Rock physics based facies classification from seismic inversion results in unconventional
reservoirs Zakir Hossain* and Stefano Volterrani, ION
Summary
The objective of this study is to demonstrate the power of
integrating rock physics theory, measurement and
simulation to improve facies prediction in an
unconventional limestone and shale reservoir. Reliable
facies prediction is a challenge in unconventional reservoir
characterization because of complex geological
heterogeneities. Both deterministic and probabilistic
approaches are commonly used in facies classifications that
use well and seismic data. Bayes’ theory with
uninformative priors is often used for probabilistic facies
classification. We provide a case study that uses Bayes’
theory with informative priors for facies classification from
pre-stack simultaneous elastic inversion results in an
unconventional reservoir. In the proposed methodology, we
integrate rock physics based theory, measurements and
simulation with Bayesian statistical techniques where the
prior probability represents our knowledge about rock
properties, and is consistent with our geological
knowledge, rock physics theory and measured data. We
evaluate four facies classification methodologies:
deterministic method, probabilistic method with
uninformative priors, probabilistic method with
uninformative priors and training facies defined from
simulation, and probabilistic method with informative
priors and training facies defined from simulation. This
study indicates that, in probabilistic facies classification
(Method 2), if uninformative priors are used, results are
sub-optimal compared to deterministic methods involving a
Rock Physics Template (RPT) workflow (Method 1).
Additionally, probabilistic facies classification can be
further improved if we use uninformative priors and
training facies defined from Monte Carlo simulation
(Method 3). Probabilistic facies prediction improves if we
use informative priors and training facies defined from
Monte Carlo simulation (Method 4).
Introduction
Reliable facies prediction is an essential problem in
reservoir characterization. Predicted facies properties are
important engineering inputs for drilling and production.
For reservoir facies characterization, two different methods
are commonly used: deterministic approach (Doyen, 1988;
Loertzer and Berkhout, 1992; Avseth et al., 2005; Hossain
et al. 2015) and probabilistic approach (Gastaldi, et al.
1998; Gouveia, 1996; Takashashi, 2000; Mukerji et al.,
2001; Hossain and Mukerji, 2011; Bachrach et al., 2004;
Bachrach, and Dutta, 2004; Grana et al., 2012; Hossain et
al. 2015). For deterministic facies classification we use an
RPT workflow, while for probabilistic facies prediction, we
can use Bayes’ theory:
n
i
ii
iii
cpcxp
cpcxpxcp
1
|
|| (1)
where, p(ci) is the prior probability,
p(ci|x) is the posterior probability of our observation,
p(x|ci) is the likelihood of obtaining our particular
observation ci, under the supposition that any of the
possible states of the variable x was actually the case.
For seismic based facies prediction, the above expression
can be written as:
priorfaciestrainingseismicp
seismicfaciespredictedp
)_|(
)|_( (2)
From this expression, we observe that the training facies
influence the predicted facies, but the prior probabilities are
more heavily influenced by the predicted facies. Hossain et
al. (2015) demonstrated the role of prior belief of Bayesian
statistics by using three types of priors: uninformative
priors, informative priors, and continuous priors and found
that for uninformative prior, the posterior remains
unchanged, while for informative priors, the posterior is
increased.
For uninformative priors, equation (1) becomes:
Figure 1: The relationship between rock physics theory, rock
physics measurement and rock physics based simulation for facies
prediction.
Page 2911© 2016 SEG SEG International Exposition and 86th Annual Meeting
Dow
nloa
ded
11/1
7/16
to 2
04.2
7.21
3.16
2. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
Rock physics based facies classification
ii cxpxcp || or
liklihoodposterior (3)
The objective of this study is to demonstrate the power of
integrating rock physics theory, measurement and
simulation to improve facies prediction for reservoir
characterization.
