Sensitivity of observational dataset to CO2 flux inversion

Takashi Maki, Kazumi Kamide

Atmospheric Environment Division

Japan Meteorological Agency

Backgrounds

Current carbon cycle analysis, GLOBALVIEW (NOAA/CMDL) is the standard dataset.

In GLOBALVIEW, CO2 data are smoothed, interpolated and extrapolated.

In higher resolution analysis in time and space, we tend to need well selected raw data.

We tried to compare estimated flux by smoothed data and raw data.

MethodsUsing smoothed dataset (from WDCGG

analysis or GLOBALVIEW) and raw data (from WDCGG) with the same scenarios.

The scenarios consist of annual mean inversion and time dependent inversion.

We did not analyze absolute value of estimate flux and flux uncertainties, but analyze sensitivity of observational dataset.

WDCGG: World Data Centre for Greenhouse Gases (JMA)

http://gaw.kishou.go.jp

Our Model (JMA CDTM)

Off-line transport model based upon JMA operational Global spectral model (GSM9603).

Resolution 2.5 x 2.5 deg (Horizontal)

32 layers (Surface to 10hPa)

Advection Semi-Lagrangian (Horizontal)

Box-scheme (Vertical)

Diffusion Cumulus convection (Kuo)

Turbulent diffusion (Mellor-Yamada)

Shallow convection (Tiedtke)

Annual mean inversion casePeriods 1997 – 2001(1996 is spin-up)

Winds JMA operational analysis(1997-2001)

Data From WDCGG (selected) monthly data

Smoothed or raw data

Uncertainty Residual from smoothed data

Site Selected by inversion (see later page)

Prior flux As in TransCom 3 Level 1(Background)

Previous year (each region)

Inversion As in TransCom 3 Level 1

Terrestrial and Oceanic regions

From TransCom 3 HP(http://transcom.colostate.edu/TransCom_3/transcom_3.html)

Site Selection We adopt sites where the misfit of inversed data

and observational data is smaller than 2ppm in every year. This is the first selection.

Also we select sites where average misfit is smaller than 0.75ppm from 1997 to 2001.

We continue these selections until we reject no site.

From these selections, we can select sites where there are small effect of anthropogenic and local sources.

Finally, we choose the same 71 sites (from 91 sites) for annual mean scenarios. 　

Selected sites (Organization)

Organizations sites Organizations sites

NOAA/CMDL 41 CAMM 1

MRI 13 INM 1

CSIRO 6 ISAC 1

JMA 3 NIPR 1

NIES 2 SAWS 1

UBA 1

Thank you very much for submitting data to WDCGG!

Selected sites

We can make use of higher altitude (Tokyo – Sydney) data!

Our model can represent vertical profile of CO2 concentrations.

Selected sites Region N Constrain Region N Constrain

Bor. N. America 1 1.25 N. Pacific 15* 28.14 Temp. N. America 3 3.60 W. Pacific 11* 26.20 Trop. America 0 0.00 E. Pacific 3 5.75 South America 1 2.75 S. Pacific 4 9.34 Tropical Africa 1 3.82 Northern Ocean 4 7.06 S. Africa 0 0.00 N. Atlantic 4 6.52 Boreal Eurasia 0 0.00 Tropical Atlantic 2 5.76 Temp. Eurasia 3 2.75 S. Atlantic 1 4.31 S.E. Asia 1* 3.82 Southern Ocean 8** 45.67 Australasia 4* 2.75 Trop. Indian Ocean 1 2.13 Europe 3 3.82 S. Indian Ocean 1 6.85

Constrain is defined as sum of (1/uncertainty) in the region.* : contain aircraft data, **:contain Antarctic sites

Estimated flux from last year (Raw)

In 1998, Tropical land regions are remarkable source of CO2.

We analyze tropical land flux in our presentation.

Difference between smoothed and raw data (annual mean)

Largest standard deviation from 1997 – 2001.

Station Region Unit : ppm

1 Tae-ahn Peninsula Temp. Eurasia 0.272

2 Cape Ferguson Australasia 0.228

3 Pacific Ocean(5N) W. Pacific 0.213

4 Shemya Island N. Pacific 0.190

5 Pacific Ocean(20N) N. Pacific 0.180

6 Estevan Point Temp. N. America 0.173

7 Schauinsland Europe 0.171

8 Pacific Ocean(15N) N. Pacific 0.171

The difference appears relatively constrained regions!

Estimated flux shows some difference between smoothed and raw data (see later page).

Estimated flux uncertainties are completely same in all regions (Data uncertainties are same)!

Difference of estimated flux between smoothed and raw data

Flux uncertainty

Model Transport

Prior flux uncertainty

Data uncertainty

Estimated flux variability

Smoothed analysis tend to show larger inter-annual variability!

