Discrimination of Discrete Fracture Network models using structural and flow data
Résumé
Fractures are key elements governing permeability and flow paths in crystalline rocks and sedimentary layers with low matrix porosity. The ability to properly predict those properties, which is crucial for some industrial applications, such as the risk assessment in nuclear waste management, strongly relies on our ability to properly describe the fracture network characteristics. A usual method is to define discrete fracture network (DFN) models, where a set of 3D fractures represents the geological environment. It primarily consists in combining various data, thanks to stereological relationships, on the fracture properties (geometry, transmissivity) and on hydraulic properties (e.g. borehole flow logs), in order to produce 3D statistical distributions and upscaling functions. However, this approach is not univocal and various DFN models can match the observations. We present the results of a study made for the Swedish Nuclear Fuel and Waste Management Company (SKB) through an application to the Forsmark site in Sweden. We perform an analysis of structural data and flow logging based on the PFL tool [1] to characterize the geological environment in terms of fracture density, permeability, and flow channeling by focusing on their scaling and variability. Then, we define candidate DFN models. A critical component of the DFN is the fracture size distribution, which is the upscaling function required to extrapolate fracture densities between data gaps, from borehole cores up to site scale. Another important feature of DFN models lays in the spatial correlations between fractures, which is neglected by the commonly used Poissonian (i.e. spatially random) models. We recently developed a new DFN modelling approach (further referenced as UFM for Ubiquitous Fracture Model), which mimics the geological processes of fracturing [2, 3]. For the same distribution of fracture size and orientations, networks from the UFM model are much less connected than their random equivalent [4]. Assigning fracture transmissivities is also an important issue in the modeling process, especially because only 20-25% of the total fracture surface is considered open (the rest is clogged). In addition, the transmissivity of permeable fractures is likely dependent on size and stress conditions. We perform flow simulations on these DFN models and reproduce the flow data. We compare the hydraulic properties of the synthetic (DFN) media to those of the natural environment to assess their prediction capabilities. Our conclusions are that networks produced by the UFM model are capable to properly reproduce the scaling behavior of the natural flow without any specific calibration, unlike Poissonian models. We also point out the importance of defining the dependency of fracture transmissivity on fracture size and orientation (including the role of the applied stress on fracture transmissivity). Indeed, such a dependency changes the scaling of hydraulic properties, their spatial variability and increases the channeling. [1] A. Öhberg, P. Rouhiainen, "Posiva Groundwater flow Measuring Techniques," (Posiva, 2000). [2] P. Davy, R. Le Goc, C. Darcel, A model of fracture nucleation, growth and arrest, and consequences for fracture density and scaling. Journal of Geophysical Research: Solid Earth 118, 1393-1407 (2013). [3] P. Davy et al., A likely universal model of fracture scaling and its consequence for crustal hydromechanics. J. Geophys. Res. 115, 1-13 (2010). [4] J. Maillot, P. Davy, R. L. Goc, C. Darcel, J. R. d. Dreuzy, Connectivity, permeability, and channeling in randomly distributed and kinematically defined discrete fracture network models. Water Resour. Res. 52, 8526-8545 (2016).