You are on page 1of 10

Improved lithofacies characterization using the SNESIM algorithm of

multiple-point geostatistics
Nagendra Babu M1, Rajesh R Nair1
Computational Petroleum Geomechanics Laboratory, Department of Ocean Engineering, Indian
Institute of Technology, Madras, India.
Corresponding author: Nagendra Babu M (oe16d022@smail.iitm.ac.in)
Abstract
We present a novel, integrated approach to improving reservoir lithofacies characterization. Our
methodology combines three significant techniques. The first is that simultaneous prestack inversion
for inverting the prestack gathers into seismic attributes. Then a cross-plot of P-impedance and V P/VS
ratio from well log data used to classify the different reservoir lithofacies for well log data. The
second is that these lithofacies are incorporated by Bayesian rule for converting the seismic attributes
into lithofacies volume and the probability volumes of each type of lithofacies. The third technique is
the multiple-point geostatistical simulation (MPS) using the SNESIM algorithm applied with the help
of training image and probability volumes for a better lithofacies model. The feasibility of this
approach is verified using the real seismic data and well data. This results from this integrated study
proved that MPS could improve in the characterization of reservoir lithofacies hence reduction in
uncertainty.
Highlights
• Prestack seismic data converted into elastic properties P-Impedance (ZP), S-
Impedance (ZS), Density (Dn), and VP/VS ratio.
• The Bayesian classification used to convert the seismic elastic properties into
probability volumes of lithofacies.
• The SNESIM algorithm used to better reservoir characterization in multi-point
geostatistics constrained by well data and seismic lithofacies
Keywords
Prestack Inversion, Angle-dependent wavelets, Non-parametric statistical classification, Bayesian
classification, Multiple-point geostatistics
1.Introduction
A goal of the reservoir characterization is to obtain a reliable reservoir geological model that has the
highest consistency to reality concerning geological shapes and their facies through the integration of
the data from various sources (Yu et al., 2011). Many sources of information such as geophysical,
geological, and petrophysical disciplines, etc. are required to study of the reservoir. The lithofacies
characterization of a reservoir plays an essential role in the identification of prospects and
development studies. Traditional methods such as seismic inversion and rock physics analysis are
understated geological shapes in many reservoirs hence uncertainty in discretizing of lithology. The
seismic inversion procedure has gained the name as an essential technique that used to obtained
elastic properties of the subsurface and petrophysical properties of the reservoir lithologies. It
provides the spatial continuity of the reservoir elastic properties between the wells. The rock physical
study provides the knowledge to link the elastic properties from seismic inversion to petrophysical
analysis of well logging data such as lithofacies classification (cluster analysis) (Avseth et al., 2005;
Doyen, 2007; Mavko et al., 2009).
The main issues in the seismic interpretation are their limited vertical resolution with the relation of
geological and petrophysical properties. The widely recognized issue in many reservoirs is that
uncertainty generally associated with the identification of lithology and fluid properties from seismic
data. The first source of uncertainty arises from non-uniqueness solutions of seismic inversion in the
relationship between the measured seismic amplitudes and elastic properties, and many alternative
models can produce the same seismic response. The second uncertainty source is the link between the
elastic properties and petrophysical properties. The uncertainty analysis in rock physics modeling has
an essential part in quantitative seismic interpretation (Bachrach, 2006; Kjønsberg et al., 2010;
Sengupta & Bachrach, 2007).
Traditionally, mainstream statistical analysis methods in geomodelling are pixel-based two-point
simulation and object-based statistical simulation. In two-point geostatistics, facies structures
characterized by using the variograms, but it cannot capture the curvilinear structures and shapes of
geological depositions such as meanders and the quantity of well data insufficient to obtained reliable
variogram for 3D simulation (Journel, 1993; Sebastien B. Strebelle1, 2002; Strebelle, 2002). Object-
based techniques can reproduce of lithofacies geometry, but they need high CPU demand and
limitations in the integration of secondary data (Haldorsen & Damsleth, 1990).
