4D-Var data assimilation of atmospheric CO from infrared ... · 4D-Var Data Assimilation 4-dimensional variational data assimilation is in principle a least-squares fit in 4 dimensions

University of Leeds 2 March 2006

4D-Var data assimilation of atmosphericCO2 from infrared satellite sounders

Richard Engelen European Centre for Medium-Range Weather Forecasts

Thanks to: Soumia Serrar, Yogesh Tiwari, Frédéric Chevallier and many others.


Outline

• COCO project – 1st attempt with a relativelysimple data assimilation system

• GEMS project – towards a full 4-dimensional greenhouse gas dataassimilation system

• Outlook & Conclusions


COCO

COCO (Measuring CO2 from space exploiting plannedmissions 2001 - 2004) was an European Union fundedIntegrated Project (IP) within the Fifth FrameworkProgramme.

The purpose of the COCO project was to takeadvantage of already planned satellite missions todevelop, evaluate and apply methods for the estimationof CO2 column inventories from space and subsequentlyto estimate CO2 emissions and CO2 surface exchangefluxes.


4D-Var Data Assimilation

4-dimensional variational data assimilation is in principle a least-squares fit in 4 dimensions between the predicted state of theatmosphere and the observations.

The adjustment to the predicted state is made at time To, whichensures that the analysis state (4-dimensional) is a model trajectory.





X0





CO2 is added tothe state vectoras a troposphericcolumn amountfor each AIRSobservation.

X0


CO2 column estimates

370 380 1.0 6.0

Mar2003

Mar2004

Sep2003

Mar2003


Comparison with in-situ observations

Japanese flight data kindly provided by H. Matsueda, MRI/JMA

370

380


Comparisons with models


Comparisons with models


GEMS

GEMS (Global and regional Earth-system Monitoringusing Satellite and in-situ data) is an European Unionfunded Integrated Project (IP) within the SixthFramework Programme.

The project will create a new European operationalsystem for global monitoring of atmospheric chemistryand dynamics and an operational system to produceimproved medium-range & short-range air-chemistryforecasts, through much improved exploitation ofsatellite data.


GEMS organisation

Validation

ReactiveGases

GreenhouseGases

AerosolRegional Air

Quality


Development of RTmodels and bias

correction methods

AIRS, IASI,CrIS,

Sciamachy,OCO, GOSATobservations

4D VARdata assimilation

system

Definition ofbackground errorcovariance matrix

CO2 & CH4analysis dataBuilding of surface

fluxparameterization/

model

CO2 & CH4Flux Inversions

Validation

Greenhouse gas activities



In the 4D-Var version, CO2 is added to the state vector X0. Thismeans that only changes to the initial CO2 field can be made to fitthe observations within the assimilation window.

X0

CO2


Ocean

CO2 surface fluxes - climatology


Natural Biosphere



Anthropogenic



Using climatological fluxes (CASA, Takahashi,and Andres) we have made a 2 year run totest the system at resolution T159 (~ 1.125˚).

CO2 in ECMWF forecast model


Using climatological fluxes (CASA, Takahashi,and Andres) we have made a 2 year run totest the system.

CO2 in ECMWF forecast model


ECMWF model compared to surface flasks

Comparisons between CMDL surface flasks and the free-running ECMWF model show good agreement for the north-south gradients.

Southern hemisphere model values are slightly too low(missing biomass burning??)


ECMWF model compared to surface flasks

Comparisons between CMDL surface flasks and the free-running ECMWF model show good agreement for theseasonal cycle.

Northern hemisphere summer model values are slightly toohigh (missing land sink??)


A negative offset and a15h filtering is applied

to observations

SF6 high frequency comparisons


CO2 4D-Var setup

• T159L60 (1.125˚ x 1.125˚ with 60 levels)• 6-hour assimilation window• Background covariance:

Each layer only correlated with 2 layers directlyabove and below

Horizontal correlation length of 500 km Standard deviation of 2 ppmv

• Operational AIRS bias correction• Operational AIRS cloud detection


First CO2 4D-Var analysis results

After 31 days of 4D-Var, theanalysis has increased theglobal mean value as well asthe spatial gradients.

The increments in anyanalysis cycle are within ± 3ppmv.

369

387

-3.1

3.2


Zonal mean CO2 distributions

367

383

60 S 60N

The effect of assimilating AIRSradiances is mainly to increase CO2mixing ratios in the upper troposphereand reduce mixing ratios in the SHstratosphere.

However, a very simple backgrounderror matrix was used!!!

100

1000

0.35

-0.55


CO2 flux inversionsWeekly fluxes

Monthly fluxes

Simulated flux inversions forOCO data show error reductionsbetween 0 and 20 % over theocean and between 10 and 40 %over land.

The difference is caused by thesmall a priori flux errors overocean compared to the landfluxes.

These estimates assume thereare no significant systematicerrors.

Thanks to Frédéric Chevallier


Near-future improvements

• Use of diurnal biosphere fluxes• Possible use of flask optimized fluxes• Better specification of background

covariance matrix• 12 hour assimilation window• Different AIRS channel selection• Use of IASI radiances• Implement CH4


Conclusions

• First relatively simple implementation of CO2variable in operational data assimilation systemproved successful

• Work in progress to build a full 4D-Vargreenhouse gas data assimilation system thatcan combine observations from various satellitesensors to estimate atmospheric CO2

• These 4D atmospheric fields will then hopefullycontribute to a better quantification andunderstanding of the carbon surface fluxes.

Documents

4D-Var data assimilation of atmospheric CO from infrared ... · 4D-Var Data Assimilation 4-dimensional variational data assimilation is in principle a least-squares fit in 4 dimensions