1. Adjoint inversion of CO sources using combined MOPITT, SCIAMACHY and AIRS CO columns
Monika Kopacz, Daniel Jacob, Jenny Fisher, Meghan Purdy
Michael Buchwitz, Iryna Khlystova, John Burrows, (SCIA Bremen) Annemieke Gloudemans, Jos de Laat (SCIA SRON), W. Wallace McMillan (AIRS)
COSPAR
Montreal, July 18, 2008

2. Need better understanding of CO sources
tracer of combustion pollution
precursor to tropospheric O3 (smog)
correlations between CO and CO2 can help improve CO2 flux estimates
Suntharalingam et al. [2004]
IPCC [2007]
CO emissions contribute to radiative forcing (indirect greenhouse gas)

3. Observational constraints on CO sources
Top-down constraints
Bottom-up emission estimates
Chemical Transport Model (CTM)
model
EMISSIONS
CONCENTRATIONS ?=
observations
Inverse model

4. Satellite instruments providing CO data
SCIAMACHY
AIRS
MOPITT
100
200
200
200
400
400
400
Pressure (mb)
600
600
600
800
800
800
1000
1000
1000
Averaging kernels
Averaging kernels
Averaging kernels
validated data product (~5% high bias)
sensitive throughout the column
extremely dense coverage
advantages
v7.4 (SRON retrieval), v0.6 (Bremen retrieval)
v5 retrieval
v5 retrieval

5. satellite CO column data
MOPITT: 1999 – 2007, SCIAMACHY: 2003 – 2005, AIRS: 2002-present
May 2004 averages (on 2° x 2.5° resolution)
MOPITT
AIRS
SCIA Bremen
SCIA SRON
1018molec/cm2
CO columns expected to be different due to different vertical sensitivity, but are they consistent?

6. Chemical Transport Model (CTM): the comparison platform
satellite 1
satellite 2
satellite 3
SATELLITE DATA
in situ observations
TRUTH
but very sparse in time and space

7. global Chemical Transport Model, comparison platform
Chemical Transport Model (CTM): the comparison platform
satellite 1
satellite 2
satellite 3
SATELLITE DATA
global Chemical Transport Model, comparison platform
in situ observations
TRUTH
but very sparse in time and space

8. Chemical Transport Model (CTM): the comparison platform
developed at Harvard University
Eulerian model, solves continuity equation for individual gridboxes
GEOS-Chem CTM
contains detailed chemical/aerosol mechanism
horizontal resolution is 2° x 2.5° and vertical resolution is ~ 1 km, temporal resolution is 15 min
uses NASA/Goddard data assimilated meteorology

9. Satellite consistency (via GEOS-Chem CTM):4 3 2 1
r2 = 0.65
r2 = 0.73
GEOS-Chem
May 2004 – May 2005 global daytime columns (averaged to 2°x2.5° resolution)
MOPITT
AIRS 4 3 2 1
Green line: 1:1
Red line: Reduced Major Axis regression
r2 = 0.24
r2 = 0.29
Model/satellite slope:
MOPITT: 0.73
AIRS: 0.76
SCIA Bremen: 0.61
SCIA SRON: 0.63
GEOS-Chem
SCIA Bremen
SCIA SRON
All units: 1018 molec/cm2

10. Satellite consistency (via GEOS-Chem):4 3 2 1
r2 = 0.65
r2 = 0.73
GEOS-Chem
May 2004 – May 2005 global daytime columns (averaged to 2°x2.5° resolution)
MOPITT
AIRS 4 3 2 1
Green line: 1:1
Red line: Reduced Major Axis regression
r2 = 0.24
r2 = 0.56
r2 = 0.34
Model/sat slope:
MOPITT: 0.73
AIRS: 0.76
SCIA Bremen: 0.61
SCIA SRON: 0.63
GEOS-Chem
MONTHLY MEAN
MONTHLY MEAN
SCIA Bremen
SCIA SRON
All units: 1018 molec/cm2

11. Seasonal variability of CO in datasets
Averaged CO columns over N. America
3.0
MOPITT
SCIA Bremen (5d mean)
2.5
Buchwitz et al. [2007]
AIRS
1018 molec/cm2
2.0
MOPITT and AIRS are consistent in seasonal variation, except in spring
1.5
SCIA SRON (5d mean)
1.0
May
July
Sept
Nov
Jan
Mar May

12. Estimating CO sources with an inverse model
GEOS-Chem CO column: F( • )
satellite CO column: y ≠
1018molec/cm2
a priori sources: xa + εa
a posteriori sources
May 2004
satellite data (0 – 60N) : y + εo
model concentrations: F(xa) + εm
observation error: εe
minimize
total mismatch
(model-obs)2
(opt source – a priori source)2

13. Schematic of adjoint source inversion
4°x5° resolution
start
forward model (GEOS-Chem)
MOPITT, SCIA, AIRS obs , y
initialization
MOPITT May 1, 2004
Improved x
end
L-BFGS optimization algorithm
adjoint model
a priori state vector xa
4°x5° resolution

14. CO emission inventories (xa)
Fossil fuel
May 2004
Anthropogenic: EDGAR (global), BRAVO (Mexico), EMEP (Europe), NEI99 (US), Streets (Asia) Biomass burning: GFED2 (global)
Biomass burning
Biofuel
1012 molec/cm2/s
0.50
1.00
1.50
2.00

15. Observational errors RRE = MOPITT SCIAMACHY Bremen AIRS
Relative Residual Error (RRE)
RRE =
MOPITT
Observations used: 0-60N %
SCIAMACHY Bremen
AIRS % %

16. Inversion results: CO emission constraints
min
Do combined datasets provide better CO constraints than individual ones?
m=28,590 obs
m=2,416 obs
m=13,138 obs
a posteriori MOPITT-model mismatch is lower in a joint inversion
SCIAMACHY data does not contribute to model bias improvement

17. Constraints from individual datasets
AIRS
MOPITT
Major features of correction factors consistent in all inversions:
increase in California and N. Mexico
increase in India
increase in E. China
decrease of W. Siberia biomass burning
SCIAMACHY (Bremen)
0.0
0.5
1.0
1.5
2.0
overestimate
underestimate

18. Joint inversion constraints
a priori model bias (w/ MOPITT)
Joint inversion constraints
Joint MOPITT-AIRS inversion
a posteriori model bias (w/ MOPITT)
0.0
0.5
1.0
1.5
2.0
Conclusions:
AIRS provides valuable information over outflow/ocean region
few MOPITT obs. over India and SE Asia  could explain stronger emiss. corrections from AIRS
SCIAMACHY constraints consistent with AIRS and MOPITT
inversion with MOPITT data only
a posteriori model bias (w/ MOPITT)
inversion with MOPITT + AIRS data
-0.50
-0.25
0.0
0.25
0.50

19. Acknowledgements: MOPITT data provided by the MOPITT retrieval team
Funding provided by
END

20. Temporal variations in data and GEOS-Chem CTM
___SCIA SRON
___SCIA Bremen
___AIRS
___MOPITT
Buchwitz et al. [2007]
GEOS-Chem columns over N. America
Lines: GEOS-Chem + averaging kernels
Symbols: daily averaged satellite data
___SCIA SRON
___SCIA Bremen
___AIRS
___MOPITT
___ no AK
averaging kernels decrease the amplitude of seasonal cycle
MOPITT and AIRS appear consistent
May 1, 2004
Nov. 1, 2004
May 1, 2005