You are on page 1of 7

Achievements and challenges in petroleum geostatistics

Olivier Dubrule1 and Eivind Damsleth2


1
TotalFinaElf, CSTJF Avenue Larribau, 64000 Pau, France (email: olivier.dubrule@totalfinaelf.com)
2
Norsk Hydro HTP, 0246 Oslo, Norway (email: eivind.damsleth@hydro.com)

ABSTRACT: The current use of geostatistics in the petroleum industry is reviewed


and the main issues that need to be tackled before the potential of geostatistics is
fully realized are highlighted. The paper reviews and discusses three main topics: (1)
geostatistics and geology; (2) multidisciplinary data integration; and (3) uncertainty
quantification with multiple realizations. Our main message is that geostatistics has
come a long way and reached maturity. In the years ahead, geostatisticians should
focus less on the development of new algorithms and more on the training of
geoscientists and the development of new work flows for decision support with
geostatistics as the core.

KEYWORDS: reservoir model, risk analysis, history matching, uncertainty, decision

INTRODUCTION environments have been modelled, but experience remains


In the last twenty years, the development of petroleum geosta- weak with carbonates, due to the lack of clear relationships
tistics has been spectacular. Many companies now recognize between the different scales of heterogeneity present. Also,
geostatisticians as professionals of similar importance to geol- although it is technically straightforward to generate a large
ogists, geophysicists or reservoir engineers. Geostatistical mod- number of reservoir realizations, the practical use of the models
elling studies have become widespread and major vendors offer has, in most cases, been limited to generating one single
geostatistical software in their portfolio. At the same time, it is realization to carry forward.
fair to say that the potential of geostatistics is not yet fulfilled.
Many geoscientists still regard it as a black box approach – little
more than coin tossing to decide which lithology is present at Parameter inference or parameter choice?
specific locations. Geophysicists and reservoir engineers have
been interested in geostatistical developments in seismic inver- Geostatistics came from the mining world, where the high
sion or history matching, but have seldom adopted a multi- density of boreholes, shafts and drifts in several directions
realization paradigm into their work flows. The Bayesian provide sufficient data (Journel & Huijbregts 1978) to ‘infer the
formalism that they often use tends to limit itself to the search 3-D variogram model’. In the oil industry, it quickly became
for a unique ‘most likely’ solution, regardless of the uncertainty. apparent that little quantitative information about the lateral
Multi-realization approaches are generally not used in project extent of bodies could be statistically derived from wells which
economics or after project sanction once development has are usually too scarce and too widespread. Analogue informa-
started. tion provided an alternative: geologists and geostatisticians
needed to combine their efforts to derive statistical information
about the size of modelled objects from outcrops or mature
GEOSTATISTICS AND GEOLOGY fields. This generated a fruitful dialogue between disciplines,
and many outcrop studies were much better focused thanks to
this need to fill quantitative databases. As a result, many major
Successes oil companies now have their own quantitative databases,
On the basis of papers published during the past twenty years, which are used to input quantitative geometrical information
petroleum geostatistics has mainly been used for the construc- into geostatistical simulations (Fig. 1).
tion of 3D reservoir heterogeneity models. Conditional stochas- Strangely enough, the approaches have remained somewhat
tic simulations have been used to generate heterogeneous 3D different between users of object-based and variogram-based
reservoir models where the discontinuities between wells reflect techniques. Users of object-based techniques usually acknowl-
geological assumptions about the various geological bodies. edge the fact that they cannot obtain reliable lateral information
These heterogeneous reservoir models have been successfully from well data alone. The unfortunate result is that sometimes
used by engineers to evaluate and understand hydrocarbon their input geometrical information is not compatible with the
production and fluid flow in realistic heterogeneous reservoirs. well data. On the other hand, many users of variogram-based
Object-based models, Sequential Indicator Simulation, Trun- techniques still define their variogram models on the basis of
cated Gaussian Simulation or Sequential Gaussian Simulation noisy experimental variograms. Although there are some exam-
(Haldorsen & Damsleth 1990; Deutsch & Journel 1992), are ples where a reasonable model can be obtained from wells
now frequently used, sometimes in conjunction with each alone (Dubrule 1998), the quantification of lateral variation is
other, sometimes independently, depending on the scale of the above all a geological problem. Input parameters must often be
problem, or on the depositional environment. Thanks to the chosen, rather than inferred, according to the geological
variety and flexibility of available techniques, most depositional knowledge available about specific depositional environments.

