Interactive seismic facies classification using textural attributes

and neural networks

BRIAN P. WEST, STEVE R. MAY, JOHN E. EASTWOOD, and CHRISTINE ROSSEN, ExxonMobil Upstream Research Company, Houston, Texas, U.S.

I n this study, we present an application of textural analy-

Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

sis to 3D seismic volumes. Specifically, we combine image

textural analysis with a neural network classification to
quantitatively map seismic facies in three-dimensional data.
Key advantages of this approach are:

1) it produces a detailed 3D facies classification volume

(whereas manual seismic facies classifications are typi-
cally 2D maps),
2) it enables rapid and quantitative analysis of the increas-
ingly large seismic volumes available to the interpreter,
and
3) it eliminates many time-consuming tasks, thereby free-
ing the interpreter to focus on determining seismic facies
and integrating them into a geologic framework.

Finally, we extend our textural analysis-based seismic

facies classification technique to interpretation of AVO
attribute volumes, such as “A + B” (AVO intercept + gradi-
ent), to reduce the inherent nonuniqueness of seismic facies
to geologic and lithologic facies, and simplify the facies
analysis of complex, mixed-impedance reservoirs.

Seismic facies analysis. Seismic facies analysis is a power-

ful qualitative technique used in stratigraphic analysis from
seismic data and in hydrocarbon exploration. Seismic facies
are groups of seismic reflections whose parameters (such as
amplitude, continuity, reflection geometry, and frequency) Figure 1. Examples of seismic facies and potential associated geologic
differ from those of adjacent groups. Seismic facies analy- fill. A seismic facies can be defined as a stratigraphic region in the
sis involves two key steps—(1) seismic facies classification seismic data volume that has a characteristic reflection pattern distin-
(i.e., seismic facies are defined, and lateral/vertical extents guishable from those of other areas on the basis of reflection amplitude,
continuity, geometry, and/or internal configuration of reflectors.
delineated) and (2) interpretation (i.e., analysis of verti- Inherent in a seismic facies analysis, however, is the nonunique rela-
cal/lateral associations, map patterns, and calibration to tionship between seismic data, seismic facies, environment of deposition
wells) to produce a geologic and depositional interpretation. (EOD), and rock property relationships. Here, typical deepwater seis-
This interpretation step is required because there is a mic facies have been interpreted to represent the stacking patterns
nonunique relationship between seismic data, seismic facies, illustrated beneath each seismic example.
and depositional environment or rock property relation-
ships (Figure 1).
In the past, the seismic facies mapping or classification
step has occurred through time-consuming, manual meth-
ods. Seismic facies are conventionally delineated in the con-
text of mapped horizons (i.e., the interpreter analyzes seismic
facies that occur between mapped horizons). This is done
by examining successive vertical sections through the seis-
mic volume to determine the dominant seismic facies that
occurs between the mapped horizons, and posting this infor-
mation on a map. The output of this step is therefore a 2D
map that generalizes the distribution of seismic facies ver-
tically within a mapped interval. In large and complex areas,
it may be difficult to map different seismic facies consistently.
Manual seismic facies mapping, although time-con- Figure 2. GLCMs for identical subregions with differing azimuths
suming and qualitative, has proven extremely useful for results in very different GLCMs. When constructing GLCMs, compar-
hydrocarbon exploration and reservoir characterization, ison orientation is a significant variable that must be considered.
even when seismic facies cannot be uniquely related to
physical properties. A skilled interpreter’s knowledge and preter skill and experience. We believe that volume-based
experience contribute greatly to the success of seismic facies seismic facies mapping using textural image analysis and
analysis. However, with increasingly large 3D seismic vol- an interpreter-trained neural network is an important
umes, a more efficient and quantitative, 3D or volume-based advance toward that goal.
approach is required but one which still incorporates inter-

