You are on page 1of 34

The impact of fine-scale turbidite channel architecture on deep-water reservoir performance

Faruk O. Alpak, Mark D. Barton, and Stephen J. Naruk

AUTHORS Faruk O. Alpak  Shell International Exploration and Production Inc., 3737 Bellaire Boulevard, Houston, Texas; Omer.Alpak@shell.com Faruk O. Alpak is a senior reservoir engineer in the IUP/ICP Research and Development and Subsurface Modeling Team, Shell International Exploration and Production Inc. He holds a Ph.D. in petroleum engineering from The University of Texas at Austin. Before joining Shell, Alpak worked at the Schlumberger-Doll Research Center as a visiting scientist on mathematical modeling and inversion projects and at The University of Texas at Austin as a research assistant. His specialization areas are reservoir simulation, upscaling, inverse problems, and computational electromagnetics. Mark D. Barton  Shell International Exploration and Production Inc., 3737 Bellaire Boulevard, Houston, Texas; Mark.Barton@shell.com Mark D. Barton is a senior geologist in the Clastic Research Team, Shell International Exploration and Production Inc. His research interests include sedimentology and stratigraphy and their application to the characterization of hydrocarbon reservoirs. Work experience involves a variety of projects ranging from deep-water developments to oil sands mining. He worked at the Bureau of Economic Geology at The University of Texas, Austin before joining Shell in 1998. Stephen J. Naruk  Shell International Exploration and Production Inc., 3737 Bellaire Boulevard, Houston, Texas; Stev.Naruk@shell.com Stephen J. Naruk has more than 20 years of experience with Shell, in multiple exploration, production, and research assignments. He is currently the principal structural geologist of Shell and leads both the Structural Geology Research Team of Shell and the Reservoir Modeling Center of Expertise within the Expertise and Deployment organization of Shell.

ABSTRACT This article concentrates on the question, Which parameters govern recovery factor (RF) behavior in channelized turbidite reservoirs? The objective is to provide guidelines for the static and dynamic modeling of coarse reservoir-scale models by providing a ranking of the investigated geologic and reservoir engineering parameters based on their relative impact on RF. Once high-importance (H) parameters are understood, then one can incorporate them into static and dynamic models by placing them explicitly into the geologic model. Alternatively, one can choose to represent their effects using effective properties (e.g., pseudorelative permeabilities). More than 1700 flow simulations were performed on geologically realistic threedimensional sector models at outcrop-scale resolution. Waterflooding, gas injection, and depletion scenarios were simulated for each geologic realization. Geologic and reservoir engineering parameters are grouped based on their impact on RF into H, intermediate-importance (M), and low-importance (L) categories. The results show that, in turbidite channel reservoirs, dynamic performance is governed by architectural parameters such as channel width, net-to-gross, and degree of amalgamation, and parameters that describe the distribution of shale drapes, particularly along the base of channel elements. The conclusions of our study are restricted to light oils and relatively highpermeability channelized turbidite reservoirs. The knowledge developed in our extensive simulation study enables the development of a geologically consistent and efficient dynamic modeling approach. We briefly describe a methodology for

ACKNOWLEDGEMENTS
Copyright 2013. The American Association of Petroleum Geologists. All rights reserved. Manuscript received June 15, 2011; provisional acceptance August 16, 2011; revised manuscript received January 22, 2012; final acceptance April 2, 2012. DOI:10.1306/04021211067

We thank Marc Alberts, Ru Smith, and Mark Hempton for supporting this work and Shell International Exploration and Production Inc. for

AAPG Bulletin, v. 97, no. 2 (February 2013), pp. 251 284

251

permission to publish this paper. Ciaran OByrne, Bradford Prather, and Carlos Pirmez were instrumental in the collection of data presented in the study as well as in the development of ideas presented here. The authors also thank David G. Morse and two anonymous reviewers for helping us enhance the quality of the manuscript. The AAPG Editor thanks the two anonymous reviewers and David G. Morse for their work on this paper. EDITORS NOTE Color versions of Figures 1618 may be seen in the online version of this paper.

generating effective properties at multiple geologic scales, incorporating the effect of channel architecture and reservoir connectivity into fast simulation models.

INTRODUCTION In recent years, operators have made many discoveries in the upper Tertiary deep-water provinces of the Gulf of Mexico, Niger Delta slope, and northwest Borneo slope. Discovery of significant oil and gas in deep-water depositional settings began an era of research into turbidite reservoirs to develop ways of setting reservoir performance expectations realistically (Chapin et al., 2002; Berg and Kjarnes, 2003). Early well performance from these discoveries has, in many cases, met and, in some cases, surpassed expectations established at the time of project sanction (Kendrick, 2000; Hampton et al., 2006). Operators in west Africa reported good connectivity based on interference testing and early production from channelized turbidite reservoirs that fill large submarine valleys (Humphreys et al., 1997; Bouchet et al., 2004). Many of these highperforming fields and reservoirs, however, have not reached a level of maturity that permits accurate estimation of their ultimate recovery factor (RF). In other cases, operators have encountered severely impaired reservoirs among these recent developments. The shortfalls are attributed in large part to reservoir compartmentalization. Operators of channelized fields, such as the Schiehallion field, west of Shetlands; the Ram Powell field in the Gulf of Mexico; and the Bittern field in the North Sea, suspect that channel architecture and the presence of shale drapes have a significant influence on hydrocarbon recovery (Govan et al., 2006; Barton et al., 2010). Individual channel stories and their shale architecture introduce significant uncertainty into static reservoir models because they are too small to resolve on conventional three-dimensional (3-D) seismic data used in field developments. Thus, dynamic effects of the subsurface risks and uncertainties related to these features must be understood in advance (both in terms of upside and downside risks) so that plans can be made to acquire data during appraisal and development sufficient to reduce their uncertainty. A vast literature exists on channelized turbidite reservoirs and flow-simulation studies conducted for such reservoirs. Providing a comprehensive list of such studies is beyond the scope of our article. Having stated that, we describe below a

252

Deep-Water Reservoir Performance

few references relevant to our study. Larue and Yue (2003) conducted a comparative reservoir database study concentrating on deep-water reservoirs to evaluate the influence of stratigraphy on oil recovery. Larue (2004) used outcrop models to quantify the uncertainty in the prediction of volumes and recovery in deep-water channelized slope reservoirs. Larue and Friedmann (2001, 2005) and Larue and Hovadik (2006) assessed, through novel staticconnectivity evaluation methods and dynamic waterflood simulations, the stratigraphic connectivity characteristics of generic channelized sector models constructed via a combination of Boolean, variogrambased geostatistics, and multipoint-statistics approaches. De Jager et al. (2009) evaluated the relationship between geologic parameters and the flow behavior of channelized reservoirs with the aid of an experimental design approach. Using the example of 3-D shale drapes attached to channel bodies in a deep-water depositional setting, Stright (2006) developed a methodology in which shale drapes are accurately upscaled and history-matched to production data while honoring the geologic concept that describes the drape geometry. Li and Caers (2011) focused specifically on the geostatistical modeling of shale drapes and perturbed the location of channels and erosional holes in the shale drapes within the context of a geologically based historymatching workflow. Recent work also focused on the definition, derivation of controlling factors, validation, and application of flow-based effective properties (connectivity factors [CFs]) to the dynamic modeling of channelized turbidite reservoirs (Alpak et al., 2010, 2011). The sensitivity structure of the CFs were revealed via flow simulations conducted on sector models at multiple geologic resolutions using fine-scale parent and stepwise-coarsened offspring models in Alpak et al. (2010). The main focus of the work was derivation of CFs, which were then converted to pseudorelative permeability (flowbased effective property) functions to mimic the dynamic effects of fine-scale channelized turbidite architecture in relatively coarser fast-running reservoir-simulation models. Subsequently, Alpak et al. (2011) presented a novel method, which uses flow-based effective properties to retain geologic realism in history-matching workflows. An applica-

tion of the method was conducted on a data set from a west African channelized turbidite reservoir. The complex multiscale shale architecture of channelized turbidite deposits requires models that resolve the full architectural detail. Only then can the accurate flow and transport behavior of the stratigraphic architecture be computed via flow simulations. This is why the sensitivity studies conducted for flow-based effective properties in Alpak et al. (2010) involved fine-scale sector model simulations. Because the focus of the work was on the development of effective properties, the sensitivitystudy results were reported briefly and only for the CFs as the dynamic modeling outcome. The sensitivity structure of the CF answers the question: Which stratigraphic parameters govern the necessary correction for a coarse-scale dynamic model such that it will approximate the dynamic behavior of a fine-scale model? The answer inevitably depends on the stratigraphic resolution of the coarsescale model. This question, however, differs from Which parameters govern RF behavior? which is posed only for fine-scale models. In this article, we analyze in detail the fine-scale sector model simulations and conduct a study to prioritize the parameters that affect the RF behavior. We predominantly leave aside the derivation of scaleappropriate flowbased effective properties covered in our previous publications. Only a brief summary of our effective property approach has been presented in this work because it is an important factor in the way we validate our simulations against field observations of recovery performance. The objective is to provide guidelines for the static and dynamic modeling of coarse reservoir-scale models by providing a ranking of the investigated geologic and engineering parameters based on their relative impact on RF. Hereafter, we will refer to the geologic and reservoir engineering parameters together as (uncertainty) matrix parameters for brevity. Simulations are conducted on detailed 3-D sector models constructed at the outcrop-scale resolution by representing accurately the full-detail channel architecture up to decimeter scale. Waterflooding, gas-injection, and gasdepletion scenarios are simulated for each geologic realization. For simplicity, we will hereafter refer to the gas-depletion recovery mechanism as the
Alpak et al. 253

Figure 1. Schematic diagram illustrating common scales of channelization observed from outcrop and nearsea-floor analog data sets of turbidite channel systems. (A) Channel story consists of a relatively conformable set of beds or bed sets bounded at the base by an erosion surface. Based on bed stacking patterns, two types of channel story elements are recognized: aggradational and laterally accreting. (B) A channel story set consists of a set of channel story elements bounded at the base by an erosion surface. (C) A channel complex is defined as a set of channel story sets bounded at the base by an erosion surface. A distinct facies association, referred to as the bypass facies association, commonly overlies the erosive channel base. It can occur at all levels of channelization but is best developed at the scale of the channel story set and channel complex. 1 m (3.3 ft).

depletion recovery mechanism whenever appropriate. The effect of structural discontinuities, such as faults and fractures, remains beyond the scope of these studies.

TURBIDITE CHANNEL ARCHITECTURE An overview of turbidite channel architecture is provided. It represents the view of the authors based on the analysis of numerous outcrop and nearsea254 Deep-Water Reservoir Performance

floor analogs acquired from a range of depositional settings and locations. Our main observations are that the (1) channel forms occur, with varying character (tributary, trunk, and distributary forms), along profiles from submarine slopes to basin floors; (2) turbidite channel architecture is commonly multiscale, displaying a hierarchical organization of smaller channel elements confined within the boundaries of larger channel elements; and (3) mudstone elements, which potentially represent low-permeability baffles or barriers to flow, are commonly distributed along the base and margin of the channel elements.

