Application of Geostatistics For Reservoir Characterization - Accomplishments and Challenges

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Application of Geostatistics for

Reservoir Characterization
Accomplishments and
Challenges
Mohan Kelkar has been a faculty member at the University of Tulsa since 1983.
Currently, he is a professor of petroleum
engineering. He holds B.S. in chemical
engineering from the University of
Bombay, India, M.S. in petroleum engineering and Ph.D. in chemical engineering from the University of Pittsburgh, and
J.D. from the University of Tulsa. His
research interests include reservoir characterization, production optimization, and
risk and uncertainty analysis. He is currently involved in several
research projects in the area of reservoir characterization which
are jointly funded by the Department of Energy and oil companies. He has authored or co-authored over 35 technical articles
and has presented his research findings at numerous meetings
and symposia. He is also writing a book on Application of
Geostatistics for Reservoir Characterization, to be published by
the Society of Petroleum Engineers. He is currently serving as a
technical editor for SPE Reservoir Evaluation. He is a member
of SPE, SEG, American Institute of Chemical Engineers, and
Oklahoma Bar Association.

eostatistics has been used to improve the reservoir characterization process for the last fifteen years. This article
briefly discusses the accomplishments so far, and discusses the future challenges.

We have better and faster computers so that we can handle


more complex algorithms, and can generate detailed reservoir descriptions in a reasonable amount of time. PC revolution has brought the computer power to small players, making it feasible for even small operators to harness the power
of detailed reservoir description processes.
Several new algorithms have been developed in the last fifteen years which allow integration of various types of data in
an easier fashion.
As the reservoirs get more mature, there is more need to
describe the reservoir in more details so as to locate the
remaining hydrocarbons.
There is an increasing recognition that representation of heterogeneities is as important as representing correct physics
in the flow simulators.

What Is Geostatistics?
Geostatistics is based on a simple principle that geological data
are spatially correlated. This spatial correlation is quantified and is
utlilized to determine the weights assigned to the nearby samples
to estimate the value at the unsampled location. Geostatistics has
been used in mining industry for several years. Its use in petroleum industry is of relatively recent origin. To address the unique
problems encountered in petroleum engineering, several new
geostastical techniques have been established in the past several
years. Some of these techniques are discussed below.

Accomplishments to Date
What Is a Reservoir Characterization
Process?
Reservoir characterization is a process of integrating various
qualities and quantities of data in a consistent manner to describe
reservoir properties of interest at inter well locations. Appropriate
weight should be given to the quality and the scale of the data, and
data should be integrated such that we can predict the future performance of the reservoir. Our goal in reservoir characterization is
not to seek the truth about reservoir; instead, to build a reasonable
reservoir model which is adequate to predict the future performance. The model will not only be dependent on the type of data
available, but it should also be dependent on the type of flow
process we are trying to simulate. More complex the flow process
(e.g., CO2 process), more detailed will be the reservoir description; more simple the flow process (e.g., dry gas reservoir), simpler will be the reservoir description.
Reservoir description process is not new; since the first oil discovery, oil companies have used all the available data techniques
to describe the reservoir so that the next well to be drilled will be
based on more information than a prior well. However, several
changes have taken place in the last fifteen years. Some of these
changes are listed below:
July 2000, Volume 39, No. 7

Several papers have been published in the literature to demonstrate the application of geostatistics. It will be impossible to list
all the successes of various geostatistical techniques. Instead, an
effort is made to highlight the accomplishments which are widely
used and embraced by oil companies.

Sequential Simulation Procedures


Conventional geostatistical estimation procedures (e.g., kriging) are filtering techniques. These techniques filter out high frequency information and only capture low frequency data. Many of
the reservoir properties exhibit short scale variability (e.g., permeability), which may not be captured by conventional geostatistical
techniques. Sequential simulation procedures, in contrast, capture
these small scale variabilities. See Figure 1 for example. Journel
and his co-workers(1 2) first established these techniques, and
subsequently developed several alternative applications of these
techniques.
The two principal sequential simulation techniques are sequential indicator simulation (SIS) (1) , and sequential Gaussian
simulation (SGS)(2). SIS transforms the data into discrete indicator
variable before simulation, whereas, SGS transforms the data into
continuous Gaussian domain before simulation. Although difficult
25

(a) Fracture network

FIGURE 1: Comparison of kriging estimates and conditional


simulation realization for porosity.

to generalize, SIS has been found to be most useful in simulating


geological facies or rock types, whereas, SGS has been found to
be most useful in simulating continuous variables such as porosity, permeability, saturation, etc. Mixed applications, where both
techniques are used to improve the description have also been
proposed(3).
Sequential simulation procedures have the advantage of generating multiple images of reservoirs, thus capturing the uncertainties in reservoir description. These procedures are very efficient,
and are able to generate multi-million cell models in a matter of
minutes. Once a novelty, these procedures are rountinely used
today in many oil companies.