Annual estimated flux

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1997 1998 1999 2000 2001

Year

GtC

/y

L03_RawL05_RawL06_RawL08_RawL09_RawL03_SmoothedL05_SmoothedL06_SmoothedL08_SmoothedL09_Smoothed

Standard deviation of estimated flux

from 1997 to 2001.

Smoothed analysis tend to show larger inter-annual variability!

1997-2001, Unit: GtC/y Smoothed Raw

L03: Tropical America 0.355 0.301

L05: Tropical Africa 0.247 0.181

L06: South Africa 0.240 0.173

L08: Temperate Eurasia 0.079 0.064

L09: Southeast Asia 0.320 0.312

Average of all regions 0.133 0.122

Estimated flux difference in each region

Averaging from 1997 – 2001. Unit is GtC/y.Red regions are larger than (average + 1 sigma) of all regions.

Smoothed-Raw St. Dev. Smoothed-Raw St. Dev.

L01 0.005 0.010 O01 0.004 0.019

L02 -0.055 0.018 O02 0.001 0.006

L03 0.027 0.067 O03 -0.003 0.015

L04 0.105 0.027 O04 -0.053 0.021

L05 -0.049 0.074 O05 -0.008 0.003

L06 -0.086 0.077 O06 -0.006 0.002

L07 -0.081 0.028 O07 0.011 0.008

L08 0.104 0.029 O08 0.018 0.010

L09 -0.034 0.009 O09 0.002 0.012

L10 -0.003 0.014 O10 -0.013 0.015

L11 0.091 0.031 O11 -0.015 0.008

Average Correlation coefficient 0.978

Summary of annual mean case

We can select site using inverse method.

Sensitivity of dataset is not so large. Average correlation coefficients is about 0.98.

Smoothed data analysis shows larger inter-annual variability than raw data analysis.

Difference between estimated flux by smoothed and raw data appears in less constrained regions and local source dominant regions.

Difference of each site does not always affect the estimated flux in the region.

Time dependent casePeriods 1990 – 2000 (1988, 1989 is spin-up)Winds JMA operational analysis (1997)Data From GLOBALVIEW 2002

(Smoothed)→ As in TransCom 3 Level 3From WDCGG monthly data (Raw)

Uncertainty As in TransCom 3 Level 3Site As in TransCom 3 Level 3 (76 Sites)Flux As in TransCom 3 Level 3Inversion As in TransCom 3 Level 3

Used sites (as in T3 L3) Region N Constrain Region N Constrain

Bor. N. America 2 1.87 N. Pacific 14* 20.41 Temp. N. America 7 5.75 W. Pacific 9* 16.93 Trop. America 0 0.00 E. Pacific 2 3.46 South America 0 0.00 S. Pacific 1 1.89 Tropical Africa 0 0.00 Northern Ocean 7 6.33 S. Africa 0 0.00 N. Atlantic 4 3.99 Boreal Eurasia 0 0.00 Tropical Atlantic 2 3.27 Temp. Eurasia 5 2.97 S. Atlantic 0 0.00S.E. Asia 1* 2.04 Southern Ocean 8** 22.86 Australasia 2 3.74 Trop. Indian Ocean 1 1.27 Europe 6 4.53 S. Indian Ocean 5 13.28

Constrain is defined as sum of (1/uncertainty) in the region.* : contain aircraft data, **: contain Antarctic sites

Difference between smoothed and raw data (monthly mean)

Largest standard deviation sites from 1990 – 2001.

Station Region Unit: ppm

1 Gosan Temp. Eurasia 2.68

2 Schauinsland Europe 2.60

3 Grifton Temp. N. America 2.24

4 Estevan Point Temp. N. America 1.06

5 Plateau Rosa Europe 0.90

6 Baltic Sea Europe 0.89

7 Ulaan Uul Temp. Eurasia 0.80

8 Cold Bay Bor. N. America 0.79

Large difference appears almost in land regions.

Estimated flux variability from climate

Climate : Average from 1990 to 2001 monthly flux.

Raw analysis data tend to show larger seasonal variability.

-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

0.4

J ul-97 Nov-97 Mar-98 J ul-98

L03_SmoothedL05_SmoothedL06_SmoothedL08_SmoothedL09_SmoothedL03_RawL05_RawL06_RawL08_RawL09_Raw

GtC

/Month

Standard deviation of estimated flux from each climate (1990-2000)

Raw data analysis tend to show higher seasonal variability!

1990-2000, Unit: GtC Smoothed Raw

L03: Tropical America 0.056 0.058

L05: Tropical Africa 0.085 0.097

L06: South Africa 0.058 0.061

L08: Temperate Asia 0.103 0.122

L09: Southeast Asia 0.078 0.094

Total flux 0.204 0.250

Standard deviation between fluxes

Less constrained or local source dominant regions tend to differ!Red regions are larger than (average + 1 sigma) of all regions.

Smoothed-Raw Flux unc. Smoothed-Raw Flux unc.

