Some Challenges Posed by Coal Bed Methane Regional Assessment Modeling by Catherine R. Moore1,2 , John Doherty3,4 , Stephen Howell5 , and Leon Erriah5
Abstract Coal measures (coal bearing rock strata) can contain large reserves of methane. These reserves are being exploited at a rapidly increasing rate in many parts of the world. To extract coal seam gas, thousands of wells are drilled at relatively small spacing to depressurize coal seams to induce desorption and allow subsequent capture of the gas. To manage this process effectively, the effect of coal bed methane (CBM) extraction on regional aquifer systems must be properly understood and managed. Groundwater modeling is an integral part of this management process. However, modeling of CBM impacts presents some unique challenges, as processes that are operative at two very different scales must be adequately represented in the models. The impacts of large-scale gas extraction may be felt over a large area, yet despite the significant upscaling that accompanies construction of a regional model, near-well conditions and processes cannot be ignored. These include the highly heterogeneous nature of many coal measures, and the dual-phase flow of water and gas that is induced by coal seam depressurization. To understand these challenges, a fine-scale model was constructed incorporating a detailed representation of lithological heterogeneity to ensure that near-well processes and conditions could be examined. The detail of this heterogeneity was at a level not previously employed in models built to assess groundwater impacts arising from CBM extraction. A dual-phase reservoir simulator was used to examine depressurization and water desaturation processes in the vicinity of an extractive wellfield within this fine-scale model. A single-phase simulator was then employed so that depressurization errors incurred by neglecting near-well, dual-phase flow could be explored. Two models with fewer lithological details were then constructed in order to examine the nature of depressurization errors incurred by upscaling and to assess the interaction of the upscaling process with the requirement for adequate representation of near-source, dual-phase processes.
Introduction Extraction of coal bed methane (CBM, also referred to as “coal seam gas”) is expanding rapidly in a number of countries, including the United States, Australia, China, and India (Moore 2012). Depressurization of coal beds that is necessary for extraction of CBM is achieved by pumping water from these beds until the hydrostatic pressure is sufficiently lowered to induce desorption of methane from the coal matrix. The resulting drawdown in groundwater levels is likely to propagate laterally, 1 Department of Groundwater, Institute of Environmental Science and Research Ltd, Porirua 5240, New Zealand. 2 Corresponding author: CSIRO Land & Water, Australia Ecosciences Precinct, 41 Boggo Road, Dutton Park, QLD 4102, Australia; [email protected] 3 National Centre for Groundwater Research and Training, Flinders University, GPO Box 2100, Adelaide, SA 5001, Australia; [email protected] 4 Watermark Numerical Computing, Brisbane, Australia. 5 QGC – A BG Group Business, Groundwater project, Level 18, 69 Ann Street, Brisbane, QLD 4000, Australia; [email protected]; [email protected] Received December 2013, accepted January 2014. © 2014, National Ground Water Association. doi: 10.1111/gwat.12276
NGWA.org
and possibly vertically to other hydrogeological strata, where it could affect extraction from water supply wells, and reduce flow to groundwater-dependent ecosystems (CWIMI 2008; DNRM 2013). Suitable management practices, such as strategic reinjection of extracted water, and the deepening of existing production wells, can be used to mitigate these impacts (e.g., Myers 2009; Freij-Ayoub 2012). Regional groundwater models can be used to assess the potential for these and other impacts of CBM development and to provide a basis for regional groundwater management. However, because of the rapid and relatively recent growth of the CBM industry, only a limited number of these types of models have been reported in the literature (Harrison et al. 2000; Myers 2009; QWC 2012; QGC 2013). The modeling of regional groundwater drawdown induced by CBM extraction poses a number of significant challenges. On the one hand, these models must often encompass large domains and must include significant aquifers and aquitards that prevail in the large sedimentary basins wherein coal sequences are found. They must also encompass the locations where the adverse consequences of CBM extraction may be felt, which can be far removed from extraction sites (QGC 2013). On the other
Vol. 53, No. 5–Groundwater–September-October 2015 (pages 737–747)
737
hand, adequate representation of hydrogeological detail near extraction sites may be important. This is because individual coal layers within an extensive coal measure sequence may be of limited lateral extent, thus reducing connectivity to regional groundwater systems and limiting regional drawdown impacts. The challenges of representing these different scales is made more complicated by the necessity for these models to accommodate, if not explicitly simulate, dualphase flow of water and gas to extraction wells (Moore et al. 2013). Unfortunately, explicit representation of dual-phase flow imposes a significant computational burden on a regional model. Hence, usually a simpler single-phase model such as MODFLOW (Harbaugh 2005), which runs relatively quickly, is used for CBM regional impact assessments (Myers 2009; QWC 2012, among others). Reduced model run times assist model calibration and calibration-constrained uncertainty analysis, both of which may require many thousands of model runs. However, as is discussed in the section titled “Near-Well Desaturation and Drawdown Propagation,” the use of a single-phase model to estimate regional drawdown induced by CBM extraction can result in overprediction of these drawdowns. Simulation of near-well, dual-phase flow or a singlephase substitute for dual-phase flow (Moore et al. 2013) requires that a suitable relationship between relative permeability and water saturation be developed for the upscaled representation of simulated coal sequences. These relationships can have a critical effect on the simulation of propagation of drawdown outward from these wells. A further requirement is the employment of suitable “dynamic pseudo-relative permeability” functions, or “pseudos” as these are often called in the petroleum literature (Kyte and Berry 1975; Stone 1991; Barker and Thibeau 1997; Hewett and Archer 1997; Darman et al. 2002; Gasda and Celia 2005; Nordbotten et al. 2010). To the authors’ knowledge these matters have received no attention in either the CBM reservoir modeling or CBM regional impact assessment modeling literature. The study reported herein uses detailed numerical simulations of near-well processes utilizing a dual-phase simulator that is employed throughout the oil and gas industry for modeling oil, water, and gas production, namely the ECLIPSE reservoir simulator (Schlumberger 2011). The study focuses on the development and progression of depressurization and water desaturation in the vicinity of a CBM extraction wellfield. Because of the high level of lithological detail represented in the model (which, as far as the authors know, is unprecedented in a study of this kind), the pressure and saturation states of the system can be simulated at a fine-scale. Examination of the magnitudes and changing nature of these states provides unique insights into the issues that any upscaled representation of these processes in a regional assessment model must address. These issues include the nature of drawdown and desaturation propagation through a highly heterogeneous sequence of coal seams, the spatial and temporal nature of errors incurred when simulating CBM 738
C.R. Moore et al. Groundwater 53, no. 5: 737–747
extraction using a single-phase model, the consequences of upscaling on calculation of dynamic pseudo-relative permeabilities, and appropriate upscaling of coal measures in a regional impact assessment model. This article is organized as follows. First, the details of the model that forms the basis of the case study reported in this article are presented. The following section describes the simulation of water desaturation and drawdown propagation accompanying CBM extraction at a fine scale. It also presents an examination of the differences between drawdown predictions made using a single-phase model and those made using a dual-phase model at this same fine scale. Upscaling issues are then addressed, with particular attention paid to the effects of two different upscaling strategies on dynamic pseudo-relative permeability functions. The first upscaling strategy is that most commonly employed in reservoir and groundwater simulation whereby an upscaled model layer represents all fine-scale model layers that lie within its boundaries. The second involves segregation of coal seams into separate layers. The advantages of the latter upscaling methodology are demonstrated. Finally, the outcomes of this work are discussed, and conclusions are drawn for the representation of near-well processes in regional modeling.
Case Study The case study used for this article is based on a section of the Walloon Coal Measures (WCM) in the Surat Basin, Queensland, Australia. The WCM are approximately 350-m thick along the northern and eastern margins of the basin, and thicken to more than 500 m in the south-west of the basin. Their coal content varies with location. At the study location, about 10% of their thickness is composed of coal, this being typical of many CBM exploration and production areas in the Surat Basin. Coal occurs as thin, discontinuous seams, with individual coal plies ranging in thickness between 0.05 and 1.7 m. Up to 45 individual seams can be recognized in places. They are separated by siltstone, mudstone, and fine-to-medium grained lithic sandstone interburden. The lateral extent of individual seams varies considerably and is thought to range from about 500 m to more than 3000 m (Ryan et al. 2012). The WCM are overlain by the Springbok Sandstone that is composed of low permeability, fine-grained, often clay-rich, sandstone, siltstone, and mudstone (Ryan et al. 2012; Esterle et al. 2013). A stochastic realization of coal and interburden lithofacies within the WCM and overlying Springbok Sandstone was generated for use by the model (Figure 1a). This was conditioned to borehole petrophysical data within the study area; its stochasticity is based on extensive company datasets acquired in this and neighboring development areas. The geological block model populated with this realization comprises the simulation domain used in the present investigation (Figure 1b), the details of which are given in Table 1. Neuman (no-flow) boundary conditions surround the entire block model. Initial hydraulic heads NGWA.org
Figure 1. (a) Stochastic realization of lithofacies within the Walloon Coal Measures and Springbok Sandstone. Coal is colored green (note that a small amount of coal occurs within the Springbok Sandstone). (b) Model grid and location of CBM extraction wells. Vertical exaggeration is 15:1.
