You are on page 1of 6

A Critical Review of the Use of

Pseudorelative Permeabilities
for Upscaling
John W. Barker, SPE, and Sylvain Thibeau, Elf Aquitaine Production

Summary Pseudorelative Permeability Generation. The role of pseudorela-


The properties and limitations of different dynamic pseudorelative tive permeabilities is to determine the flow rate of each phase out of
permeability methods are summarized. Severe difficulties common a gridblock. They relate the flow rate to the pressure gradient be-
to all methods are discussed: choosing the number and locations of tween the gridblock and its neighbor, given the average saturation
the coarse-grid rock types, defining the simulations from which the in the gridblock (assuming that single-point upstream weighting is
pseudos are generated, and the dependence of the pseudos on well used). Both the flow rate and the pressure gradient depend on the de-
rates and positions. It is concluded that, in practice, pseudos cannot tails of the saturation distribution within the gridblock. Thus, to
be used reliably to scale up from a “fine-grid” geological model to a compute a pseudorelative permeability curve, it is necessary to de-
“coarse-grid” fluid-flow model except for cases where capillary or termine the saturation distribution within the gridblock for any giv-
gravity equilibrium can be assumed at the coarse-gridblock scale. en average saturation.
Scaling up from the core scale to the geological model is more likely When the flow is dominated by gravity or capillary forces or a
to be possible because capillary forces are more important at smaller combination of the two, the saturation distribution may be deter-
scales. A practical approach to the dynamic upscaling problem is out- mined by assuming capillary/gravity equilibrium. For the case
lined, but one should not expect that the effects of the detail in the geo- where the phases are locally segregated under gravity, “vertical-
logical model will be captured more than qualitatively. equilibrium” pseudos may be calculated.4,5 For the capillary-domi-
nated case, pseudos may be calculated by scaling up the permeabil-
ity of each phase (kk r ) and dividing by the scaled-up value of the
Introduction
absolute permeability (k).6 Each of these quantities can be scaled up
Multiphase fluid-flow simulations of oil reservoirs are computation- by solution of a Laplace equation with periodic lateral boundary
ally very intensive. With currently available computers, most oil com- conditions7; however, the dependence of the results on the boundary
panies cannot afford to run routine fluid-flow simulations with more conditions and the tensorial nature of the scaled-up quantities8
than about 105 gridblocks. This implies an average gridblock size on should be recognized. In some cases, the imposition of local bound-
the order of 100 m areally and perhaps 1 to 10 m vertically. Each grid- ary conditions may introduce unacceptable errors; but, if care is tak-
block thus represents a part of the reservoir that is heterogeneous. His- en, it should be possible to obtain reasonably reliable results with
torically, little information on the structure of the reservoir at this scale this “capillary-equilibrium” method in most cases.
was available, so there was little motivation to do anything but ignore In many reservoirs, however, viscous forces cannot be neglected.
the heterogeneity not explicitly represented in the model. In these cases, the saturation distribution is not uniquely determined
Nowadays, modern reservoir imaging techniques and advances by the average saturation but also depends on the production history
in geological modeling are providing more detailed reservoir de- [i.e., the positions of the wells and their rates (mathematically, on the
scriptions. Geological reservoir models are being built with up to boundary conditions)]. One way to determine the details of the satu-
107 gridblocks (the size of these models is also limited by computa- ration distribution for a specific production history is by running a
tional constraints). These models may be used directly for fluid-in- simulation with either a fine grid or a dual-scale grid.9 Several meth-
place and connectivity calculations with little computational diffi- ods for calculating pseudorelative permeabilities from the results of
culty. Single-phase flows, such as well tests or depletion of dry gas such a simulation have been proposed; some of these are discussed
reservoirs with no aquifer influx, may possibly be simulated on in the next section. Pseudorelative permeabilities calculated from
grids of this size. But for multiphase flows, which occur in the ma- the results of a simulation are generically known as dynamic
jority of hydrocarbon reservoirs, the detailed geological informa- pseudorelative permeabilities.
tion must be incorporated into a coarser, fluid-flow simulation mod-
el by means of some upscaling technique. Dynamic Pseudorelative Permeability Methods
Is It Enough To Scale Up the Absolute Permeability? It is some- In this section, we consider the ability of various dynamic pseudore-
times suggested that scaling up of absolute permeability alone is lative permeability methods to reproduce the results of a given fine-
enough to capture the effects of heterogeneity on two-phase fluid- grid simulation on a coarse grid. Implicit in our discussion is the as-
flow simulations. Previous publications1,2 and our own unpub- sumption that a different set of pseudos will be generated for each
lished work, however, both suggest that this is not true if the correla- coarse gridblock and each flow direction, using the results of the
tion length of the heterogeneity not represented on the fluid-flow given fine-grid simulation. Of course, this is not what would happen
simulation grid is significant compared with the well spacing. This in practice, but it is useful for understanding the properties of the
often happens when long, thin, high-permeability channels; thin, various methods. The relevant references give details of the various
high-permeability layers; or extensive, but thin, shale barriers are methods discussed. Ref. 10 gives a full analysis of the properties of
present in the reservoir. the various methods, which are merely summarized here.
In these cases, a multiphase upscaling technique is generally re-
quired. The most obvious of these is the use of pseudorelative per- Kyte and Berry11 and Similar Methods. The Kyte and Berry
meabilities (pseudos), which has been suggested by many authors.3 method is very well-known and widely used, but is also widely be-
In this paper, we take a critical look at the suitability and reliability lieved to be unreliable, although there is little published evidence
of various pseudorelative permeability methods for use in scaling up for this. A similar method is the “pore-volume-weighted” method,12
from a fine-grid geological model to a coarse-grid fluid-flow model. which differs from Kyte and Berry only in the use of a different for-
mula to determine average pressures.
Copyright 1997 Society of Petroleum Engineers
In both these methods, average pressures for each coarse grid-
block and total flow rates of each phase between each pair of adja-
Original SPE manuscript received for review 14 May 1996. Revised manuscript received 17
December 1996. Paper peer approved 4 March 1997. Paper (SPE 35491) first presented at
cent coarse gridblocks are calculated from the fine-grid simulation
the 1996 European 3D Reservoir Modeling Conference, Stavanger, 16–19 April. results. These values are substituted in the coarse-grid Darcy equa-

