Seismic Geomorphology-Based Facies Classification: Interpreter'S Corner

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

INTERPRETERS CORNER

Coordinated by Rebecca B. Latimer

Seismic geomorphology-based facies classification


HONGLIU ZENG, Bureau of Economic Geology, Austin, Texas, U.S.

eismic facies analysis is a powerful, yet largely subjective,


methodology for hydrocarbon exploration and development.
The term seismic facies means different things to different
people. For this study, however, seismic facies is defined as a
seismic description of depositional facies confined in a seismically thin bed at a normal seismic sampling rate (e.g., 4 ms),
making this studys seismic facies analysis different from classical seismic facies analysis of much thicker depositional
sequences. As a subjective interpretation of seismic attributes,
seismic facies analysis is often laborious and inconsistent,
especially if multiple mapping units are involved. How to
automate the process via computer pattern recognition so
that consistency could be improved and the 3D cycle time
reduced has been a hot topic in recent years.
Depending on the scale of geologic objects being explored,
there are three different approaches to pattern recognition of
seismic data: waveform classification, texture mapping, and
seismic geomorphology imaging. Waveform classification is
the analysis of the shape of the amplitude trace (or other
attribute traces) at individual sample points and the relating
of different trace shapes to different depositional facies.
Currently most widely used commercial software applies this
approach. Texture classification started from seismic stratigraphy. Mitchum et al. (1977) defined a classic seismic facies as
a group of seismic reflections whose parameters (configuration, amplitude, continuity, apparent frequency, and interval
velocity) differ from those of adjacent groups. Attempts have
been made in recent years (e.g., West et al., 2001) to quantify
seismic facies by grouping texture attributes derived from the
gray-tone co-occurrence matrix (Haralic, 1979). Seismic geomorphology (Posamentier, 2001) is the study of ancient landforms mapped on depositional surfaces using seismic data.
Geomorphology of a depositional system can be revealed
from 3D seismic data by extracting seismic attributes on interpreted depositional surfaces. However, geomorphology has
so far received little attention for automated facies analysis.
The purpose of this study was to show that seismic geomorphology indeed provides independent and useful information for automated facies recognition. An ideal seismic
facies analysis tool should be able to handle information at
all three scales using seismic data. For a geomorphologybased seismic facies analysis, seismic image maps (stratal
slices) must be picked at depositional surfaces so that maximum preservation of the plano-form morphology of the depositional systems is achieved (Zeng et al., 1998).
Geomorphology as a seismic facies indicator. It is common
knowledge that a well-data-based depositional facies analysis should integrate studies at four different scales: (1) sedimentary texture, mineralogy, and bedding features from core
observation; (2) lithology, thickness, and vertical-grain-size
trend from wireline-log patterns; (3) lateral facies association
and cross-sectional geometry from closely spaced log sections;
and (4) plan-view geometry of regional framework facies. The
data for each scale may contain independent information for
facies analysis that is not included by the others. The same
should hold true for seismic facies analysis. From seismic
644

THE LEADING EDGE

JULY 2004

Figure 1. Morphology of
fluvial depositional systems.
From left to right: straight
(bed load), sinuous (mixed
load), and anastomosing
(suspended load). Modified
from Galloway and Hobday
(1983).

Figure 2. Morphology-based
facies classification of fluvial
systems. Facies 1 = anastomosing (suspended load)
channel. Facies 2 = floodplain. Facies 3 = straight
(bed load) channel. Facies 4
= sinuous (mixed load)
channel.

Figure 3. Amplitude stratal slices of three fluvial systems and related


channel waveforms: (a) straight, (b) meandering, and (c) anastomosing.
Although their channel morphologies are different, the three fluvial systems are characterized by a similar seismic waveform.

data, scale 1 information is typically not available because of


limited data resolution. Scale 2-4 information, however, can
be extracted from 3D seismic data by waveform analysis, texture mapping, and seismic geomorphology imaging, respectively.
Assuming a similar waveform/texture combination, which
occurs commonly within same-facies groups, plan-view morphology can be vital in differentiating various depositional systems. For example, in the fluvial-facies group, there are three
types of fluvial systemsbed load, mixed load, and suspended loadeach characterized by different channel-fill
composition, internal structure, and geometry (Galloway and
Hobday, 1983), implying different reservoir characteristics.
However, all three show a similar upward-fining grain-size
trend, which generates similar seismic waveforms, given similar thickness and wavelet. Their cross-sectional bedding styles
and dimensions may or may not be the same, however, lead-

Figure 4. Waveform-based facies classification of three fluvial systems.


Supervised neural network training was performed in a 16-ms window
centered at the slice in Figure 3. Facies -1 = anastomosing channel network. Facies 0 = floodplain. Facies 1 = straight channel. Facies 2 = sinuous channel.

Figure 5. Morphology-based facies classification of three fluvial systems.


Facies labels same as in Figure 4.

Figure 6. Morphology-based volume classification of incised valley fill


(good reservoir), fluvial channel (marginal reservoir), and floodplain
(seal).

ing to inconclusive texture mapping. In fact, a more useful


parameter for seismic facies analysis is their plan-view geometry (Figure 1). Shapes of paleo-channel elements (straight, sinuous, and anastomosing) are directly related to fluvial types

(bed load, mixed load, and suspended load, respectively).