Table 1 Block Model Dimension and Discretization Details Block Model Detail
Dimensions
Horizontal dimensions Thickness Grid
17,700 m × 19,500 m 600–700 m 131 rows, 119 columns, 614 layers Layers 1–162 Layers 163–614
Springbok sandstone Walloon coal measures
were assigned a constant value of 262 m, ensuring fully saturated and confined conditions throughout the model domain. Lithofacies within the stochastic realization that forms the basis of this study were populated with a stochastic realization of hydraulic properties that are representative of those prevailing within the study area. Ranges of some of these properties as they pertain to different model stratigraphic subdivisions are provided in Table 2. Certain properties, particularly those related to gas, are omitted from this table because of their proprietary nature. The authors wish to point out, however, that failure to provide these properties does not detract from the generality of this study, as the conclusions that are drawn from it are not specific to any particular set of properties, but are applicable in all contexts of regional coal seam gas (CSG) impact assessment. The most salient feature of the data provided in Table 2 is that the hydraulic conductivity (both horizontal and vertical) of coal is considerably greater than that of both the interburden material with which it is closely associated and the material comprising the overlying sandstone/siltstone/shale of the Springbok Sandstone. Flow of gas and water toward extraction wells is thus predominantly horizontal. As stated previously, the ECLIPSE reservoir simulator (Schlumberger 2011) was used to simulate gas production and movement of water and gas through the model domain. Coal is represented as a dual-porosity medium in which fluid moves through cleats (natural fractures in NGWA.org
coal); meanwhile, gas desorbed from the coal matrix first diffuses through the matrix and then moves through the cleat system as a separate phase from that of water. Gas desorbs as the pressure is lowered, this being governed by a Langmuir isotherm. A uniform Langmuir isotherm was assigned to all coal within the model domain, this being typical of that employed in gas company reservoir modeling. This is based on coal sample laboratory measurements and on local reservoir modeling experience. Use of a uniform isotherm allows this study to focus on the ramifications of lithological and hydraulic property heterogeneity. Water and gas are produced through a total of 34 wells. These are placed 750-m apart; each is screened across the entirety of the WCM (Figure 1b). Pumping is simulated for 20 years as water and gas are produced from each well subject to a total water flow constraint of 2000 stb/day (i.e., stock tank barrels/day where a barrel is 159 L) or 318 m3 /day, and a minimum bottom-hole pressure constraint of 90 psia (i.e., 620.5 kPa). (The reference level for this pressure is the elevation of the center of the uppermost layer from which extraction takes place.) Water extraction rates for three wells over the 20-year simulation period are shown in Figure 2. In accordance with the pumping specifications stated earlier, peak water extraction for each well is about 300 m3 /day. This peak water production rate prevails within the first 300 days of pumping, after which the rate of water production declines (Figure 2). The decline accelerates after about 1600 days. This is an outcome of the fact that coal connectivity restricts the lengths of connected coal geobodies to about 3 km; hence, the expanding cone of pressure depression encounters a low hydraulic conductivity boundary. Similar water production patterns are exhibited by other wells. Gas production peaks after about 300 days; it declines after about 1600 days as gas within coal seams is depleted and as access to further gas is inhibited by lack of coal connectivity. Gas production is not plotted for proprietary reasons. A uniform relationship between relative permeability and water saturation is applied to all cells within the model domain, regardless of lithology; this relationship is depicted in Figure 3. C.R. Moore et al. Groundwater 53, no. 5: 737–747
739
Table 2 Stratigraphic Zones, Lithology, and Hydraulic Property Ranges Assigned to the Block Model Stratigraphy
Fine-Scale Model Layers
Upper Springbok Mid Springbok Lower Springbok Basal Springbok Upper Juandah
1–70 71–135 136–159 160–162 163–287
Lower Juandah
288–432
Tangalooma
433–522
Taroom
523–614
Dominant Lithology Sandstone/siltstone/shale Sandstone/siltstone/shale Sandstone/siltstone/shale Sandstone/siltstone/shale Coal2 Interburden1 Coal2 Interburden1 Coal2 Interburden1 Coal2 Interburden1
Hydraulic Conductivity (Horizontal, m/day)
Hydraulic Conductivity (Vertical, m/day)
6.34 × 10−5 –6.45 × 10−2 6.34 × 10−5 –6.45 × 10−2 6.37 × 10−5 –6.838 × 10−3 6.37 × 10−5 –9.997 × 10−3 3.056 × 10−3 –8.71 × 10−1 1.27 × 10−6 –1.97 × 10−3 5.09 × 10−3 –7.72 × 10−1 1.273 × 10−6 –1.97 × 10−3 1.02 × 10−4 –1.961 × 101 1.27 × 10−6 –1.4 × 10−3 1.019 × 10−4 –1.96 × 101 1.27 × 10−6 –1.4 × 10−3
2.55 × 10−5 –2.04 × 10−2 2.55 × 10−5 –2.04 × 10−2 3.82 × 10−10 –3.82 × 10−8 1.27 × 10−5 –3.158 × 10−3 1.273 × 10−8 4.839 × 10−8 1.273 × 10−8 4.839 × 10−8 1.273 × 10−8 4.839 × 10−8 1.273 × 10−8 4.839 × 10−8
Porosity4 0.001–0.16 0.001–0.16 0.001–0.15 0.001–0.22 0.013 0.001–0.18 0.013 0.001–0.18 0.013 0.001–0.18 0.013 0.001–0.18
Note: As shown in the table, the Walloon Coal Measures are subdivided into the Upper and Lower Juandah Formations, the Tangalooma Sandstone, and the Taroom Coal Measures. 1 Interburden comprises siltstone, silty sandstone, and sandstone. 2 The lateral extent of any coal seam is considered to average about 3000 m. 3 Coal porosities were assigned a uniform value of 0.01, which was assumed to characterize cleat volume fraction. 4 Specific storage was calculated based on porosity, and coal measure and Springbok compressibilities of 2.0e−5 and 3.0e−6 /psi, respectively, using Terzaghi’s consolidation theory (Terzaghi et al. 1996).
Figure 2. Water extraction rates over the 20-year simulation period for three representative wells. The locations of these wells are shown in Figure 5.
Figure 3. Relative permeability vs. water saturation relationship used by ECLIPSE for all model cells.
Near-Well Desaturation and Drawdown Propagation Propagation of drawdown and water desaturation within the heterogeneous coal measure sequence that forms the basis of this study is now explored. The errors in simulated drawdown incurred through singlephase representation of these fundamentally dual-phase processes are then examined. As far as the authors are aware, there are no previously published studies that examine propagation of drawdown and water desaturation induced by CBM extraction in such detail, partly because the focus of most reservoir modeling is on water and 740
C.R. Moore et al. Groundwater 53, no. 5: 737–747
gas production rather than on pressure and desaturation propagation within the subsurface flow domain. Nor are the authors aware of any studies that attempt to quantify errors that are incurred through use of a traditional singlephase groundwater model in place of a reservoir model for calculation of drawdowns in this context. Vertical profiles of coal occurrence and head and water saturation calculated by the ECLIPSE model for a simulation time of 20 years (Figure 4) are shown for model cells (row, column) (52, 62) and (78, 78) (see Figure 5 for their locations). These indicate that, for this simulation time, water desaturation is predominantly restricted to coal layers. It is noteworthy, however, that some temporary, localized interburden water desaturation occurs at earlier times as gas desorbed from coal migrates upward before being captured by pumping-induced hydraulic gradients. CBM extraction incurs greater drawdown in coal than in the interburden, this being an outcome of the much higher horizontal hydraulic conductivity of the former than the latter. Drawdown in the overlying Springbok Sandstone is minimal, this reflecting the low vertical hydraulic conductivity awarded to this formation in the present model. Simulated water desaturation after 1, 3, 8, and 20 years of extraction confirms that desaturation is largely restricted to coal seams (Figure 6). However, as stated earlier, a close inspection reveals small and temporary pockets of desaturation in interburden. The small, expanding island of water desaturation in the upper right of the model domain is of interest. This occurs in a highly permeable but particularly thin coal seam (less than 0.1 m in thickness), surrounded both vertically and horizontally by low hydraulic conductivity interburden. Drawdown in the coal seam is sufficient to induce gas desorption. However, the hydraulic gradient in the seam is insufficient to capture the desorbed gas. Movement of gas to the wellfield would require partial water desaturation of NGWA.org
Figure 4. Vertical profiles of coal occurrence, water saturation, and head (in meters relative to sea level) at (a) column 62, row 52 and (b) column 78, row 78 after a simulation time of 20 years. The locations of these cells are shown in Figure 5. Note that the conversion of pressure to head uses the formula 1 psi = 2.30 ft of water; conversion of psi to psia requires subtraction of 14.695 psi to account for atmospheric pressure.
the surrounding interburden material in order to increase its relative permeability to gas. Not enough gas is released in the thin coal seam for this to happen.
Drawdowns With and Without Dual-Phase Flow Simulation Drawdowns computed using ECLIPSE are now compared with those computed by a traditional single-phase groundwater simulator MODFLOW-2005 (Harbaugh 2005). Corresponding cells in the ECLIPSE and MODFLOW models have the same hydraulic properties. Water extraction rates provided to MODFLOW wells are those computed by ECLIPSE for corresponding wells; recall that wells in the ECLIPSE model have target extraction rates that are subject to a bottom-hole pressure constraint. This strategy of assigning ECLIPSE-calculated water extraction rates to the MODFLOW model eliminates errors incurred by differences in pumping rates that would occur if MODFLOW were provided with the same target extraction rates and bottom-hole pressure constraints as those provided to ECLIPSE but under single-phase conditions. (Note that before comparing heads under NGWA.org
CBM extractive conditions, heads calculated by ECLIPSE under single-phase conditions were compared with those calculated by MODFLOW to ensure model parameter equivalence; these were found to be identical. Singlephase conditions were simulated with ECLIPSE by assigning zero initial gas concentrations to all coal cells.) Head drawdown profiles calculated by the two simulators down vertical columns of the model domain (see Figure 5 for their locations) are compared after 1, 3, 5, 10, and 20 years of continuous CBM extraction (Figure 7). It is apparent that use of a single-phase simulator leads to overprediction of drawdown. This results from the inability of the single-phase model to simulate loss of water from pore storage as it is displaced by gas (this representing an effective near-well source of water) and from reduction in water relative permeability following coal water desaturation. Drawdowns computed under single- and dual-phase conditions are compared along a cross section of the model domain after 1, 3, 8, and 20 years of extraction (Figure 8). These are consistent with the responses observed in Figure 7. It is apparent that at small times after the initiation of pumping (less than a year), C.R. Moore et al. Groundwater 53, no. 5: 737–747
741
Figure 5. Plan view of the model grid showing locations of the 34 CBM extraction wells as black squares. Wells 1, 2, and 3 are those depicted in Figure 2. Gray ellipses denote cells depicted in Figure 4. White ellipses denote the locations of vertical drawdown profiles provided in Figure 7.
drawdown is almost entirely confined to coal layers. As time increases, however, drawdowns propagate into neighboring interburden layers. A comparison between the left and right sides of Figure 8 illustrates the systematic overprediction of drawdown by MODFLOW. MODFLOW’s overprediction of drawdown is most pronounced in the vicinity of the extraction wellfield and at early pumping times. These differences diminish with increasing distance from the wellfield and with increasing time. After a year of pumping, MODFLOW drawdown overprediction ranges from less than a meter at a distance of 7 km from the wellfield to about 150 m around the margins of the wellfield and up to 300 m at points within the wellfield itself. After 20 years of pumping, differences between heads calculated by the two models are reduced; however, they affect a larger volume of the aquifer, including both the coal and the interburden. At this time, MODFLOW drawdown overprediction ranges from about 15 m at a distance of 7 km from the wellfield to 60 m at the margins of the wellfield to about 150 m within the wellfield itself.