138 SPE Reservoir Engineering, May 1997


tions to infer the coarse-grid (pseudo) relative permeability values from the fine-grid simulation rather than being calculated assuming
that would be required to reproduce the fine-grid flows. capillary equilibrium.
From the way the pseudos are constructed, it is clear that both It can be shown that scaling up (averaging) of phase permeability
these methods should, in principle, allow the fine-grid solution to be is valid only in the quasisteady-state case.10 Implicit in the method
reproduced exactly on a coarse grid.13 Thus, these are in fact “ideal” is the neglect of the time derivative of saturation, ēS/ēt. In the ab-
methods that give the “correct” pseudos for any particular case. sence of capillary pressure, this time derivative is infinite at a satura-
However, in practice, many things can go wrong: it is possible for tion front and cannot be neglected. The method can thus fail very
the net flow of a phase to be in the opposite direction to the average badly in viscous-dominated cases, which generally involve dis-
pressure gradient, leading to a negative pseudorelative permeabil- placements of a frontal nature. Only when the saturation fronts are
ity; there can be a nonzero net flow when the average pressure gradi- smoothed out by capillary forces is the method reliable. An attrac-
ent is zero, leading to an infinite pseudorelative permeability; and tion of the method is that the pseudos are generally smooth curves
the same average saturation may occur more than once in a given with values between zero and one, so there are no problems with
coarse gridblock, leading to a multivalued pseudorelative perme- their actual use in the coarse-grid simulation. This, however, does
ability function (if this occurs, it is a difficulty for any method). not justify the method’s use outside its range of validity.
Thus, the pseudos generated by these methods may be unusable.
A further problem with the Kyte and Berry method is that the defi- Weighted-Relative-Permeability Method. This is a widely used
nition of average pressure differs according to whether one is consid- method, for two reasons. First, it is one of the methods available in
ering x-, y-, or z-direction flow. Thus, in multidimensional problems, PSEUDO,12 the upscaling program sold with GeoQuest’s popular
one may have two or three different pseudocapillary pressures (even ECLIPSE reservoir simulation software. Second, it generally gives
if there is no capillary pressure in the fine-grid problem). The pore- smooth pseudos with values between zero and one. The pseudorela-
volume-weighted method has only one pseudocapillary pressure, and tive permeability of a phase is obtained simply as an average of the
this is zero when there is no fine-grid capillary pressure. Thus, this relative permeability values for that phase in certain blocks in the
method is preferable to the original Kyte and Berry method. fine grid. Despite its popularity, the method has little justification
because it is consistent with the coarse-grid Darcy equations only
Stone’s14 Method. Stone was the first to use total mobility as a way under very restrictive conditions and it neglects the coarse-grid
of avoiding the problems associated with estimating the average gravity term. Thus, there is no guarantee that it will reproduce the
pressures in the Kyte and Berry method. He suggested computing fine-grid solution on the coarse grid.
an average total mobility and a net fractional flow. The pseudorela-
tive permeabilities can be calculated easily from these two quanti- Other Methods. Several other dynamic pseudorelative permeabil-
ties if gravity and capillary pressure are neglected, as they were by ity methods have been proposed,21-23 but, in our opinion, these do
Stone, but neglect of gravity imposes a severe limitation on the use not offer anything better than the methods we have just discussed
of the method. Also, Stone’s formula for average total mobility is (see Ref. 10).
Obtaining pseudos by history matching24,25 (i.e., by adjusting
inadequate when there are significant variations in total mobility.15
them until the coarse-grid results match the fine-grid results) is diffi-
Thus, in cases with significant gravity (or capillary pressure) effects
cult to do when many sets of pseudos are used on the coarse grid.
or with significant variations in total mobility, the method can give
