An Optimal Estimation An Optimal Estimation Approach to AIRS CO - - PowerPoint PPT Presentation
An Optimal Estimation An Optimal Estimation Approach to AIRS CO Approach to AIRS CO Retrievals Retrievals Stuart MacCallum Stuart MacCallum Chris Merchant Chris Merchant School of Geosciences School of Geosciences The University of
s.maccallum@ed.ac.uk
xa = a-priori estimate of CO profile (US Standard profile)
Sa = a-priori covariance matrix (Modelled CO)
K = CO Jacobians (From forward model, RTTOV)
Sε = error covariance matrix (Measurement errors)
y = observed radiances (Radiances observed by AIRS)
xl = linearisation profile (Most recent estimation of CO in the iterative scheme)
yl = simulated radiances for (Modelled radiances for our most linearisation profile recent estimate of CO)
a l l T a T a a
1
Maximum a-posteriori (MAP)1
Gauss-Newton iterative scheme
Forward Model Forward Model – – RTTOV RTTOV
NWP SAF
NWP SAF
Comparable with RFM
Comparable with RFM
http://www.metoffice.gov.uk/
http://www.metoffice.gov.uk/ research/interproj/nwpsaf/ research/interproj/nwpsaf/
Prior covariance, Prior covariance, S Sa
a STOCHEM + manual adjustment
STOCHEM + manual adjustment
Prior CO, Prior CO, x xa
a AFGL US standard
AFGL US standard
AIRS level 2 products AIRS level 2 products
Cloud cleared radiances
Cloud cleared radiances
~50 channels
~50 channels
4.67
4.67 µ µm - CO fundamental m - CO fundamental
T, H
T, H2
2O, T
O, Tsurf
surf,
, ε ε
Errors contribute to
Errors contribute to Sε
Averaging kernels Averaging kernels
Peak around 500 hPa
Peak around 500 hPa
~1 degree of freedom
~1 degree of freedom
Sources
NOAA
NOAA – – CMDL CMDL
http://www.cmdl.noaa.gov/
http://www.cmdl.noaa.gov/
6 sites
6 sites
Sept 02
Sept 02 – – Dec 04 Dec 04
FAAM
FAAM – – EAQUATE EAQUATE
http://www.faam.ac.uk/
http://www.faam.ac.uk/
UK Met Office and NERC
UK Met Office and NERC
Sept 2004
Sept 2004
NASA
NASA – – AVE AVE
http://www.espo.nasa.gov/
http://www.espo.nasa.gov/
3 missions
3 missions
Jan 04, Oct/Nov 04, Jun
Jan 04, Oct/Nov 04, Jun 05 05
Coincidence criteria Coincidence criteria
± 6 hours
± 6 hours
± 1° Latitude/longitude
± 1° Latitude/longitude
Retrievals on 9 levels Retrievals on 9 levels
x xa
a = AFGL US Standard
= AFGL US Standard
x
smooth
a
m
a)
x
xm
m = aircraft CO
= aircraft CO
Black Black – – x xa
a
Green Green
SNM Red Red
Aircraft Orange Orange -Smoothed Aircraft
Dark Blue Dark Blue – – AIRS AIRS Light Blue Light Blue – – Smoothed AIRS Smoothed AIRS
SNM vs Aircraft (smoothed) SNM vs Aircraft (smoothed)
% a-priori < 50% % a-priori < 50%
Negative bias Negative bias
Good Correlation Good Correlation
Larger errors at levels towards Larger errors at levels towards top and bottom of atmosphere top and bottom of atmosphere
Fewer pixels with Fewer pixels with % a-priori < 0.5 % a-priori < 0.5
Larger contribution from Larger contribution from x xa
a at
at these levels these levels
SNM vs AIRS (smoothed) SNM vs AIRS (smoothed)
% a-priori < 50% % a-priori < 50%
High correlation High correlation
Negative bias Negative bias
Same behaviour as Aircraft Same behaviour as Aircraft vs vs SNM at top and bottom of SNM at top and bottom of atmosphere atmosphere
All data included (any % priori) All data included (any % priori)
Aircraft Aircraft vs vs SNM SNM
Good correlation
Good correlation
Problems with high CO values
Problems with high CO values
Aircraft Aircraft vs vs AIRS AIRS
Good correlation
Good correlation
Better with high CO values
Better with high CO values
All data included (any % priori) All data included (any % priori)
Aircraft Aircraft vs vs SNM SNM
Good correlation
Good correlation
Problems with high CO values
Problems with high CO values
AIRS AIRS vs vs SNM SNM
Excellent correlation
Excellent correlation
Similar at high CO values
Similar at high CO values
Smoothed Smoothed
% a-priori < 50%
% a-priori < 50%
Aircraft Aircraft vs vs SNM SNM
Excellent correlation
Excellent correlation
Still problems with high CO
Still problems with high CO values values
AIRS AIRS vs vs SNM SNM
Excellent correlation
Excellent correlation
Problems with high CO values
Problems with high CO values
All levels (not just in common with aircraft) All levels (not just in common with aircraft)
% a-priori < 50% % a-priori < 50%
High Correlation High Correlation
Total Column CO
Good correlation between SNM and Aircraft
Good correlation between SNM and Aircraft
Comparable to AIRS product
Comparable to AIRS product
Good correlation with AIRS product for full column
Good correlation with AIRS product for full column
Negative bias in SNM relative to AIRS
Negative bias in SNM relative to AIRS
CO Profiles
Good correlation on mid-troposphere levels but not so good at
Good correlation on mid-troposphere levels but not so good at top and bottom top and bottom
Heavy reliance on prior at top and bottom of the profile
Heavy reliance on prior at top and bottom of the profile
Larger errors at these levels
Larger errors at these levels
th Aug 2004
http://rapidfire.sci.gsfc.nasa.gov http://mopitt.eos.ucar.edu/mopitt/data/plots/mapsv3.html
SNM AIRS
SNM AIRS All QF < 2 All QF < 2
SNM AIRS
SNM AIRS
SNM AIRS
SNM AIRS
SNM AIRS
SNM AIRS
SNM AIRS
Alternative retrieval scheme Alternative retrieval scheme
Optimal estimation - MAP Optimal estimation - MAP
RTTOV RTTOV
S Sa
a
– – STOCHEM + manual STOCHEM + manual adjustment adjustment
Validation study Validation study
Independent aircraft projects Independent aircraft projects
Variety of locations Variety of locations
Over 100 profiles Over 100 profiles
Good correlation between SNM, Good correlation between SNM, Aircraft and AIRS Aircraft and AIRS
Regional comparison Regional comparison
Broadly similar to AIRS and Broadly similar to AIRS and MOPITT MOPITT
Significant differences between Significant differences between AIRS and SNM in some areas AIRS and SNM in some areas
Quite reasonable error structure Quite reasonable error structure (vertical and horizontal) (vertical and horizontal)
Problems Problems
Large residuals ( Large residuals (y y – – y yl
l) in some
) in some areas. areas.
Low radiance error estimates Low radiance error estimates
Dubious retrievals
Dubious retrievals
Need better understanding of Need better understanding of AIRS L2 products AIRS L2 products
Future Work Future Work
Further comparison with AIRS (V4 Further comparison with AIRS (V4 and V5) and MOPITT and V5) and MOPITT
Explain the differences between Explain the differences between AIRS and SNM AIRS and SNM
Application of retrieval scheme in Application of retrieval scheme in CO case studies. CO case studies.
s.maccallum@ed.ac.uk