You are on page 1of 16

The controversy concerning stratigraphic architecture of channelized

reservoirs and recovery by waterflooding


D. K. Larue1 and Francois Friedmann2
1
ChevronTexaco Energy Technology Company, 9525 Camino Media, Bakersfield, CA, USA
(e-mail: dkla@chevrontexaco.com)
2
Consultant, 917 Johnson Street, Manhattan Beach, CA, 90266, USA

ABSTRACT: There is a subtle controversy in the petroleum industry regarding the


relationship between reservoir architecture and recovery efficiency from petroleum
reservoirs. Stratigraphers tend to believe that facies architecture and geo-
metric shapes strongly govern recovery, whereas engineers stress factors such as
permeability heterogeneity and anisotropy. To understand the relationship between
architecture and recovery better, a suite of conceptual models of different
channelized clastic reservoir architectures was constructed and pore-volume replace-
ment waterflood simulations were performed on each. Three 11-member suites of
models were constructed at net: gross values of 35%, 60% and 85%. The reservoir
architecture features that were varied were channel width to thickness, thickness,
sinuosity, stacking patterns, orientation and reservoir element type (point bar vs.
channel). The three sets of models appear visually very different from a reservoir
architecture standpoint. Permeability and porosity were simulated using geostatistical
techniques but share the same conditioning population. The same water saturation,
well count (110 acre spacing), fluid properties and relative permeability were used in
each waterflood simulation. In this situation, in which the only significant variable
was reservoir architecture, only a few percent spread in recovery was noted for each
net: gross suite. Variation in recovery efficiency can be shown to be associated with
reservoir connectivity, which is generally high for all models but shows some
variation. Additional studies were made to address the influence of well count,
mobility ratio, permeability heterogeneity and geostatistical seed number on recovery
efficiency.
Two stratigraphic factors are shown to influence recovery efficiency: reservoir
connectivity and permeability heterogeneity. Both factors influence the volumetric
sweep efficiency of reservoirs. Models that appear to be visually different, but have
similar connectivity, well count and permeability heterogeneity, have similar sweep
efficiencies and, therefore, recovery efficiencies. The practical importance of this
observation for development studies is that building and flow-simulating suites of
models with different reservoir architectures may not help to characterize recovery
uncertainty unless the models are constructed to emphasize differences in sweep
efficiency.
KEYWORDS: reservoir architecture, connectivity, waterflooding, permeability heterogeneity, channel
architecture, deep-water channel deposits

INTRODUCTION relatively simple to quantify because volumes are relatively easy


to define, at least mathematically. Dynamic uncertainty is more
Capital investments for oil and gas field development can be difficult to define because it depends on a multitude of factors,
extremely costly and can exceed billions of dollars in remote including geological factors, time, pressures, well count and
areas. The greatest economic uncertainties in these billion- orientation, fluid factors and reservoir management practices.
dollar projects typically are geological and concern uncertainty In this paper, the origins and causes of dynamic uncertainty are
in reservoir volume and uncertainty attached to the quantities examined in more detail.
and rates of oil or gas production (for example, Dromgoole There is a largely unrecognized controversy regarding the
et al. 2000). These uncertainties are referred to commonly as degree to which stratigraphic architecture influences recovery
‘static’ and ‘dynamic’, or ‘volumetric’ and ‘producibility’ (Larue efficiency. As discussed below, geologists tend to believe
& Friedmann 2000). In concept, volumetric uncertainty seems that stratigraphic architecture and geometric shapes influence
Petroleum Geoscience, Vol. 11 2005, pp. 131–146 1354-0793/05/$15.00  2005 EAGE/Geological Society of London
132 D. K. Larue & F. Friedmann

recovery strongly, based on connectivity and continuity argu-


ments. Because of this, there has been a tendency in the
geological community to make more and more complex and
realistic facies models. Engineers, while acknowledging the
importance of connectivity and continuity, are more prone to
stress permeability heterogeneity and permeability anisotropy
on recovery, as well as dynamic reservoir characteristics such as
fluid types and well architecture.

PROBLEM STATEMENT
The goals of the study were to: (1) find out how much
stratigraphic architecture controls recovery efficiency at the
field development scale; and (2) find which stratigraphic
features or characteristics, if missed, could cause disastrous
consequences to a large development project. The study con-
centrates on channelized reservoirs because previous workers
have argued that they are harder to produce (Tyler & Finley
1991; Tyler et al. 1984) and because many new deep-water
discoveries are in channelized reservoirs. The emphasis here is
on stacking patterns and geometries of channelized reservoirs,
with implications for both deep-water and fluvial channels.
The simplest and most direct method of analysing the
relationship between stratigraphic architecture and recovery
efficiency is by computer simulating conceptual models in Fig. 1. Recovery efficiency plotted for a number of different
depositional environments, with drive and/or recovery mechanism
which certain properties can be held constant while others are or mechanisms noted (redrafted from Tyler et al. 1984). From this
varied in a systematic way. Underlying assumptions are that plot, it can be argued that recovery efficiency from clastic reservoirs
computer simulation models of channel architecture are predic- is a function of both depositional environment and recovery/drive
tive of real flow performance and that no geological rules or mechanism, although it is difficult to discern the relative importance
observations are violated by holding properties constant while of each parameter.
varying other parameters.
One of the most widely known and cited examples relating
stratigraphy to recovery is the study of Tyler et al. (1984;
PREVIOUS STUDIES: THE CONTROVERSY updated in 1991 by Tyler & Finley). In their study, 450 major
CONCERNING STRATIGRAPHY, DRIVE Texas petroleum reservoirs were evaluated (Galloway et al.
MECHANISM AND RECOVERY EFFICIENCY 1983). They showed that average recovery efficiency could be
Associating reservoir producibility with stratigraphic architec- tied closely to depositional environment and recovery mech-
ture is not a new topic. However, it is not recognized widely by anism (Fig. 1). Tyler et al. (1984) and Tyler & Finley (1991) were
the geological or engineering community that there is a contro- careful to point out the importance of drive type on recovery
versy concerning the relationship between reservoir pro- efficiency, which is also shown in Figure 1. It is not clear from
ducibility and stratigraphic architecture, or what aspects of Figure 1 whether recovery efficiency is affected more by
stratigraphic architecture affect producibility. There is tacit depositional environment or drive mechanism. Reservoir
agreement that stratigraphic characteristics affecting recovery engineers (for example, Thakur & Satter 1998) have stressed
efficiency include reservoir geometry, connectivity, continuity, the importance of reservoir and fluid characteristics, such as
heterogeneity and anisotropy. The quantitative definition of fluid mobility ratio, permeability heterogeneity, the relationship
these terms and the assignation of quantitative importance to between vertical and horizontal permeability, and reservoir
them is more nebulous. management practices (Fig. 2). Recently, Larue & Yue (2003)
Many of the current concepts about the importance of evaluated the relationship between depositional environment,
reservoir architecture and connectivity are derived from field- drive mechanism and recovery efficiency using a number of
based studies of waterflooding in West Texas and other fields in reservoir databases, including the database used by Tyler et al.
the USA (George & Stiles 1978; Barber et al. 1983; Gould & (1984) (Fig. 3a, b). To isolate the effect of drive mechanism,
Sarem 1989; Stiles & Magruder 1992). These workers stress that recovery efficiency is plotted against depositional environ-
unless there is continuity between injection and production ment for reservoirs produced by waterflooding or water drive
wells during a waterflood, the reservoir will be swept incom- (aquifer support) only. There is no obvious relationship
pletely. Infill drilling increases both the rate of depletion in a between depositional environment and recovery efficiency
field by adding more production wells, and the reserves by (Fig. 3a) observable in several databases. Instead, there seems to
intersecting reservoir elements previously unconnected to pro- be a median recovery (P50) of about 33%, with a considerable
duction wells. In the Means San Andres unit (George & Stiles spread. However, they were able to show a clear relationship
1978; Barber et al. 1983), infill drilling from 40-acre to 20-acre between drive mechanism and recovery efficiency (Fig. 3b).
to 10-acre spacing increased reserves by 5% to 10% of the Geological studies emphasizing 3D reservoir characterization
OOIP through the production of discontinuous reservoir and flow simulation have addressed specific relationships
elements. However, it should be noted that this connectivity between stratigraphic architecture and recovery efficiency.
study was 2D and the reservoirs are characterized by complex Jones et al. (1995) identified parameters that influence water-
permeability heterogeneity, which makes it unclear as to which flood performance for low net: gross fluvial reservoirs. They
factors led to increased reserves. stressed that net: gross, the stacking pattern of channel belts,
Channelized reservoirs and waterflooding 133