With such pseudos, the fine-grid solution is reproduced at the wells
poor results.
but not necessarily for each block (unless interblock flows and aver-
age block pressures are also history matched, which makes the
Other Total Mobility Methods. Methods similar to Stone’s but that
matching even more difficult). The pseudos are thus less likely to be
use better methods to find the average total mobility have been pro-
valid if the well rates or positions are changed than are pseudos ob-
posed by several authors.15-17 The best approach is probably by tained by one of the constructive methods described previously.
solution of a Laplace equation.15 Coarse-grid gravity effects can be Nonetheless, such a step may have to be performed as a final tuning
allowed for if the total flow rate is also obtained from the fine-grid of pseudos generated by the quasisteady-state or weighted-relative-
results17 (solution of a quadratic equation for the pseudorelative permeability methods.
permeabilities is necessary). Authors from Heriot-Watt U.8 have advocated relating the scal-
The resulting method, which we call the total-mobility method, ing-up step more closely to the geological structures present in the
is more robust than the Kyte and Berry or pore-volume-weighted reservoir and also the use of tensorial pseudorelative permeabili-
method in the sense that infinite values can be avoided and negative ties.8 However, the techniques they use for calculating the pseudos
values occur less frequently (but can still occur if the net flows of are essentially the same as the methods already described.
the two phases are in opposing directions). However, in solving the Finally, in renormalization-based methods,17 the key idea is to
Laplace equation, the gravity term is neglected and local boundary perform scaling up in a series of steps rather than in a single step.
conditions are necessarily imposed. These two approximations, Thus, a model is scaled up from a fine grid to a slightly coarser grid,
while generally not severe, mean that the fine-grid solution will not then again to an even coarser grid, and so on until the desired coarse
be exactly reproduced on the coarse grid. There is no point in trying grid is reached. At each scaling-up step, the methods used to calcu-
to “improve” the method by eliminating these two assumptions be- late the pseudos needed for the next coarser grid are again the same
cause this would lead back to the pore-volume-weighted pseudos, as those described previously. Note that the repeated scaling up of
which, while guaranteeing reproduction of the fine-grid results in relative permeabilities compounds the difficulties described later,
principle, are often unusable in practice. which may limit the applicability of the renormalization method in
The attraction of the total-mobility method is that it may be useful this context.
in some cases where the pore-volume-weighted pseudos are unus-
able. While it guarantees neither reproduction of the fine-grid re- Summary. In summary, all the methods have their drawbacks: the
sults nor usable pseudos, it makes a minimum of approximation and ideal method (pore-volume-weighted) in theory allows a fine-grid re-
is thus a compromise between rigorous, but unreliable, and robust, sult to be reproduced on a coarse grid, but in practice often produces
but approximate, methods. pseudos that are unusable. Other methods (quasisteady-state,
weighted-relative-permeability) usually produce pseudos that vary
Quasisteady-State Method. This method, in which the permeabil- smoothly between zero and one but that do not reproduce the fine-grid
ity (kk r ) of each phase is scaled up (by solution of a Laplace equa- result. The total-mobility method is a compromise that should
tion, for example), has appeared in several guises.18-20 The pseudo- approximate the fine-grid result more accurately than the quasistea-
relative permeabilities are obtained by dividing the scaled-up value dy-state or weighted-relative-permeability method and that is more
of kk r by the scaled-up value of absolute permeability (k); i.e., the likely to produce usable pseudos than the pore-volume-weighted
method is the same as the capillary-equilibrium method described method, but which guarantees neither. The main message of this pa-
in the Introduction except that the saturation distribution is taken per, however, is that whichever method is used (even if there were a

