Model selection in fracture mapping from elastostatic data

In this paper, the inverse problem of imaging internally pressurized cracks from elastostatic measurements is investigated with special attention to the question of model choice. The selection of the most probable model from among a finite set of fracture geometry and loading model is carried out using Bayes factors. The modelling error variance is also estimated during the inversion procedure. This Bayesian model selection method also produces a known limit for the resolution of fracture dimensions, which depends on the configuration of the measurements. Both synthetic and real field examples in hydraulic fracture mapping applications are presented.