Vous êtes sur la page 1sur 15

REE180917 DOI: 10.

2118/180917-PA Date: 2-April-16

Stage:

Page: 1

Total Pages: 15

Application of Grid-Free
Geostatistical Simulation to a Large
Oil-Sands Reservoir
Yevgeniy Zagayevskiy and Clayton V. Deutsch, University of Alberta

Summary
Geostatistical simulation is performed for reservoir characterization to depict local variability in the modeled properties. The
conventional simulation methods are implemented in a griddependent manner that makes regridding of realizations, refinement of existing grids, and the simulation on irregular grids
challenging. The grid-free-simulation (GFS) method has been
recently developed for flexible reservoir characterization. The geostatistical realizations of a reservoir are expressed as an analytical
function of the coordinates of the simulation locations and, thus,
are infinitely resolvable. The resulting model is conditioned to primary scattered point-scale hard data and secondary exhaustively
sampled block-scale soft data. The former data are sampled along
wells, whereas the latter data are from seismic surveys. The GFS
methodology is applied to the Firebag oil-sands thermal project
operated in northern Alberta, Canada. The conditioning data are
point-scale core measurements, log observations, and block-scale
acoustic impedance (AI). The models of correlated porosity, permeability, and water saturation attributes are constructed on three
different grids by facies and are consistent with each other. These
models are intended for resource estimation, reserves estimation,
and subsequent-flow simulation, respectively. The modeling
results of the grid-independent simulation method are promising
for industrial application to petroleum reservoir characterization.
Introduction
Petroleum reservoir characterization is essential. Multiple aspects
of the reservoir are examined from a geomodeling perspective to
assist reservoir management. The reservoir architecture, stratigraphy, facies distribution, and geological and petrophysical properties are static elements of the reservoir model. These attributes are
required for reserves estimation, well placement, and flow prediction of the reservoir fluids. They are usually modeled with geostatistical methods (Journel and Huijbregts 1978; Goovaerts 1997;
Chiles and Delfiner 2012; Pyrcz and Deutsch 2014).
Geostatistical simulation methods produce multiple models or
realizations of properties of natural phenomena that attempt to
depict reality as close as possible from a probabilistic point of
view. The regridding and refinement of current realizations and
simulation on irregular grids is increasingly used in practice (Aziz
1993; Pyrcz and Deutsch 2014). For instance, the reserves evaluation model can be constructed on a coarse grid, but detailed
inspection of part of the reservoir may be required at higher resolution. Another example is when part of the reservoir, which is
more homogeneous or more informed, is modeled on a structured
grid. But areas around wells or more-heterogeneous regions are
modeled with irregular grids or on finer-resolution grids that do
not align with the rest of the model. Unfortunately, most of the
conventional geostatistical simulation methods of continuous
properties, such as sequential Gaussian simulation, are implemented on a grid of evenly spaced simulation nodes and, therefore, are grid-dependent (Deutsch and Journel 1998). Current
implementation of the conventional simulation methods does not
C 2016 Society of Petroleum Engineers
Copyright V

Original SPE manuscript received for review 13 April 2015. Revised manuscript received for
review 30 January 2016. Paper (SPE 180917) peer approved 15 February 2016.

guarantee that two or more models constructed on various sets of


the simulation nodes are consistent with each other.
The geologic properties of the reservoir model are constructed
conditional to the available data. The geological data come from
limited point-scale core analysis, log measurements, and exhaustively sampled block-scale seismic observations. The scales and
quantity of the data are different and must be reflected properly in
the model. In addition, petroleum reservoir characterization
involves modeling of several correlated attributes.
A new GFS method has been recently developed that can be
used in petroleum reservoir characterization for construction of
models of various node configurations and densities (Zagayevskiy
and Deutsch In press). The simulated unconditional realizations of
multiple correlated continuous properties are expressed as an analytical function of the coordinates of the simulation nodes at point
scale. The data values are honored by subsequent kriging-based
conditioning. The spatial structure of the modeled system is preserved. By doing so, the realizations are infinitely resolvable in the
entire simulation domain, and newly integrated data would change
current realizations only locally within the correlation range.
This paper demonstrates the practical application of GFS to a
large oil-sands reservoir, the Firebag oil-sands thermal project
operated by Suncor in northern Alberta, Canada. The data were
taken from public sources. The work was undertaken independently of the operator. The realizations of porosity, permeability,
and water saturation are generated on grids of various resolution
by facies to support future models for resource evaluation,
reserves estimation, and flow simulation. The reservoir is bounded
vertically by impermeable shale layers and is restricted horizontally by a lease area. The simulation model is conditioned to scattered point-scale hard core and log data, as well as exhaustively
sampled block-scale soft data in the form of AI. The intrinsic
cokriging (ICK) approach with projection, block-matrix inversion
(BMI), and pseudopoint-scale block (PSB) value representation
are used to assimilate block-scale seismic data with GFS to reduce
computational time and avoid artifacts in the realizations
(Zagayevskiy and Deutsch 2015). The final model of key petrophysical variables is reported at point scale.
The proposed GFS method requires that modeled continuous
geological properties be transformed from original units to normal
scores for simulation and the simulated values be returned from
normal scores back to original units (Deutsch and Journel 1998).
This normal scores transformation (NST) procedure matches
probability quantiles of data distribution and the normal distribution to find equivalent values in both distributions.
Modeling within stationary domains is assumed. This implies
that spatial distribution parameters (global mean and covariance
function) of a modeled property are constant throughout the stationary domain. In petroleum reservoirs, given that the geological
structure is well-defined, the most heterogeneity would come
from the facies distribution. Therefore, the facies define stationary
domains. The conventional simulation method for categorical
variable modeling, sequential indicator simulation (SIS), is used
to produce categorical models of stationary facies domains.
This paper is organized as follows. First, the workflow for
modeling static correlated geological properties with GFS is presented. Second, the theory of GFS is highlighted. Third, the case
study demonstrates modeling of reservoir architecture and

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:35 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Define primary and secondary variables to model.

Choose appropriate data and note the scale difference.

Transform chosen debiased data to normal scores.

Compute and model variogram for all variables and


their interactions in normal scores.

Define unconditional realizations in analytical form


as a function of the coordinates of the simulation
locations with proper target spatial structure.

Condition realizations to data by dual kriging and


express them in analytical form.

Generate simulated values at required locations.

Check statistics of the generated realizations.

Back transform realizations from normal


scores to original units.

Check data and variogram reproduction in


original units.

If required, scale up realizations to required scale.

Need regridding, grid


refinement or simulation at
additional locations?

Yes

No

Stop

Fig. 1Workflow for modeling deposits with GFS.

geological properties in a grid-free manner. Finally, conclusions


and future work conclude the paper.
Modeling of Natural Phenomena With GFS
The geomodeling workflow for natural phenomena characterization with GFS is consistent with conventional geomodeling procedures (Deutsch and Journel 1998; Pyrcz and Deutsch 2014).
The difference lies in how the output model is reported and how
post-processing proceeds. With conventional simulation methods, the model is constructed on a structured grid, whereas a set
of scattered distribution nodes of any configuration can be used
in GFS to generate a model of natural phenomena. Additional
simulation locations can be queried at any time with limited computational expense.
The proposed workflow is summarized in Fig. 1. The
improvements of GFS in comparison with conventional simulation methods are shown in colored boxes. The realizations are
2

Stage:

