A Model To Estimate Carbon Dioxide Injectivity and Storage Capacity

Ryan W. J. Edwards , Florian Doster , Michael A. Celia , and Karl W. Bandilla .... Daniele Costa , João Jesus , David Branco , Anthony Danko , Antón...
0 downloads 0 Views 1MB Size
Download PDF
Article pubs.acs.org/est

A Model To Estimate Carbon Dioxide Injectivity and Storage Capacity for Geological Sequestration in Shale Gas Wells Ryan W. J. Edwards,*,† Michael A. Celia,† Karl W. Bandilla,† Florian Doster,‡ and Cynthia M. Kanno†,∞ †

Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey 08544, United States Institute of Petroleum Engineering, Heriot-Watt University, Edinburgh EH14 4AS, United Kingdom

‡

S Supporting Information *

ABSTRACT: Recent studies suggest the possibility of CO2 sequestration in depleted shale gas formations, motivated by large storage capacity estimates in these formations. Questions remain regarding the dynamic response and practicality of injection of large amounts of CO2 into shale gas wells. A twocomponent (CO2 and CH4) model of gas flow in a shale gas formation including adsorption effects provides the basis to investigate the dynamics of CO2 injection. History-matching of gas production data allows for formation parameter estimation. Application to three shale gas-producing regions shows that CO2 can only be injected at low rates into individual wells and that individual well capacity is relatively small, despite significant capacity variation between shale plays. The estimated total capacity of an average Marcellus Shale well in Pennsylvania is 0.5 million metric tonnes (Mt) of CO2, compared with 0.15 Mt in an average Barnett Shale well. Applying the individual well estimates to the total number of existing and permitted planned wells (as of March, 2015) in each play yields a current estimated capacity of 7200−9600 Mt in the Marcellus Shale in Pennsylvania and 2100−3100 Mt in the Barnett Shale.

■

INTRODUCTION Deep geological formations have been proposed and investigated as potential reservoirs in which to inject CO2 for long-term storage. Sequestration of CO2 in this manner is considered one of the most feasible near-term options to reduce emissions to the atmosphere.1 CO2 has also been routinely injected into geological formations for the purpose of enhanced hydrocarbon recovery for several decades.2 Injection for both CO2 sequestration and enhanced hydrocarbon recovery has involved relatively permeable formations such as depleted conventional oil and gas reservoirs and deep saline aquifers. Despite extremely low permeability, other favorable characteristics of organic-rich gas shales have prompted investigation of injection of CO2 into these formations for both purposes.3−5 Booming production of natural gas (the main constituent of which is CH4) from organic-rich shales over the past decade has been enabled through the technologies of horizontal drilling and high-volume hydraulic fracturing. These technologies enable access to a larger volume of a target rock layer by a single well and significantly enhance flow of gas to the well.6 Figure 1 shows a schematic of a typical hydraulically fractured horizontal well. Natural gas exists in organic-rich shales as free gas in pores and as adsorbed gas on solid surfaces. Adsorption primarily occurs in the organic matter within the rock rather than the inorganic minerals.7 Research suggests that a significant fraction (20−60%8) of CH4 stored in gas shales is adsorbed under © 2015 American Chemical Society

natural conditions, and only a relatively small portion of this is produced during the lifetime of a shale gas well.9 Studies have shown that significantly more CO2 adsorbs to the organic-rich shales than CH4,4,10−12 and that CO2 preferentially adsorbs over CH4 when present in a mixture.13−16 CO2 injected into a shale gas reservoir would likely preferentially adsorb to the organic matter and displace CH4 into the gas phase. This characteristic, along with the large number of shale gas wells drilled in the past decade and associated volume of accessible pore-space, has led to suggestions of possible injection of CO2 into depleted shale gas reservoirs for geological sequestration and/or enhanced gas recovery. Recent studies have reported large capacity estimates for storage of CO2 in shale gas reservoirs. Tao and Clarens5 estimated that 10.4−18.4 gigatonnes (Gt) (1 Gt = 1015 g) of CO2 could be stored in the Marcellus Shale by 2030; Godec et al.4 estimated 55 Gt could technically be sequestered in the Marcellus, with this number revised to 49 Gt in a subsequent study;17 and Nuttall et al.18 estimated that 28 Gt could be sequestered in the Devonian shales underlying Kentucky. Given that United States annual anthropogenic CO2 emissions are approximately 6.5 Gt per year,19 approximately 50% of which are from fixed-point sources and could potentially be captured, Received: Revised: Accepted: Published: 9222

April 19, 2015 July 6, 2015 July 7, 2015 July 17, 2015 DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229