Petroleum Geoscience, Vol. 7 2001, pp. S1–S7 1354-0793/01/$15.00 2001 EAGE/Geological Society of London
S2 O. Dubrule and E. Damsleth

· Low sinuosity 2.6


2 Meandering 1h
Channel 0.0
Braided w= 7
morphology 2.0
∆ Anastomosed 5h
0.8
5
1.8
x Channel morphology w= 2 .1h
=1
4
not known 1.5
x w x
· 2 .6h 1.
36
x 64 3h
x = 51
· x w
· x 2x w
=
x x
x 2
x 2 ·
· 2 x · ·
2
·
· x x x
· x
x x 2· · 22 2
2xx · 2 x· ·x x 2
x 2 2 2 2
xx ·x x x
∆ 2xx x 2· x · x 2 2x
·x x xxxxx x x 2 2 · · 2·
22
x x x x 2 x ·
2 ·
x xx 2 2 2
xx xx xx
22
· xx x · x x
·
· x x 2x x x ·
x xx x ∆ x x · 2 2
2 · x2 2 x x x2 2
∆x
x 2 x x
x x x x xx x
x x
x
x ·
x

·
· ·
·

Fig. 1. Example of compilation of


geometrical information for various
channel morphologies (Fielding &
Crane 1987).

Meaning of geostatistical parameters geostatistical software, including the GSLIB programs


It is clear that significant effort is required to explain the (Deutsch & Journel 1992), constructed 3D geostatistical realiz-
meaning of variogram parameters: what does the range mean? ations in a rectangular box which was intended to represent the
And the sill? What is the relation between variogram models geological formation at the time of deposition, before folding
and transition probabilities? Geologists will not use these and faulting. With these ‘sugar-box’ models the link with the
concepts if they do not fully understand them. It is still structural context was difficult to represent. Now, most com-
common to read papers where the indicator variogram range is mercial ‘Earth Modelling’ programs use a similar three-stage
identified with the size of heterogeneities. This is obviously work flow (e.g. Dubrule et al. 1997):
wrong as, in a zero or one situation, the variogram of the ‘zero’ (1) construction of the geometrical model, incorporating
variable is identical to the variogram of the ‘one’ variable. The surfaces picked on seismic data, large-scale faults and
size of heterogeneities is a function of both the variogram stratigraphic surfaces correlated between wells;
range, the variogram model and the spatial proportion of the (2) construction of the 3D stratigraphic grid controlled by the
indicator variogram (Fig. 2). Also, in multivariate situations, the geometrical model. Each cell of this grid is identified by
indicator cross-variogram can be interpreted in terms of tran- two types of coordinate: the (x,y,z) structural or absolute
sition probabilities, which is another way of making the coordinates and the (u,v,w) relative or stratigraphic
variogram model more meaningful from a geological point of coordinates (Fig. 3);
view (Carle & Fogg 1996). (3) generation of the property model within the stratigraphic
grid. The model is equivalent to the sugar box model in
the (u,v,w) coordinates, but is automatically positioned
Geostatistics and earth modelling tools in the structural space with the (x,y,z) coordinates.
The emergence of 3D earth modelling software has greatly
enhanced the realism of 3D geological realizations. Pioneer

Variogram

P L

P(1-P)

Distance

Fig. 2. The average length of a given facies along the variogram Fig. 3. The construction of a stratigraphic grid between strati-
direction is a function of the slope of the variogram at the origin, and graphic surfaces provides a structural framework at the scale of the
of the proportion p of this facies (Carle & Fogg 1996). individual grid cell.
Achievements and challenges S3

Fig. 4. Fine scale model of


permeability distribution within a tidal
bundle (Wen et al. 1998). The
dimension of this model is
1 m1 m1 m. Up-scaling from core
to geological grid scale requires similar
assumptions about the distribution of
permeability values within the geological
grid cell.

