MODEL SIMULATIONS

1
Exploiting observed CO:CO 2 correlations in Asian outflow to invert simultaneously for emissions of CO and CO 2 Observed correlations between trace gases provide constraints to chemical inverse problems but have been largely unexploited. We use correlations between CO and CO 2 in Asian outflow from the NASA TRACE-P aircraft campaign, together with the GEOS-CHEM global 3-D model driven by time-dependent emission inventories for these gases, to invert simultaneously for the emissions of CO and CO 2 using a maximum likelihood technique. We show that these CO:CO 2 correlations provide unique information to distinguish emissions from different countries in eastern Asia and also to constrain source types. MODEL SIMULATIONS MODEL SIMULATIONS 1. Observations of CO and CO 2 The NASA TRACE-P two-aircraft mission, based in Hong Kong and Japan, was conducted in March-April 2001 to characterize Asian outflow over the NW Pacific and relate it quantitatively to sources. It included high-frequency measurements of CO and CO 2 : Paul I. Palmer ([email protected]), Parvadha Suntharalingam, Dylan B. A. Jones, and Daniel J. Jacob Atmospheric Chemistry Modeling Group, Harvard University 5. Modeling correlations between CO and CO 2 sources 2. Forward Modeling Overview The GEOS-CHEM global 3-D model (resolution of 2 o x 2.5 o ) is used to simulate fields of CO and CO 2 using the ‘tagged’ approach. It is also used to determine the magnitude and distribution of OH, the main sink of CO. The following emission inventories are used: CO CO 2 Fossil Fuel Asia: Streets et al., 2002 Global: Logan et al. Asia: Streets et al., 2002 Global: Marland et al., 2001 Biofuel Asian : Streets et al., 2002 Global : Yevich and Logan, 2002 Asian : Streets et al., 2002 Global : Yevich and Logan, 2002 Biomass Burning Asian: Heald et al., 2002 Global: Yevich and Logan 2002 Asian: Heald et al., 2002 Global: Yevich and Logan 2002 Biosphere Duncan et al., 2003 CASA ecosystem model Potter et al., 1993 Ocean Takahashi et al., 1999 3. Inverse Modeling Overview Offshore China Over Japan Slope (> 840 mb) = 22 R 2 = 0.45 Slope (> 840 mb) = 51 R 2 = 0.76 Japan China Correlations between observations of CO and CO 2 during TRACE-P help distinguish airmasses from different countries within Asia: Not accounting for the “missing” sink(s) of CO 2 causes a positive model bias We correct for this bias by calculating the difference between the modeled and observed “background” (defined as the 10th percentile) as a function of latitude, and subtracting these values (4-5 ppm), from the modeled fields. The resulting model/observation comparison is shown: Latitude [deg] CO 2 [ppm] A priori Observation CO [ppb] X s = a posteriori state vector (CO and CO 2 fluxes) X a = a priori state vector (CO and CO 2 fluxes) S a = error covariance of the a priori CO and CO 2 fluxes (including correlations between CO and CO 2 ) K = forward model operator S y = observation error covariance = instrument error + model error (uncoupled) + representation error x s = x a + (K T S y -1 K + S a -1 ) -1 K T S y -1 (y – Kx a ) S S = (K T S y -1 K + S a -1 ) -1 Observation vector y State vector (Emissions x) y = Kx a + Forward model Inverse model 4. Uncoupled Inverse Model Results STATE VECTOR CH = CHINA KR = KOREA JP = JAPAN SEA = SOUTHEAST ASIA ROW = REST OF WORLD BFFF = BIOFUEL + FOSSIL FUEL BB = BIOMASS BURNING BS = NET BIOSPHERE The a posteriori correlation matrix S s provides insight into the interdependency of individual retrieved state vector elements. In an ideal inversion, S s would be the identity matrix. The figure below clearly shows strong a posteriori correlations between Korean and Japanese emissions of CO and between the CHBFFF and CHBS elements of the CO 2 state vector. China Japan Over Japan Offshore China E = A F Activity rate (kg of fuel burned per unit time) Emission factor (g C released/kg fuel burned) Emission of gas X (g C per unit time) 1- uncertainties in activity rates A and emissions factors F are scaled by a population of normally distributed random numbers (mean of zero and unit standard deviation). A large population (10 4 ) of emissions of CO and CO 2 are generated and correlated. Uncertainties assumed for anthropogenic sources of CO are typically >50% of the source (Streets et al, JGR, 2003); and uncertainties for anthropogenic sources of CO 2 are typically ~20% of the source. The biospheric source of CO 2 has an uncertainty of 75% (Gurney et al, GBC, 2003). Correlation between emissions of CO and CO 2 can help separate the anthropogenic and biospheric signals of CO 2 . CO STATE VECTOR CO 2 STATE VECTOR Calculated correlation coefficients r: CHBFF = 0.47; KR = 0.39; JP = 0.39; SEA = 0.41; CHBB=0.30* * Estimated independently of method described above CO A PRIORI A POSTERIORI Sector i E CO = (A i + A,i A,i )(1 – (F CO2,I + F,CO2,i CO2,i )) E CO2 = (A i + A,i A,i )(F CO2,i + F,CO2,i CO2,I ) Sector i (Assuming F CO + F CO2 = 1) CO CO 2 Glen Sachse and Stephanie Vay NASA Langley David Streets and Qingyan Fu Argonne National Lab. 6. Coupled Inverse Model Results Coupling the errors in emissions of CO and CO 2 mainly impact the a posteriori CO 2 state vector (see below), and consequently reduces its uncertainty. (see panel 4 for units) Uncorrela ted Correlate d S s A posteriori uncertainty of uncorrelated and correlated inversions 0.0 % 11.4 % - 9.4% 5.5% -43.7% -5.0% -9.7% 7.0.% -11.4% 6.7% 0.1% 0.9% -2.0% -1.9% -5.8% 0.1% Absolute (red bars) and relative % difference of the state vector to the uncorrelated inversion 7. Future Directions •Better representation of model error, including correlations between the errors in CO and CO 2 , will increase the importance of the correlations between CO and CO 2 in the inversion. This will help decouple CHBFFF and CHBS in the CO 2 state vector. Include Europe and boreal Asia in the CO 2 state vector. This will help decouple eastern CO 2 NET BIOSPHERE ANTHROPOGENIC CO 2