Upscaling and Preservation of Relative Permeability Functions Because of the computational burden, models that are built to explore the effects of CBM extraction on regional groundwater systems do not represent coal measures with the same level of detail as the model described in the previous section. Instead, the entire coal measure sequence is typically approximated using a limited number of model layers; regional model layers above and below these coal measure layers are likely to represent aquifers and aquitards occurring within what may be a thick sedimentary sequence (QWC 2012). Results presented earlier suggest that, notwithstanding its simplified representation in a regional model, the 742
C.R. Moore et al. Groundwater 53, no. 5: 737–747
desaturation of coal that accompanies its depressurization should be taken into account in order to avoid overprediction of drawdown propagation within that model. This could be done by using a dual-phase model for regional simulation (QGC 2013); alternatively, it may be possible to use a single-phase model that is modified to allow desaturation to occur as pressures are reduced within coal seams (Moore et al. 2013). In either case, account must be taken of the reduction in relative permeability that occurs with reduction of water saturation. This requires that a dynamic pseudo-relative permeability function (or “pseudos”) be determined for upscaled coal measure layers in the regional model. The consequences for dynamic pseudo-relative permeability functions of two different coal measure upscaling methods were investigated as part of this study. In the first method, all lithologies occurring between certain fine model layers are combined into a single upscaled model layer that replaces all of the corresponding fine layers. The second method segregates coal from interburden material before undertaking layer amalgamation; coal and interburden are, thus, consigned to separate layers in the upscaled model. In both cases, the total thickness of the coal measures as represented in the upscaled model is the same as that of the coal measures itself. Where upscaling is restricted to the vertical direction only (i.e., where each of the above-described upscaled models possesses the same number of rows and columns as the fine-scale model), calculation of approximate upscaled hydraulic properties from fine-scale hydraulic properties can be achieved through vertical averaging using the formulas provided in Table 3. Averaging takes place over all fine-scale model layers that are represented in a single upscaled model layer, with segregation of lithologies taken into account if appropriate. Approximations to the system states that would be computed by the upscaled models can also be calculated through vertical averaging from those computed by the fine-scale model (Table 3). The vertical averaging processes that are listed in this table include calculation of upscaled hydraulic conductivity as it depends on the water saturation of an upscaled model cell. From this, dynamic pseudo-relative permeability functions can be computed; see Kyte and Berry (1975) and Darman et al. (2002) for further details. It should be noted that the above-mentioned method for calculation of pseudos, relying only on properties and states computed by the fine-scale model, is approximate, as an upscaled model is not actually constructed. The integrity of the methodology relies on the fact that upscaling takes place in the vertical direction only. Hence, coal seams that facilitate flow of water in the horizontal direction and clay layers that impede flow in the vertical direction will tend to be continuous across any one model cell (recall that for the model that is the subject of the present investigation cell dimensions are 200 m by 200 m—this being far shorter than the average length of coal seams). The continuity of conductive and resistive features within any cell suggests the use of arithmetic (in the horizontal direction) and harmonic (in the NGWA.org
Figure 6. Simulated water desaturation at times of (a) 1 year and (b) 20 years after commencement of pumping (vertical exaggeration is 15:1).
vertical direction) averaging as implemented implied by the equations of Table 3, in favor of geometric averaging; the latter would be more appropriate in an environment where heterogeneity is more random. As is shown later, upscaling strategies based on lithological amalgamation on the one hand and lithological segregation on the other hand result in very different pseudo-relative permeability functions. Errors incurred by the aforementioned approximations required for calculation of these functions are far outweighed by these differences. It should be noted, however, that the authors have built and run upscaled dual-phase models (upscaling takes place also in the horizontal direction in these models) to explore these issues further as an extension of this study. The outcomes of these runs support the conclusions presented herein and will be the subject of future publications.
Upscaling Through Lithological Amalgamation The first upscaling strategy that we examine follows the method that we refer to as “lithological averaging” in which each upscaled model cell represents all finescale model cells that lie within its boundaries (Figure 9). The upper four layers of the upscaled model represent the Springbok Sandstone, while the lower four layers represent the WCM. In implementing this upscaling method, hydraulic properties ascribed to each coarse model cell are averaged over both coal and interburden cells (refer to Table 3 for averaging methods). The sharp contrasts between interburden and coal permeabilities are therefore not represented in the upscaled model. The largest cell horizontal hydraulic conductivity in the upscaled model is 0.116 m/d; contrast this with a maximum value of 0.87 m/d for the fine-scale model. The highest porosity in the fine-scale model is 0.22, whereas that in the upscaled model is 0.16. A scatterplot of relative permeability vs. water saturation for all upscaled model cells for simulation times spanning 1 to 20 years (Figure 10a) shows that the relationship between dynamic relative permeability and water saturation for the upscaled model differs considerably from that employed by the fine-scale model (Figure 10b). The difference in these relationships is a consequence of NGWA.org
the fact that coal is responsible for most of the horizontal hydraulic conductivity of each upscaled model cell. At the same time, coal is the lithology that undergoes most water desaturation as gas is desorbed. However, there is far less coal than interburden in any upscaled model cell. Hence, while the relative permeability of an upscaled cell is reduced considerably as gas is desorbed, its upscaled water saturation is reduced very little. Another salient feature illustrated by Figure 10a and 10b is that uniformity of the real relative permeability curve does not lead to uniformity of the pseudo-relative permeability curve; the latter is obviously cell (and probably time) dependent and shows considerable scatter.
Upscaling Through Lithological Segregation As explained earlier, the second upscaling method separates coal from interburden while preserving total coal and interburden thickness within any vertical grid column. Thus, the multitude of discontinuous coal seams that occur within the coal measure formation are migrated vertically such that they comprise a small number of coal-only layers separated by correspondingly segregated interburden material. In calculating upscaled properties and system states, the vertical averaging process described in Table 3 becomes lithology specific for those layers that represent a single lithology. With lithological segregation forming the basis for upscaling of the WCM, the upscaled model now has 12 individual layers, with each of the four lower layers in the previous model now separated into individual interburden-only and coal-only layers (Figure 9). The dynamic relative permeability scatterplot emerging from this upscaling method is shown in Figure 11. It is apparent that the dynamic relative permeability relationship for the lithologically segregated upscaled model follows the true permeability relationship well; furthermore, scatter about this curve is reduced. If the horizontal hydraulic conductivity of interburden is sufficiently low, flow will be predominantly vertical within it. Drawdown induced by water and gas extraction will therefore diffuse from coal into interburden material. Results presented in the previous section show that this C.R. Moore et al. Groundwater 53, no. 5: 737–747
743
Figure 7. Vertical profiles of simulated single- and dual-phase drawdowns after 1, 5, and 20 years of continuous CBM extraction. Profiles are for cells (row, column) (a) (65, 30) and (b) (65, 60). See Figure 5 for the locations of these cells.
is indeed the case. Under these circumstances, the same upscaling outcome as lithological segregation can be achieved through use of a dual-porosity representation of flow within each upscaled coal measure layer (Warren and Root 1963). Dual-porosity concepts are often employed to represent flow through fractured rock (see Landereau et al. 2001 and references cited therein). In coal measures where high hydraulic conductivity coal layers are subparallel, it is anticipated that assignment of properties that govern diffusive flow from matrix to conductive pathways (composed of coal seams) within the dual-porosity medium will be much easier than in the more general case of fractured rock where high-conductivity flow paths are less ordered. 744
C.R. Moore et al. Groundwater 53, no. 5: 737–747
Discussion and Conclusions This study comprises detailed numerical simulations that explore the movement of water and gas induced by CBM extraction. In these simulations, the coal measure sequence from which extraction takes place is represented in a realistic manner with a level of lithological detail that is rare for groundwater models built to explore drawdown impacts on overlying aquifers. This study has provided insights into near-source processes; these insights may be important for adequate representation of these processes in upscaled regional models wherein the effects of CBM extraction on regional groundwater systems are assessed. NGWA.org
(a)
(d)
(b)
(c)
(e)
(f)
Figure 8. Simulated drawdown after 1 year (a and b), 8 years (c and d), and 20 years (e and f) of extraction. Dual-phase flow is assumed in computing the drawdowns depicted in a, c, and e while only single-phase flow is assumed in computing the drawdowns depicted in b, d, and f.
Table 3 Details of Vertical Averaging Applied to Calculation of Upscaled Hydraulic Properties and System States Parameter
Description of Upscaling
Horizontal hydraulic conductivity
Upscaled horizontal hydraulic conductivity is calculated as the saturated thickness-weighted arithmetic mean of fine-scale hydraulic conductivity, thus preserving transmissivity
Vertical hydraulic conductivity
Upscaled vertical hydraulic conductivity is calculated using thickness-weighted harmonic averaging, thus preserving resistance. Note, however, that vertical hydraulic conductivity is not employed in any pseudo-calculations
tu Kvu
Upscaled porosity is calculated as the thickness-weighted arithmetic mean of fine-scale porosities, this preserving overall void space
θu =
Porosity
Pressure head
Upscaled pressure is calculated by (1) converting fine-scale pressures to heads; (2) undertaking transmissivity-weighted arithmetic averaging of fine-scale heads; (3) converting upscaled heads to pressures
Saturation
Upscaled water saturation is calculated by summing thickness-weighted fine-scale saturations, thus conserving water
Relative permeability
Upscaled relative permeability is water saturation-dependent permeability divided by permeability at full saturation
Equations Khf tf Sf Khu = tu Su =
tf Kvf
θf tf
tu
hf Khf tf Ku tu
hu =
Su = krhu (Su ) =
Sf tf
tu khu (S u ) khu (Su =1)
Notes: In the formulas, the subscripts “f” and “u” indicate “fine-scale” and “upscaled”; K h and K v the horizontal and vertical hydraulic conductivities; k the permeability; t the layer thickness; θ the porosity; S the water saturation; and h the head.