Theory and/or Method
The real nature is too difficult to understand because of the
large number of properties. The scientist’s goal is to create
an understanding of physical properties and processes of
nature that are as complete as possible, making use of the
perfect control of experimental conditions in the simulation
experiment and of the possibility to examine every aspect
of system configurations in detail (Landau and Binder,
2000). Only theory or only experiment or only simulation is
not good enough to create an understanding of physical
properties and the processes of nature. Landau and Binder
(2000) presented the relationship between theory,
experiment, and simulation as being similar to those of the
vertices of a triangle, as shown in Figure 1: each is distinct,
but each is strongly connected to the other two. In this
paper, our objective is to define the facies from seismic
data by integrating rock physics theory, rock physics
measurement and rock physics based simulation for
Figure 2: ( a) Petrophysical analysis of studied well George 1-23 from Pink limestone to Wilcox, (b) Petrophysical analysis for zone of interest
(ZOI) from Mississippi Lime to Woodford shale, (c) Defiined facies from rock physics template (RPT): silica-rich limestone (cyan), clay-rich
limestone (black), lower kerogen-rich shale (magenta) and higher kerogen-rich shale (red), (d) ) 2D PDFs for each facies using attributes P-
impedance and Vp/Vs,
Page 2912© 2016 SEG SEG International Exposition and 86th Annual Meeting
Dow
nloa
ded
11/1
7/16
to 2
04.2
7.21
3.16
2. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
Rock physics based facies classification
reservoir characterization of an unconventional limestone
and shale reservoir in the Mississippi Lime Play, north-
central Oklahoma.
We used well log measurement results of well George 1-23
from the studied region. We performed petrophysical
analysis calibrated with available laboratory measurement
results (Figure 2b). The zone of interest (ZOI) in this study
is from the top of the Mississippi Lime to the bottom of the
Woodford shale (Figure 2b). We used elastic attribute
volumes (P-Impedance, S-Impedance) obtained from the
pre-stack seismic inversion. We defined four facies based
on petrophysics and rock physics analysis. Defined facies
are: silica-rich limestone, clay-rich limestone, lower
kerogen-rich shale and higher kerogen-rich shale (Figure
2b). We used an RPT (Figure 2a after Hossain et al. 2015)
for deterministic facies prediction. Furthermore, for seismic
reservoir characterization, well data along with RPT are
used to define the prior probability. For seismic
applications, one of the central issues for stochastic
simulation is to use a statistical model rather than a rock
physics model:
yuncertaintmodel) lstatistica( Model
simulation Carlo Monte
We replaced the statistical model by a rock physics model,
addressed the uncertainly defined from rock physics
analysis, and included rock physics based upper bound and
lower bound to constrain the simulation results:
boundupper
boundlower yuncertaint
model) physicsRock ( Model
simulation Carlo Monte
Results and discussions
Figure 4a shows the deterministic facies predictions
involving an RPT workflow. Overall facies predictions are
moderate, but there are many under predicted and over
predicted intervals. To improve these predictions we
performed probabilistic facies predictions. It is commonly
assumed that probabilistic predictions are always better
than deterministic predictions. Unfortunately, this is not
always true if uninformative priors are used and training
facies are defined from well log data (Figure 4b). Defined
training facies from well log data are not good enough to
capture the seismic data away from the well (Figure 2c).
Therefore, under-predicted facies, away from well,are
mainly due to the training facies defined from the well log.
To make a better match for the entire seismic data,
including areas away from the well, we used defined
training facies from Monte Carlo simulation (Figure 3).
Hence, probabilistic facies classification can be further
improved if we use uninformative priors and training facies
defined from Monte Carlo simulation (Figure 4c).
However, better probabilistic facies prediction can be
obtained if we use informative priors and training facies
defined from Monte Carlo Simulation (Figure 4d).
Conclusions
For seismic reservoir characterization, we provided a case
study for facies predictions from pre-stack simultaneous
elastic inversion results in an unconventional reservoir.
This study indicates that, in probabilistic facies
classification, if uninformative priors are used, results are
sub-optimal compared to deterministic methods involving a
Rock Physics Template (RPT) workflow. Additionally,
probabilistic facies classification can be further improved if
we use uninformative priors and training facies defined
from Monte Carlo simulation. Probabilistic facies
prediction improves if we use informative priors and
training facies defined from Monte Carlo simulation.
Acknowledgments
The authors thank EnerVest for allowing this work to be
published. There are a number of individuals at ION that
contributed to this work, including Howard Rael, who did
the petrophysical analysis and Shihong Chi did inversion.