The multiscale stratigraphic architecture of turbidite channel systems is schematically illustrated in Figure 1. The smallest scale channel-form element is referred to as the channel story. It is composed of a relatively conformable set of beds or bedsets bounded at the base by an erosional surface. A channel story is the stratigraphic product of a geomorphic channel and scaled to the size of the sediment gravity flows that passed through it. The term is comparable to that used by Friend et al. (1979), Bridge and Diemer (1983), and Willis (1993) to describe the architecture of ancient fluvial sandstone bodies. Turbidite channel stories are typically 3 to 20 m (9.865.6 ft) in thickness and 50 to 400 m (1641312.3 ft) in width. Larger scale channel elements include channel story sets (informally referred to as meander belts) and channel complexes. A channel story set consists of a set of genetically related channel stories and associated inner-levee deposits bounded at the base by an erosion surface. Their dimensions typically range from 15 to 60 m (49.2196.9 ft) in thickness and 300 to 1000 m (984.33280.8 ft) in width. At an even larger scale, a channel complex consists of a group of channel story sets and related inner-levee deposits bounded by an erosion surface. Although differences in terminology exist, the hierarchical classification scheme is in essence similar to that proposed by some previous authors (Sprague et al., 2002; Mayall and OByrne, 2002; Mayall et al., 2006). By comparison, the usage of the term channel story in this study is similar to the usage of channel element in the scheme proposed by Sprague et al. (2002). In a similar fashion, the term channel story set is equivalent to channel complex, and channel complex is equivalent to complex set. Channel story sets and complexes represent erosional features that are significantly larger in scale than the flows responsible for their formation. They are analogous to a valley in that their development is the cumulative product of erosion by a smaller channel (Sylvester et al., 2011). The different scales of channelization (channel stories, story sets, and complexes) are interpreted to record repeated adjustments (relative falls and rises) in the slope equilibrium profile of the channel system through time. Downslope avulsions and subsequent

knickpoint erosion and retreat represent a mechanism that could cause repeated adjustments in the equilibrium profile at multiple scales. Other processes that could cause fluctuations in the slope equilibrium profile include tectonics associated with a mobile substrate and high variability in sediment discharge and flow capacity (Pirmez et al., 2000). Channel story fills display distinctive bed and/or bed-set stacking patterns that range from vertically aggrading to laterally accreting (Figure 1A). Beds and/or bed sets within vertically aggrading fills display convergent to onlapping relationships with adjacent channel margins. Convergent margins are composed of thin-bedded successions of sandstone and mudstone, whereas onlap margins are composed of thick-bedded successions of amalgamated sandstone. Outcrop observations suggest that convergent margins tend to occur on the inside bend of the channel, whereas onlap margins are common on the outside channel bend. In vertically aggrading fills, channel erosion (channel deepening), sediment bypass (channel widening), and channel filling occur in separate phases. Turbidite channel fills characterized by laterally accreting beds have also been observed from outcrop exposures (Lien et al., 2003; Arnott, 2007) and in high-resolution seismic data from nearsea-floor analogs (Abreu et al., 2003; Kolla et al., 2007). This type of channel fill is interpreted to record lateral channel migration related to the erosion of sediment from the outside bend of the channel followed by redeposition of sediment on the inside bend during the same flow event. The product of this process is lens-shaped point-bar deposits similar to those observed in fluvial systems. A distinct facies association, referred to as the bypass facies association, commonly separates the channel fill (represented by the aggradational stack of beds and/or bed sets) from the basal erosion surface (Beaubouef et al., 2000; Gardner and Borer, 2000). The bypass facies association is typically an extremely heterolithic facies association (composed of uncommonly coarse- and fine-grained sediments) characterized by features suggestive of significant sediment reworking and erosion (Barton et al., 2007b; Barton et al., 2010). The mixture of coarse and fine materials is interpreted to represent selective deposition from the head and tail of
Alpak et al. 255

Figure 2. (A) Photograph panel of the Popo channel outcrop looking east. The exposure is 600-m (1970-ft) long and 80-m (262-ft) thick. The north-south orientation of the outcrop is approximately perpendicular to the primary paleoflow direction, which is to the east and southeast. (B) Bedding diagram of the Popo channel outcrop. The outcrop can be subdivided into a series of channel story sets (color coded on diagram) that are 25-m (82-ft) thick and 350-m (1150-ft) wide. Internally, the channel story sets are composed of three to five channel story elements. Channel story sets 1 to 4 appear to infill a larger erosional container that displays more than 50 m (148 ft) of relief (figure from Barton et al., 2007a; used with permission from AAPG). The episodic nature of the downcutting phase is recorded as a series of small erosional terraces and associated remnant channel margins along the left margin of channel story set 4 (green fill). Siltstone drapes located along the base of the channel story sets are highlighted in red.

numerous sediment gravity flows that mostly passed through the channel during downslope transport. An outcrop example exhibiting multiple scales of erosion and filling are illustrated from the Popo channel outcrop, west Texas (Figure 2). Other outcrop examples include the Beacon channel complex, Brushy Canyon Formation, west Texas (Pyles et al., 2010); Buena Vista channel complex, Brushy Canyon Formation, west Texas (Rossen and Beaubouef, 2007); San Clemente channel complex, Capistrano Formation, California (Chapin and Keller, 2007); channel complexes in the Pab Range, Pakistan (Eschard et al., 2003); and the Condor channel complex, Cerro Toro Formation, Chile (Barton et al., 2007b). Mudstone and heterolithic strata commonly overlie the base of the channels. These elements are informally referred to as channel-base drapes and represent a configuration of strata that overlie the channel-base erosion surface and that may act as a potential baffle or barrier to flow. They are common at all scales of channelization and may be
256 Deep-Water Reservoir Performance

formed by several different processes (Barton et al., 2010). (1) Abandonment drapes consist of hemipelagic mudstones and fine-grained sediments derived from suspension fallout or dilute turbidity flows that have spilled into the channel from another source. They could potentially form if the channel became temporarily abandoned and subsequently reoccupied (Figure 3A). (2) Convergentmargin drapes develop as the channel progressively fills. Coarse-grained material is deposited in the axis of the channel whereas fine-grained material is deposited along the margins, converging laterally to form a margin drape. The deposit is the direct result of density-flow height and channel-margin relief. Laterally, the convergent-margin drapes interfinger with sandstone beds infilling the axis of the channel (Figure 3B). (3) Bypass drapes are muddy and/or heterolithic deposits left behind by turbidity flows that mostly pass through the channel and continue downslope (Figure 3C). They differ from convergent-margin drapes in that they are deposited before the main phase of channel filling and

Figure 3. Schematic illustrating the distinct types of channel-base drapes that have been observed from outcrop exposures of turbidite channel deposits (Barton et al., 2010). (A) Abandonment drapes (gray) form when the channel becomes abandoned after initial incision and is subsequently reoccupied. They consist of hemipelagic mudstones and finegrained sediments derived from suspension fallout or dilute turbidity flows that have spilled into the channel from another source. (B) Convergent-margin drapes (blue) consist of deposits that preferentially accumulate on the margin of the channel as the channel progressively fills. (C) Bypass drapes (blue) are deposited from the muddy tails of turbidity flows that pass through the channel before the main phase of channel infilling. They are commonly interstratified with facies suggestive of erosion and sediment reworking such as coarsegrained sandstones and shale-clast conglomerates.

are found in close association with facies suggestive of significant erosion and sediment reworking such as coarse-grained lag deposits, cross-stratified sandstones, and shale-clast conglomerates. An outcrop photograph of an interpreted convergent-margin drape lining the base of a channel story element is shown in Figure 4A. Convergentmargin drapes are generally less than a meter in thickness and composed of a bedded mudstone that thins toward the axis of the channel. A bypass drape associated with a channel story set is shown in Figure 4B. Bypass drapes tend to be relatively thick, are composed of many different facies and events, and commonly thicken into the axis of the

channel. Bypass and convergent-margin drapes are collectively referred to as shale drapes for simplicity when dynamic modeling is discussed in the text. Field-based observations from numerous turbidite channel exposures indicate that abandonment drapes are uncommon, whereas convergentmargin and bypass drapes are commonly observed and may occur in the same channel element. A model showing how the two end-member drape types are potentially linked along a slope equilibrium profile through the channel system is shown in Figure 5. The slope equilibrium profile is a proxy for the downslope transfer of sediment by sediment
Alpak et al. 257

Figure 4. (A) Outcrop example of a channel story element bounded at the base by a thin succession of bedded mudstones. Some of the mudstone beds interfinger with sandstone beds infilling the channel and are interpreted as a type of convergent-margin drape. The channel element is approximately 8 m (26 ft) in thickness. Note that sandstones are light colored and mudstones are dark colored in the photograph. The photograph is from the Tres Pasos Formation, Chile (uninterpreted photograph courtesy of Stanford Project on Deep Sea Depositional Systems Consortium, Stanford University). (B) Outcrop photograph showing a relatively thick succession of mudstones (see bracket) lining the base of a channel story set that is approximately 25 m (82 ft) in thickness. The basal erosion surface is highlighted by the red dashed line. The succession thickens toward the axis of the channel where the mudstones interfinger with lenticular sandstone beds (highlighted by green dashed lines). The mudstone-dominated succession is interpreted to have formed during a prolonged phase of sediment bypass through the channel. The upper half part of the channel fill is composed mostly of sandstone. The photograph is from the Isaac Formation, British Columbia, Canada.

gravity flows (Figure 5A). Erosional processes dominate updip parts of the profile. As the slope decreases in a downstream direction, depositional processes take precedence (Pirmez et al., 2000). The type of drape (e.g., bypass, mixed, or convergent margin) that develops depends on the degree that each of these two processes are operative during the formation of the channel (Figure 5B). Erosion and sediment reworking in updip parts of the profile favor the development of bypass drapes, whereas preservation of deposits associated with highly depositional flows in downdip parts of the
258 Deep-Water Reservoir Performance

profile favor the development of convergent-margin drapes. A schematic diagram illustrating the facies architecture and sequential development of a channelbase drape associated with a story set is illustrated in Figure 6. A fall in the equilibrium profile of the channel results in the flows eroding and downcutting into underlying strata (Figure 6A). Elements associated with the downcutting phase include overbank or levee deposits, remnant channel margins isolated by bedded mudstones, and debrisflow and slump deposits within the axis of the

Figure 5. Diagram illustrating the changes in channel architecture and drape morphology along an equilibrium profile longitudinal to the channel system. (A) Diagram showing the distribution of sediments along a longitudinal channel profile after deposition from a sediment gravity flow or series of sediment gravity flows (stratigraphic product is a bed or a bed set). In a downstream direction, processes associated with erosion and sediment reworking are gradually replaced by processes associated with deposition and sediment preservation. Fine-grained sediment (mud) settles out slowly after the turbidity flow has passed, potentially forming a thin mudstone drape across the length of the profile (modified from Mutti et al., 2003). (B) Schematic illustrating resulting channel architectures following a series of aggrading and backstepping events. Channels in updip parts of the profile potentially develop mudstone drapes along the base of the channel, which are related to sediment bypass, whereas channels in equivalent downdip parts of the profile develop convergent-margin drapes as the channel fills. Both styles of drapes (mixed) are seen in intermediate positions of the profile.

