Production Decline Curve Analysis — The Road Not Taken


Robert Frost’s poem “The Road Not Taken” concludes with the lines:

Nibbelink - fig 1 decline curve

“two roads diverged in a wood and I—I took the one less travelled by, and that has made all the difference.”

Having just returned from my Dartmouth Class of ’71 45th reunion I suppose I can be forgiven for tapping into Frost’s poetry, given his connection to my alma mater.

But this is where Decline Curve Analysis algorithms are these days—at a fork in the road, and the one you choose to evaluate your significant acquisition or divestiture candidates will make “all the difference”.

Whether you are a completions engineer who is trying to gauge the outcomes of new completion techniques, or you are reporting to your board of directors with forecasts of the cash flow of operated properties, getting your numbers right has never been more important in these times of tight margins.

With high initial decline rates and maybe long term transient flow characteristics, unconventional decline curve analysis can be especially challenging. Do it too early and you risk “pessimizing” your EURs with algorithmic assumptions that are too pessimistic about future behavior.

Nibbelink - fig 2-2 decline curve

Use an inappropriate algorithm—for example exponential vs hyperbolic—and your answers will be widely different. For the lease below, the difference is just over 440,000 BO and 1.7 BCF of EUR. That represents an unacceptable error range of production estimation.


Nibbelink - fig 3-2 decline curve

Or this?

Nibbelink - fig 4-2 decline curve

Our new approach within Production workspace has been to introduce probabilistic estimation into decline curve analysis, so that the algorithm we use is sensitive to outliers in the data and doesn’t rely on brute force curve fitting to a collection of points. Moreover, it calculates the probability of any EUR value actually being correct (P10,P50,P90).

Here’s the type curve for Delaware Basin Bone Springs wells that have produced at least 100,000 BO.

It would imply that production in the play is reasonably predictable.

Nibbelink - fig 5 decline curve

But Mother Nature is never this predictable. This well is a challenge to traditional decline curve analysis
Nibbelink - fig 6 decline curve

By pre-selecting  the Logistic Growth Model algorithm option, The Decline Curve analysis in Production Workspace computes bin volume estimates, assigns them a probability, and re-iteratively calculates fresh inputs into the algorithm to arrive at this:

Nibbelink - fig 7 decline curve

…with these P10-P90 probabilistic reserves.

Nibbelink---fig-8 decline curve

An “out-of-the box” (read: Black Box) algorithm from a third party calculates the EUR for this well at nearly 1,200,000 BO-1,770,000 (depending on method), which is nearly  2x-3X the P90  from Production Workspace Decline Curve analysis.

The point is this: No matter what software or consultant  you use to estimate EURs, it’s essential to know the underlying algorithmic models that are embedded in the software. It’s even more important to know, if you can, their error ranges.

Taking the wrong road at the fork can, at best, get you lost.

At worst, it can  lead to ruin…

Your Turn

What do you think? Leave a comment below.

The following two tabs change content below.

Mark Nibbelink

Enverus Co-Founder, Director of University Outreach. Before co-founding Enverus (formerly Drillinginfo) in 1999, Mark had a long career as a prospect geologist at Gulf Oil before beginning work as an independent geologist. Mark is responsible for quality control and data integrity. He received his Bachelor of Arts in geology and his master’s in geology and geophysics from Dartmouth College.