1042 THE LEADING EDGE OCTOBER 2002

 ple, 8-bit seismic data have 256 gray levels, and a GLCM con-
structed from 8-bit data will have 256 rows and 256 columns.
In many instances even less dynamic range is needed for effi-
cient GLCM calculation. For example, in many cases 5-bit
data (32 ǂ 32 GLCM) are sufficient for seismic facies analy-
sis. GLCMs are constructed by comparing pixel relationships
within a subregion for a specific distance (e.g., 4 pixels apart),
in a specific direction (e.g., subparallel to a datumned time
Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

surface). Each element of the GLCM represents the relative

frequency of occurrence of two pixels, the “co-occurance,”
within the subregion. If pixel A, for example, has value i and
is at defined distance D, strike σ, and dip θ, from pixel B with
value j, then the GLCM matrix location i,j (row, column) will
be incremented by one. If another example within the subre-
gion has the same relationship, then that GLCM element is
incremented again. This process of “tallying” relative occur-
rence is performed for each existing pixel pair within the sub-
region to produce the final GLCM for that area. For a sufficient
description of the data region, several GLCMs, with varying
orientations and “look distances,” are constructed.
Consider a two-dimensional image, taken from a syn-
thetic “checkerboard” data volume (Figure 2). Constructing
a GLCM, with a distance of two pixels and a diagonal com-
parison direction, begins with the comparison between pixel
(“A,” 1, 1 - white) and pixel (“B,” 2, 2 - white). Pixel A has a
value of 1 and pixel B also has a value of 1; thus the matrix
Figure 3. GLCMs calculated from the extracted seismic data subre- element in the first row and first column will be incremented
gions for the four different seismic facies of Figure 1. The left GLCM by one. This process continues until all possible transitions at
for each seismic facies class is calculated along the direction of dip at a this distance and azimuth are considered.
user-defined distance. The right GLCM is calculated in the direction of Next, consider a different possible orientation, a distance
dip at twice that distance to effectively characterize reflection continu- of three pixels and a horizontal azimuth. In this case, pixel
ity. As illustrated in this figure, a number of trends emerge. First, “A” (1, 1) again has a value of 1 while pixel “B” (1, 4) has a
continuous regions exhibit a tighter distribution along the upper left,
lower right diagonal than do semicontinuous regions (the high ampli- value of 8. With this transition, the last row and first column
tude continuous versus high amplitude semicontinuous examples). of the GLCM will be incremented by one and the process will
Second, higher amplitude regions have a greater extent along this same again continue until all possible pixel transitions for that
diagonal (the moderate amplitude versus high amplitude semicontinu- azimuth and distance have been recorded into the GLCM.
ous examples). These characteristics are quantified through textural
attributes and used for the seismic facies classification. Finally, within Seismic-based GLCMs. The structure of seismic-based
a specific seismic facies, the GLCM resulting from greater comparison
distance always exhibits a less continuous character (more dispersed GLCMs is quite simple to understand. Homogeneous regions,
along the lower left, upper right diagonal). This characteristic illus- or instances where the GLCM is calculated in direction of max-
trates the similarity of a GLCM calculation to the production of seismi- imum continuity (e.g., parallel to continuous high signal-to-
cally defined variograms. noise reflections) exhibit a tight distribution along the diagonal
(top left to bottom right) of the GLCM. This results from com-
Textural analysis and grey-level co-occurrence matrices parison of pixels with similar values, effectively a cross-plot
(GLCMs). Textural analysis can quantitatively describe many of correlated values. Less homogeneous or discontinuous
aspects of the classic seismic facies analysis performed by the regions will have more occurrences (counts) farther away
interpreter. Stratigraphically-steered, seismic texture is a quan- from the GLCM diagonal, resulting from disparate pixel com-
titative, multitrace attribute suite, that mimics the visual parisons. Pixel value magnitude is also captured in the GLCM.
process of the interpreter more effectively than traditional, sin- Regions of low amplitude have GLCMs with values clustered
gle trace-based attribute analyses. For textural analysis, the near the center (effectively the zero crossing). Regions that
interpreter examines an ensemble of traces as an image to ren- include higher amplitudes will have more broadly distributed
der a classification rather than examining only one or two adja- values within the GLCM either along the diagonal for con-
cent traces at a time. The goal of textural analysis is to tinuous textures, or throughout the GLCM in more discon-
mathematically describe the distribution of pixel values within tinuous textures (Figure 3).
a subregion of data, effectively quantifying the spatial orga- Careful inspection of the GLCMs in Figure 3 demonstrates
nization of seismic reflections. Just as the physical texture of that despite the variability in the character, there is always sym-
a object can be loosely described as relatively smooth or rough, metry about the upper left-lower right diagonal. In the cur-
the visual “texture” of a region can be intuitively thought of rent study, GLCMs are symmetric matrices about the upper
as smooth or rough, continuous or discontinuous. The tech- left-lower right diagonal due to transitions from pixel a to pixel
nique commonly used to quantify data texture employs a b being considered equally important as a transition from b
transformation that results in “grey-level co-occurrence matri- to a. Further research is exploring the possibility of exploit-
ces” or GLCMs. GLCMs, also referred to as transition proba- ing anisotropic GLCMs in the context of depositional “grains”
bility matrices or “tally” matrices, quantitatively describe the to extract geometries with geologic significance. Geometries
spatial relationships and relative occurrence of pixel values such as “onlap,” “mounding” from differential compaction,
within a defined region. and “shingles” representing progradation are of particular
GLCMs have dimensions N ǂ N; N is the number of gray interest.
levels (dynamic range) used to quantify the data. For exam-