Many geological problems in exploitation and production have become more sophisticated, new
technology and methods are widely using for reservoir characterization. In this paper, we have
addressed a workflow to overcome the limitations in the characterizing of lithology heterogeneity and
reducing the uncertainty in the geological model for complicated reservoir settings. This methodology
integrates the prestack inversion technique, probabilistic modeling approach with Single Normal
Equation Simulation (SNESIM) algorithm based multiple-point geostatistics method (MPS) to
overcome limitations and provides an improvement in the lithofacies model (Guardiano & Srivastava,
1993; Strebelle, 2002).
2.Methods
In this methodology, the aim is that it integrates three main techniques for improving lithofacies
characterization. The three techniques are (1) simultaneous prestack inversion for generating the
seismic elastic properties, (2) probabilistic rock facies estimated by establishing the link between well
log and seismic elastic properties through Bayesian classification, and (3) multiple-point geostatistics
based the SNESIM algorithm for improved lithofacies characterization.
2.1. Simultaneous prestack inversion
In this study, simultaneous prestack inversion (SPSI) has used for inverting the prestack seismic
migrated angle gathers to determine P-impedance (Z P), S-impedance (ZS), density (Dn), and VP/VS
ratio (Cataldo & Leite, 2018; D. P. Hampson et al., 2005; Simm & Bacon, 2014). The entire seismic
inversion approach involves wavelet estimation, well to seismic tie, generate initial guess model, and
seismic inversion based on the reformulated Aki-Richards equation (Fatti et al., 1994; Maver &
Rasmussen, 1995; Mrinal K. Sen, 2006).
Two wavelet extraction methods do this procedure. Statistical angle-dependent wavelets are extracted
by the autocorrelation method for angle stacks (5°-15°), (15°-25°), and (25°-35°) of seismic prestack
angle gathers at well locations (Yi et al., 2013) The reflectivity from the well log has convolved with
these statistical wavelets for generation synthetic seismogram at well locations for correlating with
real data (de Macedo et al., 2017). Using the synthetic seismogram, seismic to well tie operation has
conducted at each well location to relates well stratigraphic markers to stratigraphic markers of
seismic data for correcting to time to depth relation (Wang et al., 2018). After achieving a better
correlation between real seismic data and synthetic seismic data, the angle-dependent wavelets using
the well logs extracted by finding a time-domain operator that shapes the well log reflectivity that
convolved with the real seismic data at every well locations (D. Hampson & Galbraith, 1981).). Then
initial guess low-frequency models of P-impedance, S-impedance, and density were generated for the
seismic data from well logs and interpreted horizons. This low-frequency information is incorporated
into the seismic inversion algorithm reformulated by (Fatti et al., 1994). The prestack inversion results
are the sections of ZP, ZS, Dn, VP/VS ratio.
2.2. Bayesian classification
The Bayesian approach has used to generate the spatial distributions (Probability volumes) from
seismic derived elastic properties through the link with well-derived lithofacies. Lithofacies types
from well are integrated with seismic elastic properties to generated lithofacies volume and
probabilistic facies volumes by the Bayesian classification method. The well-based lithofacies (true
lithofacies) derived via cluster analysis (cross plot) of elastic properties (Z P and VP/VS ratio) from well
data independently. The non-parametric approach used to generate probability density functions
(PDFs) by smoothing all data points in the cross-plot with an operator that is transforming discrete
points into a continuous density function. These PDFs were used as a priori information in a Bayesian
classification to estimate the posterior lithofacies probability from elastic properties of simultaneous
seismic inversion (Mavko et al., 2009; Reverón & Roomer, 2014). After proper QC measures of non-
parametric PDF at well locations, the Bayesian classification can then be applied to the full seismic
volume to generate probability volumes for each lithofacies, and a lithofacies volume. Bayesian
theory is a statistical procedure. Bayesian decision theory is provided the essential methodology to the
classification problem (Richard O. Duda, Peter E. Hart, 1998). In this context, the use of the Bayes
method for lithofacies classification problems is closely related data corresponding to different
domains. By using Bayes' theorem, the prior knowledge of each lithofacies PDFs incorporates seismic
attributes to produce the most probability lithofacies volumes (Nieto et al., 2013). Bayes' theorem
used to include prior information in probability estimates. This theorem assumes that the probability
of an event is a function of new evidence (the likelihood) and prior probability (Teixeira et al., 2007).