Environmental Science & Technology

Article

■

METHODS Mathematical Model. The model presented herein of gas flow to a hydraulically fractured horizontal well is based on analyses of natural gas production data in the Barnett Shale.20,21 Patzek et al.20 analyzed data from 8284 wells in the Barnett and observed that scaled gas production rates initially decline inversely proportional to the square root of time before transitioning to exponential-in-time decline at later times. This behavior is consistent with the solution to a one-dimensional diffusion equation in a finite domain, with a fixed-pressure boundary condition at one end of the domain allowing for outflow and a no-flow boundary at the other end. Barnett Shale production data were shown to fit very well to this model.20 The implication is that, despite complicated fracture geometries, highly heterogeneous rock structure at the small scale, and the presence of two phases (gas and brine), at the wholewell scale the flow of gas in the Barnett Shale behaves as simple single-phase, one-dimensional Darcy flow in an effectively homogeneous medium, between parallel, planar hydraulic fractures. Figure 1 shows the implied geometry of the model and is broadly representative of the typical structure of shale gas wells: a horizontal well with perforations at regular intervals and approximately planar fractures perpendicular to the well. The well and hydraulic fractures both have essentially infinite permeability compared to the shale matrix and therefore pressure in the well and fractures is effectively equalized, resulting in the horizontal one-dimensional flow behavior between adjacent fractures. The following nonlinear diffusion equation describes the one-dimensional gas flow:20

Figure 1. Schematic of a horizontal shale gas well (top), with the model representation shown below. The horizontal well has hydraulic fractures at regular intervals that are represented in the model as planar fractures perpendicular to the horizontal well (middle), with onedimensional gas flow between the fracture planes (bottom). The red arrows represent flow into the shale in an injection scenario. Not to scale.

b ∂ ∂ ⎛ κρ (p) ∂p ⎞ ⎜⎜ b ⎟⎟ = 0 [ϕSgρ b (p) + ρa (p)] − ∂t ∂x ⎝ μ (p) ∂x ⎠

(1)

where ϕ is the rock porosity (volume of voids per total volume of rock), Sg is the gas saturation (volume of gas per volume of voids), ρb is the bulk free gas density (mass of gas per volume of gas), ρa is the adsorbed gas density (mass of adsorbed gas per total volume of rock), κ is the effective permeability (relative permeability multiplied by intrinsic permeability), μb is the bulk gas viscosity, and p is the pore gas pressure. The gas density and viscosity, and the adsorbed density, are dependent on pressure (and temperature, which is assumed to be constant), leading to the nonlinear nature of the equation. Gas saturation (and therefore relative and effective permeability) is assumed to be constant. The equation describes flow of a bulk gas in the shale matrix, accounting for both the bulk gas phase and adsorbed phase. The domain is between one hydraulic fracture and the plane of symmetry that occurs at half the distance to the adjacent fracture. This domain is repeated by symmetry along the length of a horizontal well (Figure 1). The applicability of the single-phase model implies that the brine phase is essentially immobile (after a short initial period where fracturing fluids flow back out of the well, which is not modeled). Rock compressibility is assumed to be negligible compared to the summed effect of gas compressibility and adsorption. Gas flow either out of the shale reservoir (production) or into the reservoir (injection) may be simulated with different model initial and boundary conditions. In the case of CO2 injection into a depleted shale gas well, natural gas will initially be present in the shale reservoir at relatively low pressure due to the prior production of the gas. CO2 is then injected at high pressure into the well. The gas in

these capacity estimates for shales are significant. Although they identify a substantial total capacity for CO2 storage, the aforementioned studies do not thoroughly investigate the dynamics (rates and time scales) of CO2 injection into individual shale wells. Regardless of total capacity, CO2 must be cost-effectively injected into the wells at the required rates (the rate of CO2 capture) for shale gas formations to be an attractive sequestration target. This study aims to improve on the previous assessments of CO2 sequestration potential by developing a well-scale model of gas flow in a shale reservoir that can easily be applied to investigate CO2 injection dynamics in shale gas wells. The mathematical model is based on natural gas production data and the associated models consistent with those data. This model captures the most important physical processes for CO2 injection (bulk gas flow, component transport, and adsorption of CO2 and CH4), while being relatively simple and readily applied to any shale gas play by choosing appropriate parameters and calibrating to natural gas production data. The model enables investigation of injection dynamics and allows for different injection scenarios, providing transient system responses and total (ultimate) capacity for individual wells. 9223

DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229

Article

Environmental Science & Technology

where n is the amount of gas adsorbed (number of moles, or mass), nmax is the maximum amount that is adsorbed when all adsorption sites are occupied, and K is the Langmuir constant, which is the inverse of the pressure at which half the adsorption sites are occupied. The Langmuir function calculates the absolute amount of gas adsorbed. For systems where two gas components are present in a mixture and competing for adsorption sites, the two-component Langmuir model can be used to estimate adsorption of each component:25

the pores becomes a mixture of CO2 and CH4, and the properties of the bulk gas now depend not only on the pressure, but also on the composition of the bulk gas (the fraction of CO2 in the gas). The movement of both the CO2 and natural gas components within the bulk gas, and particularly their competition for adsorption sites, are now important in the system dynamics. The single-phase, singlecomponent model is therefore extended to a single-phase, twocomponent model for CO2 injection modeling. Natural gas is represented by only its main component, CH4, since CH4 constitutes approximately 84% of natural gas on average. The model involves a system of two equations, one for each of the two components. We write these two equations as a pressure equation (eq 2), which is the sum of the two component equations, and the CO2 component transport equation (eq 3): ∂ [ϕSgρ b (ωc , p) + ρca (ωc , p) + ρma (ωc , p)] ∂t b ∂ ⎡ κρ (ωc , p) ∂p ⎤ ⎢ b ⎥=0 − ∂x ⎢⎣ μ (ωc , p) ∂x ⎥⎦

n1 = n1max

(2)

