The Marmousi Model

The Marmousi data set was generated at the Institute Francais du Petrole (IFP), and was used for the workshop on practical aspects of seismic data inversion at the 1990 EAEG meeting in Copenhagen, where different groups (contractors, universities, and oil companies) applied their proffered imaging tools on this data set. Detailed accounting of what transpired at the workshop is given by () and (). The original Marmousi data set is generated using a 2-D acoustic finite-difference modeling program.

One of the biggest contributions of the Marmousi model is that it demonstrated the limitations of first-arrival traveltimes in imaging complex media. Specifically, () showed that to properly image the Marmousi model we need multi-arrival traveltimes. In portions of the Marmousi model, or any other complex model, the first arrival is not necessarily the most energetic. Therefore, reflected energy from key horizons, such as the top of the reservoir, are not properly imaged by using only first arrival traveltimes.

The original Marmousi model was built to resemble an overall continental drift geological setting. Numerous large normal faults were created as a result of this drift. The geometry of the Marmousi is based somewhat on a profile of the North Quenguela through the Cuanza basin (). The target zone is a reservoir located at a depth of about 2500 m. The model contains many reflectors, steep dips, and strong velocity variations in both the lateral and the vertical direction (with a minimum velocity of 1500 m/s and a maximum of 5500 m/s).

Figure  shows a smoothed version of the Marmousi model displayed in the (x-z)-domain (top) and the ($x-\tau$)-domain (bottom). The smoothing operator, a 100 m box car function in 2-D, was applied in the (x-z)-domain. This same velocity is used to convert the depth axis to a time one using equation ().

 

vel
Figure 1 A smoothed version of the Marmousi velocity model displayed with the vertical axis given in depth (top) and in time (bottom). The smoothing operator consisted of a 2-D box car function with a 100 m length in each dimension.

The time section seems more complicated than the depth one, because the velocity-based depth conversion increased the apparent folding of the velocity model. Nevertheless, the output of raytracing using this new velocity model should be equivalent to the output of the conventional ray tracing using the depth velocity model.

In my paper from last fall (), I generated an anisotropic Marmousi dataset in a similar manner in which the isotropic one was generated (finite difference). I used exactly the same geometry settings used in building the the original isotropic dataset. Thus, all the parameter settings needed to perform migration on the isotropic Marmousi dataset can be used for the anisotropic one as well. The NMO and vertical velocity of the anisotropic model are the same as for the original Marmousi velocity model provided by IFP. The $\eta$ model, shown in Figure , has overall the same characteristics as the velocity model. It too has to be converted to time before usage in the migration.

 

eta
Figure 2 A smoothed version of the $\eta$ model displayed with the vertical axis given in depth (top) and in time (bottom). The smoothing operator consisted of a 2-D box car function with a 200 m length in each dimension.