This is a summary of a paper to be published in the AAPG Bulletin, entitled “The battle of Frankenstein and Gilligan and the Law of Increasing Reservoir Complexification: what matters in 3-D reservoir characterization modeling?”
Here is the problem: your company is trying to assess the volume and producibility of oil in a newly discovered deepwater reservoir. Only a handful of wells exist in the new field, perhaps as few as three or five in a field miles across, and although the 3-D seismic is helpful, it cannot be used to accurately define the geometry and lithology of the reservoir at the production scale. Wells cost hundreds of millions of dollars each and a platform much more. Everything must be done optimally and correctly, or massive business losses could occur. Is there enough oil in the reservoir for production to be profitable? Can production rates be maintained, and for how long?
If you were the project manager, what would you do in such a situation? The growth and perhaps even the survival of your company is at stake. You are dealing with massive investments, and very little, but extremely costly data.
There are two general solutions to this problem.
Frankenstein’s Model
First, and most obviously, you gather every possible expert to analyze the data, however limited it is, and give them ample time to evaluate it. Next, using interpretation, interpolation and simulation based on the available yet limited dataset, you build a 3-D reservoir characterization model. Finally, you proceed with dynamic and economic modeling and come up with the best business decision. This approach, which focuses on using what is known about the reservoir, creates a deterministic model, which is referred to here as a “Frankenstein” model.
The term “Frankenstein model” was coined by Mark Williams in 2004, then a consultant reservoir simulation engineer, who noted that building an intricate 3-D Earth model often violated timelines for major capital projects, sometimes “killing” its creator, as did Frankenstein’s monster. Even still, the reservoir model could fail during flow simulation, resulting in the need to completely rebuild the earth model, which would take even more time, further slowing the progress of the project.
Additionally, lookback studies often showed that these carefully constructed models were simply wrong or significantly in error, regardless of the time and care spent in construction. Perhaps the reservoir was modeled as a sheet when in fact it was valley-confined. Perhaps the net sand volume of the reservoir was greatly different than the modeled volume. Perhaps the reservoir was more faulted than expected or contained wet fault compartments that could not be imaged in the 3-D seismic. Perhaps the permeability was much more heterogeneous than modeled.
The Gilligan Approach
If carefully modeling what is known about a reservoir isn’t the correct approach, what else can be done? Another approach to addressing the problem is to model what is not known about the reservoir.
Instead of using limited information to create a single 3-D earth model, why not build a suite of earth models portraying every possible reservoir geometry and architecture that might exist, effectively capturing uncertainty in reservoir character? Then, after building the models, use some quantitative wizardry to assign probabilities to each potential outcome. That is, using what is known about the reservoir as conditioning information, define and characterize what is not known about the reservoir: model what could be there, what might be there.
This modeling approach is considered probabilistic, and the models are called “Gilligan” models. I created the term “Gilligan models,” also around 2004, to represent the opposite of the Frankenstein approach. The creation of simple models might seem moronic to the Frankenstein modeler, who focuses on a super-detailed workflow. Oddly enough, the Gilligan approach can be done more quickly than the Frankenstein, because it creates a variety of simple solutions, though uses available conditioning data as well. An additional benefit of Gilligan modeling is that it presents testable scenarios. If the uncertainty is, for example, whether the reservoir is sheet-like or valley-confined, and if this difference is of economic significance, then management might be prompted to consider drilling an additional well or wells to address this uncertainty. The Gilligan workflow naturally leads to management of uncertainty. Frankenstein models do not characterize or address uncertainty, at least not directly.
Reservoirs Are Understood Backwards
To summarize, there are two end-member ways to characterize a reservoir in 3-D. The Frankenstein approach builds a model based on what is known about the reservoir. The Frankenstein approach addresses uncertainty using powerful interpretation, interpolation and simulation techniques to achieve what is believed to be, or hoped to be, the best and most correct answer. The Gilligan approach builds a model, or suite of models, based on what is not known about the reservoir. The Gilligan approach creates a spectrum of possible interpretations that can be further tested or used as possible outcomes.