Fig. 2. Thakur & Satter (1998) created


a simple reservoir model and varied
geology and fluid properties prior to
simulating a waterflood. Results are
plotted as percent recovery as a
function of pore volume of water
injected. (a) The effect of waterflooding
a reservoir prior to and after reaching
bubble-point pressures. Higher
recoveries are associated with more
favourable reservoir management
strategies (earlier waterflooding relative
to reservoir bubble point). (b) The
same reservoir model was assigned
four different kv/kh ratios and each
model was simulated. Higher recoveries
are associated with more homogeneous
kv/kh ratios. (c) The same reservoir
model was assigned different
permeability heterogeneities, strongly
affecting recovery efficiency.
(d) Different fluid API values were
modelled. Better recovery was attained
by lighter, less viscous oils. (All figures
after Thakur & Satter 1998.)

the number of channels within a belt and channel width and necessary to list key characterization variables for channelized
thickness all affect waterflood performance significantly. In reservoirs. A detailed discussion of the fluvial and deep-water
high net: gross fluvial reservoirs, permeability contrast was channelized architecture is provided in Appendix A. The
found to be important (Hoimyr et al. 1993; Jones et al. 1994). model grid, well pattern and experimental design will now be
Larue & Friedmann (2000) stressed the importance of per- described, followed by a description of property and waterflood
meability heterogeneity on waterflood recovery for channelized simulation studies. Finally, definitions of significance and causal
reservoirs of both high and low net: gross. They stressed that factors will be presented.
since connectivity can be achieved at relatively low net: gross
values (35%), the channelized geological models can behave Reservoir characterization variables
like tanks. In a shallow-marine reservoir, stacking patterns of Key characteristics of both deep-water and fluvial channelized
parasequences were found to be the most important. Cook et al. reservoir architecture are net: gross (or sand fraction), channel
(1999) and Larue & Legarre (2004) noted that incorporat- width/thickness, thickness, sinuosity, stacking pattern, channel
ing high-resolution sequence stratigraphic architecture and deviation and reservoir element type. Net: gross represents the
reservoir quality trends within each parasequence of a reservoir centre point of the reservoir characterization study because of
simulation model helped improve the flow simulation history its clear relationship with reservoir connectivity (King 1990).
match and aided in identifying by-passed oil in a shoreface Three suites of models were constructed based on three net:
reservoir. For submarine channelized reservoirs, net: gross, gross assignments of 35%, 60% and 85% – the TM, SM and
permeability heterogeneity, mean permeability, mobility ratio EM suites (so-called because of the first letter of each number).
and residual oil saturation were found by Friedmann et al. A net: gross value of 35% for the TM suite was selected
(2001) to be significant factors influencing waterflood perform- because it represents a good lower boundary for reservoir net:
ance. For a tidal reservoir, the sizes of tidal bars and effective gross and channelized reservoirs tend to behave as complex
permeability of heterolithic facies were found to have the labyrinths (Weber & Van Geuns 1990).
largest impact on single-phase fluid flow (Brandsæter et al. Three channel thicknesses were used: 2 m, 5 m and 15 m
2000). (Table 1). This range covers much of the spread observed in
In summary, there is a wide range of views regarding the fluvial and deep-water deposits (for example, Fielding & Crane
degree to which stratigraphic architecture or what characteris- 1987; Clark & Pickering 1996). Ranges of channel width to
tics of stratigraphic architecture influence influence recovery thickness ratios, from 20 to 80, were derived from fluvial and
efficiency. The geological community continues to build more deep-water database studies (Fielding & Crane 1987; Bridge
and more complex facies models of a wide array of depositional 1993; Swanson 1993; Clark & Pickering 1996; Bridge & Tye
environments without demonstrating that volumetric or 2000). Channel sinuosities ranged from nearly straight to
dynamic uncertainty is being defined better or reduced, or that moderately sinuous, representative of the ranges observed both
predictions of flow performance are more accurate. in fluvial and deep-water depositional systems (Clark et al.
1992). Deviations in channel orientations were derived from
a study of subaerial and bathymetric maps of fluvial and
APPROACH deep-water channel systems.
To address how a simulation study can test in what way Stacking patterns included: (1) stochastic distributions
stratigraphic architecture affects recovery efficiency, it is of channel deposits; (2) stochastic distributions of channel
134 D. K. Larue & F. Friedmann

Table 1. Characteristics of channels used in model suite studies

Characteristic P10 value P50 value P90 value


Net: gross (%) 35 60 85
A. Width/thickness 20 50 80
B. Channel thickness (m) 2 5 15
C. Stacking pattern Bars Random Clustered at base
D. Sinuosity 1.1 1.5 23
E. Deviation of channel () 3 30 60

abandonment in deep-water systems (Prather & Keller 2000).


Finally, the implication of stochastically distributed point-bar
deposits is that channel fills are not the fundamental reservoir
element of channelized reservoirs, but point bars are (Van
Wagoner 1995; Bridge & Tye 2000; de Rooij et al. 2002). From
a continuity standpoint, it was predicted that the stochastic
distribution of point bars would result in the most disorganized
reservoir model, followed by the stochastic distribution of
channel deposits, followed by the stochastic distribution of
channel deposits with a vertical trend (Table 1). For this
modelling study, no attempt was made to make genetically-
related channel belt families (for example, Henriquez, et al.
1988; Bratvold et al. 1994). Also, aggradational channel stacking
patterns characteristic of some deep-water channel reservoir
architecture are not considered here.

Grid construction
The grid used in this study consists of horizontal layers that are
divided by area into cells of 27 m by 27 m by 0.8 m. This grid
was used for both geological characterization and computer-
based flow simulation studies. The grid contains 75 columns, 75
rows and 40 layers, totalling 225 000 cells. The dimension of
the grid is 2025 m by 2025 m by 35 m (thick). Cell size was
chosen so that the geology could be modelled without having
Fig. 3. (a) Larue & Yue (2003) re-evaluated the relationship between
depositional environment and recovery efficiency using a number of to coarsen the model for flow simulation studies.
available databases, including the databases from Tyler et al. (1984)
and Galloway et al. (1983) whose data are shown in Figure 1. Well pattern
Descriptions of the different databases are provided in Larue & Yue
(2003). They separated drive and recovery mechanism in order to Waterflood well pattern, especially well spacing, could impact
compare the relative importance of drive and depositional environ- recovery efficiency clearly. Most of the models were simulated
ment. P10, P50 and P90 values are shown for each distribution. using a 110-acre spacing. However, as discussed below,
From this plot, there seems to be no strong or obvious relationship additional cases were simulated with 250-acre, 1000-acre and
between depositional environment and recovery efficiency. Larue &
Yue (2003) noted that this could be due to a lack of other essential even greater spacing.
information such as oil gravity, viscosity, net: gross, or relative
permeability effects. Number of samples shown as N. (b) Larue & Property modelling
Yue (2003) plot recovery efficiency of deep-marine reservoirs pro-
duced by different recovery and drive mechanisms. There is a clear A Boolean modelling approach was used to create channel and
indication that primary recovery produces less than reservoirs point-bar geometries. Channel geometry and characteristics
assisted by aquifers or secondary or tertiary recovery, a fact long were specified from values in Table 1 and used by the Boolean
known in the engineering literature but substantiated by these data. program to construct realistic channel deposits. After construc-
Additional data from Tyler & Finley (1991).
tion of the channel and overbank facies, porosity, permeability
and water saturation were modelled within the facies using
deposits with vertical upward-decreasing trends; and (3) sto- sequential Gaussian simulation. Sequential Gaussian simula-
chastic distribution of point-bar deposits. The implication for tions require information about the target population and
channel deposits that are stochastically (‘randomly’) distributed spatial variance information defined using a 3D variogram,
is that multi-storey and multi-lateral channel complexes repre- which had a correlation length of 750 m along the channel axis
sent amalgamations of genetically different channel deposits. and 3 m vertically. Information about porosity, permeability
Such stacking patterns should be chaotic and disorganized. The and water saturation used in geological model construction are
implication of stochastic distributions of channel deposits with given in Table 2.
vertical upward-decreasing trends is that multi-storey and
multi-lateral channel complexes represent amalgamations of
genetically different channel fills and that there are fewer Waterflood simulation
channel deposits vertically. Such a diminution in the number Waterflood simulations were performed using a standard void-
of channel deposits has been associated with a rise in age replacement approach (injected volumes replaced produced
relative sea-level (Shanley & McCabe 1994) or progressive oil and water volumes), in which average field reservoir
Channelized reservoirs and waterflooding 135