In these equations, ωc is the mass fraction of CO2 in the bulk gas phase, ρac is the adsorbed density of CO2, ρam is the adsorbed density of CH4, Dc is the CO2 dispersion coefficient, and ub is the Darcy flux of the bulk gas phase, defined as ∂p κ μ (ωc , p) ∂x b

(4)

The primary variables are the pressure of the gas and the mass fraction of CO2. Given the mass fraction of CO2 in the bulk gas, the mass fraction of CH4 is easily calculated since both components must completely fill the available pore space: ωc + ωm = 1

Kp 1 + Kp

(7)

ne = nabs − Vaρ b e

(8) abs

where n is the excess amount adsorbed, n is the absolute (total) amount adsorbed, Va is the volume of pore space occupied by the adsorbed phase, and ρb is the density of the gas phase. From eq 8 it follows that absolute adsorption and excess adsorption are equal when the product of adsorbed phase volume and density of the bulk gas phase is negligible compared to the absolute adsorbed amount. This is usually true at relatively low pressures (less than 5 MPa) under which experimental adsorption measurements are often performed. However, in shale gas formations, pressure (typically 20−40 MPa) and therefore gas density are relatively high. Adsorbed phase volume can also be relatively high in porous media with extremely small pores and correspondingly high surface area to pore volume ratio, such as shales. These effects mean that excess adsorption is significantly lower than absolute adsorption in the conditions prevalent in shale gas formations, which has been demonstrated in high-pressure adsorption experiments on organic-rich shales.10,11 This is particularly evident for CO2, which has a sharp increase in density at 7−8 MPa as it transitions to the supercritical state. Figure 2 shows excess and absolute adsorption isotherms for pure CO2 and CH4 using adsorption data from shale gas formations. Methane deviates less from its absolute adsorption isotherm than CO2 due to its relatively lower density at high pressure compared to CO2. Figure 2 demonstrates that application of a Langmuir absolute adsorption isotherm to calculate the total mass of

(5)

The domain is finite (0 < x < d), with a no-flow boundary at the plane of symmetry (x = d) and fixed CO2 mass fraction and either a fixed pressure or mass flux at the fracture face (x = 0). The initial condition is a uniform pressure in the domain with no CO2 present. Mathematical descriptions of the initial and boundary conditions are in the Supporting Information. Fluid Properties. To solve the model equations, the density and viscosity of the CO2−CH4 gas mixture must be calculated as a function of both pressure and composition of the gas. The bulk gas density is calculated using a Peng−Robinson equation of state,22 with a mixing rule applied to determine parameters for the CO2−CH4 mixture based on the pure-component parameters and the composition of the mixture.23 The viscosity of the bulk gas is similarly calculated by applying a mixing rule24 to pressure-dependent functions of viscosity for CO2 and CH4. Adsorption Model. Sorption behavior of CH4 and CO2 in porous media has been shown to fit well to the Langmuir adsorption model,7,12 which has been used as an adsorption model in studies estimating CO2 sequestration capacity in shales.3,4,18 The standard Langmuir model is n = n max

1 + K1p1 + K 2p2

where n1 is the amount of component 1 adsorbed in a binary system of component 1 and 2. Other parameters are analogous to those for eq 6, except p1 and p2 are the partial pressures of each of the component gases. The two-component Langmuir model has been shown to provide a good empirical estimate for CO2−CH4 adsorption in porous activated carbon,13 although it is not strictly theoretically applicable for most binary systems.25 While the Langmuir model has been shown to model adsorption of CH4 and CO2 in porous media well, it is not applicable under all conditions for calculation of the total mass (free and adsorbed) of a gas in a porous medium. Since the adsorbed phase occupies a fraction of the porosity that would otherwise be filled with gas, the reduction of gas-filled porosity must be taken into account. In adsorption measurements, the results are often reported by assuming a constant porosity and introducing the concept of “excess adsorption”,12 which is the mass of a gas present in a porous medium in excess of the mass that would be present if there were no adsorbed phase and the entire porosity was filled with gas at a given density.26−28 Mathematically, a simple expression for excess adsorption is the following:

∂ ∂ b [ϕSgρ b (ωc , p)ωc + ρca (ωc , p)] + [ρ (ωc , p)u bωc] ∂t ∂x ∂ωc ⎤ ∂ ⎡ b − ⎥=0 ⎢⎣ϕSgρ (ωc , p)Dc ∂x ∂x ⎦ (3)

