Accounting for uncertainty in dual porosity descriptions of fractured systems

Fractures have crucial impact on flow in reservoirs. In naturally fractured reservoirs (NFR) the fracture network alone can be the dominant factor in fluid transport. Fractures as geological discontinuities introduce a high level of complexity into the entire reservoir modeling workflow. In current practice, the geological modeling of fractures and flow simulation are considerably disconnected, simply because current commercial simulators cannot handle the complexity of current fracture models. As a result, flow models ignore much of the geological information possibly resulting in low prediction power. A new workflow for the quantification of uncertainty of geological / geophysical parameters and engineering decisions on the flow response is proposed. This workflow spans the modeling domain, from geological discrete fracture models to flow simulations (3DSL). Multidimensional scaling (MDS) is used to quantify the distance between different geological scenarios based on the flow response and select representative fracture models. Clustering of the flow responses allows for identification of geological parameters with high impact on uncertainty. In this paper, we restrict ourselves to dual porosity modeling, although the ideas behind the workflow can be applied to a variety of flow in fracture formulations. Initially, Discrete fracture networks (DFN) are generated stochastically, e.g. in FracMan, from the input of geological parameters and constrained to geological structures (depending on the scenario). However, fluid flow simulators using the dual media paradigm require reservoir grids populated with effective properties which require upscaling. The upscaled fracture properties determine which grid cells are considered to have dual-media. When upscaling of DFNs to effective properties most of the detailed features of the fractures disappear. However, the fracture model induces patterns in the effective flow properties. These patterns represent one or more geological scenarios and can be clustered by distance-based methods. Furthermore the sensitivity to spatial uncertainty resulting from the stochastic generation of DFNs also depends on the pattern and therefore on the underlying discrete fracture model. Some initial results of these ideas are presented on complex synthetic case studies.