Page: 2

Total Pages: 15

constructed in analytical form and resolved later at required simulation locations of any configuration and density. Major points
of the multivariate GFS of natural phenomena can be summarized as follows.
Define variables to model on the basis of the project goals
and data availability.
* State primary variables for modeling.
* Study all available data, and note the scale difference in
the data.
* Determine the relationship between all variables from the
data.
* Choose secondary variables that are related to primary
variables and, thus, can be used to infer values of primary
variables
Process data.
* Note the relationship between all selected primary and
secondary variables.
* Transform data to normal scores.
Obtain spatial structure of the modeling system.
* Compute experimental variograms for all variables in all
possible directions.
* Define principal directions of the continuity.
* Fit variogram model.
Perform simulation in the normal units in a grid-free format
at the required simulation locations.
* Generate unconditional realizations as a function of the
coordinates of the simulation locations.
* Condition realizations to data values.
Retrieve conditionally simulated realizations resolved at any
set of the simulation node configurations and resolutions at
point scale.
Check target statistics in the unconditional simulation.
* Check normality of the realizations.
* Check variogram model reproduction.
* Check reproduction of the relationship between variables.
Check target statistics in the conditional simulation.
* Check data reproduction.
* Check data variogram reproduction.
* Check reproduction of the relationship between variables.
Post-process realizations.
* Transform realizations back to original units.
* Scale up point-scale realizations to a larger scale as
required.
Perform simulation at another set of locations if necessary.
Theory of GFS
Here, we present a brief description of the theoretical background
of the GFS method for modeling coregionalized variables at point
scale (Zagayevskiy and Deutsch 2015, In press). There are K correlated properties Zk (k 1, , K) to model in original units. The
point-scale data and pseudo-PSB values of large-scale data are
transformed from original units zdk;j ( j 1,, Nk; k 1,, K) to
normal scores ydk;j ( j 1,, Nk; k 1,, K), by use of the NST
algorithm, to have zero mean and unit variance and follow normal
distribution (Deutsch and Journel 1998). Here, Nk is the kth random variables number of the data. Thus, the modeled properties
are expressed as stationary random functions Yk (k 1, , K).
The pseudo-point-scale values are found from block-scale values
by linear interpolation and nodal-data adjustment. This technique
is defined as PSB-value representation by Zagayevskiy and
Deutsch (2015). The experimental variograms are computed next
in normal scores to fit the variogram model, which is later used in
the simulation. The expression for GFS is shown in Eq. 1, where
conditional realization i of cosimulated properties at location u is
stored in a vector form Yi u, as presented in Eq. 2. A total number of the realizations are denoted as NR.
Yi u A0 X0;i u

P1
X

Ap Xp;i u cY uT mi ;

p1

i 1; ; NR ;

                     1

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:35 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Yi u Y1;i u

YK;i u T ; i 1; ; NR :
                   2

Yk;i u

The expression in Eq. 1 is dependent on the linear model of


coregionalization and kriging-based conditioning (Journel and
Huijbregts 1978; Wackernagel 2003). There are P independent
random factors that are also stored in a vector form Xp;i u of size
K  1 for each realization i. The expression for these random factors is shown in Eq. 3:
Xp;i u Xp;1;i u Xp;k;i u Xp;K;i u T ;
p 0; ; P  1; i 1; ; NR :               3
The zero random factor X0,i(u) is a special one. It is called
white noise and has a nugget covariance structure; i.e., the zero
random factor is spatially uncorrelated with itself. The remaining
random factors p 1, , P1, are derived from a combination of
the turning bands and Fourier series simulation methods according to Eq. 4 (Journel and Huijbregts 1978; Chiles and Delfiner
2012; Zagayevskiy and Deutsch In press):
v
u D


Q u
L X
X
1
pqh
u1
t
dh
CX1 hcos
Xp;k;i u p
p;k:l
D
L l1 qQ D
D
pq

0
^
 cos
u  u l /p;k;l;q;i ;
D
p 1; ; P  1; k 1; ; K; i 1; ; NR :      4
The turning line processes l 1, ..., L, should be placed nearly
evenly in space. A total number of the Fourier series terms used
for the decomposition of 1D covariance function CX1 is 2Q 1.
p;k:l

The periodicity domain 2D is chosen arbitrarily, but it should be


larger than the simulation domain and the longest variogram
range. The scalar product u0  u^ l is the product between modified
simulation location u0 and unit vector u^ l of the turning line l. The
modification of the simulation location u is required to produce
anisotropic realizations with the turning bands. The stochastic nature of the simulation is defined through the fixed set of random
phases /p;k;l;q;i of finite size, distributed randomly between 0 and
2p according to uniform distribution. The lower-triangularweighting matrices Ap ( p 0, , P1) are computed from the
sill contribution matrices Bp ( p 0, , P1) of the input variogram model, as in Eq. 5:
Bp Ap ATp ;

p 0; ; P  1: . . . . . . . . . . . . . . . . 5

The dual Kriging weights mi are recomputed for each realization i according to Eq. 6, where C1
Y is the inverse of data/data covariance matrix, cY u is the data-location/simulation location u
covariance matrix, and DYdi is the vector with the difference
between data and simulated values of realization i at data locations that are also computed according to Eq. 1.
d
mi C1
Y DYi ;

i 1; ; NR : . . . . . . . . . . . . . . . . . 6

The assimilation of the point-scale data with GFS is straightforward through the kriging update. It is the last summation term
in Eq. 1. The integration of larger exhaustively sampled data sets
measured at larger block scale stipulates a need for computational
time reduction. For these reasons, ICK and BMI are implemented.
In addition to primary data, only secondary data collocated with
all primary data and simulation locations are used in conditioning
with ICK. The kriging matrix is partitioned with BMI. The removal of the boundary artifact around exhaustively sampled secondary data is carried on with projection approach to ICK. The
edge effect is addressed through PSB-value representation of the
block-scale data as an alternative to block cokriging (Zagayevskiy
and Deutsch 2015).

Stage:

Page: 3

Total Pages: 15

Therefore, the conditional realization Yi u is expressed in a


grid-free manner. The resulting simulated values are returned to
original units Zi u with the reverse procedure of NST.
Case Study: Firebag Project
A large volume of oil-sands reserves were found in northern
Alberta, Canada. The steam assisted gravity drainage (SAGD)
heavy oil extraction technique is widely implemented to extract the
bitumen when deposits are deeper than 200 m below the surface
(Butler 1991). A number of SAGD pads are built over the oil
deposits, from which several horizontal SAGD well pairs are
drilled. A SAGD well pair consists of a production well placed
roughly 5 m below an injection well. Both wells are placed as close
as possible to the reservoir base for increased oil production. Steam
is injected to lower the viscosity of the accumulated bitumen and
allow it to drain to the production well. For effective oil production, the SAGD wells should be placed in accordance with the geological setting of the reservoir. Numerical modeling of the
petrophysical properties of the entire reservoir is used for horizontal well placement (Pyrcz and Deutsch 2014). Subsequent grid
refinement or regridding is required for reserve estimation and flow
simulation. Therefore, the GFS technique can be implemented for
derivation of these models that are consistent with each other.
The geostatistical GFS algorithm is applied to the Firebag oilsands thermal project in northern Alberta, Canada, to show practical applicability of GFS for petroleum reservoir characterization
(Suncor 2013). Note that the proposed GFS method is dependent
on two-point statistics, which cannot capture complex curvilinear
features of geological depositions like a fluvial-deltaic system.
However, this method enables the building of a statistical representation of the geological deposition.
The objective of this realistic case study is to build three consistent petrophysical models of the Firebag subsurface geology.
The first model is built at a point scale for oil-reserves estimation
at coarse-resolution grid. The second model is constructed for
flow simulation around all available SAGD well pads at mediumresolution grid. The third model is simulated for flow simulation
around an individual SAGD well pad at a fine resolution. All three
models are conditioned to the same point-scale core data, log
data, and exhaustively sampled block-scale seismic measurements. Correlated realizations of porosity, permeability, and water
saturation models are built within the stratigraphic surfaces of the
reservoir by facies. All the data used in this case study were taken
from public records and processed as required. No data or information were provided by the operator.
Project Overview. The Firebag oil-sands thermal project is in
northern Alberta, Canada, approximately 100 km northeast of
Fort McMurray, as shown in the left part of Fig. 2. This project is
operated by Suncor Energy (Suncor 2013). The SAGD technology
is used to extract the bitumen. The target reservoir is in the
McMurray Formation. The lease area covers approximately
15  15 km2. The numerical data have been obtained from the
geoSCOUT database software in Log American Standard Code
for Information Interchange Standard format (geoSCOUT 2014).
The wells within the lease area are represented by 233 SAGD
well pairs, 545 delineation wells (dark blue), 180 observation
wells (light blue), and 31 miscellaneous-service wells (red), totaling to 1,222, as shown in right part of Fig. 2. There are three main
sources of information available. Fig. 3 shows well locations with
point-scale core and log data along with the block-scale seismicsurvey coverage. Not all wells have all data variables.
The stratigraphy of the lease area can be described as shown in
Fig. 4 (Suncor 2013). The major elements are the Muskeg formation; glacial till; the Clearwater Formation, which represents the
top seal of the reservoir and consists mostly of shale; and the
McMurray Formation, which is the target reservoir with good reservoir properties. It consists of shoreface, tidal flat, channel complex, and continental parts, with channel complex being the mosteconomical and feasible strata from which to produce. Note that
four facies are distinguished in the McMurray formation:

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:35 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Stage:

Page: 4

Total Pages: 15

Plan view

Alberta province

Canada

Firebag project
Miscellaneous-service
wells

Fort McMurray

SAGD wells
Observation wells

Edmonton

Delineation wells
Y
X
Z

Fig. 2Map of the geographical location of the Firebag oil-sands thermal project, which is shown by a red star (left), and well trajectories at the Firebag lease area (right). Total well number is 1,222233 SAGD well pairs (233 production wells and 233 injection
wells) shown in yellow, 545 delineation wells shown in dark blue, 180 observation wells shown in light blue, and 31 miscellaneousservice wells shown in red.

