USOO5870691A

United States Patent (19) 11 Patent Number: 5,870,691

Partyka et al. (45) Date of Patent: Feb. 9, 1999
54). SPECTRAL DECOMPOSITION FOR SEISMIC Seismic Data Processing, Society of Exploration Geophysi
INTERPRETATION cists, 1987, Chaptesr 2.2, and 2.2.1, pp. 85-94.
75 Inventors: Gregory A. Partyka; James M.
Gridley, both of Tulsa, Okla. Primary Examiner Donald E. McElheny, Jr.
73 Assignee: Amoco Corporation, Chicago, Ill. Attorney, Agent, or Firm-James A. Gabala; Robert E. Sloat
57 ABSTRACT
21 Appl. No.: 759,655
22 Filed: Dec. 6, 1996 The present invention is directed generally toward a method
of processing Seismic data So as to provide improved quan
(51) Int. Cl." ...................................................... G06F 19/00
52 U.S. Cl. ................................................................ 702/16 tification and Visualization of Subtle Seismic thin bed tuning
effects and other Sorts of lateral rock discontinuities. A
58 Field of Search ..................................... 364/421, 422;
367/68, 71, 72, 73 reflection from a thin bed has a characteristic expression in
the frequency domain that is indicative of the thickness of
56) References Cited the bed: the reflection has a periodic Sequence of notches in
U.S. PATENT DOCUMENTS its amplitude spectrum, Said notches being Spaced a distance
4,742,497 5/1988 Beasley et al. ........................... 367/52 apart that is inversely proportional to the temporal thickneSS
4,797.861 1/1989 Beasley ..................................... 367/50 of the thin bed. Further, this characteristic expression may be
5,274,605 12/1993 Hill ........................................... 367/53 used to track thin bed reflections through a 3-D volume and
5,500,832 3/1996 Berryhill ................................... 367/51 estimate their thicknesses and lateral extent. The usefulneSS
OTHER PUBLICATIONS of this invention is enhanced by a novel method of frequency
"Use of integrated energy spectra for thin-layer recogni domain whitening that emphasizes the geologic information
tion”, Geophysics, vol. 50, No. 3 (Mar. 1985), pp. 495-500. present within the Spectrum. Although the present invention
“How Thin is a Thin Bed?”, Geophysics vol. 38, pp. is preferentially applied to a 3-D Seismic Volume, it is
1176-1180, Dec. 1973. alternatively applied to any collection of Spatially related
"Stratigraphic Modeling and Interpretation-Geophysical Seismic traces.
principles and Techniques, in Seismic Stratigraphy Appli
cations to Hydrocarbon Exploration, A.A.P.G. Memoir 26,
1977, pp. 389-416. 18 Claims, 11 Drawing Sheets

SPECTRAL DECOMPOSITION

y V &
Y a
Wa n Y.

SPLIT
e
SPECTRAL DECOMPOSITION CUBE INTO INDIVIDUAL FREGUENCES
W w
e As y A
. 11
'. Yy.

INDEPENDENTLY SCALE EACH FREQUENCY MAP

y. 'y.
s Ya 1. e1
GATHER INDIVIDUAL FREGUENCES INTO SPECTRAL DECOMPOSITION CUBE
s Y. t? --
sa YA A1 ---
2. SPECTRALLY BALANCED (WHITENED)
SPECTRAL DECOMPOSITION
U.S. Patent Feb. 9, 1999 Sheet 1 of 11 5,870,691

Thick Bed Reflection COMPOSITE

SOURCE THICKBED
WAVELET R REFLECTED REFLECTION
WAVELETS

TEMPORAL THICK
THICKNESS BED

l * - INTERFACE "B"
P.
Fig. IA
Thin Bed Reflection COMPOSITE
SOURCE REFLECTED THINBED
WAVELET R WAVELETS REFLECTION

TEMPORAL THIN
THICKNESS BED

Fig. IB
U.S. Patent Feb. 9, 1999 Sheet 2 of 11 5,870,691

Stationary Spectrum
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0.00 10,00 20,00 30,00 40.00 50.00 60.00
FREQUENCY

FOURIER
TRANSFORM

Time Series

N--

FOURIER FOURIER
TRANSFORM TRANSFORM

FREQUENCY FREQUENCY FREQUENCY

Fig. 2
U.S. Patent Feb. 9, 1999 Sheet 4 of 11 5,870,691

TWHOdEL SNX|OH-1
|

žiae
qž

W N
U.S. Patent Feb. 9, 1999 Sheet S of 11 5,870,691

X
y 1 Time
1 1
3-D SEISMIC VOLUME

INTERPRET
X
y 1 Time
INTERPRETED ZONE OF INTEREST
3-D SEISMIC VOLUME

X
SUBSET
y1 Time
ZONE OF INTEREST
Fig. 5 SEISMIC VOLUME

y Fre
ZONE OF INTEREST
SPECTRAL
DECOMPOSITION
, ANIMATE
X
y Freq
FREQUENCYSLICES
THROUGHSPECTRAL
DECOMPOSITION
U.S. Patent Feb. 9, 1999 Sheet 6 of 11 5,870,691

y 41 Time
1 1
3-D SEISMIC VOLUME 2.
INTERPRET
SUBSET
y - ifreq COMPUTE
M=1

4-D SPECTRAL X> M=2

DECOMPOSITION y - Fre
X- M=3
y 41 ties M=4
y- M=5
y is #1 M = 6
Fig. 6 y-le I2; M7
y-fit,

y 41 Freq
X
y 41 Freq
X
4-D SPECTRAL v -
DEcoP6STION
ANIMATION FILE
fe
---

y 1 Freq
U.S. Patent Feb. 9, 1999 Sheet 7 of 11

WMGAONET OHE
W
=
|

9AONETOH

0-£OEIWNSTIEOSA dSTWHLOE
WNO0LEISO|d
ÅE1HOSI ÅONET H 508|EN EINT OA SE18|[0
4
U.S. Patent Feb. 9, 1999 Sheet 8 of 11 5,870,691

Fig. 8
READ
... DEFINITION OF ZONE OF INTEREST (ZOI)
-TRANSFORM WINDOWLENGTH, L
--NUMBER OF SLIDING WINDOWS, NW
--SLIDING WINDOWS INCREMENT, NS LOAD L, AVG, NW
INITIALIZE AUXILARY LOAD RECORDS FOR Mth
STORAGE AM (j, k, i) DFT WINDOW
INTO T (j, k, i)
READ NEXT TRACE

YES EXECUTE H -
(END OF FILE)YESSEE
NO
CALCULATE
AVERAGE IT (j, k, nil)
SPECTRAL VALUE - TAW G
DETERMINE WRITE FOR j = 1, J
TRACE I.D. LOCATION:
AM (j, k, i)
TO OUTPUT k = 1, K
(j,k)
CEND) CALCULATE:
T(j, k, ni) = AVG T (j, k, n)

