[en] This paper describes some new techniques for stochastic modeling of three-dimensional fracture networks. We use geostatistical simulation methods to reproduce features of the spatial structure of the rock such as the variation of fracture density and fracture orientation in space. For an example of the method we use mapped fracture data from the Fanay-Augeres mine, in Limousin, France. Two different sections of a drift wall, S1 and S2, were mapped. The S1 section is wet, and the S2 section is dry. For each case, the fractures are divided into five different sets and each set is modeled separately. The fractures in each set are represented as discs placed randomly in space. The diameter of each disc is chosen independently from a fixed probability distribution determined from the trace length distribution. For the location of discs a point process called the parent-daughter process is used. This process gives a clumping or swarming of fractures not found in the usual Poisson model. The orientation of the discs is characterized as a fluctuation about the mean orientation for the set. This fluctuation has a spatial structure that is simulated with geostatistics. Geostatistical simulations of the two fracture systems are under way. The connectivities of the simulations will be assessed to see if there is any correlation with the fact that the S1 section of drift is wet and the S2 section is dry

[en] The analysis of fluid flow through fractured rocks is difficult because the only way to assign hydraulic parameters to fractures is to perform hydraulic tests. However, the interpretation of such tests, or ''inversion'' of the data, requires at least that we know the geometric pattern formed by the fractures. Combining a statistical approach with geophysical data may be extremely helpful in defining the fracture geometry. Cross-hole geophysics, either seismic or radar, can provide tomograms which are pixel maps of the velocity or attenuation anomalies in the rock. These anomalies are often due to fracture zones. Therefore, tomograms can be used to identify fracture zones and provide information about the structure within the fracture zones. This structural information can be used as the basis for simulating the degree of fracturing within the zones. Well tests can then be used to further refine the model. Because the fracture network is only partially connected, the resulting geometry of the flow paths may have fractal properties. We are studying the behavior of well tests under such geometry. Through understanding of this behavior, it may be possible to use inverse techniques to refine the a priori assignment of fractures and their conductances such that we obtain the best fit to a series of well test results simultaneously. The methodology described here is under development and currently being applied to several field sites. 4 refs., 14 figs