Developments of the fast RT model OPTRAN for High Spectral Sounder

1
Developments of the fast RT model OPTRANfor
High Spectral Sounder

• Yoshihiko Tahara (UCAR, JMA, NCEP)
• Paul V. Delst (CIMSS, NCEP)
• John C. Derber (NCEP)
• Larry M. McMillin (NESDIS)
• Xiaozhen Xiong (NESDIS)
• Thomas J. Kleespies (NESDIS)

2
OPTRAN Meets Difficulty

• Absorption coefficient prediction
• A huge number of equations
• 900,000 equations for 1000 channels
• Difficult to validate all
• Numerous coefficients
• 41 Mbytes for 1000 channels
• A lot of computer resources
• Difficult to introduce high spectral sensors
• Difficult to keep updated
• Difficult to be more developed
• More absorbers, more predictors, more layers, etc

3
New Prediction Form

• Coefficients traced over 300 layers by
  polynomial
• Coefficients are reduced more than 95.
• 1 additional atmospheric predictor introduced
• No vertical interpolation

4
Weight in Generating Trans Coef

• Impossible to transfer all information of 300
  equations to one
• Weight transfers and preserves the essential
  information.

NOAA/HIRS Ch 3
Error in Abs Coef
Error in Trans
Weight
Trans
5
Atmospheric Predictors

• 17 atmospheric predictors
• 2 new predictors
• Q/T2 for wet continuum
• P1/4 for Ozone

6
Choice of Predictors

• 21778 predictor set possibilities for each gas
  and channel
• Known unstable sets removed in advance
• Stability is monitored by a stability index
• Transmittance prediction error is monitored to
  select the best set.

Validated sets automatically
Less than 24 hrs to find best sets for 500 chs
using 10 CPUs on IBM SP4
7
LBL Transmittances

• UMBC atmospheric profiles (for generating
  coefficients)
• 48 profiles
• 101 levels (0.005 -- 1100 hPa)
• CIMSS atmospheric profiles (for independent
  verification)
• 32 profiles
• 101 levels (0.005 -- 1100 hPa)
• Transmittances (Infrared)
• LBLRTM Ver 6.12 with HITRAN2000 AER updates
• 6 angles for each atmos profile
• Sep 2002 AIRS SRFs for AIRS
• Central frequencies generated based upon the SRFs

8
Comparison to Dependent
S.D.
Bias
9
Comparison to Independent
S.D.
Bias
10
Comparison to Real-time Data
Ch 28 656.1 SD 0.76 MN 0.16
Ch 253 722.1 SD 0.39 MN 0.29 (clr sky)
Ch 1082 1036.5 SD 0.63 MN 0.33 (clr sky)
Ch 1756 1524.4 SD 1.22 MN 0.44 (clr sky)
11
Comparison to Real-time Data
S.D.
Bias
Num
12
Impacts of New OPTRAN on Global Fcst (No AIRS,
Preliminary)
NH 1000 Z
SH 1000 Z
SH 500 Z
NH 500 Z
13
Impacts of New OPTRAN on Global Fcst (No AIRS,
Preliminary)
Trop 850 Vct Wind
Trop 200 Vct Wind
14
Conclusion

• Equations and coefficients are reduced
  significantly.
• Predictor selection and inspection is automated.
• Offline validation shows
• Around 0.1 deg or less S.D. for most channels
• 0.2 0.3 deg S.D. for ozone channels
• Small bias except ozone and WV channels in
  independent val.
• Real-time data comparison shows
• Stable calculation for AIRS 281 subset channels
  except 1
• 0.5 1 deg S.D. for dry and ozone, 0.5 1.3 deg
  S.D. for WV
• Within plus minus 1 deg bias
• Impacts of the new OPTRAN (without AIRS) on the
  NCEP global model are small positive or neutral
  over the extra tropics and neutral over the
  tropics.