STORE RECORD:
-TRACE I.D. 1 LOCATION: (j,k)
-ZO START TIME, M, NS, NW, L
- COMPLEX COEFFICIENTS IN AM (j, k, i)
U.S. Patent Feb. 9, 1999 Sheet 9 of 11 5,870,691

PLANE 1

PLANE 2

PLANE 3

PLANE 1

ANIMATION PLANE 2

Fig. 9C. <sed PLANE 3

 5,870,691
1 2
SPECTRAL DECOMPOSITION FOR SESMC acoustic impedance of the SubSurface materials. The digital
INTERPRETATION Samples are usually acquired at 0.004 second (4 millisecond)
intervals, although 2 millisecond and 1 millisecond intervals
FIELD OF THE INVENTION are also common. Thus, each Sample in a Seismic trace is
The present invention is directed generally toward a asSociated with a travel time, a two-way travel time in the
method of quantifying and Visualizing Subtle Seismic thin case of reflected energy. Further, the Surface position of
bed tuning effects. This method disclosed herein is imple every trace in a Seismic Survey is carefully recorded and is
mented by decomposing the Seismic reflection signal into its generally made a part of the trace itself (as part of the trace
frequency components through the use of a discrete Fourier header information). This allows the seismic information
contained within the traces to be later correlated with
(or other orthonormal) transform of length dependent upon Specific SubSurface locations, thereby providing a means for
the thickness of the bed which is to be resolved. After
decomposition by Said discrete transform, the coefficients posting and contouring Seismic data, and attributes extracted
obtained thereby are organized and displayed in a fashion therefrom, on a map (i.e., "mapping”). The signal that is sent
down into the earth is called a Seismic waveform or Seismic
which reveals and enhances the characteristic frequency 15 wavelet. Different Seismic waveforms are generated depend
domain expression of thin bed reflection events, making ing on whether the Source is an air gun, dynamite or Vibrator.
variations in SubSurface layer thickness Visible that other The term “Source Signature' or "Source pulse' is generally
wise might not have been observed. The present invention used to describe the recorded Seismic character of a particu
allows the Seismic interpreter to analyze and map SubSurface lar Seismic waveform.
geologic and Stratigraphic features as a function of Spatial
position, travel time, and frequency to an extent that has A Seismic Source generated at the Surface of the earth
heretofore not been possible. immediately begins to travel outward and downward from
its point of origin, thereafter encountering and passing
BACKGROUND through rock units in the SubSurface. At each interface
By most Standards exploration geophysics is a relatively between two different rock units, there is the potential for a
25 Seismic reflection to occur. The amount of Seismic energy
young Science, with Some of the earliest work in the Subject that is reflected at an interface is dependent upon the
area dating back to the 1920's and the renowned CMP acoustic impedance contrast between the units and the
approach dating from only the 1950's. In the years since its reflection coefficient is one conventional measure of that
genesis, however, it has become the oil industry's preemi contrast. The reflection coefficient can be thought of as the
nent approach to finding SubSurface petroleum deposits. ratio of the amplitude of the reflected wave compared with
Although exploration geophysics generally encompasses the the amplitude of the incident wave. In terms of rock prop
three broad Subject areas of gravity, magnetics, and Seismic, erties:
today it is the Seismic method that dominates almost to the
point of excluding the other disciplines. In fact, a simple acoustic impedance below -
count of the number of seismic crews in the field has become 35 acoustic impedance above
reflection coefficient
one accepted measure of the health of the entire oil industry. acoustic impedance below +
A Seismic Survey represents an attempt to map the Sub acoustic impedance above
Surface of the earth by Sending Sound energy down into the
ground and recording the "echoes” that return from the rock
layers below. The Source of the down-going Sound energy 40
might come from explosions or Seismic vibrators on land, where, the acoustic impedance of a rock unit is defined to be
and air guns in marine environments. During a Seismic the mathematical product of the rock density (p and pa
Survey, the energy Source is moved acroSS the land above the being the densities of the upper lower rock units,
geologic Structure of interest. Each time the Source is respectively) multiplied times the Velocity in the same rock
detonated, it is recorded at a great many locations on the 45 unit, V and V corresponding to the upper and lower rock
Surface of the earth. Multiple explosion/recording combina unit velocities. (Strictly speaking, this equation is exactly
tions are then combined to create a near continuous profile correct only when the wavelet Strikes the rock interface at
of the SubSurface that can extend for many miles. In a Vertical incidence. However, it is generally accepted in the
two-dimensional (2-D) seismic Survey, the recording loca industry that the requirement of verticality is satisfied if the
tions are generally laid out along a Straight line, whereas in 50 wavelet strikes the interface within about 20 of the
a three-dimensional (3-D) Survey the recording locations are vertical.)
distributed acroSS the Surface in a grid pattern. In Simplest Reflected energy that is recorded at the Surface can be
terms, a 2-D Seismic line can be thought of as giving a croSS represented conceptually as the convolution of the Seismic
Sectional picture of the earth layers as they exist directly wavelet with a subsurface reflectivity function: the so-called
beneath the recording locations. A 3-D Survey produces a 55 “convolutional model”. In brief, the convolutional model
data “cube' or Volume that is, at least conceptually, a 3-D attempts to explain the Seismic Signal recorded at the Surface
picture of the SubSurface that lies beneath the Survey area. as the mathematical convolution of the down going Source
Note that it is possible to extract individual 2-D line surveys wavelet with a reflectivity function that represents the reflec
from within a 3-D data volume. tion coefficients at the interfaces between different rock
A Seismic Survey is composed of a very large number of 60 layers in the SubSurface. In terms of equations,
individual Seismic recordings or traces. In a typical 2-D
Survey, there will usually be several tens of thousands of
traces, whereas in a 3-D Survey the number of individual where x(t) is the recorded seismogram, w(t) is the Seismic
traces may run into the multiple millions of traces. A Seismic Source wavelet, e(t) is the earth's reflectivity function, n(t) is
trace is a digital recording of the Sound energy reflecting 65 random ambient noise, and “*” represents mathematical
back from inhomogeneities in the SubSurface, a partial convolution. Additionally, the convolutional model requires,
reflection occurring each time there is a change in the in part, that (1) the Source wavelet remains invariant as it
 5,870,691
3 4
travels through the Subsurface (i.e., that it is stationary and FIGS. 1a and 1b illustrates in a very general way how the
unchanging), and (2) the Seismic trace recorded at the thin bed problem arises under the axioms of the convolu
Surface can be represented as the arithmetic Sum of the tional model. Consider first the “thick” bed reflection
Separate convolutions of the Source wavelet with each depicted in FIG. 1a. On the left side of this figure is
interface in the Subsurface (the principle of "Superposition', represented a Source wavelet which has been generated on
i.e., that wavelet reflectivity and propagation is a linear the Surface of the earth. The Source wavelet travels down
system.) Although few truly believe that the convolutional ward unchanged through the earth along path P1 until it
model fully describes the mechanics of wave propagation, encounters the rock unit interface labeled “A.” (Note that the
the model is Sufficiently accurate for most purposes: accu wave paths in this figure are actually vertical, but have been
rate enough to make the model very useful in practice. The illustrated as angled for purposes of clarity. This is in
convolutional model is discussed in Some detail in Chapter keeping with the general practice in the industry.) In FIG.
2.2 of Seismic Data Processing by Ozdogan Yilmaz, Society 1a, when the down-going Seismic waveform encounters
of Exploration Geophysicists, 1987, the disclosure of which Interface “A” a portion of its energy is reflected back toward
is incorporated herein by reference. the Surface along path P2 and is recorded on the Surface as
Seismic data that have been properly acquired and pro 15 the reflected event R1. Note that wavelet R1 is reversed in
cessed can provide a wealth of information to the polarity as compared with the Source wavelet, thereby
explorationist, one of the individuals within an oil company indicating a negative reflection coefficient at the “A” inter