channel. Terraced erosional cuts are common and indicate that the downcutting process was episodic. As the channel achieves a stable equilibrium profile, downcutting ceases and channel widening ensues (Figure 6B). This phase may be characterized by a prolonged period of sediment bypass as the channel acts as a conduit for the downdip transport of sediment. Resultant deposit consists of a complex assemblage of bedded mudstones; lenticular, cross-stratified, coarse-grained sandstones; slumps; slide blocks; debris flows; and imbricated shale-clast conglomerates. A rise in the equilibrium profile results in infilling of the channel with a series of channel stories (Figure 6C). Convergent-

margin drapes develop along the margin of the channel. It progressively fills with channel story elements. Channel-to-channel connectivity at the story and story-set levels can occur where (1) drapes have not been deposited, (2) previously formed drapes have been eroded, or (3) sandstone beds and lag deposits amalgamate to form a permeable body at the base of the channel. Architectural elements associated with the erosion, bypass, and filling phases are illustrated from an exposure of a channel story set at Fishermens Point, Ross Formation, Ireland (Figure 7). Photographs illustrating some of the key characteristics of convergent-margin and bypass drapes are shown
Alpak et al. 259

in Figures 8 and 9. Convergent-margin drapes are characterized by an angular basal discordance overlain by a mudstone that is interstratified with an

upward bed-thickening succession of channelmargin sandstones (Figure 8A, B). Bypass indicators such as shale-clast conglomerates are commonly

260

Deep-Water Reservoir Performance

Figure 7. (A) Photograph panel of Fishermens Point, Ross Formation, Ireland. The exposure is 100-m (328-ft) long and 20-m (65-ft) thick. (B) Bedding diagram of the exposure illustrating architectural elements commonly observed at the scale of a channel story set, including (1) a remnant channel margin (green fill) formed during the downcutting phase; (2) a heterolithic assemblage (purple fill) of lenticular sandstones, mud-clast breccias, and mudstone drapes developed during a phase of sediment bypass; and (3) a series of channel story elements and convergent-margin drape (blue fill) deposited as the channel filled.

Figure 6. Schematic diagram illustrating the evolution and facies architecture of a channel story set. Deposits associated with the active phase are colored whereas deposits from previous phases are shaded gray. Mudstone is indicated by a blue fill, and sandstone, by shades of green, yellow, and orange. Other elements are identified on the figure. The architecture of the channel story set can be subdivided into three phases that correspond to erosion, sediment bypass, and channel filling. (A) A fall in the equilibrium profile of the channel initiates a phase of erosion and knickpoint retreat. Downdip of the knickpoint, slumps, slide blocks, and debris-flow deposits are preserved in the axis of the channel. If the downcutting phase is episodic, thin-bedded inner-levee and remnant channel margins may be preserved on the flank of the channel. (B) As a graded equilibrium profile is achieved, the channel acts as a conduit for the downslope transport of sediment and undergoes a potentially prolonged phase of channel widening and sediment bypass. Deposits include crudely bedded coarse-grained sandstones and mud-clast conglomerates; lenses of cross-stratified sandstone; thin-bedded sandstones and mudstones with wavy-, flaser-, and lenticular-bedding types; and bedded mudstones and siltstones. (C) A rising equilibrium profile results in channel filling by a series of aggrading channel story elements and genetically related overbank or inner-levee deposits. Convergentmargin drapes are commonly developed on the margin of the channel story elements. Note that the diagram represents a composite of features observed from several outcrops. For illustration purposes, not all features are represented to scale. The approximate dimensions of the channel story set are 20 to 25 m (6582 ft) in thickness and 300 to 500 m (9841640 ft) in width. Alpak et al. 261

262

Deep-Water Reservoir Performance

not present. In contrast, drapes formed during the bypass phase display complex architectures and are associated with deposits that display features suggestive of erosion and sediment reworking. Facies that are indicative of sediment bypass include isolated beds of cross-stratified sandstone (Figure 8C, D); starved ripples and flaser-laminated sandstones (Figure 8E, F); imbricate mud-clast conglomerates (Figure 9A, B); and multiple, closely spaced erosion surfaces (Figure 9C, D). In general, the bypass facies indicators are most likely to be observed within the axis of the channel. However, deepening of the channel through time can result in nested lag deposits high on the channel margins. Indicators of slope readjustment and lateral bank collapse within the channel include small-scale (0.1 1.0-m [0.33-ft] thick) slumps and rotated slide blocks. Thin debris-flow and slurry-bed deposits are also common components of the drape element. In contrast to the complexity observed within the bypass facies association, overlying beds that infill the channel generally consist of thick repetitive successions of massive-to-graded sandstone beds. In subsurface settings, heterolithic bypass deposits are commonly overlooked as potential baffles or barriers affecting channel-to-channel connectivity for several reasons. One reason is that they are commonly misidentified as channel-margin or channel-overbank deposits and therefore are not evaluated in the correct stratigraphic context. Criteria to differentiate thin-bedded bypass deposits from thin-bedded overbank deposits using borehole data are discussed in Barton et al. (2010). Thin-bedded overbank deposits generally lack bypass indicators

and are commonly dominated by beds that display waning-flow sequences such as well-developed Bouma sequences, diffuse graded bedding, or climbing ripples. A second reason is that the bypass deposits are commonly composed of a relatively high proportion of sandstone, as illustrated in the photographs from Figures 8 and 9, and therefore are not viewed as a potential baffle or barrier to flow. Cases where the sandstone beds and lag deposits amalgamate to form a permeable body at the base of the channel definitely exist. However, our outcrop field work on the internal architecture of such deposits indicates that the sandstone beds are commonly extremely lenticular and bounded by or interstratified with mudstone beds that are relatively continuous. As a result, the unit as a whole essentially behaves as a barrier or baffle to flow. A third reason is the assumption that mudstone drapes are only present on the margin of the channel element and not along the base. This is commonly the case with convergent-margin drapes, but outcrop studies on the distribution of bypass drapes indicate that they commonly extend across the base of the channel. To assess the impact that channel-base drapes could have on channel-to-channel connectivity, the distribution of channel-margin and bypass-related drapes were mapped in detail from several channelized turbidite outcrops. The percentage of channelbase erosion surfaces overlain by convergent-margin and/or muddy-to-heterolithic bypass deposits is proportional to the total length of convergent-margin and bypass drapes divided by the total length of channel-base erosion surfaces (Figure 10A). When

Figure 8. (A) Convergent-margin drape composed of bedded siltstones overlying an erosive channel story base, Brushy Canyon Formation, west Texas. Laterally, toward the axis of the channel, the siltstones interfinger with channel-margin sandstones. (B) Bedding diagram of the previous photograph. (C) Interpreted bypass facies association overlying erosive channel story set base, Isaac Formation, British Columbia, Canada. Lenticular beds of cross-bedded sandstone overlying an erosive channel story set base. Beds are capped by thin but laterally extensive mudstone drapes. A thin succession of flaser- to lenticular-bedded sandstones and mudstones overlies the cross-bedded sandstones. The green dashed line near the top of the photograph marks the base of a succession of thick-bedded sandstones that compose the channel fill. (D) Closeup photograph of cross-bedded sandstone and mudstone drapes. See the dashed blue outline on Figure 8C for the location. (E) Closeup photograph of flaser-bedded sandstones and mudstones from the previous photograph (Figure 8C). The width of the photograph is 50 cm (20 in.). (F) Closeup photograph of lenticular- and wavy-bedded sandstones and mudstones from the previous photograph (Figure 8C). Note the lack of waning-flow sequences. The width of the photograph is 45 cm (18 in.). (G) Interpreted bypass facies association overlying erosive channel story set base, Skoorsteenberg Formation, Tanqua Karoo, South Africa. (H) Bedding diagram of the previous photograph. The element consists of lenticular coarse-grained sandstones, cross-bedded sandstones, and mud-clast conglomerates interstratified with bedded mudstones. Alpak et al. 263

Figure 9. (A) Interpreted bypass facies association overlying erosive channel story set base, Isaac Formation, British Columbia, Canada. (B) Bedding diagram of the previous photograph. The element consists of bedded mudstones interstratified with lenses of sandstone and mudclast breccias. Although the sandstones and mud-clast breccias have erosive bases, they commonly do not fully erode underlying bedded mudstones. (C) Interpreted bypass facies association overlying erosive channel story set base, Brushy Canyon Formation, west Texas. (D) Bedding diagram of the previous photograph. The element consists of truncated successions of thin-bedded sandstone and is bounded by thin siltstone beds. The thin-bedded sandstones are discontinuous and do not appear to provide connectivity. 264 Deep-Water Reservoir Performance

Figure 10. The distribution of channel-margin and bypass drapes were mapped in detail from several channelized turbidite outcrops. Mapped outcrops are from a variety of locations and depositional settings including the Brushy Canyon Formation, west Texas; the Bell Canyon Formation, west Texas; the Jackfork Group, Arkansas; the Capistrano Formation, California; the Isaac Formation, British Columbia; the Ross Formation, Ireland; the Ainsa Formation, Spain; the Loma de Los Baos Formation, Spain; and the Skoorsteenberg Formation, South Africa. The channel-base drape coverage is calculated as the percentage of channel-base erosion surfaces overlain by a mudstone element (bedded mudstone, muddy debris-flow deposit). It is proportional to the total length of mudstone elements (L) divided by the total length of channel-base erosion surfaces (W). (A) An example showing how the average drape coverage was calculated for a hypothetical outcrop composed of three channel elements. The elements are named channel 1 (Ch-1) to channel 3 (Ch-3) in descending stratigraphic order. The distribution of mudstone drapes are indicated by blue fill. Drapes are absent on the margins of Ch-1 and in the axis of channel 2 (Ch-2), and continuous across Ch-3. (B) Plot of the average channel-base drape coverage for each outcrop studied. The measurements are organized by channel type (story and story set) as well as inferred depositional setting (toe of slope or basin floor, lower slope, mid- to upper slope). CS = channel story; CSS = channel story set. Alpak et al. 265

Figure 11. Diagram illustrating stratigraphic elements within the sector models. A surface-based modeling program (Wen et al., 1998; Wen, 2005) was used to construct the sector models. Models are hierarchical in scale and are composed of channel story sets and channel stories. (A) Channel complex composed of multiple channel story sets. In the figure, channel story sets are colored red, orange, and yellow in ascending stratigraphic order. The channel-base drape for the last channel story set is shown in purple. The regions where the drapes are absent are indicated by white areas. No bias of drapes toward axis or margin exists. (B) Channel story sets composed of multiple channel stories. In the figure, channel story sets are colored red, orange, yellow, and green in ascending stratigraphic order. The channel-base drape for the last channel story is shown in blue; the gaps in the drape are indicated by white areas. A weak bias in the model exists for drapes to be placed toward the margins of the channel story.

estimating channel-base drape coverage, note that two-dimensional measurements from outcrop exposures are not biased in a positive or negative manner from the coverage that exists in three dimensions. The same can be said for a series of one-dimensional measurements obtained from borehole data. The criteria to recognize and quantify the distribution of channel-base drapes using core and image logs are discussed in Barton et al. (2010). A plot of the average channel-base drape coverage for each outcrop is shown in Figure 10B. Measurements are organized by channel type (story and story set) as well as inferred depositional setting (toe of slope or basin floor, lower slope, mid to upper slope). Inferences of slope setting were based on the type of deposits associated with the channels: toe of slope channels were flanked by lobe deposits (e.g., Willow Mountain outcrop, Bell Canyon Formation), lower slope channels were flanked by low-relief channel-overbank deposits (e.g., Popo channel complex, middle Brushy Canyon Formation), and mid- to upper slope channels were flanked
266 Deep-Water Reservoir Performance

by slope mudstones or high-relief levees (Condor channel complex, Cerro Toro Formation). Channel-base drape coverage is extremely variable and commonly exceeds 60%, with a range extending from less than 5% to more than 90%. As expected, channel-base drape coverage appears to increase in depositional settings and stratigraphic situations where sediment bypass is more prevalent. Slope settings display much higher drape coverage than basin-floor settings (Barton et al., 2010). Channel story sets display drape coverages that are, on average, 20% higher than channel stories. Preservation may also be important in drape coverage but it is more difficult to predict. Several slope settings, such as the Windermere (British Columbia, Canada) and Cerro Toro (southern Chile) outcrops, exhibit significantly high drape coverage at the channel story set level but low drape coverage at the channel story level. Channel story drapes appear to have been formed but were subsequently eroded and replaced by thick shale-clast breccias (Barton et al., 2010).