One major challenge: scale change + the mapping of deep-water environments, thanks to the
It is surprising that, until now, geoscientists have been more excellent quality of 3D seismic data.
concerned about how to up-scale geostatistical models from As a result of these developments, geostatisticians are now
geological to reservoir simulation scale rather than from core faced with new challenges.
plug to geological grid scale. It is still common to read papers + How to incorporate sedimentologists’ knowledge about
where many pages are spent explaining the up-scaling from high-resolution sea-level variation into the models? Sedi-
geological to dynamic grid scale, but where cells of the mentary environment maps at significant chronostrati-
geological grid are given their value from one single core plug graphic surfaces, as produced by sedimentologists, can be
that is believed to be representative. This happens in spite of used as a constraint in the generation of 3D sedimentary
the fact that the latter change of support is much more drastic models between wells (Fig. 5). Macdonald et al. (1992) are
than the former. This is one of the major weaknesses of the great supporters of constraining geostatistical models by
stochastic modelling work process. The ‘mini 3D’ approach has using sequence stratigraphic concepts, for depositional en-
been developed to deal with this up-scaling problem: very vironments where these concepts can be applied. This
detailed models are generated at plug scale within a group of interaction between sequence stratigraphy and geostatistics
geological grid cells (Alabert & Massonnat 1990; Wen et al. will become even more important in the future.
1998). These models are then up-scaled to the geological grid + The excellent seismic quality in the deep-water clastic fields
scale and the geostatistical parameters (histogram and vari- of offshore West Africa has led to the construction of
ogram) associated with the up-scaled model are calculated. high-resolution geological models on the basis of seismic
These parameters are then used as the input to the generation data alone (Fig. 6). Now, new questions are raised: is the
of the model at geological grid scale. In this approach, the geological interpretation correct? How can we go one step
geostatistical distribution of plug-scale values within the geo- further, and use the seismic-derived information as a con-
logical grid cells has an important impact: up-scaling, too often straint in the construction of 3D petrophysical property
regarded as a problem for the reservoir engineer only, is now models?
also rightly understood as a problem of imaging the heterogen-
eity distribution at very small scale, in order to understand its New methods
impact on heterogeneity distribution at geological grid scale It is the authors’ view that no radically new geostatistical
(Fig. 4). Hurst et al. (2000) argue that the heterogeneity algorithm has been developed in the recent past. The main
distribution can only be modelled deterministically in deposi- objective of recent developments has been to adapt exist-
tional environments where repetitive features can be identified ing algorithms so that they better support the work of the
through a range of scales (fluvial, aeolian or shallow marine) petroleum geologist.
whereas, in other depositional environments (sand-rich tur- The improvement of object-based models continues. With
bidites, carbonates), up-scaling relationships are elusive and an increasing focus on ancient deep-water systems, there is a
heterogeneity can only be modelled geostatistically. shift away from modelling fluvio-deltaic systems (Holden et al.
1998) to turbidite systems (Portella & Lanzarini 1999).
New modelling techniques are being developed that are
New challenges: the progress in sequence stratigraphy better at handling the relationships between various facies.
and seismic imaging Examples of these developments are the work of Carle & Fogg
Recently, geological science has made strong progress in the (1996) or that of Haas & Dubrule (1999), who try to account
following areas: for transition probability information in the modelling of facies
+ sequence stratigraphy, particularly for shallow-marine architecture. Recent developments with the Pluri-Gaussian
deposits; technique (Le Loc’h & Galli 1997) go in the same direction.
S4 O. Dubrule and E. Damsleth

Fig. 5. This cross-section is extracted


from a 3D sedimentary model
generated in two steps. First, the
sedimentologist produced a depositional
environment maps at the level of the
two bounding chronostratigraphic
surfaces. Then the depositional
environment was interpolated between
the two surfaces (Fontaine et al. 1998).

Wells

Turbiditic channels
picked from seismic

Main erosional
surface
Fig. 6. This model was built entirely
from seismic data acquired over a
turbidite deposit in West Africa
(Wonham et al. 2000). The main
features are the erosive envelope of the
channel complex, and the individual
channels picked from detailed attribute
analysis of the 3D seismic dataset.

