Merging Outcrop Data and Geomechanical Information in Stochastic Models of Fractured Reservoirs

Abstract

Abstract According to recent estimates, the potential for oil production from fractured reservoirs in North America is of the order of tens of billion of barrels. With domestic production depending more and more on mature fields, better technology for characterizing fracture flow paths, especially in deep, non-conventional plays and in carbonate rocks is key to producing hydrocarbons economically. Fracture transmissivity (or permeability) can enhance oil production, or on the other hand, result in early water breakthrough and consequently early well abandonment. However, spatial characteristics of the fracture system cannot be known deterministically in the subsurface reservoir. Instead, stochastic characterisation of fracture systems is usually attempted. The development of a stochastic modeling approach that yields realistic field-scale model of fracture networks consistent with patterns observed on an outcrop and adhere to a mechanical basis for fracture propagation is presented in this paper. Introduction Fracture patterns are generally characterized by statistics such as fracture spacing, density, orientation and statistical distributions of width. Differences in variation of fracture orientations and spacing are important for distinguishing between different fracture types. Data for inferring distributions of orientation, spacing and density are generally obtained after detailed outcrop characterization. Figure 1a depicts the regional fracture patterns found in Jurassic Navajo sandstone, Lake Powell, southeastern Utah (Nelson, 1976). The dominant fracture orientation can be gauged visually in that figure. Figure 1b depicts a set of conjugate shear fractures in an outcrop from Wyoming. This set is reflective of tectonic fractures. Figure 1c is a photograph of desiccation cracks observed in mud (Nelson, 1979). It is evident that the three fracture systems exhibit widely different pattern characteristics. Statistical tools that are capable of capturing the fracture pattern statistics have to be formulated and tested first. Once pattern statistics have been reliably calibrated, a spatial interpolation scheme has to be devised that can take into consideration any reservoir/well data specific to the reservoir under study as well as the pattern statistics inferred from analogs. Such an interpolation scheme has to yield realistic fracture patterns consistent with that observed in outcrops and other analogs. In order to constrain the interpolation even more, auxiliary data such as that inferred from geomechanical models for fracture propagation have to be used. The stochastic models of fracture networks have to be consistent with the physical criteria for fracture growth observed in the geomechanical models. Since in most cases the data available to model the fractured reservoir is sparse and information such as seismic maps and production response are related imprecisely to the fracture pattern characteristics, a probabilistic approach to fracture characterization is necessary. In the object-based modeling approaches, fractures are represented as objects defined by their centroid, shape, size and orientation. In "Random Disk" models (Baecher et al., 1977), fractures are represented as two-dimensional convex circular disks located randomly in space. The radii of the disks are drawn from a lognormal distribution whose parameters are inferred from the fracture trace-length distributions observed in outcrops. The distribution and the disk location, radii and orientation are assumed uncorrelated from one disk to the next. It is difficult to model the clustering of fractures accurately assuming random placement of disks. A spatial density function can be utilized to represent such clustering of fractures (Billaux, 1987; Chiles, 1989). Seed locations for fractures are drawn based on this spatial density function and fracture sets are simulated over a pre-defined volume around the seed location. The resultant parent-daughter fracture sets will exhibit clustering.