SECTOR MODEL High-resolution sector models are constructed using a surface modeling program (Wen et al., 1998; Wen, 2005). Layers with a surface model are generated by migrating a surface through a series of time steps. The migration of the surface mimics processes associated with erosion and deposition. Surface models have some advantageous attributes. They can construct models at the bed scale, create stratigraphically realistic layering (layers or elements are organized from oldest to youngest), and represent hierarchical relationships. Because many types of shales or mudstones occur at the interface between beds or stratigraphic elements, surface modeling allows for a more realistic representation of stratigraphic heterogeneity. An example of a sector model used in this study is shown in Figure 11. The largest scale elements are channel story sets (Figure 11A). Elements that may impact connectivity at this scale (i.e., between channel story sets) are mudstone drapes deposited along the base and margins of the channel story set and the degree of amalgamation between individual channel story sets. Internal to the channel story sets are channel stories (Figure 11B). Elements that may impact connectivity at this scale are channel-base drapes, mudstone-filled channels, and the degree of amalgamation between individual stories. Within the channel story elements, the frequency and lateral extent of intrachannel shales were varied.

DEFINITIONS, RANGES, AND SIMULATION LEVELS FOR MATRIX PARAMETERS Reservoir engineering parameters resemble offshore west African data sets. In terms of fluids, the focus is on relatively light oils and gas. Three fundamentally different recovery mechanisms are considered for sector model simulations: (1) waterflooding in the absence of movable gas, (2) immiscible gas injection in the absence of mobile water, and (3) primary depletion of gas. Aquifer and gas cap energy are assumed to be negligible within the sector model. Hereafter, we will refer to the immiscible gas-injection recovery

mechanism simply as the gas-injection recovery mechanism. The primary depletion scenario for a large spectrum of liquid hydrocarbons involves the decrease of the reservoir pressure below the bubble point in the absence of external pressure support. The resulting evolution of gas from liquid hydrocarbons can potentially cause significant complications for analyzing the sensitivity structure. Of course, one can claim that, before reservoir pressure decreases below bubble-point pressure, in general, a pressuremaintenance activity such as water flooding or gas injection is initiated. However, our goal here is not to derive RFs for a particular recovery optimization scenario. Instead, our objective is to quantify the hydrocarbon trapping in the reservoir caused by the existence of fine-scale geologic features. Thus, we choose to investigate the primary depletion scenario by assuming a dry-gas reservoir and focus on the RF behavior (as an indicator of hydrocarbon trapping and reservoir connectivity), with the purpose of understanding the first-order physics. A comprehensive list of matrix parameters subject to investigation is shown in Tables 1 and 2. The matrix parameters are abbreviated for practicality in documentation and data management. Hereafter, in the text, we will adhere to the naming convention of Tables 1 and 2. We also report whether a given parameter is discrete or continuous by definition (characteristic 1) as well as whether a given parameter is qualitative or quantitative by nature (characteristic 2) in Tables 1 and 2. The units of the matrix parameters are given in the last column of the table. Tables 3 and 4 document investigated levels for the matrix parameters. These parameters are sensitized separately for each recovery mechanism. Base-case parameter settings are displayed in the V0 column. Levels of variation are listed under the following column titles: V3, V2, V1, V1, V2, V3, , and so on. Upper and lower bounds of variation define the investigated physical range for each parameter. We emphasize that some of the reservoir engineering parameters represent combined outcomes of several subparameters. The parameter description column in Table 4 elaborates on such relationships. Additionally, important characteristic features of sector models that
Alpak et al. 267

Table 1. Description of the Geologic Parameters Parameter Number 1 2 3 4 5 6 7 8 9 10 11 12 Matrix Geologic Parameter BNO BNG BNS BSC BRS BSG BSH BDA CDT CWD CSN CCN Parameter Description Number of meander belts Meander-belt channel-to-nonchannel ratio (meander-belt net-to-gross) Meander-belt channel-to-nonchannel ratio shift parameter Meander-belt shale drape coverage Random sequence number for the meander-belt shale drape realization Meander-belt shale drape geometry Meander-belt shale drape hole size Meander-belt degree of amalgamation Channel depth Channel widthtodepth ratio Channel sinuosity Channel-continuity number (frequency of mud plugs internal to the channel) Channel shale drape coverage Random sequence number for the channel shale drape realization Channel shale drape geometry Channel shale drape hole size Channel degree of amalgamation Channel-infill architecture Channel-infill shale frequency (thin-bedded channel margins) Channel-infill shale drape coverage Random sequence number for the channel-infill shale drape realization Channel-infill shale drape geometry Channel-infill shale drape hole size Characteristic 1 Discrete Continuous Continuous Continuous Discrete Discrete Continuous Discrete Continuous Continuous Continuous Discrete Characteristic 2 Quantitative Quantitative Quantitative Quantitative Quantitative Qualitative Quantitative Qualitative Quantitative Quantitative Quantitative Quantitative Unit Dimensionless Dimensionless Dimensionless Percent (%) Dimensionless Dimensionless Element width Dimensionless Meter (m) Dimensionless Dimensionless Number of plugs per kilometer Percent (%) Dimensionless Dimensionless Element width Dimensionless Dimensionless Dimensionless Percent (%) Dimensionless Dimensionless Element width

13 14 15 16 17 18 19 20 21 22 23

CSC CRS CSG CSH CDA FAR FSF FSC FRS FSG FSH

Continuous Discrete Discrete Continuous Discrete Discrete Discrete Continuous Discrete Discrete Continuous

Quantitative Quantitative Qualitative Quantitative Qualitative Qualitative Quantitative Quantitative Quantitative Qualitative Quantitative

remain unchanged across sensitivity studies exist. Such features are listed in Table 5.

WELL CONFIGURATION The principal direction of fluid displacement is assumed to occur parallel to the channel axes (X direction) for waterflooding and gas injection. A single injector-producerpair well configuration is considered for the sensitivity-study simulations. Wells are placed in close proximity to opposing
268 Deep-Water Reservoir Performance

bounding edges of the sector model that are perpendicular to the channel axes. Wells are placed preferentially in (local) high net-to-gross zones consistent with field development practices. One production well is placed in the vicinity of the model center for depletion simulations. Only vertical and fully penetrating wells are considered in this study. Figure 12A displays a realization for the waterflooding recovery mechanism. Superimposed well locations are shown on saturation and net-togross maps. Figure 12B illustrates analogous panels for the depletion recovery mechanism.

Table 2. Description of the Reservoir Engineering Parameters Parameter Number 1 Matrix Reservoir Engineering Parameter BXL Parameter Description* Well-spacing parameter (fx) Well-spacing parameter (lx) Well-spacing parameter (fy) Well-spacing parameter (ly) Dip angle in X direction Dip angle in Y direction Rock compressibility Reference pressure (Pref.) Reference depth (Href.) Initial reservoir pressure at the reference depth (Pini) PVT parameter: Hydrocarbon fluid density (APIvalue) PVT parameter: average oil density (densO) PVT parameter: average water density (densW) SCAL parameter: acid number SCAL parameter: asphaltene percentage SCAL parameter: water salinity SCAL parameter: critical gas saturation SCAL parameter: residual oil saturation to gas Horizontal permeability (of sands) Permeability anisotropy Production rate Injection rate (inRat = 1.2 prRat) Model size in the vertical direction Characteristic 1 Discrete Discrete Discrete Discrete Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Continuous Characteristic 2 Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Quantitative Unit Dimensionless Dimensionless Dimensionless Dimensionless Degree () Degree () One over pound per square inch (psi1) Pounds per square inch (psi) Meter (m) Pounds per square inch (psi) Degree API (API) Pounds per cubic foot (lb/ft3) Pounds per cubic foot (lb/ft3) Dimensionless Percent (%) Parts per million (ppm) Fraction Fraction Millidarcy (md) Dimensionless Reservoir barrels per day (rbbl/day) Reservoir barrels per day (rbbl/day) Meter (m)

2 3 4 5

dipX dipY CR Pref.

APIvalue

7 8 9 10 11 12 13 14

acidNb asph sal sgc sorg permX kvTokh prRat

15

BVS

*PVT = pressure-volume-temperature; SCAL = special core analysis. The SCAL parameters determine the correlation-based relative permeability and capillary pressure relationships.

CHARACTERISTICS OF THE BASE CASE A 3-km (9842.5-ft) (X) by 2-km (6561.7-ft) (Y) box with a total thickness of 50 m (164 ft) houses a typical sector model. Hereafter, in the text, (1)

channel story sets are also informally referred to as meander belts and (2) channel stories are also referred to as channels for simplicity. The base-case sector model contains three meander belts exhibiting a medium degree of amalgamation. Overall,
Alpak et al. 269

270 Deep-Water Reservoir Performance

Table 3. Levels of Variation for the Geologic Parameters* Parameter Number 1 2 3 4 5 6 7 8 9 10 11 12 Parameter BNO BNG BNS BSC BRS BSG BSH BDA CDT CWD CSN CCN Parameter Description Number of meander belts Meander-belt channel-to-nonchannel ratio (meander-belt net-to-gross) Meander-belt channel-to-nonchannel ratio shift parameter Meander-belt shale drape coverage Random sequence number for the meander-belt shale drape realization Meander-belt shale drape geometry Meander-belt shale drape hole size Meander-belt degree of amalgamation Channel depth Channel widthtodepth ratio Channel sinuosity Channel-continuity number (frequency of mud plugs internal to the channel) Channel shale drape coverage Random sequence number for the channel shale drape realization Channel shale drape geometry Channel shale drape hole size Channel degree of amalgamation Channel-infill architecture Channel-infill shale frequency (thin-bedded channel margins) Channel-infill shale drape coverage Random sequence number for the channel-infill shale drape realization Channel-infill shale drape geometry Channel-infill shale drape hole size Unit Dimensionless Dimensionless Dimensionless Percent Dimensionless Dimensionless Element width Dimensionless Meter (m) Dimensionless Dimensionless Number of plugs per kilometer Percent Dimensionless Dimensionless Element width Dimensionless Dimensionless Dimensionless Percent Dimensionless Dimensionless Element width 0 30 0 30 0 30 45 V3 25 V2 35 V1 0 50 V0 (Base Case) 3 65 No variation (65-65-65) 60 1 Random 0.6 Medium 12 20 1.5 0 V1 6 75 Variation (90-60-30) 75 2 Margin 1.5 High 18 40 2 1 V2 90 V3 100

90 3

100

0.2 Low 6 10 1.2

24

13 14 15 16 17 18 19 20 21 22 23

CSC CRS CSG CSH CDA FAR FSF FSC FRS FSG FSH

45

60 1 Random 0.6 Low Convergent 5 60 1 Margin 0.6

75 2 Margin 1.5 Medium Accretionary 15 75 2 Random 1.5

90 3

100

0.2 Layered 0 45

High

Variable

90 3

100

0.2

*The unit element width is the average meander-belt width for erosional holes in meander-belt drapes. However, the unit element width is the average channel width for erosional holes in channel and channel-infill drapes. A circular shape is assumed for the erosional holes in shale drapes in this study. Thus, the hole diameter is equal to the hole size (width) by definition. Base-case parameter settings are displayed in the V0 column. The levels of variation are listed under the following column titles: V3, V2, V1, V1, V2, V3, , and so on.