OCTOBER 2002 THE LEADING EDGE 1043

 Dip-steering the analysis. Seismic facies must be consid-
ered within the structural/stratigraphic framework of a
data set and the desired scale for interpretation. Textural
analysis applied to seismic data must follow the strati-
graphic dip of the reflections to produce satisfactory results.
Texture analysis and construction of a subregion’s GLCM
is extremely sensitive to the “look-direction” or azimuth in
which the pixels within the region are related. Following
Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the stratigraphic dip in a GLCM calculation, dynamically

changing θ depending on the stratigraphy, maximizes the
continuity of the image as expressed in the GLCM. The
process of guiding a calculation by stratigraphic dip is called
dip-steering.
Dip-steering of a GLCM calculation requires quantita- Figure 4. Generalized diagram of the workflow for interactive seismic
tive knowledge of the orientation of seismic reflections at facies classification with iterative quality control.
all points within an image. Many methods for dip-steering
calculations are available. “Coherency” or “discontinuity” valid comparisons to produce a representative GLCM.
based dip-steering methods follow stratrigraphic frame-
works through calculated trace to trace shifts in the lag dis- Textural attributes. GLCMs are not efficiently interpreted
tance. Others are based on an initial “skeletonization” of the directly; instead summary statistics calculated from the GLCM
reflections within a data volume. To further exploit the are exploited. Textural attributes are divided into first- and
image-based nature of textural analysis techniques, we adopt second-order descriptors. First-order statistics quantify the
a gradient-based dip-steering algorithm. The multitrace global distribution of pixel values within an image and can
nature of the technique is thus exploited, and dips within be calculated directly using standard statistical techniques
an image are estimated with high-accuracy. The first step in without an intermediate GLCM transformation. Within the
this process requires calculation of the horizontal (dx) and region of interest, average absolute amplitude values and
vertical (dy) gradient of pixel values within the image. standard deviation of amplitude values can be considered first-
The local apparent dip of the reflectors is then calculated order textural attributes and useful in delineating amplitude
via: anomalies and reflection strength. Derived attributes such as
instantaneous amplitude, phase, and frequency can also be
used to produce first-order statistics. Second-order statistics
With dy/dx having units of time per CDP; however, for of an image quantify the spatial relationships of pixels within
convenience, these units can be ignored and the dip can be the image and are calculated via the intermediate transform
expressed in terms of “pseudo-radians” or “pseudo-degrees” to the GLCM. Second-order GLCM statistics can capture trace
relative to a horizontal time slice. For fully three-dimensional shape characteristics, reflection geometry, and reflection con-
analyses, two orthogonal apparent dip calculations are per- tinuity, in addition to amplitude strength. Second-order sta-
formed and the results used to estimate the true dip and tistics of a subregion are a multitrace attribute, which allows
azimuth for dip and directional steering of the GLCM cal- reflection geometry and continuity to be captured through
culation. analysis of the dip-steered GLCM.
Once reflection dip and azimuth are known everywhere, The utility of various textural attributes for seismic facies
this information guides the dip and azimuth of the look analysis continues to be a topic of research. A general textural
direction for the GLCM calculation. Due to the discrete description can be achieved with four commonly used tex-
nature of a seismic image (CDPs and time samples), a con- tural attributes: homogeneity, inertia, entropy, and energy.
