Résumé
Today, one of the main challenges in the oil industry is the exploitation of new resources in naturally fractured reservoirs often located in structurally complex areas such as plate boundaries, mountain ranges or near salt structures. While numerical models of rock deformation based on continuum mechanics are becoming industry standard in providing efficient means for modelling natural fractures in reservoirs, the knowledge of past perturbed stress fields, which is referred as the type (normal, wrench or reverse), orientation and magnitude of the regional or local tectonic stresses that needs to be applied as boundary condition in numerical simulations, is often unknown or approximated.The essence of this thesis is to improve the recovery of the paleostress field needed to constrain numerical simulations. I first described and discuss through a comprehensive mechanically-based multi-parametric study how far we can rely on the commonly used paleostress inversion methods based on fault slip data and the Wallace and Bott assumptions. I then compared this method with a new generation technique based on geomechanics using field observations. Since slip markers on faults are hardly observed in core or image log, I used observed natural fracture data as main drivers for the inversion of the paleostress using geomechanically-based method. I demonstrated, through various outcrop and subsurface examples, how this can efficiently be done. Finally, I aimed at addressing the problem of polyphase fracture data sets with unknown mechanical type (joints, faults, stylolites ...) and expanded the mechanical stress inversion to the separation of tectonic phases.