Table 4. Levels of Variation for the Reservoir Engineering Parameters* Parameter Number 1 Parameter BXL Parameter Description** Well-spacing parameter (fx) Well-spacing parameter (lx) Well-spacing parameter (fy) Well-spacing parameter (ly) Dip angle in X direction Dip angle in Y direction Rock compressibility Reference pressure (Pref.) Reference depth (Href) Initial reservoir pressure at the reference depth (Pini) PVT parameter: hydrocarbon fluid density (APIvalue) PVT parameter: average oil density (densO) PVT parameter: average water density (densW) SCAL parameter: acid number SCAL parameter: asphaltene percentage SCAL parameter: water salinity SCAL parameter: critical gas saturation SCAL parameter: residual oil saturation to gas Horizontal permeability (of sands) Permeability anisotropy Production rate Injection rate (inRat = 1.2 prRat) 15 271 BVS Model size in the vertical direction Unit Dimensionless Dimensionless Dimensionless Dimensionless Degree Degree One over pound per square inch (psi1) Pounds per square inch (psi) Meter (m) Pounds per square inch (psi) Degree API (API) Pounds per cubic foot (lb/ft3) Pounds per cubic foot (lb/ft3) Dimensionless Percent Parts per million (ppm) Fraction Fraction Millidarcy (md) Dimensionless Reservoir barrels per day (rbbl/day) Reservoir barrels per day (rbbl/day) Meter (m) V 2 V1 V0 (Base Case) 1 100 1 100 10 0 2 105 4500 3000 4500 30 f(P, APIvalue) Reference value = 67.77; f(p) 1.0 0.727 120,000 0.05 0.1 1000 1.000 15,000 18,000 50 V1 10 90 V2 20 80 V3 25 76 V4 30 70 V5 37 64 V6 40 60

2 3 4 5

dipX dipY CR Pref.

5 1 105 3000 2000 3000 20

20 10 3 105 6000 4000 6000 40

20

30

APIvalue

7 8 9 10 11 12 13 14

acidNb asph sal sgc sorg permX kvTokh prRat

0.465

0.7 0.7 40,000 0.02 0.05 250 0.100 10,000 12,000

1.3 1.4 200,000 0.08 0.15 4000 20,000 24,000 75

2.5

0.001 5000 6000

Alpak et al.

25,000 30,000

*Base-case parameter settings are displayed in the V0 column. The levels of variation are listed under the following column titles: V3, V2, V1, V1, V2, V3, , and so on. **PVT = pressure-volume-temperature; SCAL = special core analysis. The SCAL parameters determine the correlation-based relative permeability and capillary pressure relationships.

Table 5. Common Characteristics of the Sector Models that Remain Unchanged over the Course of Simulations Parameter Number 1 Parameter Model size Parameter Description X direction total size Y direction total size X direction grid-block size Y direction grid-block size Z direction grid-block size Rate constrained Well geometry, producer Well geometry, injector Displacement (or flow) direction Sand porosity Unit Kilometer (km) Kilometer (km) Meter (m) Meter (m) Meter (m) Dimensionless Fraction V0 (Base Case) 3 2 30 20 Variable Yes Vertical Vertical X 0.25

Wells

3 4

flowDir Porosity

meander-belt net-to-gross is assumed as 65%, with channels exhibiting a low degree of amalgamation (Figure 13A). Meander-belt, channel, and channelinfill drapes are at 60% coverage level as shown in Figure 13A, B, and C, respectively. Meander-belt and channel drapes are distributed randomly,

whereas channel-infill drapes are concentrated along the margins of the channels. Each channel contains five channel-infill drapes. Commonly observed (in

Figure 12. Geology and (recovery mechanism driven) wellplacement strategy used in flow simulations. Two example simulation snapshots are shown: one for waterflooding and the other for primary gas depletion. The former snapshot is taken subsequent to water breakthrough in the producer well. The latter snapshot is taken right after the initialization of the dynamic model but before the producer well is opened for production. Color scheme: ([A] oil saturation [So], upper panel) So = 0.093, dark blue; So = 0.91, red; ([A] net-to-gross [NTG], lower panel) NTG = 0.22, orange; NTG = 1.0, red; ([B] mass of gas in the model [mg; pore volume gas density], upper panel) low mg, dark blue; high mg, green; ([B] NTG, lower panel) NTG = 0.22, orange; NTG = 1.0, red. In the above figure panels, the text boxes point to the exact locations of the injector (INJ) well and the producer (PRD) well. 272 Deep-Water Reservoir Performance

Figure 13. (A) Base-case sector model with three meander belts. Horizontal slices taken from this model by peeling off the stratigraphic surfaces (and the layers between these surfaces) are used to illustrate the drape layers and the erosional holes in these layers. (B) Plan view of the first meander-belt base layer (from the top of the model) illustrating the geometry of an example meander-belt drape (light green) with 60% coverage (BSC = meander-belt shale drape coverage), including randomly distributed erosional holes (dark green). (C) Plan view of the first channel-base layer (from the top of the model within the first meander belt) illustrating the geometry of an example channel drape (light blue) with 60% coverage (CSC = channel shale drape coverage) and randomly distributed erosional holes (darker blue patches in the channel track). (D) Plan view of the first channel-infill layer (from the top of the model within the first channel story, which is in turn within the first meander belt) illustrating the geometry of an example channel-infill drape (orange) with 60% coverage (FSC = channel-infill shale drape coverage), including erosional holes (red) distributed preferentially near the axis of the channel bed.

outcrops) convergent channel-infill architecture is adapted in the base case. The entire detail of the shale architecture is modeled explicitly. The model is initially saturated with a 30API oil at irreducible water saturation for waterflooding and gas-injection recovery mechanisms and with dry gas for the depletion recovery mechanism. An initial reservoir pressure of 4500 psi (31 MPa) is taken as the reference pressure for pressure, volume, and temperature (PVT) calculations. Only the isothermal flow of in-situ and injected fluids is investigated. A conventional black-oil formulation with standard correlation-based PVT relationships is used to describe reservoir fluids as a function of (varying) reservoir pressure during simulation. A homogeneous and isotropic sandstone permeability of 1000 md and a homogeneous sandstone porosity of 0.25 are considered in the base case to more clearly distinguish between the effects of shale architecture and petrophysical properties. Shales of all types are assumed impervious. Being such, they may potentially form prevalent barriers to flow in the reservoir depending on the architectural characteristics of a specific geologic realization. A total liquid (or gas) production rate of 15,000 reservoir barrels per day (rbbl/day) (2385 reservoir m3 per day [rm3/day]) is used as the production rate constraint. Injection rate constraint is set at 18,000 rbbl/day of water (2862 rm3/day) (or equivalent gas) for displacement-based recovery mechanisms. An injectiontoproduction rate ratio of 1.2 is maintained consistently even when the production rate is varied in the sensitivity study. In addition to rate constraints, wells are also constrained with realistic maximum and minimum bottomhole pressure bounds. Long periods of recovery allow reasonably accurate estimation of the ultimate unswept hydrocarbons caused solely by the effects of stratigraphy. All simulations are run for 50 yr. Depletion simulations typically terminate earlier because of the minimum bottomhole pressure constraint imposed on the production well. The injection well is placed downdip and the production well is placed updip in the reservoir for the waterflooding recovery mechanism in all simulations that involve a dipping sector model. The well-placement

configuration is reversed for the simulations performed for the gas-injection recovery mechanism. In the sensitivity studies, the effect of the spatial randomness is only evaluated for the placement of erosional holes in the shale drapes. Three realizations of the base case are generated for each drape type. The parameters BRS, CRS, and FRS represent the erosional holeplacement randomness for meanderbelt, channel, and channel-infill drapes, respectively. The effect of the variability in the spatial locations of other reservoir elements between wells (e.g., channel stories) on RF has been investigated via simulations of a very limited number of additional geomodel realizations. In these realizations, channel story locations are different than those in the models used for the sensitivity study, whereas all other characteristics of the sector models remain unchanged. Clearly, the erosional-hole placement varies simultaneously with channel story locations in these one-off realizations. We observed that the channel-story randomness effects on RF are, in general, quite comparable to those obtained for the erosional holeplacement randomness. This observation stems from a nonextensive study. A different combination of channel story and erosional-hole placement not captured in our limited study may yield a different result.

RESULTS OF SENSITIVITY STUDIES A dimensionless normalized impact value, NI, is defined to establish a quantitative sensitivity structure among sensitized matrix parameters. The definition of NI is given by NIi where j maxRFi minRFi j RFbase; i i 1; . . . ; n: 1

In the above equation, RF denotes the vector of RFs obtained from the sensitivity simulations conducted for the variation of a single parameter; RF base, the base-case RF for a given sensitivity study; and n, the number of parameters investigated within a given

Alpak et al.

