Near-surface geophysical imaging and natural hazards : new methodological developments applied to landslide and active fault imaging. ; Imagerie géophysique de proche surface et aléas naturels : nouveaux développements méthodologiques appliqués à l'imagerie des glissements de terrains et failles actives.

Although different, the geological hazards associated with landslides and earthquakes share a common dynamic due to sliding along a weak surface. Characterizing the geometry of these surfaces is an essential prerequisite for estimating sliding rates and, thus, assessing landslide and seismic hazards. Geophysical methods are easy to deploy, non-destructive, and low-cost, providing images at depths of up to several hundred meters. Among these methods, electrical resistivity tomography (ERT) provides images of the subsurface and associated information on lithology, fluid content, and structure. However, interpreting these images can be challenging due to the regularization required to solve the inverse problem, which reduces their resolution and limits their ability to image localized contrasts necessary for defining the geometry of slip surfaces.As part of this thesis, new analysis and inversion tools have been developed to improve the interpretation and "quality" of subsurface images. The PyMERRY post-inversion code calculates ERT profile coverage based on sensitivity and electrode distance. This coverage defines a mask that hides unconstrained areas of the image, considering the resistivity heterogeneities, and limits over-interpretation. PyMERRY also allows for calculating uncertainties on the electrical resistivity of each cell of the resistivity model. The second development is a nested inversion of electrical resistivity data, progressively incorporating data from profiles with smaller electrode spacing. This technique leads to better locate potential lithological contours and anomalous contacts. Finally, we used a joint stochastic approach to simultaneously invert ERT, seismic, gravity, and topographic data. This preliminary approach provides an a posteriori probability distribution of the parameters studied without the need for smoothing.These developments were applied to two different contexts. On the Pégairolles-de-l'Escalette landslide, the processing of 2D near-surface geophysical data (ERT, ...