Firebag Well Locations with Core Data

6,352,000

Firebag Well Locations with Log Data

862,000

877,000

East (m)

6,352,000

Firebag Siesmic-Data Locations

North (m)

6,367,000

North (m)

6,367,000

North (m)

6,367,000

862,000

877,000

East (m)

6,352,000

862,000

877,000

East (m)

Fig. 3Plan view of data locations of 302 wells with core data, 224 wells with log data, and seismic coverage resolution for all facies. Not all shown wells have measurements of the entire set of modeled geologic reservoir properties.

impermeable shale, permeable sand, semipermeable inclined heterolithic strata (IHS), and permeable breccia. The final elements
are the eroded Devonian carbonate depositions below the McMurray formation.
On the basis of this reservoir description and data availability,
the key primary and secondary variables for modeling are the
following.
Geological facies: sand, shale, breccia, and IHS
Structural geology: reservoir top and base of the McMurray
formation
Petrophysical properties: porosity / by facies, permeability
(horizontal permeability j) by facies, fluid saturation by fa-

cies (water saturation s, where oil saturation is the remaining


fraction), and the volume fraction of shale mineral (Vshale)
by facies
Petroelastic rock attribute in the form of seismic AI
Workflow. A geomodeling workflow of modeling the Firebag
reservoir with GFS is presented. The first module involves data
analysis. The second module is related to actual generation of the
geostatistical reservoir model in a grid-free manner.
The essential parts of the data analysis are explained here, taking into account data availability for the Firebag oil-sands thermal

McMurray
Formation
Top

Surface

Delineation, observation,
or service vertical well

Shale

Sand

IHS

Clearwater Formation
(mainly shale)

Breccia

McMurray Formation
(shale, sand, IHS, breccia)

Devonian Age
(mainly shale)

Horizontal SAGD well pair

McMurray
Formation
Base

Fig. 4Stratigraphy of the Firebag project area in cross-sectional view and schematic of wells (left), and facies classification
example of Well 1AA102009405W400 (right), where GR- and PE-log data are used for the facies classification.
4

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:35 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

project. The geological attributes to model are conditioned to


available primary data and secondary seismic attributes. Measurements of porosity, horizontal permeability, and water saturation
are available as core data. Gamma ray (GR), photoelectric effect
(PE), and porosity logs are provided for modeling reservoir boundaries, facies, petrophysical properties, and seismic attributes.
Note that the core data are direct measurements of the petrophysical properties of the rock. On the other hand, the log data are indirect observations of the petrophysical properties. Because core
data are more reliable than the available log data, the log data are
corrected with the core data and used in GFS. No seismic data are
available, but are generated synthetically.
Step 1: Perform facies classification on the basis of the welllog data around approximate boundaries of the McMurray
formation. Define facies with distinct petrophysical properties and use as separated stationary modeling domain.
Step 2: Define McMurray formation top and base elevations
from facies classification to establish vertical boundaries of
the modeling domain.
Step 3: Check core data of porosity, permeability, and water
saturation by facies and look for potential measurement
errors and outliers within the modeling domain.
Step 4: Derive Vshale values from GR-log data.
Step 5: Select log-porosity data for each facies, and use
Vshale data to correct log-porosity data.
Step 6: Infer pseudo-log permeability and water saturation
values from log-corrected log-porosity data, assuming permeability and water saturation have similar spatial structure
as porosity and by use of the core-data relationship between
porosity and permeability and porosity and water saturation.
Step 7: Generate synthetic seismic data (AI) for part of the
modeling domain by use of the Gassmann fluid-substitution
model (Kumar 2006). Note the scale difference between log
and seismic data. The seismic data are supposed to be
sampled at a larger block scale than the log data.
Step 8: Transform data to normal scores. Consider keeping
the relationship between the variables after the simulation is
performed.
Step 9: Construct variogram models for all variables.
The simulation and model checking are performed after the
data analysis, as explained here.
Step 1: Perform simulation of reservoir-top and -bottom elevations. Define boundaries of the reservoir.
Step 2: Perform simulation of geological facies within the
reservoir boundaries. SIS is used here.
Step 3: Simulate key petrophysical properties within corresponding facies in normal space by use of GFS, conditional
to all log data and AI.
Step 4: Back transform the simulated normal scores to the
original units.
Step 5: Validate the data reproduction, statistical and spatial
properties (variogram for two-point statistics) reproduction,
preservation of the variables relationship in normal scores
and original units, and consistency of the numerical model
with the conceptual model.
Step 6: Potentially post-process the model to meet the study
objectives (e.g., estimate reserves, evaluate economic feasibility of the area, or run the model through the flow
simulation).
Data Analysis. The available data are analyzed according to the
workflow explained previously. More details are provided in the
following subsections.
Facies Classification. The GR and PE logs are used to define
the facies types (Ellis and Singer 2008). The sand facies is characterized by low-GR and low-PE values. The shale facies has high
readings of GR and PE. The IHS facies is closer to sand, and breccia facies is closer to shale in their log properties. On the basis of
these characteristics, the following algorithm is applied to assign
facies dependent on the GR- and PE-log values. If GR reading is
lower or equal to 22.0 R, the sand facies is assigned to the corre-

Stage:

Page: 5

Total Pages: 15

sponding interval of the rock. If GR is larger than 65.0 R, the


shale facies is identified. For the GR readings between 22.0 and
65.0 R, PE measurements are used in addition to classify the facies types. If the PE reading is less than or equal to 2.10, the sand
facies is defined. The IHS facies is assigned to a rock interval of
PE reading between 2.10 and 2.35. The breccia facies assumingly
has PE between 2.35 and 3.20. Higher PE readings correspond to
shale facies.
When the proposed-facies classification algorithm is applied to
the wells, a possible stratigraphy is defined along the well bores.
As an example, facies defined in Well 1AA102009405W400 are
shown in Fig. 4. They appear to be in a compliance with the stratigraphic conceptual model presented in Fig. 4. The global-facies
proportions are 0.30, 0.54, 0.13, and 0.03 for shale, sand, IHS,
and breccia, respectively. The horizontal and vertical experimental variograms of sand facies derived from all classified-facies
data with corresponding variogram model are shown in Fig. 5.
The variograms of other facies are similar to the sand-facies variogram. The fitted variogram models in the form of the covariance
function are depicted in Eq. 7 for all facies. These variogram
models are used to generate the facies model with SIS for subsequent by-facies simulation of petrophysical properties of the petroleum reservoir.
8
Cshale h 0:10sph r 400 m h
>
>
h
>
>
>
>
>
r
v 2:6 m
>
>
>
>
h

0:08exp
>
>
rh 1300 m
>
>
>
>
>
rv 25 m
>
>
>
>
>
h

0:12sph
h
C
sand
>
rh 300 m
>
>
>
>
>
rv 2:5 m
>
>
>
>
<
0:10exp r 1000 m h
h
:
Cfacies h
>
>
rv 30 m
>
>
>
>
>
> CIHS h 0:07sph rh 500 m h
>
>
>
>
>
rv 1:5 m
>
>
>
>
>
h

0:04exp
>
rh 550 m
>
>
>
>
>
rv 12 m
>
>
>
>C
>
>
breccia h 0:024exp r 300 m h
>
h
>
>
:
rv 2 m