273

Figure 14. (A) Recovery factor and (B) water-cut profiles simulated as a function of the channel shale drape coverage (CSC) parameter for the waterflooding recovery mechanism. The basecase CSC value (60%) corresponds approximately to the most likely (P50) value in the cumulative probability distribution shown in Figure 10B. FC indicates that the fine-scale case (full-detail model without upscaling) is flow simulated. STOIIP = stock tank oil initially in place.

sensitivity study. Example RF, water-cut, and gas-oil ratio profiles for waterflooding- and gas injection based hydrocarbon recovery mechanisms are respectively shown in Figures 14 and 15 as a function of the channel shale drape coverage (CSC) parameter. The CSC parameter significantly affects the recovery profile beyond approximately 60% coverage level for waterflooding and 75% coverage level for gas injection. The ranking of the matrix parameters in terms of their influence on RF follows from their NI values (Figures 1618). We emphasize that sensitivity charts are not Tornado charts. For the variation of a given parameter, a sensitivity chart does not reflect a downside or upside variation of outcomes with respect to a base case. Instead, it displays a normalized range of outcomes in response to the variations of the
Figure 15. (A) Recovery factor and (B) gas-oil ratio profiles simulated as a function of the channel shale drape coverage (CSC) parameter for the gas-injection recovery mechanism. The basecase CSC value (60%) corresponds approximately to the most likely (P50) value in the cumulative probability distribution shown in Figure 10B. FC indicates that the fine-scale case (full-detail model without upscaling) is flow simulated. STOIIP = stock tank oil initially in place. MSCF/STB = 1000 standard cubic foot per stock-tank barrel. 274 Deep-Water Reservoir Performance

matrix parameter under investigation. The process is repeated for each matrix parameter subject to the sensitivity study. The magnitude of the normalized range serves as the criterion for determining the relative ranking of a given parameter. Analysis is then simple: the larger the spread, the more sensitive the matrix parameter is. For waterflooding and gas injection, RF sensitivities are analyzed at the time of breakthrough and at the time of ultimate recovery, a sufficiently long time allowing the assessment of the hydrocarbon volume left behind because of stratigraphic reasons (typically 50 yr in this study). The time of ultimate recovery is also referred to as the trapped-hydrocarbon time in the remainder of this article. Recovery factor sensitivities are evaluated at the trapped-hydrocarbon

Figure 16. (A) Sensitivity of the breakthrough-time recovery factor for waterflooding simulations. (B) Sensitivity of the trapped hydrocarbontime recovery factor for waterflooding simulations. See Tables 1 and 2 for descriptions of the parameters. (Median [p = 0.50; dashed line in the left-hand side]; [p = 0.85; dashed line in the right-hand side]).

time in the sensitivity study conducted for the depletion recovery mechanism. Sensitivity charts are displayed in Figure 16 for waterflooding, in Figure 17 for gas injection, and in Figure 18 for depletion. Statistical information that emphasizes the central tendency (mean and median) and significance (p = 0.85) of simulation results is superimposed on sensitivity charts. A joint analysis of sensitivity charts for RF at the breakthrough time (RFbt) and RF at the trapped-

hydrocarbon time (RFth) yields a grouping of the matrix parameters into high-importance (H), intermediate-importance (M), and low-importance (L) categories for a given recovery mechanism (Table 6). In this analysis, a parameter is assigned H if it exceeds the p = 0.85 significance criterion either in the RFbt or the RFth sensitivity chart. In other words, the joint set of matrix parameters that exceed p = 85 significance level in RFbt or RFth sensitivity charts constitutes the set of H parameters.

Figure 17. (A) Sensitivity of the breakthrough-time recovery factor for gas-injection simulations. (B) Sensitivity of the trapped hydrocarbontime recovery factor for gas-injection simulations. See Tables 1 and 2 for descriptions of the parameters. (Median [p = 0.50; dashed line in the left-hand side]; [p = 0.85; dashed line in the right-hand side]). Alpak et al. 275

namely, waterflooding and gas injection. Next, we analyze the simulation results in further detail for every recovery mechanism evaluated in our study. Waterflooding The number of meander belts (BNO), BNG, BSC, BDA, channel-continuity number (CCN, which refers to the frequency of mudstone plugs internal to the channel), CSC, FSF, channel-infill shale drape coverage (FSC), and well spacing constitute the set of H parameters for waterflooding. Note that, hereafter, mudstone plugs are simply referred to as mud plugs in the text. Several matrix parameters play an M function in the RF behavior (Table 6). As an interesting result, reservoir engineering parameters that influence the balance between gravity and viscous forces have a relatively more subtle effect on RF compared to geologic parameters. This is because relatively light oils are considered in our study. More specifically, the displacement of 20 to 40API oils by injected water leads to favorable to very mildly unfavorable mobility ratio displacements. Gas Injection Gas-injection sensitivity-study results are similar to the waterflooding case in many aspects. As an important difference, in addition to geologic parameters and well spacing, several reservoir engineering parameters appear as key controlling factors, including the dip angle (reservoir dip along the channel direction), production rate, horizontal permeability, and fine-scale permeability anisotropy. The interplay between gravity and viscous forces arises as a crucial factor in characterizing the recovery efficiency. The sensitivity charts exhibit a relatively more stair steptype behavior, indicating a sensitivity structure that is less polarized. Although some H geologic parameters for waterflooding drop here to M level (e.g., BNO, CCN, and FSC), the set of M parameters is larger for the gas-injection recovery mechanism compared to waterflooding. Depletion The depletion recovery mechanism differs from displacement-based recovery processes. Fewer matrix parameters account for the large part of the RF

Figure 18. Sensitivity of the trapped hydrocarbontime recovery factor for depletion simulations. See Tables 1 and 2 for descriptions of the parameters. (Median [ p = 0.50; dashed line in the left-hand side]; [ p = 0.85; dashed line in the right hand side]).

Intermediate importance parameters are selected analogously but using the median as the criterion. The remaining parameters form the set of L factors.

DISCUSSION OF RESULTS Sensitivity Structure It is crucial to underline that the set of important matrix parameters that affect the recovery efficiency are also a strong function of the recovery mechanism applied to extract hydrocarbons from the subsurface. Therefore, before constructing a static model, one has to pay special attention to the type of fluidextraction process that will be simulated on a geologic model. Meander-belt channel-to-nonchannel ratio (BNG, also known as meander-belt net-to-gross), meander-belt shale drape coverage (BSC), and CSC constitute the set of important matrix parameters common to all investigated recovery mechanisms. In addition, the meander-belt degree of amalgamation (BDA) and channel-infill shale frequency (FSF) emerge as important parameters common to displacement-based recovery processes,
276 Deep-Water Reservoir Performance

Table 6. Categorization of the (Uncertainty) Matrix Parameters Based on their Impact on the Recovery Factor

response. Except for the CSC, geologic big-hitter parameters are associated with the meander-belt scale architecture, namely BNO, BSC, and BNG. The foremost reservoir engineering parameter is the reference (initial) reservoir pressure. Note the strong influence of abandonment conditions on the RF response in a depletion recovery mechanism. Although not investigated here in detail, RF obtained from a depletion-dominated recovery mechanism can also be governed by well and facilities constraints. Therefore, we recommend the investigation of the impact of H and, possibly, some M subsurface parameters on the RF behavior by coupling the subsurface model to a facilities model in field-specific studies. Coarse-Scale Modeling Requirements Analysis results indicate that not all of the elements of the detailed channel architecture need to be included in full-field models. Moreover, two alternative approaches for incorporating the effects of important stratigraphic parameters exist:

(1) explicit modeling, and (2) use of effective properties (see Alpak et al., 2010, and Alpak et al., 2011 for the details of the latter approach) The answer to the question, Which parameters need to be modeled explicitly and which ones can be left to effective properties? depends very much on the objectives of the dynamic modeling practice as well as on the available hard data at hand. It is desirable to model explicitly every seismically imaged reservoir element as a high-level rule for appraisal and development-phase models. The effects of seismically invisible reservoir elements can be accounted for with an explicit multiscenario modeling approach or by use of a multiscenario effective-property approach (e.g., low-, mid-, and high-case pseudorelative permeabilities), again depending on the objectives of dynamic modeling. For example, appropriate effective properties commonly perform accurately in predicting the cumulative recovery, RF, and field rates. However, multiscenario explicit modeling of certain subseismic elements (e.g., channel stories) may be necessary to decipher well-by-well responses and target left-behind hydrocarbons.
Alpak et al. 277

Figure 19. Cross sections from a fine, outcrop-scale parent model and the coarse-scale offspring models. See text for full explanation.

One has to recognize that explicit-modeling and effective-property approaches are not mutually exclusive. Contemporary detailed reservoir-scale models are far from having the level of stratigraphic architectural detail that exists in the high-resolution sector models simulated in our studies. Therefore, even when some subseismic elements are modeled explicitly (e.g., channel stories), the use of a set of effective properties may be needed to represent the effects of missing finer-scale architectural details. For example, one can build a channel storylevel, detailed full-field model but, still, the impact of channel-infill shale drapes may need to be included in a dynamic model (especially, if core and log data indicate an infill drape coverage >6070%). In such a case, one can calculate scale-appropriate effective properties (suitable for the detail level of the fullfield model) and incorporate them into the dynamic model via pseudorelative permeability functions. However, exploration-scale reservoir models are constructed typically at a coarse resolution (e.g., tank-type or layered models), and the available turnaround time for simulations is commonly very short. In such cases, one can resort to the sole use of ef278 Deep-Water Reservoir Performance

fective properties by partnering them to relatively simple reservoir-scale models.

THREE-DIMENSIONAL CONNECTIVITY METHOD: A SUMMARY The dynamic behavior of subseismic turbidite channel architecture in coarse reservoir-scale models can be represented by means of CFs implemented in terms of effective properties within reservoir simulation models (Alpak et al., 2010). Definition of the Connectivity Factor Connectivity factors are derived from representative full-detail and coarse-scale sector model simulations. High-resolution sector model simulations are performed to capture an accurate representation of the geologic detail. Coarse-scale and high-resolution sector model simulations are used together for quantifying the impact of channel architecture in terms of CFs. The CF quantifies the influence of model parameters at different model complexities. It is

defined as the ratio of the fine-scale RF to the coarsescale RF at a given point in time during production: RFt ; fine-scale 3-D model RFt ; coarse-scale 3-D model

CFt

Connectivity factors are computed for two crucial points in production time. The first type of CF is calculated using fine- and coarse-scale sector model RFs at their respective breakthrough times (t = bt). The resulting CF is defined as the breakthrough-time CF, CFbt. The second type of CF is calculated using fine- and coarse-scale sector model RFs at the time of ultimate recovery, also called the trapped-hydrocarbon time (t = th). Typically, a sufficiently late abandonment time is selected to quantify the hydrocarbon trapping effect of stratigraphic architecture. Ensuing CF is coined as the trappedhydrocarbon CF, CFth. The stronger the hydrocarbon trapping effect of channel architecture on recovery, the closer the CF value is to 0.0. For reservoirs where the imprint of channel architecture is weak, CF value is closer to 1.0. This observation is valid for both types of CFs. Note the difference between the CF and RF. The CF should be perceived as a correction factor for the effect of unmodeled stratigraphic detail on RF. Therefore, it is sensitive to a relatively more restricted set of parameters. The effects of many parameters that influence RF cancel out in the calculation of the CF. This is because they are already accounted for in the coarse-scale model. The CF depends strongly on the level of geologic detail in the coarse modeling scale and the recovery mechanism (waterflooding, gas injection, depletion, and others). Simulations for Characterizing Dynamic Connectivity The high-resolution fine-scale sector model simulations described previously are enriched with a twostage approach: 1. The high-resolution simulations are repeated for five coarsened versions of the fine-scale model,

which are designed to mimic the recovery behavior of coarse reservoir-scale models constructed at various resolutions in terms of channel architecture. A sensitivity analysis is conducted to identify the high-impact parameters that significantly influence the CF behavior. These parameters include BNG, BDA, BNO, BSC, CSC, FSC, FSF, and CCN. We also conclude that, albeit relatively weak with respect to geologic parameters, average well spacing is also a key parameter for the waterflooding recovery mechanism (Alpak et al., 2010). The above findings are still restricted to relatively light (2040 API) hydrocarbon-bearing turbidite channels, with sandstone permeabilities ranging from 2504000 md. 2. Flow simulations evaluating the simultaneous variation of multiple parameters are conducted in the second stage. Separate sets of simulations are run for every recovery mechanism and for one fine and five coarse modeling scales described below. Such simulations are performed only for the parameters with a noteworthy imprint on the CF behavior. The latter study amounted to a large number of simulations (31,200) on channelized turbidite models at multiple geologic resolutions (Alpak et al., 2010). Dynamic Modeling Scales Figure 19 shows cross sections from an example fine, outcrop-scale parent model and its coarsescale offspring models. The L0 models contain the full ground-truth architectural detail. The models have been coarsened by merging the boundary layers and the architecture together through upscaling. In L1 models, the layers of channel-infill architecture are upscaled into a single channel-infill layer. In L2 models, the channel-infill layers are merged with their respective channel-base drapes. In L3 models, the channels within a meander belt are combined into a single unit. In L4 models, the meander-belt base drapes are lumped with their respective meander belts, and in L5 models, all meander-belt layers are upscaled into a single layer. The simulation study described in Alpak et al. (2010) demonstrates that modeling the full stratigraphic detail of the
Alpak et al. 279

