observational data screening technique using transport model and inverse model in estimating co2...

OBSERVATIONAL DATA SCREENING OBSERVATIONAL DATA SCREENING TECHNIQUE USING TRANSPORT TECHNIQUE USING TRANSPORT MODEL AND INVERSE MODEL IN MODEL AND INVERSE MODEL IN ESTIMATING CO2 FLUX HISTORYESTIMATING CO2 FLUX HISTORY

Takashi MAKI ,Kazumi KAMIDE and Yukitomo Takashi MAKI ,Kazumi KAMIDE and Yukitomo TSUTSUMITSUTSUMI

Japan Meteorological AgencyJapan Meteorological Agency

TransCom 3 workshop in Paris, 15TransCom 3 workshop in Paris, 15 thth June 2005 June 2005

OutlineOutline

1. We have developed a new data selection 1. We have developed a new data selection technique which can treat technique which can treat raw (not smoothed) raw (not smoothed) observational dataobservational data in inter-annual inversion. in inter-annual inversion.

2. We have determined proper threshold value to 2. We have determined proper threshold value to data selection in inversion.data selection in inversion.

3. We have performed several sensitivity tests to 3. We have performed several sensitivity tests to evaluate this technique. evaluate this technique.

BackgroundBackgroundWe should use raw (not We should use raw (not smoothed) observational smoothed) observational data when we try to estimate data when we try to estimate CO2 flux.CO2 flux.

However, the representatives However, the representatives of each observational data of each observational data (not site) is different and the (not site) is different and the observational network observational network changes year by year.changes year by year.

We should treat each We should treat each observational data in observational data in accordance with their accordance with their representatives.representatives.

Monthly mean CO2 concentration reported to World Data Centre for Greenhouse Gases (WDCGG)

Problems in using raw observational dataProblems in using raw observational data

1. The latest data are not reported

2. The lack of observational data

3. Affected by local sources??

4. We have to treat observational data as their representative.

How to resolve these problemsHow to resolve these problemsThe latest data are not reported, the lack of observational data.

Affected by local sources??

Creating dummy data (given large uncertainties and reject from analysis in actual) by smoothing, interpolating and extrapolating.

Each observational data are selected by using inverse model.

We have to treat observational data as their representative.

Data uncertainty (representative) are determined by the difference between raw data and smoothed data (as in TC3L2).

Inversion setupInversion setupTranspoert model JMACDTM, as in TC3L2Meteorological condition JMA analysis (1997), as in

TC3L2Observational data WDCGG monthly mean (1988-

2003), 133 sites (surface + JAL)Data uncertainty Standard deviation of difference

between raw and smoothed dataBackgroud Flux Fossil fuel (1990,1995), CASA,

Takahashi, as in TC3L2*Prior flux As in TC3L3(Houweling et al)Prior flux uncertainty As in TC3L3(Houweling et al)Inversion code As in TC3L2(Rayner et al)Data selection Using inversion (next slide)

*We did not use fossil fuel correction as Baker et al 2005.

Data selection by inversionData selection by inversionPrinciple

1. The majority of raw observational data has a sub-continental scale representative.

2. CO2 fluxes are determined by the decision by majority of raw data.

3. Therefore, we can select raw observational data using transport model and inter-annual inversion (IAI).

Merit of this method

1. We can treat raw data without developing a new data selection technique.

2. We could adopt this method to several kind of observational data (surface, upper air and column data).

3. We do not need huge computational resources (we need only a few iteration of IAI).

Data selection procedureData selection procedureMethod

1. We use all raw observational data in IAI at first.

2. We reject observational data which the data mismatch is larger than the threshold value.

3. We use selected observational data in inversion with same condition　 (prior flux and prior flux uncertainty).

4. We repeat process 2 and 3 until we have no rejected data.Ryori case (Threshold value = 1.5ppm)

350

355

360

365

370

375

380

J - 96 J - 97 J - 98 J - 99 J - 00

InversedRaw dataRejected data

How do we determine the threshold value?How do we determine the threshold value?1. In smaller threshold value, we reject many observational data

including back ground data.

2. In larger threshold value, we use many observational data including local representative data.

We have determined the proper threshold value by checking data selection rate of Global Atmosphere Watch (GAW) global stations (back ground site).

Data selection rate in back ground sitesData selection rate in back ground sites

To select 95% raw data in back ground sites, the threshold value (TV) should be larger than 2ppm 　 (without Zugspitze). We choose this TV as the control case.