                   7
Reservoir Surfaces Definition. The target petroleum reservoir comprises the McMurray Formation rock. The stratigraphic
top and base elevations of the reservoir are found by use of facies definition. It is deemed that the McMurray Formation is
bounded by two impermeable layers of shale deposition above
and below. These shale intervals are represented by the Wabiskaw (or Clearwater) Formation overlying the McMurray Formation and by eroded Devonian rock underlying the McMurray
Formation. An example of these shale intervals in Well
1AA102009405W400 is shown in Fig. 4. A total of 224 wells
with classified facies are available. The picks of top and base
are performed for all wells in the same way as for Well 1AA102
009405W400.
The target reservoir top is easier to distinguish from the facies
because of the uninterrupted geological deposition. The interpreted elevation of the reservoir top and reservoir thickness (reservoir-top elevation minus reservoir-base elevation) are treated as
surface data. Once this elevation and thickness are modeled with
GFS, the modeled elevation of reservoir base is found by subtracting modeled thickness from the modeled reservoir-top elevation.
Fig. 6 contains the interpolated Kriging maps of three structural
elements of the reservoir (elevations of reservoir top and base and
reservoir thickness) to better understand the formation.
The data histograms of the structural elements are presented in
Fig. 7. The correlation coefficients between McMurray Formation

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:35 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Stage:

Page: 6

Horizontal Variogram

Total Pages: 15

Vertical Variogram

0.30

0.30

0.25

0.25

0.20

0.20

0.15

0.15

0.10

0.10

0.05

0.05
0.00

0.00
0

500

1000

1500

2000

2500

3000

0.0

4.0

8.0

Distance (m)

12.0

16.0

20.0

Distance (m)

Fig. 5Experimental horizontal and vertical variograms and associated model for McMurray formation sand facies; experimental
variograms are shown by red dots, and variogram model is presented by solid red line.

McMurray Top Elevation (m)

McMurray Base Elevation (m)

McMurray Formation Thickness (m)

6,367,000

6,367,000
305.00

100

330

286.25

75

310
290

267.50
248.75

270
6,352,000
862,000

50
25

230.00
6,352,000
862,000

877,000

North (m)

350

North (m)

North (m)

6,367,000

East (m)

0
6,352,000
862,000

877,000

East (m)

877,000

East (m)

Fig. 6Kriging maps of the McMurray formation top elevation, base elevation, and thickness.

Frequency

0.05

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.04
0.03
0.02
0.01
0.00
296

306

316

326

336

348.98
339.85
334.36
325.50
296.30

346

McMurray Formation Thickness


0.06

Number of data 224


Mean 272.49
Standard deviation 13.73

0.06

Frequency

Number of data 224


Mean 331.54
Standard deviation 10.92

0.06

McMurray Formation Base Elevation


0.07
0.05

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.04
0.03
0.02
0.01

0.05

Frequency

McMurray Formation Top Elevation


0.07

302.75
283.40
273.47
262.10
234.70

224
5.92
17.21

Maximum
Upper quartile
Median
Lower quartile
Minimum

93.80
71.68
59.20
48.10
13.12

0.04
0.03
0.02
0.01

0.00
234 244 254 264 274 284 294 304

Elevation (m)

Number of data
Mean
Standard deviation

0.00
13 23 33 43 53 63 73 83 93

Elevation (m)

Thickness (m)

Fig. 7Histograms of the McMurray formation top elevation, base elevation, and thickness data.

top elevation and formation thickness are 0.60 m and 0.54 in normal scores, respectively. The experimental direct and cross variograms for normal scores of reservoir-top elevation and thickness
are depicted in Fig. 8. The associated variogram model is
expressed in Eq. 8:

Variogram of McMurray Top


Elevation, Normal Scores

Variogram of McMurray
Thickness, Normal Scores

1.00

1.00

0.80

0.80

0.60

0.60

0.40

0.40

0.20

0.20

0.00

8
>
< CYtop h 0:20expr3000 m h
Csurfaces h CYtop Ythickness h 0:09expr3000 m h :
>
:
CYthickness h 0:70expr3000 m h

1000 2000 3000 4000 5000

Distance (m)

Cross-Variogram of McMurray Top


Elevation and Thickness, Normal Scores
0.40
0.30

0.20
0.10

0.00
0

                   8

1000 2000 3000 4000 5000

Distance (m)

0.00
0

1000 2000 3000 4000 5000

Distance (m)

Fig. 8Experimental omnidirectional direct and cross variograms shown by red dots and resulting variogram model presented by
red line for the McMurray formation top elevation and thickness.
6

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:35 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Maximum
Upper quartile
Median
Lower quartile
Minimum
100

1,000

0.02

0.189

Permeability (md)

Maximum
Upper quartile
Median
Lower quartile
Minimum

13544.0
8120.0
5369.9
2717.5
101.0

100

1,000

10,000 100,000

0.04

0.10

0.20

Maximum
Upper quartile
Median
Lower quartile
Minimum
100

1,000

10000.0
7055.0
4010.0
2185.0
15.2

0.46
0.36
0.35
0.33
0.00

0.02

0.00
0.01

0.06

0.44
0.36
0.34
0.31
0.10

0.00
10

100

1,000

10,000

0.41

0.61

0.81

0.01

IHS - Water Saturation Core Data

Mean 0.39
Standard deviation 0.23

0.04
0.03

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.02
0.01

0.20

0.296

0.03

0.396

0.99
0.58
0.34
0.19
0.03

Mean 0.32
Standard deviation 0.04
Maximum
Upper quartile
Median
Lower quartile
Minimum

0.04
0.02

0.42
0.35
0.33
0.30
0.17

0.00
0.172 0.222 0.272 0.322 0.372 0.422

Permeability (md)

0.43

0.63

0.83

0.103

Breccia - Water Saturation Core Data

Mean of data 246

0.06

0.23

Water Saturation

Frequency

10000.0
7830.0
4240.0
2642.5
51.7

0.21

0.95
0.36
0.20
0.14
0.01

Mean of data 1296

Breccia - Porosity Core Data

Frequency

Frequency

Maximum
Upper quartile
Median
Lower quartile
Minimum
1

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.04

Porosity

Mean 4916.2
Standard deviation 2998.5

0.04

Mean 0.28
Standard deviation 0.20

0.06

0.02

0.08

0.08

Mean of data 2017

0.00
0.196

Mean of data 35

0.12

0.90

Water Saturation

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.04

Breccia - Permeability Core Data

0.70

Sand - Water Saturation Core Data

0.08

0.40

Mean 0.33
Standard deviation 0.04

0.06

0.00
0.096

0.50

0.05

0.08

10,000

0.30

Mean of data 1722

Permeability (md)

0.16

0.10

IHS - Porosity Core Data

0.00
10

0.30

0.10

Frequency

Frequency

Mean 4527.7
Standard deviation 2975.9

0.00
0.10

Porosity

Mean of data 227

0.04

0.02

0.99
0.78
0.65
0.50
0.10

Water Saturation

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.08

IHS - Permeability Core Data

0.08

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.03

0.01

0.389

Mean 0.34
Standard deviation 0.03

Permeability (md)

0.12

0.41
0.33
0.30
0.27
0.09

Mean of data 3891

0.12

0.00
0.00

0.00
10

0.289

0.16

Frequency

Frequency

Mean 5418.2
Standard deviation 3133.6

0.04

Mean 0.61
Standard deviation 0.22

0.04

Sand - Porosity Core Data

Mean of data 561

0.08

Mean 0.30
Standard deviation 0.04

Porosity

Sand - Permeability Core Data


0.12

Mean of data 428

0.05

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.03

0.00
0.089

0.06

Frequency

10

0.04

0.01

0.00
1

0.05

Shale - Water Saturation Core Data

Mean of data 888

Frequency

0.04

489.0
133.0
85.1
27.00
6.0

0.06

Frequency

Frequency

0.12

0.08

Total Pages: 15

Shale - Porosity Core Data


0.07

Mean of data 44
Mean 120.6
Standard deviation 129.9

Page: 7

Frequency

Shale - Permeability Core Data

Stage:

0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0.00
0.05

Porosity

Mean of data 224


Mean 0.37
Standard deviation 0.24
Maximum
Upper quartile
Median
Lower quartile
Minimum
0.25

0.45

0.65

0.97
0.54
0.30
0.16
0.05

0.85

Water Saturation

Fig. 9Histograms of core data of porosity, permeability, and water saturation for each facies.

Because the McMurray formation top is relatively undisturbed


and the McMurray formation base is deposited on an eroded surface, the adjustments to vertical coordinates are made for all subsequent calculations relative to the McMurray formation top to
make sure that data at similar adjusted elevations correspond to
the similar stratigraphic time within each well (Schneider and
Grobe 2013). The vertical coordinate z of a location u(x, y, z) in
the space is adjusted to coordinate z0 , as shown in Eq. 9, where
ztop(x, y) is the elevation of the McMurray formation just above
location u. Note that the relative vertical coordinates z0 are negative. To get original elevations z back, Eq. 10 is used.
z0 z  ztop x; y; . . . . . . . . . . . . . . . . . . . . . . . . . . 9
z z0 ztop x; y: . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Thus, the vertical boundaries of the model are defined to be
within the McMurray formation, and horizontal boundaries are
defined by the lease area extent. The modeling is performed relative to the McMurray formation top.

