In a climate change context, many studies are dealing with the observation of numerous physical parameters such as the soil temperature or the vegetation. The soil moisture is one of the observed variables for monitoring the evolution of the water cycle on Earth. Satellites help to meet this need. Among them, the first spatial mission specially designed for soil moisture monitoring at the global scale is the SMOS mission launched on November 2, 2009. The first part of this thesis deals with the validation of the SMOS soil moisture product. A comparison between soil moistures retrieved from satellite observations (SMOS, AMSR-E, and ASCAT), modeled soil moisture (ECMWF) and measurements is realized over four watersheds in the U.S. In a second phase, the SMOS soil moisture product is evaluated at the global scale by using the triple collocation method. The global SMOS error map is then linked to different parameters by using the methods of ANOVA and CART to investigate its origin. Thereafter, the SMOS soil moisture product is put into a temporal context. Two statistical methods are presented in order to link the SMOS and the AMSR-E soil moisture datasets over the four watersheds: CDF matching and the copulas. Finally, the last part comes back to the SMOS algorithm and to his innovations compared to those already existing. A simpler algorithm has been implemented and adapted to the L-band multi-angular SMOS observations. A comparison between the retrieved soil moisture from this simple algorithm and the SMOS product has allowed us to better understand the improvements that SMOS is capable of and how to improve the SMOS product.
