- Shale gas has revolutionized US energy including prices, consumption, & emissions.
- Key questions remain including environmental concerns & extraction efficiencies.
- New discoveries are identified through data mining & analysis of 20,000 wells.
- Findings include learning-by-doing, refracturing, & long-term tail production.
- We hypothesize that manipulating tail production could re-revolutionize shale gas.
Richard S. Middleton, Rajan Gupta, Jeffrey D. Hyman, Hari S. Viswanathan
Los Alamos National Laboratory, PO Box 1663, MS T003, Los Alamos, NM 87545, USA
Received 23 January 2017 Accepted 15 April 2017
Shale gas and hydraulic refracturing has revolutionized the US energy sector in terms of prices, consumption, and CO2 emissions. However, key questions remain including environmental concerns and extraction efficiencies that are leveling off. For the first time, we identify key discoveries, lessons learned, and recommendations from this shale gas revolution through extensive data mining and analysis of 23 years of production from 20,000 wells.
Discoveries include identification of a learning-by-doing process where disruptive technology innovation led to a doubling in shale gas extraction, how refracturing with emerging technologies can transform existing wells, and how overall shale gas production is actually dominated by long-term tail production rather than the high-profile initial exponentially-declining production in the first 12 months. We hypothesize that tail production can be manipulated, through better fracturing techniques and alternative working fluids such as CO2, to increase shale gas recovery and minimize environmental impacts such as through carbon sequestration.
Shale gas technology—a combination of horizontal drilling, high-pressure fracturing, multiple fracturing stages, and more efficient working fluids—has opened up vast reservoirs of shale gas for cost-effective production . Shale gas and hydraulic fracturing has had a profound impact on US prices , carbon intensity, and energy independence with reserves projected to last for around 100 years , , , as well as energy security across the globe, , , , , .
The increased availability of natural gas from shale, which produces roughly half the CO2 than coal for the same heat output, is facilitating the transition from coal-fired to gas-fired power generation and is consequently responsible for US emissions falling to 1990 levels . Natural gas also plays an important role as a bridge fuel to renewables since gas power plants can respond quickly to changes in load and renewable generation unlike coal-fired and nuclear plants .
In addition, depleted shale gas formations offer significant opportunity for CO2 storage due to potentially enormous capacity, ability to sequester or trap CO2 physically, and reduced sequestration costs by leveraging existing infrastructure , . However, the wide spread use of hydraulic fracturing (often referred to as just fracturing or “fracking”) is still plagued by barriers such as inefficient extraction ,  and environmental concerns such as drinking and groundwater contamination , , , , , induced seismicity due to wastewater disposal , , , and fugitive emissions , .
Improving extraction efficiency through enhanced fracturing and unconventional fracturing fluids  could reduce the environmental footprint of shale gas production through reduced water use and improved disposal practices, as well as fewer infrastructure impacts. The findings presented in this data-driven study illustrate several key recommendations that could potentially improve extraction efficiency while reducing environmental impacts.
Through novel data mining coupled with subject matter expertise, we quantitatively explore the history of hydraulic fracturing using production data from 20,000 wells over 23 years in the Barnett Shale (Fig. 1). The Barnett Shale play is an ideal microcosm to explore the shale gas revolution because it contains the entire history of the shale gas development in the United States: early pioneering technology, rapid technological development, massively-increased shale gas production rates, dramatic fluctuations in oil and gas prices and its impact on investment, and the recent decline in field-level production.
Fluctuations in hydrocarbon price significantly impact greenhouse gas emissions and sustainability . Using publicly-available production data—compiled and released for the first time in a single dataset in the Supplementary Information (SI)—we have uncovered four critical aspects of development, which drove shale gas technology and exploration. Barnett Shale gas production displays a learning-by-doing process between the first technically successful wells in 1998 to mature technology around 2011.
This developing discipline resulted in a more than doubling in per-well production rates between 2007 and 2011 alone. These increased efficiencies were largely the result of a trial-and-error approach due to a lack of utilizing first principles to guide production strategies. The refracturing phase—when existing conventional and poor-performing wells were refractured between 2000 and 2004—demonstrates the effects of bringing new technologies online and into the field. This refracturing phase resulted in older, pre-1998 wells, to out-perform wells that were drilled between 1998 and 2006.