The simulations also demonstrate the spatial and temporal characteristics of drawdown and water desaturation propagation from a pumping center to other parts of a coal measure sequence. Initially, pressure reduction is confined to coal layers as it propagates rapidly along these layers. As this occurs, gas is desorbed from the coal matrix and enters the cleat system of the coal. This desorption NGWA.org
slows the propagation of pressure drawdown compared to that which would be induced without gas desorption with the same amount of water extraction. Two factors contribute to this slowing of drawdown propagation. The first is that water is removed from pore storage in addition to elastic storage, thereby providing a temporary source of water which must be transported to extraction wells. The C.R. Moore et al. Groundwater 53, no. 5: 737–747
745
Figure 9. Eight-layered upscaled model structure.
Figure 10. (a) Scatterplot of relative permeability vs. water saturation where upscaling involves lithological amalgamation. (b) The horizontal scale is expanded to show the true water saturation vs. permeability relationship in comparison.
Figure 11. Scatterplot of relative permeability vs. water saturation where upscaling involves lithological segregation.
second is that the relative permeability of coal in the vicinity of extraction sites is considerably reduced as methane replaces water in coal cleats, and must thereby be transported with it to extractive sites. Meanwhile, over time, pressure reduction propagates vertically from coal layers into the neighboring interburden layers. This leads to a more even vertical distribution of drawdown throughout the whole coal measure sequence. 746
C.R. Moore et al. Groundwater 53, no. 5: 737–747
Insights provided by this study have a number of repercussions for modeling of the impact of CBM extraction on regional groundwater systems. The first is that, despite the localization of water desaturation to the vicinities of extraction sites, there are dangers in ignoring nearwell desaturation in regional modeling. Even under the comparatively benign conditions presented in this article wherein ECLIPSE-calculated water extraction is provided to a single-phase model, drawdown overprediction by the single-phase model can be significant. At early pumping times (1 year), overprediction of drawdown by the singlephase model is as high as 300 m in the vicinity of wells and as high as 150 m at the margin of the extraction wellfield, falling to 1 m at a distance of 7 km from the extraction center. At later times, peak drawdown overprediction is reduced, but prevails over a larger rock volume, as it occurs in both coal and interburden material. Thus, at a simulation time of 20 years, drawdown overprediction is as high as 15 m at a distance of 7 km from the well extraction center, rising to 60 m at the margins of the wellfield, and rising again to 150 m within the extraction wellfield itself. Moore et al. (2013) demonstrate that in intervening years, locations of maximum drawdown overprediction tend to form a halo around an expanding ring of coal water desaturation that propagates outward from the pumping center. Because the neglect of water desaturation can lead to overprediction of drawdown within coal measures from which extraction takes place, this can also lead to drawdown overprediction in underlying and/or overlying aquifers as well, depending on the geological context. The extent of this overprediction will depend, of course, on the hydraulic conductivity of intervening aquitard material (or lack thereof). A second implication for regional modeling, wherein coal measures (and the physical processes associated with CSG extraction) must be represented in a highly upscaled manner, is that coal should be either explicitly or implicitly segregated from interburden in such a model. Coals can be explicitly represented by devoting specific upscaled model layers to the representation of either coal or interburden, but not to both. Alternatively, coals can be implicitly represented through assigning dual-porosity status to upscaled layers from which CBM extraction takes place. In either case, it is likely that the relationship between relative permeability and water saturation will be unaffected, or only mildly affected, by the upscaling process, and so requires little or no upscaling itself. It has yet to be determined whether it is better to represent coal and interburden layers individually or through a dualporosity mechanism. Factors not considered in this short study, such as correct apportionment of well extraction between coal and interburden layers, may need to be taken into account in making this judgment. Moore et al. (2013) demonstrate the potential importance of well-lithology interactions in the overall upscaling process. Issues that are only touched on in this article, but that are deserving of further research in order that a better basis for regional simulation of CBM extraction impacts can be provided, include the following: NGWA.org
•
Parameterization of upscaled representations of coal and interburden where regional models have a large domain and model cells are large. • Use of a single-phase model that allows water desaturation near extraction points as a substitute for dual-phase simulation. • Upscaling of extraction processes in a regional model. These issues are being researched by the authors of this article.
Acknowledgments The authors would like to thank the Queensland Gas Company (QGC) and CSIRO for co-funding the project on which this article is based, and to staff within those organizations who assisted the authors in this larger project, including Dan de Verteuil (QGC), and Tao Cui (CSIRO). The authors would also like to acknowledge the reviewers contribution to this article.
References Barker, J.W., and S. Thibeau. 1997. A critical review of the use of pseudorelative permeabilities for upscaling. SPE Reservoir Engineering 12, no. 2: 138–143. Centre for Water in the Minerals Industry (CWIMI). 2008. Scoping study: Groundwater impacts of coal seam gas development—assessment and monitoring. Report Document Reference P08-010-002. Brisbane, Australia: Department of Infrastructure and Planning. Darman, N.H., G.E. Pickup, and K.S. Sorbie. 2002. A comparison of dual-phase dynamic upscaling methods based on fluid potentials. Computational Geosciences 6, no. 1: 5–27. DOI:10.1023/A:1016572911992. DNRM. 2013. Groundwater risks associated with coal seam gas development in the Surat and southern Bowen basins. In Healthy HeadWaters CSG Water Feasibility Study, Activity 5 Groundwater Risks Associated With Coal Seam Gas: Development in the Surat and Southern Bowen Basins. Prepared by Worley Parsons for the Department of Natural Resources and Mines (DNRM), Queensland, Australia. Esterle, J.S., S.K. Hamilton, V. Ward, S. Tyson, and R. Sliwa. 2013. Scales of geological heterogeneity within the Walloon Subgroup and its coal measures, February. Final Report of Activity 1.3 of the Healthy HeadWaters Coal Seam Gas Water Feasibility Study. Uniquest project no. 00572. Department of Natural Resources and Mines, University of Queensland. Freij-Ayoub, R. 2012. Opportunities and challenges to coal bed methane production in Australia. Journal of Petroleum Science and Engineering 88–89: 1–4. DOI:10.1016/ j.petrol.2012.05.001. Gasda, S.E., and M.A. Celia. 2005. Upscaling relative permeabilities in a structured porous medium. Advances in
NGWA.org
Water Resources 28, no. 5: 493–506. DOI:10.1016/j.advw atres.2004.11.009. Harbaugh, A.W. 2005. MODFLOW-2005. The U.S. Geological Survey Modular Ground-Water Model – Ground-Water Flow Process. Reston, Virginia: U.S. Geological Survey Techniques and Methods 6-A16. Harrison, S.M., J.W. Molson, H.J. Abercrombie, J.F. Barker, D. Rudolph, and R. Aravena. 2000. Hydrogeology of a coalseam gas exploration area, southeastern British Columbia, Canada: Part 1. Groundwater flow systems. Hydrogeology Journal 8, no. 6: 608–622. Hewett, H.A., and R.A. Archer. 1997. Scale-averaged effective flow properties for coarse-grid reservoir simulation. In 14th SPE Reservoir Simulation Symposium, June 8–11. Paper SPE 37988. Dallas, Texas. Kyte, J.R., and D.W. Berry. 1975. Pseudo functions to control numerical dispersion. Society of Petroleum Engineers Journal 15, no. 4: 269–275. Landereau, P., B. Noetinger, and M. Quintard. 2001. Quasisteady two-equation models for diffusive transport in fractured porous media: Large-scale properties for a densely fractured system. Advances in Water Resources 24, no. 8: 863–876. Moore, T.A. 2012. Coalbed methane: A review. International Journal of Coal Geology 101: 36–81. Moore, C.R., J.E. Doherty, S. Howell, L. Erriah, D. de Verteuil, and T. Cui. 2013. Challenges for Upscaling Hydraulic Properties and Processes in the Coal Seam Gas Context. Melbourne, Australia: CSIRO. Myers, T. 2009. Groundwater management and coal bed methane development in the Powder River Basin of Montana. Journal of Hydrology 368, no. 1–4: 178–193. DOI:10.1016/j.jhydrol.2009.02.001. Nordbotten, J.M., J.P. Nogues, and M.A. Celia. 2010. Appropriate choice of average pressure for upscaling relative permeability in dynamic flow conditions. SPE Journal 15, no. 1: 228–237. DOI:10.2118/113558-PA. QGC: A BG Group Business. 2013. Developing robust water management tools. http://www.qgc.com.au/media/233081/ 3.0_-_developing_robust_water_management_tools.pdf (accessed September 10, 2014). QWC. 2012. Underground Water Impact Report for the Surat Cumulative Management Area. Queensland, Australia: Queensland Water Commission http://www.dnrm. qld.gov.au/__data/assets/pdf_file/0016/31327/undergroundwater-impact-report.pdf (accessed September 10, 2014). Ryan, D.J., A. Hall, L. Erriah, and P.B. Wilson. 2012. The Walloon coal seam gas play, Surat Basin, Queensland. APPEA Journal 52: 273–289. Schlumberger. 2011. Manual for ECLIPSE Reservoir Simulation Software. Houston, Texas: Schlumberger. Stone, H.L. 1991. Rigorous black oil pseudo functions. In 11th SPE Reservoir Simulation Symposium, February 17–20. Paper SPE 21207. Anaheim, California. Terzaghi, K., R.B. Peck, and G. Mesri. 1996. Soil Mechanics in Engineering Practice, 3rd ed. New York: WileyInterscience. Warren, J., and P. Root. 1963. The behaviour of naturally fractured reservoirs. SPE Journal 3: 245–255.