ub =

K1p1

(6) 9224

DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229

Article

Environmental Science & Technology

approximation is cell-centered, with flux parameters upstreamweighted. A modified Picard iteration is applied to iterate the nonlinear coefficients and to enforce mass balance following the method of Celia et al.31 The pressure and transport equations are solved simultaneously and implicitly. This scheme was found to be necessary to prevent mass balance errors due to the nonlinear nature of the equations. The numerical solution is implemented in MATLAB. Further details and equations for the numerical solution scheme are included in the Supporting Information. Parameter Selection and History-Matching. In order to use the shale reservoir model to perform realistic simulations of CO2 injection into shale gas wells and compare different shale plays, model parameters were history-matched to natural gas production data. Two major U.S. shale gas plays were selected for analysis: the Barnett Shale in Texas, and the Marcellus Shale in Pennsylvania (PA). The Barnett Shale was selected because it was the first shale gas play to be commercially developed using horizontal wells and high-volume hydraulic fracturing, has the longest production history, and was the subject of the analyses that form the basis for this modeling approach.20,21 The Marcellus Shale was selected because it is the largest shale play, with the greatest potential CO2 sequestration capacity, and a number of large stationary CO2 sources (coal-fired power plants) overlie the formation. The Marcellus was split into two different regions for this analysis, southwest and northeast PA, because the formation is much thicker in the northeast than southwest and consequently wells in the northeast are significantly more productive. The delineation of each region is included in the Supporting Information. Natural gas production data from all producing wells in each of the regions that began production within a certain time span (2007−2011 for the Barnett; 2010−2014 for the Marcellus) were obtained through Drilling Info, Inc.32 The data were then filtered to remove vertical wells, failed wells with very little production and wells with discontinuous production history; approximately 12% of wells were removed from the total data set. The final sample included 8951 wells in the Barnett Shale, 1449 wells in southwest PA, and 2516 wells in northeast PA. Production data were averaged among wells commencing production within particular subset time periods in each region and model reservoir parameters were history-matched to the averaged data. Wells were grouped by age because shale horizontal well design has evolved (e.g., longer laterals, closer fracture spacing) leading to more productive wells over time. A range of parameter fits was made to account for the changing average well design and uncertainty in reservoir parameters. More detailed descriptions of the data selection and filtering and parameter history-matching procedures are included in the Supporting Information. The parameter history-matching was performed by running the model in single-component mode (ωc = 0) for natural gas production. Natural gas fluid properties were used and boundary and initial conditions set appropriately for a gas production scenario (i.e., low pressure at the fracture boundary and high initial shale reservoir pressure). The model was fitted by adjusting two groups of fitting parameters: an “interference time”, τ, that dictates the ease of fluid flow through the shale (eq 11), and an ‘original mass of gas in place’ term, 4 , that dictates the size of the reservoir and amount of gas accessible by the horizontal well (eq 12):20

Figure 2. Absolute (solid lines) and excess (dashed lines) adsorbed mass per cubic meter of rock as a function of gas pressure for CO2 (red) and CH4 (blue) using typical parameters from samples of shale gas formations.7,11,18

CO2 in a shale gas formation without adjusting the gas-filled porosity leads to a significant overestimation of CO2 adsorption and therefore total sequestration capacity (50−80% for our simulations). An excess adsorption function must be used to ensure the total mass in the domain is calculated consistently. Most experimental CO2 adsorption measurements on samples from U.S. shale gas formations have been performed at relatively low pressure (0−5 MPa),12,18,29 below the CO2 critical point, over which range the excess adsorption isotherms often appear to follow standard Langmuir isotherm curves. Langmuir function parameters fitted to these isotherms and extrapolated to shale-formation pressures are more reflective of absolute adsorption than excess adsorption. Assuming the reported Langmuir parameters describe absolute adsorption isotherms, the two-component Langmuir isotherm equation can be modified to estimate excess adsorption as follows: ρ1e = ρ1max

⎛ ω ρ b (ω1 , p) ⎞ ⎜⎜1 − 1a ⎟⎟ 1 + K1p1 + K 2p2 ⎝ ω1 (ω1 , p)ρa ⎠ K1p1

(9)

ρe1

where is the excess adsorbed mass of component 1 per unit volume of rock. Adsorption is now expressed as a bulk density rather than an amount (as in eq 6 and eq 7). In eq 9, ρb is the density of the bulk gas and ρa is the adsorbed phase density, which is assumed to be constant. The adsorbed phase density can be estimated from experimental excess adsorption data.11 ωa1 is the adsorbed mass fraction of component 1: ω1a =

ρ1abs (ω1 , p) ρ1abs (ω1 , p) + ρ2abs (ω1 , p)

(10)

ρabs 1

where is the absolute adsorption of component 1 (eq 7) per volume of rock. Eq 9 is applied for each of the two components in the system. Details of the derivation of eq 9 beginning with eq 8 are given in the Supporting Information. This method for modifying the Langmuir model has been shown to fit measurements of excess adsorption on organic-rich shales reasonably well,11,30 and is applied as a model of twocomponent, competitive excess adsorption in our shale reservoir model. Numerical Solution. The governing eqs 2 and 3 are solved numerically using a finite difference approximation. The 9225

DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229

Article

Environmental Science & Technology

τ=

⎡ ∂ρ b d 2μ b ⎢⎣ϕSg ∂p +

∂ρa ⎤ ⎦ ∂p ⎥

kρ b

(11)

4 = (n f 2d)a f (ϕSgρ b + ρa )

(12)

where d is half the distance between adjacent hydraulic fractures, nf is the number of hydraulic fractures, and af is the surface area of each hydraulic fracture plane. All fluid parameters are evaluated at initial reservoir pressure and temperature. The term nf2d represents the total length of the horizontal well lateral, which is known, so in practice nf is set by the fracture spacing and horizontal lateral length. Most parameters were either set to typical play parameters obtained from peer-reviewed and industry literature (reservoir pressure, temperature, porosity, gas saturation and adsorption), governed by physical properties of the fluids (density and viscosity), or set from known well data (well lateral length). The remaining parameters (permeability, fracture spacing and fracture surface area) were varied to fit the data, although these too were guided by typical values from the literature. Some selected parameters include porosity of 6% in all regions, effective permeability of 45 nanoDarcy in the Barnett and 20− 25 nanoDarcy in the Marcellus, and average horizontal well length of 872 m in the Barnett, 1556 m in southwest PA, and 1675 m in northeast PA. More detailed information about data sources and all selected parameters are included in the Supporting Information. The fitted shale reservoir models were able to closely match averaged production data from each of the three different regions analyzed. Figure 3 shows examples

