From field data to fracture network modeling: An example incorporating spatial structure

This paper describes a technique for processing field data for a fracture network model which accounts for the observed spatial variability. This has been done by generating a network subregion by subregion where the properties of each subregion are predicted through geostatistics. Once the geometry of a particular realization is specified, flow through the network is studied. We develop the method for a two‐dimensional analysis based on data from Fanay‐Augeres, a uranium mine in France. We plan to extend the analysis to three dimensions and compare the results with in situ test results. In particular, we have focused on the data collected in a long section of a drift where fractures have been mapped and steady state permeability tests have been performed in 10 boreholes. In order to generate fractures in a statistically heterogeneous region we first divided the region into statistically homogeneous subregions. In each subregion and for each fracture set we must specify the areal fracture density and the orientation, length, and aperture distributions. We divided the fractures into five sets based on tectonic history and observed that for each set, fractures spaced close together tended to have similar orientations. This was built into the simulation. An estimate of the aperture distribution for each set was made by assuming that the hydraulic apertures of the fractures intersecting a well test zone were proportional to the fracture opening observed in the core. Data input to the geostatistical analysis for each set consisted of 16 values of mean length and density of fractures in each 5 m by 2 m section of the drift wall. The results of the simulation are tables of values for mean length and fracture density for each of the five sets. The value of density simulated for each subregion was used directly to determine the number of fractures to be generated of that set in the subregion. The value of mean length was used as the mean of the length distribution in the subregion. The local coefficient of variation for length was the same in every subregion. Using the data described above, we generated a 100 m by 100 m fracture network in a series of 100 statistically homogeneous subregions. The fractured region generated has a total of 65,740 fractures. A 70 m by 70 m region was isolated for directional permeability testing. Results show that the system is barely connected. About 0.1% of the fractures essentially control permeability. In the language of percolati