C.R. Moore et al. Groundwater 53, no. 5: 737–747
747
Abstract Coal measures (coal bearing rock strata) can contain large reserves of methane. These reserves are being exploited at a rapidly increasing rate in many parts of the world. To extract coal seam gas, thousands of wells are drilled at relatively small spacing to depressurize coal seams to induce desorption and allow subsequent capture of the gas. To manage this process effectively, the effect of coal bed methane (CBM) extraction on regional aquifer systems must be properly understood and managed. Groundwater modeling is an integral part of this management process. However, modeling of CBM impacts presents some unique challenges, as processes that are operative at two very different scales must be adequately represented in the models. The impacts of large-scale gas extraction may be felt over a large area, yet despite the significant upscaling that accompanies construction of a regional model, near-well conditions and processes cannot be ignored. These include the highly heterogeneous nature of many coal measures, and the dual-phase flow of water and gas that is induced by coal seam depressurization. To understand these challenges, a fine-scale model was constructed incorporating a detailed representation of lithological heterogeneity to ensure that near-well processes and conditions could be examined. The detail of this heterogeneity was at a level not previously employed in models built to assess groundwater impacts arising from CBM extraction. A dual-phase reservoir simulator was used to examine depressurization and water desaturation processes in the vicinity of an extractive wellfield within this fine-scale model. A single-phase simulator was then employed so that depressurization errors incurred by neglecting near-well, dual-phase flow could be explored. Two models with fewer lithological details were then constructed in order to examine the nature of depressurization errors incurred by upscaling and to assess the interaction of the upscaling process with the requirement for adequate representation of near-source, dual-phase processes.
Introduction Extraction of coal bed methane (CBM, also referred to as “coal seam gas”) is expanding rapidly in a number of countries, including the United States, Australia, China, and India (Moore 2012). Depressurization of coal beds that is necessary for extraction of CBM is achieved by pumping water from these beds until the hydrostatic pressure is sufficiently lowered to induce desorption of methane from the coal matrix. The resulting drawdown in groundwater levels is likely to propagate laterally, 1 Department of Groundwater, Institute of Environmental Science and Research Ltd, Porirua 5240, New Zealand. 2 Corresponding author: CSIRO Land & Water, Australia Ecosciences Precinct, 41 Boggo Road, Dutton Park, QLD 4102, Australia; [email protected] 3 National Centre for Groundwater Research and Training, Flinders University, GPO Box 2100, Adelaide, SA 5001, Australia; [email protected] 4 Watermark Numerical Computing, Brisbane, Australia. 5 QGC – A BG Group Business, Groundwater project, Level 18, 69 Ann Street, Brisbane, QLD 4000, Australia; [email protected]; [email protected] Received December 2013, accepted January 2014. © 2014, National Ground Water Association. doi: 10.1111/gwat.12276
NGWA.org
and possibly vertically to other hydrogeological strata, where it could affect extraction from water supply wells, and reduce flow to groundwater-dependent ecosystems (CWIMI 2008; DNRM 2013). Suitable management practices, such as strategic reinjection of extracted water, and the deepening of existing production wells, can be used to mitigate these impacts (e.g., Myers 2009; Freij-Ayoub 2012). Regional groundwater models can be used to assess the potential for these and other impacts of CBM development and to provide a basis for regional groundwater management. However, because of the rapid and relatively recent growth of the CBM industry, only a limited number of these types of models have been reported in the literature (Harrison et al. 2000; Myers 2009; QWC 2012; QGC 2013). The modeling of regional groundwater drawdown induced by CBM extraction poses a number of significant challenges. On the one hand, these models must often encompass large domains and must include significant aquifers and aquitards that prevail in the large sedimentary basins wherein coal sequences are found. They must also encompass the locations where the adverse consequences of CBM extraction may be felt, which can be far removed from extraction sites (QGC 2013). On the other
Vol. 53, No. 5–Groundwater–September-October 2015 (pages 737–747)
737
hand, adequate representation of hydrogeological detail near extraction sites may be important. This is because individual coal layers within an extensive coal measure sequence may be of limited lateral extent, thus reducing connectivity to regional groundwater systems and limiting regional drawdown impacts. The challenges of representing these different scales is made more complicated by the necessity for these models to accommodate, if not explicitly simulate, dualphase flow of water and gas to extraction wells (Moore et al. 2013). Unfortunately, explicit representation of dual-phase flow imposes a significant computational burden on a regional model. Hence, usually a simpler single-phase model such as MODFLOW (Harbaugh 2005), which runs relatively quickly, is used for CBM regional impact assessments (Myers 2009; QWC 2012, among others). Reduced model run times assist model calibration and calibration-constrained uncertainty analysis, both of which may require many thousands of model runs. However, as is discussed in the section titled “Near-Well Desaturation and Drawdown Propagation,” the use of a single-phase model to estimate regional drawdown induced by CBM extraction can result in overprediction of these drawdowns. Simulation of near-well, dual-phase flow or a singlephase substitute for dual-phase flow (Moore et al. 2013) requires that a suitable relationship between relative permeability and water saturation be developed for the upscaled representation of simulated coal sequences. These relationships can have a critical effect on the simulation of propagation of drawdown outward from these wells. A further requirement is the employment of suitable “dynamic pseudo-relative permeability” functions, or “pseudos” as these are often called in the petroleum literature (Kyte and Berry 1975; Stone 1991; Barker and Thibeau 1997; Hewett and Archer 1997; Darman et al. 2002; Gasda and Celia 2005; Nordbotten et al. 2010). To the authors’ knowledge these matters have received no attention in either the CBM reservoir modeling or CBM regional impact assessment modeling literature. The study reported herein uses detailed numerical simulations of near-well processes utilizing a dual-phase simulator that is employed throughout the oil and gas industry for modeling oil, water, and gas production, namely the ECLIPSE reservoir simulator (Schlumberger 2011). The study focuses on the development and progression of depressurization and water desaturation in the vicinity of a CBM extraction wellfield. Because of the high level of lithological detail represented in the model (which, as far as the authors know, is unprecedented in a study of this kind), the pressure and saturation states of the system can be simulated at a fine-scale. Examination of the magnitudes and changing nature of these states provides unique insights into the issues that any upscaled representation of these processes in a regional assessment model must address. These issues include the nature of drawdown and desaturation propagation through a highly heterogeneous sequence of coal seams, the spatial and temporal nature of errors incurred when simulating CBM 738
C.R. Moore et al. Groundwater 53, no. 5: 737–747
extraction using a single-phase model, the consequences of upscaling on calculation of dynamic pseudo-relative permeabilities, and appropriate upscaling of coal measures in a regional impact assessment model. This article is organized as follows. First, the details of the model that forms the basis of the case study reported in this article are presented. The following section describes the simulation of water desaturation and drawdown propagation accompanying CBM extraction at a fine scale. It also presents an examination of the differences between drawdown predictions made using a single-phase model and those made using a dual-phase model at this same fine scale. Upscaling issues are then addressed, with particular attention paid to the effects of two different upscaling strategies on dynamic pseudo-relative permeability functions. The first upscaling strategy is that most commonly employed in reservoir and groundwater simulation whereby an upscaled model layer represents all fine-scale model layers that lie within its boundaries. The second involves segregation of coal seams into separate layers. The advantages of the latter upscaling methodology are demonstrated. Finally, the outcomes of this work are discussed, and conclusions are drawn for the representation of near-well processes in regional modeling.
Case Study The case study used for this article is based on a section of the Walloon Coal Measures (WCM) in the Surat Basin, Queensland, Australia. The WCM are approximately 350-m thick along the northern and eastern margins of the basin, and thicken to more than 500 m in the south-west of the basin. Their coal content varies with location. At the study location, about 10% of their thickness is composed of coal, this being typical of many CBM exploration and production areas in the Surat Basin. Coal occurs as thin, discontinuous seams, with individual coal plies ranging in thickness between 0.05 and 1.7 m. Up to 45 individual seams can be recognized in places. They are separated by siltstone, mudstone, and fine-to-medium grained lithic sandstone interburden. The lateral extent of individual seams varies considerably and is thought to range from about 500 m to more than 3000 m (Ryan et al. 2012). The WCM are overlain by the Springbok Sandstone that is composed of low permeability, fine-grained, often clay-rich, sandstone, siltstone, and mudstone (Ryan et al. 2012; Esterle et al. 2013). A stochastic realization of coal and interburden lithofacies within the WCM and overlying Springbok Sandstone was generated for use by the model (Figure 1a). This was conditioned to borehole petrophysical data within the study area; its stochasticity is based on extensive company datasets acquired in this and neighboring development areas. The geological block model populated with this realization comprises the simulation domain used in the present investigation (Figure 1b), the details of which are given in Table 1. Neuman (no-flow) boundary conditions surround the entire block model. Initial hydraulic heads NGWA.org
Figure 1. (a) Stochastic realization of lithofacies within the Walloon Coal Measures and Springbok Sandstone. Coal is colored green (note that a small amount of coal occurs within the Springbok Sandstone). (b) Model grid and location of CBM extraction wells. Vertical exaggeration is 15:1.
