6.4 Ranking geomodels for the purpose of flow simulation
The previous section describes the fundamentals of creating the geometry of a flow simulation grid and how to transfer facies and petrophysical properties. One big issue remains. What do we do when we have created hundreds of geological models, each one made of a 3D distribution of facies and of 3D distributions of petrophysics. Do we give only one to the simulation engineer? Do we give several realizations? And, depending upon the approach, which of the hundreds shall we give? Do we transfer all the realizations? Answering the last question is the easiest one. No, we do not transfer all the realizations (much to the relief of our engineers!). If each realization takes hours, days or even weeks to run, one can only imagine how long it would take to run all of our realizations.
In that case, should we send just one then? This approach is still used a lot. It tends to be part of some discussion between the geomodeler and the engineer which goes like this: “I have hundreds of realizations, which ones do you want?”... “Hundreds!?!?! I can’t run flow simulation on hundreds of realizations! Give me just one, the one you want!” And it is up to the geomodeler to pick one. Of course, it is possible to pick just one. We can use the concept of ranking that we’ll cover in the next few paragraphs. But, is this acceptable? Providing only one to the engineer means that we suddenly ignore all the uncertainty we have identified with the geoscientists. The model is now completely deterministic, giving a false sense of certainty to the engineers, to the team and to the decision-makers of our companies.
How tempting it might be to send only one model to flow simulation; the team and the company will make more grounded decisions about the reservoir if the geological uncertainties are taken into account in flow simulation. We can’t just send one realization. We need to send several of them to sample adequately the geological uncertainty space. The question is to know which realizations.
The most common approach is to use the in-place volume, specific to each realization as a guide. The use of geomodels in reserve computations will be detailed in the next paper of this series. Hereafter, the topic is covered in very simple terms.
Figure 1 represents the base case realization of a sand-shale reservoir (a variation of the example used in section 6.2). The thickness of the shale separating the sands 2 and 3 is uncertain. It could be thinner (Figure 2A) or thicker (Figure 2B). In the first case, the in-place volume is higher than in the reference model, while it is lower in the second case (a thicker shale means less sand in sands 2 and 3). The in-place of the reference model is in the middle of the range. If a single model is sent to flow simulation, it might be tempting to send the reference case; if the in-place volume is an average of the in-place volumes of all the realizations, it is tempting to assume that this specific realization is also going to behave in an “average” way in terms of flow simulation. If we have hundreds of realizations and the two examples shown in Figure 2A and Figure 2B are two extreme in-places, then, following the same logic, we might send the three realizations to flow simulation. We might expect that the model showing a low in-place volume will show less production than the reference case which itself will be less performant than the model showing a high in-place volume.
Unfortunately, ranking realizations based on in-place volumes can be misleading because in-place volumes tell nothing about the connectivity in the reservoir. Connectivity is a key controller in flow simulation.
Let’s consider some new variations around the reference case. This time, the uncertainty is in terms of the continuity of the different shales. Do we really have continuous shales isolating completely the different sands, as in the reference case (Figure 1)? Or, are the shale discontinued and the sands connected? Figure 3 shows two scenarios where the sands are more (Figure 3A) and more (Figure 3B) connected. These two new realizations might have in-place volumes very similar to the in-place volumes of the reference case. Nevertheless, in terms of flow simulation, they will behave completely differently from the reference case. Without digging more about the consequence in term of flow simulation, we can easily imagine that the sands in the reference case will need to be produced independently one from the other, while connected sands might be produced as one block.
Now, we have five realizations. The reference case, (Figure 1 as well as upper central model in Figure 4), a realization of isolated sands with a lot of shale (Figure 2B as well as upper left model in Figure 4), a realization of isolated sands with less shale (Figure 2A as well as upper right model in Figure 4) and lastly, two models of connected sands, where the level of connection keeps increasing from model to model (Figure 3 as well as the middle and lower central models in Figure 4). Which ones shall we send to flow simulation? Changes in connection have a larger impact than changes in in-place volumes. It is preferable to send the reference case and the two models of connected sands. The reference case is expected to be a pessimist case in terms of flow simulation, because of the low level of connectivity, while the highly-connected model is expected to be the more optimistic scenario.
Figure 4 is the opportunity to see how information, provided by the geoscientists, is translated into geomodeling constraints which will lead to different results of flow simulation. On one hand, uncertainty in the sand proportions will lead to adding uncertainty in the input facies proportions used in geostatistical algorithms such as SIS (Sequential Indicator Simulation). This uncertainty will impact mostly the output range of in-place volumes. On the other hand, uncertainty in how the sands are connected will lead to adding uncertainty in the dimensions of the variograms used in geomodeling. The space of uncertainty in this simple example is the two-dimension space with sand proportion on one axis and connectivity on the second.
Quantifying the level of connectivity and understanding its correlation to flow simulation is still a topic of active research. A bibliographic review would be in order to understand how it is applied in the type of reservoirs studied by your team.
Another approach to ranking is to run streamline simulation (Figure 5) on as many of the geomodel realizations as possible. Streamline is a type of simplified simulation which runs much faster than full, true flow simulation. Streamline simulation could be used to evaluate how the different realizations are going to behave. From there, a few realizations are picked and sent to true, intensive flow simulation computations. More approaches exist to rank based on expected behavior in flow simulation. Geomodelers should discuss the topic with their reservoir engineers to see which one(s) seem(s) more appropriate to the project at hands.