1

Using satellite data in joint CO2-CO inversion to

improve CO2 flux estimates

Helen Wang1,4,

Daniel J. Jacob1, Monika Kopacz1,

Dylan B.A. Jones2, Parvadha Suntharalingam3,

Jenny A. Fisher1

1

1. Harvard University2. University of Toronto3. University of East Anglia4. Harvard-Smithsonian Center For Astrophysics

Submitted to ACP

[Palmer et al., 2006]

Kg/m2

Jan 2006 model column CO2 Jan 2006 model column CO

g/m2

•CO and CO2 share common combustion sources and transport,indicating cross correlation in concentration and error.

•CO has stronger gradient, is more sensitive to transport error, can provide additional constraint.

Why CO2 : CO ?

•CO is relatively easy to measure from space; Multiple validated data sets have been used for CO source inversion.

•Joint CO – CO2 inversion using aircraft data in Asian outflow showed substantial improvement over CO2 – only inversion.

33

Joint CO2 – CO inversion

Coupling between CO2 and CO occurs through off-diagonal elements in error covariance matrices S

2

22

CO CO CO

COco co

S cov( , )S

cov( , ) S

2 2co co co cocov( , ) var( ) var( )r

correlation coefficient•Off diagonal elements increase information content•Negligible correlation in Sa due to CO emission factor uncertainty

Observation Vector: Column

concentration 2

CO

CO

xx

x

State Vector:

Sources

1 1( ) ( ) ( ) ( ) ( )T Ta a aJ x y Kx S y Kx x x S x x

Cost function Observational

error covariance

A priorierror covariance

4

OCO CO2 : important

MOPITT CO: dominant

I R M

Instrumenterror

Representationerror

Modelerror

Observationalerror

Components of observational error

Covaiance strucutre comes from the model error

5

1. Paired model: GEOS4 – GEOS5 (same meteorological year, same sources/sinks)

Calculating model transport error correlation

GEOS-Chem COGEOS4

GEOS5dCO

GEOS-Chem CO2GEOS4

GEOS5dCO2

monthlyerror

correlation

The same can be performed for GEOS3 – GEOS4 pair5

Mr

6

Model error correlation 1:30 PM, no averaging kernel

Large scale error correlation patterns are robust Error correlation is different from concentration correlation

7the same large scale pattern as before

GEOS5 ARCTAS 48h – 24h forecast of column CO and CO2

July 2008

2. Model error correlation from paired forecast (NMC) method

8

Analytical inversion results for Europe with globally uniform model error correlation and no instrument or representation errors

•Substantial improvements when rM > 0.6

Model error correlation rM Model error correlation rM

biosphere

combustion

December 2005

A posteriori CO2 flux uncertaintyA posteriori biosphere -

combustion CO2 flux correlation

•14 days of pseudo column CO and CO2 data from GEOS-Chem 2x2.5 simulation sampled from A-train orbit

•With OCO-like averaging kernels for column CO and CO2

Co

rrel

atio

n c

oef

fici

ent

Co

rrel

ated

in

vers

ion

flu

x u

nce

rtai

nty

/U

nco

rrel

ated

in

vers

ion

un

cert

ain

ty

Results with model error correlation mapand no instrument or representation errors

CombustionBiosphere

1.0

0.6

0.0

0.2

0.8

0.4

US Europe N.Africa S. Africa E.Asia

Dec 2005

Rel

ativ

e a

po

ster

iori

un

cert

ain

ty

Importance of decreasing I R

1.0

0.6

0.0

0.2

0.8

0.4

15 - 55% improvements in a posteriori CO2 flux in winter10 - 30% improvements in summer

July 2005

M