tinuum of angular values is not feasible. Because the cal- The mathematical expression of these GLCM attributes is:
culation is performed over multitraces, the potential angular
resolution is inversely proportional to the distance over
which the comparison is made, and the size of the analysis
subvolume. For example, a dip-accuracy of 15 pseudo-
degrees, can be approximated if the width is equal to three
pixels.
In our application, the interpreter controls the look dis-
tance, the size of the analysis subvolume, and the length scale
for dip-steering. The size of the subvolume depends on the
scale of the features the user wishes to analyze, and the com- where cij is the ith, and jth component of GLCM, c, and n is
parison distances depend on the length over which continu- the size of the matrix (squared number of gray levels within
ity must be judged. In reservoir characterization, a typical the image). The input GLCM, c is normalized such that:
subvolume might be 325 m on a side and approximately 50
m vertically. Within this subvolume, we generally calculate
three GLCMs. Two GLCMs will be calculated in the direction The GLCM characteristics that these mathematical rela-
of maximum seismic continuity (dip-steered) at distances of tionships quantify are simple to understand. Textural homo-
approximately 40 and 80 m. The third GLCM is calculated geneity measures the similarity of pixels. Homogeneity will
orthogonal to the seismically defined dip and can be used as be high for GLCMs with elements concentrated near the diag-
a metric for frequency content and shape of the seismic data onal. Conversely, low homogeneity values result from highly
within the analysis window. Statistical stability considera- contrasting pixel values in the comparison orientation. These
tions also play a factor in deciding the comparison distances characteristics make textural homogeneity particularly use-
within the subvolume. Generally, comparison distances aver- ful for quantifying reflection continuity.
age 20% of the overall subvolume dimension to ensure enough Textural inertia is also indicative of the contrast of the

1044 THE LEADING EDGE OCTOBER 2002

 Figure 5. (upper) A typical training
session showing the definition of train-
Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ing polygons on a seismic section. The

training regions are provided directly
by the interpreter. (lower) Same seismic
section showing local dip of the seis-
mic. Yellow represents dips down to the
left; grays represent dips down to the
right.

Figure 7. During the training phase, summary characteristics of the

digitized training polygons (denoted by numbers on the chart) are
calculated from the full suite of textural attributes and are plotted to
assist the interpreter in training the probabilistic NN.

ment or set of elements, corresponding to a particular pixel

Figure 6. Examples of two seismic facies training polygons from Figure to pixel transition, is favored.
5, representing high-amplitude continuous (HAC) seismic facies (left)
and moderate amplitude semicontinuous (MASC) seismic facies Textural energy (sometimes called uniformity) is also
(right). Using the digitized examples, gray-level co-occurrence matrices indicative of the spatial organization. Energy is lowest when
(GLCMs, lower panels) are extracted and, from these GLCMs, textural all elements of the GLCM are equal, nearly opposite of the
attributes are calculated. Note that the data in these figures were textural entropy and very useful for highlighting regions of
rescaled to 5 bit precision (32 gray levels), thereby reducing the overall reflector continuity and geometry. Although energy and
size of the GLCM and expediting subsequent calculations. In this entropy are metrics of similar characteristics, our experience
figure, three GLCMs are produced from the polygon area, two in the
direction of dip at differing distances (upper subimages), and one has shown that their combined use leads to more satisfac-
orthogonal to the dip direction (lower subimage). tory results.