Table 1 Block Model Dimension and Discretization Details Block Model Detail
Dimensions
Horizontal dimensions Thickness Grid
17,700 m × 19,500 m 600–700 m 131 rows, 119 columns, 614 layers Layers 1–162 Layers 163–614
Springbok sandstone Walloon coal measures
were assigned a constant value of 262 m, ensuring fully saturated and confined conditions throughout the model domain. Lithofacies within the stochastic realization that forms the basis of this study were populated with a stochastic realization of hydraulic properties that are representative of those prevailing within the study area. Ranges of some of these properties as they pertain to different model stratigraphic subdivisions are provided in Table 2. Certain properties, particularly those related to gas, are omitted from this table because of their proprietary nature. The authors wish to point out, however, that failure to provide these properties does not detract from the generality of this study, as the conclusions that are drawn from it are not specific to any particular set of properties, but are applicable in all contexts of regional coal seam gas (CSG) impact assessment. The most salient feature of the data provided in Table 2 is that the hydraulic conductivity (both horizontal and vertical) of coal is considerably greater than that of both the interburden material with which it is closely associated and the material comprising the overlying sandstone/siltstone/shale of the Springbok Sandstone. Flow of gas and water toward extraction wells is thus predominantly horizontal. As stated previously, the ECLIPSE reservoir simulator (Schlumberger 2011) was used to simulate gas production and movement of water and gas through the model domain. Coal is represented as a dual-porosity medium in which fluid moves through cleats (natural fractures in NGWA.org
coal); meanwhile, gas desorbed from the coal matrix first diffuses through the matrix and then moves through the cleat system as a separate phase from that of water. Gas desorbs as the pressure is lowered, this being governed by a Langmuir isotherm. A uniform Langmuir isotherm was assigned to all coal within the model domain, this being typical of that employed in gas company reservoir modeling. This is based on coal sample laboratory measurements and on local reservoir modeling experience. Use of a uniform isotherm allows this study to focus on the ramifications of lithological and hydraulic property heterogeneity. Water and gas are produced through a total of 34 wells. These are placed 750-m apart; each is screened across the entirety of the WCM (Figure 1b). Pumping is simulated for 20 years as water and gas are produced from each well subject to a total water flow constraint of 2000 stb/day (i.e., stock tank barrels/day where a barrel is 159 L) or 318 m3 /day, and a minimum bottom-hole pressure constraint of 90 psia (i.e., 620.5 kPa). (The reference level for this pressure is the elevation of the center of the uppermost layer from which extraction takes place.) Water extraction rates for three wells over the 20-year simulation period are shown in Figure 2. In accordance with the pumping specifications stated earlier, peak water extraction for each well is about 300 m3 /day. This peak water production rate prevails within the first 300 days of pumping, after which the rate of water production declines (Figure 2). The decline accelerates after about 1600 days. This is an outcome of the fact that coal connectivity restricts the lengths of connected coal geobodies to about 3 km; hence, the expanding cone of pressure depression encounters a low hydraulic conductivity boundary. Similar water production patterns are exhibited by other wells. Gas production peaks after about 300 days; it declines after about 1600 days as gas within coal seams is depleted and as access to further gas is inhibited by lack of coal connectivity. Gas production is not plotted for proprietary reasons. A uniform relationship between relative permeability and water saturation is applied to all cells within the model domain, regardless of lithology; this relationship is depicted in Figure 3. C.R. Moore et al. Groundwater 53, no. 5: 737–747
739
Table 2 Stratigraphic Zones, Lithology, and Hydraulic Property Ranges Assigned to the Block Model Stratigraphy
Fine-Scale Model Layers
Upper Springbok Mid Springbok Lower Springbok Basal Springbok Upper Juandah
1–70 71–135 136–159 160–162 163–287
Lower Juandah
288–432
Tangalooma
433–522
Taroom
523–614
Dominant Lithology Sandstone/siltstone/shale Sandstone/siltstone/shale Sandstone/siltstone/shale Sandstone/siltstone/shale Coal2 Interburden1 Coal2 Interburden1 Coal2 Interburden1 Coal2 Interburden1
Hydraulic Conductivity (Horizontal, m/day)
Hydraulic Conductivity (Vertical, m/day)
6.34 × 10−5 –6.45 × 10−2 6.34 × 10−5 –6.45 × 10−2 6.37 × 10−5 –6.838 × 10−3 6.37 × 10−5 –9.997 × 10−3 3.056 × 10−3 –8.71 × 10−1 1.27 × 10−6 –1.97 × 10−3 5.09 × 10−3 –7.72 × 10−1 1.273 × 10−6 –1.97 × 10−3 1.02 × 10−4 –1.961 × 101 1.27 × 10−6 –1.4 × 10−3 1.019 × 10−4 –1.96 × 101 1.27 × 10−6 –1.4 × 10−3
2.55 × 10−5 –2.04 × 10−2 2.55 × 10−5 –2.04 × 10−2 3.82 × 10−10 –3.82 × 10−8 1.27 × 10−5 –3.158 × 10−3 1.273 × 10−8 4.839 × 10−8 1.273 × 10−8 4.839 × 10−8 1.273 × 10−8 4.839 × 10−8 1.273 × 10−8 4.839 × 10−8
Porosity4 0.001–0.16 0.001–0.16 0.001–0.15 0.001–0.22 0.013 0.001–0.18 0.013 0.001–0.18 0.013 0.001–0.18 0.013 0.001–0.18
Note: As shown in the table, the Walloon Coal Measures are subdivided into the Upper and Lower Juandah Formations, the Tangalooma Sandstone, and the Taroom Coal Measures. 1 Interburden comprises siltstone, silty sandstone, and sandstone. 2 The lateral extent of any coal seam is considered to average about 3000 m. 3 Coal porosities were assigned a uniform value of 0.01, which was assumed to characterize cleat volume fraction. 4 Specific storage was calculated based on porosity, and coal measure and Springbok compressibilities of 2.0e−5 and 3.0e−6 /psi, respectively, using Terzaghi’s consolidation theory (Terzaghi et al. 1996).
Figure 2. Water extraction rates over the 20-year simulation period for three representative wells. The locations of these wells are shown in Figure 5.
Figure 3. Relative permeability vs. water saturation relationship used by ECLIPSE for all model cells.
Near-Well Desaturation and Drawdown Propagation Propagation of drawdown and water desaturation within the heterogeneous coal measure sequence that forms the basis of this study is now explored. The errors in simulated drawdown incurred through singlephase representation of these fundamentally dual-phase processes are then examined. As far as the authors are aware, there are no previously published studies that examine propagation of drawdown and water desaturation induced by CBM extraction in such detail, partly because the focus of most reservoir modeling is on water and 740
C.R. Moore et al. Groundwater 53, no. 5: 737–747
gas production rather than on pressure and desaturation propagation within the subsurface flow domain. Nor are the authors aware of any studies that attempt to quantify errors that are incurred through use of a traditional singlephase groundwater model in place of a reservoir model for calculation of drawdowns in this context. Vertical profiles of coal occurrence and head and water saturation calculated by the ECLIPSE model for a simulation time of 20 years (Figure 4) are shown for model cells (row, column) (52, 62) and (78, 78) (see Figure 5 for their locations). These indicate that, for this simulation time, water desaturation is predominantly restricted to coal layers. It is noteworthy, however, that some temporary, localized interburden water desaturation occurs at earlier times as gas desorbed from coal migrates upward before being captured by pumping-induced hydraulic gradients. CBM extraction incurs greater drawdown in coal than in the interburden, this being an outcome of the much higher horizontal hydraulic conductivity of the former than the latter. Drawdown in the overlying Springbok Sandstone is minimal, this reflecting the low vertical hydraulic conductivity awarded to this formation in the present model. Simulated water desaturation after 1, 3, 8, and 20 years of extraction confirms that desaturation is largely restricted to coal seams (Figure 6). However, as stated earlier, a close inspection reveals small and temporary pockets of desaturation in interburden. The small, expanding island of water desaturation in the upper right of the model domain is of interest. This occurs in a highly permeable but particularly thin coal seam (less than 0.1 m in thickness), surrounded both vertically and horizontally by low hydraulic conductivity interburden. Drawdown in the coal seam is sufficient to induce gas desorption. However, the hydraulic gradient in the seam is insufficient to capture the desorbed gas. Movement of gas to the wellfield would require partial water desaturation of NGWA.org
Figure 4. Vertical profiles of coal occurrence, water saturation, and head (in meters relative to sea level) at (a) column 62, row 52 and (b) column 78, row 78 after a simulation time of 20 years. The locations of these cells are shown in Figure 5. Note that the conversion of pressure to head uses the formula 1 psi = 2.30 ft of water; conversion of psi to psia requires subtraction of 14.695 psi to account for atmospheric pressure.
the surrounding interburden material in order to increase its relative permeability to gas. Not enough gas is released in the thin coal seam for this to happen.
Drawdowns With and Without Dual-Phase Flow Simulation Drawdowns computed using ECLIPSE are now compared with those computed by a traditional single-phase groundwater simulator MODFLOW-2005 (Harbaugh 2005). Corresponding cells in the ECLIPSE and MODFLOW models have the same hydraulic properties. Water extraction rates provided to MODFLOW wells are those computed by ECLIPSE for corresponding wells; recall that wells in the ECLIPSE model have target extraction rates that are subject to a bottom-hole pressure constraint. This strategy of assigning ECLIPSE-calculated water extraction rates to the MODFLOW model eliminates errors incurred by differences in pumping rates that would occur if MODFLOW were provided with the same target extraction rates and bottom-hole pressure constraints as those provided to ECLIPSE but under single-phase conditions. (Note that before comparing heads under NGWA.org
CBM extractive conditions, heads calculated by ECLIPSE under single-phase conditions were compared with those calculated by MODFLOW to ensure model parameter equivalence; these were found to be identical. Singlephase conditions were simulated with ECLIPSE by assigning zero initial gas concentrations to all coal cells.) Head drawdown profiles calculated by the two simulators down vertical columns of the model domain (see Figure 5 for their locations) are compared after 1, 3, 5, 10, and 20 years of continuous CBM extraction (Figure 7). It is apparent that use of a single-phase simulator leads to overprediction of drawdown. This results from the inability of the single-phase model to simulate loss of water from pore storage as it is displaced by gas (this representing an effective near-well source of water) and from reduction in water relative permeability following coal water desaturation. Drawdowns computed under single- and dual-phase conditions are compared along a cross section of the model domain after 1, 3, 8, and 20 years of extraction (Figure 8). These are consistent with the responses observed in Figure 7. It is apparent that at small times after the initiation of pumping (less than a year), C.R. Moore et al. Groundwater 53, no. 5: 737–747
741
Figure 5. Plan view of the model grid showing locations of the 34 CBM extraction wells as black squares. Wells 1, 2, and 3 are those depicted in Figure 2. Gray ellipses denote cells depicted in Figure 4. White ellipses denote the locations of vertical drawdown profiles provided in Figure 7.
drawdown is almost entirely confined to coal layers. As time increases, however, drawdowns propagate into neighboring interburden layers. A comparison between the left and right sides of Figure 8 illustrates the systematic overprediction of drawdown by MODFLOW. MODFLOW’s overprediction of drawdown is most pronounced in the vicinity of the extraction wellfield and at early pumping times. These differences diminish with increasing distance from the wellfield and with increasing time. After a year of pumping, MODFLOW drawdown overprediction ranges from less than a meter at a distance of 7 km from the wellfield to about 150 m around the margins of the wellfield and up to 300 m at points within the wellfield itself. After 20 years of pumping, differences between heads calculated by the two models are reduced; however, they affect a larger volume of the aquifer, including both the coal and the interburden. At this time, MODFLOW drawdown overprediction ranges from about 15 m at a distance of 7 km from the wellfield to 60 m at the margins of the wellfield to about 150 m within the wellfield itself.
