For many years, the ARA team has been designing and maintaining retrieval algorithms, based on radiative transfer modelling, for the analysis of satellite observations, at global scale and over long time periods.

Radiative transfer modelling

Forward modelling

Several generations of forward radiative transfer models, in the infrared and in the microwaves, have been developed, from the original line-by-line and layer-by-layer STRANSAC and 4A models to parametric hyper fast models. Each model is regularly validated from in-situ observations collocated with satellite observations. This acquired experience is at the basis of the recent success of the poject of the automatic validation of the radiances measured by the Infrared Imaging Radiometer (IIR) on board CALIPSO (cooperation CNES, ICARE, LATMOS) and will also be transferred to the validation of the radiances measured by IASI (under CNES contract).

The "4A" model is used by numerous international groups (research, industry) and has recently been "labelled" by CNES for IASI further developments. In particular, an operational version of 4A,-4A/OP-, has been developed by the French company Noveltis in cooperation with LMD. Recently, 4A has been coupled with the Discrete Ordinates Radiative Transfer (DISORT), model for the simulation of radiation in scattering medium, thus allowing studies of clouds and aerosols. Also, the domain of application of 4A has been extended to the processing of data from limb viewing sounders as MIPAS/Envisat or ACE-FTS/Scisat.


Inverse modelling

Retrieving climate variables from radiances measured by space borne instruments goes through the inversion of the radiative transfer equation. The inference algorithms developed in the ARA team either rely on methods making a large use of a priori information ( Bayesian system) - this is the case of the Improved Initialization Inversion ( 3I) method -, or on non-linear, non-Gaussian methods as, for example, the neuronal inference methods.

The choice of the proper method actually depends on the characteristics of the variable to be retrieved. Close cooperation with specialized research groups (as, for example Ceremade at Paris-Dauphine University) systematically accompanies the evolution of these methods.


Hyperfast forward radiative models

The launch of high spectral resolution vertical infrared sounders like the Advanced Infrared Sounder (AIRS) on board of EOS-Aqua (May 2002) or of the Infrared Atmospheric Sounder Interferometer (IASI) on board of the Meteorological Operational Satellite METOP (October 2006) have opened promising perspectives for remote sensing applications as the improvement of temperature and water vapor profile retrieval, cloud and surface characteristics retrieval, or retrieval of greenhouse gases (CO2 and CH4 for example). The availability of an accurate forward radiative transfer model is the key to all these applications.

Fast line-by-line models like the 4A model are able to produce accurate results but remain too slow for the treatment of huge amounts of data from these new instruments. In order to fill this gap, we have developed two hyperfast codes devoted to the simulation of the reduced set of 324 AIRS channels distributed by NESDIS. These two models rely on the availability of the TIGR-AIRS dataset of brightness temperatures, transmission functions, temperature and gas mixing ratio analytical Jacobians, calculated for all the atmospheric situations of the TIGR thermodynamic database using our fast line-by-line model 4A. The first model is based on a multilayer perceptron trained using supervised learning techniques on the TIGR-AIRS database as the learning set. The second model is based on thermodynamic profile pattern recognition in the TIGR database and linearization of the radiative transfer equation. Computation time is of the order of 0.02 sec/atmosphere for the neural network approach and of 0.2 sec/atmosphere for the pattern recognition approach. Similar developments are planned for IASI.

Calibration & Validation

To be fully useful for weather, climate and environmental applications, satellite observations must be qualitatively and quantitatively controlled during the instruments lifetime: any radiometric systematic error not identified in the level1 radiances may propagate as errors in the retrieved variables. At LMD, the technique for inter-calibration has been initially developed for the calibration of Meteosat, based on space and time collocations with instruments on the NOAA series (J. Appl. Meteor., vol 21, 1982). Another technique is based on statistical studies of residuals between simulated and observed brightness temperatures. see more...

Statistical inference

Statistical inference consists in extracting regularities (space/time patterns, relationships between variables and samples, physical processes) in a large dataset of samples. Modern statistical tools like Neural Network inference, Independant Component Analysis, Classification and Clustering techniques have been developed to solve many real-world problems related to meteorology or climatology.


Last update : 2011/01/25