MULTIDISCIPLINARY DATA INTEGRATION formal integration of seismic data as well as production data
For multidisciplinary data integration, two approaches are during geological model building.
possible. The first one, which could be called the ‘soft’
approach, relies on statistical relationships between the simu-
lated parameter and another attribute, already simulated or The soft approach
measured. In the geostatistical literature, this other attribute is
often called the secondary variable. The second approach, more This approach is often used to predict permeability using
ambitious, consists of running a stochastic inversion approach, information from an already simulated porosity realization, or
assuming that there is a functional (deterministic) relationship for predicting porosity from seismic information. The first step
(the forward model) linking the simulated parameter with the consists of calibrating a statistical relationship between the two
secondary variable. This approach finds applications both in parameters. This is most often a linear relationship. The second
seismic and production data integration. In relation with this step consists of using this calibrated formula in conjunction
approach, Omre and his colleagues have, over the years, with kriging or conditional simulation to generate a realization
developed an elegant formalism for integration of data from of one parameter constrained by the other. A great variety of
multiple sources with varying quality, based on a likelihood techniques have been developed in the last decade: Markov–
concept (for a brief summary, see Damsleth & Omre 1997). Bayes, external drift, cosimulation, cloud-transform or Bayesian
This approach has contributed to a better understanding of the approaches (Dubrule 1998). It now seems that the collocated
Achievements and challenges S5

cokriging approach, reformulated by Doyen et al. (1996) in the behaviour of large numbers of realizations, in order to
Bayesian framework, is becoming widely adopted. recognize the extremes within a realistic time frame. It now
seems like the streamtubes approach (Datta Gupta et al.
Stochastic inversion techniques 2001; Brandsaeter et al. 2001) is providing a satisfactory
solution to this problem. However, the generation and
In multidisciplinary data integration, multi-realization storage of multiple realizations is still a computer-intensive
approaches such as seismic geostatistical inversion are used operation.
more and more commonly (Dubrule et al. 1998). They tend + The need for history-matching methods that are adapted to
to compete with Bayesian-type seismic inversion techniques. multiple realizations from a geostatistical model. Currently,
Unfortunately, geoscientists familiar with the latter often tend computer-aided history-matching techniques still assume
to be suspicious of the former. Bayesian formalism does not that there is only one single model to be history matched. As
usually include generation of realizations, but simply leads to mentioned above, the approach that is used is Bayesian,
expressions for the posterior distribution. Its application is, in but almost always limited to the calculation of a MAP
most cases, reduced to the calculation of the most likely or (maximum a posteriori) model.
maximum a posteriori (MAP) model. Actually, the formalisms of + Reservoir engineers have traditionally focused on optimizing
stochastic and Bayesian inversion are closely related. Only when recovery using a single ‘base case’ model. Rather, they
this is clearly recognized and explained will geophysicists fully should look for robust developments and well locations that
accept the stochastic formalism. Computer time is also an issue, perform well for a range of plausible realizations. A change
since the generation of a large number of 3D realizations that is needed in the way engineers address uncertainty in their
are constrained by seismic data still takes too much time to be decision-making.
done routinely. Development of techniques such as Fast The use of earth modelling tools is crucial in the generation of
Fourier Transform (FFT) simulation should contribute to make multi-realizations incorporating all major uncertainties. It is
these techniques more acceptable (Le Ravalec et al. 2001). only lately that commercially available tools have become close
to being capable of incorporating all the major uncertainties
History-matching techniques (Fig. 7). Only recently have we seen complete studies where the
uncertainty in the seismic interpretation, depth conversion,
History-matching techniques used in the industry tend to focus stratigraphic interpretation, faults (both interpreted and sub-
on the calculation of one single model matching the production seismic) is combined with the uncertainty in the facies distri-
data. Most of these techniques also work at the scale of the bution and petrophysical properties (Corre et al. 2000). In this
up-scaled dynamic model. This model is updated in order to respect, it is crucial that not only the uncertainty induced by
match the production data, but seldom are the modifications generating multiple realizations with the same set of input
applied to the original geological model. These two issues have parameters is considered, but also that the parameter uncer-
been major obstacles to the use of geostatistical techniques by tainty is taken into account. The two authors do not quite agree
reservoir engineers. with each other as to whether the use of 3D modelling tools for
The non-uniqueness topic is discussed in this issue of integrating all these major uncertainties is the best way to
Petroleum Geoscience by Floris et al. (2001). The pilot-point address the full uncertainty quantification. If the objective is the
method, or the method of gradual simulations proposed by Hu quantification of in-place volume or reserve uncertainty, the
et al. (2001), allows the generation of a large number of alternative is to use Monte Carlo (MC) simulation of more or
up-scaled geostatistical realizations that are consistent with less refined global models. However, the input to the various
production data. components in the MC-model and their interrelationships
To solve the problem of linking the up-scaled model and the should be derived from geostatistical (sub-) models.
geological model, Chevron uses a ‘down-scaling’ approach.
They history match a number of realizations using streamline
simulation (for streamline simulation, see Brandsaeter et al. Multiple realizations vs. discrete scenarios:
(2001) or Datta Gupta et al. (2001), both in this issue). Min/Median/Max
Then they down-scale, assuming that a power-law relates In most operating companies, uncertainty quantification still
up-scaled permeabilities of the reservoir simulation grid to the consists of calculating, for each parameter or each map, a
permeabilities of the geological grid (Tran et al. 1999). minimum, a most likely and (although less important in the
The techniques proposed to history match at the scale of the decision-making and in the reporting process) a maximum
geological model while quantifying uncertainties will not be value. Then these optimistic, most likely and pessimistic sce-
accepted immediately. There is a fundamental need for reser- narios associated with each variable are combined to derive
voir engineers to address history matching in a different way, if minimum, median and maximum values for the associated
geostatistical techniques are to be used to their full benefit. economic parameters. This approach has the advantage of
being simple to use in many different situations. However, it is
UNCERTAINTY QUANTIFICATION WITH well known that combining a minimum with another minimum
MULTIPLE REALIZATIONS to calculate a combined minimum is over-pessimistic. Also,
approaches used for deriving minimums, medians or maxi-
Still unestablished mums are often subjective and, in many cases, no probability
Initially, geostatistics, with its toolbox of multi-realization figure is associated with the different scenarios. On the other
simulation techniques, promised to be ideally suited to address- hand, the multi-realization approach documents probabilities at
ing the quantification of uncertainty. However, only recently each step of the process and translates these probabilities on
has the use of multiple realizations proven to be operationally pdfs of the output parameters. A stochastic approach can thus
practical. Reasons for the slow progress of multi-realization be combined with one based on discrete scenarios, as long as
approaches are as follows. probabilities are quantified (Corre et al, 2000). This move from
+ Computer time and storage. Many years of research have the ‘Min/Median/Max’ paradigm to a fully probabilistic for-
been spent looking for tools that screen quickly the dynamic malism is taking place slowly in the industry, but it will take
S6 O. Dubrule and E. Damsleth