Table 2. Fluid and rock characteristics

Sorwf (residual oil saturation after waterflooding) 0.2


Oil viscosity (cp) 1.95
Swir (irreducible water saturation)
Shales 0.97
Sandstones 0.3
Sandstones with permeability >1500 mD 0.2
Mobility ratio (M) 1
kz/kx 0.05
Pattern size (acres) 110
Correlation length (m) 750
Porosity (mean) 0.25
kx (mean, mD) 500

pressure is maintained. Fluid properties used in the models,


including residual oil saturation (Sorwf), viscosity and mobility
ratio (M), are summarized in Table 2.
Cumulative recovery as a percent of OOIP (recovery
efficiency) was plotted against time in days for each waterflood
simulation run. However, the results could not be compared
easily between different model runs if certain geological or
engineering factors were changed: mean permeability;
permeability thickness at the injection and production wellbores
(which influences injection and production rates); and well
count or well orientation. To make the analysis applicable more
generally, reservoir engineers (for example, Craig 1971; Willhite
1986; Thakur & Satter 1998; Friedmann et al. 2001) typically
evaluate recovery efficiency as a function of pore volume of
water injected (PVI) for each simulation. PVI can be converted
easily into time in days if the reservoir pore volume and Fig. 4. Definitions of static and path-length connectivity. (a) For a
injection rate are known. Plots of recovery efficiency against waterflood, static connectivity is defined as the proportion of the
PVI minimize the effects of differences between models reservoir connected to both production and injection wells. Text on
with regard to mean permeability, permeability thickness at the left describes individual and group well connectivity: for
wellbores, well count or well orientation. example, producer p1 is connected to 20% of the reservoir volume,
whereas producer p2 is connected to 78% of the total reservoir
volume. Note that in this case the connectivity of the reservoirs to
Experimental design: one variable at a time the production or injection wells (both 78%) is different than the
total reservoir connectivity to both production and injection wells
Perhaps the simplest type of experimental design is referred (58%). (b) Path-length connectivity is defined as the proportion of
to as ‘one variable at a time’. One model was constructed the reservoir connected to the wells less than a given path-length
using only P50 (fiftieth percentile, or median) characteristics from both production and injection wells. Path-length connectivity
(Table 1). Then, additional models were constructed by chang- clearly depends on other properties, such as time and oil viscosity,
ing one characteristic to a P10 or P90 value (tenth and ninetieth and thus is more difficult to use.
percentiles) and keeping all other characteristics at the P50
value. P10, P50 and P90 values were determined from geologi- assumes that every cell in the connected reservoir has an equal
cal analysis of field datasets. In this way, 11 models were potential of being produced. Path-length connectivity (Fig. 4b)
constructed for each of the three model net: gross suites (a total assumes that some positions in the reservoir are more likely to
of 33 models). Although this experimental design is conceptu- be swept than others and that unfavourable positions can be
ally simple and allows comparison of the relative significance of discounted. Path-length connectivity also represents a measure
variables, it does not consider the relationship between vari- of reservoir tortuosity. Connectivity studies were compared
ables, and response surfaces are not created. Tornado diagrams with waterflood simulation studies to address the importance of
are used to characterize the results. Table 1 summarizes the reservoir connectivity in affecting recovery efficiency at a given
reservoir characterization variables used in this conceptual time or PVI.
modelling study. Although rock properties are similar and were conditional to
the same parent population, differences in permeability hetero-
geneity in reservoir characterization may occur. To assess such
Causal analysis differences, measures of permeability heterogeneity were made
Previous studies that considered the relationship between in 3D for each of the constructed models. In this case, the
stratigraphic architecture and waterflood recovery efficiency did Dykstra–Parsons coefficient was calculated in 3D, although
not evaluate the cause of differences in waterflood simulation other parameters could have been used (see Jensen et al. 1997
results. Because the same rock property and variogram charac- and subsequent discussion).
teristics are used in each model, the significant difference
between models is the spatial distribution of rock properties, RESULTS
reservoir connectivity and tortuosity. Two types of reservoir
connectivity are considered in this study, static and path-length TM model suite
connectivity. Static connectivity is a measure of the percentage The TM model suite is shown in Figure 5a. Note the significant
of reservoir connected to the wells (Fig. 4a). Static connectivity visual difference in reservoir architecture between the displayed
136 D. K. Larue & F. Friedmann

Fig. 5. The TM model suite. (a) Images of the porosity of the TM models suite at 35% net: gross. Other geological characteristics are discussed
in Table 1. ‘A’ models are characterized by P10 to P90 values of channel width/thickness ratio (10A vs. 90A) with all other characteristics held
constant at P50 values, as defined in Table 1. B, C, D and E models are also defined by using P10 and P90 values shown in Table 1, holding
other properties constant at P50 values. (b) Results of waterflood simulation of the TM suite of models, where recovery efficiency is plotted
against PVI. The total spread of recovery efficiencies is about 5–7% in these models. When measuring the spread in recovery efficiencies,
allowance was made to include spread associated with the TM10E model, which performed most poorly and stopped producing before attaining
1 PVI. (c) Results of waterflood simulation of the TM suite of models, where recovery efficiency is plotted against time in days. Because mean
permeability and permeability distributions are similar, results are comparable. Note that the spread of recovery efficiencies is about 8–10% in
these models, where differences in recovery are probably due to differences in injection and production rates.

models, including relative width/thickness, thickness, sinuosity, to the model so that the resulting net: gross of the reservoir
channel orientation, stacking geometry and the shape of reser- characterizations is 60%. Model characteristics are shown in
voir elements. Waterflood simulation results for the TM model Table 1 and the workflow for constructing the channels was
suite are shown in Figure 5b. In Figures 5b, c, the initial identical to that for the TM suite.
increase in cumulative recovery is linear, followed by a flatten- A waterflood simulation was performed using the same
ing in cumulative recovery subsequent to water breakthrough at workflow as that for the TM models and well locations are
the production wells. Water breakthrough in any production identical. There is a subtle increase of a few percent in recovery
well can be complex, because different sandstone intervals in efficiency at 1 PVI relative to that observed in the TM models
the wellbore may experience different breakthrough times. (from 45–47% in Fig. 6b). There is a more substantial increase
Even though there is a profound variability in the appear- in recovery efficiency at 10 000 days (Fig. 6c) from 42% to 50%
ance of the geological models (Fig. 5a), there are modest because of the increase in permeability thickness in the injec-
differences in recovery efficiency, whether considering time in tion wells that accompanies the higher net: gross of the SM
days (9% variation for the times shown) or the pore volume of models. The spread in recovery efficiency with time or with
water injected (4–7% for the PVI values shown) (Fig. 5a, b). PVI ranges from 5% to 2%, respectively (Fig. 6b, c).
The spread in recovery efficiency for time measured in PVI is
less than that for time measured in days, because the former
units address differences in injectivity; i.e. because there is a EM model suite
variation in permeability thickness at the wellbore of the The EM model suite depicts a sandstone-rich channelized
injection wells and, because this is a voidage replacement reservoir consisting of 85% net: gross (Fig. 7a). The EM model
waterflood simulation, recovery efficiency measured in time in suite was constructed using the same random seed number and
days is a function of injectivity. The spread in recovery geological characteristics as the TM and SM model suites, hence
efficiency can be considered as a function of average recovery channel deposits in the EM models were additive from the TM
efficiency (about 45%), such that the relative spread of the and SM model suites.
recovery efficiency is 20% measured in time or 13% measured A waterflood simulation was performed using the same
in PVI. In the subsequent discussion, only the actual spread of workflow as that for the TM and SM models and well locations
recovery efficiency is discussed at a specified PVI or time in are identical. An increase in the average recovery efficiency and
days. Well locations are shown in Figure 9a, for the nine-pattern further diminishment in the variation in range of recovery
five-spot case at 110-acre spacing. efficiency from that observed in the TM and SM model suites
was expected. At 1 PVI, the average recovery efficiency
increased from 45% for the TM models to 46% for the SM
SM model suite models and to 47% for the EM models. At 10 000 days, the
The SM model suite (Fig. 6a) is identical in all respects to the average recovery efficiency increased from 42% to 50% to 52%
TM model suite, except that channel deposits have been added for the respective model groups. These differences in recovery
Channelized reservoirs and waterflooding 137

Fig. 6. The SM model suite. (a) Images


of porosity in the SM models suite at
60% net: gross. Other geological
characteristics are described in Table 1.
(b) Results of waterflood simulation of
the SM suite of models, where recovery
efficiency is plotted against PVI. The
total spread of recovery efficiencies is
about 2% in these models. (c) Results
of waterflood simulation of the SM
suite of models, where recovery
efficiency is plotted against time in days.
Note that the spread of recovery
efficiencies is about 5–6% in these
models.

