Geology-driven modeling: A new probabilistic approach for incorporating uncertain geological interpretations in 3D geological modeling

Combining different sources of information about the subsurface is an inherent challenge in the process of making realistic geological and hydrostratigraphic models. Often the available geological and hydrological data from boreholes or outcrops are sparse and modeling is supplemented spatially with geophysical data to obtain a better understanding of the 3D lithological, structural, and hydrological relations in the study area. In traditional geological modeling, the modeler combines all this information during modeling and consider several factors like e.g., distance to neighboring data, consistency between different information, data uncertainty and geological environment when assigning uncertainties to the interpretation points. However, the assigned uncertainty is subjective and can only be communicated qualitatively. The benefit of a probabilistic model is that it enables a more quantifiable approach to subsurface modeling, but probabilistic models are usually difficult to set up, computationally demanding as well as difficult to interpret for the geologist and decision makers. Moreover, there is little tradition for including geological knowledge/information directly in probabilistic approaches. In the following, we utilize the interpretations from a traditional manually interpreted (cognitive) 3D hydrostratigraphic layer model as input for a probabilistic model. A realization of the subsurface is created from a geological or hydrostratigraphic model by geostatistical simulation of each interpreted layer based on the geologist's interpretation points with corresponding uncertainties. By compiling all the simulated layers, a 3D structural model is obtained. By studying a sample of such 3D realizations, the interpretation uncertainty in the cognitive structural model can be derived. We name this methodology geology-driven modeling (GDM) as it is based on geological interpreted data rather than the geophysical data directly. The methodology is tested using sequential Gaussian simulation on a cognitive hydrostratigraphic model from Denmark. Our results show that GDM successfully allows transforming the static cognitive model into a full probabilistic model and enables the uncertainties to be communicated to further modeling or decisionmakers. The proposed methodology allows updating pre-existing 3D geological and hydrostratigraphic layer models in a geologically intuitive stochastic framework or be directly incorporated into the current modeling framework.