Climatology and meteorology are mainly based on numerical models, but they also need independent data from in situ measurements or satellite observations. In this thesis, we attempt to optimize the use of the satellite observations in order to globally retrieve atmospheric profiles of temperature and water vapor. Knowledge about the impact of land surfaces on the radiation measured by a satellite is crucial to be able to determine the quality of the retrieved profiles. A Bayesian estimator has been used to invert the radiative equation, leading to a simultaneous retrieval of surface temperature and emissivity in the infrared, based on IASI measurements. An operational algorithm has been built. It has allowed the creation of a surface emissivity and temperature database from 2007 to today. Those surface retrievals have been used in an atmospheric inversion scheme, which led to a global decrease in the error on the retrieval of temperature and water vapor profiles, especially in the troposphere, which is the most important in meteorology. The neural network-based algorithm used for the retrievals needs a representative learning database. To build such datasets, we created a multi-variate sampling method able to compute numerous non-homogeneous variables. Finally, we have shown that the simultaneous use of infrared and microwave observations is a promising way to increase the quality of the satellite retrievals. The synergy between instruments like IASI, AMSU-A and MHS on board MetOp decreases the error of the retrieved atmospheric profiles of temperature and water vapor.