Fig. 7. The EM model suite. (a)


Images of porosity in the EM models
suite at 85% net: gross. Other
geological characteristics are described
in Table 1. (a) Results of waterflood
simulation of the EM suite of models,
where recovery efficiency is plotted
against PVI. The total spread of
recovery efficiencies is about 3% in
these models. (b) Results of waterflood
simulation of the EM suite of models,
where recovery efficiency is plotted
against time in days. Note that the
spread of recovery efficiencies is about
3% in these models, where differences
in recovery are probably due to
differences in injection and production
rates.

efficiency at 10 000 days are correlated positively with differ- plotted against PVI. Note that the SM and EM suites clearly
ences in permeability thickness in the wells, which accompany achieve higher recovery efficiencies of about 5% relative to the
the higher net: gross of the SM and EM models. For the PVI TM suite, but there is little difference between the SM and EM
values indicated in Figures 5–7, the TM model suite showed a results.
range in variation of recovery efficiency from 4% to 7%, the Even though 33 models with wide range of net: gross
SM model suite showed a range of recovery efficiency of about were considered here, the total spread in recovery efficiency at
2%, whereas the EM model suite showed a range of recovery 0.5 PVI is only about 5%. All the stratigraphic uncertainty
efficiency of about 3%. The spread of recovery efficiencies for represented by the wide range of geometries in the models did
the EM model suite is, therefore, similar to the spread observed not have disastrous affects on recovery efficiency.
for the SM suite and clearly less than the spread of recovery A tornado diagram is presented in Figure 8b, showing the
efficiencies noted for the TM model suite. possible influence of geological factors on simulated recovery
from waterflooding. Each of the points in the tornado diagram
represents recovery efficiencies at 0.5 PVI from the waterflood
TM, SM and EM model suites: analysis of results simulation, using geological models assigned P10, P50 and P90
Figure 8a shows a comparative plot for the results of water- values (Table 1). Difference values showing the spread of
flooding the three model suites, in which recovery efficiency is recovery efficiency are shown to the right in Figure 8b. Instead
138 D. K. Larue & F. Friedmann
Channelized reservoirs and waterflooding 139

Fig. 8. Summary plots for the TM, SM and EM models. (a) Plot of recovery efficiency against PVI for the three model groups. Note that the
recovery efficiencies of the SM and EM model suites are very similar, but the recovery efficiencies of the TM model suite are slightly less.
(b) Tornado-type plot showing the relationship between geological factors and recovery at 0.5 PVI. The spread of points is associated with the
P10, P50 and P90 runs for each model. The three model suites, TM, SM and EM, are represented. The width of the tornado bar is shown as
differences and represents the differences between the highest and lowest recovery at 0.5 PVI for a given simulation run. Stars (*) indicate results
that are out of sequence: the progression of points is inconsistent from P10 to P50 to P90. Note that in some cases P90 values perform worse
than P10 values: perhaps the original perception of P90 and P10 was in error (for example, thickness in the EM model suite). (c) Tornado-type
plot showing the relationship between geological factors and ranges of static connectivity for the TM, SM and EM models. (d) Tornado-type
plot showing the relationship between geological factors and ranges of path-length connectivity for the TM, SM and EM models.
(e) Tornado-type plot showing the relationship between geological factors and ranges for vertical Dykstra–Parsons coefficient for the TM, SM
and EM models. (f) Tornado-type plot showing the relationship between geological factors and ranges for horizontal Dykstra–Parsons coefficient
for the TM, SM and EM models. (g) Comparison of static connectivity of the TM, SM and EM models with recovery efficiency at 0.5 PVI. Note
that the TM models are connected more poorly, which apparently caused the reduced recovery efficiency. The static discount field is the amount
of reservoir ‘removed’ by static connectivity, i.e. if the recovery should be 40% and the reservoir is connected only 90%, then the discounted
recovery is 36% (90% of 40%). (h) Comparison of path-length connectivity of the TM, SM and EM models with recovery efficiency at 0.5 PVI.
The path-length discount field is defined as being similar to the static discount field, but the actual cut-off value is a function of parameters such
as time, mobility ratio and so on. The path-length discount field is, therefore, defined relatively poorly.

of using the typical tornado plot in which the width of the bars although it is interesting to speculate that the moderate spread
increases upward, the ordering of the variables from Table 1 of recovery efficiency for the thickness factor in the EM model
was maintained for the three net: gross suites to ease compari- suite (Fig. 8b) may be due to horizontal permeability hetero-
son of the properties. Note the general shifting of the bar geneity (Fig. 8f), although this is clearly not the case for the TM
locations from lower to higher recovery efficiencies from the and EM models (Fig. 8f). The definition and description of the
TM to the SM and EM model suites. Interpreting the tornado Dykstra–Parsons coefficient is provided later.
diagram is more difficult. From the spread of recovery effi- In Figure 8g, the static connectivity of each reservoir
ciency, deviation, sinuosity, stacking pattern and channel architecture model is plotted against the recovery efficiency at
width/thickness appear important in the TM models. However, 0.5 PVI. Static connectivity is defined as the volume of rock
it is unclear which of these factors, if any, are statistically connected to both injection and production wells. If the
significant from this plot. Moreover, it is important to note that reservoir is unconnected to both production and injection
the order of the models (from P10 to P50 to P90) is not wells, the waterflood will be ineffective and recovery efficiency
preserved in all cases (sometimes the P90 model recovery lies will be reduced. If the reservoir is only 80% connected then it
between the P10 and P50 cases). Cases in which order is not cannot produce much more than 80% of a 100% connected
preserved are annotated with a star (*) in Figure 8b. Finally, reservoir. The dashed yellow line on Figure 8g shows the
factors that are seemingly important for the TM model suite do theoretical reduction in recovery efficiency associated with a
not seem important for the SM and EM suites. For example, loss in static connectivity, above which is the ‘static discount
thickness has the greatest spread in recovery efficiency for the field’ (recovery discounted by poor connectivity). On Figure 8g,
EM models but the smallest spread for the TM models. the TM model suite results from a poorly defined linear trend
Deviation seems significant for the TM models but not for the that plots generally below the static discount field, indicating
EM models. that recovery efficiency is worse than that predicted by static
Several other tornado diagrams were constructed to assess connectivity. Note that the SM and EM models are essentially
possible factors affecting recovery, using static connectivity 100% connected and form point clouds.
(Fig. 8c), path-length connectivity (Fig. 8d), permeability In a waterflood, all connected volumes cannot be produced.
heterogeneity (Fig. 8e) and horizontal permeability hetero- For example, if parts of the reservoir do not lie between an
geneity (Fig. 8f). Comparing Figure 8b and c, one could injector and a producer they may not be swept effectively. In
conclude that the range in recovery observed in the TM suite Figure 8h, path-length connectivity, a simple measure of
for the deviation, sinuosity and channel stacking patterns could dynamic connectivity, is shown. This provides information
be related to static connectivity. Clearly, static connectivity is about the distance of any connected cell in the reservoir to the
unimportant for the SM and EM models. There is also some distances from both the producer and injector. To define
indication in Figure 8d that path-length connectivity con- path-length connectivity, a distance must be defined beyond
tributed to the worse recovery efficiency of the TM models. which connectivity is ineffective. This critical path-length is a
Permeability heterogeneity (represented by the Dykstra– function of fluid properties and time and is, thus, not as well
Parsons coefficient), whether vertical or horizontal, may not defined as in the case of static connectivity. Based on previous
have contributed much to differences in recovery efficiency, unpublished studies, the cut-off distance selected was 1500 m.
140 D. K. Larue & F. Friedmann

Fig. 9. Comparison of recovery efficiency against PVI for the TM models suite with different well configurations. (a) Images of the different
well configurations, with water saturation after water injection shown. Nine patterns correspond to a well spacing of 110 acres, a four-pattern
five-spot corresponds to a well spacing of 250 acres and a one-pattern five-spot corresponds to a well spacing of 1000 acres. In addition, a
two-well line drive is shown. (b) Comparison of recovery efficiency for the TM models suite with different well configurations against PVI.
Examples of ranges in recoveries are shown to the right (RE). (c) Static connectivity of the models using the different well configurations
plotted against recovery efficiency at 0.5 PVI.