Fig. 1. Map of the Barnett Shale formation and production. The map illustrates the county-level spatial distribution of shale gas production (main map) and average per-well production rate (inset). Production is heavily focused on Tarrant County (the city of Forth Worth) and surrounding counties. The average well production curve for the Barnett shows the typical production pattern: high initial production (just over 1650 Mcf/d for the Barnett) followed by an exponential decline and an extended production tail. The average production curve is based on more than 18,000 wells over a 23-year period (1993–2015). The map was made using ESRI’s ArcMap 10.3 software (http://desktop.arcgis.com/en/arcmap/) and the production data is available in the supplementary information.
In fact, the data over the same time period show that refractured existing wells were as productive as virgin wells; an observation that has major implications for deploying future technologies to existing wells around which vast amounts of hydrocarbon still remain. We identify unique characteristics of high- and low-producing wells, showing that despite increasing extraction efficiencies with time, high- and low- producers behave fundamentally differently independent of the year examined.
High-producing wells typically produce an order of magnitude more than low performing wells in the Barnett. We hypothesize that high-producing wells are the result of hydraulically-generated fractures becoming highly connected to the pre-existing natural fracture network . Finally, although industry typically focuses on production in the first 12 months, motivated by a quick return on the capital-intensive investment, 75% or more of well production in the first ten years actually occurs after the first 12 months showing the importance of long-term tail production.
We hypothesize that the tail in production is controlled by a different set of geochemical, geomechanical, and hydrologic processes. These processes can be manipulated with alternate working fluids such as supercritical CO2, or reservoir management strategies such as cyclic pressure injection, huff’n’puff, or refracturing  to further increase long term production. Thus, understanding and enhancing the tail production is a potentially significant research area.
Results and discussion
History of the Barnett
Shale gas production in the Barnett formation is unevenly distributed across more than 20 counties in north-central Texas, with production heavily focused in and around the Fort Worth urban area (Fig. 1). Fossil energy resources, e.g., coal; oil; and natural gas, in the region were identified and developed long before the emergence of large municipalities. Consequently, modern urban shale gas and oil production has faced numerous environmental , , , political , and public awareness ,  challenges, including a production moratorium in the Marcellus formation , that conventional oil and natural gas never had to address.
The history of shale gas exploration in the Barnett formation is summarized in the field-level production data shown in Fig. 2 with multiple key points indicated in the comprehensive figure caption. Very few wells were drilled before 2000 (blue bars, primary y-axis, in lower chart in Fig. 2) and these wells—a combination of conventional-like, vertical fractures, and emerging horizontal fractures—were typically poor performing (red line, secondary y-axis, in upper chart of Fig. 2). The first technically successful wells in the Barnett were drilled in 1998 .
A moderate increase in the number of new wells drilled occurred for the next few years, though the average well production did not improve until around 2004. The period of 2004–2011 saw a dramatic increase in both the average per-well production and in the number of new leases, which resulted in a quadrupling of total field production (grey bars, primary y-axis, in upper chart of Fig. 2). This dramatic increase was driven by a combination of high natural gas and oil prices (orange lines, secondary y-axis, in lower chart in Fig. 2) and improving technology.
Fig. 2. History of the Barnett Shale. Top chart: total field production (grey bars; primary y-axis) and average cumulative well production in the first 12 months (red line; secondary y-axis). Bottom chart: number of new producing leases for each month (blue bars; primary y-axis) and normalized oil (straight) and natural gas (dotted) prices (orange lines; secondary y-axis). Prices for oil (Western Texas Intermediate price) and natural gas (commercial price) were individually normalized between 0 and 1 based on their minimum and maximum prices in the period 1993–2015. A full description of key events—labeled A to L in the chart—is available in Fig. S1 in the SU. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
The global recession and the rapid decline of oil and gas prices in 2008 and 2009 resulted in the number of new leases falling by more than two-thirds between April 2008 and September 2009. The recovery in oil prices in 2011 led to a mini resurgence in new leases. However, the opening up of other formations such as the Marcellus along with persistent low and declining natural gas prices led to reduced investment in the Barnett field, even though individual well efficiency was as high as it has ever been.