description

Japan. China. Slope (> 840 mb) = 22 R 2 = 0.45. Slope (> 840 mb) = 51 R 2 = 0.76. Offshore China. Over Japan. Glen Sachse and Stephanie Vay NASA Langley. - PowerPoint PPT Presentation

Transcript of MODEL SIMULATIONS

Page 1: MODEL SIMULATIONS

Exploiting observed CO:CO2 correlations in Asian outflow to invert simultaneously

for emissions of CO and CO2

Observed correlations between trace gases provide constraints to chemical inverse problems but have been largely unexploited. We use correlations between CO and CO2 in Asian

outflow from the NASA TRACE-P aircraft campaign, together with the GEOS-CHEM global 3-D model driven by time-dependent emission inventories for these gases, to invert simultaneously for the emissions of CO and CO2 using a maximum likelihood technique. We

show that these CO:CO2 correlations provide unique information to distinguish emissions

from different countries in eastern Asia and also to constrain source types.

MODEL SIMULATIONS

MODEL SIMULATIONS

1. Observations of CO and CO2

The NASA TRACE-P two-aircraft mission, based in Hong Kong and Japan, was conducted in March-April 2001 to characterize Asian outflow over the NW Pacific and relate it quantitatively to sources. It included high-frequency measurements of CO and CO2:

Paul I. Palmer ([email protected]), Parvadha Suntharalingam, Dylan B. A. Jones, and Daniel J. Jacob Atmospheric Chemistry Modeling Group, Harvard University

5. Modeling correlations between CO and CO2 sources

2. Forward Modeling OverviewThe GEOS-CHEM global 3-D model (resolution of 2o x 2.5o) is used to simulate fields of CO and CO2 using the ‘tagged’ approach. It is also used to determine the magnitude and distribution of OH, the main sink of CO. The following emission inventories are used:

CO CO2

Fossil Fuel Asia: Streets et al., 2002

Global: Logan et al.

