Uncertainties in geomechanical models – A novel concept to data-driven stress modelling

The undisturbed stress state of the subsurface is a key parameter for the assessment of geological barrier integrity in deep geological repositories. A robust knowledge of the stress state is required for forward simulations that predict the evolution of the deep geological repository during building, canister emplacement, and post-closure phase. However, stress magnitude data will always be sparse and point-wise and thus not sufficient to allow a continuous characterization of the stress field. 3D geomechanical-numerical models are incorporated to mitigate this shortcoming. The significance of such a model is tied to the stress magnitude data records whose availability is generally low and subject to measurement errors. In turn, the uncertainties of the modelled 3D stress state are large which lowers the predictive values of the ensuing forward models. Here, we present a novel approach to reduce the uncertainties of a modelled stress state (Ziegler and Heidbach, 2023).

In a first step we estimate all possible and realistic stress states with respect to the available stress magnitude data and their uncertainties. This provides an overall quantification of the range of stress states that are supported by data. In a second step we limit this usually very large range of stress states using additional indirect stress information to determine upper and lower limits. Formation integrity tests (FITs) are used as a lower limit for the stress state. Observed borehole wall failures (drilling induced tensile fractures and borehole breakouts) provide an estimate of the quality of a modelled stress state in relation with estimations of the rock strength. Observed seismicity and observed seismological quiescence in connection with a failure criterion is used as an upper constraint of the stress state. These additional data allow to identify the probability of a modelled stress state based on its agreement with observations. A weight is estimated for each modelled stress state in a formalized Bayesian approach. This allows an improvement in the significant interpretation of the initial stress state in terms of its robustness and transparency.

Results from an explanatory model and a case study in the Bavarian Molasse Basin show an improved significance in terms of clearly reduced model uncertainties. The amount of uncertainty reduction, however, depends significantly on the quality, suitability and assessment of the additional information.

 

Reference:

Ziegler, M. O., & Heidbach, O. (2023). Bayesian quantification and reduction of uncertainties in 3D geomechanical-numerical models. Journal of Geophysical Research: Solid Earth, 128, e2022JB024855. https://doi.org/10.1029/2022JB024855