Marcellus Wells: Ultimate Production Accurately Predicted From Initial Production

In this work, the authors perform automatic decline analysis on Marcellus Shale gas wells and predict ultimate recovery for each well.

Gas well location in the Marcellus Shale
Getty Images

In this work, the authors perform automatic decline analysis on Marcellus Shale gas wells and predict ultimate recovery for each well. A minimal model is used that captures the basic physics and geometry of the extraction process. A key discovery is that wells can have their estimated ultimate recovery (EUR) predicted early in life with surprising accuracy.


There are many challenges to production in the Marcellus, ranging from the regulatory and price environment to the highly variable responses of rock to hydro­fracture treatment. The complete paper’s primary goal is to generate scenarios for future gas production from the Marcellus by use of varying gas prices and other assumptions.

The first step in this process is determining technically recoverable reserves, which requires deciding how much each well will produce over its lifetime. To do this, the authors use recovery-­factor curves that describe total recovery in terms of two parameters: total gas in the stimulated reservoir volume (SRV) and the characteristic time to boundary-dominated flow (BDF).

Most wells produce following these recovery curves, but not all. Upon examining production data, the authors found that several hundred wells have very slow decline for the first several years, slower than the production decline expected from linear flow. In discussions with multiple operators, the authors found that this is attributable to production choking.

There have been several analytical and numerical models developed to predict production from horizontal, hydrofractured shale gas wells. This paper uses the scaling model developed by Patzek et al. (2013) to build recovery-factor curves to describe production for Marcellus wells. Fitting production of 5,275 wells, 404 wells are found that experience boundary-dominated flow. Among these wells, 175 are horizontally drilled. The original gas inside the SRV for these wells is compared with initial production, and a very high correlation is found. This correlation is used to estimate gas production for the remaining wells. Among the entire set of wells, an average EUR of 3.6 Bscf is estimated, with the middle 60% having EURs between 6.6 and 1.8 Bscf. The model for flow to the wellbore is detailed in the complete paper.


Production reports from wells in Pennsylvania are sparse, which makes decline analysis from publicly accessible data a difficult endeavor. Most of the analysis used in this paper relies upon having monthly production data, so semiannual and yearly production had to be converted into monthly production.

Many Marcellus wells in northeastern Pennsylvania produce at rates far higher than those of any other shale gas field, with EURs predicted to be twice that of the best wells in the Haynesville Shale. In addition, there are infrastructure limits in some parts of the field because pipeline capacity has not kept up with production potential in Pennsylvania. In order to protect gathering equipment and not overwhelm available pipeline space, the production of these wells for the first 1 to 4 years is limited by use of a choke at the surface. This limits how quickly the flowing tubing pressure drops to its final value.

There are two sweet spots in the Marcellus. In northeastern Pennsylvania, production is driven by a thick formation of dry gas. In southwestern Pennsylvania and West Virginia, production is driven by high porosity filled with high-Btu gas. There is also liquids production in southwestern Pennsylvania.

Desorbing gas is another contributor to production. There is some debate about whether gas adsorption follows the Langmuir isotherm or the ­Brunauer-Emmett-Teller (BET) isotherm, but Langmuir isotherms are far more readily available for the Marcellus and have been used as the industry standard.

The BET isotherm allows desorption in large quantities at far higher pressures than the Langmuir isotherm, owing to the ability of gas molecules to adsorb on multilayers with the BET model; the Langmuir model allows only a single layer to adsorb.

Regardless of which isotherm is better suited for describing gas desorption, desorption contributes in several ways to gas production. First, it delays BDF. Second, production declines more slowly once it has entered BDF because gas desorption acts as an additional production drive, allowing pressure to fall at a slower rate as gas is produced than it otherwise would. Both of these effects increase the EUR.

The authors create recovery-factor curves with 500-psi fracture-face pressures, initial reservoir pressures varying from 2,000 to 13,000 psi, Langmuir isotherms, and the two gas compositions specified in the complete paper. The curves were calculated by solving the pseudopressure diffusion equation and using Darcy’s law at the fracture face.

Next, the recovery-factor curves are fitted to cumulative production from wells, ignoring the first 4 months for unchoked wells. For choked wells, production is fitted to the period after they come off the choke, or at the end of their third year if they have not yet come off the choke, by use of the following method.