L01 0.022 0.333 O01 0.024 0.309

L02 0.045 0.467 O02 0.017 0.204

L03 0.025 0.711 O03 0.011 0.206

L04 0.013 0.627 O04 0.029 0.401

L05 0.048 0.675 O05 0.007 0.126

L06 0.024 0.724 O06 0.006 0.202

L07 0.054 0.485 O07 0.005 0.202

L08 0.077 0.516 O08 0.005 0.256

L09 0.051 0.384 O09 0.012 0.225

L10 0.012 0.172 O10 0.023 0.309

L11 0.076 0.397 O11 0.012 0.214

Average Correlation coefficient 0.950

Summary of time dependent caseSensitivity of dataset appeared. Average

correlation coefficients is about 0.95.

Smoothed data analysis shows smaller seasonal variability than raw data analysis in time dependent inversion.

Difference between estimated flux by smoothed and raw data appears in less constrained or local source dominant regions.

Summary of our presentationSensitivity of dataset (smoothed or raw) is not so large in

annual mean inversion. The sensitivity become larger in time-dependent inversion.

→　 If we tried to estimate higher resolution in time and space, the importance of data quality become clear.

Difference between estimated flux by smoothed and raw data appears not only local source dominant regions but also less constrained regions .

→　We have to put new observational sites in less constrained region and use good model in order to avoid making less constrained region as source dump.

We have an option to select site using inversion.

Our future plans

Using inter-annual analyzed wind in the time dependent inversion.

Using higher resolution version (If possible, On-line version) of CDTM.

Using more regions than now.

Joint experiment with FRSGC using Earth Simulator (From this year)!

We hope this experiment could contribute TransCom 3 (4?)!

We need more computational resources.

References (Data)GLOBALVIEW-CO2, Cooperative Atmospheric Data Integration Pr

oject - Carbon Dioxide, CD-ROM, NOAA CMDL, Boulder, Colorado, 2002.

WMO WDCGG Data Summary. GAW Data Vol. IV - Greenhouse Gases and Other Atmospheric Gases. WDCGG No. 27. World Meteorological Organization, Global Atmosphere Watch, World Centre for Greenhouse Gases (WDCGG), Japan Meteorological Agency, Tokyo 2003.

Conway, T. J., P. P. Tans, L. S. Waterman, K. W. Thoning, D. R. Kitzis, K. A. Masarie, and N. Zhang, Evidence for interannual variability of the carbon cycle from the National Oceanic and Atmospheric Administration/Climate Monitoring and Diagnostics Laboratory global air sampling network, J. Geophys. Res., 99, 22831-22855, 1994.

References (Data)

Matsueda, H., H. Inoue, Y. Sawa, Y. Tsutsumi, and M. Ishii, Carbon monoxide in the upper troposphere over the western Pacific between 1993 and 1996, J. Geophys. Res., 103, 19093-19110, 1998.

Rayner, P. J., I. G. Enting, R. J. Francey and R. Langenfelds, Reconstructing the recent carbon cycle from atmospheric CO2,d13C and O2/N2 observations, Tellus, 51B, 213-232, 1999.

Watanabe, F., O. Uchino, Y. Joo, M. Aono, K. Higashijima, Y. Hirano, K. Tsuboi, and K. Suda, 2000: Interannual variation of growth rate of atmospheric carbon dioxide concentration observed at the JMA's three monitoring stations: Large increase in concentration of atmospheric carbon dioxide in 1998. J. Meteor. Soc. Japan, 78, 673-682.

References (Inversion)

Enting, I. 2002. Inverse Problems in Atmospheric Constituent Transport. Cambridge University Press, Cambridge, U. K.

Gurney, K., Law, R., Rayner, P., and A.S. Denning, "TransCom 3 Experimental Protocol," Department of Atmospheric Science, Colorado State University, USA, Paper No. 707, 2000.

Tarantola, A. (1987), The least-squares (12-norm) criterion, in Inverse Problem Theory: Methods for Data Fitting and Parameter Estimation, chap. 4, pp. 187– 287, Elsevier Sci., New York.

Thank you very much for providing level 1 – 3 inversion codes!!

Appendix: WDCGG Data policyWMO WDCGG Data

The WDCGG acknowledges the support of the organizations and individual researchers that provide their measurement data for greenhouse and other related gases. Such data contributors should receive fair credit for their work. When you use and publish data and information in the WDCGG's publications or on this web site, please make reference properly to the contributors and data source and inform both WDCGG and the data contributors. If your work substantially depends on the data, it is recommended to contact the data contributors at an early stage to discuss co-authorship and other necessary arrangements. The following is an example of the citation when you need to cite data from the WDCGG website as a reference:

Tsutsumi, Y, M. Yoshida, S. Iwano, O. Yamamoto, M. Kamada, H. Morishita, Atmospheric CO2 monthly mean oncentration, Ryori, WMO WDCGG , JMA, Tokyo, 21 May 2003.