF
500H
SLP
500H
SLP
850T
850U
850T
850U
250U
850RH
250U
850RH
10
Statistical Consistency at 48hr (Nov. 2005)
UPA Temp

• Mean Squared Error/VarianceError
• best 1 (Eckel and Mass, 2005)
• SREF-21 improved over SREF-15
• WRF subset yields lowest SC compared to Eta
  subsets

UPA Winds
11
SREF PerformanceStatistical Consistency (Nov.
2005)
2 m Temp

• Mean Squared Error/VarianceError
• best 1 (Eckel and Mass, 2005)
• SREF-21 improved over SREF-15
• WRF subset yields lowest SC compared to Eta
  subsets

10 m Winds
12
(No Transcript)
13
(No Transcript)
14
Outlier (chance analysis falls out of ensemble
fcst range
SLP
500H
850T
850U
850RH
250U
15
12h-apcp scores (15mem vs 21mem, 8/25/05-9/17/09)
ETS
RPSS
16
Outstanding Issues

• For upper-air fields including SLP and precip,
  the number 1 priority of the SREF is no longer
  the improvement to under-dispersive nature after
  the 6-WRF member added but the bias correction
• For near surface elements such as 2m T and
  humidity, 10m wind, improvement in ensemble
  spread is, however, desperately needed (PBL, LSM
  )

17
(No Transcript)
18
SREF Operational PerformanceTalagrand Ranked
Histograms 48 h forecasts (November 2005)
850 mb Temp SREF/15
850 mb Temp SREF/21
Poorer performance over Western Region
19
SREF Operational PerformanceTalagrand Ranked
Histograms 48 h forecasts (November 2005)
250 mb winds SREF/21
250 mb winds SREF/15
Poorer performance over Western Region
20
SREF-related other projects

• DTRA project (Jeff, Dusan, Hui-Ya) drive
  dispersion model with hi-res SREF learn to
  perturb PBL, LSM and urban model (with PSU)
• Beijing 2008 Olympic project (Jun, Geoff) run
  real-time SREF during the game (a) relocation
  capability of regional ensemble system (b)
  assimilating radar and mesonet non-conventional
  obs into mesoscale EPS (with CMA visitors)
• Coupling CMAQ (air quality model) with SREF (Jun,
  Pius and Jeff) future direction of NCEP
  operational CMAQ (with a UMD PhD student)

21
Learn from MREF Group

• Ensemble Transform (ET) technique
• Stochastic physics
• Others