The results of Bayesian classification used as constraints in multi-point geostatistical simulation.
2.3. Single Normal Equation Simulation (SNESIM) of multiple-point geostatistics
The concept of multiple-point geostatistics is using the training image (Ti) proposed by (Guardiano &
Srivastava, 1993). It has introduced to overcome limitations of existing statistical methods such as
pixel-based algorithms and object-based algorithms. The SNESIM algorithm in MPS used to reduce
the massive demand for memory requirements through avoiding scanning of the training image for
every node simulation (Strebelle, 2000). The MPS using the SNESIM algorithm can have benefits of
both pixel and object algorithm for dealing the complicated geological settings without an explicit
non-gaussian model (Journel, 1993).
In the MPS, a training image represents a numerically defined database of geological patterns, and
their associated facies, which provide multiple statistics/conditioning probability values for SNESIM
simulation at the central node of the search/simulation template (T D/devs(u)) to a lithofacies type
given any multiple-point conditioning data event. A Ti has generated with the help of previous
information about facies types and their proportion from the cross-plot analysis of well logging data
and accordance with the deposition environment. By using the concept of Training image (Ti), the
SNESIM algorithm can reproduce the lithofacies channels in the simulation as retaining the
advantages of pixel and object algorithms. In the SNESIM algorithm, the whole Ti scanned once, and
all possible statistics stored as a search tree. This search tree will reduce the high CPU demanding
during the simulation. The detailed MPS procedure explained by (Sebastien B. Strebelle1, 2002;
Strebelle, 2002).
The main steps of the SNESIM algorithm process follow as
1. Create the Ti with the help of previous geological depositional knowledge with the exact
proportion gained from the cross-plot analysis at least 1.5 times of the simulation grid.
2. Scan the Ti with the search template (3 x 3), identify all possible physical patterns/statistics
values a central node, and construct a searching tree as a database.
3. Create a simulation grid.
4. Hard data (Well litho-logs) placed in the simulation grid act as defined facies (Hard data
conditioning).
5. Define the random path visiting all locations to simulated.
6. Find the conditional data event devj(u).
7. Retrieve the conditional probability distribution from training image P(Z(u)=k|dev s(u)).
8. The soft data (probability volumes of each facies) are integrated with conditional probability
distribution in simulation.
9. A simulated value has drawn from that conditional probability (Realization).
3.Application & Results
This methodology applied to field data from the Upper Assam basin, India. The field data contains 3D
prestack migrated gathers in the offset domain, a total of five well logging data, and pre interpreted
horizons are available. The HRS-9 and SGeMS software are used for this study.
3.1. Estimation of elastic properties from simultaneous prestack inversion
Elastic properties are essential in identifying litho-bodies of a reservoir. In this study, simultaneous
prestack inversion has used invert the seismic migration gathers into elastic properties. The seismic
data has preconditioned to improve the signal to noise ratio and converted into an angle domain (angle
gathers) for the convenience of the prestack inversion algorithm (Huang et al., 2015; Zhang et al.,
2015). Two types of wavelet extraction procedures (Statistical/Deterministic) have applied while
doing the seismic to well tie for wells. Once we get the best time to depth relation with a good
correlation coefficient at two well locations, an initial model of elastic properties with low-pass
filtered well-logging data is created. The simultaneous prestack inversion uses these initial models for
perturbing iteratively with a selected deterministic wavelet to obtain the elastic properties (ZP, ZS, Dn
& VP/VS ratio) by achieving the good match between synthetic and real data. The prestack inversion
has conducted for angle range 5°-35°of seismic data. Figure 1(a) & 1(b) shows an arbitrary line of ZP
& VP/VS ratio from prestack inversion analysis via all five wells. The results of simultaneous
prestack inversion have cross verified with well-logging data, including the wells which have not
involved in the inversion process. The insightful matching between seismic elastic properties and well
elastic properties indicates the quality of inversion results.