Extraction of shale gas via hydraulic fracturing benefitted from disruptive technologies innovation—a combination of high-pressure fracturing, horizontal drilling, multiple fracturing stages, and improved low-viscosity working fluids—that led to a remarkable improvement in extraction efficiencies. This innovation occurred in the middle of an extended learning-by-doing process in the Barnett (Fig. 3) that can be divided into three key time periods: early exploration (1981 to 1998), technical success (1999 to 2004), and high-volume production (2005 to present).
Learning-by-doing increased well performance across the three key time periods, though with diminishing returns from 2011 onwards (Fig. 2). Arguably, the Barnett would require another wave of disruptive technology innovations, such as fluid-free fracturing or non-aqueous fracturing fluids , in order to transform shale gas extraction efficiencies. Such a revolution would also impact national hydraulic fracturing practices where the majority of fields are either flat or in decline  (Fig. S2 in the SI).
Fig. 3. Stacked histogram of production in the Barnett Shale formation. The chart is broken down by the number of wells (primary vertical axis; blue) and total field production (secondary y-axis; red). All production data are based on production in only the first 12 months of each well. The x-axis is divided into 81 production bins representing the amount of gas produced in the first 12 months of production; 2015 is omitted to ensure all wells had at least 12 months of production, leaving a total of 18,657 wells. The number of wells in each production bin (primary y-axis) is the sum of three time periods: “Learning” (wells drilled before 1998; dark blue), “Technical success” (1999–2004; medium blue), and “High-volume production” (2005–2014; light blue). The total field production (secondary y-axis) includes the entire 1993–2014 period, and is the sum of the production of all wells in a given bin. The green numbers at the top of the chart show the number of wells in each quintile of production; for example, the top 20% of total field production is driven by 1383 wells (7% of all wells). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 3 also illustrates that total field production is largely driven by higher-performing wells. For example, the highest number of wells are in the 100,000–125,000 Mcf bin (peak in the blue bars in Fig. 3), whereas the peak in overall field production (red line in Fig. 3, secondary y-axis) occurs at the 350,000–375,000 Mcf bin. Similarly, the top 20% of field production is driven by only 7% of the wells (1383 wells in green text at the top of Fig. 3).
In addition to field-level data, individual wells corroborate the learning-by-doing concept. For example, production in first month of a representative well jumps from approximately 1000 Mcf/d in 2004 to 1400 Mcf/d in 2005 and 2100 McF/d in 2011 (Fig. 4). Although average production rates plateau around 2011 (Fig. 2; red line), top-performing individual wells still exhibit increased extraction efficiencies (Fig. 4).
Fig. 4. Evidence for rapid learning-by-doing and success of refracturing. The representative wells were selected from a set of the top five performing wells in the chosen years (1993–2015). The names of the seven wells/leases can be found in the SI Workbook. Each letter-in-circle refers to a different event: A: The few wells producing in the early 1990s perform poorly, especially up until refracturing post-1998. B: Wells in the mid-1990s illustrate marginally improved performance. C: By 1998 wells show clear signs of successful virgin fracturing. D: Refracturing of early, low-producing wells leads to significant relative production increases. Here, a well that was already producing in 1993 (A) sees a production increase from 325 Mcf/d (peak production 1993–2001) to 1550 Mcf/d post-refracture in 2001. E: A successfully-fractured well in 1998 (C; 1500 Mcf/d peak production) is refractured in 2003. The refracture produced both a higher peak (1725 Mcf/d) and more production in the tail. F: By the mid-2000s, virgin-fractured wells perform very well in terms of initial and tail production. G: Like the two previous refracture examples (D and E), this refracture sees a significant increase in peak production and an elevated tail in production compared with the initial fracture (F). H: This 2009 well has yet-again improved initial production with evidence of stronger production in the tail. I: Although average well production asymptotes from 2011 onwards, high-performing wells still show significant improvement. J: 2014–2015 wells are still high-producers, though there are few wells being drilled compared to the late 2000s.