To predict choked wells accurately, wells were fitted only after they returned to normal decline. The first step in this process is identifying choked wells. A well is choked if the second year’s total gas production is more than 75% of the first year’s production or its third or fourth year’s production is more than 80% of the production of the preceding year.

There are 304 wells that have been choked for 3 years; cumulative-­production fitting starts at Month 34. There are 869 wells that have been choked for at least 2 years, and production fitting starts at Month 22. For wells that have not yet come off choke, constant production at the last reported rate is assumed until they reach 36 months of age; then, they are allowed to decline normally following the appropriate recovery-factor curve.

A recovery-factor curve is selected for each well to match its reservoir pressure and presumed fluid composition. Then a least-squares minimizing package is used to minimize the error function, making it as close to zero as possible.

The authors found 404 wells in BDF. Among those wells, there is a very high correlation between initial productivity and total gas in place.

Fitting leads to the recognition that gas in place scales linearly with initial production. Each well’s recovery has two fitting parameters, and with gas in place and initial productivity known, it is a simple matter to calculate the characteristic time to BDF. With the latter parameter and the SRV determined from each well’s initial productivity and the regression equation provided in the complete paper, production can be forecast. If one uses the assumption that gas in place actually scales linearly with initial production, then this means a constant characteristic time of 4 years (48.4 months). Keep in mind that this does not mean that all wells in the Marcellus will enter BDF at 48.4 months. There is a wide probability distribution around this value, but the average value remains 48 months regardless of initial productivity. Stochastically, it is predicted that the average time to BDF will be 4 years for the field, but that of an individual well cannot be predicted until the interference is observed.


Ultimate recovery has been estimated for each Marcellus well that has more than 18 months of production data. The EUR and time to interface for each predicted well are given in Fig. 1. The mean well is expected to produce 3.62 Bcf over its 25‑year lifetime and have a time to recovery of 3.9 years. The standard deviations around these averages are 3.62 Bcf and 0.87 years, respectively.

Fig. 1: EURs and time to interference for each well in the Marcellus. The upper histogram shows time to interference and is cut off at 200 wells. The highest bar, at 3.9 years, is responsible for approximately 3,500 wells. The right-hand histogram describes EUR. Orange dots and bars denote wells in BDF, while blue dots and bars represent wells that are not yet interfering.

The authors forecast production for 5,275 horizontal Marcellus wells in Pennsylvania and West Virginia. There are two groups of wells that were fitted: those that have and those that have not entered BDF. Among the wells that have entered BDF, the most influential predictor of gas in place, and therefore EUR, is initial production. This knowledge can be leveraged into making production predictions by regression onto a very simple equation. From this, production was predicted for wells that have not entered BDF. The median EUR for all wells was 3.9 Bcf.

This has several implications. First, wells in a mature field can have their production predicted with reasonable uncertainty from an early age. Second, and more importantly, the time to BDF does not vary systematically with initial production. Time to BDF depends upon hydraulic diffusivity and distance between adjacent producing fractures, so this likely means that the distance between fractures does not systematically effect time to BDF, nor vice versa. If this is the case, then the largest driver of EUR is the extent of the producing fracture network.

Previously, it could be reasonably assumed that generating a more-extensive fracture network would increase the initial production but decrease time to interference, leading to roughly the same EUR. However, that trend was not observed in this study.

This result could mean that there is a minimum interfracture distance that is achieved by many wells in the field. Further fracture creation, then, leads to a larger, not more densely fractured, SRV. This, in turn, leads to larger initial production with the same time to BDF and, therefore, to a larger EUR.


Patzek, T.W., Male, F., and Marder, M. 2013. Gas Production in the Barnett Shale Obeys a Simple Scaling Theory. Proc., National Academy of Sciences 110 (49): 19,731–19,736.

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 180234, “Marcellus Wells: Ultimate Production Accurately Predicted From Initial Production,” by Frank Male, Michael P. Marder, John Browning, and Svetlana Ikonnikova, The University of Texas at Austin, and Tad Patzek, King Abdullah University of Science and Technology, prepared for the 2016 SPE Low-Permeability Symposium, Denver, 5–6 May. The paper has not been peer reviewed.