3.2. Integration of seismic inversion results with well data to generate the probability volumes
using Bayesian classification
Probability volumes for lithofacies are generating from seismic elastic properties obtained in the
previous step. The Bayesian approach used integrates the seismic elastic properties with lithofacies
such as shale, brine sand, and prospective sand (HC sand) are identified by cross plot analysis of Z P &
VP/VS ratio from log data of selected wells (Calabrese et al., 2011; Coulon et al., 2006; Doyen, 2007).
The main facies are classified as shale (green), water-bearing sand zone (yellow), and prospective-
bearing sand represents HC zone (red) with different ranges of Z P & VP/VS ratio. Figure 2(a) shows
the cross-plot analysis for identifying the three lithofacies. After cross plot analysis, a multivariate
Probability distribution functions (PDF) has been prepared based on non-parametric statistical
classification using kernel analysis in the cross-plot space (Węglarczyk, 2018). For the kernel
analysis, facies proportions for hydrocarbon-bearing sand, water-bearing sand, and shale estimated
from logging data. These proportions of each facies can apply while generating the Ti in the MPS
procedure. Figure 2(b) shows the PDFs are generated from non-parametric kernel analysis. By using
Bayes' theorem, the prior knowledge of each lithofacies incorporates with PDFs to produce the
probability volumes for each lithofacies type and lithofacies volume (Nieto et al., 2013). Figure 2(c)
shows the QC measure applied to the Bayesian results using the confusion matrix at well locations.
The confusion matrix measures the mismatch between actual lithofacies (litho-log generates by the
cross plot) and estimated Bayesian lithofacies from seismic attributes. The diagonal values of the
confusion matrix should be large values for the quality of prediction lithofacies and visual inspection
on comparison of the actual litho log with the predicted litho log from Bayesian classification. Figure
2(d) shows the visual inspection of predicted lithofacies log from seismic data and true lithofacies
from well data. Figure 3(a) the arbitrary line of final lithofacies volume along the well locations.
Figure 3(b),3(c) & 3(d) shows the 3D horizontal slice of the Z P, VP/VS ratio and lithofacies volume.
3.3. Improved lithofacies from the SNESIM algorithm
The MPS procedure has started with the generation of training images (Ti). The training image has
generated with three facies representing shale, water-bearing sand, hydrocarbon-bearing sand with the
help of depositional knowledge in the study basin, and proportions of lithofacies from the cross plot
analysis (Fadlelmula F. et al., 2016). As per geological knowledge of the study area, Ti generated with
parametric shapes given to each facies using the SGeMS software (Remy et al., 2009). The un-
conditional realizations of object-based algorithms used to generate Ti with sinusoidal channels
(sand), elliptical shape geometries (HC), and the remaining area used as background (shale). Figure 4
shows the training image for providing multiple statistics in MPS simulation.
Two main steps involved in the SNESIM algorithm: one is constriction of a searching tree, and the
second is the simulation part. The prepared whole training image can scan once with a pre-defined
search template (TD) to identify the different lithofacies patterns with a value at the center node: all
these patterns and their possibility categories stored in the form search tree before simulation. In the
simulation part, the hard data lithofacies values assigned in the simulation grid according to the well's
position. The SNESIM algorithm works as a sequential simulation at each node/cell of the simulation
grid along a random path through the simulation grid. At each visit of a cell, search for closest nearby
well data and earlier simulated cells according to its specific conditioning data event (template equal
to TD). The probability values/ patterns are retrieved from the search tree, and value is simulated and
assigned a new conditional probability value from it with constraints of soft data (probable volumes).
