Vers une nouvelle génération de modèles de source via une prise en compte réaliste des incertitudes

Perhaps the biggest obstacle to significant progress in observational earthquake source modeling arises from imperfect predictions of geodetic and seismic data due to uncertainties in the material parameters and fault geometries used in our forward models - the impact of which are generally overlooked. In this study, we develop a physically based stochastic forward model to treat the model prediction uncertainty and show how to account for inaccuracies in the Earth model elastic parameters. Based on a first-order perturbation approach, our formalism relates prediction error to uncertainties on the elastic parameters of different regions (e.g. crust, mantle, etc.). We apply this formalism to infer the co-seismic slip distribution of the 2013 Balochistan earthquake. We use a massively parallel Bayesian sampler named ALTAR. This Bayesian solver is based on GPU-accelerated computing and allows determining the full posterior probability density function of high-dimensional source models. Our results suggest that the rupture nucleated on a subvertical segment, branching out of the Chaman Fault system, and grew into a major earthquake along a north-dipping fault with significant along-strike curvature.