Asia: Streets et al., 2002

Global: Marland et al., 2001

Biofuel Asian : Streets et al., 2002

Global : Yevich and Logan, 2002

Asian : Streets et al., 2002

Global : Yevich and Logan, 2002

Biomass Burning Asian: Heald et al., 2002

Global: Yevich and Logan 2002

Asian: Heald et al., 2002

Global: Yevich and Logan 2002

Biosphere Duncan et al., 2003 CASA ecosystem model

Potter et al., 1993

Ocean Takahashi et al., 1999

3. Inverse Modeling Overview

Offshore China

Over Japan

Slope (> 840 mb) = 22

R2 = 0.45Slope (> 840 mb) = 51

R2 = 0.76

JapanChina

Correlations between observations of CO and CO2 during TRACE-P help distinguish airmasses from different countries within Asia:

Not accounting for the “missing” sink(s) of CO2 causes a positive model bias We correct for this bias by calculating the difference between the modeled and observed “background” (defined as the 10th percentile) as a function of latitude, and subtracting these values (4-5 ppm), from the modeled fields. The resulting model/observation comparison is shown:

Latitude [deg]

CO

2 [p

pm

]

A priori

Observation

CO

[p

pb

]

Xs = a posteriori state vector (CO and CO2 fluxes)

Xa = a priori state vector (CO and CO2 fluxes)

Sa = error covariance of the a priori CO and CO2 fluxes (including correlations between CO and CO2)

K = forward model operator

Sy = observation error covariance

= instrument error + model error (uncoupled) + representation error

xs = xa + (KTSy-1K + Sa

-1)-1 KTSy-1(y – Kxa)

SS = (KTSy-1K + Sa

-1)-1

Observation vector y

State vector (Emissions x)

y = Kxa +

Forward model

Inverse model

4. Uncoupled Inverse Model Results

STATE VECTOR

CH = CHINA KR = KOREA JP = JAPAN SEA = SOUTHEAST ASIA ROW = REST OF WORLD

BFFF = BIOFUEL + FOSSIL FUEL BB = BIOMASS BURNING BS = NET BIOSPHERE

The a posteriori correlation matrix Ss provides insight into the interdependency of individual retrieved state vector elements. In an ideal inversion, Ss would be the identity matrix. The figure below clearly shows strong a posteriori correlations between Korean and Japanese emissions of CO and between the CHBFFF and CHBS elements of the CO2 state vector.

China Japan

OverJapan

OffshoreChina

E = A F

Activity rate (kg of fuel burned per unit time)

Emission factor (g C released/kg fuel burned)

Emission of gas X (g C per unit time)

1- uncertainties in activity rates A and emissions factors F are scaled by a population of normally distributed random numbers (mean of zero and unit standard deviation). A large population (104) of emissions of CO and CO2 are generated and correlated.

Uncertainties assumed for anthropogenic sources of CO are typically >50% of the source (Streets et al, JGR, 2003); and uncertainties for anthropogenic sources of CO2 are typically ~20% of the source. The biospheric source of CO2 has an uncertainty of 75% (Gurney et al, GBC, 2003).

Correlation between emissions of CO and CO2 can help separate the anthropogenic and biospheric signals of CO2.

CO STATE VECTOR CO2 STATE VECTOR

Calculated correlation coefficients r: CHBFF = 0.47; KR = 0.39; JP = 0.39; SEA = 0.41; CHBB=0.30*

* Estimated independently of method described above

CO

A PRIORI

A POSTERIORI

Sector i

ECO = (Ai + A,iA,i)(1 – (FCO2,I+F,CO2,iCO2,i))

ECO2 = (Ai + A,iA,i)(FCO2,i + F,CO2,iCO2,I ) Sector i

(Assuming FCO + FCO2 = 1)

CO CO2

Glen Sachse and Stephanie Vay NASA Langley

David Streets and Qingyan Fu Argonne National Lab.

6. Coupled Inverse Model Results

Coupling the errors in emissions of CO and CO2 mainly impact the a posteriori CO2 state vector (see below), and consequently reduces its uncertainty.

(see

pan

el 4

for

un

its)

Uncorrelated Correlated

Ss

A posteriori uncertainty of uncorrelated and correlated inversions

0.0%11.4%

-9.4%

5.5%

-43.7%-5.0%

-9.7%

7.0.%

-11.4%

6.7%

0.1%0.9%

-2.0%-1.9%-5.8%

0.1%

Absolute (red bars) and relative % difference of the state vector to the uncorrelated inversion

7. Future Directions•Better representation of model error, including correlations between the errors in CO and CO2, will increase the importance of the correlations between CO and CO2 in the inversion. This will help decouple CHBFFF and CHBS in the CO2 state vector.

•Include Europe and boreal Asia in the CO2 state vector. This will help decouple eastern Asian sources and ROW.

CO2

NET BIOSPHEREANTHROPOGENIC

CO2