Tanya L. Inks from IS Interpretation Services, Inc and Paul
Brettwood from ION acknowledged for edits.
Figure 3: Simulated results for silica-rich limestone (cyan), clay-
rich limestone (black), lower kerogen-rich shale (magenta), higer
kerogen-rich shale (magenta) and seismic inversion results (red).
Page 2913© 2016 SEG SEG International Exposition and 86th Annual Meeting
Dow
nloa
ded
11/1
7/16
to 2
04.2
7.21
3.16
2. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
Rock physics based facies classification
Figure 4: (a) Deterministic facies predictions involving an RPT workflow. Probabilistic facies prediction using Bayes theory: (b) if
uninformative priors are used, (c) if uninformative priors and training facies defined from Monte Carlo simulation are used, and (d) if
informative priors and training facies defined from Monte Carlo simulation are used.
Page 2914© 2016 SEG SEG International Exposition and 86th Annual Meeting
Dow
nloa
ded
11/1
7/16
to 2
04.2
7.21
3.16
2. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
EDITED REFERENCES Note: This reference list is a copyedited version of the reference list submitted by the author. Reference lists for the 2016
SEG Technical Program Expanded Abstracts have been copyedited so that references provided with the online metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web.
REFERENCES Avseth, P., T. Mukerji, and G. Mavko, 2005, Quantitative seismic interpretation: Applying rock physics
tools to reduce interpretation risk: Cambridge University, http://dx.doi.org/10.1017/CBO9780511600074.
Bachrach, R., and N. Dutta, 2004, Joint estimation of porosity and saturation using stochastic rock physics modeling: 66th Annual International Conference and Exhibition, EAGE, Extended Abstracts, http://dx.doi.org/10.2118/89991-MS.
Bachrach, R., M. Beller, C. Liu, J. Perdomo, D. Shelander, and N. Dutta, 2004, Combining rock physics analysis, full wave form prestack inversion and high resolution seismic interpretation to map lithology units in deep water: A Gulf of Mexico case study: The Leading Edge, 23, 378–383, http://dx.doi.org/10.1190/1.1729224.
Doyen, P. M., 1988, Porosity from seismic data: A geostatistical approach: Geophysics, 53, 1263–1275, http://dx.doi.org/10.1190/1.1442404.
Gastaldi, C., D. Roy, P. Doyen, and L. Den Boer, 1998, Using Bayesian simulations to predict reservoir thickness under tuning conditions: The Leading Edge, 17, 539–543, http://dx.doi.org/10.1190/1.1438008.
Gouveia, W., 1996, Bayesian seismic waveform data inversion: Parameter estimation and uncertainty analysis: Ph.D. thesis, Colorado School of Mines.
Grana, D., M. Pirrone, and T. Mukerji, 2012, Quantitative log interpretation and uncertainty propagation of petrophysical properties and facies classification from rock physics modeling and formation evaluation analysis: Geophysics, 77, no. 3, WA45–WA63, http://dx.doi.org/10.1190/geo2011-0272.1.
Hossain, Z., and T. Mukerji, 2011, Statistical rock physics and Monte Carlo Simulation of seismic attributes for greensand: 73rd Annual International Conference and Exhibition, EAGE, Extended Abstracts, http://dx.doi.org/10.3997/2214-4609.20149674.
Hossain, Z., S. Volterrani, and F. Diaz, 2015, Integration of rock physics template to improve Bayes’ facies classification: 85th Annual International Meeting, SEG, Expanded Abstracts, 2760–2764, http://dx.doi.org/10.1190/segam2015-5900545.1.
Landau, D. P., and K. Binder, 2000, A Guide to Monte Carlo Simulations in Statistical Physics: Cambridge University Press.
Lörtzer, G. J. M., and A. J. Berkhout, 1992, An integrated approach to lithologic inversion: Part 1, Theory: Geophysics, 57, 233–244, http://dx.doi.org/10.1190/1.1443236.
Takahashi, I., 2000, Quantifying information and uncertainty of rock property estimation from seismic data: Ph.D. thesis, Stanford University.
Page 2915© 2016 SEG SEG International Exposition and 86th Annual Meeting
Dow
nloa
ded
11/1
7/16
to 2
04.2
7.21
3.16
2. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/