SPE Reservoir Engineering, May 1997 139


perfect method with no drawbacks), there are serious practical diffi-
culties that prevent the use of dynamic pseudos for scaling up reliably
from a fine-grid geological model to a coarse-grid fluid-flow model.

Practical Difficulties
We have discussed the ability of dynamic pseudos to reproduce the
results of a particular fine-grid simulation on a coarse grid when the
pseudos are generated from the fine-grid results and a different set
of pseudos is used for every coarse gridblock and flow direction. Im-
plicit in this is that, if the well rates or positions are changed, the
pseudos would have to be regenerated from the results of a new fine-
grid simulation (with the new well rates and positions). In practice,
this means that the following difficulties arise.
1. We cannot have a different set of pseudos for every coarse grid-
block, but must group the coarse gridblocks into a limited number
of “rock types,” each of which has just one set of pseudos (or one
set for each flow direction).
2.ĂWe cannot afford to run the full fine-grid simulation (if we could,
there would be no need to use pseudos at all), so we must choose oth-
er, smaller fine-grid simulations to run to generate the pseudos. Fig. 1—Example of dual-scale grid.
3. We cannot afford to regenerate the pseudos for every different
simulation (with different well positions, rates and other such pa- The invention of the dual-scale-simulation (DSS) method9 has
rameters) we want to run but must be able to run several sensitivity gone some way toward solving this difficulty. In this approach, the
cases using the same pseudos. pseudos are generated from a dual-scale model of the whole (or a
large part) of the reservoir. This model uses a fine grid in and around
Choice of Coarse-Grid Rock Types. Each coarse gridblock re- the coarse gridblock for which the pseudos are to be generated and
quires its own set of pseudos because the pseudos depend both on a coarse grid elsewhere. A 2D example is illustrated in Fig. 1. In
the heterogeneity within the gridblock and on the position of the such a model, the details of the flow in and around the gridblock of
gridblock within the reservoir. Two identical coarse gridblocks (in interest are generally very similar to those that would be obtained
terms of their absolute permeability distribution) may require differ- with the full model, but the cost of the simulations is much reduced.
ent pseudos because of their different positions with respect to the Once the coarse-grid rock types have been defined, DSS models
wells or initial fluid contacts or because of the influence of differing could be set up, one for each rock type. The only difficulties would
permeability distributions in neighboring blocks (i.e., because they be in choosing the exact extent of the model (how large a part of the
are subjected to different boundary conditions). field should be modeled) and the extent of the refined region, bear-
In practice, we cannot afford to have a different set of pseudos for ing in mind that the total number of gridblocks must be small enough
each coarse gridblock, both because of the work required to generate so that the model does not take too long to run, especially if there are
so many sets and because of the amount of memory required to store many coarse-grid rock types. Software to set up and run such models
them during the coarse-grid simulation. Thus, we must assign each in a relatively user-transparent manner would be necessary for a
coarse gridblock to one of a limited number of rock types in such a practical application.
way that all blocks within the same rock type have “similar” pseudos.
However, because the shape of the pseudos depends on so many vari- Dependence on Well Positions and Rates. In principle, pseudos
ables, there are at present no guidelines for making this assignment depend on well positions and rates. They should therefore be gener-
a priori (i.e., as a function of the heterogeneity of the gridblock and ated from a fine-grid or DSS simulation in which the same well rates
its position in the reservoir). Even after the pseudos have been gener- and positions are used as will be used in the coarse-grid model. They
ated for all the coarse gridblocks, there are no established rules for de- should be regenerated for each coarse-grid simulation in which well
termining when two sets of pseudos are sufficiently similar to be positions or rates are changed. There is some evidence that one can
treated as the same (some rules have been suggested recently20,26 but obtain acceptable results for a range of similar production scenarios
their usefulness has not yet been extensively tested). without having to regenerate the pseudos.9 However, there are at
In the past, engineers have used their own judgment to define the present no guidelines for deciding whether two production scenar-
rock types, but this has usually been done on the basis of simple geo- ios are sufficiently similar for regeneration of the pseudos to be un-
logical criteria for models involving relatively simple geological necessary. As for the choice of coarse-grid rock types, it seems un-
structure (usually layering). Position with respect to wells and/or the likely that a few simple guidelines will be sufficient for all cases.
water/oil or gas/oil contact may also have been considered. The va-
lidity of the resulting grouping has rarely been tested. As more and
Implications
more complex geological models are developed, it seems likely that
greater numbers of coarse-grid rock types will be necessary. Assign- We have argued that to be sure of reliable results when using dynam-
ment of gridblocks to the different rock types will become corre- ic pseudorelative permeabilities, the following procedure must be
spondingly more difficult, and less likely to be achievable in an a followed.
priori manner with a few simple guidelines. 1. Divide the coarse gridblocks into a sufficient number of rock
types according to their internal heterogeneity and position in the
Choice of Fine-Grid Simulation. Historically, pseudorelative per- reservoir.
meabilities for three-dimensional (3D) full-field reservoir models 2. Set up and run a number of DSS models (or fine-grid cross-sec-
have been generated from finely gridded two-dimensional (2D) verti- tion, sector, or pattern-element models) that are sufficiently repre-
cal cross-section models or possibly from 3D sector or pattern-ele- sentative of the different geological structures and fluid-flow pat-
ment models. The number of such models used in any one case has terns encountered in the reservoir.
usually been very limited. Given the complex dependence of pseudos 3. Generate a set of pseudos for each rock type from the results
on reservoir geology and geometry and on well rates and positions, of these simulations. Any method may be used if it is verified that
it is likely that this approach has often been inadequate. In practice, the pseudos allow the DSS/fine-grid results to be reproduced on the
its applicability has never really been tested because it has not been corresponding coarse grid (if they do not, the pseudos must be ad-
possible to run a full fine-grid simulation to check the results. justed until they do allow this).