Figure 4. History-matched model fit (blue line) of cumulative natural gas production (billion standard cubic feet) versus time compared with averaged production data for Marcellus Shale wells in southwest PA commencing production in Q1 2011 (orange circles), Q1 2012 (red squares) and Q2 2012 (green triangles).

injection scenarios were considered: injection of CO2 at a constant maximum allowable pressure, and injection at a constant rate. The constant-pressure injection would maximize the mass sequestered by a single well for a given time, but injection rate declines strongly with time. Constant-rate injection would enable the matching of the injection rate to the steady output of CO2 from an industrial emission source, with pressure at the injection well increasing with time, and eventually exceeding the maximum allowable pressure at the well. Both injection scenarios were simulated. It was assumed that the maximum allowable injection pressure would be selected to be as high as possible to maximize the mass that could be sequestered in each well, while minimizing the risk of refracturing the rock. To meet these criteria, we chose the original natural gas reservoir pressure. Model Limitations. While the CO2 injection model is based on a simplified mathematical model consistent with production data from shale gas wells and parameters that have been selected by history-matching the model to production data, parametric and physical uncertainties remain due to the relative lack of understanding of shale formations. The largest uncertainties in the model are associated with adsorption: reliable adsorption data are scarce in the literature and, to our knowledge, no experimental data for CO2 adsorption on samples of U.S. shale gas formations at high (supercritical) pressure are available. The applicability of the adsorption model is therefore also uncertain. Although excess adsorption does not significantly increase total storage capacity and therefore any error would not fundamentally change the results for total capacity, we note that the absolute adsorbed mass is in fact a majority of the total stored CO2. Plots showing the contributions of excess and absolute adsorbed CO2 to total mass stored are included in the Supporting Information. The large amount of absolute CO2 adsorption may possibly lead to substantial decrease in open pore volume and associated decrease in permeability, which would impact the dynamics of CO2 injection. This phenomenon is known as “swelling”. Studies on the potential injection of CO2 into coal bed methane reservoirs show that the organic matter in coal swells upon CO2 adsorption and reservoir permeability consequently decreases significantly.33,34 While shale gas formations have much lower organic matter content than coals (2−10% compared with 75−90%) and associated lower adsorption, it

Figure 3. History-matched model fit of natural gas production rate (thousand standard cubic feet per day) versus time for an average Barnett Shale well (red line) compared with averaged production data from wells commencing production in 2008 (blue circle), 2009 (purple square) and 2010 (green diamond).

for production rate from the Barnett Shale, and Figure 4 displays cumulative production from southwest PA. These figures show data from age groups of wells with similar average well lengths, such that one parameter fit could be shown to match the data adequately (although the more recently drilled wells are visibly more productive due to increasing lateral lengths). Simulations of CO2 injection were performed using the history-matched reservoir and well parameters for each of the three regions. Boundary conditions at the fracture face were an important consideration in the simulations. Two feasible 9226

DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229

Article

Environmental Science & Technology has been shown that adsorption causes volumetric strain in some of the major mineral components of shale gas formations.12 However, studies have yet to show whether swelling in the relatively small fraction of organic matter in shales yields a significant effect on overall rock permeability, so these effects were not included in our model. Also, while adsorption has the most uncertain parameters, selection of all parameters in the model would benefit from comprehensive data on the properties of shale gas formations: the parameter selections in the current model were guided by data from disparate sources.

■

RESULTS AND DISCUSSION Figure 5 shows the simulated constant-pressure injection rates of CO2 into single, average horizontal shale gas wells in the

Figure 6. Simulated time of constant-rate CO2 injection before exceeding the maximum allowable pressure versus injection rate for average wells in the Marcellus Shale in northeast (green) and southwest (blue) PA for 35 MPa maximum pressure, and in the Barnett Shale (red) for 25 MPa maximum pressure.

Figure 5. Simulated constant-pressure CO2 injection rate with time for average wells in the Marcellus Shale in northeast (green) and southwest (blue) PA at 35 MPa injection pressure, and in the Barnett Shale (red) at 25 MPa injection pressure. The inset graph shows rates at early time that are truncated in the main graph.

Figure 7. Simulated cumulative mass of CO2 injected with time for constant-pressure injection in average wells in the Marcellus Shale in northeast (green) and southwest (blue) PA at 35 MPa injection pressure, and in the Barnett Shale (red) at 25 MPa injection pressure. The gray shaded areas represent the bound of upper and lower historymatched parameter scenarios based on changing well design with time and reservoir parameter uncertainty (See Supporting Information for detail of bound matching methods).

