Static and Dynamic Assessment of DFN Permeability Upscaling

Abstract Nearly half of the remaining petroleum reserves are contained in naturally fractured reservoirs (NFR). An accurate estimate of the effective fracture permeability tensor is a key to the successful prediction of oil recovery from NFR. Standard workflows nowadays employ discrete fracture network (DFN) modeling and analytical or flow-based methods to upscale fracture permeabilities. However, DFN modeling imposes some important challenges, which can cause great uncertainty in the effective permeability tensor and subsequent recovery prediction: Analytical upscaling methods, which are commonly used due to computational efficiency, are inaccurate for poorly connected fracture networks. Flow-based upscaling methods depend on boundary conditions and are computationally expensive. Defining the optimum grid size for either method is also very difficult. In addition, DFN upscaling is often driven by practical issues such as time constrains and computational limitations, leaving little room to investigate the effects of upscaling methods and grid size. In this paper we utilize features in leading DFN simulators employed in standard industry workflows for computing effective permeability tensors with flow-based and analytical methods. We use two realistic dataset from fractured formations of onshore reservoirs in our assessment. Not surprisingly, there is up to three orders of magnitude variation in the effective permeability based on the chosen upscaling method and perceived optimum grid cell size. This has tremendous impact on predicted recovery rates and ultimate recovery; ultimately uncertainty in upscaling can mask uncertainty in the geological model. We hence introduce a new simulation technique, Discrete Fracture and Matrix (DFM) modeling, which accounts accurately for flow in the fractures and rock matrix as an efficient alternative for computing effective permeability tensors as it allows us to assess the accuracy of classical DFN upscaling approaches, which all help reducing uncertainty in recovery prediction.