DARE : Data Assimilation in Readiness for Envisat

William Lahoz and Roger Brugge
Centre for Global Atmospheric Modelling,
Department of meteorology, University of Reading

DARE is a European Union funded concerted action programme among whose objectives are:

to co-ordinate the interaction between European groups in order to build a capability in data assimilation to process Envisat data for the atmosphere, and thereby to improve the quality and cost-effectiveness of the data.

to identify data products that can be produced by data assimilation to facilitate the exploitation of Envisat data by a diverse user community.

Data assimilation is a technique whereby observational data are combined with forecast fields from a numerical model, in order to produce an optimal representation of the evolving state of the atmosphere (or ocean or other system). The technique lies at the heart of present-day weather forecasting and climate research.

Envisat-1, will be launched in May 2000 by the European Space Agency (ESA). Envisat-1 is an advanced polar-orbiting Earth observation satellite which will provide measurements of the atmosphere, ocean, land and ice over a five year period.

The Envisat-1 satellite has a payload that will ensure the continuity of the data measurements of the ESA ERS satellites. The Envisat-1 data will support earth science research and allow monitoring of environmental and climatic change evolution.

The benefits of data assimilation

Information from instruments (e.g. satellite measurements) and short-range forecasts of a numerical model are combined in a manner that allows for the errors in the observations and also those in the forecasts produced by the model.

Error statistics for the model and observations can be used to derive the most probable state of the atmosphere, consistent with both observations and model forecast. The procedure allows data void regions, in both space and time, to be filled. Continual ingestion of the observational data by the model helps to keep the (imperfect) model in a state that is in close agreement with the observations - the model ensures that the data are combined in a manner consistent with the equations that govern the evolution of the model.

Both the model and the measurements can profit from the assimilation procedure. Modelling problems can be identified, as they will manifest themselves in the form of large forecast errors. Similarly, the assimilation may provide information about the quality of the instruments and retrieval code.

The ozone hole of 11 September 1996. Above, the southern hemisphere GOME total ozone data collected over 24 hours, and used in the data assimilation. Right, the assimilated field at 1200 GMT. Blue denotes low ozone values, green the intermediate values, while high ozone values are shaded red. Uncertain model values are not shaded. Note how the non-polar data void regions have been filled.

New methods in data assimilation

Most assimilation techniques analyse measurements sequentially in time. When a measurement is imported into a model, the new (analysed) model field value will, generally, lie somewhere in between the observation and the forecast. After the measurement is assimilated the model field is propagated forwards in time until the next observation, and the procedure is repeated. In this way only measurements prior to the model time are assimilated.

Another approach known as four dimensional variational (4dvar) data assimilation is now increasingly being used. Here, a penalty function that describes the fit of the observations to the model forecast over a time interval [0,T] is minimised. A first-guess model forecast is computed by running the model from the analysis at time 0 to time T. During this run the mismatches between the forecast and all measurements in the interval are recorded. Then a second (`adjoint') model is used which combines all the mismatches in the time interval in order to modify the analysis at time t=0, thereby allowing an improved forecast to be made. Typically, the above procedure is repeated several times to converge the analysed field towards the true state.

The evolution of the model field in the time interval is described purely by the model, and no jumps occur. Note that all measurements at all spatial locations in the interval [0,T] are taken into account in the single analysis. This results in a smooth behaviour of the forecast fields in the [0,T] interval.

Schematic illustration of 4dvar technique for a single iteration. The dashed curve is the first guess forecast, while the broken-arrowed lines show the mismatches between this forecast and (circled) observations. A backward-running `adjoint model' (lower solid curve) incorporates these mismatches (the vertical steps in this curve) to produce a field of corrections at t=0 which are used to modify the analysed state at time 0. This provides the starting point for a forecast (solid curve) that smoothly includes information from all the observations over the time interval. For instance the first measurement would suggest a positive correction, but the actual correction made is negative, once measurements at later times are incorporated.