140 SPE Reservoir Engineering, May 1997


4. Regenerate the pseudos every time the well rates or positions need to worry about compensating for numerical dispersion at this
are changed significantly. level because the numerical dispersion associated with the geologi-
We have also argued that criteria for performing Step 1, for choos- cal model will be negligible if the grid is fine enough.
ing the models (other than DSS models) in Step 2, and for determin- Thus, the prospects for scaling up from the core scale to the geolog-
ing what changes are “significant” in Step 4, are not available. Further ical model scale are much brighter than for scaling up from the geo-
research may produce some guidelines that may be valuable; howev- logical model to a coarsely gridded reservoir simulation model. Note
er, it is unlikely that general criteria of this kind can ever be adequately that today, however, most geological models are probably not yet fine
validated, given the wide range of possible geological models. Fur- enough for capillary equilibrium to be assumed. Eventually, of
ther, even if such criteria were available, the procedure would be too course, computers will become so powerful that it will be possible to
time consuming to be feasible in practice. At a minimum, the whole run fluid-flow simulations on grids whose blocks are small enough
process would have to be automated, although automation of the that capillary equilibrium can be assumed within each block. Then the
verification procedure in Step 3 would be difficult. Even with au- capillary-equilibrium method can be used to scale up from any finer
tomation, the approach might not be practical because the number geological model, no matter how many gridblocks it has.
of rock types needed to obtain an accurate result may simply be too
large and/or because the amount of computer time required to run A Practical Approach for Today. In the meantime, when presented
the fine-grid/DSS simulations may be prohibitive. Thus, we con- with a geological model that is too large to be used directly for multi-
clude that dynamic pseudos cannot currently be used to scale up reli- phase flow simulations, a first step is to establish the type of hetero-
ably from a geological to a fluid-flow model, nor is it likely that this geneity present in the model. The key features to look for are prob-
will be possible in the foreseeable future. ably long, thin, high-permeability channels; thin, high-permeability
layers; and extended flow barriers, such as shales.
Discussion If such features are not present (i.e., if the correlation length of the
heterogeneity is much shorter than the interwell spacing), then the
Alternative Approaches. An alternative to generating pseudos heterogeneity may have little effect on the flow and it may be suffi-
from simulation results is to attempt to scale up the relative permea- cient to scale up the absolute permeability alone. Pseudos may still
bilities analytically. One such approach is the large-scale averaging be required to compensate for numerical dispersion or if there are
method.27-29 As with most analytic methods, however, quite severe small-scale flow phenomena, such as viscous fingering or thin grav-
assumptions are necessary. To date, results have only been obtained ity tongues, that the grid is too coarse to resolve.
for systems where the heterogeneity has a very short correlation If some of these geological features are present, the first thing to
length and for quasisteady-state (capillary-dominated) cases. Other do is to try and represent them explicitly by a careful choice of grid-
authors30 have used extended forms of the well-known Dietz theory, ding or the use of local grid refinement. If there are too many such
but this approach assumes vertical equilibrium and results have features for this to be possible with an acceptable number of grid-
been obtained only for stratified systems. blocks, pseudorelative permeabilities will be necessary. A check
The alternative to using pseudorelative permeabilities to incorpo- should be made to see whether the flow is gravity dominated (in
rate the effect of heterogeneity on fluid-flow simulations is to avoid which case vertical-equilibrium pseudos can be used) or capillary
scaling up altogether and to make some kind of fluid-flow simula- dominated (in which case capillary-equilibrium pseudos can be
tion directly on the geological model grid. It may be possible to re- used). These checks may not be straightforward because of the diffi-
duce the cost of these simulations through one of the following. culty of evaluating representative dimensionless numbers for a
1. Modeling only a part of the reservoir (i.e., a sector or pattern- complex, heterogeneous reservoir (see Appendix A).
element model). At this stage, it may be worth considering whether a full-field
2. Using adaptive gridding techniques and local grid refinement model is really necessary or whether a sector or pattern-element
to minimize the number of gridblocks used while still obtaining ac- model, in which the fine grid of the geological model could be re-
ceptable results. This approach has much to recommend it,1,31 but tained, could be used instead. This will depend on the objectives of
existing criteria for defining the grid neglect gravity effects, and, in the simulation study, of course.
some cases, it may simply be a fact that the number of gridblocks If all these possibilities are ruled out, dynamic pseudos will have
required to give an accurate result is prohibitively large. to be generated. In this case, the procedure outlined in the Implica-
3. Using a coarse grid to solve the pressure equation while retain- tions section should be followed as closely as possible, paying par-
ing a fine grid for solving the saturation equation.32 ticular attention to the extended flow channels and barriers. Pseudo-
4. Using steamtube-based techniques.33 capillary pressure can probably be neglected in most cases (see
However, all these methods (except the first) require further R&D Appendix B). It should be recognized that the results will probably
before they can be considered for use on an industrial basis. not be very reliable, and it should not be expected that the detailed
information in the geological model will be captured in the fluid-
Scaling Up From Core to Geological Model. Note that there is a flow simulation. If care is taken, however, the effects of the princi-
second scaling-up problem, that of moving from relative permeabi- pal heterogeneities (i.e., the channels, thief zones, and barriers)
lities measured on the core scale (typically 10 cm) or smaller (e.g., should be captured in a qualitative manner.
laminated rocks with millimeter- to centimeter-scale laminations) to
those that should be assigned to gridblocks in the geological model. Uses for Fine-Grid Geological Models. It should not be concluded
This problem is likely to be less difficult because capillary forces be- from the arguments advanced in this paper that the construction of
come more important at smaller length scales. If the geological highly detailed geological models is not useful. As noted in the
model gridblocks are small enough that the saturation distribution introduction, they may be used directly for fluid-in-place and con-
within them can be approximated by assuming capillary equilibri- nectivity calculations. They may give valuable insight into the het-
um, the capillary-equilibrium method described in the Introduction erogeneity that will significantly affect fluid flow in the reservoir
can be used. This avoids most of the problems with dynamic pseu- and, hence, may guide the choice of grid for the fluid-flow simula-
dos that we have described. tions. The effect of the heterogeneity on the fluid flow may be quan-
In this case, the dependence of the pseudos on the boundary tified through use of sector or pattern-element models even if this
conditions is reduced, because the saturation distributions are con- is only qualitatively possible in a full-field model. A detailed model
trolled locally by capillary forces rather than by “external” factors whose permeability has been scaled up to a coarser grid may provide
(i.e., well rates and positions or the permeability distribution in the a useful starting point for a history match that could not easily be ob-
surrounding gridblocks). The rock types can thus be chosen purely tained in any other manner.
on the basis of the internal heterogeneity of each gridblock. Addi- It should also be noted that the accuracy of fluid-flow simulations
tionally, details of the flow in a particular grid cell have less signifi- has improved greatly over the years largely as a result of the inclu-
cance for the overall field or well performance, so retaining all the sion of more and more geological detail in the models. This is turn
details of each set of pseudos is less crucial. Finally, there is less has been made possible by the increases in computing power, which