whose job it is to locate potential drilling sites. For example, face. This polarity reversal is offered by way of example
a Seismic profile gives the explorationist a broad view of the only and those skilled in the art recognize that reflection
Subsurface Structure of the rock layers and often reveals coefficients of either polarity are possible.
important features associated with the entrapment and Stor The remainder of the down-going energy (after the partial
age of hydrocarbons Such as faults, folds, anticlines, reflection at the interface “A”) continues through the thick
unconformities, and Sub-Surface Salt domes and reefs, bed until it strikes Interface “B” at the base of the thick
among many others. During the computer processing of lithographic unit. Upon reaching the “B” interface, part of
Seismic data, estimates of SubSurface Velocity are routinely 25 the wavelet energy continues deeper into the earth along
generated and near Surface inhomogeneities are detected and path P5, while the remainder of its energy is reflected back
displayed. In Some cases, Seismic data can be used to to the Surface along path P4 where it is recorded as reflection
directly estimate rock porosity, water Saturation, and hydro R2. Note that the reflection from interface “B” occurs at a
carbon content. LeSS obviously, Seismic waveform attributes later point in time than the reflection from interface “A”. The
Such as phase, peak amplitude, peak-to-trough ratio, and a exact time Separation between the two events depends on the
host of others, can often be empirically correlated with thickness and Velocity of the intervening layer between the
known hydrocarbon occurrences and that correlation applied two interfaces, with thick layerS and/or slow Velocities
to Seismic data collected over new exploration targets. In creating a greater time Separation between the top and base
brief, Seismic data provides Some of the best SubSurface reflections. The temporal thickness of this layer is the time
Structural and Stratigraphic information that is available, 35 that is required for the Seismic waveform to traverse it.
short of drilling a well. On the surface, the composite thick bed reflection-the
Seismic data are limited, through, in one crucial regard: expression actually recorded-is the arithmetic Sum
rock units that are relatively “thin' are often not clearly (Superposition) of the two returning reflections, taking into
resolved. In more particular, whereas Seismic reflection data account the time Separation between the events. Because the
can provide a near "geologic croSS Section” representation of 40 two returning wavelets do not overlap in time, the recorded
the SubSurface when the lithologic layers are relatively Seismic record clearly displays both events as indicative of
“thick', the Seismic image that results when the layers are two discrete horizons. (Note that the time separation
“thin' is much less clear. This phenomenon is known to between the two reflected events has not been accurately
those skilled in the art as the Seismic resolution problem. portrayed in this figure. Those skilled in the art know that the
Seismic resolution in the present context will be taken to 45 time Separation should actually be twice the temporal thick
refer to vertical resolution within a Single Seismic trace, and ness of the layer.)
is loosely defined to be to the minimum Separation between Turning now to FIG. 1b, wherein a thin bed reflection is
two Seismic reflectors in the SubSurface that can be recog illustrated, once again a Source wavelet is generated on the
nized as Separate interfaces-rather than as a Single com surface of the earth which then travels along path P6 until it
posite reflection-on the Seismic record. By way of 50 encounters the rock unit interface labeled “C.” (As before,
explanation, a SubSurface unit is ideally recognized on a the wave paths in the figure are actually vertical.) AS is
Seismic Section as a combination of two things: a distinct illustrated in FIG. 1b, when the down-going seismic wave
reflection originating at the top of the unit and a Second form encounterS Interface “C” a portion of its energy is
distinct reflection, possibly of opposite polarity, originating reflected back toward the surface along path P7, where it is
from its base. In the ideal case, both the top and the bottom 55 recorded as reflection R3. The remainder of the down-going
of the unit appear on the recorded Seismogram as distinct energy continues through the thin bed until it Strikes Inter
and isolated reflectors that can be individually “time picked” face “D’. Upon reaching the “D' interface, part of the
(i.e., marked and identified) on the Seismic Section, the wavelet energy continues deeper into the earth along path
Seismic data within the interval between the two time pickS P10, while the remainder of its energy is reflected back to the
containing information about the intervening rock unit. On 60 surface along path P9 where it is recorded as reflection R4.
the other hand, where the Seismic unit is not Sufficiently Note once again, that the reflection from interface “D’
thick, the returning reflections from the top and the bottom occurs at a later time than the reflection from interface “C”,
of the unit overlap, thereby causing interference between the however the temporal Separation between the two reflections
two reflection events and blurring the image of the SubSur in the case of a thin bed is less because the distance the
face. This blurred image of the SubSurface is one example of 65 waveform must travel before being reflected from interface
the phenomena known to those skilled in the art as the “thin “D” is less. In fact, the time separation between the two
bed” problem. reflections is So Small that the returning (upward going)
 5,870,691
S 6
wavelets overlap. Since the composite thin bed reflection is in the proper circumstances, provide Some information
once again the arithmetic Sum of the two returning related to the thickness of the thin bed.
reflections, the actual recorded signal is an event that is not The pioneering work of Widess in 1973 (Widess, How
clearly representative of either the reflection from the top or thin is a thin bed, Geophysics, Vol. 38, p. 1176-1180) has
the base of the unit and its interpretation is correspondingly given birth to one popular approach to thin bed analysis
difficult. This indefinite composite reflected event exempli wherein calibration curves are developed that rely on the
fies the typical thin bed problem. peak-to-trough amplitude of the composite reflected thin bed
Needless to Say, the thickness of a SubSurface exploration event, together with the peak-to-trough time Separation, to
target is of considerable economic importance to the oil provide an estimate of the approximate thickness of the
company explorationist because, other things being equal, “thin” layer. (See also, Neidell and Poggiagliomi, Strati
the thicker the lithographic unit the greater the Volume of graphic Modeling and Interpretation-Geophysical prin
hydrocarbons it might potentially contain. Given the impor ciples and Techniques, in SEISMIC STRATIGRAPHY
tance of accurately determining layer thickness, it should APPLICATIONS TO HYDROCARBON EXPLORATION,
come as no Surprise that a variety of approaches have been A.A.P.G. Memoir 26, 1977). A necessary step in the cali
bration proceSS is to establish a "tuning amplitude for the
employed in an effort to ameliorate the thin bed problem. 15 thin bed event in question, Said tuning amplitude occurring
A first technique that is almost universally applied is at the layer thickness at which maximum constructive inter
Shortening the length of the Seismic wavelet, longer wave ference occurs between the reflections from the top and base
lets generally offering worse resolution than shorter ones. of the unit. In theory at least, the tuning thickneSS depends
During the data processing phase the recorded Seismic only on the dominant wavelength of the wavelet, ), and is
waveform may often be shortened dramatically by the equal to 2/2 where the reflection coefficients on the top and
application of well known signal processing techniques. By bottom of the unit are the same sign, and 2/4 where the
way of example, it is well known to those skilled in the art reflection coefficients are opposite in Sign.
that conventional predictive deconvolution can be used to Because of the flexibility of calibration-type approaches,
whiten the Spectrum of the wavelet, thereby decreasing its they have been used with Some Success in rather diverse
effective length. Similarly, general wavelet processing 25 exploration Settings. However, these amplitude and time
techniques, including Source signature deconvolution and based calibration methods are heavily dependent on careful
any number of other approaches, might alternatively be Seismic processing to establish the correct wavelet phase and
applied to attempt to reach a similar end result: a more to control the relative trace-to-trace Seismic trace ampli
compact waveform. Although any of these processes could tudes. Those skilled in the Seismic processing arts know,
