From pointwise in-situ data to stress prediction in 3-D

For a deep geological repository (DGR) for radioactive waste the characterization of the in-situ stress field is  critical for the repository design and evaluation of long-term safety. It is crucial to obtain a spatial 3-D description of the stress state (both orientation and magnitudes) in the host rock formation covering the potential repository area, and ideally also in the under- and overlying formations. For the prediction in 3‑D geomechanical-numerical models are used. The geological structure of these models is derived from the interpretation of 3‑D seismic surveys and borehole logs. After assigning the rock properties to the different lithologies, the model is calibrated using in-situ data of the horizontal stress magnitudes. One of the issues encountered during the calibration process is, that stress measurements might sample the mechanical variability of the subsurface units that is not explicitly represented by the geomechanical model. Another one is that stress measurements are typically reported as a best estimate without any uncertainty attached to it.

As part of the site evaluation for the final disposal of radioactive waste in Switzerland, three siting regions were explored. Between 2019 and 2022 an extensive and integrated campaign of measurements of the horizontal stress magnitudes was carried out in eight deep boreholes using Micro-Hydraulic Fracturing (MHF) and (dry) Sleeve Re-opening (SR) tests. Out of 139 successful MHF tests, 121 estimates of the magnitude of the minimum horizontal stress S_hmin and 65 estimates of the magnitude of the maximum horizontal stress S_Hmax were obtained. These data are used for the calibration of 3-D geomechanical-numerical models of the three siting regions. To achieve a best-fit with respect to the in-situ stress data, lateral displacements can be determined automatically provided that the problem is an elastically linear system. The models have a lateral side length between 10 and 15 km, represent 17 or 18 lithologies, and contain 6 to 13 faults that are implemented as contact surfaces allowing relative displacement to each other. The resolution in the vertical is at best 5 m and laterally between 50-150 m using up to 8 × 10⁶ finite elements.

We show that a best-fit can be achieved with our model workflow and that the prediction of the 3-D stress state in the larger volume is primarily controlled by the variability of the rock stiffness. As the rock stiffness is a probability distribution in each lithological layer, the prediction of the horizontal stress magnitudes in the larger volume is as well. The result of our workflow are bandwidths of the predicted 3-D stress field that can be represented for example with the P05-P95 probability range.