3/5/01 7

Fig. 7. Three realizations of a geological model combining a sampling of the various uncertainties affecting the model: each realization has a
different hydrocarbon–water–contact, a different top structure and a different heterogeneity distribution.

many years to become widely accepted. To achieve this, the model itself changes, uncertainty has no reason to decrease as
‘scenario’ approach and the probabilistic approach must be the amount of data increase. In a multi-realization approach, it
recognized as complementary and combined. is crucial to emphasize that any uncertainty quantification is
conditional to a set of basic assumptions, e.g. the assumption
that the overall conceptual model is correct. These assumptions
Multi-realizations and economic uncertainty analysis must be made clear and documented. This issue is not one of
Geostatistical uncertainty evaluations are often too limited in developing further geostatistical algorithms, but of making the
their scope. Only one development plan is considered, and the work flow and its assumptions very clear to everyone involved
influence of subsurface uncertainties on this single develop- in the decision-making process.
ment scheme is quantified. This is obviously too limited. For
instance, the choice of well locations should be derived from
the geostatistical uncertainties, to achieve a choice of locations
that maximize production while minimizing risk. It is likely that,
in the future, subsurface uncertainties will be integrated in a
more general work flow where the impact of subsurface
uncertainties on NPV (or other economic characteristics), or on
the optimization of development decisions, will be integrated
(Fig. 8). A major revision is required of the approaches
currently used by petroleum companies.