result in dramatic changes to the character of the Seismic however, how difficult it can be to produce a Seismic Section
Section and may shorten the length of the wavelet that truly maintains relative amplitudes throughout. Further,
Significantly, it is often the case that further Steps must be the calibration based method disclosed above is not well
taken. Suited for examining thin bed responses over a large 3-D
Even the best Signal processing efforts ultimately only Survey: the method works best when it can be applied to an
postpone the inevitable: no matter how compact the wavelet 35
isolated reflector on a Single Seismic line. It is a difficult
is, there will be rock layers of economic interest that are too enough task to develop the calibration curve for a Single line:
thin for that wavelet to resolve properly. Thus, other broad it is much more difficult to find a calibration curve that will
approaches have been utilized that are aimed more toward work reliably throughout an entire 3-D grid of Seismic data.
analysis of the character of the composite reflection. These Heretofore, as is well known in the Seismic processing
approaches are based generally on the observation that, even 40
and Seismic interpretation arts, there has been a need for a
when there is only a single composite reflection and the method extracting useful thin bed information from conven
thickness of the layer cannot be directly observed, there is tionally acquired Seismic data which does Suffer from the
still information to be found within the recorded seismic above problems. Further, the method should also preferably
data that may indirectly be used to estimate the actual provide attributes for Subsequent Seismic Stratigraphic and
thickness of the lithographic unit. 45 Structural analysis. Accordingly, it should now be
By way of example, FIG. 4a illustrates a familiar "pinch recognized, as was recognized by the present inventors, that
out Seismic model, wherein the Stratigraphic unit of interest there exists, and has existed for Some time, a very real need
(here with its thickness measured in travel time rather than for a method of Seismic data processing that would address
in length) gradually decreases in thickness until it disappears and solve the above-described problems.
(i.e., "pinches out”) at the left most end of the display. FIG. 50 Before proceeding to a description of the present
4b is a collection of mathematically generated Synthetic invention, however, it should be noted and remembered that
Seismograms computed from this model that illustrate the the description of the invention which follows, together with
noise free convolution of a Seismic wavelet with the inter the accompanying drawings, should not be construed as
faces that bound this layer. Notice that at the right most edge limiting the invention to the examples (or preferred
of FIG. 4b, the composite signal recorded on the first trace 55
embodiments) shown and described. This is so because
shows that the reflector is clearly delimited by a negative those skilled in the art to which the invention pertains will
reflection at the top of the unit and a positive reflection at its be able to devise other forms of this invention within the
base. Moving now to the left within FIG. 4b, the individual ambit of the appended claims. Finally, although the inven
reflections at the top and base begin to merge into a single tion disclosed herein may be illustrated by reference to
composite reflection and eventually disappear as the thick 60
various aspects of the convolutional model, the methods
neSS of the interval goes to Zero. Note, however, that the taught below do not rely on any particular model of the
composite reflection Still continues to change in character recorded Seismic trace and work equally well under broad
even after the event has degenerated into a single reflection. deviations from the Standard convolutional model.
Thus, even though there is little direct visual evidence that SUMMARY OF THE INVENTION
the reflection arises from two interfaces, the changes the 65
reflections exhibit as the thickness decreases Suggest that The present inventors have discovered a novel means of
there is information contained in these reflection that might, utilizing the discrete Fourier transform to image and map the
 5,870,691
7 8
extent of thin beds and other lateral rock discontinuities in is because long analysis windows encompass a great deal of
conventional 2-D and 3-D Seismic data. In more particular, geological variation, Said variations tending over the long
the invention disclosed herein is motivated by the observa term to exhibit a “white” (or random and uncorrelated)
tion that the reflection from a thin bed has a characteristic reflectivity function, which has a “flat” amplitude spectrum.
expression in the frequency domain that is indicative of the Thus, the shape of a frequency spectrum calculated from an
thickness of the bed: a homogeneous thin bed introduces a entire Seismic trace is largely dependent on the frequency
periodic Sequence of notches into the amplitude Spectrum of content of the Source wavelet. (See, for example, Chapter
the composite reflection, Said notches being Spaced a dis 2.2.1 of Seismic Data Processing by Ozdogan Yilmaz,
tance apart that is inversely proportional to the temporal Society of Exploration Geophysicists, 1987, the disclosure
thickness of the thin bed. Further, if the Fourier transform of which is incorporated herein by reference.) On the other
coefficients are properly displayed this characteristic expres hand, where the analysis window is So short that the earth
sion may be exploited by the interpreter to track thin bed reflectivity function is non-white, the resulting Fourier Spec
trum contains influences from both the wavelet and local
reflections through a 3-D volume and estimate their thick geology. Over Such Small windows, geology acts as a filter,
neSSes and extent to a degree not heretofore possible. More attenuating the spectrum of the Source wavelet and thereby
generally, the method disclosed herein may be applied to 15 creating non-stationary short-window spectra.
detect and identify vertical and lateral discontinuities in the The foregoing ideas are illustrated generally in FIG. 2,
local rock mass. Also, the usefulness of the present invention wherein a typical Seismic trace and Some frequency Spectra
is enhanced by a novel method of frequency domain whit calculated therefrom have been plotted. At the top of that
ening that emphasizes the geologic information present figure is the Fourier transform frequency spectrum of the
within the Spectrum. Finally, the instant invention is also entire Seismic trace. The appearance of that spectrum is
directed toward uncovering Seismic attributes that can be generally that of a typical field wavelet. However, the
correlated with SubSurface Structural and Stratigraphic fea Spectra calculated over Shorter windows, pictured at the
tures of interest, thereby providing quantitative values that bottom of FIG. 2, are nonstationary, tending to reflect the
can be mapped by the explorationist and used to predict underlying geology which may potentially change dramati
Subsurface hydrocarbon or other mineral accumulations. 25 cally over fairly short intervals.
By way of general background, the present invention The importance of this observation for the present inven
preferably utilizes a relatively short discrete Fourier trans tion is illustrated in FIG. 3, wherein two representative
form to determine the frequency components of a Seismic Spectra are generically pictured. The left frequency spectrum
trace. AS is known to those skilled in the art, calculation of (FIG. 3A) represents that of a typical broad bandwidth
a Fourier transform of a time Series, even one that is source wavelet. The right spectrum (FIG. 3B), however,
exclusively real valued, results in a collection of Fourier represents in a general way the frequency domain expression
transform coefficients that are complex data values of the of a composite reflection from a thin bed. In this later case,
form "A+Bi”, where "i” represents the “imaginary number the geology of the thin bed has tended to act as a frequency
or the Square root of a negative one. Further, it is well known domain filter and has produced its own mark on the fre
that the expression A+Bi may equivalently written as: 35 quency content of the reflected wavelet. AS is generally
illustrated in FIG.3B, the present inventors have discovered
A + Bi = re', that a homogeneous thin bed affects the amplitude spectrum
of the reflection event by introducing “notches”, or narrow
where, bands of attenuated frequencies, into it, thereby producing a
40 characteristic appearance. A homogeneous bed is one that
r = A + Bi = A2 + B2 , has a constant Velocity and density throughout its extent.
Further, the distance between the notches So introduced is
and equal to the inverse of the “temporal thickness” of the thin
bed layer, temporal thickness being the length of time that it
0-in (f). A.