Core Data Analysis. The core data are analyzed by facies.


The variables to be analyzed from core and log data are porosity,
permeability (horizontal permeability, because there are no reliable data on vertical permeability), and water saturation. No gas
is assumed to be present in the reservoir, and therefore the oil saturation is simply the remaining rock porous volume after water
saturation is subtracted.
The by-facies histograms of cleaned porosity, permeability,
and water saturation are shown in Fig. 9. The cleaning involves
removal of the outliers and suspiciously large or small data values. The sand facies, breccia, and IHS have the highest permeability. The shale facies has lower flow quality relative to the other
three facies. The average porosity values are approximately the
same for all four facies. The water saturation is the highest in the
shale, the lowest in the sand, and in between in the IHS and breccia. The porosity and permeability core data are linearly correlated for all facies with correlation coefficients of 0.71, 0.70, 0.88,
and 0.87 for shale, sand, IHS, and breccia facies, respectively.
The porosity and water saturation core data inversely correlated

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:35 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Frequency

Frequency

0.20

Page: 8

Total Pages: 15

Sand - Neutron-Grouped-Corrected-Porosity-Log Data


Number of data 1289
0.16
Mean 0.34
Standard deviation 0.03

Shale - Neutron-Grouped-Corrected-Porosity-Log Data


Number of data 51
Mean 0.12
Standard deviation 0.03
0.30
Maximum 0.16
Upper Qurtile 0.14
Median 0.13
Lower Quartile 0.12
Minimum 0.05

Stage:

Maximum 0.40
Upper quartile 0.36
Median 0.34
Lower Quartile 0.31
Minimum 0.13

0.12

0.08

0.10
0.04
0.00
0.000

0.100

0.200

0.300

0.400

0.00
0.000

0.500

Porosity

0.12

Number of data 793


Mean 0.32
Standard deviation 0.04

0.300

0.400

0.500

Breccia - Neutron-Grouped-Corrected-Porosity-Log Data

0.08

Maximum 0.42
Upper Qurtile 0.35
Median 0.32
Lower Quartile 0.29
Minimum 0.23

0.04

Number of data 275


Mean 0.31
Standard deviation 0.04

0.12

Frequency

Frequency

0.200

Porosity

IHS - Neutron-Grouped-Corrected-Porosity-Log Data

0.00
0.000

0.100

Maximum 0.42
Upper Qurtile 0.34
Median 0.30
Lower Quartile 0.28
Minimum 0.22

0.08

0.04

0.100

0.200

0.300

0.400

0.500

0.00
0.000

0.100

0.200

0.300

0.400

0.500

Porosity

Porosity

Fig. 10Histograms of grouped corrected neutron porosity log data by facies.

with correlation coefficients 0.58, 0.22, 0.37, and 0.55 for


shale, sand, IHS, and breccia facies, respectively.
Log Data Analysis. The log data are selected for GFS-model
conditioning because of wider coverage, greater data number, and
denser sampling rate along the wells. The data to derive from log
measurements represent porosity, permeability, and water saturation variables.
First, the neutron porosity / is corrected by Vshale, as shown in
Eq. 11. The Vshale is derived from GR as expressed in Eq. 12,
where GRmin and GRmax are the minimum and maximum readings
of the gamma ray along the wellbore, respectively.
/0 /1:0  Vshale ; . . . . . . . . . . . . . . . . . . . . . . . 11
Vshale

GR  GRmin

: . . . . . . . . . . . . . . . . . . . . . 12
GRmax  GRmin

Then, corrected-neutron-porosity data /0 are classified by facies


and calibrated to core data discussed previously. The mean of the
corrected-porosity-log data is matched to the mean of the porosity
core data for all facies except for shale. The mean of the porositylog data for shale is intentionally made lower because it is believed
that core data are preferentially sampled in better-quality rock.
Given a high sampling density and large number of the log data,
single porosity value from each 5.0-m interval is selected randomly
to be a representative of the entire interval. Corresponding by-facies histograms of the porosity-log data are shown in Fig. 10.
Only log-porosity data are provided. The permeability and
water saturation (unsampled variables) are derived from the porosity-log data by use of the cloud transformation/P-field simulation technique (Srivastava 1992). In cloud transformation, values
are drawn from the conditional distributions between porosity and
permeability (and porosity and water saturation) derived from the
core data. To do so, first correlated probabilities are simulated at
the well locations honoring spatial structure of the unsampled
variable. Second, the simulated values of the unsampled variables
are drawn from the conditional distributions by use of these correlated probabilities and collocated values of the porosity-log data.
The probability fields for permeability and water saturation are
generated with the GFS method by use of variogram models com8

puted from the core data of permeability and water saturation for
sand facies.
The distribution parameters of core and log data match each
other quite well for all facies except shale, where log-porosity values are lower than core-porosity values and are deemed more realistic. The derived pseudo-log data of permeability and water
saturation are not unique.
The experimental direct and cross variograms for these three logdata types of sand facies in normal scores are presented in Fig. 11 for
the horizontal direction. The variogram model has the form shown
in Eq. 13. It is an exponential variogram type with isotropic horizontal component with a range of 600 m and a vertical range of 8 m.

CY h

8
CY/ h 0:93exp r 600 m h
>
>
h
>
>
>
>
rv 8 m
>
>
>
>
>
CY/ Yj h 0:69exp r 600 m h
>
>
h
>
>
>
>
>
rv 8 m
>
>
>
>
h

0:02exp
h
C
>
Y
Y
/ s
>
rh 600 m
>
>
>
<
r 8m
v

>
CYj h 0:93exp r 600 m h
>
h
>
>
>
>
>
rv 8 m
>
>
>
> C h 0:09exp
>
h
>
Y
Y
j
s
rh 600 m
>
>
>
>
>
>
rv 8 m
>
>
>
>
h

1:00exp
h
C
>
>
rh 600 m
> Ys
>
:

rv 8 m

                   13
Seismic Data Generation. A seismic survey was conducted
over part of the lease area, but these data are not available to the
public. Synthetic seismic data is generated to make the case study
more realistic in the use of exhaustively sampled secondary data.
The Gassmann fluid-substitution model is adopted to generate
synthetic AI on the basis of the porosity, water saturation, and
Vshale log data (Kumar 2006; Zagayevskiy 2012). These variables
2016 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 02:36 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Horizontal Variogram of Sand Porosity


Normal Scores

1.20

Stage:

1.20

1.00

1.00

0.80

0.80

0.60

Total Pages: 15

Horizontal Variogram of Sand Permeability


Normal Scores

0.60

0.40

0.40

0.20

0.20

0.00

Page: 9

0.00
0

500

1000

1500

2000

2500

3000

500

Distance (m)

1000

1500

2000

2500

3000

Distance (m)

Horizontal Variogram of Sand Water Saruration


Normal Scores

1.20

Horizontal Variogram of Sand Porosity Permeability Normal Scores

1.00
0.40
0.80

0.00

0.60
0.40

0.40
0.20
0.80

0.00
0

500

1000

1500

2000

2500

3000

500

Distance (m)

1000

1500

2000

2500

3000

Distance (m)

Horizontal Variogram of Sand Porosity Water Saruration Normal Scores

0.200

Horizontal Variogram of Sand Permeability Water Saruration Normal Scores

0.0200
0.100
0.0100

0.0000

0.000

0.0100
0.100
0.0200
0.0300

0.200
0

500

1000

1500

2000

2500

3000

Distance (m)

500

1000

1500

2000

2500

3000

Distance (m)

Fig. 11Experimental direct and cross variograms as red dots and associated variogram model shown by red line for normal
scores of porosity, permeability, and water saturation of sand facies in horizontal direction.