GLCM. Whereas homogeneity will be low for a highly con- Classification of textural attributes: probabilistic neural
trasted data, textural inertia will be high. “Inertia” is a mea- networks. Attribute classification is the mechanism by which
sure of the distance-weighted density of points away from a continuum of quantitative seismic characteristics is related
the GLCM diagonal, similar to the physical property iner- to a discrete number of classifications within the volume,
tia. Although textural inertia and homogeneity are related, effectively a “dimension reduction exercise” to simplify the
experiments with dip-steered calculations demonstrate that interpretation of the attributes. Neural networks are often
together they provide more seismic facies discrimination used for statistical analysis and data classification. A neural
than either alone. network is an interconnected assembly of simple process-
Textural entropy measures the organization of pixels. ing elements. The processing ability of the network is stored
Entropy is large when the values of the GLCM are uniform, in the connection strengths, or weights, obtained by a process
corresponding to a scenario when all transitions are equally of adaptation to, or learning from a set of examples. One
probable. Textural entropy is low when a particular ele- advantage of neural networks is the ability to “train” or mod-

OCTOBER 2002 THE LEADING EDGE 1045

 Figure 8. A typical
seismic facies classi-
fication using the
interpreter trained
probabilistic neural
network, where
multiple seismic
facies classes have
been identified. The
Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

seismic classifica-
tion scheme on the
right consists of
high amplitude
(HA), moderate
amplitude (MA),
low amplitude (LA),
continuous (C) and
semicontinuous
(SC) seismic facies.

analysis, a seismic facies classification can occur in princi-

ple with as little as one good example per facies class with
the textural attributes of the training images supplying the
weight vectors in the first layer of the network.
When an input pattern is presented to the probabilistic
Figure 9. A relative confidence section resulting from the extraction of
neural network, the first layer computes distances from the
probabilities from the neural network output. Low confidence values can input vector to the training input vectors, and produces a
be observed in and near fault zones and in areas of facies transition. vector whose elements indicate how close the input is to a
training input. The second layer sums these contributions
for each class of inputs to produce, as its net output, a vec-
tor of probabilities. A second advantage of the PNN in pat-
tern recognition is the ability to extract classification
probabilities directly from the second, “hidden” layer, in
addition to the classification of the maximum probability
from the output layer. The resultant data quantifies the con-
fidence of the classification output based on the input train-
ing set. Finally, a “competitive transfer function” (a function
which simply selects the maximum value) on the output of
the second layer picks the maximum of these probabilities,
and produces a “1” for the chosen class and a “0” for the
other classes.

Interactive training and iterative QC. Figure 4 shows the

workflow for seismic facies classification using textural
analysis and neural networks. Two extremely important
aspects of the technique are: (1) interactive training of mul-
Figure 10. Result of the textural analysis using neural networks is a
seismic classification volume (upper left). This volume can be examined in tiple seismic facies classes simultaneously and (2) the capa-
vertical section or horizon slice mode, and compared to other seismic bility to do iterative training and quality control checking
attribute volumes, in addition to available well or core data, in order to of results between the seismic interpreter and the neural net-
develop a geologic interpretation of the seismic facies classification. In this work classification prior to full data analysis. This iterative
example from a channelized deepwater reservoir, a slice from the seismic approach for multiple, user-defined classes enhances the
facies classification volume (middle image) is compared to an equivalent ability of a probabilistic neural network to reproduce the
slice from the seismic discontinuity volume (lower right). Calibration of
these features to available well data results in the environment of deposi- geologically significant classifications.
tion interpretation map shown in Figure 11. Training the neural network begins with the definition
of textural analysis parameters, such as calculation volume
ify the connection strengths within the network to produce size, dip-steering window size, bit depth of the data, and
desired results. Computationally, the connectivity of the analysis distance. Specification of these parameters is based
nodes within a general neural network (the weights), mod- on the experience of the interpreter, and actual parameter
ify an input “vector” of attributes, and pass the modified values are dependent on the frequency and spacing of the
values on to the next layer of the network. Once sufficiently data, as well as the scale of the analysis. To train the neural
trained on a number of “calibration” images, the neural net- network, the interpreter selects one or several key seismic
work can then be applied to the remaining images in a data lines, and digitizes polygons of a consistent scale that are
volume. examples of different seismic facies (Figure 5, upper). The
Probabilistic neural networks (PNNs) can efficiently per- computer automatically calculates “on-the-fly” the data and
form pattern classification. Mathematically, these neural dip/azimuth information associated with the training poly-
networks are similar to kriging, where proximity to known gon (Figure 5, lower) and performs a textural analysis on
points (attribute values in the training set), guide the clas- these subregions (Figure 6). The interpreter inspects the
sification and prediction of unknown points. Probabilistic resulting GLCMs and can either reject the training polygon
neural networks do not require extensive training. In our or accept it and classify it as a specific seismic facies. The