Object-Based Modelling
Sequential simulation processes represent simulation methods
which are grid-based. That is, values are assigned at individual
grid locations. In contrast, object-based models generate geological objects (Figure 2) . Complex geological, crisp, patterns can be
created using object-based models. Over the last ten years,
marked point processes have been extensively used to generate
such patterns.(4 5) In addition to geological objects, the methods
have also been used to generate sub-seismic fault maps as well as
other discrete objects. Successful applications of these methods
for many sandstone reservoirs have made the object-based modeling a routine practice.

Integration Algorithms
One of the critical needs in reservoir description is to integrate
data from various sources. Geological, geophysical, and engineering data need to be integrated in a consistent manner so as to
generate as complete description as feasible. Significant strides
have been made towards this goal.
We have already discussed object-based modeling for geological data integration. For grid-based simulation, both SIS as well as
SGS can be used to integrate geological data by first quantifying
the geological information.
Seismic data integration is no longer restricted to defining the
26

(b) Channel sands


FIGURE 2: Examples of stochastic modelling(4).

structural maps. Because of increasing resolution of seismic data,


it is frequently used to describe reservoir properties such as geological facies and porosity. For the integration of structural information, traditional geostatistical approaches such as co-kriging
and external drift method work quite well.(6) For estimating facies
or porosity using seismic data, various methods such as Beysian
updating (7), proportion curves (8), simulated annealing (9), and
sequential co-simulation(10) are proposed. Although the utility of a
specific procedure depends on a specific application, many of
these methods work very well in accomplishing the goal of integrating seismic data. If seismic data needs to be integrated on the
same scale of resolution as the well data, methods such as stochastic simulation can be used(11). It is not an exaggeration to state that
the technology for integration of static data is very mature today
and is rountinely applied. This is also reflected in several new
commercial softwares which are capable of integration of various
sources of data using many of the methods discussed above.

Simplified Flow Simulation


As geostatistical methods become increasingly sophisticated,
the reservoir can be described in much greater details than was
possible before. How to use these detailed descriptions has
become an important issue. For conventional flow simulators, it is
difficult to use multi-million cell grid blocks. That means that the
fine scale grid block scales will have to be up-scaled to a coarser
Journal of Canadian Petroleum Technology

FIGURE 3: Summary of reservoir modeling process showing impact of geologic and rock properties uncertainties on field performance
predictions.

level. Although many sophisticated techniques have been developed to upscale various reservoir properties, any upscaling
involves a loss of information. One way to overcome this problem
is to use flow simulators with simplified physics. Streamline simulators have filled this gap in the last ten years.(12) Streamline
technology has advanced to a level that 3D simulations with
changing well rates and conditions, as well as incorporating gravity effects, can be conducted using streamline simulations. Because
of simplified physics, multi-million cells can be used in the
streamline simulations, thus avoiding the loss of information commonly observed in conventional simulations. Streamline simulators can also be used for ranking various reservoir descriptions
generated by simulation processes. In addition, streamline simulator can also be used to optimally up-scale reservoir descriptions
such that critical information is not lost.

Summary
To summarize the accomplishments in reservoir characterization, in the last fifteen years, we have developed better algorithms,
which can capture uncertainties in reservoir description, and can
integrate various types of static data in a consistent manner. We
also realized that accurate representation of reservoir heterogeneities is critical to capture the reservoir performance. To that
end, we have developed simplified flow simulation process, which
can capture more detailed levels of heterogeneities.

Future Challenges
Although lot has been accomplished in the area of reservoir
characterization, several challenges remain. We discuss some of
them below.

Integration of Uncertainties in Various


Disciplines
Typically, in a history-matching process, geologists provide the
July 2000, Volume 39, No. 7

information on the correlation between wells, and the geophysicists provide the information on structure. This information is carried forward to the engineer who populates the interwell areas
with petrophysical properties to describe the reservoir, eventually
for flow simulation purposes. During the history-matching phase,
the engineer will most likely adjust the petrophysical properties of
the reservoir to match the historical data. Rarely, the geophysical
or geological model will be altered to match the historical data.
Instead, we concentrate on changing engineering data only.
Similar to the uncertainty in engineering properties, uncertainty
also exists in the geological and geophysical interpretations.
Typically, the scale over which geological and geophysical models
are constructed, is much bigger than the scale over which
engineering properties are estimated. When only one geological
and geophysical model is provided to the engineer, we have
assumed that the uncertainty in the geological and geophysical
model is not that critical, although, in reality, the scale over which
these uncertainties are defined has much bigger impact on the
reservoir performance. As a result, when the engineer tries to
match the historical performance, the solution domain which he or
she can explore is already reduced. This problem can be avoided if
we can make the loop bigger (see Figure 3). That is, we understand
the importance of uncertainties in the geological and geophysical
models, and encourage both the geologists and engineers to
develop alternate models, which match the observed data. These
models should capture the uncertainties at the appropriate scale to
be meaningful. We know that interpretation of the data, whether
geophysical or geological, is subjective, and different persons can
develop alternate descriptions. The goal here is to develop
sufficiently different alternate models so that reality can be
bounded. If such integration of uncertainties is achieved, the
engineering properties will not have to be arbitrarily adjusted to
match the historical production performance. Instead, using a
larger solution domain, the engineer would be able to describe a
reservoir, which is consistent with geological, geophysical and
engineering information.
27