were calculated for all facies, because seismic surveys measure


acoustic properties independent of facies.
The compressional wave (P-wave) AI is computed from the
elastic bulk modulus, elastic shear modulus, density of rock minerals and constituent fluids, and rock porosity. The saturated and
dry rocks are distinguished. It is assumed that the rock matrix is a
binary mixture of the clay and quartz minerals of varying concentrations defined by Vshale. It is also assumed that no gas is present
in the reservoir; the reservoir fluid comprises brine reservoir water
and heavy viscous oil. The log data locations used to generate the
AI are shown in Fig. 12 in a plan view. The horizontal and vertical experimental variograms are computed from log porosity,
Vshale, and water saturation-log data.
The simulation of porosity, Vshale, and water saturation is performed with the GFS method in normal space at a point scale to
provide the inputs to the Gassmann fluid-substitution model. The
seismic coverage area has been shown previously in Fig. 3. The
simulation is performed on a grid with adjusted vertical coordinates to align with the formation top.
First, the seismic attributes are computed at a point scale from
the simulated input variables according to the Gassmann fluid-

substitution model. The number of the point-scale seismic data


grid is 200,000 (100  100  20) blocks of 50  50  5-m node
spacing. Then, this point-scale AI is arithmetically averaged over
200  200  20-m3 cubes to obtain block-scale seismic data. Note
that in reality, AI is not an additive attribute. Thus, it is assumed
that the resulting AI is resolved at a block scale of 200 
200  20-m3 volume.
The histogram of the output AI data at a block scale is shown
in Fig. 13, which is a normal distribution. The AI is highly inversely related to porosity. The correlation between AI and Vshale or
water saturation is weak. The relationships between input variables are preserved and compared with the data. The derived AI
data are shown in 3D space in original units at block scale in Fig.
12. The experimental variogram of the synthetic seismic attributes
at a point scale in normal scores for vertical and horizontal directions and the suggested variogram model are presented in Fig. 14.
The variogram structure derived for the key petrophysical properties also fit the AI in normal scores.
The final covariance model for the GFS simulation of the porosity, permeability, water saturation, and AI in normal scores is
presented in Eq. 14. The contributions of cross variograms of AI

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:36 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Well Locations Used in


Seismic Data Generation

Stage:

Page: 10

Total Pages: 15

Synthetic Seismic Dats


at Block Scale
6

6,367,000

10 Pas/m
Lease area

6.5
6.0

Seismic coverage

North (m)

5.5
5.0
4.5
Y
Z

6,352,000

877,000

862,000

East (m)
Fig. 12Plan view of well locations, log data from which are used for the synthetic AI generation (left), and a plan view of synthetic
seismic data in original units at block scale (right).

with other variables are chosen approximately on the basis of the


correlation coefficients between AI and the rest of the variables.
8
CY/ h 0:93exp r 600 m h
>
h
>
>
>
>
>
rv 8 m
>
>
>
>
>
CY/ Yj h 0:69exp r 600 m h
>
h
>
>
>
>
>
rv 8 m
>
>
>
>
>
CY/ Ys h 0:02exp r 600 m h
>
h
>
>
>
>
>
rv 8 m
>
>
>
>
>
C
h

0:90exp
h
Y
Y
>
/ AI
rh 600 m
>
>
>
>
>
rv 8 m
>
>
>
>
>
C
h

0:93exp
h
Yj
>
rh 600 m
>
>
>
<
rv 8 m
CY h
:
>
CYj Ys h 0:09exp r 600 m h
>
h
>
>
>
>
>
rv 8 m
>
>
>
>
>
CYj YAI h 0:50exp r 600 m h
>
h
>
>
>
>
>
rv 8 m
>
>
>
>
>
h

1:00exp
h
C
Y
s
>
rh 600 m
>
>
>
>
>
rv 8 m
>
>
>
>
>
h

0:20exp
h
C
Ys YAI
>
rh 600 m
>
>
>
>
>
rv 8 m
>
>
>
>
>
C
h

1:00exp
h
>
rh 600 m
> YAI
>
:
rv 8 m

                   14

Frequency

0.20
0.15
0.10

Number of data 200,000


Mean 5.33
Standard deviation 0.29
Maximum
Upper quartile
Median
Lower quartile
Minimum

6.82
5.51
5.32
5.13
4.23

0.05
0.00
3.74 4.74 5.74 6.74 7.74 8.74 9.74 10.74

AI (106 Pas/m)
Fig. 13Histogram of derived AI data at block scale.
10

Simulation Results. The key petrophysical properties including


porosity, permeability, and water saturation are modeled in a gridfree manner within the McMurray formation reservoir, conditional to corrected log data presented at point scale and synthetic
AI resolved at a block scale. The modeling is performed within
each facies independently.
Three grids are generated to meet the objectives of the case
study. The key point is to generate consistent realizations of the
Firebag reservoir model at various resolutions that would be part
of the same random function. A single realization is shown for
illustration purposes for each grid type. The first model is resolved
at coarse resolution for reserve estimation. The second model is
constructed at medium resolution for flow simulation within the
high-quality reservoir zone. The third model is built at fine resolution around a SAGD well pad in the upper-right corner. The corresponding grid resolutions are 200  200  20 m, 100  100 
10 m, and 20  50  3 m, respectively. The projection of the grid
nodes on a plan view is shown in Fig. 15.
The number of the turning lines used in the simulation is 161.
The number of the Fourier coefficients, 3,161, is selected to keep
the covariance approximation error less than 1%.
Reservoir Surfaces Simulation. Reservoir-top elevation and
reservoir thickness are generated with the GFS method at the
grid-node projections shown in Fig. 15. The variogram model of
the correlated reservoir-top elevation and thickness were presented previously in Eq. 8 and Fig. 8. The reservoir base is
derived from simulated reservoir top and thickness. The resulting
elevations and thickness are shown in Fig. 16. These boundaries
are used to limit the simulation grids in the vertical dimension.
The simulation of key geological variables is performed relative
to the simulated reservoir top. Later, vertical coordinates z0 are
adjusted back to original elevations z, according to Eq. 10.
Facies Simulation. A realization of facies is generated with
SIS according to the variograms in Eq. 7 to obtain four stationary
domains for petrophysical-properties simulation. The resulting
conditional simulation is presented in Fig. 17. Because the SIS
method considers only two-point statistics, complex curvilinear
features that may be expected in a channelized formation are not
prominently seen in the realization (Suncor 2013).
Geological Properties Simulation. The simulation of the porosity, permeability, and water saturation at a point scale and
complementary AI at a block scale is performed by facies on the
selected three grids. The models are conditioned to point-scale
log data and blocks-scale seismic data with the covariance structure shown in Eq. 14. The difference in the scale of the seismic
data is addressed through the PSB-value-representation method.
The ICK with data and simulation locations projection is used to
assimilate exhaustively sampled AI. The simulation is performed
2016 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 02:36 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Horizontal Variogram of
Synthetic Acoustic Impedance

1.20

Stage:

1.00

0.80

0.80

0.60

0.60

0.40

0.40

0.20

0.20

0.00
1000

2000

3000

4000

0.00
0.0

5000

Total Pages: 15

Vertical Variogram of
Synthetic Acoustic Impedance

1.20

1.00

Page: 11

4.0

8.0

Distance (m)

12.0

16.0

20.0

Distance (m)

Fig. 14Experimental horizontal and vertical variograms shown by red dots of the synthetic AI data at a point scale in normal
scores and suggested variogram model shown by red line.

water saturation. Resulting models of the porosity are shown in


Fig. 21 by facies and for all facies at once. Simulated values
within the seismic data look more continuous than in the other
parts of the simulation domain because of the continuous nature
of the soft block-scale seismic data. The hard data have been honored. The permeability, water saturation, and AI models are
resolved in the same way as the porosity models. The computational time to run a single realization of these four variables for
sand facies at 7,285 nodes conditional to 6,469 point-scale data
and 3,125 block-scale data is approximately 1 hour by use of a
64-bit Windows machine with Intel i7 processor of 2.8 GHz and
24.0 GB RAM.

