Multivariate Clustering Analysis of the ECRB Cross Drift Discontinuities, Yucca Mountain Project

A great deal of effort has been made to characterize the nature of the discontinuities of the Yucca Mountain proposed nuclear waste repository. Discontinuity sets were identified, using orientation only as the basis of cluster analysis. Other discontinuity characteristics such as lithology fracture frequency, continuity, roughness, aperture, and infilling have been analyzed, but separate and divorced from the clustering analysis. The across drift data has been re-analyzed using a multivariate clustering analysis algorithm developed by the authors (CYL). This type of analysis represents a relatively recent development in characterizing the structure of rock masses. It characterizes discontinuities into subsets according to multiple parameters, such as orientation, spacing, and roughness, where rather than considering one variable at a time, a number of parameters can be treated simultaneously, so that the interactions between parameters are taken into account. The comprehensive algorithm has been developed into a software package. It enables fully automated multivariate clustering analysis and offers various visualization tools, such as a three dimensional stereonet, a stereoscopic view, a statistical table, and pie charts relating the other factors such as lithology continuity, roughness, aperture, and infilling back to each cluster. Fig. 1. Plan view of the Yucca Mountain site, showing the cross drift (Mongano et al. [1]). Multivariate clustering has been proposed in the literature [3-9] in the last few years. In these algorithms clustering is done on basis of not just orientation, but also physical location roughness, and other quantifiable parameters. In the algorithms previously presented by the authors of this paper [37], a “3 dimensional” stereonet (concept shown in Figs. 2 and 3) where discontinuity normals are plotted on individual “stacked” stereonets, each normal is plotted with respect to its own stereonet, and each stereonet is plotted in a linear position that corresponds to the position where the discontinuity corresponding to that discontinuity normal intersects the bore hole or mapping scanline. This stereonet is the ideal device for visualization of clustering that is based on orientation and position, as shown in Fig. 4. The other parameter that can be used for clustering is roughness which cannot be visualized so easily. The clustering methods are best described in [4] and examples can be found in [5-7]. Fig. 2. Top: A lower hemisphere stereonet with four discontinuity normals (poles), each pole ostensibly from a different depth along an imaginary vertical bore hole. Middle: Each discontinuity normal (pole) is plotted on an individual stereonet. Bottom: The individual stereonets are stacked, with each spacing in proportion to the spacing between discontinuities in the borehole [7]. 2 3