"Non-Uniqueness" of Reservoir Simulation

When someone asserts that reservoir simulation generally isn't worth the time and effort because the results are "non-unique", they show to me how shallow their experience and insight really are.

First and foremost it tells me that they have never tried to come up with one of those (supposedly) many solutions. It can be $@<<> hard to find even a single history match! Sine the person making the complaint has little or no experience with modeling, then it is easy to see why they do not comprehend good model-building or the value it generates.

In a century of our discipline, we have only identified six methods for the estimation of reserves. Simulation incorporates and thus constrains the answer by all of the relevant physics and thus has fewer degrees of freedom in aggregate. It is true that simulation requires many inputs and that some much have large uncertainties. On the other hand, though, sits Arps decline curve analysis which has very few constraints. The history-match and then sense-check of a simulation against actual history uses the same processes as Arps, AND it honors a cabinet of other parameters about the nature of the rock and the fluid which ARE known. Rate-transient analysis has more constraints and thus more uniqueness than historical decline analysis, and it may feel more comfortable. But even RTA is constructed from a model, and that model with more simplifications than inherent to full reservoir simulation.

Moreover, we can sometimes quantify the effects of the non-uniqueness. The power of new algorithms, software and hardware has enabled work processes to generate sometimes multiple history-matches, and below is just one examples showing the distribution of forecast results when multiple history-matches have been achieved. Interesting to note that the distribution of results are more narrow than for decline curve analysis.


Regardless of the inherent uncertainty, even with a single history-match or none at all, a prudent process of simulation can benefit reservoir development in ways no other technique can.

All models are wrong, but some are useful.

So goes the aphorism attributed to mathematician George Box. The apologist made the assertion in the context of expounding on the wisdom of what makes for a good model, in this case "parsimony."

Creating a model requires the engineer to examine consciously and quantitatively every characteristic of a reservoir system; every aspect becomes an input to the model. Then, in the running and testing of the model, the user can calculate and not merely guess at the significance of each uncertainty and about what truly matters to our plans for the reservoir. Modeling in general and reservoir simulation in particular should not be treated as a plug-and-chug, black-box technique. It is a tool for understanding not an algorithm for an answer. Almost regardless of the outcome of the simulation, the process of deconstructing and testing the reservoir leads to greater insights and thus better development decisions.

It should also not be treated as the world's most expensive curve-fitting tool. Excessive elaboration is, in the words of Dr. Box, "often the mark of mediocrity." Less pejoratively, excessive elaboration is a rookie mistake which violates established wisdom of model-building. The insight of William of Occam has endured since the 14th-century, namely answers should favor solutions which require fewer prerequisites to be true. To find and focus on the truly key parameters, to explain the dynamics simply and elegantly, those are the requisites to a successive development campaign.

In the ultimate step, simulation allows for a well or a field to be developed virtually in any number of ways before it is developed in real life. It is by far the most economical venue for experimentation and optimization in this industry without prototypes or do-overs.

Perhaps the uninsightful use of simulation has earned a poor reputation and created dismissive attitudes. Perhaps software designers have intimidated engineers and priced the products out of reach. For sure, simulation is not the appropriate tool in every case, but it probably should be used much more often than we do.

(For additional reading, the pithiest resource I know in our industry is the classic paper by Dr. Keith Coats, "Use and Misuse of Reservoir Simulation Models". Similar concepts of model-building are also explored in "Judgment in Probabilistic Analysis".)

Please subscribe to my newsletter for updates and articles!