Figure 5 shows a comparison of 3D horizontal slices of lithofacies volumes and the MPS realization.
Figure 5(a) & 5(b) shows a comparison of the 3D horizontal slices of lithofacies from Bayesian
classification results and MPS results. The improvement in HC lithofacies and brine sand can be
observed (arrow symbols) in these figures. Figure 5(c) & 5(d) shows the improvement in lithofacies
observed by comparison the lithofacies from Bayesian classification with MPS simulation. Figure
5(e) & 5(f) shows the comparison of confusion matrices of predicted lithofacies of both
methodologies (Bayesian & MPS) with true lithofacies (well litho-log generated by cross-plot
analysis) at two wells. It was clearly showing improving the diagonal elements in the confusion
matrix of MPS simulation than the Bayesian simulation. The quantitatively represents the correlation
of predicted lithofacies with true lithofacies in MPS simulation quite better than Bayesian correlation.
The quality of the prediction in the confusion matrix of this methodology is higher than the confusion
matrix of Bayesian classification. The percentages of the diagonal elements of the confusion matrix
have increased due to a reduction in the mismatch between true lithofacies from well log data and
lithofacies resulted from the MPS method. From these confusion matrices, we find that the present
methodology warranting the reliable lithofacies prediction.
The results of the MPS simulation substantiated that the SNESIM algorithm successfully reproduced
the missing lithofacies, which could not be characterized in the conventional Bayesian classification.
With real data, the present methodology promising to improve lithofacies characterization. Therefore,
the integrated workflow used in this study proved that it could provide a better controlled and reliable
lithofacies characterization.
4. Conclusions
In this study, we have worked on an integrated perspective that integrates seismic inversion, Bayesian
classification, and the SNESIM algorithm to be improved lithofacies characterization in a reservoir.
The simultaneous prestack seismic inversion used to obtain the elastic properties from seismic data.
The Baye's rule has been applied to seismic derived elastic properties to obtain probabilities of each
type lithofacies through creating a link between the seismic elastic volumes and cross plot of well
logs. The training images (Ti) hs generated for the facies identified in cross plot analysis with
geological knowledge of the depositional environment. The SNESIM algorithm based MPS used for
reproducing the geological facies with the assist of multi-statistics from Ti and constrained by soft
data (lithofacies probabilities) and hard data to control the development of facies. The results have
compared with the Bayesian method results with MPS results. Hence the SNESIM algorithm has
proven that as an effective technique for improved reservoir characterization.
Acknowledgment
Special thanks to SGeMS developers for providing this open-source software to this integrated study.
Data Availability Statement
For data that are restricted for by industry policies and regulations and are not accessible to the public
research community. Industry has given permission to our institute for piloting this methodology and
publication purpose only. We don’t have data sharing rights.
References
Avseth, P., Mukerji, T., & Mavko, G. (2005). Quantitative Seismic Interpretation. Cambridge
University Press. https://doi.org/10.1017/CBO9780511600074
Bachrach, R. (2006). Joint estimation of porosity and saturation using stochastic rock-physics
modeling. GEOPHYSICS, 71(5), O53–O63. https://doi.org/10.1190/1.2235991
Calabrese, M., Pirera, F., Massimo, R., Scaglioni, P., & Tosoratti, F. (2011). Integration Of Seismic
Lithology Data In The Reservoir Model. In SPE EUROPEC/EAGE Annual Conference and
Exhibition. Society of Petroleum Engineers. https://doi.org/10.2118/143126-MS
Cataldo, R. A., & Leite, E. P. (2018). Simultaneous prestack seismic inversion in a carbonate
reservoir. REM - International Engineering Journal, 71(1), 45–51. https://doi.org/10.1590/0370-
44672016710024
Coulon, J., Lafet, Y., Deschizeaux, B., Doyen, P. M., & Duboz, P. (2006). Stratigraphic elastic
inversion for seismic lithology discrimination in a turbiditic reservoir. In SEG Technical
Program Expanded Abstracts 2006 (pp. 2092–2096). Society of Exploration Geophysicists.
https://doi.org/10.1190/1.2369949
Doyen, P. (2007). Seismic Reservoir Characterization: An Earth Modelling Perspective (EET 2).