Threshold value (ppm) 3.0 2.5 2.0 1.5 1.0Alert 1.00 1.00 1.00 0.98 0.93Point Barrow 1.00 1.00 0.97 0.89 0.70Cape Grim 1.00 1.00 1.00 1.00 0.99Izana 1.00 1.00 0.99 0.97 0.87Mace Head 0.98 0.96 0.92 0.83 0.63Mauna Loa 1.00 1.00 1.00 0.99 0.94Minamitorishima 1.00 1.00 1.00 0.99 0.94Pallas-Sodankyla 0.79 0.71 0.69 0.54 0.35Samoa 1.00 1.00 1.00 1.00 0.99Ushaia 1.00 1.00 1.00 1.00 0.97Mt. Waliguan 0.99 0.96 0.91 0.83 0.59Ny Alesund 1.00 0.98 0.94 0.91 0.81Zugspitze 0.62 0.51 0.36 0.23 0.13Other 5 stations 1.00 1.00 1.00 1.00 1.00Average 0.97 0.95 0.93 0.90 0.82Average (ex. Zugspitze) 0.99 0.98 0.97 0.94 0.87

Data selection rate for all raw dataData selection rate for all raw dataSelected and rejected data number

0

20

40

60

80

100

120

1990 1992 1994 1996 1998 2000 2002

Year

Num

ber

of d

ata

Rejected dataSelected data

About 82% of the all input data are selected in control case.

Recently, the data selection rate tend to smaller?

TV Rate3.0ppm 0.8712.5ppm 0.8512.0ppm 0.8241.5ppm 0.7791.0ppm 0.694

Data availability* rate (1990-2002)Data availability* rate (1990-2002)

The site which continues observation longer period could contribute well in this analysis.

*Availability rate = selected data / analysis period (12*years)

Data selection* rate (1990-2002)Data selection* rate (1990-2002)

Remote sites show better selection rate than continental sites.

*Selection rate = selected data / input raw data

Iteration of inversion (TV=2ppm) Iteration of inversion (TV=2ppm)

Fluxes are remarkably changed from 1st to 2nd inversion.

Temprate North America

-2.5

-2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

1990 1992 1994 1996 1998 2000

year

CO

2 Fl

ux (G

tC/y

)

1st2nd3rd4th

Temperate Africa

-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

2.5

1990 1992 1994 1996 1998 2000

year

CO

2 Fl

ux (G

tC/y

)

1st2nd3rd4th

Temperate Asia

-1.5

-1.0

-0.5

0.0

0.5

1.0

1990 1992 1994 1996 1998 2000

year

CO

2 Fl

ux (G

tC/y

)

1st2nd3rd4th

Europe

-3.5

-2.5

-1.5

-0.5

0.5

1.5

2.5

1990 1992 1994 1996 1998 2000

year

CO

2 Fl

ux (G

tC/y

)

1st2nd3rd4th

Sensitivity testsSensitivity testsWe have performed several sensitivity testsWe have performed several sensitivity tests

1. Threshold value (TV) sensitivity1. Threshold value (TV) sensitivity

We have changed TV from 1.0 to 3.0ppm.We have changed TV from 1.0 to 3.0ppm.

2. Observational network sensitivity2. Observational network sensitivity

We have rejected specific site (Crozet, Izana, We have rejected specific site (Crozet, Izana, Mt. Waliguan) which have been shown by K. Mt. Waliguan) which have been shown by K. R. Gurney at last TransCom workshop from R. Gurney at last TransCom workshop from observational network. observational network.

In global scale, each result shows good agreement with each other except restrict selection (threshold value = 1ppm) case.

1990 1995 2000

Baker et al., 2005

Total CO2 Flux

- 6

- 4

- 2

0

2

4

6

1990 1992 1994 1996 1998 2000

GtC

/y

Land_1.0Land_1.5Land_2.0Land_2.5Land_3.0Ocean_1.0Ocean_1.5Ocean_2.0Ocean_2.5Ocean_3.0

Sensitivity to the global CO2 fluxSensitivity to the global CO2 flux

Boreal Asia Flux

- 1.0

- 0.5

0.0

0.5

1.0

1.5

1990 1992 1994 1996 1998 2000

Year

Flu

x (G

tC/y)

TV1.0TV1.5TV2.0TV2.5TV3.0wo_WLG

Temperate Asia Flux

- 2.0

- 1.5

- 1.0

- 0.5

0.0

0.5

1.0

1.5

2.0

2.5

1990 1992 1994 1996 1998 2000

Year

Flu

x (G

tC/y)

TV1.0TV1.5TV2.0TV2.5TV3.0wo_WLG

Sub-continental case (1)Sub-continental case (1)

In Asian region (relatively less constrained region), Mt. Waliguan has a large role.

Europe Flux

- 2.0

- 1.5

- 1.0

- 0.5

0.0

0.5

1990 1992 1994 1996 1998 2000

Year

Flu

x (G

tC/y)

TV1.0TV1.5TV2.0TV2.5TV3.0

Temperate North America Flux

- 2.5

- 2.0

- 1.5

- 1.0

- 0.5

0.0

0.5

1.0

1990 1992 1994 1996 1998 2000

Year

Flu

x (G

tC/y)

TV1.0TV1.5TV2.0TV2.5TV3.0wo_IZO


Relatively many number of observational data region (North America and Europe), loose selection (TV=2.5ppm and 3ppm) tend to enlarge inter-annual CO2 flux variability.

Izana has some roles in Temperate America.


Less constrained region, loose selection tend to enlarge inter-annual CO2 flux variability.