Morphology-based neural net classifier. A feed-forward
neural network was selected for supervised morphology classification. Each input node of the neural network takes on the
values of one morphologic attribute, and each output node
represents one facies class. Neural networks are considered
superior to standard statistical classifiers in this study because
they do not assume class distribution and can better handle
the nonlinearity of seismic data.
What is key is providing the neural network with morphologic attributes that can effectively describe different morphology groups (facies) in a statistical sense. Examples include
(but are not limited to) size, orientation, geometric tendency
(e.g., line or circle), and curvature. Supervised learning
involves converting sample depositional facies (visually picked
by geologists) into facies categories labeled by arbitrary numbers (1, 2, 3, etc.). If results are satisfactory, the rule learned
from the training can then be applied to the whole seismic
data set for automated depositional facies mapping.
Figure 2 shows the morphology-based classification of
fluvial systems. Results clearly identify systems having
straight, sinuous, and anastomosing patterns as different fluvial facies (Figure 2). In processes, waveform (amplitude) is
assumed insensitive to morphologic variations across facies
spectra, and no texture attributes are independently analyzed.
Real seismic example: three fluvial systems. In an offshore
Louisiana 3D seismic volume, interpreters identified three
different fluvial channels on the basis of geomorphology on
stratal slices (Figure 3): straight, meandering, and anastomosing. The task was to classify them as different seismic facies
so that a more quantitative evaluation of the depositional-morphology-reservoir-quality relationship could be performed.
Unfortunately, channels in different facies are characterized
by a similar amplitude range. Worse, waveforms of wiggle
traces penetrating different-type channels are, in fact, statistically identical. Their texture attributes (continuity and cooccurrence matrix) are also statistically similar. Consequently,
a waveform classification and texture analysis generated poor
results (e.g., the waveform classification results in Figure 4).
Only a morphology-based neural-network classification
(Figure 5) correctly identified the three types of channels as
three different facies. The result is not perfect, though, because
some minor misclassifications remain in each of the facies
maps. The ambiguity that occurs in automated facies mapping has apparently been reduced, but it has not been eliminated.
Facies subvolumes: IVF, fluvial channel, and floodplain. One
advantage of morphology-based, automated, seismic facies
analysis is its ability to trace depositional (stratigraphic) features in a 3D seismic mass at great speed and in great detail,
even if they do not stand out as seismic amplitude, waveform,
or texture anomalies that are sensitive to interpreting eyes. If
a stratigraphic feature is related to potential production, the
tool is also useful for developing play and trapping concepts.
For example, in a seismic stratal-slice volume (Figure 6a),
incised valley-fill (IVF) sandstones, fluvial-channel sandstones,
and floodplain shale exhibit different geomorphologic characteristics. Amplitude and waveform were not conclusive in
our attempt to distinguish sandstones of IVF origin from those
of fluvial origin. Instead, a slice-by-slice (at a 4-ms sample rate)
implementation of the rules learned from morphology-based
neural-network training helped us to quickly generate a seismic lithofacies volume (Figure 6b) including three seismic
(Continued on p. 688)
JULY 2004

THE LEADING EDGE

645

(Zeng, from p. 645)

facies, which was then easily subdivided into three subvolumes representing the three lithofacies (Figures 6c-e).
Considering that 90% of the hydrocarbon reserves in the region
have been produced from the IVF and other lowstand reservoirs, these results will be beneficial for future exploration and
infill drilling in the area. TLE
Suggested reading. Neural net generated seismic facies map and
its application in various geologic environments: examples from
south Texas basin by Addy (www.paradigmgeo.com, on waveform classification in Stratimagic). Interactive seismic facies classification of stack and AVO data using textural attributes and
neural networks by West et al. (SEG 2001 Expanded Abstracts).
Seismic geomorphology and depositional systems of deep water
environments; observations from Offshore Nigeria, Gulf of
Mexico, and Indonesia by Posamentier (AAPG 2001 Abstracts).
Stratal slicing, Part I: realistic 3-D seismic model; and Part II: real
seismic data by Zeng et al. (GEOPHYSICS, 1998). Terrigenous Clastic
Depositional Systems, Applications to Petroleum, Coal, and Uranium
Exploration by Galloway and Hobday (Springer-Verlag, 1983). TLE
Acknowledgments: I am grateful to Lesli J. Wood for her encouragement
and support of this study and Dingshan Zhou for his contribution in neuralnetwork computing. Charles Kerans, Eric C. Potter, and David C. Jennette
are also graciously acknowledged for their helpful comments and technical
input. The study was funded by the U.S. Department of Energy (DOE),
under Cooperative Agreement No. DE-FC26-98FT40136. However, any
opinions, findings, conclusions, or recommendations expressed herein are
those of the author and do not necessarily reflect the views of the DOE.
ChevronTexaco contributed the well log, production, and 3D seismic data.
Landmark Graphics provided software via the Landmark University Grant
Program for interpretation and display of seismic data. Published by permission of the director, Bureau of Economic Geology.
Corresponding author: hongliu.zeng@beg.utexas.edu

688

THE LEADING EDGE

JULY 2004

You might also like