45 takes for a wavelet to traverse the layer in one direction (the
thickness of the layer divided by the its velocity). Thus,
attenuated frequencies in the amplitude Spectrum may be
The quantity 0 is known as the phase angle (or just the used to identify a thin bed reflection and to gauge its
“phase”) of the complex quantity A+Bi, the quantity “r” its thickness.
magnitude, and the expression A+Bi is the mathematical 50 Turning now to FIGS. 4a–4c, the results suggested in the
notation for the magnitude of a complex valued quantity, previous paragraph are extended to the analysis of a Sim
also called its absolute value. A frequency spectrum is plistic 2-D geological model, wherein the frequency domain
obtained from the Fourier transform coefficients by calcu expression of a thin bed is investigated. In FIG. 4a, a typical
lating the complex magnitude of each transform coefficient. "pinch out” reflectivity function (geological model) is pre
Further, the numerical size of each coefficient in the fre 55 Sented. FIG. 4c contains a grey-Scale display of the Fourier
quency Spectrum is proportional to the Strength of that transform frequency spectrum amplitudes calculated from
frequency in the original data. Finally, after application of a this model. This display was produced by creating a discrete
Fourier transform to Some particular time Series, the result time Series at fifty equally spaced positions acroSS the model,
ing Series of complex coefficients are Said to be in the each of which has only two non-Zero values: one corre
frequency domain, whereas the untransformed data are 60 sponding to the reflection coefficient at the top of the layer,
referred to as being in the time domain. and the other to the reflection coefficient at its base. Each of
Returning now to a discussion of the instant invention, the the non-zero values is placed within the otherwise Zero-filled
invention disclosed herein relies on the general observation trace in positions that reflect the location in time of the top
that a frequency spectrum calculated from a whole trace and bottom of the reflector, respectively. A Standard discrete
Fourier transform tends to resemble the spectrum of the 65 Fourier transform was then calculated for the time Series,
Source wavelet, whereas Shorter window spectra tend to followed by calculation of the complex magnitude of each
reflect more of the underlying geological information. This coefficient.
 5,870,691
9 10
In FIG. 4c, the lighter portions correspond to larger values The Selection and extraction of the data corresponding to
of the amplitude spectra, whereas the darker portions rep the Zone of interest is known as “subsetting” the data (FIG.
resent Small values. Thus, “notches' in the amplitude Spectra 5). One criterion that guides the selection of the Zone of
are represented by the darker values in the plot. This figure interest is the desire to keep the Zone as short (in time) as
displays, in a most literal Sense, the Fourier transform of the possible. This is in keeping with the general philosophy
geology and, more particularly, the characteristic Signature espoused above regarding the tendency of long-window
that is impressed on the wavelet by this event. What is most Fourier transform spectra to resemble the wavelet and
Significant about this plot relative to the instant invention is Short-window Fourier transform Spectra to contain more
that, as the thickness of the model decreases, the Spacing geologically related information. Note that there is a “hid
between the notches increases. Further, for a given model den' window enlargement that is often applied automati
thickness the notches are periodic, with a period equal to the cally and unthinkingly to Fourier transform windows: exten
reciprocal of the temporal thickness of the layer. Thus, if this Sion of the window Size to a length that is a power of two.
Signature-periodic frequency domain notches-can be This lengthening of the window is done for purposes of
located within a Seismic Survey, it is strong evidence that a computational efficiency, as window lengths that are powers
thin bed is present. 15
of two are candidates for the Fast Fourier Transform (FFT)
According to one aspect of the present invention there has algorithm. However, the present inventorS Specifically coun
been provided a System for interpreting Seismic data con Sel against this industry-wide practice and prefer to use a
taining thin bed events, wherein the data are decomposed more general, if leSS computationally efficient, discrete
into a series of Fourier transform 2-D lines or 3-D volumes, Fourier transform algorithm, thereby keeping the length of
thereby providing enhanced imaging of the extent of Said the analysis window to its minimum possible value. Given
thin bed layers. The instant embodiment utilizes a Single the computational power of computers today, there is little
Fourier transform window which is Separately applied to the reason not to transform only the data within the Zone of
portion of each Seismic trace that intersects a Zone of interest.
interest. This embodiment is illustrated generally in FIG. 5 In FIG. 5, the “COMPUTE" step, as applied to the present
as applied to 3-D Seismic data, but those skilled in the art embodiment, consists of at least one operation: calculating
will realize that the Same method could also be practiced to 25 a discrete Fourier transform over the Zone of interest. The
advantage on a 2-D collection of Seismic traces to enhance resulting coefficients, the spectral decomposition of the Zone
the visibility of thin bed reflections contained therein. of interest, are then Stored as part of the output spectral
AS a first, Step a collection of Spatially related Seismic decomposition volume ("tuning cube') for Subsequent view
traces are assembled. These traces might be, for purposes of ing. Note that there will be one trace (i.e., collection of
illustration only, one or more shot records, a constant offset Fourier transform coefficients) in the output tuning cube
gather, a CMP gather, a VSP survey, a two-dimensional Volume for each Seismic trace processed as part of the input.
Seismic line, a two-dimensional Stacked Seismic line Also note that in this presently preferred output
extracted from a 3-D seismic Survey or, preferably, a 3-D arrangement, horizontal plane Slices through the Volume
portion of a 3-D Seismic Survey. Furthermore, the present contain coefficients corresponding to a Single common Fou
invention might also be applied to a 2-D or 3-D survey 35 rier frequency.
wherein the data have been transposed, i.e., where an Optionally, the “COMPUTE" step may contain additional
“offset” or spatial axis (“X” or “Y” axis for 3-D data) has operations which have the potential to enhance the quality of
been oriented So as to replace the vertical or “time' axis. the output Volume and Subsequent analysis. First, a weight
More generally, any 3-D Volume of digital data may be function may be applied to the Seismic data within the Zone
processed by the methods disclosed herein. That being Said, 40 of interest prior to calculating the transform. The purpose of
for purposes of clarity the vertical axis will be referred to as the weight function is to taper or Smooth the data within the
the “time” axis hereinafter, although those skilled in the art Fourier analysis window, thereby lessening the frequency
understand that the digital Samples might not be separated domain distortions that can arise with a "box-car type
by units of time. Whatever the choice, the invention dis analysis window. The use of a weight function prior to
closed herein is most effective when applied to a group of 45 transformation is well known to those skilled in the art. The
Seismic traces that have an underlying spatial relationship preferred weighing function for the invention disclosed
with respect to Some SubSurface geological feature. Again herein is Gaussian in shape and is in many ways optimal for
for purposes of illustration only, the discussion that follows this application. That being Said, note that many other
will be couched in terms of traces contained within a Stacked weight functions could also potentially be used.
3-D Survey, although any assembled group of Spatially 50 Additionally, Since it is usually the amplitude Spectrum
related Seismic traces could conceivably be used. that is of greatest interest to the explorationist, the amplitude
As is illustrated generally in FIG. 5, a Zone of interest is Spectrum may be calculated from the transform coefficients
next selected within a particular 3-D volume. The Zone of as they are moved into an auxiliary Storage area.
interest might be, by way of example, the undulating region Alternatively, a phase spectrum, or Some other derived
bounded by two picked reflectors as is pictured in FIG. 5. In 55 attribute, can be calculated from the transform coefficients
that case, the reflector is preferentially flattened or before Storage and, indeed, these Sorts of calculations have
datumized (i.e., made flat by shifting individual traces up or been made by the present inventors.