1046 THE LEADING EDGE OCTOBER 2002

 Figure 11. Environment of deposition and
net:gross distribution map for slices shown
in Figure 10. This discontinuity slice high-
lights the lateral edges of the channel (red
lines), a broad, older sinuous element (1)
and a narrower, younger sinuous element
(2). Comparison with the seismic facies slice
shows that sinuous element (1) is composed
of HAC to MASC seismic facies, whereas
Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

sinuous element 2 is primarily composed of

HAC seismic facies.Because the textural
analysis seismic facies classification is a
volume, this type of analysis can be applied
at numerous stratigraphic levels within an
interval of interest, whether or not these
intervals are bounded by mapped horizons.
Combining these results with the conceptual
relationships illustrated in Figure 1, the net
result of the analysis is effectively an "envi-
ronment of deposition volume" where 3D
regions can delineate differing depositional
and geologic properties.

Figure 12. Example of seismic data for

mixed impedance sands for near, far, full,
and A+B volumes.

interpreter should specify several examples of each seismic modified through deletion of existing polygons and/or
facies class, and these examples should be distributed addition of new polygons. The neural network is then recre-
throughout the 3D seismic area of interest. In the initial ated with the modified training set, and again checked, until
training phase, plots of summary attributes, such as ampli- the interpreter is satisfied with the results.
tude and reflection continuity, assist the interpreter in pro- After training and QC (i.e., the interpreter is satisfied that
viding consistent training examples to the neural network the algorithms are producing the facies classification which
(Figure 7). would have been arrived at through manual interpretation),
Once initial training is complete, a probabilistic neural the network and textural analysis algorithms can be applied
network is constructed from the textural attributes and their to the rest of the data volume. The result is a seismic facies
associated classifications. This initial neural network is then classification volume which has a seismic facies classifica-
used to classify a portion of the data set (typically several tion for every trace and sample within the volume (Figure
key seismic lines) to facilitate interactive quality control and 10) and an associated relative confidence volume. Key
analysis. Determining whether the PNN has been trained advantages of the neural net classification volume over a
adequately occurs in the quality control phase of the work- manual classification are: (1) the neural network classifica-
flow. In this step, the interpreter asks “is the computer doing tion is volume-based (rather than map-based for manual
what I would have otherwise done manually and is it geo- classifications), (2) the neural network classification is typ-
logically reasonable?” The interpreter can judge the per- ically much more detailed than any manual classification
formance of the result supplied by the neural network simply and can be produced in a fraction of the time, (3) the clas-
by examining representative classification and confidence sification is based on quantitative criteria and is reproducible
profiles (Figures 8 and 9) prior to classification of larger data given a particular training set. These characteristics promote
sets. If results are unsatisfactory, the training set can be improved efficiency in the analysis of large data volumes,
OCTOBER 2002 THE LEADING EDGE 1047
 and improved capture of lateral and vertical heterogeneities reflection (LAC) on the near offset volume and a high-ampli-
in the seismic volume that are due to stratigraphic or other tude-continuous (HAC) on the far offset volume. Conversely,
factors. a poorly sorted, coarse-grained class I sand may be high-
Although seismic facies classification using textural amplitude continuous on the near offsets and dim to a low-
analysis and neural networks techniques greatly improves or moderate-amplitude, continuous package on the far-off-
the efficiency of seismic classification, the output classifi- set volume. The full stack data would be some average of
cation volume is not the final product. This involves trans- the two end-members, complicating the interpretation.
lation of the amplitude and geometric characteristics of the In a properly constructed AVO attribute volume, lateral
Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