EAGE Publications bv. https://doi.org/10.3997/9789073781771
Fadlelmula F., M. M., Killough, J., & Fraim, M. (2016). TiConverter: A training image converting
tool for multiple-point geostatistics. Computers & Geosciences, 96, 47–55.
https://doi.org/10.1016/j.cageo.2016.07.002
Fatti, J. L., Smith, G. C., Vail, P. J., Strauss, P. J., & Levitt, P. R. (1994). Detection of gas in
sandstone reservoirs using AVO analysis: A 3-D seismic case history using the Geostack
technique. GEOPHYSICS, 59(9), 1362–1376. https://doi.org/10.1190/1.1443695
Guardiano, F. B., & Srivastava, R. M. (1993). Multivariate Geostatistics: Beyond Bivariate Moments
(pp. 133–144). https://doi.org/10.1007/978-94-011-1739-5_12
Haldorsen, H. H., & Damsleth, E. (1990). Stochastic Modeling. Journal of Petroleum Technology,
42(04), 404–412. https://doi.org/10.2118/20321-PA
Hampson, D., & Galbraith, M. (1981). Wavelet extraction by sonic log correlation. J. Can. Soc. Expl.
Geophys., 17, 24–42.
Hampson, D. P., Russell, B. H., & Bankhead, B. (2005). Simultaneous inversion of prestack seismic
data. In SEG Technical Program Expanded Abstracts 2005 (pp. 1633–1637). Society of
Exploration Geophysicists. https://doi.org/10.1190/1.2148008
Huang, H.-D., Wang, Y.-C., Guo, F., Zhang, S., Ji, Y.-Z., & Liu, C.-H. (2015). Zoeppritz equation-
based prestack inversion and its application in fluid identification. Applied Geophysics, 12(2),
199–211. https://doi.org/10.1007/s11770-015-0483-3
Journel, A. G. (1993). Geostatistics: Roadblocks and Challenges (pp. 213–224).
https://doi.org/10.1007/978-94-011-1739-5_18
Kjønsberg, H., Hauge, R., Kolbjørnsen, O., & Buland, A. (2010). Bayesian Monte Carlo method for
seismic predrill prospect assessment. GEOPHYSICS, 75(2), O9–O19.
https://doi.org/10.1190/1.3339678
de Macedo, I. A. S., da Silva, C. B., de Figueiredo, J. J. S., & Omoboya, B. (2017). Comparison
between deterministic and statistical wavelet estimation methods through predictive
deconvolution: Seismic to well tie example from the North Sea. Journal of Applied Geophysics,
136, 298–314. https://doi.org/10.1016/j.jappgeo.2016.11.003
Maver, K., & Rasmussen, K. (1995). Seismic Inversion for Reservoir Delineation and Description. In
Proceedings of Middle East Oil Show. Society of Petroleum Engineers.
https://doi.org/10.2523/29798-MS
Mavko, G., Mukerji, T., & Dvorkin, J. (2009). The Rock Physics Handbook. Cambridge: Cambridge
University Press. https://doi.org/10.1017/CBO9780511626753
Mrinal K. Sen. (2006). Seismic Inversion. Society of Petroleum Engineers.
Nieto, J., Batlai, B., & Delbecq, F. (2013). Seismic lithology prediction: a Montney shale gas case
study. CSEG Recorder, 38(February), 34–41.
