An integrated data set containing detailed three-dimensional ground-penetrating radar surveys, outcrop mapping, and both in situ and lab-based petrophysical measurements provides an unusually detailed foundation for simulating fluid flow through the three-dimensional permeability structure of a channel sandstone reservoir analog within a study volume 40 m x 20 m (130 ft x 60 ft) (in plan) and 15 m (45 ft) deep. Permeability (k) data (more than 1100 data points ranging from 0.5-300 md) are obtained from probe permeameter measurements made on core plugs collected from five vertical cliff-face transects and on slabbed drill-hole core from four, 14-m-deep (42-ft) holes drilled 10-20 m (30-60 ft) behind the outcrop. Analytical petrography and comparison of outcrop k values with those from drill-hole core show that weathering has caused a factor of 2 to 5 increase in cliff-face permeability with a significantly different univariate k distribution from that of the behind-the-outcrop drill-hole core. The permeability differences are accounted for by "correcting" the weathered cliff-face values to mimic the distribution of k values obtained behind the outcrop using an inverse transformation method. The field-based k data are integrated into two classes of three-dimensional, geostatistically generated, stochastic permeability models. One model type uses ground-penetrating radar data solely to define important bounding surfaces with sandstone permeability extrapolated directly from the k data. The other model type uses the ground-penetrating radar data to explicitly constrain a sandstone permeability structure and to define the geometry, thickness, and distribution of shale/mudstone units.

A series of two-phase fluid-flow simulations within 15 m x 15 m (45 ft x 45 ft) (in plan) by 10.8 m (35 ft) deep model domains, using grid blocks 1 m² (9 ft²) (in plan) and 0.2 m (0.7 ft) thick, yield several conclusions. First, replacing the entire volume with a single k value computed with arithmetic, geometric, or harmonic mean of all k values in the volume yields a poor approximation of oil production

End_Page 405------------------------

computed using the original detailed, heterogeneous k structure. Second, in most cases upscaling using a vertically averaged geometric mean (as opposed to using arithmetic or harmonic means) provides a reasonable match to the oil production computed using the original detailed, heterogeneous k structure. Third, it is important to distinguish between fully continuous versus discontinuous shale unit geometries, and to be able to estimate angle of shale unit dip, when selecting methods for upscaling outcrop permeability values for use in reservoir simulators. Integrating three-dimensional ground-penetrating radar surveys with outcrop-based sedimentological and petrophysical data can provide the estimates of size, dip, continuity, and distribution of shale units needed at the scale of individual simulator grid blocks to help improve our approaches for deriving the upscaled k values used as input to production-scale reservoir simulators.