Over the past decade, a number of geostatistical methods have emerged for modelling depositional environments. Although these methods enjoy wide acceptance, they have rarely been tested with realistic reservoir models. One reason for this is lack of quantitative data. Another is the lack of consensus on what constitutes a valid test. The present study addresses both of these problems.
Fluvial environments offer a particular challenge for geostatistical modelling. The key to unlocking recovery in such reservoirs lies in a better understanding of how reservoir architecture and connectivity affect sweep efficiency. The present study entails constructing a "deterministic" model of one such reservoir. This model describes six channels within a twenty meter interval of the well-characterized Gypsy sandstone of Northeastern Oklahoma. Detailed spatial distributions of reservoir properties (permeability, porosity, and lithology) have been obtained from extensive sampling and mapping of the geological units of this formation as it is exposed by strike and dip oriented roadcuts. In addition, three-dimensional data has been obtained from a grid of twenty-two boreholes, with oriented core, drilled within three hundred meters behind the primary strike oriented outcrop. This model is as complete and as densely sampled as any model of a producing reservoir is ever likely to be.
Simulations on the Gypsy Outcrop Model show that waterflood recovery efficiencies can be highly variable, depending on the choice of well placement and transmissibility multipliers. Recovery efficiencies are presented for the case of unit mobility ratio displacements. An important focus for comparing various geostatistical methods should be to determine whether they can mimic the variability in recovery efficiencies that is displayed by models such as the present one. The criteria for evaluating the relative merits of competing approaches should emphasize recovery efficiency, rather than mere visualizations of heterogeneities in porosity and permeability.
Just as with reservoir simulation, there needs to be an SPE "Comparative Solution Project" in geostatistics. The Gypsy Outcrop model presented herein can serve as one of several "ground truth" models for evaluating both existing and future geostatistical methods. Comparison of flow modeling results using stochastic realizations constructed from such deterministic models will permit evaluations of the sensitivity of geostatistical methods to the type and amount of data available.