Our perception of uncertainty improves as more data are


gathered
Often, the acquisition of new data leads to a fundamental
revision of models built before these data were acquired. This is
because, at appraisal stage, when only a few wells and lower
quality seismic data are available, it is very difficult to forecast
all the possible scenarios. As more data are acquired, faults Fig. 8. The uncertainty on reserves, obtained using such approaches
appear, or permeability barriers, which complicate the defini- as that illustrated on Figure 7, must be incorporated with other
tion of the reservoir model, become apparent. As a result, if the economic parameters to provide a support to decision-making.
Achievements and challenges S7

THE FUTURE Dubrule, O., Thibaut, M., Lamy, P. & Haas, A. 1998. Geostatistical Reservoir
Characterization constrained by 3D Seismic Data. Petroleum Geoscience, 4,
What is the future of petroleum geostatistics? To us, there are 121–128.
few areas where more algorithms need to be developed. A Fielding, C. R. & Crane, R. C. 1987. An Application of Statistical Modelling
crucial development though is that of history matching under to the Prediction of Hydrocarbon Recovery Factors in Fluvial Reservoir
uncertainty. As long as reservoir engineers do not have a Sequences. In: Ethridge, F. C., Flores, R. M. & Harvey, M. D. (eds)
Recent Developments in Fluvial Sedimentology. SEPM Special Publication, 39,
method to handle multiple realizations at the geological model 321–327.
scale, they will remain uninterested in uncertainty quantification Floris, F. T., Bush, M. D., Cuypers, M., Roggero, F. & Syversveen, A. R.
using geostatistics. 2000. Comparison of production forecast uncertainty quantification
Apart from this crucial area of research, the message is to methods – an integrated study. Petroleum Geoscience, 7, S87–S96.
focus less on the development of new algorithms, but rather to Fontaine, J. M., Dubrule, O., Gaquerel, G., Lafond, C. & Barker, J. 1998.
train geoscientists, and develop new work flows based on Recent Developments in Geoscience for 3D Earth Modelling. Paper SPE
geostatistics. We need to train geophysicists and reservoir 50568 presented at the SPE European Petroleum Conference, The Hague,
The Netherlands.
engineers, especially those working on seismic inversion, reser- Haas, A. & Dubrule, O. 1999. Petroleum Geostatistics, from the Stone Age
voir geophysics, reservoir characterization and computer-aided to Industrial Times. In: Lippard, et al. (ed.) Proceedings of the 5th Annual
history matching. We need to work more closely with geol- Conference of the International Association for Mathematical Geology. Tapir,
ogists, who must understand that geostatistics is simply a Trondheim, Norway, 485–492.
toolbox that can be tailored to the geological problem at hand. Haldorsen, H. H. & Damsleth, E. 1990. Stochastic Modelling. Journal of
We also need to tackle the issue of reserves reporting, prob- Petroleum Technology, April, 404–412.
abilistic economic evaluation and their integration with the Holden, L., Hauge, R., Skare, O. & Skorstad, A. 1998. Modelling of fluvial
reservoirs with object models. Mathematical Geology, 30.
geostatistical multi-realization approach.
Homewood, P., Guillocheau, F., Eschard, R. & Cross, T. A. 1992.
Corrélations Haute-Résolution et Stratigraphie Génétique: Une Démarche
The authors would like to express their thanks to the management of Intégrée. Bulletin Centre Recherche Exploration-Production Elf-Aquitaine,
TotalFinaElf and Norsk Hydro ASA for permission to publish this 57–381.
paper. However, the views and opinions presented in this paper are Hu, L. Y., Le Ravalec, M. & Blanc, G. 2000. Gradual Deformation
those of the authors only. and Iterative Calibration of Truncated Gaussian Simulations. Petroleum
