Abstract:
Fractures, such as joints, faults and veins, strongly influence the transport of fluids through rocks by either enhancing or inhibiting flow. Especially in rocks with negligible permeability fractures can act as major for fluid conduits. Therefore, the contribution of fracture networks to the overall flow behavior is of high interest to, for example, the hydrocarbon industry, the power generation using deep geothermal systems, the sustainable management of fractured rock aquifers, the planning of high risk waste repositories and geotechnical projects situated in fractured rock. The fluid transport through a fractured rock mass can be simulated by continuum, discrete fracture network (DFN) or hybrid models. Latter combines aspects of the continuum and the DFN approaches. Especially the DFN approach relies on a detailed knowledge of fracture network characteristics. In the subsurface fractures and fracture networks are typically characterized studying well cores and image logs, whereas at the surface outcropping subsurface analogues are studied. Especially outcrops provide valuable information such as fracture length and length distribution, which are important parameters for fluid transport simulations. The fracture networks encountered at outcrops are commonly analyzed applying the scanline sampling, window sampling, or the circular scanline and window method. These methods vary in their application and the parameters they provide. Therefore, each method has its specific advantages and limitations, which are summarized in a critical review.
In order to compare the application of the scanline sampling, window sampling, and the circular scanline and window methods, natural fracture networks outcrops of (1) Ignimbrites at Craghouse Park, UK, (2) the Wajid Sandstone in Saudi Arabia, and (3) Miocene limestone in the Oman Mountains, Oman are analyzed. Although, sampling biases such as orientation, truncation and size bias were accounted for, the network parameters calculated from different sampling methods show significant differences. Two plausible explanations for those differences exist: (1) a lack of measurements to adequately define the fracture network parameters, and (2) the influence of censoring bias on the estimated network parameters. Artificially generated orthogonal two-dimensional fracture networks (AFNs) with known input parameters were used two evaluate (1) the required minimum number of measurements for each sampling method, and (2) quantify the influence of censoring bias on the evaluation of fracture network parameters. The large numbers of sampling areas investigated during this process were analyzed using the novel software FraNEP (Fracture Network Evaluation Program), which was developed as part of this thesis.
The lowest minimum number of measurements to adequately capture the statistical properties of fracture networks was found to be 110 for the window sampling method, followed by the scanline sampling method with 225. For the application of the circular scanline and window method at least 860 fractures should be present in a sampling area. Although, these numbers are not universally applicable, they may serve as a first guideline for the analyses of fracture populations with similar length distributions. The latter resemble those reported for typical natural fracture networks. Furthermore, the window sampling method proved to be the method that is least sensitive to censoring bias. Reevaluating the natural fracture networks proved that the existing percentage of censored fractures significantly influences the accuracy of inferred fracture network parameters.