Learning-by-doing is even more prominent when wells are grouped by the first year they started production (Fig. 5, left panel). For instance, the average 2007 well took seven years to produce a billion cubic feet (1 Bcf) whereas wells in 2011 took only three years (Fig. 5, right panel). Starting in 2011, the production curves become indistinguishable as shown in Fig. 5, suggesting fewer technological improvements or better technologies being cancelled out by fewer production sweet-spots.
Fig. 5. Average well daily (left) and cumulative (right) production sorted by starting year. Wells starting in 1993 are excluded.
Our data mining analysis revealed that refracturing existing wells with new technology can transform them into high-performing wells with production characteristics of a high-performing virgin well. This observation has profound implications in the potential revitalization of the hundreds of thousands of shale gas wells across the United States. Refracturing existing wells drastically reduces environmental impacts by using the existing footprint.
Existing wells in the Barnett started to be refractured around 2000 and this procedure continued until around 2004 and production increase is evident in the humps in Fig. 5 (left panel) and the production hump in the average Barnett well production (inset of Fig. 1). Refracturing has a clear benefit both in the averaged-by-year data (Fig. 5) as well as individual cases (jumps at D, E, and G in Fig. 4). For example, the wells first prior to 1999 that were refractured in 2000–2004, cumulatively outperform virgin wells first drilled between 1999 and 2004 (Fig. 5, right panel). The refracturing likely stopped by 2004 because existing profitable candidates for refracturing were exhausted, especially since relatively few wells were drilled before 2000.
An important observation is that the production from 1994 to 1999 wells did not hinder post-refracturing production. Quantitatively, the wells refractured between 2001 and 2003 perform like virgin fractured wells (Fig. 5, right panel) even though they had already been producing for several years. This underscores an exciting potential for future fracturing technologies—such as using non-aqueous working fluids including supercritical CO2 and natural gas —being applied to wells that have reached a low-level asymptotic production.
Refracturing is particularly important because current fracturing technologies leave behind the majority of shale gas resources. For example, the average US shale play recovers only around 13% of original gas-in-place, with the Barnett likely around 6–8% , , . Thus there is massive potential for restimulating existing wells, particularly once oil and gas prices rebound. Moreover, restimulating existing wells is cost effective because it eliminates the capital costs of a new well while providing a smaller environmental footprint.
Unique characteristics of top- and bottom-producing wells
There is a substantial difference in performance between high- and low-producing wells, which has far-reaching implications for field-level production and increasing our understanding of the mechanisms that control the efficiency of hydraulic fracturing. Mean field production is dominated by a few high-performing wells. Partitioning the data into quintiles provides a more comprehensive and informative view of production.
Fig. 6. Average cumulative per-well production by quintile for wells starting production in 2006–2015. The ten years of data are split into quintiles ranging from high-performers (top 20%; Q1) to low-performers (bottom 20%; Q5). The top 20% refers to the top 20% performing number of wells (i.e., same number of wells in each quintile) based on the initial 12-month production. The chart (left) shows cumulative production curves. Fig. 7 shows the quintile plots for the years 2006 through 2015; each of these plots demonstrates the same relationship as here. The table (right) shows: (A) Average cumulative production for wells (Mcf) in each quintile for months ranging from 3 to 120 months (i.e., 10 years). For example, a well in the 0–20% quintile (Q1) cumulatively produces 802,980 Mcf of gas by month 12. (B) Fraction of production in each month. For example, wells in the 0–20% quintile (Q1) produce 33% of their production by month 12. (C) The percentage of production across all quintiles that is produced within each quintile. For example, wells in the 0–20% quintile (Q1) account for 42% of all production by month 12.
The top-performing wells not only produce significantly more gas than the other quintiles, they exhibit different scaling behavior for short- and long-term behavior Fig. 7 (left panel). We use two mathematical models to show that high-performing production rates are fundamentally different from low-performing wells. The observed daily production (Mcf/d) for the five quintiles are fit using a power law and a two-exponential model Fig. 7 (right panel); see Methods and SI for more details.
