1. Assessment of the Potential of Observations from TES
to Constrain Emissions of Carbon Monoxide
Dylan B. A. Jones, Paul I. Palmer, Daniel J. Jacob
Division of Engineering and Applied Sciences,
and Department of Earth and Planetary Sciences,
Harvard University, Cambridge, MA
Kevin W. Bowman, John Worden
California Institute of Technology, Jet Propulsion Laboratory, Pasadena, CA
Ross N. Hoffman
Atmospheric and Environmental Research, Inc.
Lexington, MA
November 19, 2002

2. Tropospheric Emission Spectrometer (TES)
an infrared, high resolution Fourier spectrometer
covers the spectra range cm-1 ( mm)
has 2 viewing modes: a nadir view and a limb view
Products
Nadir
Limb
Temperature X
Surface temp.
Land surf. emissivity O3 CO
H2O
CH4 NO
HNO3
will be launched in early 2004
makes orbits per day
repeats its orbit every 16 days (or 233 orbits)

3. Apply averaging kernels for TES to simulate nadir retrievals of CO
OBJECTIVE: Determine whether nadir observations of CO from TES will have enough information to reduce uncertainties in our estimates of regional sources of CO
APPROACH:
Assume that we know
the true sources of CO
Use GEOS-CHEM 3-D model to simulate “true” CO concentration field
Apply averaging kernels for TES to simulate nadir retrievals of CO
Obtain a posteriori sources and errors;
How successful are we at finding the true source and reducing the error?
Make a priori
estimate of CO sources
by applying errors
to the “true” source
Use an optimal
estimation
inverse method

4. Source Distribution Other Sources: oxidation of methane and other VOCs
Fossil Fuels
Biofuels
Biomass Burning
Other Sources: oxidation of methane and other VOCs
Main sink: reaction with atmospheric OH

5. Inversion Analysis State Vector
EUFF
NAFF
ASFF
ASBB
AFBB
CHEM
SABB
RWFF
RWBB
FF = Fossil Fuel + Biofuel
All sources include contributions from oxidation of VOCs
OH is specified
Use a “tagged CO” method to estimate contribution from each source
Use inverse method to solve for annually-averaged emissions
“True” Emissions (Tg/yr)
NAFF: SABB:
EUFF: RWBB: 98.0
ASFF: RWFF:
ASBB: CHEM:
AFBB:
Total = 2270 Tg/yr

6. Modeled CO Level 13 (6 km)
15 Mar. 2001, 0 GMT
North American fossil fuel CO tracer

7. Simulating the Data
Sample the synthetic atmosphere along the orbit of TES
Assume that the TES nadir footprint of 8 x 5 km is representative of the entire GEOS-CHEM 2º x 2.5º grid box
We consider data only between the equator and 60ºN
CO at hPa for March 15, 2001

8. Simulating the Data
The retrieved profile is: yret = ya + A(F(xt) – ya)
ya = a priori profile
xt = true emissions
yt = F(xt)
A = averaging kernels
F(x) = GEOS-CHEM
P > 250 mb
yret ya

9. Methodology xa = a priori estimate of the CO sources (state vector)
Minimize a maximum a posteriori cost function
Gauss-Newton iterative method
xa = a priori estimate of the CO sources (state vector)
xi = estimate of the state vector for the ith iteration
yret = retrieved profile
ymod = forward model simulation of xi
K = Jacobian (from tagged CO method)
Ŝ = a posteriori error covariance
Sx = error covariance of sources
Sy = error covariance of observations
= instrument error (Si) + model error (Sm) representativeness error (Sr)

10. } The Forward Model Gein = instrument noise
We apply the same ya and A used for the retrieved profile to GEOS-CHEM (forward model) }
The retrieved profile and the forward model must be consistent
Gein = instrument noise
G = gain matrix associated with A
ei = NESR of instrument Si = G<(ein)(ein)T>GT
n = Gaussian white noise with unity variance
sn = model noise based on variance of (Sm + Sr)
(Sm + Sr) = s2I

11. Constructing the Covariance Matrix for the Model Error
The NMC method (Parrish and Derber, 1992):
assume that the difference between forecasts on different timescales, valid at the same time, is representative of the forecast error structure
Assimilation Model
t0+24hr
forecast 2 t0
forecast 1
t0+48hr
assume that the dominant error structures for the 24-hr period are representative of the errors in the assimilation

12. Forecast Error
root mean square error based on the difference of 89 pairs of 48-hr and 24-hr forecasts of CO during TRACE-P, Feb-April 2001
Model Level 16 (8 km)
CO Mixing Ratio (ppb)

13. Scaling the Forecast Errors
Compare model with observations of CO from TRACE-P and estimate the relative error
Assume mean bias in relative error is due to emission errors
Remove mean bias and assume the residual relative error (RRE) is due to transport

14. Scaling the Forecast Errors
Compare model with observations of CO from TRACE-P and estimate the relative error
Assume mean bias in relative error is due to emission errors
Remove mean bias and assume the residual relative error (RRE) is due to transport
Mean forecast error
RRE
RRE  2-3 x mean forecast error

15. Model Error (15 March 2001, 0 GMT)
Model Level 16 (8 km)
Relative error
Model Level 8 (1.5 km)
Relative error

16. 2x2.5o cell Representativeness Error (Sr) TRACE-P GEOS-CHEM
Use TRACE-P aircraft observations to estimate the subgrid-scale variability of CO
GEOS-CHEM
2x2.5o cell
TRACE-P
Estimate 1s about 2x2.5o mean value of TRACE-P observations
sr = 5%

17. Assumptions
Assume that sources are uncorrelated with uncertainty (Sx)jj = 50% (for the chemistry source we assume 25%)
Assume that the transport errors are spatially uncorrelated
Cloud free conditions
A posteriori results are based on statistics from an ensemble of 100 realizations of the instrument noise and model noise

18. Inversion Results (5 days of TES data, Mar. 10-15th, 2001)
a priori
a posteriori
true
Model successfully retrieves the true emissions for all sources.
Caveat: perfect Gaussian error statistics with zero bias

19. Resolution Matrix (KTSy-1K + Sx-1)-1KTSy-1K Sensitivity
of a posteriori
solution to
true state

20. How well do we constrain the sources with only upper tropospheric (P < 500 hPa) observations?
Good constraint (because of large volume of data), but errors are larger  lower tropospheric retrievals provide useful additional information

21. Conclusions
TES retrievals of CO have the potential to constrain regional sources of CO, but proper error characterization will be crucial – particularly model error; real data will not have unbiased Gaussian error statistics which will reduce the information we can extract
Upper tropospheric retrievals alone may provide considerable information to constrain the sources
Since most of the CO information is in the lower troposphere, TES retrievals below 500 hPa will be useful in constraining the sources, despite the instrument’s reduced sensitivity in the lower troposphere