Using this cut-off, a path-length connectivity discount field was efficiency increases slightly with lower well count owing to less
defined in Figure 8h. Note that the TM model suite points interference between wells and the spread in recovery increases
form a crude linear trend and that there is some scatter of significantly with lower well count owing to problems in sweep
SM and EM points. Clearly, much of the difference in efficiency. Note that a comparison between models with
recovery efficiency noted in Figures 5–7 is associated with static different well spacing can be made only because recovery
connectivity and path-length connectivity. efficiency is plotted against dimensionless PVI. Clearly, if
recovery efficiency had been plotted as a function of time in
TM model suite: effect of well count on recovery days, a strong positive correlation would be seen between well
efficiency count and recovery efficiency. The result that all well patterns
attained essentially the same recovery was due to the high
In the TM, SM and EM geological model suites, waterflood connectivity of the reservoir for all well groups (even at 35%
simulations were performed using nine five-spot patterns at net: gross) and limited sweep efficiency problems. To test this
110-acre spacing. Such a well spacing is unlikely in the offshore conclusion, static connectivity is plotted against the recovery
or deep-water environment where costs are prohibitive. A study efficiency at 0.5 PVI. The resulting plot shows a strong linear
was performed to test the effect of well count on recovery relationship between static connectivity and recovery efficiency
efficiency (Fig. 9a). The relationship between well count at 0.5 PVI. Points fall just below the static discount field line,
and recovery efficiency is described commonly as the ‘infill indicating that recovery is affected both by static and dynamic
problem’. Clearly, infill drilling allows a reservoir to be depleted connectivity.
more rapidly. However, does infill drilling also find new
reserves? George & Stiles (1978) argued that new reserves were
discovered with infill drilling. The discovery of new reserves Effect of mobility ratio on recovery efficiency
with infill drilling implies that reservoir connectivity is increased The mobility ratio, which is the ratio of the mobility of the
and that reservoir tortuosity or path length is reduced below displacing phase to the displaced phase (see Fig. 10), affects
some critical level. reservoir performance strongly (Wilhite 1986; Thakur & Satter
The TM model suite was used to define a wide array of 1998). To demonstrate the magnitude of effect of the mobility
reservoir architectures. The field was simulated by voidage ratio, a single geological model was flow simulated using the
replacement as previously described using nine patterns at same geological and engineering characteristics and properties,
110-acre spacing, four patterns at 250-acre spacing, one pattern changing only the mobility ratio (Fig. 10). In this experiment,
at 1000-acre spacing and just two wells in linear arrangement – recovery efficiency changes in the order of 15–20%, indicating
a producer and an injector (Fig. 9a). Waterflood simulation that the mobility ratio influences the results of reservoir
results are plotted in Figure 9b. In general, the average recovery simulation strongly. In Figure 10b, the mobility ratio is plotted
Channelized reservoirs and waterflooding 141

Fig. 10. (a) The effect of mobility ratio


on recovery. The P50 example from the
TM model suite was used to study the
effects of mobility ratio on recovery. As
shown in the figure, the mobility of oil
is defined as the ratio of the viscosity
of oil to the relative permeability of oil,
and the mobility of water is defined as
the ratio of the viscosity of water to the
relative permeability of water. Mobility
ratios of 1 to 10 were simulated in the
waterflood experiment. (b) Recovery
efficiency at 0.5 PVI plotted against the
mobility ratio for the results of the
simulation shown in (a).

against recovery efficiency at 0.5 PVI and the effect of variation An example of an extreme case of permeability heterogeneity
in the mobility ratio can be observed. Figure 10b shows the is shown in Figure 12. It was assumed that a channelized
reduction in recovery efficiency associated with increasing the reservoir contained sandstone characterized by a mean
mobility ratio for the simulation runs. The mobility ratio clearly permeability of 10 mD. However, the presence of a 2000 mD
has a large effect on recovery efficiency. However, unlike many thief zone, which comprised a thin highly permeable gravel lag
geological parameters, such as channel width/thickness or not present in core or well log, was missed. Figure 12 compares
permeability heterogeneity, which are never known precisely, recovery efficiencies for the two scenarios and emphasizes the
the mobility ratio is relatively well understood when fluids from importance of thief zones, which can lead to early water
a new field are studied. breakthrough and reduced recovery efficiencies, in this case
50%.
Effect of permeability heterogeneity on recovery
efficiency
In their classic study of waterflood performance prediction,
Dykstra & Parsons (1950) defined a dimensionless number to
characterize heterogeneity of the permeability population. That
coefficient, now called the Dykstra–Parsons coefficient (Vdp), is
defined as follows:

Vdp = (k50  k84.1)/k50 (1)

where k50 is the median reservoir permeability and k84.1 is the


permeability at the 84.1 percentile, where k84.1 is defined to be
less than k50 and is, therefore, what many would refer to as the
15.9 percentile. Alternatively, there is an abundance of differ-
ent measures of permeability heterogeneity (for example, the
coefficient of variation or the Lorenz coefficient; see Jensen
et al. 1997), but all are relatively comparable to the Vdp. A wide
variety of geological models with a variable Vdp was simulated
from the TM model suite. To do this, one of the geological
models from the TM suite was selected, then the parent
permeability population was modified by adding, for example,
thief zones or expanding the spread of permeability values. This
modified parent population was used for collocated co-kriging
with sequential Gaussian simulation of permeability in the
channel fill, where permeability was collocated with porosity.
For the geological models that were evaluated and the Fig. 11. The effect of permeability heterogeneity on recovery. The
engineering examples used, the Vdp was associated with a range P50 example from the TM model suite was used to study the effects
in recovery efficiencies of about 15% to 25% (Fig. 11). Studies of variation of permeability heterogeneity on recovery. The formu-
by Jensen et al. (1997) indicate that common values of Vdp in lation of the Dykstra–Parsons coefficient is shown at the bottom.
reservoirs are 0.6–0.95, indicating that the wide spread of values Jensen et al. (1997) have shown that in core the Dykstra–Parsons
coefficient typically ranges from 0.6 to 0.95, whereas the range
shown in Figure 11 is extreme. The range in recovery efficien- shown in the figure is from 0 to 0.95. The area outlined in black
cies from reservoirs with Vdp values from 0.6–0.95 is about shows Vdp values from 0.6 to 0.95 and the measurement bars show
15% in Figure 11. this restricted range.
142 D. K. Larue & F. Friedmann

Fig. 12. Example of how permeability


heterogeneity can affect recovery
strongly by waterflooding. Two models
are shown in this extreme case, one of
a channel containing a 10 mD
sandstone and the other containing a
channel with a 10 mD sandstone
together with a unit of 2000 mD gravel.
The 2000 mD gravel unit was not
recognized in core and, therefore, not
included in the model. The
high-permeability thief zone unit causes
earlier water breakthrough and
appreciably lower recovery efficiencies.

Fig. 13. Example of the relationship


between geostatistical seed number and
recovery efficiency by waterflooding.
(a) Nine identical models were
constructed, differing only in
geostatistical seed number, which is a
measure of the positional uncertainty of
the models. This means that the
geological models had similar net: gross,
channel width, sinuosity, direction and
stacking patterns, but the positions of
the channel deposits was allowed to
vary by seed number. (b) Results of
waterflood pore-voidage simulation
study. The difference in recovery
efficiency in all cases was relatively
small.

