Stochastic Regridding of Geological Models for Flow Simulation

Abstract

Abstract Regridding geological models to a higher resolution for flow simulation is an important problem in geostatistical modeling. For practical reasons, over a large area, models can only be built at a relatively coarse resolution. Subsequently, the resolution of specified regions of interest must be increased before upscaling for flow modeling. The construction of a high-resolution model of the entire reservoir at the beginning of the evaluation may be impractical because of computational and time constraints. It is standard practice to implement nearest neighbor interpolation to increase the resolution of models. Although it is a simple practical solution, nearest neighbor interpolation introduces spatial continuity artifacts that are often unrealistic. This paper proposes an automatic stochastic regridding approach based on simulation. The simulation is conditioned to the initial coarse resolution model/realization. The process includes the extraction of specified regions of interest, definition of corresponding local variography, and implementation of Sequential Gaussian Simulation (SGS) and/or Sequential Indicator Simulation (SIS) to characterize continuous and categorical variables, respectively. In each specified region, the local variography can be defined by either implementing automatic fitting algorithms or assigning the global variography initially used to build the coarse resolution model. The regridding process is automated. The advantage of this approach over the conventional nearest neighbor interpolation is in the improvement in the realistic spatial variability features of small scale geologic heterogeneity. The benefits of obtaining a proper regridded model are discussed in a case study of a fluvial reservoir in the McMurray formation. One of the main reasons for generating high resolution models is in the appropriate characterization of small scale impermeable geobodies such as remnant shales. The coarse resolution models are not able to properly characterize the small scale geologic features of the shales; more amount of information is required to characterize smaller scale features. The metric of performance considered is the effective vertical permeability. The automated stochastic regridding workflow described in this paper is available on a Fortran platform with additional scripting which will be distributed upon request. Note that the terms “regridding” and “stochastic regridding” are used interchangeably and both refer to the proposed workflow of modeling at higher resolution.