3dvar may be considered as the limiting case of 4dvar when there is no time-dependency in the data. Advantages of 3dvar over other simpler (or `more traditional') sequential techniques are:

3dvar does not require any data selection; it solves a global minimisation problem, using all the observations,

it allows for easier assimilation of satellite-measured radiances without the need for algorithms to retrieve, for example, temperature or wind.

Perceived advantages of 4dvar over 3dvar are:

a large number of observations can be analysed simultaneously,

additional information from future observations is made available to the analysis,

observations of one quantity will influence other quantities via the model (e.g. ozone observations may have a positive impact on lower stratospheric winds).


Much of the illustrative information on data assimilation used here was taken from Variational data assimilation: How to extract more information from GOME total ozone data by H.J. Eskes, A.J.M. Piters, P.F. Levelt, M.A.F. Allaart, and H. Kelder of KNMI, which is available at http://www.knmi.nl/~eskes/eoq/EOQ_article.html.


Sources of instrumental research satellite data for data assimilation

NASA Upper Atmosphere Research Satellite (UARS):
Microwave Limb Sounder (MLS) - ozone, temperature
Improved Stratospheric and Mesospheric Sounder (ISAMS) - temperature
High Resolution Doppler Interferometer (HRDI) - winds

Global Ozone Monitoring Experiment (GOME) - ozone

ESA Envisat:
Global Ozone Monitoring by Occultation of Stars (GOMOS)
Michelson Interferometer for Passive Atmospheric Sounding (MIPAS)
SCanning Imaging Absorption spectroMeter for Atmospheric CartograpHY (SCIAMACHY)
- these three instruments give data on ozone, temperature and atmospheric chemical tracer constituents.

NASA Earth Observing System, EOS-CHEM:
Earth Observing System MLS:
- these instruments give data on ozone, temperature and atmospheric chemical tracer constituents.

Related scientific projects funded by the European Union

MSDOL - Monitoring of Stratospheric Depletion of Ozone Layer
SODA - Studies of Ozone Distribution based on Assimilated satellite measurements

Some Websites pertinent to DARE

ESA ESTEC: http://www.estec.esa.nl
ESA's Envisat programme: http://envisat.estec.esa.nl
NASA's EOS programme: http://eos-chem.gsfc.nasa.gov
UARS: http://umpgal.gsfc.nasa.gov/uars-science.html
SODA: http://www.knmi.nl/soda/

DARE Group activities and contact names

Coordinators: William Lahoz , Alan O'Neill

NERC Centre for Global Atmospheric Modelling,
The University of Reading, UK

Using the UKMO data assimilation system for assimilation of temperature and ozone from the UARS MLS instrument.
William Lahoz , http://ugamp.nerc.ac.uk/cgam/cgam.htm

U. K. Meteorological Office, Bracknell,
UK UARS assimilations (temperature, ozone and winds).
Partner in the SODA project.
Andrew Lorenc , http://www.meto.gov.uk

Université Pierre-et-Marie Curie, Paris, France
Chemical data assimilation (stratospheric profiles, tropospheric columns) with chemistry transport models and trajectory models; GOMOS project investigator.
A partner in the MSDOL project.
Gérard Megie , http://www.ipsl.jussieu.fr

Koninklijk Nederlands Meteorologisch Instituut, De Bilt,
The Netherlands

Ozone chemical data assimilation with a global circulation model and a chemistry transport model.
SCIAMACHY instrument principal investigator.
Coordinator of the SODA project.
Hennie Kelder , http://www.knmi.nl/onderzk/atmosam

Institut für Geophysik und Meteorologie der Universität zu Köln,

4dvar chemical data assimilation with a chemistry transport model.
Hendrik Elbern , http://www.uni-koeln.de/math-nat-fak/geomet/index.html

For further details visit the DARE website at http://ugamp.nerc.ac.uk/~wal/dare.html