Crozet has some roles in South Africa.

Tropical Africa Flux

- 1.5

- 1.0

- 0.5

0.0

0.5

1.0

1.5

2.0

2.5

1990 1992 1994 1996 1998 2000

Year

Flu

x (G

tC/y)

TV1.0TV1.5TV2.0TV2.5TV3.0wo_CRZ

South Africa Flux

- 1.5

- 1.0

- 0.5

0.0

0.5

1.0

1.5

2.0

2.5

1990 1992 1994 1996 1998 2000

Year

Flu

x (G

tC/y)

TV1.0TV1.5TV2.0TV2.5TV3.0wo_CRZ

Averaged flux uncertainty (1990-2002)Averaged flux uncertainty (1990-2002)

This work could reduce estimated flux uncertainty especially in formerly less-constrained regions.

This selection method could contribute more homogeneous uncertainty in global carbon cycle than T3L2.

Regions T3L3(JMA) This work Regions T3L3(JMA) This workBoreal N. America 0.87 1.00 Boreal Asia 1.26 1.30

Temp. America 1.25 1.29 Temp. Asia 1.31 1.35Trop. S.America 3.42 2.32 Trop. Asia 1.27 0.97

Temp. S. America 2.41 2.04 Australia 0.46 0.46Tropical Africa 2.25 2.11 Europe 0.93 1.02Temp. Africa 2.61 2.10

Threshold value = 2ppm

ConclusionsConclusionsWe have tried to use raw observational data in inter-anual inversio

n (IAI).

We have developed data selection technique using IAI.

In this selection method, we can use 80% of the raw observational data (from WDCGG) in IAI.

In global scale, the difference in threshold value and observational network is small.

In sub-continental scale, some specific site can effect less constrained region. Therefore, we need more observational data especially less constrained regions.

We can reduce estimated uncertainty in using raw data.

We can use raw observational data consistently in IAI!

Future plansFuture plansWe should have good model to data selection1. Upgrade model and region spatial and temporal resolution.2. From off-line model to on-line model?

We should determine threshold value by statistical analysis1. Calculate chi-square (selected data).2. 2ppm is too large (comparing with Baker et al) ??

We should know the reason which makes the difference between this work and T3L2.

1. Inversion set-up (prior flux uncertainties)?2. Observational data network?

TV χ 2(All) χ 2(Selected)3.0 1.300 1.5172.5 1.191 1.4252.0 1.063 1.3201.5 0.891 1.1761.0 0.631 0.941

Acknowledgement and ReferencesAcknowledgement and ReferencesAcknowledgementWe thank to TransCom project for providing us the protocol, analy

sis, inversion codes and many products.We also thank to WDCGG data contributors. (http://gaw.kishou.g

o.jp:/wdcgg/pdf_contributors.php).ReferencesGurney, K.R., R.M. Law, A.S. Denning, P.J. Rayner, B. Pak, and t

he TransCom 3 L2 modelers, “Transcom 3 Inversion Intercomparison: Control results for the estimation of seasonal carbon sources and sinks”, Global Biogeochemical Cycles,18, GB1010, doi:10.1029/2003GB002111, 2004.

Baker, D. F., R. M. Law, K. R. Gurney, P. Rayner, P. Peylin, A. S. Denning, et al., “TransCom 3 inversion intercomparison: Impact of transport model errors on the interannual variability of regional CO2 fluxes, 1988-2003”, Global Biogeochemical Cycles, 2005, in review.

Appendix 1 : WDC* data policyAppendix 1 : WDC* data policyFrom Minutes of GAW From Minutes of GAW WWorld orld DData ata CCentre Managers entre Managers

Meeting, March 2005.Meeting, March 2005.WDCGG (JMA) proposal - Requirement to acknowledge data cont

ributors with co-authorship is dis-incentive where large scale use of data made in modeling study. Therefore either acknowledge data centre, or have online acknowledgement board.

*WDC: World Data Centre, AREP: Atmospheric Research and Environment Program, CAS: Commission for Atmospheric Sciences, EPAC: Environmental

Pollution and Atmospheric Chemistry, PC:Precipitation Chemistry

Acknowledgment of data originator is too variable. We forward option 2 of the JMA proposal to AREP and CAS-EPAC together with the recommendation from Albany (WDCPC) for further action. “The WDC managers recommended that the secretariat investigate means of ensuring that Journals require data originators and data centres be properly acknowledged in publications using their data”, with the request for advice and/or a standard acknowledgment/citation for use of WDC data.

Appendix 2 : TransCom continuous Appendix 2 : TransCom continuous experiment (JMA experiment)experiment (JMA experiment)

We have submitted our result at Apr. 2005.The futures of our model (JMACDTM)1. The model calculate vertical mixing coefficient

(cumulus convection, turbulent diffusion and shallow convection) making use of our GCM code.

2. The model does not use PBL height and cloud cover rate explicitly.

3. The model has artificial parameter (minimum diffusion coefficient) in near surface layers.

Thanks to Wouter, Christain, Rachel and Lori for coordinating!!