In section 3.3, it is shown that auto-correlation curves are theoretically calculated from the dispersion curves. Classically, obtaining the 
profile at one site is a two-stage processing: derivation of the dispersion curve from the auto-correlation curves with a least-square scheme (e.g. Bettig et al. (2001)) and inversion of the dispersion curve to determine the
profile. Recently, Asten et al. (2004) proposed to merge them into a single inversion based on least-square optimisation (Herrmann (1994)), allowing the determination of 
directly from the auto-correlation curves. The approach proposed here is conceptually the same except that we make use of the neighbourhood algorithm (section 2.3) for the inversion. It allows an exploration of nearly all equivalent minima in terms of the misfit function and thus enables additionally an improved uncertainty analysis when compared to classical linearized inversion schemes (least-squares). Shapiro (1996) showed, that the solutions obtained from classical surface wave inversion schemes are too restrictive and uncertainties are not correctly estimated.
The text and the figures of this section are extracted from a paper we submitted to the Bulletin of Seismological Society of America in October 2004. This is why the reference model utilized here below differs from the one used in other sections.