SPE Reservoir Engineering, May 1997 141


have allowed more and more gridblocks to be used. What we are 3. Lake, L.W., Kasap, E., and Shook, M.: Pseudofunctions—The Key to
saying in this paper is that dynamic pseudos do not provide a reliable Practical Use of Reservoir Description,” North Sea Oil and Gas Reser-
alternative to the explicit representation of heterogeneity in the voirs—II, Graham and Trotman (1990).
fluid-flow simulation grid. 4. Coats, K.H., Dempsey, J.R., and Henderson, J.H.: “The Use of Vertical
Equilibrium in Two-Dimensional Simulation of Three-Dimensional
Reservoir Performance,” SPEJ (March 1971) 63.
Other Views of Pseudos. We have discussed the difficulty of using
5. Rosenzweig, J.J., Abdelmalck, N.A., and Gochnour, J.R.: “The Devel-
pseudos to reproduce the results of a fine-grid simulation on a coarser opment of Pseudo Functions for Three-Phase Black-Oil Simulators,”
grid. However, given the uncertainty usually associated with fine- Reservoir Characterization, L.W. Lake and H.B. Carroll (eds.), Aca-
scale reservoir descriptions, it might be more desirable to find demic Press, San Diego, California (1985).
pseudos that reproduce in some average sense the results of many 6. Lemouzy, P., Kruel Romeu, R., and Morelon, I.: “A New Scaling-Up
fine-grid simulations, each of which is a different realization of the Method To Compute Relative Permeability and Capillary Pressure for
geological model (although this would not help in estimating the un- Simulation of Heterogeneous Reservoirs,” paper SPE 26660 presented
certainty resulting from the existence of multiple possible realiza- at the 1993 SPE Annual Technical Conference and Exhibition, Houston,
tions). Alternatively, it may be enough to reproduce the differences 3–6 October.
in production associated with different operating strategies even if 7. Durlofsky, L.J.: “Numerical Calculations of Equivalent Gridblock
Permeability Tensors for Heterogeneous Porous Media,” Water Re-
there is a bias in the actual production predictions. A third possibility
sources Research (1991) 27, No. 5, 699.
might be to find the pseudos that minimize the difference between 8. Pickup, G.E. and Sorbie, K.S.: “The Scale-up of Two Phase Flow Using
coarse- and fine-grid results for a range of operating scenarios. Which Permeability Tensors,” paper presented at the 1994 European Confer-
(if any) pseudos might be useful in any of these circumstances is a ence on the Mathematics of Oil Recovery, Roros, Norway, 7–10 June.
question that has received little attention to date; again it is a question 9. Thibeau, S., Barker, J.W., and Souillard, Ph.: “Dynamical Upscaling
that would be difficult to answer with any generality. Techniques Applied to Compositional Flows,” paper SPE 29128 pre-
sented at the 1995 SPE Symposium on Reservoir Simulation, San Anto-
Conclusions nio, Texas, 12–15 February.
10. Barker, J.W. and Dupouy, P.: “An Analysis of Dynamic Pseudo Relative
1. Pseudorelative permeabilities cannot be used to scale up reli- Permeability Methods,” paper presented at the 1996 European Conference
ably from a fine-grid geological model to a coarse-grid fluid-flow on the Mathematics of Oil Recovery, Leoben, Austria, 3–6 September.
model except in cases where capillary or gravity equilibrium can be 11. Kyte, J.R. and Berry, D.W.: “New Pseudofunctions To Control Numeri-
assumed at the coarse-gridblock scale. cal Dispersion,” SPEJ (August 1975) 269.
2. Except in the cases where capillary or gravity equilibrium can 12. PSEUDO Reference Manual, Intera Information Technologies Ltd.
be assumed at the coarse-gridblock scale, the effects on fluid flow 13. Ekrann, S. and Dale, M.: “Averaging of Relative Permeability in Hetero-
of the correlated heterogeneities in the geological model can only geneous Reservoirs,” The Mathematics of Oil Recovery, P.R. King, Ox-
be captured qualitatively unless they are explicitly resolved on the ford U. Press, Oxford, U.K. (1992).
14. Stone, H.L.: “Rigorous Black-Oil Pseudofunctions,” paper SPE 21207
fluid-flow simulation grid.
