Let’s assume that geologists are (now ) all convinced that modelers will be able to respect their interpretation while building their models. We are still in a situation where the interpretation is completed before the modeling is done. It is not necessarily wrong to do so, but it’s also possible to get better results by using the modeling to at least test and if needed improve the interpretation. An example illustrates this point (Figure 1 and Figure 2).

Figure 1 shows two wells A and B in cross-section, again in a sand-shale reservoir. They are part of a complete 3D dataset. The geologist decided to correlate the sand A1 of well A with the sand B1 of well B, as well as the sand A2 and B2. The sand A3 is not correlated to any sand in well B. From there, the geologist created a thickness map for the upper sand (A1-B1) as well as a thickness map for the lower sand (A2-B2) and the team starts making plans on how to develop this field. In parallel, a reservoir model was created to be fed to flow simulation (Figure 2). The two wells A and B were used in the 3D modeling, but not the geological interpretation itself due to limited communications between the geologist and the modeler: the team assumed that the model would necessarily match the geologist’s interpretation, so minimal time was spend on this. The result is drastically different from what the geologist had in mind though: in the geomodel, the sand A1 is not connected to well B, while we have two massive sands A2-B1 and A3-B2. If the model was to be sent as is to flow simulation, the team would now have two different representations of their reservoir: the one expressed in the early sand thickness maps created by the geologist and now the 3D model.

From our personal experience, this situation occurs more often than not and this example illustrates two important points.

Firstly, as was mentioned earlier, a model needs to be built from the data (the wells) and the geological interpretation, not just the data. The next section will give more details about this. Secondly, the model can be used to test the geological interpretation. In this example, maybe the data, once looked through the prism of geostatistics and 3D gridding algorithms, make it more geologically reasonable to connect the sands as per Figure 2 and not as the geologist first thought. A 3D model will of course not always improve on what geologists interpreted up-front. Most of the time, the model will simply concur with their analysis. But at least the model should be used to validate the geological hypotheses. Then, after looping through interpretation and modeling several times and when a unique representation of the reservoir is agreed upon by the team, it is then time to extract some useful maps (for example) to help guide the development of our field.