different seismic facies classes into a geologic, depositional transitions between classes I, II, and III can be mitigated. Using
environment, or rock property model. This stage of analy- an AVO attribute volume as the basis for seismic facies clas-
sis still requires strong interpreter input, and typically sification can thus minimize the complexity of interpreting
involves integration of numerous techniques including: multiple stacks of data to identify sand-prone intervals. For
analysis of vertical and lateral seismic facies associations on the location described above, the A + B volume contains a sin-
vertical sections through the classification volume, deter- gle high-amplitude continuous reflection denoting a single
mination of map patterns on horizontal and stratigraphic sand body regardless of AVO class (Figure 12, lower). The
sections through the volume, comparison to the input seis- inclusion of the AVO classification information can be exploited
mic and other seismic attribute volumes, and finally, cali- to further discriminate high-impedance sands from low-
bration to available well-log and core data (Figure 11). impedance sands for rock property assignment.
At the final integration stage, the improved efficiency
and quality of the seismic facies classification volume pro- Potential pitfalls of quantitative facies analyses. As with
duced using textural analysis and neural networks tech- any quantitative analysis, data quality weighs heavily into the
niques translates into more time for the interpreter to focus confidence of the interpretation. In cases where data quality
on interpretation of the classification volume. This enables is suspect, the potential subjectivity of a traditional, manual
the interpreter to produce a more robust geologic analysis. seismic facies analysis can be an asset when interpreters are
aware of the limitations of their data. For example, where data
Seismic facies interpretation applied to AVO attribute vol- are known to be nonstationary across a reservoir interval,
umes. Mixed impedance reservoirs, typical of some deep- manual seismic interpretation can compensate for this vari-
water depositional environments, are composed of sands and ability. Automated seismic facies methods, however, may not
shales, which are not uniquely defined with only a near-off- necessarily take this data variability into account unless spa-
set volume or a full stack volume. In these environments, an tially variant training is employed.
integrated multidisciplinary workflow is especially impor- Further, the use of computer generated facies data with-
tant in delineating the seismic-scale distribution of sand and out appropriate geologic insight and interpreter understand-
shale for reservoir characterization and input to volume- ing of the technology is dangerous. Such application of
based geologic models. Seismic interpretation, including technology can give automated methods a “black-box” rep-
seismic facies analysis, remains a cornerstone of the work- utation and lead to a misunderstanding of the technology being
flow. However, with increasing reservoir complexity, reliance applied, its expected benefits, and its limitations. Fundamental
on sophisticated multitrace and multivolume (including knowledge of the geology and geophysics of an exploration
AVO) attributes is necessary to capture this geologic com- or development target, as well as an appreciation of the tech-
plexity. nology applied, is critical to the successful application of any
We have extended the method for seismic facies analy- technique in an integrated workflow.
sis to these amplitude-versus-offset attribute volumes to
demonstrate how the inherent subjectivity of a facies analy- Conclusions. This article demonstrated:
sis, especially in mixed impedance deepwater environments,
can be reduced with reliable prestack seismic data. In par- 1) The combination of seismic textural analysis and neural
ticular, the AVO properties of the interval, once fluid con- network techniques can be successfully applied to auto-
tacts have been identified, can be fundamental in the mate the seismic classification step of traditional seismic
seismic-scale placement of sand and shale packages. facies analysis, with resulting improvements in the effi-
Deterministic sand placement within facies-based net-to- ciency, degree of detail, and reproducibility of the seis-
gross packages can significantly improve the accuracy of a mic facies classification product. Time-efficiencies and
volume-based reservoir characterization workflow. quality improvements result in more time for the inter-
Among the advantages of using a volume AVO attribute preter to interpret the classification results, and to trans-
is the reduction in nonuniqueness in the seismic data. For late it into a robust geologic framework.
example, using an “A + B” AVO attribute cube, high-ampli- 2) Two- and three-dimensional textural analysis provides a
tude seismic facies can be directly interpreted as likely HC- significantly different approach to seismic facies classifi-
charged sands” whereas low-amplitude facies can be cation over single, trace-based seismic attributes in that
interpreted as shale-prone and/or water-wet. Simplified seismic texture is a multitrace, volume-based attribute that
well-to-facies ties are also a result of this type of application. provides a quantitative measure of the reflection ampli-
We illustrate the predictive power of an AVO attribute tude, continuity, and internal configuration of reflectors.
facies analysis with an example from a stratigraphically com- Seismic textural analysis thus provides a means to quan-
plex, mixed-impedance clastic reservoir. In a mixed imped- tify the description of elements contained in seismic facies
ance scenario, the seismic expression of sand-prone intervals, analysis.
in terms of amplitude and reflection continuity, will vary with 3) The use of neural networks to classify textural attributes
AVO class of the sands and with the stack of data examined provides an efficient method by which the seismic inter-
(Figure 12). Laterally amalgamated sands may also have preter can interactively “teach” the computer the seismic
varying responses depending on the stack of data examined characteristics of interest. Probabilistic neural networks
(Figure 12). For example, a well-sorted, fine-grain class III also offer the advantage of providing a quantitative mea-
sand may be associated with a low-amplitude-continuous sure of the confidence placed in each facies classification,