Fig. 15Plan view of the simulation grid nodes projections at


coarse (Grid 1), medium (Grid 2), and fine (Grid 3) resolutions
with well trajectories.

in the normal space, before which the data values are transformed
to the normal space by use of NST.
As an example, the simulation results of porosity, horizontal
permeability, water saturation, and AI for the sand facies over the
sparsest simulation domain (Grid 1) are shown in Figs. 18 and 19
in original units. The univariate and bivariate distributions of the
variables are honored. The variogram model (red line) and experimental variograms computed from simulated normal scores
resolved on Grid 1 in the horizontal plane are shown in Fig. 20.
The variogram model is well reproduced except for the cross variogram between porosity and water saturation, which could be
explained by the low covariance value between porosity and
Reservoir-Top Elevation

Discussion. The GFS approach is demonstrated to be a practical


method for petroleum reservoir characterization. In this case
study, the developed simulation method was applied in three different circumstances. The probabilities within the wells for calculation of pseudo-log permeability and water saturation values,
target reservoir-top and -base elevations, and geological properties resolved at various resolutions were computed in a grid-free
manner. The simulation remains consistent regardless of the simulation grid density and orientation. The data values and data distribution are honored in final realizations. The variogram is
reproduced quite accurately. The modeling results presented in
this work prove the potential of the GFS technique for application
to real reservoirs.
Conclusions
GFS has been applied to the Firebag oil-sands thermal project in
northern Alberta, Canada, for modeling vital geological properties
of the reservoir. The porosity, permeability, water saturation, and
AI have been simulated by facies on three grids of various resolutions conditional to point-scale core and log data and block-scale
seismic attributes.

Reservoir-Base Elevation

350

Reservoir Thickness

350

330

330

330

310

310

310

290

290

290

Y
Z

350

Y
X

Y
X

Fig. 16Grid-free realizations of the McMurray formation top elevation, base elevation, and reservoir thickness resolved on three
grids.
2016 SPE Reservoir Evaluation & Engineering
ID: jaganm Time: 02:36 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

11

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Facies Model in 3D

Shale

Sand

IHS
Z
Breccia

Y
X

Fig. 17Single realization of the categorical facies model for


stationary-domains definition.

All three Firebag models of reservoir architecture and petrophysical properties at coarse, medium, and fine scales are consistent with each other. This is an outstanding feature of the GFS
method in comparison with conventional geostatistical simulation
methods. Regridding and refinement of realizations is straightforward to perform with GFS. For instance, the finer-scale third
model provides more insight into the understanding of the reservoir properties and reservoir architecture around the selected
SAGD pad because of the increased resolution. It is a magnified
version of previously simulated coarse-scale models. Also, simultaneous conditioning to point-scale and block-scale data allows
for the GFS method to stand out from the conventional geostatistical methods.
Thus, the GFS method has significant potential to be applied
extensively in the practice of petroleum reservoir characterization.
Future Work
The computational cost of the GFS method in comparison with
conventional simulation methods should be reduced. In general,

Stage:

Nomenclature
2D 1D size of periodical domain in Fourier series-based
decomposition
2Q 1 total number of Fourier series terms
AI acoustic impedance, kg/(sm2) or Pas/m, Rayl
Ap lower-triangular matrix of linear model of coregionalization coefficients for pth vector of independent random factors
Bp sill contribution matrix of linear model of coregionalization coefficients for pth vector of independent random factors
cY u data/simulation location covariance matrix
Ch covariance function
C1
Y inverse of data/data covariance matrix
CX1 1D covariance function of lth turning line process that
p;k:l
corresponds to pth independent random factor of kth
coregionalized random function

0.080

0.42
0.36
0.34
0.32
0.21

0.040

0.000
0.000

69853.19
5654.50
2988.25
1417.39
59.44

0.040

0.000
0.100

0.200
0.300
Porosity

0.400

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.060
0.040

0.91
0.37
0.21
0.14
0.02

0.200

Frequency

0.080

10

100
1000 10000 100000
Permeability (mD)

Simulated-Sand-Facies
Acoustic Impedance on Grid 1

Number of data 7,285


Mean 0.28
Standard deviation 0.20

0.100

Frequency

Maximum
Upper quartile
Median
Lower quartile
Minimum

0.080

Simulated-Sand-Facies
Water Saturation on Grid 1

0.150
0.100

Number of data 7,285


Mean 5.33
Standard deviation 0.31
Maximum
Upper quartile
Median
Lower quartile
Minimum

6.44
5.53
5.31
5.12
4.19

0.050

0.020
0.000
0.00

Number of data 7,285


Mean 4328.65
Standard deviation 5158.14

0.120

Frequency

Frequency

Simulated-Sand-FaciesPermeability on Grid 1

Number of data 7,285


Mean 0.34
Standard deviation 0.03
Maximum
Upper quartile
Median
Lower quartile
Minimum

Total Pages: 15

this computational cost depends on the number of coregionalized


variables in the modeled system, number of simulation nodes, and
number and type (point scale or block scale) of the conditioning
data. The GFS algorithm is inherently parallel in that multiple
locations and realizations can be processed independently. Thus,
parallel programming allows for the use of multiple processors to
simulate several realizations at multiple locations simultaneously.
The GFS method can be easily coupled with other geostatistical techniques, such as GFS of categorical variables. By doing
so, all reservoir components, such as structural elements, compartmentalization, geological attributes, and trends, will be simulated at once. The entire reservoir will be simulated and
visualized on the fly. The GFS of categorical variables can be
performed with truncated Gaussian simulation within the GFS
paradigm, where simulated continuous random functions are
truncated at some thresholds to obtain distribution of categories
in space (Goovaerts 1997).

Simulated-Sand-Facies Porosity
on Grid 1

0.120

Page: 12

0.000
0.20

0.40

0.60

0.80

Water Saturation

1.00

3.50 4.50 5.50 6.50 7.50 8.50 9.50 10.50 11.50


6

AI (10 Pas/m)

Fig. 18Histograms of simulated porosity, permeability, water saturation, and AI in original units for sand facies over simulation
Grid 1.
12

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:36 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Stage:

Simulated-Sand-Facies Porosity Permeability on Grid 1

Page: 13

Total Pages: 15

Simulated-Sand-Facies Porosity Water Saturation on Grid 1

100,000

1.00

Number of data 7,285

Number of data 7,285


Correlation 0.54

Water Saturation

Permeability (md)

10,000
1,000
100
10
1
0.000

0.100

0.200

0.300

Correlation 0.02

0.80
0.60
0.40
0.20
0.00
0.000

0.400

0.100

Porosity

0.200

0.300

Simulated-Sand-Facies Porosity Acoustic Impedance on Grid 1

Simulated-Sand-Facies Permeability Water Saturation on Grid 1


100,000

10.50

Number of data 7,285

9.50

Correlation 0.93

8.50
7.50
6.50
5.50

Number of data 7,285


Correlation 0.05

10,000

Permeability (md)

AI (106 Pas/m)

0.400

Porosity

1,000
100
10

4.50
3.50
0.000

0.100

0.200

0.300

1
0.00

0.400

Porosity

0.80

1.00

Number of data 7,285


10.50

Correlation 0.41

1,000
100
10

Correlation 0.19

9.50
6
AI (10 Pas/m)

Permeability (md)

0.60

Simulated-Sand-Facies Water Saturation


Acoustic Impedance on Grid 1

Number of data 7,285

10,000

0.40

Water Saturation

Simulated-Sand-Facies Permeability Acoustic Impedance on Grid 1


100,000

0.20

8.50
7.50
6.50
5.50
4.50

1
3.5 4.5 5.5 6.5 7.5 8.5 9.5 10.5

3.50
0.00

AI (106 Pas/m)

0.20

0.40

0.60

0.80

1.00

Water Saturation

Fig. 19Scatter plots of simulated porosity, permeability, water saturation, and AI in original units for sand facies over simulation
Grid 1.

exp(h) exponential-covariance-function model


GR gamma ray measurement, coulomb/kg, R
GRmax maximum gamma ray measurement along the wellbore, C/kg, R
GRmin minimum gamma ray measurement along the wellbore, C/kg, R
h lag-separation distance
h lag-separation vector
i realization index
j data index
k index of coregionalized random functions
K total number of coregionalized random functions
l index of turning line processes
L total number of turning line processes
Nk total number of data of kth coregionalized random
function
NR total number of realizations
p index of independent random factors

P
q
r
rh

rv
s
sph(h)
T

u
u0
^
u
l
Vi

V(u)
Vshale
x

total number of independent random factors


index of Fourier series terms
variogram or correlation range, L, m
variogram or correlation range in horizontal plane, L,
m
variogram range in vertical direction, L, m
water saturation
spherical covariance function model
vector or matrix transpose operator
location vector in space
modified location vector in space to account for anisotropy in realizations
unit vector of lth turning line process
matrix of dual kriging weights of ith realization of
coregionalized random functions
volume or block scale centered at location u
volume fraction of shale minerals in the rock
coordinate in x-direction of horizontal plane

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:36 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

13

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Horizontal Variogram of Sand Porosity