Fig. 7. Observed (left) and modeled (right) production curves by quintile (2006–2015). The ten years of observed daily production (Mcf/d)—left panel—show the characteristic exponential decline for all quintiles. The top 20% wells, on average, produce at seven-to-ten times the rate compared to the bottom 20% in the first year, dropping to four to five times the rate after ten years. By the tenth year of data, only wells that started producing in 2006 are represented and the production curves become noisier. The modeled production data—right panel—are fits to the observed records using a (i) power law (dotted line) and (ii) a two-exponential model (dashed line) laid on top of the original production data (solid line). Only the top, middle, and bottom quintile are displayed in color for visualization purposes. The parameters for both models are presented in Fig. S3 in the SI.
This analyze suggests that production from high- and low-performing wells are driven by different underlying physical mechanisms. Karra et al.  hypothesized that early production from a top producing well is dominated by large hydraulic fractures that effectively connect the well to the pre-existing fracture network. The mechanisms that control hydraulic fracturing have been extensively studied and are likely the key to enhancing shale gas recovery , . Validation of these mechanisms through focused research is critical to increasing the probability of developing high-producing wells, and could be achieved through coupling field data with mechanistic modeling .
Contrary to common perception, total shale gas production from a well is actually dominated by long-term tail production, rather than the initial flush. All wells in the data set reach an asymptote in production where they continue to produce for many years at a flat or slowly-declining rate. Thus, the cumulative production is mostly made up of production after the first 12 months. For example, over a 10-year period Barnett wells produce 22% (low-performing wells) to 33% (high-performing wells) of their gas in the first 12 months with the remaining volumes coming from the production tail (Fig. 6, right panel).
That high-performing wells produce so much more gas in the first 12 months suggests the presence of different physical mechanisms governing production from high and low producing wells. More importantly, as shown above, high-producing wells outperform lower-producers during all stages of production including the extended tail production, perhaps indicating that a highly connected fracture network not only increases short-term production but also late-term production and strongly impacts the profitability and efficiency of a well .
Therefore, research focusing on tail production could massively increase shale gas extraction and re-revolutionize hydraulic fracturing. Typically, shale gas producers focus research on early production to get a return on capital investments as quickly as possible, move to a new lease, and then profitably sell-on the well rights.
The dominance of tail production in terms of overall production indicates a research gap that should be addressed by less short-term profit-driven institutions including academia and the federal government. Key gains of increasing tail production include enhancing longer-term profitability, increased energy security, and a reduction in CO2 emissions. One possible mechanism for increasing tail production is enhanced desorption as it alone could lead to a 20–30% increase in production over 5 to 10 years .
Depleted shale gas fields have been identified as potential repositories for sequestering CO2 due to their enormous storage capacity to potential to safely store CO2. Moreover, their use would reduce costs and logistics by using existing infrastructure such as well bores, pipelines, and access facilities , , . The majority of the storage potential is predicated by assuming that CO2 will swap with adsorbed methane in the formation. However little, if any, research has focused on linking sequestration potential with production metrics.
For example, do high-performing wells offer preferential storage since they have created a larger, better-connected fracture network that increases available storage space? Or are they poor candidates because a higher proportion of the methane has been produced or because high-performing wells are likely to continue being more profitable as gas-producers rather than storing CO2? Contrariwise, are low-performing wells more likely to be classed as depleted and abandoned by their operator, thereby becoming available for CO2 sequestration.
Or, do these low-producing wells offer lower sequestration potential due poorer connections with the pre-existing formation? In the case of the Barnett, the data indicate that wells continue to produce for decades, with the majority of total production in the tail.
Consequently, it is not clear when a shale reservoir should be considered depleted or when sequestration economics would provide operators sufficient incentive to store CO2 rather than produce natural gas. Questions also remain regarding the long-term suitability of storing CO2 due to issues such as fractures closing up or being flooded with water, as well as the potential absence of trapping mechanisms such as an impermeable caprock.