Satellite Bias Correction for CFSRR

1
Satellite Bias Correctionfor CFSRR

• Haixia Liu, Russ Treadon, Robert Kistler, John
  Derber, Suru Saha and Hua-lu Pan
• Nov. 7, 2007
• with input from Jack Woollen, Quanhua Liu,
  Moorthi Shrinivas, Cathy Thiaw and others

2
radiance used
radiance NOT used

(Reference)
3
Whats needed to assimilate radiance?

• CRTM model, CRTM coeff files
• satinfo which contains the usage flag, obs error
  and Instrument/observation characteristics
  (microwave/IR, etc.)
• Radiance obs bias correction is needed.

4
Why bias correction (BC)?

• RTM are subject to systematic errors.
• Radiance obs contain systematic errors.
• Magnitudes of these errors are comparable with
  radiance changes corresponding to typical errors
  in the T from short-range forecast model.
• Try to use radiance to improve the accuracy of
  the T field during DA process.

5
Satellite Bias Correction in GSI
Observation bias
Air mass bias (dependent on atmospheric state)
Constant Zenith angle Cloud liquid water Square
of T lapse rate T lapse rate etc.
is the coefficients of predictors
6
Adaptive BC in GSI
Allow temporal variation of the coefficients
using data assimilation
Variational bias correction
VS
Find the optimal b to minimize the cost function
J through the variational procedure.
7

• 2 BC files are required to run gsi
• Coefficients for predictor part of BC
  (biascr.gdas.)
• Updated within inner loop of analysis
• Slowly evolving angle dependent part of BC
  (satang.gdas.)
• Updated in separate job step following analysis
  using Tb innovation file
• Method
• Start from guess values and run a training
  period. Experiments were conducted to see if this
  procedure can lead to convergence to the
  operational values (truth) after the training
  period.

8
Preliminary Results
AMSU
Blue EXP Red CNTL
9
HIRS
Blue EXP Red CNTL
10
Scan bias and air mass mean bias compensate to
each other. Combine the scan bias and mean term
together to avoid the compensation between them.
Remove predictor zenith angle.
11

• The coefficients for predictor cloud liquid water
  should not start from 0 since this term evolves
  very slowly. Only applied to amsua data available
  after 199810. Will start from average value used
  in operation.

12
GSI codes and scripts ready for CFSRR

• Modify GSI code and analysis script to handle
  pre-NOAA-14 satellites/sensors.
• Modification made in GSI and link to new CRTM
  which considers co2 leaking in cell pressure
  modulator which ssu data suffers.
• GSI is basically frozen for CFSRR now.

13
Generating satinfo historical files

• The satinfo historical files are generated every
  6 hours from 1979 to present. The
  satellite/sensor/channel use flag is turned
  on/off in satinfo files based on the GMAO
  blacklist and data availability at NCEP.
• Amsua_aqua channel 4 is going to be used as QC
  purpose rather than assimilated. Amsua_n16 chn4
  is in monitor mode.
• Questions hirs2_n11 in 2nd period is extended to
  20000426 and ssu_n11 is extended to 20040616
  (GMAO ends using them at Feb 1999 and ERA40 ends
  using them in Sep 1998) ssu-n12 data is going to
  be used extending from 19970530 to the end of
  1998.

14
Bias correction exp list

• For sensors/satellites before NOAA 15, the bias
  correction spin up must be done ahead of CFSRR
  production.
• combine all the spun up BC for all
  channels/sensors/satellites together to formulate
  a huge biascr.gdas. and satang.gdas. file which
  will be used in CFSRR production.