down in time) before analysis, and possibly also palinspas Finally, as part of the computation Step, an individual
tically reconstructed. More conventionally, a Specific frequency Scaling may be applied to each plane (i.e.,
bounded time interval (for example, from 2200 ms to 2400 60 frequency) in the output volume. As illustrated generally in
ms) might be specified, thereby defining a “cube' or, more FIG. 10, the inventors have found it preferable to separately
accurately, a “box” of seismic data within the 3-D volume: Scale each frequency Slice in the output volume to have the
a sub-volume. Additionally, the lateral extent of the Zone of Same average value before viewing it. This Scaling is just
interest may be limited by Specifying bounding “in-line' and one of many that might be applied, but the inventors prefer
“cross-line” trace limits. Other methods of specifying the 65 this method because it tends to enhance the geological
Zone of interest are certainly possible and have been con content of the Stored frequency spectra at the expense of the
templated by the inventors. common wavelet information.
 5,870,691
11 12
Once the Spectra have been calculated and Stored, they are mapping may result in a Seismic data cube or rectangular
ready to be used in the geophysical exploration for thin beds. piece of an individual Seismic line.
Note that it is critical that, when the data are Subsequently In the present embodiment, rather than applying a single
displayed, each Spectrum be organized and viewed in the window Fourier transform to each trace, instead a Series of
Same Spatial relationship with the other spectra as the traces overlapping short window Fourier transforms are utilized.
from which they were calculated. That is, Spatial relation The length of the window and the amount of overlap varies
ships that are present in the untransformed data must be with the particular application, but once again the window
preserved in the arrangements of the transform coefficients. length need not be equal to a power of “2” but rather should
The presently preferred method of viewing the transform be chosen So as to best image the underlying geology. Note
coefficients is to begin by forming them into a 3-D “volume” that a weight might optionally be applied to the data within
(tuning cube), provided, of course, that the input data were each short window before transformation and, as before, a
originally taken from a 3-D volume. Note, however, that the Gaussian weight is the preferred choice.
vertical (“Z”) axis is no longer “time” as it was before the AS indicated in FIG. 6, as each short-window Fourier
transformation, but rather now represents, by convention, transform is calculated, the coefficients resulting therefrom
units of frequency, as Fourier transform coefficients are 15
are Separately Stored within an individual tuning cube that
Stored therein.
The tuning cube, as illustrated in the last step in FIG. 5, remains associated with the short-window that gave rise to
may now be viewed in any manner that could be used to it. Note that in the instant case there will be as many tuning
view a conventional 3-D volume of seismic data. That being cubes produced as there were overlapping windows in the
Said, the present inventors have discovered that viewing analysis. Scaling, if it is applied, is applied Separately to
Successive horizontal slices through the Volume of coeffi each frequency plane in each tuning cube.
cients is a preferred way to locate and Visualize thin bed Each short-window tuning cube produced by a sliding
effects. Note that in the tuning cube arrangement, a hori window may now be individually examined in exactly the
Zontal Slice represents all of the coefficients corresponding Same fashion as that proposed previously for first embodi
to a single Fourier frequency and, thus is a constant fre 25
ment. Once again, each cube is preferably viewed in hori
quency croSS Section. Further, as an aid in the analysis of the Zontal Slices or constant frequency images, thereby provid
data contained within this Volume, the inventors preferably ing a means for visualizing geological changes with
animate a Series of horizontal views through the Volume. In frequency. Further, Since there is now a collection of tuning
the event that the Zone of interest is a portion of an individual cubes calculated at different time points in the trace, in effect
Seismic line rather than a Volume, the resulting display, a collection of tuning cubes have been produced that span a
being a collection of Spatially related Seismic trace Fourier range of depths in the SubSurface.
transform Spectra displayed in their original Spatial Finally, according to a third aspect of the present inven
relationship, will still be referred to as a tuning cube herein, tion there has been provided a System for processing Seismic
even though, technically, it may not be a “cube' of data. data to enhance the appearance on the Seismic record of thin
Animating Successive horizontal slices through the Spec 35 bed events, wherein the data are decomposed into a Series of
tral Volume is the preferred method of Viewing and analyZ Fourier transform 2-D lines or 3-D volumes by using a
ing the transform coefficients, Said animation preferably Short-window Fourier transform and are then reorganized
taking place on the computer monitor of a high Speed into Single frequency tuning cubes, thereby providing
WorkStation. AS is well known to those skilled in the art, enhanced imaging of thin bed layers.
animation in the form of interactive panning through the 40 As is generally illustrated in FIG. 7, the first steps in this
Volume is a fast and efficient way to view large amounts of present embodiment mirror those of the previous two
data. The data volume might be viewed in horizontal, embodiments: the data are first interpreted then Subsetted.
Vertical, or oblique Slices, each of which provides a unique Thereafter, a Series of Overlapping short window Fourier
view of the data. More importantly, however, in the context transforms are calculated from the Seismic data within the
of the present invention rapidly viewing Successive horizon 45 Zone of interest, optionally preceded by the application of a
tal Slices in Succession provides a diagnostic means to weight or taper within each window before calculating the
Survey a large Volume of data and identify the thin bed transform. AS in the previous embodiment, the coefficients
reflections therein, the details of which will be discussed from each short window transform are accumulated. In the
below. Note that it is preferable for the method disclosed instant case however, rather than Viewing the calculated
herein that the Slices be ordered in terms of frequency (either 50 Fourier transform coefficients as tuning cubes, the data are
Strictly increasing or decreasing) when they are animated reorganized into Single frequency energy cubes which can
and viewed. thereafter be examined either in a horizontal or vertical
According to a Second aspect of the present invention, plane for evidence of thin bed effects.
there has been provided a System for processing Seismic data In more particular, the reorganization contemplated by the
to enhance the appearance on the Seismic record of thin bed 55 present inventors conceptually involves extracting from all
events, wherein the data are decomposed into a Series of of the tuning cubes every horizontal slice that corresponds to
Fourier transform 2-D lines or 3-D volumes by using a series a particular frequency. Then, these individual Same
of overlapping short-window Fourier transforms, thereby frequency Slices are "stacked” together, the topmost Slice
providing enhanced imaging of thin bed layers. This containing coefficients calculated from the topmost sliding
embodiment is illustrated generally in FIG. 6 as applied to 60 window, the next Slice containing coefficients calculated
3-D seismic data, but those skilled in the art will realize that from the first sliding window below the top, etc. Note that,
the same method could also be practiced to advantage on a after reorganization, the Volume of coefficients is organized
2-D seismic line to enhance the visibility of thin bed into units of “x-y' and time. This is because the vertical axis
reflections contained therein. AS indicated in FIG. 6, and as is ordered by the “time” of the sliding window that gave rise
disclosed previously, the first Step in the present embodiment 65 to a particular coefficient.
involves the interpreter mapping the temporal bounds of the To utilize the information with the Single frequency
Seismic Zone of interest. AS was described previously, the tuning cubes constructed by the previous Step, a Seismic
 5,870,691
13 14
interpreter would Select a frequency and the Seismic Volume
corresponding thereto (e.g., he or she might Select the L-1
coefficient Volume corresponding to 10hz, and/or the Vol Aina"(n)- (-k)
ume for 11 hz, etc.). Each constant-frequency cube may be
Viewed in plan or horizontal view, or in any other manner,
thereby providing a means for Visualizing geological
changes with lateral extent for a particular frequency. where