FIGURE 4: Different sources of reservoir data from different


scales(2).

Scale Consistency
In achieving the integration of the data, we need to be aware
that the scales of reservoir properties are different. See Figure 4.
We can discuss several examples where the scale consistency
becomes an issue. It is not only related to the fact that different
reservoir parameters are measured over different scales. For
example, well test permeability vs. core permeability, or core
porosity vs. log porosity, or core porosity vs. seismic-derived
porosity. The problem is also related to the fact that crosscorrelations established based on one scale may not be valid on
the other scale. This is because the attributes scale differently at
different scales. For example, if we establish a relationship
between permeability and porosity on a core level, how do we
know that such a relationship will also hold for a grid block scale?
In relating multiple attributes to each other and generating the
reservoir description, we need to correctly address both these
issues. Although the issue has been addressed for static properties,
we still do not have methods for effectively integrating dynamic
measurements of the same attribute. For example, the integration
of core-derived permeability and well test permeability is largely
done on an ad-hoc basis. We need more robust techniques for generating reservoir descriptions, which can properly include the core
permeability and well test permeability in a consistent manner.

Scale Resolution
One of the issues, which needs to be addressed in the integration process is what scale is most appropriate and needs to be
resolved to properly describe the reservoir. With improved tools
and computer speed, there is an increasing tendency to describe
the reservoir on a finer and finer scale. Is this search for detail
really adding to an improved resolution? As the details increase,
so does the uncertainty in describing the reservoir properties. See
Figure 5. We need to develop some guidelines about what is the
most appropriate scale for properly describing a particular reservoir and a particular flow process.

Inverse Problem
In any reservoir description process, we need to solve an
inverse problem. That is, knowing an output, we need to determine the input. This is especially true for integrating production
data in the reservoir description process. Traditional historymatching solves an inverse problem to integrate production data;
however, the procedure is much more cumbersome. Changes in
the reservoir properties are made, largely based on experience,
and based on subjective judgement of the simulation engineer, till
a reasonably satisfactory match is obtained. What is considered a
reasonably satisfactory match is also very subjective. However,
once a match is obtained, the same reservoir description is used to
predict the future performance.
The difficulty in using this traditional approach is that the
28

FIGURE 5: Porosity variance and spread versus averaging


scale(5).

process takes a long time to match the performance. As a result,


very rarely, multiple descriptions are constructed which can match
the historical performance and, at the same time, are able to predict the uncertainty in the future performance. If we have a desire
to match the historical performance, and to be able to predict the
uncertainty in the future performance, we need to automate the
procedure so that multiple descriptions can be easily constructed.
This will allow us to match the historical performance, and predict
the uncertainty in the future performance as well.
Over the last ten years, a lot of progress has been made in
automating the history matching procedure.(13-17) Several new
techniques have emerged; however, the procedures developed so
far are still in research stages. Large scale, practical applications
are still a few years away.

Data Management
The management of data coming from various disciplines is not
a truly technical issue; however, in practice, the lack of data management can hamper the integration of data. Geophysicists, geologists, and engineers typically use different types of software to
process the raw data. Although in recent years, it is changing, the
use of different software is more the norm than the exception. A
consensus is emerging to build software where data from different
sources can be integrated in a seamless fashion. This is a concept
advanced by Petroleum Open Software Corporation(18), and supported by many oil companies. See Figure 6. As our ability to
store large amounts of information increases, it is more critical
than ever that we establish data transparency procedures so that
data used by one discipline can easily be used by other discipline
in a meaningful way.

Summary
In this section, we have summarized some of the challenges in
describing reservoirs in a consistent manner. Chief among them
are the integration of various data in a scale consistent manner,
developing procedures for optimizing appropriate scale for
reservoir description, integrating production data in an optimum
manner, and the ability to use data from various sources in a transparent manner. I am confident that most of these questions will be
satisfactorily answered in the next ten years.

