The starting point of our project was building a reservoir model, shown in Figure , whose lithology and structure resembles a North Sea reservoir. We then simulated the multiphase fluid flow and production history of the reservoir over a period of 3 years. We estimated the effective seismic properties of the reservoir combining the lithologic model and the fluid properties. Finally, we simulated multi-offset 3-D seismic surveys at different times during the production history. This process is shown as Steps A-D in Figure .

Figure 1

Figure 2

While the macro-scale of our model reservoir is constant ( km) the sampling of the reservoir physical parameters varies greatly between steps. The main scales that we used are described below:

**Core Scale**- is about m.
Much of what is known
about rock physics relations among rock, fluid, and seismic properties
has come from measurements at the core scale. Description of the complete
reservoir at this
scale would yield 10
^{13}(10,000 billion) voxels and is not practical. Nevertheless, in any reservoir, a subset of core data can yield seismic-to-reservoir properties relations that are critical to the seismic modeling. **Reservoir Geological Modeling Scale**- is nominally m. Ideally, the geological model would be at the resolution of the best information, i.e., the core data. This is not practical and, instead, a scale is chosen small enough to capture the significant geological heterogeneities and large enough to yield a tractable number of cells or voxels. The scale chosen here yields 200 million voxels which is ambitious but practical for reservoir modeling. There are about 50,000 core scale voxels within a geological modeling cell.
**Reservoir Flow Simulation Scale**- The scale of the grid blocks in fluid flow simulation is a compromise
between the desire to minimize artificial numerical artifacts,
accounting for sufficient geological detail, and available computational
resources. Full-field simulation, requiring order 10
^{2}grid blocks between wells can easily lead to systems with 10^{6}-10^{8}grid blocks, which remains well beyond current computational possibilities. In this study, a discretization of the macro-scale into nx=132 ny=140 nz=12 resulted in grid blocks with an areal scale of m in the coarse region and m in the refined region. The vertical size of the grid blocks, on average, was 8.3 m. **Seismic Imaging Scale**- is the scale of the resolution of the seismic images, that is linked with the wavelength of the seismic signal recorded at the surface ( m in our model). We described the seismic properties of the reservoir on a meters scale. Description at this scale of the whole reservoir would yield about 50 million voxels. We have used a subset of the whole reservoir, for a total of about 1.5 million voxels.

11/12/1997