It is important to note that, in all of the above described 0, n z in

embodiments, the fact that the original untransformed traces 1O 6(n - m) = s
were spatially related provides additional utility to the 1, n = n
invention disclosed herein. In more particular, it is well
known that short-window Fourier transform coefficients are and A (k,n) represents the complex conjugate of A(k,n). For
inherently quite noisy and have poorer frequency resolution the discrete Fourier transform, the basis functions corre
in comparison with a longer window transform. One 15 sponding to a forward transform of length L are conven
approach that the present inventors have used to improve the tionally chosen to be the Set of complex exponentials:
reliability of the transformed values is to apply a Gaussian
weight function to the pre-transformed data values.
However, another equally important Step taken by the There are thus L basis functions (or basis vectors in this
present inventors is to display the coefficients within a case), one basis function for each value of “n”:
Volume in the Same Spatial relationship as that of the input L
data. Since the traces So displayed contain Spatially corre i =--
2 ,..., 0,... --
2 - 1.
lated information, displaying them next to each other allows
the observe to visually “smooth out the noise and perceive 25 To Summarize: each transform coefficient, X(n), calculated
the underlying coherent signal information. from a data window corresponds to a particular basis
Finally, although the present invention is discussed herein function, and a tuning Volume is formed by collecting all of
in terms of the discrete Fourier transform, in reality the the transform coefficients corresponding to a particular Zone
Fourier transform is just one of any number of discrete time of interest and Storing those coefficients in an auxiliary
data transformations that could used in exactly the same Storage area in the same Spatial arrangement as the traces
fashion. The general Steps of (1) computing a short window from which each window was computed.
transformation (2) associating the resulting transform coef By way of another specific example, those skilled in the
ficients into a volume, and (3) examining the volume for art understand that a discrete Walsh transform could be used
evidence of thin bed effects, could be accomplished with a in place of the Fourier transform and the Walsh coefficients
wide variety of discrete data transformations other than the 35 Similarly grouped, displayed, and analyzed. In the manner
Fourier. If the transformation is other than a Fourier, the disclosed above, a Walsh transform may be computed within
tuning Volumes would be formed by grouping together an overlapping Series of sliding windows and the coefficients
coefficients corresponding to the Same basis function. resulting therefrom organized and Stored into tuning cubes.
Those skilled in the art will understand that a discrete Rather than the calculated transform coefficients represent
Fourier transform is just one of many discrete linear unitary
40 ing frequency, of course, these coefficients instead represent
transformations that satisfy the following properties: (1) a similar quantity called “Sequency' by those skilled in the
they are linear operators that are (2) exactly invertible, and art. Thus, “single Sequency' tuning cubes may be formed
(3) their basis functions form an orthonormal set. In terms of from the Walsh transform coefficients in a manner exactly
equations, if X(k), k=1, L, represents a time Series, and X(n) analogous to that used in the construction of Fourier tuning
45 cubes.
its “nth' transformed value, n=1, L, then the forward trans Finally, although the discrete Fourier transform is a trans
form of the time series may be generally written for this form that is characterized by a set of orthonormal basis
class of transformations as:
functions, application of a non-trivial weight function to the
basis functions prior to computation of a transformation
50 destroys their orthonormality. Under conventional theory, a
weight function that is applied within a window is viewed as
being applied to the basis functions rather than the data,
where A(k,n) represents the forward transform kernel or thereby preserving the integrity of the underlying data.
collection of basis functions. Further, there is an inverse However, basis functions that were orthogonal before appli
transform which maps the transformed values back into the 55 cation of the weight function will generally no longer be So
original data values: thereafter. That being said, in point of fact whether the
weight function is applied to the data or to the basis
functions, the end computational result after transformation
is exactly the same.
60 One means of avoiding the minor theoretical dilemma that
arises when a weight function is used with a discrete
where B(k,n) is inverse transform kernel. The requirement orthonormal transform is to Select an orthonormal transform/
of orthonormality implies that the inner products between weight combination which is not so affected. By way of
two different basis functions must be equal to Zero, and the example, the local cosine (and local Sine) transform is a
magnitude of each basis function must be equal to unity. 65 discrete orthonormal transform wherein the weight function
This requirement may be Succinctly Summarized by the of choice is a Smooth, Specially designed taper that preserves
following equations: the orthonormality of the basis functions at the expense of
 5,870,691
15 16
Some loSS in frequency resolution. Further, the underlying increasing depth, AZ. However, the instant invention can and
rationale of the local cosine?sine transform provides a natu has been applied in exactly the same fashion to this Sort of
ral theoretical bridge to the field of general wavelet trans data. Thus, in the text that follows At (and “time') will be
forms. used in the broader Sense of referring to the Separation
between Successive digital Samples, whatever measurement
BRIEF DESCRIPTION OF THE DRAWINGS form that Separation might take.
AS a first Step, the explorationist or Seismic interpreter
FIGS. 1a and 1b is a schematic diagram that illustrates selects a Zone of interest within the 3-D volume. This might
generally the nature of the thin bed problem. be done, by way of example only, by digitizing time pickS
FIG. 2 displays a typical Seismic trace and compares long (“picking”) seismic events either on a digitizing table or,
and short window spectra computed therefrom. more commonly, at a Seismic workStation. When an event is