Barnett Shale and the Marcellus Shale. Initially, injection rates decline rapidly, before displaying a more moderate decline rate at later times. These later-time injection rates are quite low compared to expected source rates for captured CO2. For constant-rate injection simulations, pressure at the fracture face increases with time and eventually exceeds the maximum allowable pressure, with the time to exceedance related directly to the injection rate. An example plot of the injection pressure increase with time is included in the Supporting Information. The simulated characteristic exceedance times for a range of injection rates are shown in Figure 6. Exceedance times decline considerably with increasing injection rate, and are less than 10 years for rates greater than 100 tonnes of CO2 per day in all regions. Figure 7 shows cumulative CO2 injected over time for constant-pressure injection in each of the regions. An associated plot showing the contributions of adsorbed CO2 to the total cumulative mass stored is in the Supporting Information. Figures 5−7 show that there is a significant difference in injection rate and total mass able to be injected for average wells in each of the regions. Northeast PA has the highest capacity, mainly because the Marcellus Shale is significantly thicker in the northeast than the southwest. The Barnett Shale capacity is less than half that of southwest PA, which in turn is about half of the northeast PA capacity. The discrepancy between the Barnett and southwest PA is primarily due to horizontal laterals in the Barnett that are almost half the length

of those in the Marcellus on average, rather than any major difference in formation thickness or other reservoir parameters. Although not visible in the regionally averaged graphs, significant capacity variation is also observed within each play temporally (well laterals becoming longer with time), between companies (different well designs), and likely also spatially on smaller scales (due to geological differences). CO2 injection sites could be targeted within each region at wells with higher capacity than the average (depending on proximity to emission sources and cost of transportation). However, for large-scale sequestration many wells would be required and therefore the regional averages are an important consideration. The average capacity of wells is likely to continue increasing over time, if average well lengths continue to increase as in the past. While the uncertainty ranges in Figure 7 make some allowance for this, the possibility remains that the averages could exceed these ranges at some future time. The Barnett Shale is a more mature play where drilling of new wells and overall production are declining, so less variation is expected over time. 9227

DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229

Article

Environmental Science & Technology Table 1. CO2 Sequestration Capacity Estimates for Each Analyzed Shale Play Region

a

shale play region

Barnett Shale

Marcellus southwest PA

Marcellus northeast PA

Marcellus PA total

existing wells capacity (Gt) (number of wellsa) existing and permitted wells capacity (Gt) (number of wellsa)

1.9−2.9 (15 729) 2.1−3.1 (16 663)

0.5−0.7 (1731) 1.6−2.2 (5623)

1.3−1.8 (2801) 5.0−6.6 (10 449)

1.9−2.6 (4845) 7.2−9.6 (18 273)

As of March, 2015.

study by Tao and Clarens5 estimated that 10.4−18.4 Gt of CO2 could be injected in the Marcellus by 2030. Based on our results this would require approximately 22 000−39 000 wells to be depleted, available for CO2 injection and filled by 2030. With approximately 8000 wells drilled and producing in the Marcellus to date and considering the time scales for production and injection, this is an unlikely prospect. In summary, a model of gas flow to a hydraulically fractured horizontal well incorporating the most important physical processes for CO2 injection has been developed. Application of the model to the Barnett and Marcellus Shales shows that individual shale wells have small capacities and low injection rates compared with CO2 emissions from large industrial sources. Large-scale sequestration would require many shale wells. Although prior studies showed large overall capacities for CO2 sequestration in shale formations, drilled and currently permitted wells provide access to only a fraction of this capacity and are unlikely to be available for CO2 sequestration for some decades into the future. This study builds understanding of the engineering requirements and capacity availability for CO2 sequestration in depleted shale formations and will help inform assessment of the relative cost effectiveness of sequestration in shale formations compared to conventional targets such as saline aquifers.

Environmental Implications. The injection rates in average shale gas wells under constant-rate or constant-pressure scenarios (Figures 5 and 6) are relatively low compared with large industrial CO2 sources. A typical 500 MW coal-fired power plant emits approximately 3 million metric tonnes (Mt) per year,35 or 8200 tonnes per day, compared with rates of less than 100 tonnes per day for any significant duration in shale wells. The total capacities of shale wells are also relatively small: the capacity of an average well in northeast PA (the highest among the three regions analyzed) is approximately 0.6 Mt. Many shale wells would be required to sequester the emissions from a single coal-fired power plant. Considering the typical coal plant emitting 3 Mt per year and assuming a relevant time scale of sequestration to be 40 years (approximate plant lifetime), then 120 Mt are emitted, requiring approximately 250 wells in northeast PA, 400 wells in southwest PA, or 800 wells in the Barnett Shale if all the emissions were captured and injected into shale wells. In comparison, conventional targets such as deep saline aquifers will have significantly larger capacity per well. For example, the Sleipner CO2 injection in the North Sea has injected approximately 0.9 Mt per year (2500 tonnes per day) into a single well in a very permeable saline aquifer for 19 years continuously, with over 15 Mt injected to date and only a small (0.2 MPa) increase in wellhead pressure observed.36 A CO2 injection trial in the Mount Simon formation in Illinois injected 0.33 Mt per year (1000 tonnes per day) into a single well for a period of three years.37 Application of the average capacity estimates to the number of wells enables an estimation of the total CO2 sequestration capacity of the analyzed regions. The number of wells in each region is dynamic, however, increasing quickly in the Marcellus Shale but more slowly in the Barnett Shale. Two estimates have therefore been made for each play: a current capacity using all existing horizontal wells, and a future projection using all existing and permitted (approved but not yet drilled) wells. The limitations for the future projection are that average well parameters will likely change in time and more wells will continue to be drilled following those already permitted, particularly in the Marcellus Shale. The capacity estimates are shown in Table 1. The capacity of all existing and permitted Marcellus Shale wells in PA is estimated to be 7.2−9.6 Gt. Given likely average well production lifetime of 20−40 years,9,38 this capacity would not be available for CO2 sequestration for some time into the future. As a comparison, Godec et al.4,17 estimated total theoretical CO2 storage capacity for the Marcellus (in all states) to be 171 Gt (104 Gt in PA), and total technically accessible capacity to be 49 Gt. For the technically accessible estimate to be consistent with our results, approximately 100 000 wells would be required (assuming well and reservoir parameters from PA are applicable to Marcellus wells in adjacent states). At current rates of drilling in the Marcellus, it will be approximately 50 more years until 100 000 wells have been drilled, with an additional several decades of production before all those wells would be available for CO2 injection. Another

