Validating the GHG production chain at multiple levels Julia Marshall (MPI-BGC), Richard Engelen (ECMWF), Cyril Crevoisier (LMD), Peter Bergamaschi (JRC),

Validating the GHG production chain at

multiple levels

Julia Marshall (MPI-BGC), Richard Engelen (ECMWF), Cyril Crevoisier (LMD), Peter Bergamaschi (JRC), Frédéric Chevallier (LSCE), Peter Rayner (LSCE), and various data

providers (referenced throughout)

Data flow of the GHG system: CO2

Assimilation into

ECMWF system

Independent

retrievals (ANN)

4D-fields

Flux inversion system

Gridded flux fields

Biosphere

models

AIRS data

AIRS & IASI data

Data flow of the GHG system: CO2

Assimilation into

ECMWF system

Independent

retrievals (ANN)

4D-fields

Flux inversion system

Gridded flux fields

Biosphere

models

AIRS data

AIRS & IASI data

Data flow of the GHG system: CO2

Assimilation into

ECMWF system

Independent

retrievals (ANN)

4D-fields

Flux inversion system

Gridded flux fields

Other satellite data

(e.g. SCIAMACHY)

Surface-based

measurements

Flux towers

Biosphere

models

AIRS data

AIRS & IASI data

4D IFS CO2 fields

independent AIRS CO2

retrievals

gridded flux fields

Flux towers

biosphere model

AIRS data IASI data

IASI CO2

retrievals

SCIAMACHY data

SCIAMACHY CO2

retrievals

independent 4D CO2

fieldssurface-informed

gridded flux fields

surface-based

measurements

prior-informed 4D LMDZ CO2

fields

4D IFS CH4 fields

gridded flux fields

IASI data

IASI CH4

retrievals

SCIAMACHY data

independent SCIAMACHY

CH4

retrievals surface-based

measurements

optimized 4D TM5 CH4

fields

Data flow of the GHG system: CO2

Assimilation into

ECMWF system

Independent

retrievals (ANN)

4D-fields

Flux inversion system

Gridded flux fields

Surface-based

measurements

Flux towers

Biosphere

models

Surface-based

assimilation and

inversion systems

(e.g. CarbonTra

cker)

Other satellite data

(e.g. SCIAMACHY)

AIRS data

AIRS & IASI data

Data flow of the GHG system: CO2

Assimilation into

ECMWF system

Independent

retrievals (ANN)

4D-fields

Flux inversion system

Gridded flux fields

Surface-based

measurements

Flux towers

Biosphere

models

Surface-based

assimilation and

inversion systems

(e.g. CarbonTra

cker)

Other satellite data

(e.g. SCIAMACHY)

AIRS data

AIRS & IASI data

A note on the models considered here

• All data are matched to the gridbox matching the altitude of the measurement, and linearly interpolated in time

Model name

Grid resolution

Timestep

IFS assimilated

1x1 degree 6hr

IFS free-run (CASA fluxes)

1x1 degree 6hr

TM3 4D fields

4x5 degree 6hr

CarbonTracker

4x6 degree 3hr

Surface-based measurement network

• For the purposes here, this includes surface stations, ship-based measurements, aircraft data, and ground-based remote sensing (i.e. FTIR)

• 179 datasets considered at present

Some metrics to be considered:

• Based on VAL scoring document:– Modified normalized mean bias:

– Fractional gross error:

– Correlation coefficient:

• Visualization with Taylor diagrams

A brief note on Taylor diagrams

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

A brief note on Taylor diagrams

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

A brief note on Taylor diagrams

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

A brief note on Taylor diagrams

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

A brief note on Taylor diagrams

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

A brief note on Taylor diagrams

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Some results from validation with station data• Most stations

show reasonable agreement

• Standard deviation tends to be somewhat high, but scattered

Correlation coefficients

• Remote stations generally show good agreement

• Poor correlation over highly variable regions, such as Europe

Comparison for MNM Bias ( )

• Similar pattern of disagreement, showing up to a 10% positive bias over Europe

• Southern hemisphere well-constrained, slightly positive tendency in northern hemisphere

Fractional Gross Error ( )

• Similar pattern of disagreement, showing up to a 10% fractional error

• Again, low error in remote regions

A view of the errors in time and space

A view of the errors in time and space

NH summer

A view of the errors in time and space

NH summer

Increasing with time

Comparison to surface-data constrained assimilation systems:

• CarbonTracker and TM3 are not always independent

• Correlation better, but standard deviation consistently low

Comparison with CarbonTracker: considering subset of 99 independent data sets

• More similar results when comparing only independent stations

Comparison with CarbonTracker: considering only independent flight data

• Yet more comparable when looking at only flight data

Comparison with free run, i.e. the effect of the satellite data (aircraft data only)

• Improvement in variability of the model, if not correlation coefficients

A look at the total column results:

• Northern hemisphere bias seen at Park Falls, but seasonal cycle reproduced well

• Poor agreement with Darwin

R=-.33RMSE=3.6

ppm

R=0.91RMSE=5.9

ppm

Some conclusions:

• IFS 4D fields compare well with remote observations

• Positive bias and higher error seen over highly populated regions with heterogeneous fluxes

• Slight northern-hemisphere high bias, seems related to too weak seasonal cycle

• Trend shows some divergence over time• Performance when considering non-surface

data is comparable to that of an inversion system using only surface-based data

Other activities:• Further comparisons carried out

with CO2 flux output

• Comparison to independent satellite retrievals

• Similar work done for methane validation, which will be briefly discussed in the VAL session tomorrow morning

Data sources:• WMO Global Atmosphere Watch data • CarboEurope IP concentration

measurements, including flights, tall towers, and flasks

• NOAA ESRL tall towers and routine flight data

• flight data over Siberia from Machida et al.• Darwin FTIR:Deutscher et al., (in preparation) • Park Falls FTIR: Washenfelder et al., 2006• CarbonTracker 2008 results provided by

NOAA ESRL, Boulder, Colorado, USA from the website at http://carbontracker.noaa.gov.

• TM3 4D fields: from Christian Rödenbeck