REFERENCES
1. JOURNEL, A.G. and ALABERT, F., New Method for Reservoir
Mapping; Journal of Petroleum Technology, pp. 212-218, February
1990.
Journal of Canadian Petroleum Technology

History Matching of Production Data; paper SPE 38656, presented


at the SPE Annual Technical Conference and Exhibition, San
Antonio, Texas, October 5 8, 1997.
15. LANDA, J.L. and HORNE, R.N., A Procedure to Integrate Well
Test Data, Reservoir Performance History and 4D Seismic
Information into a Reservoir Description; paper SPE 38653, presented at the SPE Annual Technical Conference and Exhibition, San
Antonio, Texas, October 5 8, 1997.
16. VASCO, D.W., DATTA-GUPTA, A. and LONG, J.C.S., Integrating
Field Production History in Stochastic Reservoir Characterization;
paper SPE 36567, presented at the SPE Annual Technical
Conference and Exhibition, Denver, Colorado, October 6 9, 1996.
17. DE-MARSILY, G., LAVEDAN, G., BOCUHER, M. and
FASININO, G., Interpretation of Interference Tests in a Well Field
Using Geostatistical Techniques to Fit the Permeability Distribution
in a Reservoir Model; in Geostatistics for Natural Resources
Characterization, (Verly, G. et al., ed) D. Reidel, Norwell,
Massachusetts, pp. 831-849, 1984.
Source: Landmark Graphics Corp.

18. Petroleum Open Software Corporation , www.posc.org.

FIGURE 6: Integration concepts for shared earth model proposed


by POSC showing interaction among different disciplines and
sources of reservoir data(18).
2. ALABERT, F., The Practice of Fast Conditional Simulation Through
the LU Decomposition of the Covariance Matrix; Math Geology,
Vol. 19, No. 5, pp. 369-386, 1987.
3. XU, W. and JOURNEL, A.G., GTSIM: Gaussian Truncated
Simulation of Lithofacies; Report 6, Stanford Center for Reservoir
Forecasting, May, 1993.
4. HALDORSEN, H.H., BRAND, P J., MACDONALD, C.J., Review
of Stochastic Nature of Reservoirs; Mathematics in Oil Production;
Edwards, S. S. and King, P. R., ed., Clarendon Press, Oxford, 1988.
5. BRATVOLD, R.B., HOLDEN, L., SVANES, T. and TYLER, K.,
STORM: Integrated 3D Stochastic Reservoir Modeling Tool for
Geologists and Reservoir Engineers; paper SPE 27563 presented at
the European Petroleum Computer Conference, Aberdeen, U.K.,
March 15 17, 1994.
6. DEUTSCH, C.V. and JOURNEL, A.G., GSLIB: Geostatistical
Software Library and Users Guide; Oxford Press, New York, NY,
1992.
7. DOYEN, P.M., PSAILA, D.E., DEN BOER, L.D. AND JANS, D.,
Reconciling Data at Seismic and Well Log Scales in 3D Earth
Modeling; paper SPE 38698, presented at the SPE Annual Technical
Conference and Exhibition, San Antonio, Texas, October 5 8,
1997.
8. BAHAR, A. and KELKAR, M., Integrated Lithofacies and
Petrophysical Properties Simulation; paper SPE 38261, presented at
the SPE Western Regional Meeting, Long Beach, California, June 25
27, 1997.
9. DEUTSCH, C.V., SRINIVASAN, S. and MO. Y., Geostatistical
Reservoir Modeling Accoutning for Precision and Scale of Seismic
Data; paper SPE 36497, presented at the SPE Annual Technical
Conference and Exhibition, Denver, Colorado, October 6 9, 1996.
10. ALMEIDA, A.S., Joint Simulation of Multiple Petrophysical
Properties Using Markov-Type Coregionalization Model; Stanford
Center for Reservoir Forecasting, Report 6, Stanford, California.
11. HUANG, X. and KELKAR, M., Reservoir Characterization by
Integration of Seismic and Dynamic Data; paper SPE 35415 presented at the SPE/DOE Tenth Symposium on Improved Oil Recovery,
Tulsa, OK, April 21 24, 1996.

CATCH THE NEW WAVE


In Impr oved Oil Recover y

12. KING, M.J. and DATTA-GUPTA, A., Streamline Simulation: A


Current Perspective; In Situ, 22 (1), 1998.
13. OLIVER, D.S., Incorporation of Transient Pressure Data into
Reservoir Characterization; In Situ, Vol. 18, pp. 243-275, 1994.
14. RAHON, D., EDOA, P.G. and MASMOUDI, M., Inversion of
Geological Shapes in Reservoir Engineering Using Well Tests and
July 2000, Volume 39, No. 7

VISIT US AT: www.geoservicegroup.com


CONTACT: brett@geoservicegroup.com or khayes@bordercity.com
29

You might also like