The ATOVS assimilation at JMA

1
The ATOVS assimilation at JMA

• Kozo Okamoto1
• Masahiro Kazumori2
• Hiromi Owada2
• 1Japan Meteorological Agency/EMC visiting
  scientist
• 2 Japan Meteorological Agency

2
NWP models at JMA

• GSM Global Spectral Model
• T213L40 (top 0.4hPa)
• 90h (00UTC),216h (12UTC)
• weekly weather forecast
• RSM Regional Spectral Model
• 20km L40 (325x257, top 10hPa)
• 51h (00,12UTC)
• short range weather forecast (up to 2days)
• MSM (gt NHM FY2003)
• 10km L40 (361x289, 10hPa)
• 18h (00,06,12,18UTC)
• disaster prevention
• TYM Typhoon Model
• 24km L25 (271x271, 10hPa)
• 84h (00,06,12,18UTC)
• Ensemble Prediction Model
• T106L40 (0.4hPa), 216 (12UTC), 25Members
• weekly weather forecast

3
Analysis systems at JMA

• 3DVar for GSM (since Sep2001 gt 4DVar Mar2004)
• cut off time 2.5h (Early anal), 7.5/12.5h(Final)
• satellite assimilation ATOVS radiance, SeaWinds
  wind, Geo.sat. AMW
• 4DVar for RSM (Jun2003)
• 3.0h (00,12 UTC), 8.5h(06,18 UTC)
• ATOVS temperature, Geo.sat. AMV
• 4DVar for MSM (Mar2002)
• 50min (00,06,12,18 UTC)
• ATOVS temperature, Geo.sat. AMV, (SSMI/TMI
  TPWRR)
• 3DVar for TYM (Sep2001)
• 1.5h (06,18 UTC)
• same as global 3DVar

4
Recent developments of the satellite assimilation
at JMA

• ATOVS radiance assimilation (May2003)
• SeaWinds derived sea surface winds
• QI on METEOSAT AMW is used for QC
• SSMI/TMI derived TPWRR for Regional/Meso 4DVar
• SSMI/TMI/AMSR derived TPWsea surface wind
  velocity for global 3DVar
• GPS occultation for global 3DVar
• ground based GPS TPW for Regional/Meso 4DVar
• clear radiance from the geostationary satellite
• cloudy radiance from the IR imager/sounder

The satellite assimilation sub-group was
established in NWP division/JMA in 2000
5
SSMI/TMI RainRate TPW impact on MSM
W RR TPW
W/o SSMI/TMI
W RR
Radar
Anal
6h Fcst
12h Fcst
?
?
?
?
Init18Jun2002 18UTC
From Y.Sato
mm/h
6
ATOVS data process flow comparison
Retrieval Use
Radiance Use
1DVar as pre-processor
Pre-processor

• NESDIS/MSCretrieved T,q, dZ
• Thinning
• Conversion
• QC
• Select area lev

• NESDIS TBB
• Thinning
• QC
• Ch selection
• Bias Correction

B.C.TBB, Tskin, emissivity, dT/dp over model top
dZ( -1000hPa)
3DVar
3DVar
7
1DVar as pre-processor

• J (x-xb)TB-1(x-xb) (y-H(x))TR-1(y-H(x))
• Xanalysis variables
• T lnQ at 41 levels of GSM and Tskin
• Y observation
• HIRS/AMSUA/AMSUB TBB from NESDIS BUFR
• AMSUA is mapped to HIRS
• H observation operator
• RTTOV-6 (Saunders et al.,2000)

8
Thinning

• Constant distance
• 250km(HIRS/AMSA)
• 180km(AMSUB)
• Priority is given to
• clear radiances
• satellite closer to analysis time when they
  overlap

N.H.
S.H.
7 July 2003 00UTC
9
Cloud/Rain detection

• Detect clear/thin-cloud(IR)/thick-cloud(MW)/rain
  using only observation information (not guess)
• Cloud detection for HIRS/AMSUA J ( y-yclear
  )T C-1 ( y-yclear )
• y TBobs vector of HIRS for thin(IR) cloud
  detection AMSU-A for
  thick(MW) cloud detection
• yclear mean clear TBB , C clear TBB covariance
• J gt J0 gt cloudy
• Rain detection over sea for HIRS/AMSUA SI
  TBcal(A15) - TBob(A15)
• TBcal(A15) is calculated based on a statistical
  regression approach with predictors of AMSU1-3
• SI gt 10 gt rain
• Rain detection over sea for AMSUB SI
  TBob(B1) - TBob(B2)
• SI gt 3K gt rain

10
STD of clear TBB21Aug - 9Sep 2002
NOAA16 over Sea
NOAA15 over Sea
NOAA16 over Land
NOAA15 over Land
11
Ch selection Obs error

• Depend on 8 obs conditions
• clear/cloudy/rain sea/sea ice/land
• Obs errors are enlarged 2.3 4.5 times in 3DVar

