Penguin and the ECM Analysis Forecasts

1

Penguin and the ECM Analysis Forecasts
Jordan Alpert
Jordan C. Alpert, NOAA/NWS/EMC jordan.alpert_at_noaa.
gov 20061212 EMC Briefing
2
Tests GFS pgbanl analysis compared with ECM
analysis and GFS forecast

• ECMWF GRIB file at 1 degree pgbanl type file w/
• Ps,u,v,q,t (q from rh) at 15 levels but GFS
  pgbanl RHlt50.
• Interpolate to standard levels with ops orog and
  
• Write into BUFR file PREPQC as the only obs to
  make sanl using SSI to interpolate the sigma IC.
• In addition to ECMWF run and GFS operational
  runs,
• GFS IC from pgbanl (prgg)
• ECM (pryy) test experiment with 6-h guess and
  from GFS pgbanl IC to check Antarctic problem.
• Next 2 slides SSI pgbanl IC with GFS 5 day
  forecast.
• These show SSI is a good interpolator and method
  reproducesresonably sigma type restart for each
  model.

3
gg run NH GFS pgbanl file as IC similar (but
worse) to GFS forecast. (GFS OPS (Black), GFS
pgbanl (Green), ECMWF OPS (Red)
4
gg run SH GFS pgbanl file as IC similar (but
worse) to GFS forecast. (GFS OPS (Black), GFS
pgbanl (Green), ECMWF OPS (Red)
5
ECM Sept. (Penguin) runs

• GFS starting from pgbanl SSI files produces
  similar forecast as OPS GFS.
• Next slides are ECMWF pgbanl files with GFS
  forecasts for SEP 2006 period.
• Significant improvement from using the ECMWF IC
  (even as pgbanl 1x1) in fact, ECMWF Analysis is
  pre-condition to give best GFS forecast close to
  ECMWF forecast in height AC.
• But tropical winds are not improved.

6
ECM pgbanl IC in Green, ECMWF in Red, GFS OPS in
Black
7
ECM pgbanl IC in Green, ECMWF in Red, GFS OPS in
Black
8
SH Die off curves, note that ECM comparison is
to NCEP SSI.
9
NH Die off curves, note that ECM comparison is
to NCEP SSI.
10
RMS Die Off curves, SH NB RMS difference between
analysis (upper left at days0) and Mean (bias in
out days upper right) is improved for height.
11
RMS 200mb wind is not improved
12
RMS 850mb wind is not improved. NB large
difference in analyses.
13
ECM runs

• Changing the (surface or sigges) guess for
  siganl/sfcanl makes no difference on scores.
• Surface Temperature differences
  at f00 GFS is cooler near coastline and
  inland but warmer off shore of Antarctica.
• Greatest difference occurs at IC and decreases
  with time.

14
Diff ECM-GFS, f00, T (sig995)
15
Diff ECM-GFS, f48, T(sig995)
16
SFC Gnd Heat Fluxw/m2 nb extent of temporary
sea ice in land surface.
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
Zonal Ave T (18) Diff ECM-GFS pgbanl IC. Note
Antarctica orog
21
Zonal Ave T (18) Diff ECM-GFS (from pgbanl IC)
24-h.
22
Zonal Ave T (18) Diff ECM-GFS (from pgbanl IC) by
48-h Surface differences gone.
23
Zonal Ave UGRD (18) Diff ECM-GFS (pgbanl IC) f00.
24
Zonal Ave UGRD (18) Diff ECM-GFS (pgbanl IC) 24-h.
25
Zonal Ave UGRD (18) Diff ECM-GFS (pgbanl IC) 48-h.
26
ec/gfs Parallel

• Yuejians verification page shows other ecm
  products
• http//www.emc.ncep.noaa.gov/gmb/yzhu/html/GFS_m_m
  odel.html
• 4-way comparison of ECMWF and GFS
• GFS ---gt NCEP forecasts from T00Z
• ECM ---gt ECMWF forecasts from T12Z
• E-G ---gt GFS model runs ECMWF initial condition
  from T00Z
• G-E ---gt ECMWF model runs GFS initial condition
  from T12Z?

27
GFS/ECM Overlay runs

• Fanglin p GFS overlay ECMWF 60(50)-90S
• Alpert o ECMWF overlay GFS 60-90S

28
Fanglins verification (from e-mail)
29
As in Fanglins average of overlay runs similar
half way to ECM
30
Overlay makes no difference in NH so p like
ECMWF and o like GFS.
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
Summary

• SSI analysis from pgbanl files faithfully
  reproduces a proxy for GFS/ECMWF model analysis
  and different surface/guess files do not change
  result.
• Overlay runs indicate that the pre-conditioning
  of ECMWF analysis to give improved GFS forecasts
  is a hemispheric scale phenomenon.
• The differences in the SH do not appear to be
  directly related to local affects, eg., sea ice
  changes, or the recent operational implementation.