The example discussed here has focused on the issue of deepwater fields, with very sparse wells, limited seismic data quality and enormous costs.
Can this concept of Frankenstein and Gilligan be applied to other new discoveries, other business situations, or even mature fields?
Let’s ask the question another way: Are there surprises in development or production geology?
Surprises are a manifestation of uncertainty. Unexpected results are surprising. If there was a fully developed uncertainty model that existed prior to drilling a well, surprises would be minimized: surprises would be reduced to confirmation of specific cases. Surprises occur at scales from drilling a single horizontal well, to drilling well patterns, to full field reservoir management, to frac’ing patterns in a tight well, to bidding on a property. In each of these cases, reservoir uncertainty is present at levels that could significantly impact economic or business decisions. In each of these situations, the role of uncertainty and how it impacts business value must be addressed. Building a single deterministic “Frankenstein” model may lead to a non-optimal business result, especially in light of uncertainty.
Kierkegaard wrote, “Life can only be understood backwards; but it must be lived forwards.” You can replace “life” with “reservoirs.”
Essentially, all reservoirs are characterized by uncertainty, even if they have been produced for a hundred years. Consider the fact that wells are spaced from hundreds to thousands of feet apart, wellbores are only inches wide, core is taken relatively sparsely, well logs are subject to all sorts of interpretation issues and seismic data can never resolve reservoir properties at the well log scale. Even the most mature reservoirs are sparsely sampled. To illustrate the issue of uncertainty in ultra-mature reservoirs, typical important uncertainties in such reservoirs are: 3-D distribution and definition of saturation, distribution of saturated geobodies, and even the distribution and definition of net sand. The science of reservoir characterization is taking the limited information we have, from wells, seismic and production, and creating a 3-D model using interpretation, interpolation and simulation. The role of uncertainty in reservoir characterization should never be taken lightly or assumed to be inconsequential.
“The only certainty is that nothing is certain,” wrote Pliny the Elder.
Uncertainty is pervasive in all reservoirs, though reservoir uncertainty clearly decreases with production of the reservoir, added wells and improved seismic quality (Figure 1). The role of reservoir characterization in the oil and gas industry is always dealing with the relationship of what we know and what we don’t know, and how that impacts business decisions. There is always a decision involved with regard to whether a deterministic approach or a probabilistic approach is preferable to address the economic issue at hand. In general, uncertainty should always be included in a study of reservoir character.
Increasing Complexity
Another interesting feature of oil and gas reservoirs that should be acknowledged in reservoir characterization studies is that reservoirs tend to become more complex as information is provided by drilling additional wells and through production. This is referred to here as the “law of increasing reservoir complexification.” That is, a 3-D earth model built for an exploration play, or for early development, will likely not be useful as the reservoir is developed, more wells are drilled, reservoir trends are established and quantified and new faults are found. The Frankenstein approach cannot be valid in the case of increasing complexification.
To summarize, sometimes the Frankenstein approach is better, sometimes the Gilligan but the difference lies in the nature of the business problem. If uncertainty is the main business focus, then the Gilligan approach is clearly superior. If group discussion and visualization is the main business focus, or reservoir uncertainty is considered minimal, then the Frankenstein approach might be more useful.
Finding the correct solution to the business problem is commonly referred to as the “Goldilocks approach,” and usually Goldilocks is present somewhere between Frankenstein and Gilligan. But more fundamentally, the goal of 3-D earth modeling and reservoir characterization is always to add value to a business decision. In some cases, building a highly deterministic model adds value. In most cases, however, the characterization and quantification of uncertainty has the greatest economic value, for all reservoirs. If your company or business unit relies on Frankenstein 3-D earth models, consideration of the Gilligan approach would likely add value, on every reservoir, from appraisal to ultra-mature, from the spotting an individual well to full-field management.