Statistical methodology for discrete fracture model - including fracture size, orientation uncertainty together with intensity uncertainty and variability

Abstract

Investigations led for several years at Laxemar and Forsmark reveal the large heterogeneity of geological formations and associated fracturing. This project aims at reinforcing the statistical DFN modeling framework adapted to a site scale. This leads therefore to develop quantitative methods of characterization adapted to the nature of fracturing and data availability. We start with the hypothesis that the maximum likelihood DFN model is a power-law model with a density term depending on orientations. This is supported both by literature and specifically here by former analyses of the SKB data. This assumption is nevertheless thoroughly tested by analyzing the fracture trace and lineament maps. Fracture traces range roughly between 0.5 m and 10 m - i e the usual extension of the sample outcrops. Between the raw data and final data used to compute the fracture size distribution from which the size distribution model will arise, several steps are necessary, in order to correct data from finite-size, topographical and sampling effects. More precisely, a particular attention is paid to fracture segmentation status and fracture linkage consistent with the DFN model expected. The fracture scaling trend observed over both sites displays finally a shape parameter k{sub t} close to 1.2 with a density term (alpha{sub 2d}) between 1.4 and 1.8. Only two outcrops clearly display a different trend with k{sub t} close to 3 and a density term (alpha{sub 2d}) between 2 and 3.5. The fracture lineaments spread over the range between 100 meters and a few kilometers. When compared with fracture trace maps, these datasets are already interpreted and the linkage process developed previously has not to be done. Except for the subregional lineament map from Forsmark, lineaments display a clear power-law trend with a shape parameter k{sub t} equal to 3 and a density term between 2 and 4.5. The apparent variation in scaling exponent, from the outcrop scale (k{sub t} = 1.2) on one side, to the lineament scale (k{sub t} = 2) on the other, addresses the issue of the nature of the transition. We develop a new 'mechanistic' model that could help in modeling why and where this transition can occur. The transition between both regimes would occur for a fracture length of 1-10 m and even at a smaller scale for the few outcrops that follow the self-similar density model. A consequence for the disposal issue is that the model that is likely to apply in the 'blind' scale window between 10-100 m is the self-similar model as it is defined for large-scale lineaments. The self-similar model, as it is measured for some outcrops and most lineament maps, is definitely worth being investigated as a reference for scales above 1-10 m. In the rest of the report, we develop a methodology for incorporating uncertainty and variability into the DFN modeling. Fracturing properties arise from complex processes which produce an intrinsic variability; characterizing this variability as an admissible variation of model parameter or as the division of the site into subdomains with distinct DFN models is a critical point of the modeling effort. Moreover, the DFN model encompasses a part of uncertainty, due to data inherent uncertainties and sampling limits. Both effects must be quantified and incorporated into the DFN site model definition process. In that context, all available borehole data including recording of fracture intercept positions, pole orientation and relative uncertainties are used as the basis for the methodological development and further site model assessment. An elementary dataset contains a set of discrete fracture intercepts from which a parent orientation/density distribution can be computed. The elementary bricks of the site, from which these initial parent density distributions are computed, rely on the former Single Hole Interpretation division of the boreholes into sections whose local boundaries are expected to reflect - locally - geology and fracturing properties main characteristics. From that starting point we built Statistical Fracture Domains whose significance rely exclusively on fracturing statistics, not including explicitly the current Fracture Domains or closeness between one borehole section or the other. Theoretical developments are proposed in order to incorporate the orientation uncertainty and the fracturing variability into a resulting parent distribution density uncertainty. When applied to both sites, it comes that variability prevails in front of uncertainty, thus validating the good level of data accuracy. Moreover, this allows to define a possible range of variation around the mean values of densities. Finally a sorting algorithm is developed for providing, from the initial elementary bricks mentioned above, a division of a site into Statistical Fracture Domains whose internal variability is reduced.