presented at the 1991 SPE Symposium on Reservoir Simulation, Ana-
3. Scaling up from the core scale to the geological model is more heim, California, 17–20 February.
likely to be possible because capillary forces are more important at 15. Barker, J.W. and Fayers, F.J.: “Transport Coefficients for Compositional
smaller scales. Simulation With Coarse Grids in Heterogeneous Media,” SPE Ad-
vanced Technology Series (April 1994) 103.
Nomenclature 16. Hewett, T.A. and Behrens, R.A.: “Scaling Laws in Reservoir Simulation
and Their Use in a Hybrid Finite Difference/Streamtube Approach to
g+ gravitational acceleration Simulating the Effects of Permeability Heterogeneity,” Reservoir Char-
k+ absolute permeability acterization II, L.W. Lake, H.B. Carroll, and T.C. Wesson (eds.), Aca-
kr + relative permeability demic Press, San Diego, California (1991) 402–441.
Pc + capillary pressure 17. Christie, M.A. et al.: “A Renormalization-Based Upscaling Technique
S+ saturation for WAG Floods in Heterogeneous Reservoirs,” paper SPE 29127 pre-
t+ time sented at the 1995 SPE Symposium on Reservoir Simulation, San Anto-
v+ Darcy velocity nio, Texas, 12–15 February.
Dh+ depth variation 18. Alabert, F.G. and Corre, B.: “Heterogeneity in a Complex Turbiditic
Reservoir: Impact on Field Development,” paper SPE 22902 presented
Dl+ gridblock size
at the 1991 SPE Annual Technical Conference and Exhibition, Dallas,
D ò+ density difference 6–9 October.
m+ viscosity 19. Smith, E.H.: “The Influence of Small-Scale Heterogeneity on Average
Relative Permeability,” Reservoir Characterization II, L.W. Lake, H.B.
Acknowledgments Carroll, and T.C. Wesson (eds.), Academic Press, San Diego, California
(1991) 52–76.
This paper would not have been possible without the benefit of simula-
20. Saad, N., Cullick, A.S., and Honarpour, M.M.: “Effective Relative
tion studies run by Ian Culverwell and Philippe Dupouy of the Elf Geo- Permeability in Scale-Up and Simulation,” paper presented at the 1995
science Research Centre in London. Discussions with Dominique SPE 29592 Joint Rocky Mountain Regional Meeting/Low Permeability
Monfrin, Lisette Quettier, and Philippe Tessier of Elf Aquitaine Pro- Reservoirs Symposium, Denver, Colorado, 20–22 March.
duction in Pau have also been helpful. We thank Elf Aquitaine Produc- 21. Jacks, H.H., Smith, O.J.E., and Mattax, C.C.: “The Modeling of a Three-
tion and Elf Enterprise Caledonia for permission to publish this paper. Dimensional Reservoir With a Two-Dimensional Reservoir Simula-
The views expressed in this paper are those of the authors and do not tor—The Use of Dynamic Pseudofunctions,” SPEJ (June 1973) 175.
necessarily reflect those of Elf Aquitaine or any of its subsidiaries. 22. Guzman, R.E. et al.: “Evaluation of Dynamic Pseudo Functions for Res-
ervoir Simulation,” paper SPE 35157 presented at the 1996 SPE Annual
Technical Conference and Exhibition, Denver, Colorado, 6–9 October.
References
23. Bedrikovetsky, P. and Bruining, J.: “A Percolation Based Upscaling Tech-
1. Durlofsky, L.J., Jones, R.C., and Milliken, W.J.: “A New Method for the nique for Viscous Force Dominated Waterflooding in Uncorrelated Heter-
Scale Up of Displacement Processes in Heterogeneous Reservoirs,” pa- ogeneous Reservoirs,” paper presented at the 1995 EAPG European Sym-
per presented at the 1994 European Conference on the Mathematics of posium on Improved Oil Recovery, Vienna, Austria, 15–17 May.
Oil Recovery, Roros, Norway, 7–10 June. 24. Killough, J.E. and Foster, H.P.: “Reservoir Simulation of the Empire
2. Muggeridge, A.H.: “Generation of Effective Relative Permeabilities Abo Field: The Use of Pseudos in a Multilayered System,” SPEJ (Octo-
from Detailed Simulation of Flow in Heterogeneous Porous Media,” ber 1979) 279.
Reservoir Characterization II, L.W. Lake, H.B. Carroll, and T.C. Wes- 25. Tan, T.B.: “Estimating Two and Three Dimensional Pseudorelative Per-
son (eds.), Academic Press, San Diego, California (1991) 197–225. meabilities With Nonlinear Regression,” paper SPE 29129 presented at