HIRS
AMSU-A
AMSU-B
12
TBB Bias Correction

• PredictorsTPW,Tskin,TBobs(A5),TBobs(A7),
  TBobs(A10)
• Defined every scan position
• The coefficients are produced by TBobs-TBcal
  dataset.
• Radio sonde observations are used for TBBcal due
  to the presence of a model bias
• Guess is used to complement the lack of the upper
  stratosphere temperature and moisture.
• Not bias-correct AMSU-A12-14, AMSU-B, HIRS11-12
• due to large model bias

13
dTB w/wo BC at each scan position dTBTBobs-TBges
A4 A5 A6
A7
A8 A9 A10
A11
A12 A13 A14
27June2002
14
Bias grows (1)
MEAN
RMS

• Mean and RMS temperature difference between
  analysis/guess and RAOB
• In the upper tropics, temperature rises through
  assimilation cycle

NH
Trp
SH
15
Bias grows (2)
MEAN
RMS

• Use of TBsnd instead of TBges for the regression
  coefficients stopped the bias from growing.

NH
Trp
SH
16
Example of Bias Correction TBobs-TBges (1)
A4
A5
A6
A7
A8
A9
A10
A11
A12
A13
A14
BluewoBC / RedwBC
Range -5K-5K
17
Example of Bias Correction TBobs-TBges (2)
w/o BC
A4
A5
Range -6K-6K
A7
A6
A8
A9
A10
A12
A11
A13
A14
A4
A5
A7
A6
A8
A9
A10
w BC
A12
A11
A13
A14
18
Impact Experiments

• Forecast model T213L40 (top0.4hPa)
• Analysis incremental 3DVar (inner T106L40)
• 6 hourly cycle and 9-day forecast at 12 UTC
• Test
• NOAA1516 ATOVS radiance assimilation from 120km
  BUFR
• Modified convective scheme
• Modified 3DVar background error statistics
• Cntl
• NOAA1516 ATOVS retrieval assimilation from
  SATEM120kmBUFR
• GMS-derived moisture bogus
• Period
• Jul2002 06/27/2003 08/09/2002
• Dec2001 11/27/2001 01/09/2002

19
1-month statistics for Analysis/Guess Temperature
verified against RAOB for the Dec2001 experiment
20
1-month statistics for Analysis/Guess Temperature
verified against RAOB for the Jul2002 experiment
21
The upper stratospheric temperature
0.4
02Aug2003 00UTC
1
2
Zonal Mean
5
10
1
20
Test (Jul2002)
2
hPa
5
0.4
20
0.4
2
3
2
5
7
Cntl
10
20
100
90N
90S
22
Total Precipitable Water (TPW) Comparison
between Test/Cntl guess and SSM/I (1)
Test SSMI TPW
Test TPW
Cntl TPW
Cntl SSMI TPW
Test-Cntl
SSMI TPW
23
TPW Comparison between Test/Cntl guess and SSM/I
(1)
24
Forecast Scores Anomaly Correlation 500Z
Jul2002
Dec2001
Global
NH
Global
NH
SH
Trop
SH
Trop
25
Forecast Scores RMSE 500Z
Jul2002
Dec2001
Global
NH
SH
Trop
26
Forecast Score BIAS 500Z
Jul2002
Dec2001
Global
NH
Global
NH
SH
SH
Trop
Trop
27
Zonal Mean of RMSE 500Z Test-CntlJul2002
1-day fcst 5-day
fcst
Test Cntl
Test Cntl
Test - Cntl
Test - Cntl
90S
90N
90S
90N
28
Mean difference between Test and Cntl NH 500Z
RMSE for 1-day forecast
Test
Cntl
Jul2002
Test-Cntl
1-day fcst
29
Mean difference between Test and Cntl NH 500Z
RMSE for 5-day forecast
Test
Cntl
Jul2002
Test-Cntl
5-day fcst
30
Forecast Score RMSE/BIAS 850T
Jul2002
RMSE
BIAS
Global
NH
Global
NH
SH
Trop
SH
Trop
31
RMSE 850T Test-Cntl FT1 and FT5

• 1-day fcst 5-day
  fcst

32
Mean difference between Typhoon forecast and best
track
33
Summary

• JMA has been developing the ATOVS radiance
  assimilation in the global 3DVar data
  assimilation system.
• Cycle experiments showed improvements in analysis
  and forecast in terms of the analyzed temperature
  and moisture, the stratospheric structure and
  mean forecast skill, Especially short term
  forecasts are greatly improved.
• JMA implemented ATOVS radiance assimilation
  operationally since 28 May 2003.

34
Problem and Plan

• The bias correction scheme does not work well,
  especially for the higher latitude in the SH,
  where there are few RAOB.
• Improvement of the QC and thinning procedures.
  Especially the investigation is under way that
  either clear or cloudy radiance should be used
  favorably.
• Assimilate Level 1B data, instead of Level 1D
  data, which vanish complicated errors due to
  mapping.
• Assimilation of radiances from Advanced sounder,
  MW imagers and geostationary satellite imagers.

35
Thanks
36
Improvement ratio for Z500 RMSE
better
Better

• (RMSECntl RMSETest)/ RMSEcntl
• refQuikSCAT/AMW impacts

QuikSCAT
better
METEOSAT-AMW (QI)
37
Improvement Ratio for 500Z AC

• Not draw for AC lt 60

better
better