Bayesian Quantification and Reduction of Uncertainties in 3D Geomechanical-Numerical Models

The distance to failure of the upper crustal rock in the prevalent stress field is of importance to better understand fault reactivation by natural and induced processes as well as to plan and manage georeservoirs. In particular, the contemporary stress state is one of the key ingredients for this assessment. To provide a continuous description of the 3D absolute stress state geomechanical-numerical models are used. However, stress magnitude data for model calibration are sparse and incomplete and thus, the resulting model uncertainties are large. In order to reduce the uncertainties, we incorporate additional constraints on stress magnitudes to check the plausibility of different data-based stress states. We use formation integrity tests, borehole breakouts, drilling induced fractures, and observations of seismicity and distinct seismological quiescence. This information is weighted according to its confidence and the agreement with the different modeled stress states is assessed. The information is introduced to a Bayesian approach to estimate weights of the modeled stress states and thereby identify their plausibility. A case study in southern Germany shows the ability of the approach to identify from a wide range of stress states a small number of plausible ones and reject implausible stress states. This significantly reduces the number of stress states and thus lowers the model uncertainties.