The objective of this work is the improvement of existing inversion techniques in the context of ambient vibration methods in order to obtain 
. A special attention has been paid to the reliability of the inverted profiles and to the possibility of including information from other types of experiments in the inversion process.
The derivation of one-dimensional shear-wave velocity profiles from surface wave dispersion curves is a classical inversion problem in geophysics, generally solved using linearized methods (Tarantola (1987); Nolet (1981)). The inversion of dispersion curves is known to be strongly non-linear and is affected by non-uniqueness, i.e. various models may explain the same data set with an equal misfit. Linearized methods behave poorly in such contexts and a Monte Carlo like approach has been chosen here. During this thesis, we have developed a new code using the neighbourhood algorithm (Sambridge (1999a)) for inverting dispersion curves. The software allows the inclusion of prior information on the different parameters and a major effort has been made to optimize the computation time at the different stages of inversion. In particular, we have re-implemented the dispersion curve computation in C++ language using Dunkin's formalism (Dunkin (1965)). The code is tested on synthetic cases as well as on real data sets, combining ambient vibrations and active-source data. In both cases, the role of a prior information for constraining the solution is emphasized. Moreover, specific methods are proposed to invert the auto-correlation curves to obtain directly the ground structure, to identify and to invert higher modes, and to include frequency information measured with the H/V technique.
Aside, a software package has been developed for preparing array campaigns, storing, visualizing and analysing the recorded signals (open source project, GEOPSY). The techniques for processing raw signals were revisited and the corresponding algorithms were implemented in this unified platform dedicated to seismic prospecting.