Upscaling and Preservation of Relative Permeability Functions Because of the computational burden, models that are built to explore the effects of CBM extraction on regional groundwater systems do not represent coal measures with the same level of detail as the model described in the previous section. Instead, the entire coal measure sequence is typically approximated using a limited number of model layers; regional model layers above and below these coal measure layers are likely to represent aquifers and aquitards occurring within what may be a thick sedimentary sequence (QWC 2012). Results presented earlier suggest that, notwithstanding its simplified representation in a regional model, the 742
C.R. Moore et al. Groundwater 53, no. 5: 737–747
desaturation of coal that accompanies its depressurization should be taken into account in order to avoid overprediction of drawdown propagation within that model. This could be done by using a dual-phase model for regional simulation (QGC 2013); alternatively, it may be possible to use a single-phase model that is modified to allow desaturation to occur as pressures are reduced within coal seams (Moore et al. 2013). In either case, account must be taken of the reduction in relative permeability that occurs with reduction of water saturation. This requires that a dynamic pseudo-relative permeability function (or “pseudos”) be determined for upscaled coal measure layers in the regional model. The consequences for dynamic pseudo-relative permeability functions of two different coal measure upscaling methods were investigated as part of this study. In the first method, all lithologies occurring between certain fine model layers are combined into a single upscaled model layer that replaces all of the corresponding fine layers. The second method segregates coal from interburden material before undertaking layer amalgamation; coal and interburden are, thus, consigned to separate layers in the upscaled model. In both cases, the total thickness of the coal measures as represented in the upscaled model is the same as that of the coal measures itself. Where upscaling is restricted to the vertical direction only (i.e., where each of the above-described upscaled models possesses the same number of rows and columns as the fine-scale model), calculation of approximate upscaled hydraulic properties from fine-scale hydraulic properties can be achieved through vertical averaging using the formulas provided in Table 3. Averaging takes place over all fine-scale model layers that are represented in a single upscaled model layer, with segregation of lithologies taken into account if appropriate. Approximations to the system states that would be computed by the upscaled models can also be calculated through vertical averaging from those computed by the fine-scale model (Table 3). The vertical averaging processes that are listed in this table include calculation of upscaled hydraulic conductivity as it depends on the water saturation of an upscaled model cell. From this, dynamic pseudo-relative permeability functions can be computed; see Kyte and Berry (1975) and Darman et al. (2002) for further details. It should be noted that the above-mentioned method for calculation of pseudos, relying only on properties and states computed by the fine-scale model, is approximate, as an upscaled model is not actually constructed. The integrity of the methodology relies on the fact that upscaling takes place in the vertical direction only. Hence, coal seams that facilitate flow of water in the horizontal direction and clay layers that impede flow in the vertical direction will tend to be continuous across any one model cell (recall that for the model that is the subject of the present investigation cell dimensions are 200 m by 200 m—this being far shorter than the average length of coal seams). The continuity of conductive and resistive features within any cell suggests the use of arithmetic (in the horizontal direction) and harmonic (in the NGWA.org
Figure 6. Simulated water desaturation at times of (a) 1 year and (b) 20 years after commencement of pumping (vertical exaggeration is 15:1).
vertical direction) averaging as implemented implied by the equations of Table 3, in favor of geometric averaging; the latter would be more appropriate in an environment where heterogeneity is more random. As is shown later, upscaling strategies based on lithological amalgamation on the one hand and lithological segregation on the other hand result in very different pseudo-relative permeability functions. Errors incurred by the aforementioned approximations required for calculation of these functions are far outweighed by these differences. It should be noted, however, that the authors have built and run upscaled dual-phase models (upscaling takes place also in the horizontal direction in these models) to explore these issues further as an extension of this study. The outcomes of these runs support the conclusions presented herein and will be the subject of future publications.
Upscaling Through Lithological Amalgamation The first upscaling strategy that we examine follows the method that we refer to as “lithological averaging” in which each upscaled model cell represents all finescale model cells that lie within its boundaries (Figure 9). The upper four layers of the upscaled model represent the Springbok Sandstone, while the lower four layers represent the WCM. In implementing this upscaling method, hydraulic properties ascribed to each coarse model cell are averaged over both coal and interburden cells (refer to Table 3 for averaging methods). The sharp contrasts between interburden and coal permeabilities are therefore not represented in the upscaled model. The largest cell horizontal hydraulic conductivity in the upscaled model is 0.116 m/d; contrast this with a maximum value of 0.87 m/d for the fine-scale model. The highest porosity in the fine-scale model is 0.22, whereas that in the upscaled model is 0.16. A scatterplot of relative permeability vs. water saturation for all upscaled model cells for simulation times spanning 1 to 20 years (Figure 10a) shows that the relationship between dynamic relative permeability and water saturation for the upscaled model differs considerably from that employed by the fine-scale model (Figure 10b). The difference in these relationships is a consequence of NGWA.org
the fact that coal is responsible for most of the horizontal hydraulic conductivity of each upscaled model cell. At the same time, coal is the lithology that undergoes most water desaturation as gas is desorbed. However, there is far less coal than interburden in any upscaled model cell. Hence, while the relative permeability of an upscaled cell is reduced considerably as gas is desorbed, its upscaled water saturation is reduced very little. Another salient feature illustrated by Figure 10a and 10b is that uniformity of the real relative permeability curve does not lead to uniformity of the pseudo-relative permeability curve; the latter is obviously cell (and probably time) dependent and shows considerable scatter.
Upscaling Through Lithological Segregation As explained earlier, the second upscaling method separates coal from interburden while preserving total coal and interburden thickness within any vertical grid column. Thus, the multitude of discontinuous coal seams that occur within the coal measure formation are migrated vertically such that they comprise a small number of coal-only layers separated by correspondingly segregated interburden material. In calculating upscaled properties and system states, the vertical averaging process described in Table 3 becomes lithology specific for those layers that represent a single lithology. With lithological segregation forming the basis for upscaling of the WCM, the upscaled model now has 12 individual layers, with each of the four lower layers in the previous model now separated into individual interburden-only and coal-only layers (Figure 9). The dynamic relative permeability scatterplot emerging from this upscaling method is shown in Figure 11. It is apparent that the dynamic relative permeability relationship for the lithologically segregated upscaled model follows the true permeability relationship well; furthermore, scatter about this curve is reduced. If the horizontal hydraulic conductivity of interburden is sufficiently low, flow will be predominantly vertical within it. Drawdown induced by water and gas extraction will therefore diffuse from coal into interburden material. Results presented in the previous section show that this C.R. Moore et al. Groundwater 53, no. 5: 737–747
743
Figure 7. Vertical profiles of simulated single- and dual-phase drawdowns after 1, 5, and 20 years of continuous CBM extraction. Profiles are for cells (row, column) (a) (65, 30) and (b) (65, 60). See Figure 5 for the locations of these cells.
is indeed the case. Under these circumstances, the same upscaling outcome as lithological segregation can be achieved through use of a dual-porosity representation of flow within each upscaled coal measure layer (Warren and Root 1963). Dual-porosity concepts are often employed to represent flow through fractured rock (see Landereau et al. 2001 and references cited therein). In coal measures where high hydraulic conductivity coal layers are subparallel, it is anticipated that assignment of properties that govern diffusive flow from matrix to conductive pathways (composed of coal seams) within the dual-porosity medium will be much easier than in the more general case of fractured rock where high-conductivity flow paths are less ordered. 744
C.R. Moore et al. Groundwater 53, no. 5: 737–747
Discussion and Conclusions This study comprises detailed numerical simulations that explore the movement of water and gas induced by CBM extraction. In these simulations, the coal measure sequence from which extraction takes place is represented in a realistic manner with a level of lithological detail that is rare for groundwater models built to explore drawdown impacts on overlying aquifers. This study has provided insights into near-source processes; these insights may be important for adequate representation of these processes in upscaled regional models wherein the effects of CBM extraction on regional groundwater systems are assessed. NGWA.org
(a)
(d)
(b)
(c)
(e)
(f)
Figure 8. Simulated drawdown after 1 year (a and b), 8 years (c and d), and 20 years (e and f) of extraction. Dual-phase flow is assumed in computing the drawdowns depicted in a, c, and e while only single-phase flow is assumed in computing the drawdowns depicted in b, d, and f.
Table 3 Details of Vertical Averaging Applied to Calculation of Upscaled Hydraulic Properties and System States Parameter
Description of Upscaling
Horizontal hydraulic conductivity
Upscaled horizontal hydraulic conductivity is calculated as the saturated thickness-weighted arithmetic mean of fine-scale hydraulic conductivity, thus preserving transmissivity
Vertical hydraulic conductivity
Upscaled vertical hydraulic conductivity is calculated using thickness-weighted harmonic averaging, thus preserving resistance. Note, however, that vertical hydraulic conductivity is not employed in any pseudo-calculations
tu Kvu
Upscaled porosity is calculated as the thickness-weighted arithmetic mean of fine-scale porosities, this preserving overall void space
θu =
Porosity
Pressure head
Upscaled pressure is calculated by (1) converting fine-scale pressures to heads; (2) undertaking transmissivity-weighted arithmetic averaging of fine-scale heads; (3) converting upscaled heads to pressures
Saturation
Upscaled water saturation is calculated by summing thickness-weighted fine-scale saturations, thus conserving water
Relative permeability
Upscaled relative permeability is water saturation-dependent permeability divided by permeability at full saturation
Equations Khf tf Sf Khu = tu Su =
tf Kvf
θf tf
tu
hf Khf tf Ku tu
hu =
Su = krhu (Su ) =
Sf tf
tu khu (S u ) khu (Su =1)
Notes: In the formulas, the subscripts “f” and “u” indicate “fine-scale” and “upscaled”; K h and K v the horizontal and vertical hydraulic conductivities; k the permeability; t the layer thickness; θ the porosity; S the water saturation; and h the head.