142 SPE Reservoir Engineering, May 1997


the 1995 SPE Symposium on Reservoir Simulation, San Antonio, Tex- can be assumed in determining the saturation distributions within
as, 12–15 February. gridblocks). In practice, it is not easy to determine the required “rep-
26. Christie, M.A.: “Upscaling for Reservoir Simulation,” JPT (November resentative” values for complex heterogeneous reservoir models
1996) 1004. nor the threshold values below which capillary or gravity equilibri-
27. Ahmadi, A., Labastie, A., and Quintard, M.: “Large-Scale Properties for um applies. For example, Rosenzweig et al.5 found that the thresh-
Flow Through a Stratified Medium: Various Approaches,” SPERE (Au-
old value of thier “equilibrium numbers” varied according to wheth-
gust 1993) 214.
28. Amaziane, B. and Bourgeat, A.: “Effective Behaviour of Two-Phase
er the system had one or three layers (the only cases studied).
Flow in heterogeneous Reservoirs,” Numerical Simulation in Oil Re-
covery, IMA Volumes in Mathematics and its Applications, M.F. Wheel- Appendix BĊPseudocapillary Pressure
er (ed.), Springer-Verlag, New York City (1988) 1. We have not made much mention of the use of pseudocapillary pres-
29. Quintard, M. and Whitaker, S.: “Two-Phase Flow in Heterogeneous Po- sure in conjunction with dynamic pseudorelative permeabilities.
rous Media, the Method of Large-Scale Averaging,” Transport in Po-
Except in the Kyte and Berry9 method, none of the methods discussed
rous Media (1988) 3, 357.
30. Ingøy, P., Gauchet, R., and Lake, L.W.: “Pseudofunctions and Extended
uses a pseudocapillary pressure if there is no capillary pressure in the
Dietz Theory for Gravity-Segregated Displacement in Stratified Reser- fine-grid model. In all cases, the pseudocapillary pressure will not be
voirs,” SPERE (February 1994) 67. greater than the fine-grid capillary pressure. The importance of capil-
31. Durlofsky, L.J. et al.: “Scale Up of Heterogeneous Three Dimensional lary pressure decreases as the length scale increases. In cases where
Reservoir Descriptions,” paper SPE 30709 presented at the 1995 SPE dynamic pseudorelative permeabilities are required, viscous forces
Annual Technical Conference and Exhibition, Dallas, 22–25 October. are at least comparable with capillary forces even at the gridblock
32. Guérillot, D.R. and Verdiere, S.: “Different Pressure Grids for Reservoir scale. All these factors indicate that, unless the rock capillary pressure
Simulation in Heterogeneous Reservoirs,” paper SPE 29148 presented is unusually large, it is unlikely that the pseudocapillary pressure
at the 1995 SPE Symposium on Reservoir Simulation, San Antonio, would have any significant effect on the coarse-grid simulations
Texas, 12–15 February. when dynamic pseudorelative permeabilities are required.
33. Thiele, M.R., Blunt, M.J., and Orr, F.M.: “A New Technique for Predict-
ing Flow in Heterogeneous Systems Using Streamtubes,” paper SPE
27834 presented at the 1994 SPE/DOE Symposium on Enhanced Oil SI Metric Conversion Factor
Recovery, Tulsa, Oklahoma, 17–20 April. in. 2.54* E)00 +cm
*Conversion factor is exact. SPERE
Appendix AĊCriteria for Capillary
or Gravity Domination
The ratio of capillary to viscous forces is given by the dimensionless John W. Barker is a reservoir engineer at Elf Aquitaine Production
in Pau, France. Previously, he worked on upscaling and comĆ
capillary number.
positional reservoir simulation at the Elf Geoscience Research
kP c Centre in London; before that, he was at BP Research. He holds
Nc + , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) a BA degree in mathematics from Cambridge U. and a PhD deĆ
mnDl gree in applied mathematics from the California Inst. of
where Pc , k, m and v are representative values of capillary pressure, Technology. Barker is a member of the Editorial Review CommitĆ
permeability, viscosity, and Darcy flow velocity respectively, and tee and was 1993ć94 chairman and a 1990ć93 member of an
Annual Meeting Technical Committee. Sylvain Thibeau is a reĆ
Dl is the gridblock size. Similarly, the ratio of gravity to viscous search engineer working on upscaling methods at Elf Aquitaine
forces is given by the dimensionless gravity number. Production in Pau, France. He holds a degree from Ecole NorĆ
male Supérieure in Paris and a PhD degree in numerical fluid
k vgDòDl
Ng + , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2) mechanics from the U. of Orsay.
mnDh
where D ò and Dh are representative values of density difference
and depth variation, respectively, and g is the gravitational accelera-
tion. Alternative forms of these dimensionless numbers have been
proposed elsewhere.4,5
In principle, evaluation of these dimensionless numbers allows
one to determine whether a flow is capillary or gravity dominated
at the gridblock scale (i.e., whether capillary or gravity equilibrium Barker Thibeau

SPE Reservoir Engineering, May 1997 143

You might also like