We address the question of what grid-size is necessary to adequately simulate oil recovery from a variety of heterogeneous facies found within a fluvial-deltaic system. We start by generating geological models of core-plug-scale poro-perm heterogeneities based on a geological understanding of the spatial distribution of lithotypes within a facies. This fine scale distribution is then re-scaled to different reservoir simulation grids which coarsen both horizontally and vertically. The finest simulation grid was such that the effect of re-scaling was negligible and thus the simulation model was almost an exact representation of the "true" reservoir. Equivalent homogeneous models, using the global effective properties of the fine-scale reservoir description, were also run. This enabled us to compare purely numerical effects with the effects of re-scaling the heterogeneous description. Waterflood and gasflood simulations were then run on each grid.
For the heterogeneous systems considered, the coarser grids under-estimated recovery - by up to 16%. Also, for the homogeneous models, waterflood recovery always decreased as the grid coarsened; gasflood recovery was relatively insensitive to block size. Further, recoveries were always higher in the heterogeneous model compared to the homogeneous model at any given grid size.
Finally, recovery and PI were grossly under-estimated at gridblock sizes that were approximately the same as that of significant heterogeneities - finer or coarser grids did a better job. The global effective permeability of the re-scaled grid was much lower for these inadequate grid sizes. Global effective permeability can thus be used to assess grid adequacy with respect to heterogeneity prior to performing simulations.