The simulations also demonstrate the spatial and temporal characteristics of drawdown and water desaturation propagation from a pumping center to other parts of a coal measure sequence. Initially, pressure reduction is confined to coal layers as it propagates rapidly along these layers. As this occurs, gas is desorbed from the coal matrix and enters the cleat system of the coal. This desorption NGWA.org
slows the propagation of pressure drawdown compared to that which would be induced without gas desorption with the same amount of water extraction. Two factors contribute to this slowing of drawdown propagation. The first is that water is removed from pore storage in addition to elastic storage, thereby providing a temporary source of water which must be transported to extraction wells. The C.R. Moore et al. Groundwater 53, no. 5: 737–747
745
Figure 9. Eight-layered upscaled model structure.
Figure 10. (a) Scatterplot of relative permeability vs. water saturation where upscaling involves lithological amalgamation. (b) The horizontal scale is expanded to show the true water saturation vs. permeability relationship in comparison.
Figure 11. Scatterplot of relative permeability vs. water saturation where upscaling involves lithological segregation.
second is that the relative permeability of coal in the vicinity of extraction sites is considerably reduced as methane replaces water in coal cleats, and must thereby be transported with it to extractive sites. Meanwhile, over time, pressure reduction propagates vertically from coal layers into the neighboring interburden layers. This leads to a more even vertical distribution of drawdown throughout the whole coal measure sequence. 746
C.R. Moore et al. Groundwater 53, no. 5: 737–747
Insights provided by this study have a number of repercussions for modeling of the impact of CBM extraction on regional groundwater systems. The first is that, despite the localization of water desaturation to the vicinities of extraction sites, there are dangers in ignoring nearwell desaturation in regional modeling. Even under the comparatively benign conditions presented in this article wherein ECLIPSE-calculated water extraction is provided to a single-phase model, drawdown overprediction by the single-phase model can be significant. At early pumping times (1 year), overprediction of drawdown by the singlephase model is as high as 300 m in the vicinity of wells and as high as 150 m at the margin of the extraction wellfield, falling to 1 m at a distance of 7 km from the extraction center. At later times, peak drawdown overprediction is reduced, but prevails over a larger rock volume, as it occurs in both coal and interburden material. Thus, at a simulation time of 20 years, drawdown overprediction is as high as 15 m at a distance of 7 km from the well extraction center, rising to 60 m at the margins of the wellfield, and rising again to 150 m within the extraction wellfield itself. Moore et al. (2013) demonstrate that in intervening years, locations of maximum drawdown overprediction tend to form a halo around an expanding ring of coal water desaturation that propagates outward from the pumping center. Because the neglect of water desaturation can lead to overprediction of drawdown within coal measures from which extraction takes place, this can also lead to drawdown overprediction in underlying and/or overlying aquifers as well, depending on the geological context. The extent of this overprediction will depend, of course, on the hydraulic conductivity of intervening aquitard material (or lack thereof). A second implication for regional modeling, wherein coal measures (and the physical processes associated with CSG extraction) must be represented in a highly upscaled manner, is that coal should be either explicitly or implicitly segregated from interburden in such a model. Coals can be explicitly represented by devoting specific upscaled model layers to the representation of either coal or interburden, but not to both. Alternatively, coals can be implicitly represented through assigning dual-porosity status to upscaled layers from which CBM extraction takes place. In either case, it is likely that the relationship between relative permeability and water saturation will be unaffected, or only mildly affected, by the upscaling process, and so requires little or no upscaling itself. It has yet to be determined whether it is better to represent coal and interburden layers individually or through a dualporosity mechanism. Factors not considered in this short study, such as correct apportionment of well extraction between coal and interburden layers, may need to be taken into account in making this judgment. Moore et al. (2013) demonstrate the potential importance of well-lithology interactions in the overall upscaling process. Issues that are only touched on in this article, but that are deserving of further research in order that a better basis for regional simulation of CBM extraction impacts can be provided, include the following: NGWA.org
•
Parameterization of upscaled representations of coal and interburden where regional models have a large domain and model cells are large. • Use of a single-phase model that allows water desaturation near extraction points as a substitute for dual-phase simulation. • Upscaling of extraction processes in a regional model. These issues are being researched by the authors of this article.
Acknowledgments The authors would like to thank the Queensland Gas Company (QGC) and CSIRO for co-funding the project on which this article is based, and to staff within those organizations who assisted the authors in this larger project, including Dan de Verteuil (QGC), and Tao Cui (CSIRO). The authors would also like to acknowledge the reviewers contribution to this article.
References Barker, J.W., and S. Thibeau. 1997. A critical review of the use of pseudorelative permeabilities for upscaling. SPE Reservoir Engineering 12, no. 2: 138–143. Centre for Water in the Minerals Industry (CWIMI). 2008. Scoping study: Groundwater impacts of coal seam gas development—assessment and monitoring. Report Document Reference P08-010-002. Brisbane, Australia: Department of Infrastructure and Planning. Darman, N.H., G.E. Pickup, and K.S. Sorbie. 2002. A comparison of dual-phase dynamic upscaling methods based on fluid potentials. Computational Geosciences 6, no. 1: 5–27. DOI:10.1023/A:1016572911992. DNRM. 2013. Groundwater risks associated with coal seam gas development in the Surat and southern Bowen basins. In Healthy HeadWaters CSG Water Feasibility Study, Activity 5 Groundwater Risks Associated With Coal Seam Gas: Development in the Surat and Southern Bowen Basins. Prepared by Worley Parsons for the Department of Natural Resources and Mines (DNRM), Queensland, Australia. Esterle, J.S., S.K. Hamilton, V. Ward, S. Tyson, and R. Sliwa. 2013. Scales of geological heterogeneity within the Walloon Subgroup and its coal measures, February. Final Report of Activity 1.3 of the Healthy HeadWaters Coal Seam Gas Water Feasibility Study. Uniquest project no. 00572. Department of Natural Resources and Mines, University of Queensland. Freij-Ayoub, R. 2012. Opportunities and challenges to coal bed methane production in Australia. Journal of Petroleum Science and Engineering 88–89: 1–4. DOI:10.1016/ j.petrol.2012.05.001. Gasda, S.E., and M.A. Celia. 2005. Upscaling relative permeabilities in a structured porous medium. Advances in
NGWA.org
Water Resources 28, no. 5: 493–506. DOI:10.1016/j.advw atres.2004.11.009. Harbaugh, A.W. 2005. MODFLOW-2005. The U.S. Geological Survey Modular Ground-Water Model – Ground-Water Flow Process. Reston, Virginia: U.S. Geological Survey Techniques and Methods 6-A16. Harrison, S.M., J.W. Molson, H.J. Abercrombie, J.F. Barker, D. Rudolph, and R. Aravena. 2000. Hydrogeology of a coalseam gas exploration area, southeastern British Columbia, Canada: Part 1. Groundwater flow systems. Hydrogeology Journal 8, no. 6: 608–622. Hewett, H.A., and R.A. Archer. 1997. Scale-averaged effective flow properties for coarse-grid reservoir simulation. In 14th SPE Reservoir Simulation Symposium, June 8–11. Paper SPE 37988. Dallas, Texas. Kyte, J.R., and D.W. Berry. 1975. Pseudo functions to control numerical dispersion. Society of Petroleum Engineers Journal 15, no. 4: 269–275. Landereau, P., B. Noetinger, and M. Quintard. 2001. Quasisteady two-equation models for diffusive transport in fractured porous media: Large-scale properties for a densely fractured system. Advances in Water Resources 24, no. 8: 863–876. Moore, T.A. 2012. Coalbed methane: A review. International Journal of Coal Geology 101: 36–81. Moore, C.R., J.E. Doherty, S. Howell, L. Erriah, D. de Verteuil, and T. Cui. 2013. Challenges for Upscaling Hydraulic Properties and Processes in the Coal Seam Gas Context. Melbourne, Australia: CSIRO. Myers, T. 2009. Groundwater management and coal bed methane development in the Powder River Basin of Montana. Journal of Hydrology 368, no. 1–4: 178–193. DOI:10.1016/j.jhydrol.2009.02.001. Nordbotten, J.M., J.P. Nogues, and M.A. Celia. 2010. Appropriate choice of average pressure for upscaling relative permeability in dynamic flow conditions. SPE Journal 15, no. 1: 228–237. DOI:10.2118/113558-PA. QGC: A BG Group Business. 2013. Developing robust water management tools. http://www.qgc.com.au/media/233081/ 3.0_-_developing_robust_water_management_tools.pdf (accessed September 10, 2014). QWC. 2012. Underground Water Impact Report for the Surat Cumulative Management Area. Queensland, Australia: Queensland Water Commission http://www.dnrm. qld.gov.au/__data/assets/pdf_file/0016/31327/undergroundwater-impact-report.pdf (accessed September 10, 2014). Ryan, D.J., A. Hall, L. Erriah, and P.B. Wilson. 2012. The Walloon coal seam gas play, Surat Basin, Queensland. APPEA Journal 52: 273–289. Schlumberger. 2011. Manual for ECLIPSE Reservoir Simulation Software. Houston, Texas: Schlumberger. Stone, H.L. 1991. Rigorous black oil pseudo functions. In 11th SPE Reservoir Simulation Symposium, February 17–20. Paper SPE 21207. Anaheim, California. Terzaghi, K., R.B. Peck, and G. Mesri. 1996. Soil Mechanics in Engineering Practice, 3rd ed. New York: WileyInterscience. Warren, J., and P. Root. 1963. The behaviour of naturally fractured reservoirs. SPE Journal 3: 245–255.
C.R. Moore et al. Groundwater 53, no. 5: 737–747
747
