The exploitation of unconventional gas reservoirs has become an ever increasing component of North American gas supply. The economic viability of many unconventional gas developments hinges on effective stimulation of extremely low permeability rock by creating very complex fracture networks that connect huge reservoir surface area to the wellbore. In addition, gas desorption may be a significant component of overall gas recovery in many shale-gas reservoirs. The widespread application of microseismic mapping has significantly improved our understanding of hydraulic fracture growth in unconventional gas reservoirs (primarily shale) and led to better stimulation designs. However, the overall effectiveness of stimulation treatments is difficult to determine from microseismic mapping, as the location of proppant and distribution of conductivity in the fracture network cannot be measured (and are critical parameters that control well performance). Therefore it is important to develop reservoir modeling approaches that properly characterize fluid flow in and the properties of a complex fracture network, tight matrix, and primary hydraulic fracture (if present) to evaluate well performance and understand critical parameters that affect gas recovery. The paper illustrates the impact of gas desorption on production profile and ultimate gas recovery in shale reservoirs, showing that in some shale-gas reservoirs desorption may be a minor component of gas recovery. In addition, the paper details the impact of changing closure stress distribution in the fracture network on well productivity and gas recovery. In shale-gas reservoirs with lower Young's modulus rock, stress dependent network fracture conductivity may significantly reduce ultimate gas recovery. The paper includes an example that contrasts the application of numerical reservoir simulation and advanced decline curve analyses to illustrate issues associated with conventional production data analysis techniques when applied to unconventional reservoirs. Selected examples from the Barnett shale are included that incorporate microseismic fracture mapping and production data to illustrate the application of the production modeling to evaluate well performance in unconventional gas reservoirs. This paper highlights production modeling and analysis techniques that aid in evaluating stimulation and completion strategies in unconventional gas reservoirs. Introduction Gas shales are organic-rich shale formations and are apparently the source rock as well as the reservoir. The gas is stored in the limited pore space of these rocks and a sizable fraction of the gas in place may be adsorbed on the organic material. The natural gas resource potential for gas shales in the USA is estimated to be from 500 to 1,000 Tcf (Arthur 2008). Typical shale gas reservoirs exhibit a net thickness of 50 to 600 ft, porosity of 2–8%, total organic carbon (TOC) of 1–14% and are found at depths ranging from 1,000 to 13,000 ft. The success of the Barnett Shale has illustrated that gas can be economically produced from rock that was previously thought to be source and/or cap rock, not reservoir rock. This revelation has led to the development of many other shale-gas reservoirs, including the Woodford, Fayetteville, Marcellus, and the Haynesville (Figure 1). Besides increasing natural gas prices (until recently), the economic development of many shale reservoirs was made possible through improved stimulation techniques and horizontal drilling.

Enhanced oil recovery in unconventional plays has been a focus of many E&P operators to increase recovery factors. Conventional displacement-based secondary (e.g. water flooding) and tertiary EOR methods are not viable options due to their low injectivity in these ultra-low permeability formations. As such, huff-n-puff (HnP) EOR techniques involving field gas injection may be the most effective EOR methods to increase the recovery factors from these shale formations. Several numerical reservoir simulation studies have showed the efficiency of CO2 HnP process in shale and tight formations (Yu et al, 2014); however, these studies show little to no field results to support the simulation predictions. This paper describes the conceptual development of the reservoir simulation models to investigate the viability of gas injection HnP in unconventional reservoirs and the results of applying those methods in Eagle Ford field test sites. Single-well and multi-well models with multi-stage hydraulic fractures were constructed and history matched using their primary production period performance. Sensitivity studies were conducted on well communication behavior/impacts, injection gas compositions, injection rates, injection/production cycling, and reservoir fluid types to optimize the pilot project well location(s) and to inform development strategies. Optimal cases from the simulation study were successfully applied to multi-well pads in the Eagle Ford formation across multiple fluid types. The simulation and the field application results were summarized and compared to provide detailed insights of unconventional HnP EOR. This study indicates the importance of confinement of the gases to afford optimal recovery factors during unconventional gas HnP EOR. An optimization engine was used in this study to optimize key operational parameters, such as injection pressures and slug sizes, to maximize recovery and efficiency. The resulting EOR designs were successfully implemented in field operations. Field recovery factors are within 10% of those predicted by simulation, indicating the value of numerical reservoir simulation prior to field trials and subsequent future development.

Analytical models available in Rate-Transient-Analysis (RTA) packages are widely used as tools for history matching and forecasting production in unconventional resources. There has also been an increasing interest in the use of numerical simulation of unconventional reservoirs. In this study, we use both methods to history match the production of hydraulically fractured unconventional wells, followed by forecasting future production to establish a well's EUR (Estimated Ultimate Recovery) for reserves determination purposes. This study's goal is to quantify the differences one might expect to encounter in a well's EUR when using analytical model-based RTA vs numerical simulation-based workflows in unconventional reservoirs.First, we consider an undersaturated shale oil reservoir as a base model for this study. The base case also satisfies all assumptions inherent to analytical solution-based methods such as homogenous reservoir properties and fully-penetrating planar fractures. An excellent match between results of both methods for the base model validates the numerical simulation approach. We then impose a series of real-world deviations from RTA assumptions and investigate reliability of EUR predictions made by both approaches. In all cases, historical data and reference EURs are derived from finely-gridded numerical simulations.Example results show that, in the presence of real-world deviations from RTA assumptions, analytical models can still match the historical production data; however, key reservoir and fracture parameters need to be modified drastically to compensate for the lack of sufficient physics in the analytical models. Results show that the analytical solution-based history-matched models are not predictive for future production, and somewhat surprisingly provide pessimistic EURs in all real-world scenarios investigated in this work. For the cases presented in this study, analytical models under-predict EURs by 6-17% when two years of production history is available for matching. The underestimation of EUR increases drastically (up to 60%) as the length of available historical data decreases from 2 years to 3 months.For all cases, we also apply an efficient numerical simulation-based workflow for probabilistic forecasting of EURs. This workflow provides multiple history-matched models that are constrained by historical production data. The probabilistic forecast method employed in this work provides P90 (conservative), P50 (most likely), and P10 (optimistic) values for EUR. In all examples, the range of P90 to P10 EUR values includes the reference EUR, and the P50 values are within 2.2% of the reference EUR.