Effect of model seed number on recovery efficiency with varying the random seed number. The SM, EM and TM
A series of models was constructed that had identical geological models are characterized by a range in recovery efficiency at 0.5
characteristics, such as net: gross (35%), channel width, thick- PVI of 3–6%. The range in recovery efficiencies for the SM and
ness, sinuosity, permeability and porosity architecture (and so EM model suites (39–41% and 39–42% at 0.5 PVI) is con-
on), but differed only in the geostatistical model seed number sidered to be very similar to that of the range in recoveries for
(Fig. 13a). A seed number is a variable that allows a stochastic the random seed suite of models. A larger range in recovery
model to be rebuilt: by knowing the random seed number, efficiency associated with the TM model suite is noted (35–
identical model properties and distributions will be created. 41%). This increase in range is associated with lower minimum
Alternately, by varying the seed number similar stochastic recovery efficiency (35%) relative to the EM and SM models,
geological models are created; only the spatial distribution of but with similar maximum recovery efficiency (about 41%).
characteristics is varied. Varying the geological model seed This decrease in minimum recovery efficiency was shown to be
number can be thought of as testing the effect of spatial associated with poorer reservoir connectivity (Fig. 8g, h). A
uncertainty (that is, all characteristics of the models are very larger range in recovery efficiency (35–46%) associated with
similar, yet the position of the channel deposits is allowed to changing well count in the TM model suite is noted. This range
vary). Because the models here do not use conditioning wells, in recovery efficiency includes both a decrease in minimum
the net: gross at individual wells is allowed to vary from model recovery efficiency and an increase in maximum recovery
to model. For the nine models built and flow simulated for efficiency at 0.5 PVI, compared with the TM model suite
waterflooding, the range in variation of recovery efficiency is produced with wells at 110-acre spacing. Presumably this larger
c. 3% or 4% (Fig. 13b). Theoretically, if millions and millions of range in recovery efficiency is associated with a larger range in
geological realizations had been built and flow simulated the volumetric sweep efficiency associated with the different well
expected spread could be significantly larger. spacing. The largest ranges in recovery efficiency are associated
with ranges in mobility ratio and permeability heterogeneity. In
the cases considered here, in which a light oil with a mobility
DISCUSSION OF RESULTS ratio of 1 represented the dominant oil type, the effect of
A tornado diagram, showing the range in recovery efficiency increasing the mobility ratio on recovery efficiency for the
after 0.5 PVI as a function of the various parameters, is shown model considered is completely negative. Similarly, most of
in Figure 14. A range of 3% in recovery efficiency is associated the models had relatively low permeability heterogeneity. By
Channelized reservoirs and waterflooding 143

Fig. 14. Tornado diagram showing the


results of study. Range of recovery
efficiencies at 0.5 PVI shown by
different categories, and percent
difference relative to lowest recovery
efficiency shown at left (as an example
for permeability heterogeneity:
(4126)/26=58%).

increasing the permeability heterogeneity, the effect on recovery permeability data from core). However, the assembled reservoir
efficiency was negative, with recovery efficiency ranging from model may have a different permeability heterogeneity based
26% to 41%. on how the model is constructed. Differences in reservoir
From this tornado diagram, it can be seen for the model connectivity seem to account for the large range in recovery
simulations discussed here that permeability heterogeneity, efficiencies associated with changes in well count.
mobility ratio and well count have the greatest impact on the Models that appear to be different visually but have similar
range in recovery efficiency. However, well count is a reservoir connectivity, well count and permeability heterogeneity, have
management decision based on economic and other factors, similar sweep efficiencies and, therefore, similar recovery effi-
and mobility ratio, once defined, has a relatively narrow band of ciencies. This observation can be generalized into an initial
uncertainty. Permeability heterogeneity stands out from a geo- hypothesis: if geological models have similar sweep efficiencies,
logical perspective as significant as it has a large impact on
they will perform in a similar fashion where sweep efficiency is
recovery efficiency. However, it is never known precisely,
a function of connectivity, permeability heterogeneity and
because cores represent such a small volume of a reservoir and
permeability does not scale simply. There is very little difference anisotropy, and reservoir path length. This hypothesis has been
in recovery between the sandier model suites (SM and EM embraced by the reservoir engineering community (Thakur &
suites) and the range in recoveries associated with changing the Satter 1998). If this hypothesis is correct, unless changing
model seed number (see also, Jones et al. 1994). reservoir architecture changes volumetric sweep efficiency,
such as connectivity and permeability heterogeneity, differ-
ent reservoir architectures do not change recovery efficiency
CONCLUSIONS significantly. Note that this is only from a field-wide recovery
The goal of this study was to investigate how differences in efficiency viewpoint: reservoir architectural characteristics
stratigraphic architecture could affect recovery efficiency. For may affect other aspects of reservoir management in terms
the 33 examples of reservoir architecture described here in the of perceptions about sandstone distribution, local field
TM, SM and EM model suites (Figs 5–7), the range in recovery performance, the volume of reservoir and the positioning of
efficiencies at any given PVI is not large when compared with wells.
the differences cited by a number of other workers for field The main practical conclusion from this study is that the
studies (Dromgoole & Speers 1997; Dromgoole et al. 2000; current trend toward creating more complex and diverse types
Thomas 1998, Larue & Yue 2003) or simulation studies of geostatistical facies models should be modified to stress that
(Thakur & Satter 1998; Fig. 2). In fact, varying model seed better models need more refined understanding of sweep
number (Fig. 13), which only affects the spatial distribution of efficiency. A more realistic model of a fluvial point bar may not
reservoir elements, has as much effect on recovery efficiency as be any more predictive than an unrealistic one, unless the sweep
changing reservoir architecture in sandy (_60% net: gross) efficiency is modelled better (for example, Li & White 2003).
reservoirs (Figs 6, 7). In the TM reservoir models, modest Measures to quantify sweep efficiency are also needed (see
differences in static and path-length connectivity have measur- examples in Larue & Legarre 2004, or in the present paper). In
able and probable economic impact on recovery efficiency general, stratigraphic insight into the origins and causes
(Fig. 8g, h). Well count, mobility ratio and permeability hetero- of sweep efficiency are needed for creating better reservoir
geneity have significant impact on recovery efficiency (Figs 9, characterization.
10, 11). As noted previously, mobility ratio and well count are
not stratigraphic characteristics of the reservoir, but were Tom Tran, of ChevronTexaco Exploration and Production
included to compare their effect on recovery. Technology Company, modified the Boolean modelling program to
Two stratigraphic factors that are clearly shown to influence capture thief zone geometries more accurately. Alan Reed, also of
recovery efficiency – reservoir connectivity and permeability CTEPTC, helped assign properties to the fluvial and estuarine
channel deposits. Kaveh Dehghani was instrumental in setting up the
heterogeneity – both influence volumetric sweep efficiency flow simulation decks and explaining how to complete the flow
(Craig 1971; Willhite 1986; Thakur & Satter 1998). Reservoir simulations and interpret the results. Danny Horowitz and Ed
connectivity and permeability heterogeneity characteristics are Wood, formerly of Exxon Production Research, provided early
derivative properties that are not modelled directly. Connec- inspiration for these studies. Discussions with June Gidman, Chris
tivity and permeability heterogeneity are measured on the Harrison, Larry Lake, Andrew Latham, Marge Levy, Julian Thorne,
assembled model. Only general guidelines exist for building a Tim McHargue, Ron Behrens, Connie Terricola, Jeff Warner and
Frank Harris were important for the progress of this study. The
well-connected or a poorly connected model (King 1990). connectivity programs were developed with Joseph Hovadik and
Similarly, permeability heterogeneity is a function of the parent Jerome Amiotte: figures from a report by Joseph Hovadik are shown
population used to construct the reservoir model (for example, in Figure 4a, b.
144 D. K. Larue & F. Friedmann