1048 THE LEADING EDGE OCTOBER 2002

 thereby increasing the efficiency of iterative training. The
confidence value can also be used to dynamically reclas-
sify until a specified confidence value is achieved.
4) Seismic facies analysis of reliable AVO attribute volumes
can significantly reduce the uncertainty and nonunique-
ness of lithologic interpretations based on seismic facies
interpretation. With appropriate data, an AVO facies
analysis may also be used to position sands in a clastic
Downloaded 03/31/15 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

reservoir model.

Suggested reading. “Segmentation of the Mid-Atlantic Ridge

south of the Azores, based on acoustic classification of TOBI
data” by Blondel (in Tectonic, Magmatic, Hydrothermal and
Biological Segmentation of Mid-Ocean Ridges, Geological Society
Special Publication 118, 1996). “Interactive AVO time-align-
ment and neural network classification” by Eastwood and West
(SEG 2001 Expanded Abstracts). An Introduction to Neural Networks
by Gurney (UCL Press Limited, London, 1997). “Statistical and
structural approaches to texture” by Haralick (Proceedings of the
IEEE, 1979). “Three-dimensional texture attributes for seismic
data analysis” by Randen et al. (SEG 2000 Expanded Abstracts).
“Seismic stratigraphy and global changes in sea level, Part 6:
Stratigraphic interpretation of seismic reflection patterns in
depositional sequence” by Mitchum (in AAPG Memoir 26, 1977).
“Digital imaging processing techniques for enhancement and
classification of SeaMarc II side-scan sonar imagery” by Reed
and Hussong (Journal of Geophysical Research, 1989).
“Probabilistic neural networks” by Specht (Neural Networks,
1990). “3D seismic texture classification” by Vinther et al. (SPE
35482, 1996). TLE

Acknowledgments: We thank the numerous ExxonMobil geoscientists who

have contributed to the development and effective use of these techniques:
in particular, J. Ardill, C. Dawson, D. Gao, L. Foreman, M. Porter, R.
Hill, R. Stephens, and L. Magennis. We also thank ExxonMobil’s affili-
ates and partners for permission to publish this work.

Corresponding author: [email protected]

OCTOBER 2002 THE LEADING EDGE 1049