■

ASSOCIATED CONTENT

S Supporting Information *

Additional material is available as indicated in the text. The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.5b01982.

■

AUTHOR INFORMATION

Corresponding Author

*Phone: 347-906-0484; e-mail: [email protected]. Present Address ∞

Civil and Environmental Engineering, Colorado School of Mines, 1500 Illinois St, Golden, CO 80401, United States

Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This work was supported in part by funding from the Carbon Mitigation Initiative at Princeton University, and the Andlinger Center for Energy and the Environment at Princeton University. Underlying natural gas production data provided by Drilling Info, Inc. and used with permission.

■

REFERENCES

(1) IPCC. IPCC Special Report in Carbon Dioxide Capture and Storage, Prepared by Working Group III of the Intergovernmental Panel on Climate Change; Metz, B., Davidson, O., H. C. de Coninck, Loos, M., Meyer, L. A., Eds.; Cambridge University Press: Cambridge, 2005. (2) Orr, F. M.; Taber, J. J. Use of Carbon Dioxide in Enhanced Oil Recovery. Science 1984, 224 (4649), 563−569.

9228

DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229

Article

Environmental Science & Technology

(25) Ruthven, D. M. Principles of Adsorption and Adsorption Processes; John Wiley & Sons: Hoboken, NJ, 1984. (26) Zhou, L.; Bai, S. P.; Su, W.; Yang, J.; Zhou, Y. P. Comparative study of the excess versus absolute adsorption of CO2 on superactivated carbon for the near-critical region. Langmuir 2003, 19 (7), 2683−2690. (27) Zhou, Y.; Zhou, L. Fundamentals of High Pressure Adsorption. Langmuir 2009, 25 (23), 13461−13466. (28) Myers, A. L.; Monson, P. A. Adsorption in porous materials at high pressure: Theory and experiment. Langmuir 2002, 18 (26), 10261−10273. (29) NYSERDA. Geologic, Engineering and Economic Evaluation of the CO2 Sequestration Capacity of New York’s Gas Shales, 12−02; New York State Energy Research and Development Authority, 2011. (30) Gasparik, M.; Ghanizadeh, A.; Bertier, P.; Gensterblum, Y.; Bouw, S.; Krooss, B. M. High-Pressure Methane Sorption Isotherms of Black Shales from The Netherlands. Energy Fuels 2012, 26 (8), 4995− 5004. (31) Celia, M. A.; Bouloutas, E. T.; Zarba, R. L. A General MassConservative Numerical-Solution for the Unsaturated Flow Equation. Water Resour. Res. 1990, 26 (7), 1483−1496. (32) Drillinginfo, Inc.. info.drillinginfo.com. (33) Mazumder, S.; Wolf, K. H. Differential swelling and permeability change of coal in response to CO2 injection for ECBM. Int. J. Coal Geol. 2008, 74 (2), 123−138. (34) van Bergen, F.; Pagnier, H.; Krzystolik, P. Field experiment of enhanced coalbed methane-CO2 in the upper Silesian basin of Poland. Environ. Geosci. 2006, 13 (3), 201−224. (35) MIT. The Future of Coal: Options for a Carbon-Constrained World; Massachussetts Institute of Technology: Cambridge, MA, 2007. (36) Chadwick, R. A.; Williams, G. A.; Williams, J. D. O.; Noy, D. J. Measuring pressure performance of a large saline aquifer during industrial-scale CO2 injection: The Utsira Sand, Norwegian North Sea. Int. J. Greenhouse Gas Control 2012, 10, 374−388. (37) Finley, R. J. An overview of the Illinois Basin - Decatur Project. Greenhouse Gases: Sci. Technol. 2014, 4 (5), 571−579. (38) Browning, J.; Ikonnikova, S.; Gülen, G.; Tinker, S. W. Barnett Shale Production Outlook. SPE Economics & Management 2013, 5 (3), 89−104.

