PSG ASSESSMENT OF ALGORITHMS CrISATMS

1
 PSG ASSESSMENT OF ALGORITHMS - CrIS/ATMS 

• JOEL SUSSKIND
• November 4, 2003 

2
CrIMMS EDR ALGORITHM

• CrIMMS EDR algorithm has been developed and
  tested with simulation data
• NOAA 88 radiosonde data set (no cloud data)
• AIRS December 15, 2001 data set - GCM derived
  fields including clouds
• The same truth data sets were used by the AIRS
  Science Team to generate
• AIRS pre-launch retrieval algorithm
• In simulation AER CrIMMS retrieval algorithm
  performs comparably to AIRS team algorithm
  modified to handle CrIMMS data
• Real data contains complications not found in
  simulation
• Further modifications to AER algorithm will be
  needed to handle real data

3
LIMITATIONS OF SIMULATION EXPERIMENTS AS PERFORMED

• Physics
• Perfect knowledge of physics was assumed
• Non-local thermodynamic equilibrium (non-LTE)
  was ignored
• Knowledge of instrument SRFS was assumed to be
  perfect
• Scenes
• Cloud and surface parameters were model
  generated
• There was no sub-pixel variability in the scenes
• clouds, surface parameters, etc.
• Variability of upper tropospheric water vapor
  and trace gases was
• unrealistic

4
EXPERIENCE OF AIRS SCIENCE TEAM WITH OBSERVED
AIRS DATA

• Differences exist between observed AIRS radiances
  and those computed from truth
• Larger differences exist for AMSU A brightness
  temperature - antenna pattern uncertainty
• Bias correction (tuning) needed for observed
  minus computed brightness temperature
• Residual tuning error must be included in
  channel noise covariance matrix
• More channels affected by non-LTE than previously
  assumed
• Cloud clearing is working well for the most part
  but
• some homogeneous low stratus clouds at night are
  undetected
• weak cirrus cloud emissivity signal is found in
  some clear column radiances
• Surface emissivity variability over land is
  causing some problems
• Rejection tests and thresholds needed to be
  modified
• Eighteen months after launch, AIRS products are
  approaching required
• accuracy
• More research is still needed

5
PREDICTION OF OBSERVED CrIS FROM OBSERVED AIRS

• are determined
  from extensive simulation using AIRS
• retrieved
  parameters as truth scenes
• Method tested successfully by predicting AIRS
  from AIRS
• Two whole days of CrIS data have been predicted -
  Sept. 6, 2002 and Jan. 25, 2003
• Observed AMSU data is appended
• Because observed AIRS data is used, most
  limitations of simulation are accounted for
• Degui Gu at NGST is studying this data set
• His recommendations for needed EDR algorithm
  work are analogous to AIRS experience

6
ADDITIONAL EDRs THAT CrIS COULD PRODUCE

• A number of additional EDRs are being derived
  from AIRS
• CrIS is capable of giving the same products
• Cloud top height and fractional cloud cover
• Useful in its own right for climate studies
• Used by Bob Atlas in data assimilation
  experiments
• Computed OLR
• Helps explain variability of OLR in terms of
  component parts
• Total O3, O3 profile
• Needed to supplement OMPS at night, polar
  winter
• Should use 9.6 mm band - not in current
  algorithm
• Trace gas (CO2, CH4) profiles and possibly
  total CO2 burden
  

7
CrIS SDR ALGORITHM

• SDR algorithm is very complex and needs review by
  science team
• Particular concern exist with regard to
• Adjusting SDRs in different detector positions
  to SDRs with a common ILS
• What is the nature of the error (noise)
  introduced by this
• procedure?
• To what extent does cloud obscuration affect
  ILS and effective SDR noise
• CrIS data predicted from AIRS observations does
  not address these factors
  

8

• BACK UP CHARTS

9
AIRS CHANNEL NOISE REDUCTION

• There are many more AIRS channel
  than independent pieces of information
  
• Use of the whole spectrum predicts more
  accurately than it can be measured
• METHOD
• Simulate noisy and noise free AIRS radiances
  
• Find regression relationship
• 
  
• A is block diagonal - LW, MW, SW
• Use as truth retrieved state (all parameters
  including clouds, trace gases,
• surface emissivity ..) for all accepted
  retrievals for
• September 6, 2002 and January 25, 2003
• Generate coefficients on 140,000 cases - test on
  280,000 cases
• Transformed channel noise covariance matrix
• where is diagonal
  with original noise contains off
  diagonal matrix elements

10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)