Geoscience, 7, S25–S30.
REFERENCES Hurst, A., Cronin, B. & Hartley, A. 2000. Reservoir modelling sand-rich
deep-water clastic: the necessity of down-scaling. Petroleum Geoscience, 6,
Alabert, F. G. & Massonnat, G. J. 1990. Heterogeneity in a Complex 67–76.
Turbiditic Reservoir: Stochastic Modelling of Facies and Petrophysical Journel, A. G. & Huijbregts, Ch. 1978. Mining Geostatistics. Academic Press,
Variability. Paper SPE 20604 presented at the 65th Annual Technical New York.
Conference and Exhibition of the Society of Petroleum Engineers, New
Orleans, Louisiana. Le Loc’h, G. & Galli, A. 1997. Truncated Pluri-Gaussian Method: Theoretical
and Practical Points of View. In: Baafi, & Schofield, (eds) Geostatistics
Brandsaeter, I., Wist, H. T., Naess, A., Lia, O., Arntzen, O. J. & Ringrose, Wollongong ’96. Kluwer, Amsterdam.
P. S. 2000. Ranking of stochastic realizations of complex tidal reservoirs
using streamline simulation criteria. Petroleum Geoscience, 7, S53–S63. Le Ravalec, M., Noetinger, B., Hu, L. Y. & Blanc, G. 2000. conditioning to
Carle, S. F. & Fogg, G. E. 1996. Transition Probability-Based Geostatistics. dynamic data: an improved zonation approach. Petroleum Geoscience, 7,
Mathematical Geology, 28, 453–476. S9–S16.
Corre, B., Thore, P., De Feraudy, V. & Vincent, G. 2000. Integrated Macdonald, A. C., Hoye, T. H., Lowry, P., Jacobsen, T., Aasen, J. O. &
Uncertainty Assessment for Project Evaluation and Risk Analysis. Paper Grindheim, A. O. 1992. Stochastic Flow Unit Modelling of a North Sea
SPE 65205. Coastal-Deltaic Reservoir. First Break, 10, 124–133.
Damsleth, E. & Omre, H. 1997. Geostatistical Approaches in Reservoir Portella, R. C. M. & Lanzarini, W. L. 1999. Impact of Heterogeneities
Evaluation. Journal of Petroleum Technology, 49, 498–502. in Horizontal Well Performance. Paper SPE 53982 presented at the
Datta-Gupta, A., Kulkarni, K. N. & Yoon, S. 2000. Streamlines, ray tracing SPE Latin American and Caribbean Petroleum Engineering Conference,
and production tomography: generalization to compressible flow. Petroleum Caracas, Venezuela.
Geoscience, 7, S75–S86. Tran, T. T., Wen, X.-H. & Behrens, R. A. 1999. Efficient Conditioning of 3D
Deutsch, C. V. & Journel, A. G. 1992. GSLIB: Geostatistical Software Library and Fine-Scale Reservoir Model To Multiphase Production Data Using
User’s Guide. Oxford University Press, New York. Streamline-Based Coarse-Scale Inversion and Geostatistical Downscaling.
Doyen, P. M., Den Boer, L. D. & Pillet, W. R. 1996. Seismic Porosity Paper SPE 56518 presented at the SPE Annual Technical Conference and
Mapping in the Ekofisk Field using a New Form of Collocated Cokriging. Exhibition, Houston, Texas.
Paper SPE 36498 presented at the SPE Annual Technical Conference and Wen, R., Martinius, A. W., Naess, A. & Ringrose, P. S. 1998. Three-
Exhibition, Denver, Colorado. dimensional simulation of small-scale heterogeneity in tidal deposits. A
Dubrule, O. 1998. Geostatistics in Petroleum Geology. AAPG Continuing process-based stochastic method. In: Buccianti, et al. (ed.) Proceedings of the
Education Course Note Series, 38. 4th Annual Conference of the International Association for Mathematical Geology.
Dubrule, O., Basire, C., Bombarde, S., Samson, P., Segonds, D. & Wonham, Ischia, Italy.
J. 1997. Reservoir Geology Using 3-D Modelling Tools. Paper SPE 38659 Wonham, J. P., Jayr, S., Mougamba, R. & Chullion, P. 2000. 3D Sedimentary
presented at the SPE Annual Technical Conference and Exhibition, Evolution of a Canyon Fill (lower Miocene age) from the Mandorove
San Antonio, Texas. Formation, Offshore Gabon. Marine and Petroleum Geology, 172, 175–197.

Received 7 February 2000; revised typescript accepted 4 September 2000

You might also like