(3) Liu, F. Y.; Ellett, K.; Xiao, Y.; Rupp, J. A. Assessing the feasibility of CO2 storage in the New Albany Shale (Devonian-Mississippian) with potential enhanced gas recovery using reservoir simulation. Int. J. Greenhouse Gas Control 2013, 17, 111−126. (4) Godec, M.; Koperna, G.; Petrusak, R.; Oudinot, A. Potential for enhanced gas recovery and CO2 storage in the Marcellus Shale in the Eastern United States. Int. J. Coal Geol. 2013, 118, 95−104. (5) Tao, Z.; Clarens, A. Estimating the carbon sequestration capacity of shale formations using methane production rates. Environ. Sci. Technol. 2013, 47 (19), 11318−11325. (6) Martineau, D. F. History of the Newark East Field and the Barnett Shale as a gas reservoir. AAPG Bull. 2007, 91 (4), 399−403. (7) Zhang, T. W.; Ellis, G. S.; Ruppel, S. C.; Milliken, K.; Yang, R. S. Effect of organic-matter type and thermal maturity on methane adsorption in shale-gas systems. Org. Geochem. 2012, 47, 120−131. (8) NETL. A Comparative Study of the Mississippian Barnett Shale, Fort Worth Basin, and Devonian Marcellus Shale, Appalachian Basin, DOE/NETL-2011/1478; National Energy Technology Laboratory, U.S. Department of Energy, 2011. (9) Cipolla, C. L.; Lolon, E. P.; Erdle, J. C.; Rubin, B. Reservoir Modeling in Shale-Gas Reservoirs. SPE Reservoir Evaluation & Engineering 2010, 13 (4), 638−653. (10) Chareonsuppanimit, P.; Mohammad, S. A.; Robinson, R. L.; Gasem, K. A. M. High-pressure adsorption of gases on shales: Measurements and modeling. Int. J. Coal Geol. 2012, 95, 34−46. (11) Weniger, P.; Kalkreuth, W.; Busch, A.; Krooss, B. M. Highpressure methane and carbon dioxide sorption on coal and shale samples from the Parana Basin, Brazil. Int. J. Coal Geol. 2010, 84 (3− 4), 190−205. (12) Heller, R.; Zoback, M. D. Adsorption of methane and carbon dioxide on gas shale and pure mineral samples. Journal of Unconventional Oil and Gas Resources 2014, 8 (0), 14−24. (13) Dreisbach, F.; Staudt, R.; Keller, J. U. High pressure adsorption data of methane, nitrogen, carbon dioxide and their binary and ternary mixtures on activated carbon. Adsorption 1999, 5 (3), 215−227. (14) Billemont, P.; Coasne, B.; De Weireld, G. Adsorption of carbon dioxide-methane mixtures in porous carbons: effect of surface chemistry. Adsorption 2014, 20 (2−3), 453−463. (15) Ottiger, S.; Pini, R.; Storti, G.; Mazzotti, M. Competitive adsorption equilibria of CO2 and CH4 on a dry coal. Adsorption 2008, 14 (4−5), 539−556. (16) Kurniawan, Y.; Bhatia, S. K.; Rudolph, V. Simulation of binary mixture adsorption of methane and CO2 at supercritical conditions in carbons. AIChE J. 2006, 52 (3), 957−967. (17) Godec, M.; Koperna, G.; Petrusak, R.; Oudinot, A. Enhanced Gas Recovery and CO2 Storage in Gas Shales: A Summary Review of its Status and Potential. Energy Procedia 2014, 63, 5849−5857. (18) Nuttall, B. C.; Eble, C. F.; Drahovzal, J. A.; Bustin, R. M. Analysis of Devonian Black Shales in Kentucky for Potential Carbon Dioxide Sequestration and Enhanced Natural Gas Production, DE-FC26− 02NT41442; Kentucky Geological Survey: Lexington, KY, 2005. (19) EPA. Inventory of US Greenhouse Gas Emissions and Sinks: 1990− 2012, EPA 430-R-14-003; U.S. Environmental Protection Agency: Washington DC, April 15, 2014. (20) Patzek, T. W.; Male, F.; Marder, M. Gas production in the Barnett Shale obeys a simple scaling theory. Proc. Natl. Acad. Sci. U. S. A. 2013, 110 (49), 19731−19736. (21) Silin, D. B.; Kneafsey, T. J. Shale Gas: Nanometer-Scale Observations and Well Modelling. J. Can. Pet. Technol. 2012, 51 (6), 464−475. (22) Gasem, K. A. M.; Gao, W.; Pan, Z.; Robinson, R. L. A modified temperature dependence for the Peng-Robinson equation of state. Fluid Phase Equilib. 2001, 181 (1−2), 113−125. (23) Sandler, S. I. Chemical, Biological, and Engineering Thermodynamics, 4th ed.; John Wiley & Sons: Hoboken, NJ, 2006. (24) Herning, F.; Zipperer, L. Calculation of the viscosity of technical gas mixtures from viscosity of the individual components. Gas Wasserfach 1936, 79, 69−73.

■

NOTE ADDED AFTER ASAP PUBLICATION This article was published July 17, 2015. Due to a production error, it was published with errors in the well number approximations noted for the Barnett and Marcellus Shales in the Environmental Implications section. The Supporting Information file was revised due to distortions to Figures S.1 and S.2 in the original file, caused in the PDF conversion during publishing. The corrected version of the article and the Supporting Information were published July 21, 2015.

9229

DOI: 10.1021/acs.est.5b01982 Environ. Sci. Technol. 2015, 49, 9222−9229