Both of the above "limitations" are not inherent to the technique, they are created by: (1)(,, improvements in Drillstem Test.design, iiiofiitcrifiq ---uu,-~~~;~~~f +--+ A-..nla"+a"+ WW-mff,,?.nc LC>L CL+U IPIIII=,II. ?I~d~, ""%-"s b.., ,-, and interpretation procedures which maximize the pro-inflexible wellsite procedures that do not allow bability of a successful test and permit well perfor-for varied response to observed behavior (3) mance predictions. inappropriate use of "standard" pressure transient test interpretation techniques, and (4) a lack of INTRODUCTIONconcern for setting and prioritizifigDST tilf07fi6tion objectives. These operational deficiencies, Drillstem Testing is a 50-year-old technique tradi-along with the common practice of using flow rate tionally used and controlled by exploration personnel data as a "quantitative" indicator of reservoir to ositivel identify reservoir fluid~during %--l-Y production rate potential (see paragraph one wil cat drl lng operations when other sources of in-above), explain why DST data has not been utilized formation have been either unavailable or unreliable. to its full potential by the oil industry, in The use of drillstem test data by engineering personnel has historically been limited to: (1) reserve esti-particular by engineering personnel. mation using extrapolated reservoir static shut-inThe application of appropriate, and in many pressures from DST shut-in periods i-unction with cases improved, DST technology in the Rocky Mounlog data (porosity and net pay) and PVT data (measured tain area of the United States during the past 2.5 or from correlations), and (2) qualitative projection years has provided the evidence that with proper of reservoir (not welij reduction rate potential by f---'-equipment, test proc.e&iFf?S iifid ifiterpretaticm _L -----combining measured DST f ow rate data with the various techniques, DST data can be used not only to identify indicators of near-well-bore transmissibility reduction (i.e. damaged formation) or improvement (i.e. stimu-reservoir fluid type during wildcat drilling operations, but also to accuratel predict well perforated formation) which are calculated by applying an mance quantitativeltiate versu~me behav-"appropriatet'pressure transient test interpretation fian be predicted as a function of the post DST technique to DST shut-in period data. initial completion design (open hole DST) or recom-The use of DST data for the engineering purposes pletion design (cased hole DST). Drillstem tests are now being used routinely as an engineering and described above, and for any other quantitative engi-management tool to access the economic impact of neering calculations, for that matter, has always been looked upon as unrealizableby many engineers and scien-expensive completion or recompletion (i.e. workover) design decisions, such as "what frac length, tists in the oil industry, including the pressure tran-prop size and prop type is economically optimum" sient theory experts in the academic conmnunity. Sup-and "what shot dens...

Analytical models available in Rate-Transient-Analysis (RTA) packages are widely used as tools for history matching and forecasting production in unconventional resources. There has also been an increasing interest in the use of numerical simulation of unconventional reservoirs. In this study, we use both methods to history match the production of hydraulically fractured unconventional wells, followed by forecasting future production to establish a well's EUR (Estimated Ultimate Recovery) for reserves determination purposes. This study's goal is to quantify the differences one might expect to encounter in a well's EUR when using analytical model-based RTA vs numerical simulation-based workflows in unconventional reservoirs. First, we consider an undersaturated shale oil reservoir as a base model for this study. The base case also satisfies all assumptions inherent to analytical solution-based methods such as homogenous reservoir properties and fully-penetrating planar fractures. An excellent match between results of both methods for the base model validates the numerical simulation approach. We then impose a series of real-world deviations from RTA assumptions and investigate reliability of EUR predictions made by both approaches. In all cases, historical data and reference EURs are derived from finely-gridded numerical simulations. Example results show that, in the presence of real-world deviations from RTA assumptions, analytical models can still match the historical production data; however, key reservoir and fracture parameters need to be modified drastically to compensate for the lack of sufficient physics in the analytical models. Results show that the analytical solution-based history-matched models are not predictive for future production, and somewhat surprisingly provide pessimistic EURs in all real-world scenarios investigated in this work. For the cases presented in this study, analytical models under-predict EURs by 6-17% when two years of production history is available for matching. The underestimation of EUR increases drastically (up to 60%) as the length of available historical data decreases from 2 years to 3 months. For all cases, we also apply an efficient numerical simulation-based workflow for probabilistic forecasting of EURs. This workflow provides multiple history-matched models that are constrained by historical production data. The probabilistic forecast method employed in this work provides P90 (conservative), P50 (most likely), and P10 (optimistic) values for EUR. In all examples, the range of P90 to P10 EUR values includes the reference EUR, and the P50 values are within 2.2% of the reference EUR.