Normal Scores from Grid 1

1.20

Stage:

1.20
1.00

0.80

0.80

0.60

0.40

0.40

0.20

0.20

0.00

Total Pages: 15

Horizontal Variogram of Sand Permeability


Normal Scores from Grid 1

1.00

0.60

Page: 14

0.00
500

1000

1500

2000

2500

3000

500

Distance (m)

1000

1500

2000

2500

3000

Distance (m)

Horizontal Variogram of Sand Water Saruration


Normal Scores from Grid 1

Horizontal Variogram of Sand Porosity Permeability Normal Scores from Grid 1

1.20
1.00

0.40

0.80

0.00

0.60
0.40

0.40
0.20
0.80

0.00
0

500

1000

1500

2000

2500

3000

500

Distance (m)

1000

1500

2000

2500

3000

Distance (m)

Horizontal Variogram of Sand Porosity Water Saruration Normal Scores from Grid 1

Horizontal Variogram of Sand Permeability Water Saruration Normal Scores from Grid 1

0.20

0.02
0.10
0.01
0.00

0.00
0.01

0.10
0.02
0.03

500

1000

1500

2000

2500

3000

Distance (m)

0.20

500

1000

1500

2000

2500

3000

Distance (m)

Fig. 20Horizontal experimental variograms shown by green lines of simulated geological properties in normal scores for sand facies resolved on Grid 1 and target variogram model presented by a red line; the reproduction of a cross variogram between porosity and water saturation is poor, because their correlation is weak in comparison with sill contributions of other variograms.

X0,i(u) ith realization vector of zero-term random factor or


white noise at location u
Xp,i(u) ith realization vector of pth independent random factors at location u
Xp,k,i(u) ith realization of pth independent random factor of
kth random function at location u
y coordinate in y-direction of horizontal plane
ydk;j jth data of kth coregionalized random function in normal scores
Yi(u) ith realization vector of coregionalized random variables at location u in normal scores
Yk kth coregionalized random function in normal scores
Yk,i(u) ith realization vector of kth coregionalized random
variable at location u in normal scores
z coordinate in vertical z-direction
z0 coordinate in vertical z-direction adjusted to the petroleum reservoir top
14

zdk;j jth data of kth coregionalized random function in original units


ztop(x, y) coordinate in vertical z-direction of petroleum reservoir top on a horizontal plane with x- and ycoordinate
Z random function in original units
Z(u) point-scale random variable at location u in original
units
Z[V(u)] V block-scale random variable centered at location u
in original units
Zi(u) ith realization vector of coregionalized random variables at location u in original units
Zk kth coregionalized random function in original units
DYdi vector of difference between data and simulated values of ith realization at data locations
j rock permeability, L2, md
/ rock porosity
2016 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 02:37 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

REE180917 DOI: 10.2118/180917-PA Date: 2-April-16

Shale

Sand

IHS

Breccia

All Facies

Porosity
0.4
0.3
0.2
0.1
0.0
Z
Y
X

Fig. 21Porosity realization for each facies domain separately


and together at a point scale.

/0 neutron-rock porosity adjusted to shale content in the


rock
/p;k;l;q;i random phase of pth independent random factor, kth
coregionalized random function, lth turning line process, qth Fourier term, and ith realization

Acknowledgments
The financial support provided by members of the Centre for
Computational Geostatistics is highly appreciated. We thank
anonymous authors for their constructive, thoughtful, and valuable comments on improvement of the quality and readability of
the present manuscript.
References
Aziz, K. 1993. Reservoir Simulation Grids: Opportunities and Problems.
J Pet Technol 45 (7): 658663. SPE-25233-PA. http://dx.doi.org/
10.2118/25233-PA.
Butler, R. M. 1991. Thermal Recovery of Oil and Bitumen. Upper Saddle
River, New Jersey: Prentice Hall.
Chiles, J.-P. and Delfiner, P. 2012. Geostatistics: Modeling Spatial Uncertainty, second edition. New York City: John Wiley & Sons.
Deutsch, C. V. and Journel, A. G. 1998. GSLIB: Geostatistical Software
Library and Users Guide, second edition. New York City: Oxford
University Press.
Ellis, D. V. and Singer, J. M. 2008. Well Logging for Earth Scientists, second edition. Dordrecht, The Netherlands: Springer.
geoSCOUT. 2014. geoLOGIC systems geoSCOUT, http://www.geologic.
com/products-services/geoscout (accessed 28 August 2014).

Stage:

Page: 15

Total Pages: 15

Goovaerts, P. 1997. Geostatistics for Natural Resources Evaluation. New


York City: Oxford University Press.
Journel, A. G. and Huijbregts, C. J. 1978. Mining Geostatistics. New York
City: Academic Press.
Kumar, D. 2006. A Tutorial on Gassmann Fluid Substitution: Formulation,
Algorithm and Matlab Code. Geohorizons 1: 412.
Pyrcz, M. J. and Deutsch, C. V. 2014. Geostatistical Reservoir Modeling:
2nd edition. New York City: Oxford University Press.
Schneider, C. L. and Grobe, M. 2013. Regional Cross-Sections of Devonian Stratigraphy in Northeastern Alberta (NTS 74D, E). Open File
Report 2013-05, Alberta Energy Regulator and Alberta Geological
Survey, Canada.
Srivastava, R. M. 1992. Reservoir Characterization with Probability Field
Simulation. Presented at the SPE Annual Technical Conference and
Exhibition, Washington, DC, 47 October. SPE-24753-MS. http://
dx.doi.org/10.2118/24753-MS.
Suncor. 2013. Suncor Firebag 2013 ERCB Performance Presentation,
Commercial Scheme Approval No. 8870, Report Period March 1,
2012 to February 28, 2013. Oral presentation given at the ERCB Performance Presentation, Calgary, 12 May.
Wackernagel, H., 2003. Multivariate Geostatistics, third edition. Berlin:
Springer.
Zagayevskiy, Y. 2012. Constraining 3D Petroleum Reservoir Models to
Petrophysical Data, Local Temperature Observations, and Gridded
Seismic Attributes with the Ensemble Kalman Filter (EnKF). Masters
thesis, University of Alberta, Edmonton, Canada.
Zagayevskiy, Y. and Deutsch, C. V. 2015. Multivariate Grid-Free Geostatistical Simulation with Secondary Data. Stoch. Env. Risk A. 121.
http://dx.doi.org/10.1007/s00477-015-1154-x.
Zagayevskiy, Y. and Deutsch, C. V. In press. Multivariate Geostatistical
Grid-Free Simulation of Natural Phenomena. Math. Geosci.
Yevgeniy Zagayevskiy is a research assistant at the Centre for
Computational Geostatistics, University of Alberta, Edmonton,
Canada, directed by Clayton V. Deutsch. His current main
research work focuses on the spectral and fractal geostatistical simulation of geological phenomena. Zagayevskiy has
also performed research on the development of a framework
of sensitivity analysis on the basis of the regression approach,
permeability modeling by use of log microimages, and construction of the impact map for assessment of new delineation-well locations. He worked as a summer geomodeling
student for Husky Energy, Calgary, in 2011, and for Statoil, Calgary, in 2012. Zagayevskiy has been an SPE student member
since 2010. He holds masters and PhD degrees in mining engineering (geostatistics) from the University of Alberta, and a
bachelors degree in petroleum engineering from Kazakh-British Technical University, Kazakhstan. Zagayevskiys masters
degree thesis concerned the application of the ensemble Kalman filter in petroleum reservoir characterization and his PhD
degree dissertation concerned multivariate grid-free geostatistical simulation.
Clayton V. Deutsch is a professor in the Department of Civil
and Environmental Engineering at the School of Mining and
Petroleum Engineering at the University of Alberta. He teaches
and conducts research into better ways to model heterogeneity and uncertainty in petroleum reservoirs and mineral
deposits. Deutsch holds the Alberta Chamber of Resources
Industry Chair in Mining Engineering and the Canada
Research Chair in Natural Resources Uncertainty Characterization. He has been an SPE member for almost 25 years.
Deutsch holds masters and PhD degrees in applied earth sciences (geostatistics) from Stanford University and a bachelors
degree in mining engineering (with distinction) from the University of Alberta.

2016 SPE Reservoir Evaluation & Engineering


ID: jaganm Time: 02:37 I Path: S:/REE#/Vol00000/160015/Comp/APPFile/SA-REE#160015

15

Vous aimerez peut-être aussi