Item

Abstract

The current crustal stress field is of key importance to understand geodynamic processes and to assess stability aspects during subsurface usage. To provide a 3-D continuous description of the stress state, linear elastic forward geomechanical-numerical models are used. These models solve the equilibrium of forces between gravitational volume forces and surfaces forces im- posed mainly by plate tectonics. The latter are responsible for the horizontal stress anisotropy and impose the inverse problem to estimate horizontal displacement boundary conditions that provide a fit best to horizontal stress magnitude data within the model volume. Ho wever , horizontal stress magnitude data have high uncertainties and they are sparse, clustered and not necessaril y representati ve for a larger rock v olume. Even w hen Bay esian statistics are incor - porated and additional stress information such as borehole failure observations or formation integrity test are used to further constrain the solution space, this approach may result in a low accuracy of the model results, that is the result is not correct. Here, we present an alternative approach that removes the dependence of the solution space based on stress magnitude data to avoid potential low accuracy . Initially , a solution space that contains all stress states that are physically reasonable is defined. Stress magnitude data and the additional stress information are then used in a Bayesian framework to e v aluate which solutions are more likely than others. We first show and validate our approach with a generic truth model and then apply it to a case study of the Molasse foreland basin of the Alps in Southern Germany. The results show that the model’s ability to predict a reliable stress state is increasing while the number of likely solutions may also increase, and that outlier of stress magnitude data can be identified. This alternative approach results in a substantial increase in computational speed as we perform most of the calculations anal yticall y.