15
satellite sensor Start date End date
tirosn Hirs2 19790101 19800121
tirosn Msu 19790101 19790626
tirosn ssu 19790101 19800129 19800121 19810205
N06 Hirs2 19790713 19830418
N06 Msu 19790701 19851015 19830418 19861118
N06 ssu Same as msu Same as msu
N07 Hirs2 19810824 19850205
N07 Msu 19810801 19850219
N07 ssu Same as msu Same as msu
N08 Hirs2 19830501 19840621
N08 Msu Same as hirs2 Same as hirs2
N08 ssu Same as hirs2 Same as hirs2
N09 Hirs2 19850225 19881102
N09 Msu 19850225 19870308
N09 ssu 19850225 19881102
N10 Hirs2 19861126 19910902
N10 msu Same as hirs2 Same as hirs2
N11 Hirs2 19881108 19941231
N11 Msu Same as hirs2 Same as hirs2
N11 ssu Same as hirs2 Same as hirs2
16
N11 2nd period Hirs2 19970715 19990226
N11 2nd period Msu Same as hirs2 Same as hirs2
N11 2nd period ssu Same as hirs2 Same as hirs2
N12 Hirs2 19910916 19911225
N12 Hirs2 19920308 19930610
N12 Hirs2 19930619 19970530
N12 Msu 19910916 19970524
N14 Hirs2 19950501 20060925
N14 Msu Same as hirs2 Same as hirs2
N14 ssu 19950119 20060505
N15 Hirs3 19981026hl1  20000607
N15 amsua 19981026 21001231
N15 amsub Same as amsua Same as amsua
N16 Hirs3 200101hl2 26 20040521
N16 amsua 20010124 20021121
N16 amsua 20021128 21001231
N16 amsub 20010124 21001231
N17 Hirs3 20020715 21001231
N17 amsua 20020715hl3  20031028
N17 amsub Same as hirs3 Same as hirs3
 hl1 Hirs3_n15 are not used any more by GDAS in
operation.  hl2 Hirs3_n16 are not used any more
by GDAS in operation.  hl3 Amsua_n17 are not
used any more by GDAS in operation.
17
N18 Hirs4 Not use
N18 amsua 20051101 21001231
N18 mhs Same as amsua Same as amsua
Metop-a Hirs4 20070521 21001231
Metop-a amsua Same as hirs4 Same as hirs4
Metop-a mhs Same as hirs4 Same as hirs4
aqua airs 20021012 21001231
aqua amsua Same as hirs4 Same as hirs4
G08 sndr 19940413 20030401
G10 sndr 19970425 20060621
G11 sndr 20060621hl1  20070529
G11 sndrD1 20070529 21001231
G11 sndrD2 20070529 21001231
G11 sndrD3 20070529 21001231
G11 sndrD4 20070529 21001231
G12 sndr 20030401 20070529
G12 sndrD1 20070529 21001231
G12 sndrD2 20070529 21001231
G12 sndrD3 20070529 21001231
G12 sndrD4 20070529 21001231
 hl1 Sndr_g11 are not used any more by GDAS in
operation.
18

• 4 Simultaneous Streams
• Jan 1979 Sep 1986 8 years
• Apr 1986 Oct 1993 7 1/2 years
• Apr 1993 Oct 2000 7 1/2 years
• Apr 2000 Feb 2010 10 years
• 6 month overlap for ocean and land spin ups
• Total of 33 years of Reanalysis

19
The bias correction must be done first for the
beginning of these 4 streams. Thus we have 4 more
exps to spin up BC.
The 3rd priority (2nd closest to the beginning
of the 4 streams) 1981080100 1988110800 199404130
0(goes08) 1995011900 1995050100 2001012400(n16) 20
02071500(n17) ......
The first priority (the beginning of the 4
streams) 1979010100 1986040100 1993040100 2000040
100 The 2nd priority (closest to the beginning
of the 4 streams) 1979070100 1986112500
The first 4 spin up exps start from the beginning
of each stream. Subsequently, we consider to spin
up BC for other sensors/satellites according to
their usage timing.
20
Future work

• 21 satellite bias correction spin up experiments
  should be done by the end of 2007.
• Sensitivity of bias to the atmospheric IC. Start
  from r2 or gdas of the previous cycle?
• Tests on amsua_n15 indicate that the bias
  correction is not very sensitive to the IC (the
  current predictors chosen can identify obs bias
  from NWP model error.)
• Possible issues related to SSU bias correction
  and assimilation.