At the start of the SEAM project, a general consensus was that a finite-difference (FD) method or related method would be the most efficient to create a synthetic seismic data set from the SEAM geologic model. This was not a requirement, and SEAM was open to alternative simulation approaches. As with all methods, FD has its specific limitations and costs associated with a particular simulation. The SEAM organization tried to anticipate many of these when choosing the geologic model size and survey specification so that the simulation could be performed in a reasonable amount of time and cost.
The main limitation is that the model is sampled at discrete points along a regular grid. Thus, it is not possible to capture jumps in material properties at arbitrary points. This leads to time shifts in output data when comparing results of simulations that use finer or coarser spatial sampling. These time shifts are a major cause of differences as discussed in “Benchmark Comparisons” and related references.
The discretization of a continuous model also has implications for the solution. In short, the spatial wavelength has to be adequately sampled (cf. the Nyquist criterion). Inadequate spatial sampling leads to what is called numerical dispersion, i.e., different frequencies propagate at different velocities. The numerical dispersion is different for FD schemes of different accuracy order. A higher-order scheme in general is less affected by numerical dispersion, which allows the use of a larger grid size for a given amount of dispersion; however, higher-order schemes require.