APPENDIX A: STRATIGRAPHIC FEATURES OF than the width to thickness ratio of deep-water channels
DEEP-WATER AND FLUVIAL CHANNEL (typically about 20: Clark & Pickering 1996). Some channel
ARCHITECTURE deposits in the deep-water system may be filled entirely with
Deep-water and fluvial channel architecture is reviewed here at debris-flow and/or slump deposits (Clark & Pickering 1996).
four scales: the bed, bar, channel and channel complex scales This is probably much rarer in the fluvial environment.
(Table A1). Key similarities between deep-water and fluvial Mudstone- or siltstone-drapes on erosional channel walls de-
channels are mostly at the channel scale and include gross posited from bypassing sediment gravity flows are present in
geometry, overall shape and sinuous form. Channel-related deep-water channel deposits, but have not been identified in
facies, such as overbank splays and levees are present in both fluvial deposits (Zelt & Rossen 1995;Beauboeuf et al. 1999).
fluvial and deep-water slope channel deposits. Key differences Deep-water channel deposits may show evidence of reworking
that have been cited between fluvial and deep-water channel by bottom currents, a feature certainly absent in subaerial
deposits are as follows. First, and most obviously, at the bed deposits (Posamentier et al. 2000). At the channel complex
scale the dominant transport and depositional mechanism in scale, deep-water channel deposits show a greater abundance of
deep-water is sediment gravity flow, whereas tractional deposi- aggradational stacking patterns than fluvial deposits (McHargue
tion dominates in fluvial deposits (Walker 1978; Friend et al. 1991; Peakall et al. 2000; Kolla et al. 2001), though aggradational
1979; Clark & Pickering 1996). These different depositional fluvial deposits have been described (for example, Rust &
mechanisms can lead to differences in bedding and permeability Legun 1983). Aggradational stacking of channels favours the
continuity as well as layering architecture. At the bar scale, there formation of multi-storey channels as opposed to multi-lateral
is a paucity of barforms in deep-water channel deposits relative channels (in the sense of Friend et al. 1979). Also at the channel
to fluviatile channel deposits. Barforms in deep-water deposits complex scale, there is a tendency for deep-water channel
are described only rarely (for example, Clark & Pickering 1996; deposits to stack in a compensatory fashion, as opposed to the
Sullivan et al. 2000; Abreu et al. 2003). Fluvial barforms ‘random’ stacking of channel deposits modelled commonly in
consisting of inclined heterolithic stratification may decrease the fluvial environment (McHargue & Webb 1986; Kolla &
horizontal continuity significantly (Thomas et al. 1987). At the Coumes 1987). Compensatory stacking occurs when younger
channel scale, there tends to be a lack of evidence for sweeping channels avoid older channel deposits, presumably due to
lateral accretion in deep-water channel deposits as is seen depositional relief associated with older deposits.
commonly in meandering fluvial channel deposits (Stelting et al. A summary of characteristics at the bedset, bar, channel and
1986), although lateral accretion deposits have been demon- channel complex scale is presented in Table A1, with a brief
strated in deep-water (Posamentier et al. 2000; Kolla et al. 2001; description of how these features were treated in the conceptual
Abreu et al. 2003). The width to thickness ratio of fluvial models developed. Only a subset of the results is presented in
channels (typically about 50: Fielding & Crane 1987) is greater this paper.

Table A1. Stratigraphic features that may affect reservoir flow performance

Description Feature Modelling technique


Bed and bedset variable rock properties use several permeability scenarios
variable length scales use several continuity lengths
onlap versus drape geometry onlap modelled for sandstone-filled channels, mudstone drapes
tested for some models
thief zones (high permeability layers between lower permeability layers) modelled by modifying permeability function
mudstone beds, fining-upward character modelled using several fining-upward trends within individual
channel deposits
variable vertical continuity modelled by changing the vertical to horizontal permeability
ratio (kz/kh)
Bar and barform point bars modelled using special in-house facies modelling option
bar continuity length modelled rock properties using bar length scale based on in-
house database studies
Channel inclined beds within bar deposit not modelled (not easily done with current software)
isolated sinuous and straight channels modelled using standard Boolean software program
sinuous and straight channels with associated facies modelled using standard Boolean software program
braided channels modelled through intersection of channel bodies
variable channel orientation modelled parallel and diversely orientated channel bodies
variable channel width and thickness modelled using standard Boolean software: values constrained
by database
Channel complex presence of local mudstone-filled channels populations of channels were assigned mudstone values based
on stochastic sampling
random distributions of channels form channel complex modelled using standard Boolean software program
clustering of channels, with fewer vertically (sequence stratigraphic modelled using standard Boolean software program with trend
stacking)
channel clustering, channel belts or channel families modelled through intersection of channel bodies (not modelled
specifically)
multi-lateral stacking modelled through intersection of channel bodies (not modelled
specifically)
multi-storey (vertical) stacking not specifically addressed (not easily done with available soft-
ware)
valley architecture modelled through construction of channels within a valley-
shaped container
Channelized reservoirs and waterflooding 145