Figure 20. Validation of the proxy-model trapped-hydrocarbon connectivity factor (CFth) predictions with field databased trapped-hydrocarbon connectivity factors. The error bars (for the vertical axis) show P85, P50, and P15 connectivity factor values (from top to bottom) estimated using our statistical method that relies on a large number of flow simulations. The connectivity factor values in the horizontal axis are estimated directly from actual production data. The cases analyzed in this study extend from low-productivity (e.g., Reservoir A) to high-productivity (e.g., Reservoir J) reservoirs (modified from Alpak et al., 2010, with added reservoirs).

turbidite channel architecture (using L0-level sector models) has two main subsurface-flow characteristics in reference to the coarser representations of the stratigraphic architecture (L1-, L2-, L3-, L4-, and L5-level models): (1) increased trapped-hydrocarbon volume caused by stratigraphic dead ends and baffles to flow, for example, shale drapes; and (2) early injected fluid breakthrough. The latter conclusion is only valid for displacement-type recovery mechanisms such as waterflooding and dry gas injection.

on CFs obtained via flow simulations on fine- and coarse-scale sector models. We developed a separate proxy function for every recovery mechanism mentioned above and for every coarse modeling scale (L1L5) in Figure 19. Proxy functions are embedded into a spreadsheet that links to a Monte Carlo sampling engine and calculates CF statistical distributions based on input geologic-parameter statistical distributions. Statistical distributions of BNG, BDA, BNO, BSC, CSC, FSC, FSF, and CCN constitute the input. The spreadsheet also admits an average well-spacing value as input for the waterflooding recovery mechanism. The CF distributions estimated by the Monte Carlo simulation process are sampled at appropriate percentiles (e.g., P15, P50, and P85). Corresponding breakthrough-time and trapped-hydrocarbon CFs are converted into corresponding pseudorelative permeability functions using the method described in Alpak et al. (2010). In turn, pseudorelative permeability functions are applied to a coarse reservoir-scale model to incorporate the effects of missing subgridscale stratigraphic architecture. The pseudorelativepermeability effective properties lend themselves naturally to the variability in the stratigraphic fabric within a reservoir-scale model. For instance, a reservoir can be composed of a region with less amalgamated channels and a region with highly amalgamated channels. The stratigraphic variability can be accounted for in a straightforward manner by assigning pseudorelative permeability functions appropriate for each region. Alpak et al. (2010) shows a field application of the 3-D connectivity method on a channelized turbidite reservoir. Alpak et al. (2011) demonstrates how the 3-D connectivity method can be used within a multistage historymatching workflow applied to a channelized turbidite reservoir in offshore west Africa.

Effective-Property Modeling Method Predictive CF functions, also called proxy functions, are developed using the outcome of the large number of simulations conducted for waterflooding, gas-injection, and depletion recovery mechanisms. Proxy functions are closed-form equations derived by training multilayer-perceptron neural networks using a backpropagation algorithm (Haykin, 1999)
280 Deep-Water Reservoir Performance

Validation Connectivity factor predictions are conducted with the proxy model for 10 turbidite channel reservoirs with varying degrees of geologic characteristics. Abandoned and late-life reservoirs are selected to decrease the uncertainty around the ultimate RF.

Special attention is dedicated for selecting cases (1) where the dynamic connectivity is predominantly controlled by stratigraphy instead of structure (e.g., faults, deformation bands) and (2) that cover the full spectrum of observed reservoir-connectivity behaviors. The selected range varies from poorquality reservoirs requiring a P50 RF reduction of approximately 55% to good-quality reservoirs requiring virtually no reduction.Case-specific geologic-parameter distributions are created for each reservoir. This process involves the geologic and reservoir-engineering analysis of all available static and dynamic data. For each reservoir, the CFth prediction of the proxy model is plotted against the field CFth, computed by dividing the actual field ultimate RF by the ultimate RF simulated using a coarse field-scale model (Figure 20). A crossplot similar to that in Figure 20 is also shown in Alpak et al. (2010). In this article, the study is extended to high-performing channelized turbidite reservoirs. Data are added to the crossplot for reservoir I and reservoir J. The L5 resolution is selected for representing the stratigraphic architecture within coarse field-scale simulations of the ultimate RF. The same resolution is used for the proxy-model CFth computations. The only exception is reservoir D, where the dynamic model is at L3 resolution. Proxy-model P50 CFth predictions are compared to field dataderived CFth values. The P50 CFth predictions exhibit a reasonably good agreement with the field data except for reservoir C. The P15 to P85 range of the proxy-model CFth forecasts are also reported in the same crossplot (Figure 20). It is evident that the P15 to P85 range exhibits variability in the range of 0.17 (reservoir E) to 0.36 (reservoir H) dimensionless CF units, excluding reservoir C, which emerges as an outlier case. Ranges of the input-parameter distributions control the range of CFth outcomes. The most likely regions of input-parameter distributions capture the geologic ground truth accurately, as indicated by the degree of accuracy of P50 CFth predictions. However, ranges of the input parameters are still large, preventing the estimation of relatively narrower ranges for the CFth, except for reservoir E. No clear trend exists in Figure 20 between the

size of the P15 to P85 CFth range and the reservoir quality.

CONCLUSIONS A detailed flow-simulation study is performed using high-resolution interwell-scale sector models to identify the most influential stratigraphic architecture parameters that govern the recovery factor (RF) behavior in channelized turbidite reservoirs. The findings of this study are restricted to relatively light (2040API) hydrocarbon-bearing channelized turbidite reservoirs, with sandstone permeabilities ranging from 250 to 4000 md. Shale drapes are assumed impermeable, and sandstone-to-sandstone hydraulic communication occurs across shale drapes only through erosional holes. The geologic parameters associated with meander belts such as number of meander belts (BNO) (the parameter that embeds meanderbelt width information), meander-belt degree of amalgamation (BDA), meander-belt net-to-gross (BNG), and meander-belt shale drape coverage (BSC) emerge consistently as significant factors among the investigated recovery mechanisms. However, channel shale drape coverage (CSC), channel-infill shale drape coverage (FSC), and channel-infill shale frequency (FSF) also emerge as equally important factors, especially for displacement-type recovery mechanisms. A nontrivial separation of spatial scales emerges among geologic parameters as a function of the recovery mechanism. Results indicate that a hierarchical view to the sensitivity structure is not generally applicable. For example, the statement that most important parameters are associated with meander belts followed by channel stories and channel infill does not entirely hold, especially for displacement-based recovery mechanisms. The sensitivity structure is more complex than a simple dimension-based classification of reservoir elements. Interestingly, although shale drapes are relatively thin features vertically, they can have profound effects on the recovery efficiency above a certain coverage level (typically >6070%). Despite that mud plugs are
Alpak et al. 281

less commonly observed in outcrops, if present, they can also have a noteworthy impact on RF. Our study shows that the shale architecture and sandstone content of meander belts dominate the recovery behavior. Specific features of the sandstone architecture that are commonly thought to be of great importance (e.g., channel size, sinuosity, and others) are less significant. Gas injection is the most sensitive recovery mechanism to stratigraphic architecture, followed by waterflooding and depletion. In the case of waterflooding and depletion, a relatively more distinct separation between highimportance (H) and intermediate-importance (M) matrix parameters emerges, whereas this boundary is less obvious for gas injection. The ranking order of sensitivities reflects a more continuous stepwise progression. Reservoir engineering parameters that affect the balance between gravity and viscous forces have a relatively more subtle effect on the RF behavior of waterflooding compared to geologic parameters. Results of the gas-injection sensitivity study are similar to the waterflooding case in many aspects. However, several reservoir-engineering parameters appear as key controlling factors in addition to geologic parameters. The interplay between gravity and viscous forces arises as a crucial factor in characterizing the recovery efficiency for gas injection. Fewer parameters account for the large part of the RF response in the depletion recovery mechanism. The sensitivity of RF with respect to geologic parameters is dominated by meander-belt scale parameters and mud plugs. Explicit inclusion of all the parameters that affect the dynamic response in a reservoir model may not be a practical approach. Several high-impact channel-architecture elements could be modeled explicitly in the reservoir-scale model depending on the desired coarseness and available information. The effects of the remaining parameters could be incorporated into the dynamic model via use of effective properties. Alpak et al. (2010, 2011) describe a novel effective-property modeling method for incorporating the dynamic effects of fine-scale stratigraphic elements into relatively coarser scale dynamic models of channelized turbidite reservoirs. A detailed validation study using field pro282 Deep-Water Reservoir Performance

duction data proves the geologic consistency of the method.

REFERENCES CITED
Abreu, V., M. Sullivan, C. Pirmez, and D. Mohrig, 2003, Lateral accretion packages (LAPs): An important reservoir element in deep-water sinuous channels: Marine and Petroleum Geology, v. 20, p. 631648, doi:10.1016/j .marpetgeo.2003.08.003. Alpak, F. O., M. D. Barton, F. F. van der Vlugt, C. Pirmez, B. E. Prather, and S. H. Tennant, 2010, Simplified modeling of turbidite channel reservoirs: Society of Petroleum Engineers Journal, v. 15, p. 480494. Alpak, F. O., M. D. Barton, and D. Castineira, 2011, Retaining geologic realism in dynamic modeling: A channelized turbidite reservoir example from west Africa: Petroleum Geoscience, v. 17, p. 3552, doi:10.1144/1354 -079309-033. Arnott, R. W. C., 2007, Stratal architecture and origin of lateral accretion deposits (LADs) and conterminous innerbank levee deposits in a base-of-slope sinuous channel, lower Isaac Formation (Neoproterozoic), east-central British Columbia, Canada: Marine and Petroleum Geology, v. 24, p. 515528, doi:10.1016/j.marpetgeo.2007 .01.006. Barton, M., C. OByrne, C. Pirmez, B. Prather, F. van der Vlugt, F. O. Alpak, and Z. Sylvester, 2010, Turbidite channel architecture: Recognizing and quantifying the distribution of channel-base drapes using core and dipmeter data, in M. Pppelreiter, C. Garca-Carballido, and M. A. Kraaijveld, eds., Dipmeter and borehole image-log technology: AAPG Memoir 92, p. 195211. Barton, M. D., C. J. OByrne, B. Prather, C. Pirmez, Z. Sylvester and D. Commins, 2007a, Deep-water channel-complex architecture, Popo fault block, Brushy Canyon Formation, Texas, U.S.A.: Part 2. Facies architecture, in T. H. Nilsen, R. D. Shew, G. S. Steffens, and J. R. J. Studlick, eds., Atlas of deep-water outcrops: AAPG Studies in Geology 56, p. 463466. Barton, M. D., G. S. Steffens, and C. J. O Byrne, 2007b, Facies architecture of a submarineslope-channel complex, Condor West channel, Cerro Toro Formation, Chile, in T. H. Nilsen, R. D. Shew, G. S. Steffens, and J. R. J. Studlick, eds., Atlas of deep-water outcrops: AAPG Studies in Geology 56, p. 149153. Beaubouef, R. T., C. Rossen, F. Zelt, M. D. Sullivan, D. Mohrig, and D. C. Jennette, 2000, Deep-water sandstones, Brushy Canyon Formation, west Texas: AAPG Hedberg Field Research Conference Field Guide, April 1520, 1999, AAPG Continuing Education Course Note Series 40, 48 p. Berg, E. A., and P. A. Kjarnes, 2003, Uncertainty, risk and decision management on the Ormen Lange gas field, offshore Norway (abs.): AAPG International Conference, Barcelona, Spain, September 21 24, 2003, http ://www.searchanddiscovery.com/abstracts/pdf/2003 /intl/index.htm#B (accessed December 2012).

