In the troposphere, the microwave electromagnetic radiation interacts with matter according to three processes: absorption by gas molecules and water, scattering due to the deflection of the waves by hydrometeors, and the emission of a natural radiation. My work has a common basis for these processes of interaction of electromagnetic waves with the atmosphere, and relates to two distinct areas: Remote Sensing of the atmosphere and Earth-satellite telecommunications. I developed a unique method to develop neural inverse algorithms for the observation of the atmosphere using passive microwave sensors. My work highlighted the ability of neural networks to retrieve certain atmospheric quantities underlying the physical phenomena involved. An improvement of the overall performance of the algorithms developed to retrieve these atmospheric quantities has been achieved by modelling non-linear relationships that allows this approach to be followed. In a geophysical context, the use of a modular neural architecture, or the use of Kohonen maps, offered multiple perspectives that are highlighted here. The evolution of telecommunications microwave satellite use requires ever increasing frequencies. At frequencies above 20 GHz the atmosphere is causing a degradation of link quality. I will present the measurement campaigns and modeling work performed in connection with the development of new Fade Mitigation Techniques. The originality of my research on the modelling of the propagation channel consists in the use of statistical models of the TARIM GARCH type, which , rather than predicting the average behaviour of the modelled process, focus on the prediction of variability and the probability of extreme behaviours.