REFERENCES net-to-gross fluvial reservoir. In: de Haan, H.J. (ed.) New Developments in
Improved Oil Recovery. Geological Society, London, Special Publications, 84,
5–18.
Abreu, V., Sullivan, M., Pirmez, C. & Mohrig, D. 2003. Lateral accretion King, P.R. 1990. The connectivity and conductivity of overlapping sand
packages (LAPs): an important reservoir element in deep water sinuous bodies. In: Buller, A.T. (ed.) North Sea Oil and Gas Reservoirs II. The
channels. Marine and Petroleum Geology, 20, 631–648. Norwegian Institute of Technology, Graham & Trotman, London,
Barber, A.H. Jr, George, C.J., Stiles, L.H. & Thompson, B.B. 1983. Infill 353–362.
drilling to increase reserves – actual experience in Nine Fields in Texas, Kolla, V. & Coumes, F. 1987. Morphology, internal structure, seismic
Oklahoma and Illinois. Journal of Petroleum Technology, 35, 1530–1538. stratigraphy and sedimentation of Indus Fan. American Association of
Beauboeuf, R.T., Rossen, C., Sullivan, M.D., Mohrig, D.C. & Jennette, D.C. Petroleum Geologists Bulletin, 71, 650–677.
1999. Deep-water sandstones, Brushy Canyon Formation, West Texas. Kolla, V., Bourges, P., Urruty, J.M. & Safa, P. 2001. Evolution of Deep-water
American Association of Petroleum Geologists, Hedberg Field Research Conference. Tertiary Sinuous Channels Offshore Angola (West Africa) and Implica-
AAPG, 68. tions for Reservoir Architecture. American Association of Petroleum Geologists
Brandsæter, I., Wist, H.T., Naess, A. et al. 2000. Ranking of stochastic Bulletin, 85 (8), 1373–1405.
realizations of complex tidal reservoirs using streamline simulation criteria. Larue, D.K. & Friedmann, F. 2000. The Relationship Between Channelized
Petroleum Geoscience, 7, 853–863. Deep-Water Reservoir Architecture and Recovery from Petroleum
Bratvold, R. B., Holden, L., Svanes, T. & Tyler, K. 1994. Storm: Integrated Reservoirs. In: Weimer, P. (ed.) Deep-water reservoirs of the world. Gulf Coast
3D Stochastic Reservoir Modeling Tool for Geologists and Reservoir Section Society of Economic Paleontologists and Mineralogists 20th
Engineers. Paper SPE 27563, presented at the SPE 2nd European Annual Research Conference, 469–472.
Petroleum Computer Conference, Aberdeen, 15–17 March. Larue, D.K. & Legarre, H. 2004. flow units, connectivity and reservoir
Bridge, J.S. & Tye, R.S. 2000. Interpreting the dimensions of ancient fluvial characterization in a wave-dominated deltaic reservoir: Meren Reservoir,
channel bars, channels, and channel belts from wireline-logs and cores. Nigeria. American Association of Petroleum Geologists Bulletin, 88, 303–324.
American Association of Petroleum Geologists Bulletin, 84, 1205–1228. Larue, D.K. & Yue, Y. 2003. How stratigraphy influences oil recovery: a
Bridge, J.S. 1993. Description and interpretation of fluvial deposits: a critical comparative reservoir database study concentrating on deepwater
perspective. Sedimentology, 40, 801–810. reservoirs. The Leading Edge, 22, 332–339.
Clark, J.D. & Pickering, K.T. 1996. Submarine Channels, processes and architecture. Li, H. & White, C.D. 2003. Geostatistical models for shales in distributary
Vallis Press, London. channel point bars (Ferron Sandstone, Utah) from ground-penetrating
Clark, J.D., Kenyon, N.H. & Pickering, K.T. 1992. Quantitative analysis of radar data to three-dimensional flow modeling. American Association of
the geometry of submarine channels: implications for the classification of Petroleum Geologists Bulletin, 87, 1851–1868.
submarine fans. Geology, 20, 633–636. McHargue, T.R. & Webb, J.E. 1986. Internal geometry, seismic facies, and
Cook, G., Chawathe, A., Larue, D., Legarre, H. & Ajayi, E. 1999. Sequence petroleum potential of canyons and inner fan channels of the Indus
stratigraphy in reservoir simulation. Journal of Petroleum Technology, 51, 60–61. Submarine Fan. American Association of Petroleum Geologists Bulletin, 70,
Craig, F.F. Jr 1971. The Reservoir Engineering Aspects of Waterflooding. SPE 161–180.
Monograph, 3. Richardson, TX. McHargue, T.R. 1991. Seismic facies, processes and evolution of Miocene
De Rooij, M., Corbett, P.W.M. & Barens, L. 2002. Point bar geometries, inner fan channels, Indus submarine fan. In: Weimer, P. & Link, M.H. (eds)
connectivity and well test signatures. First Break, 20, 755–763. Seismic facies and sedimentary processes of submarine fans and turbidite systems.
Dromgoole, P. & Speers, R. 1997. Geoscore: a method for quantifying Springer-Verlag, New York, 403–414.
uncertainty in field reserve estimates. Petroleum Geoscience, 3, 1–12. Peakall, J., McCaffrey, B. & Kneller, B. 2000. A process model for the
Dromgoole, P., Bowman, M., Leonard, A., Weimer, P. & Slatt, R.M. 2000. evolution, morphology and architecture of sinuous submarine channels.
Developing and managing turbidite reservoirs – case histories and experi- Journal of Sedimentary Research, 70, 434–448.
ences: results of the 1998 EAGE/AAPG Research Conference. Petroleum Posamentier, H.W., Meizarwin, Wisman, P.S. & Plawman, T. 2000. Deep
Geoscience, 6, 97–105. Water Depositional Systems–Ultra-Deep Makassar Strait, Indonesia. In:
Dykstra, H. & Parsons, R.L. 1950. The prediction of oil recovery by Weimer, P. (ed.) Deep-water reservoirs of the world. Gulf Coast Section Society
waterflood. Secondary Recovery of Oil in the United States (2nd edn). API, New of Economic Paleontologists and Mineralogists 20th Annual Research
York, 160–174. Conference, 806–816.
Fielding, C.R. & Crane, R.C. 1987. An application of statistical modeling to Prather, B.E. & Keller, F.B. 2000. Hierarchy of Deep-Water Architectural
the prediction of hydrocarbon recovery factors in fluvial reservoir Elements with Reference to Seismic Resolution: Implications for Reservoir
sequences. In: Ethridge, F.G., Flores, R.M. & Harvey, M.D. (eds) Recent Prediction and Modeling. In: Weimer, P. (ed.) Deep-water reservoirs of the world.
developments in fluvial sedimentology. SEPM Special Publication, 39, 321–327. Gulf Coast Section Society of Economic Paleontologists and Mineralogists
Friedmann, F., Chawathe, A. & Larue, D. K. 2001. Assessing Uncertainty in 20th Annual Research Conference, 817–835.
Channelized Reservoirs Using Experimental Designs. Paper SPE 71622, Rust, B.R. & Legun, A.S. 1983. Modern anastomosing-fluvial deposits in arid
presented at the SPE Annual Technical Conference and Exhibition, New Central Australia, and a Carboniferous analogue in New Brunswick,
Orleans, LA, 30 September–3 October. Canada. In: Collinson, J.D. & Lewin, J. (eds) Modern and ancient fluvial systems.
Friend, P.F., Slater, M.J. & Williams, R.C. 1979. Vertical and lateral building International Association of Sedimentologists, Special Publication, 6,
of river sandstone bodies, Ebro Basin, Spain. Journal of the Geological Society, 385–392.
London, 1367, 39–46. Shanley, K.W. & McCabe, P.J. 1994. Perspectives on the sequence stratigra-
Galloway, W.E., Ewing, T.E., Garrett, C.M. Jr, Tyler, N. & Bebout, D.G. phy of continental strata. American Association of Petroleum Geologists Bulletin,
1983. Atlas of major Texas oil reservoirs. Bureau of Economic Geology, Special 78, 544–568.
Publication. The University of Texas at Austin. Stelting, C.E., Droz, L., Bouma, A.H. et al. 1986. Late Pleistocene seismic
George, C.J. & Stiles, L.H. 1978. Improved Techniques for Evaluating stratigraphy of the Mississippi Fan. In: Bouma, A.H., Coleman, J.M.,
Carbonate Waterfloods in West Texas. Journal of Petroleum Technology, 30, Meyer, A.W. et al. (eds) Initial Reports of the Deep Sea Drilling Project, 96. US
1547–1554. Government Printing Office, Washington D.C, 563–576.
Gould, T.L. & Sarem, A.M. 1989. Infill Drilling for Incremental Recovery. Stiles, L.H. & Magruder, J.B. 1992. Reservoir Management in the Means San
Journal of Petroleum Technology(SPE Paper 18941), 229–237. Andres Unit. Journal of Petroleum Technology(SPE Paper 20751), 469–475.
Henriquez, A., Tyler, K. J. & Hurst, A. 1988. Characterization of Fluvial Sullivan, M., Jensen, G., Goulding, F., Jennette, D., Foreman, L. & Stern, D.
Sedimentology for Reservoir Simulation Modeling. Paper SPE 18323, 2000. Architectural Analysis of deep-water outcrops: implications for
presented at the SPE 63rd Annual Technical Conference and Exhibition, exploration and development of the Diana Sub-basin, Western Gulf of
Houston, TX, 2–5 October. Mexico. In: Weimer, P. (ed.) Deep-water reservoirs of the world. Gulf Coast
Hoimyr, O., Kleppe, A. & Nystuen, J.P. 1993. Effects of heterogeneities in a Section Society of Economic Paleontologists and Mineralogists 20th
braided stream channel sandbody on the simulation of oil recovery: a case Annual Research Conference, 1010–1031.
study from the Lower Jurassic Statfjord Formation, Snorre Field, North Swanson, D.C. 1993. The Importance of Fluvial Processes and Related
Sea. In: Ashton, M. (ed.) Advances in Reservoir Geology. Geological Society, Reservoir Deposits. Journal of Petroleum Technology, 45, 368–377.
London, Special Publications, 69, 105–134. Thakur, G.C. & Satter, A. 1998. Integrated waterflood asset management. Pennwell,
Jensen, J.L., Lake, L.W., Corbett, P.W.M. & Goggin, D.J. 1997. Statistics for Tulsa, Oklahoma.
Petroleum Engineers and Geoscientists. Prentice Hall, NJ. Thomas, J.M. 1998. Estimation of ultimate recovery for UK oil fields: the
Jones, A.D.W., Verly, G.W. & Williams, J.K. 1994. What reservoir charac- results of the DTI questionnaire and a historical analysis. Petroleum
terization is required for predicting waterflood performance in a high Geoscience, 4, 157–163.
net-to-gross fluvial environment?. In: Aasen, J.O., Berg, E. et al. (eds) North Thomas, R.G., Smith, D.G., Wood, J.M., Visser, J., Caverley-Range, E.A. &
Sea Oil and Gas Reservoirs, III. Kluwer, Dordrecht, 223–232. Koster, E.H. 1987. Inclined heterolithic stratification – terminology,
Jones, A., Doyle, J., Jacobsen, T. & Kjonsvik, D. 1995. Which sub-seismic description, interpretation and significance. Sedimentary Geology, 53,
heterogeneities influence waterflood performance? A case study of a low 123–179.
146 D. K. Larue & F. Friedmann

Tyler, N. & Finley, R.J. 1991. Architectural controls on the recovery of Walker, R.G. 1978. Deep-water sandstone facies and ancient submarine fans:
hydrocarbons from sandstone reservoirs. In: Miall, A.D. & Tyler, N. (eds) models for exploration for stratigraphic traps. American Association of
The three-dimensional facies architecture of terrigenous clastic sediments and its Petroleum Geologists Bulletin, 62, 932–966.
implications for hydrocarbon discovery and recovery. SEPM Concept in Sedimen- Weber, K.J. & van Geuns, L.C. 1990. Framework for Constructing Clastic
tology and Paleontology, 3, 1–5. Reservoir Simulation Models. Journal of Petroleum Technology, 42, 1248–1297.
Tyler, N., Galloway, W.E., Garrett, C.M. Jr & Ewing, T.E. 1984. Oil Willhite, G.P. 1986. Waterflooding. Society of Petroleum Engineers Textbook,
accumulations, production characteristics and targets for additional recovery in major oil Series 3. Richardson, Texas.
reservoirs of Texas. Geological Circular, Bureau of Economic Geology, The Zelt, F.G. & Rossen, C. 1995. Geometry and continuity of deep water
University of Texas at Austin, 84-2. sandstones and siltstones, Brushy Canyon Formation (Permian) Delaware
Van Wagoner, J.C. 1995. Sequence stratigraphy and marine to nonmarine Mountains, Texas. In: Pickering, K.T., Hiscott, R.N., Kenyon, N.H., Ricci
facies architecture of foreland basin strata, Book Cliffs, Utah, U.S.A. In: Lucci, F. & Smith, R.D.A. (eds) Atlas of deep water environments: Architectural
Van Wagoner, J.C. & Bertram, G.T. (eds) Sequence Stratigraphy of Foreland style in turbidite systems. Chapman & Hall, London, 89–93.
Basin Deposits. American Association of Petroleum Geologists Memoir, 64,
137–224.

Received 2 April 2004; revised typescript accepted 19 January 2005.

You might also like