Remy, N., Boucher, A., & Wu, J. (2009). Applied Geostatistics with SGeMS. Cambridge: Cambridge
University Press. https://doi.org/10.1017/CBO9781139150019
Reverón, J., & Roomer, J. (2014). Probabilistic facies discrimination from simultaneous seismic-
inversion results in clastic reservoirs in southwestern Venezuela. The Leading Edge, 33(7), 784–
790. https://doi.org/10.1190/tle33070784.1
Richard O. Duda, Peter E. Hart, D. G. S. (1998). Pattern Classification (p. 654). The MIT Press.
Sebastien B. Strebelle1, A. G. J. (2002). Reservoir Modeling Using Multiple-Point Statistics. AAPG
Bulletin, 86. https://doi.org/10.1306/3FEF4D07-1741-11D7-8645000102C1865D
Sengupta, M., & Bachrach, R. (2007). Uncertainty in seismic-based pay volume estimation: Analysis
using rock physics and Bayesian statistics. The Leading Edge, 26(2), 184–189.
https://doi.org/10.1190/1.2542449
Simm, R., & Bacon, M. (2014). Seismic Amplitude. Cambridge: Cambridge University Press.
https://doi.org/10.1017/CBO9780511984501
Strebelle, S. (2000). Sequential Simulation Drawing Structures from Training Images. UMI.
Strebelle, S. (2002). Conditional simulation of complex geological structures using multiple-point
statistics. Mathematical Geology, 34(1), 1–21. https://doi.org/10.1023/A:1014009426274
Teixeira, R., Braga, I., & Loures, L. G. (2007). Bayesian Characterization of Subsurface Lithofacies
and Saturation Fluid. In Proceedings of Latin American & Caribbean Petroleum Engineering
Conference. Society of Petroleum Engineers. https://doi.org/10.2523/108027-MS
Wang, Y., Deng, W., & Morozov, I. B. (2018). A Simple Wavelet-Estimation Approach for Well-Log
to Seismic Tying, 1–3.
Yi, B. Y., Lee, G. H., Kim, H.-J., Jou, H.-T., Yoo, D. G., Ryu, B. J., & Lee, K. (2013). Comparison of
wavelet estimation methods. Geosciences Journal, 17(1), 55–63.
https://doi.org/10.1007/s12303-013-0008-0
Yu, X., Ma, Y. Z., Gomez, E., Psaila, D., Pointe, P. La, & Li, S. (2011). Reservoir Characterization
and Modeling. In Uncertainty Analysis and Reservoir Modeling (pp. 289–309). American
Association of Petroleum Geologists. https://doi.org/10.1306/13301421M963458
Zhang, B., Chang, D., Lin, T., & Marfurt, K. J. (2015). Improving the quality of prestack inversion by
prestack data conditioning. Interpretation, 3(1), T5–T12. https://doi.org/10.1190/INT-2014-
0124.1
Figure captions
Figure 1. An arbitrary line sections of seismic derived elastic properties along the well locations: (a)
P-Impedance (ZP) & (b) VP/VS ratio.
Figure 2. Generation of lithofacies types to converting the seismic elastic properties into lithofacies
volume using the wells: (a) Cross-plot analysis, (b) Probability density functions in the cross plot
space, (c) confusion matrix, (d) The comparison of true lithofacies (well log data) and predicted
lithofacies (Bayesian results).
Figure 3. The Bayesian classification results and comparison of 3D slices with seismic derived elastic
properties: (a) Arbitrary line of the lithofacies section along the well logs, (b) 3D slice of the P-
Impedance, (c) 3D slice of the VP/VS ratio, (d) 3D slice of the lithofacies volume.
Figure 4. Generating the training image
Figure 5. The results of the SNESIM algorithm in MPS: (a) & (b)The comparison of the 3D slices of
the Bayesian classification and MPS simulation to identify the development of the HC lithofacies
(Arrow marks) (c) & (d) The comparison of the 3D slices of the Bayesian classification and MPS
simulation to identify the development of the brine sand lithofacies (Arrow marks), (e) & (f) The
comparison of confusion matrix for improvement of the lithofacies for two wells.

You might also like