Bouchet, R., B. Levallois, G. Mfonfu, and J.-F. Authier, 2004, Girassol field optimized development (abs): AAPG International Conference, Cancun, Mexico, October 2427, 2004, http://www.searchanddiscovery.com/documents /abstracts/2004intl_cancun/index.htm#b (accessed December 2012). Bridge, J. S., and J. A. Diemer, 1983, Quantitative interpretation of an evolving ancient river system: Sedimentology, v. 30, p. 599 623, doi:10.1111/j.1365-3091 .1983.tb00698.x. Chapin, M., and F. Keller, 2007, Channel-fill sandstones at San Clemente State Beach, California, U.S.A., in T. H. Nilsen, R. D. Shew, G. S. Steffens, and J. R. J. Studlick, eds., Atlas of deep-water outcrops: AAPG Studies in Geology 56, p. 401405. Chapin, M., P. Swinburn, R. van der Weiden, D. Skaloud, S. Adesanya, D. Stevens, C. Varley, J. Wilkie, E. Brentjens, and M. Blaauw, 2002, Integrated seismic and subsurface characterization of Bonga field, offshore Nigeria: The Leading E dge, v. 21, p . 1125 113 1, doi: 10.1 190/1 .1523745. De Jager, G., J. F. M. Van Doren, J. D. Jansen, and S. M. Luthi, 2009, An evaluation of relevant geological parameters for predicting the flow behavior of channelized reservoirs: Petroleum Geoscience, v. 15, p. 345354, doi:10.1144/1354-079309-819. Eschard, R., E. Albouy, R. Deschamps, T. Euzen, and A. Ayub, 2003, Downstream evolution of turbiditic channel complexes in the Pab Range outcrops (Maastrichtian, Pakistan): Marine and Petroleum Geology, v. 20, p. 691710, doi:10.1016/j.marpetgeo.2003.02.004. Friend, P. F., M. J. Slater, and R. C. Williams, 1979, Vertical and lateral building of river sandstone bodies, Ebro Basin, Spain: Journal of the Geological Society, v. 136, p. 3946, doi:10.1144/gsjgs.136.1.0039. Gardner, M. H., and J. M. Borer, 2000, Submarine channel architecture along a slope-to-basin profile, Permian Brushy Canyon Formation, in A. H. Bouma, and C. G. Stone, eds., Fine-grained turbidite systems and submarine fans: AAPG Memoir 72 and SEPM Special Publication 68, p. 195213. Govan, A., T. Primmer, C. Douglas, N. Moodie, M. Davies, and F. Nieuwland, 2006, Reservoir management in a deep-water subsea field: The Schiehallion experience: Society of Petroleum Engineers Reservoir Evaluation and Engineering, v. 9, p. 382390. Hampton, B. D., T. V. Wilson, and R. Crookbain, 2006, From euphoria to reality: One operators experience with reservoir performance in the deep-water Gulf of Mexico: AAPG Annual Convention Abstracts, v. 15, p. 42. Haykin, S., 1999, Neural networks: A comprehensive foundation: Upper Saddle River: New Jersey, Prentice Hall, 842 p. Humphreys, N. V., L. G. Myers, A. G. Pollin, S. Hill, and I. Treherne, 1997, Using interference tests during field startup to solve critical reservoir-management issues at the Zafiro field, offshore Equatorial Guinea: Society of Petroleum Engineers Production and Facilities, v. 12, p. 205211. Kendrick, J. W., 2000, Turbidite reservoir architecture in the

northern Gulf of Mexico deep water: Insights from the development of Auger, Tahoe, and Ram/Powell fields, in P. Weimer, R. M. Slatt, J. H. Coleman, N. C. Rosen, H. Nelson, A. H. Bouma, M. J. Styzen, and D. T. Lawrence, eds., Deep-water reservoirs of the world: 20th Annual Gulf Coast Section SEPM Foundation Bob F. Perkins Research Conference, Houston, Texas, December 36, 2000, p. 450468. Kolla, V., H. W. Posamentier, and L. J. Wood, 2007, Deepwater and fluvial sinuous channels: Characteristics, similarities and dissimilarities, and modes of formation: Marine and Petroleum Geology, v. 24, p. 388405, doi:10 .1016/j.marpetgeo.2007.01.007. Larue, D. K., 2004, Outcrop and waterflood simulation modeling of the 100-ft channel complex, Texas, and the Ainsa 2 channel complex, Spain: Analogs to multistory and multilateral channelized slope reservoirs, in G. M. Grammer, P. M. Harris, and G. P. Eberli, eds., Integration of outcrop and modern analogs in reservoir modeling: AAPG Memoir 80, p. 337364. Larue, D. K., and F. Friedmann, 2001, Stratigraphic uncertainty in field development studies: A conceptual modeling app roa ch : The Lead ing Ed ge, v. 20, p . 30 33, doi:10.1190/1.1438872. Larue, D. K., and F. Friedmann, 2005, The controversy concerning stratigraphic architecture of channelized reservoirs and recovery by waterflooding: Petroleum Geoscience, v. 11, p. 131146, doi:10.1144/1354-079304-626. Larue, D. K., and J. Hovadik, 2006, Connectivity of channelized reservoirs: A modeling approach: Petroleum Geoscience, v. 12, p. 291 308, doi:10.1144/1354 -079306-699. Larue, D. K., and Y. Yue, 2003, How stratigraphy influences oil recovery: A comparative reservoir database study concentrating on deep-water reservoirs: The Leading Edge, v. 22, p. 332339, doi:10.1190/1.1572086. Li, H., and J. Caers, 2011, Geological modeling and history matching of multiscale flow barriers in channelized rese r voirs: Methodology and application: Petroleum Geoscience, v. 17, p. 1734, doi:10.1144/1354-079309-825. Lien, T., R. G. Walker, and O. J. Martinsen, 2003, Turbidites in the Upper Carboniferous Ross Formation, western Ireland: Reconstruction of a channel and spillover system: Sedimentology, v. 50, p. 113148, doi:10.1046/j .1365-3091.2003.00541.x. Mayall, M., and C. OByrne, 2002, Reservoir prediction and development challenges in turbidite slope channels: Offshore Technology Conference, Houston, Texas, May 6 9, 2002, OTC Paper 14029, 10 p. Mayall, M., E. Jones, and M. Casey, 2006, Turbidite channel reservoirs: Key elements in facies prediction and effective development: Marine and Petroleum Geology, v. 23, p. 821 841, doi:10.1016/j.marpetgeo.2006.08.001. Mutti, E., G. S. Steffens, C. Pirmez, M. Orlando, and D. Roberts, 2003, Turbidites: Models and problems: Marine and Petroleum Geology, v. 20, p. 733755. Pirmez, C., R. T. Beaubouef, S. J. Friedmann, and D. C. Mohrig, 2000, Equilibrium profile and base level in submarine channels: Examples from late Pleistocene systems and implications for the architecture of deep-water reservoirs, in P.

Alpak et al.

283

Weimer, R. M. Slatt, J. H. Coleman, N. C. Rosen, H. Nelson, A. H. Bouma, M. J. Styzen, and D. T. Lawrence, eds., Deep-water reservoirs of the world: 20th Annual Gulf Coast Section SEPM Foundation Bob F. Perkins Research Conference, Houston, Texas, December 36, 2000, p. 782805. Pyles, D. R., D. C. Jennette, M. Tomasso, R. T. Beaubouef, and C. Rossen, 2010, Concepts learned from a 3-D outcrop of a sinuous slope-channel complex: Beacon Channel complex, Brushy Canyon Formation, West Texas, U.S.A.: Journal of Sedimentary Research, v. 80, p. 6796, doi:10.2110/jsr.2010.009. Rossen, C., and R. T. Beaubouef, 2007, Slope-channel complexes at Guadalupe Canyon, upper Brushy Canyon Formation, Texas, U.S.A., in T. H. Nilsen, R. D. Shew, G. S. Steffens, and J. R. J. Studlick, eds., Atlas of deep-water outcrops: AAPG Studies in Geology 56, p. 429431. Sprague, A. R. G., et al., 2002, The physical stratigraphy of deep-water strata: A hierarchical approach to the analysis of genetically related stratigraphic elements for improved reservoir prediction (abs.): AAPG Annual Meeting, Houston, Texas, March 1013, 2002, http://www .searchanddiscovery.com/abstracts/pdf/2002/annual /CONTENTS.HTM#S (accessed December 2012).

Stright, L., 2006, Modeling, upscaling, and history matching thin, irregularly shaped flow barriers: A comprehensive approach for predicting reservoir connectivity: Society of Petroleum Engineers Annual Technical Conference and Exhibition, San Antonio, Texas, September 24 27, 2006, SPE Paper 106528. Sylvester, Z., C. Pirmez, and A. Cantelli, 2011, A model of submarine channel-levee evolution based on channel trajectories: Implications for stratigraphic architecture: Marine and Petroleum Geology, v. 28, p. 716 727, doi:10.1016/j.marpetgeo.2010.05.012. Wen, R., 2005, 3-D geologic modeling of channelized reservoirs: Applications in seismic-attribute facies classification: First Break, v. 23, p. 7178. Wen, R., A. W. Martinius, A. Nss, and P. S. Ringrose, 1998, Three-dimensional simulation of small-scale heterogeneity in tidal deposits: A process-based stochastic method, in A. Buccianti, G. Nardi, and A. Potenza, eds., Proceedings of the 4th Annual Conference of the International Association of Mathematical Geology, Ischia, Naples, Italy, p. 129134. Willis, B. J., 1993, Ancient river systems in the Himalayan foredeep, Chinji Village area, northern Pakistan: Sedimentary Geology, v. 88, p. 4376.

284

Deep-Water Reservoir Performance