Identifying Training Images from Fracture Outcrops for MPS-based Modeling

Abstract:

Various MPS (multi-point geostatistics) based techniques have been successfully employed for modeling heterogeneous geologic features such as channel bodies. However, few studies exist on modeling fracture networks implementing MPS. Selecting a TI (training image) that is representative of the target subsurface fracture pattern is key to an effective implementation of any MPS algorithm. While recent research has employed equivalent continuum models created from Discrete Fracture Networks (DFN) to serve as TIs, the present research explores the direct usage of outcrop analogues of fracture networks for this purpose. A set of nested-fracture maps from the Devonian Sandstone of Hornelen basin created at multiple scales and resolutions is considered. These maps have been previously classified as belonging to a single fractal system characterized by a fractal dimension but having slightly different spatial organization at each scale. Our research implements unconditional image quilting in generating multiple realizations of fracture networks from these maps. This pattern-based algorithm requires a template-size to be chosen which is representative of the heterogeneity of the pattern of interest. Lacunarity is a technique that essentially quantifies the distribution of spaces or gaps in a pattern and can thus delineate scale-dependent heterogeneity. It is therefore investigated if this technique, in conjunction with Entropy (a measure for randomness), can be applied for choosing the template-size required for reproducing a desired pattern from a given TI. Finally, it will be tested if fracture networks generated from two or more TIs at different scales but belonging to the same fractal-fracture system, can capture similar ranges of uncertainty.