FIG. 3A illustrates the spectrum of a typical seismic picked, the explorationist attempts to pinpoint the same
Source wavelet and FIG. 3B illustrates the spectrum of that reflector feature (e.g., peak, trough, Zero crossing, etc.) on
same wavelet after reflection by a thin bed. every Seismic trace in which it appears, the ultimate goal
FIGS. 4a–4c contains a simple Seismic pinch out model, 15 being the production of a computer file that contains time
and Surface location information that tracks the event acroSS
the convolutional response thereto, and the frequency
domain representation of Said convolutional response. a 2-D section or through a 3-D volume. As illustrated in FIG.
11, given this information a computer program can be
FIG. 5 is a Schematic diagram that illustrates the general designed to read the picks and find the Zone of interest for
approach of a presently preferred embodiment. any trace within the data Volume, and/or perform the method
FIG. 6 contains a schematic illustration of how a presently of the present invention. Said program might be transported
preferred embodiment of the present invention is used in an into the computer, for example, by magnetic disk, tape,
exploration Setting. optical disk or CD-ROM.
FIG. 7 is a schematic illustration of another presently Alternatively, the interpreter might simply Specify con
preferred embodiment. 25 Stant Starting and ending times which bound the event of
FIG. 8 is a flow chart that illustrates a presently preferred interest throughout the entire Volume, thereby creating a
embodiment. “cube' of interest, where “cube' is used in the generic sense
FIGS. 9a–9a is a schematic illustration that describes the to represent a 3-D sub-volume of the original 3-D survey
appearance of a thin bed during animation of constant Volume. For purposes of illustration only, the discussion that
frequency Slices. follows will assume that a 3-D Sub-cube has been extracted,
FIG. 10 illustrates the general approach utilized to scale although those skilled in the art will recognize that the same
the constant frequency Slices So as to enhance the geologic techniques discussed below can easily be adapted to a
content of the transformed data. window that is not constant in time. Again, just for purposes
of illustrating the disclosed techniques, the temporal Zone of
FIG. 11 is a schematic illustration of another presently 35 interest, after extraction, will be assumed to extend from the
preferred embodiment. first Sample of the 3-D Sub-volume, to last Sample, Sample
DETAILED DESCRIPTION OF THE number “N' hereinafter. Similarly, it will also be assumed
PREFERRED EMBODIMENTS hereinafter that the Zone of interest is present on every trace
in the sub-volume, although those skilled in the art will
The instant invention provides a method of processing 40 recognize that it is often the case that the Zone of interest
Seismic data using a discrete Fourier transform, whereby its extends to only a portion of the 3-D volume.
utility as a detector of thin beds is enhanced. Given a Zone of interest, the next Step is to Select a Fourier
According to a first aspect of the present invention, there transform window length, “L” hereinafter. Generally
has been provided a method of enhancing and viewing thin Speaking, the length of the transform should be no longer
bed effects using a discrete Fourier transform wherein a 45 than is absolutely necessary to encompass the Zone of
Single Fourier transform is calculated for a window Spanning interest. Conventionally, the length of the Fourier transform
the Zone of interest and the coefficients obtained therefrom is Selected for purposes of computational efficiency and is
are thereafter displayed in a novel manner. AS is illustrated usually restricted to an integer power of 2 (e.g., 32, 64, 128,
generally in FIG. 5, let X(k,nt) represent a 3-D seismic data etc.), thereby allowing the highly efficient FFT calculation
Volume, where k=1, K, and j=1, J, represent indices that 50 algorithm to be utilized, rather than a Somewhat less efficient
identify a specific trace within a given 3-D volume. By way mixed radix Fourier transform or a much less efficient
of example only, these indices might be in-line and croSS general discrete Fourier transform. However, within the
line position numbers, although other location Schemes are context of the present invention the inventorS Specifically
also possible. The variable “nt” will be used to represent the recommend against extending the chosen window length, as
time (or depth) position of within each Seismic trace, nt=0, 55 is conventionally done, to an integer power of two: a general
NTOT-1, the total number of samples in each individual discrete Fourier transform should be used instead. That
trace. The time Separation between Successive values of being Said, in the discussion that follows, it is understood by
X(k,nt) (i.e., the sample rate) will be denoted by At, where those skilled in the art that whenever a discrete Fourier
At is customarily measured in milliseconds. Each trace in the transform is called for, an FFT will be calculated if appro
3-D volume, therefore, contains a recordation of (NTOT)* At 60 priate. Otherwise, a general discrete Fourier transform, or
milliseconds of data, the first Sample conventionally taken to Some mixed radix variant, will be Selected if the chosen
occur at “Zero' time. That being said, those skilled in the art window length is not an integer power of 2.
understand that Some Seismic data that are eminently Suit Before beginning the Fourier transformations, an auxil
able for analysis by the invention disclosed herein are not iary Storage Volume must be established in which to Store the
ordered in terms of “time”. By way of example only, seismic 65 calculated Fourier coefficients. Auxiliary Storage least as
data Samples that have been processed by a depth migration large as L computer words in extent must be provided for
program are Stored within a Seismic trace in order of each trace in which to Save the calculated transform
 5,870,691
17 18
coefficients, with even more Storage being required if the
Seismic data valueS or the transformed results are to be kept L
f, = - L(At/1000)- , n ---, ... 3 0.vs. . . . 12 - 1.
as double (or higher) precision. By way of explanation, a
Fourier transform of a real time Series of length L requires
Storage for L/2 complex data values, each of which normally It should be noted in passing that the nominal Sample rate
requires two computer words of Storage. (There are actually of a Seismic trace, At, may not be the same Sample rate at
only (L/2)-1 unique complex data values, rather than L, which the data were acquired in the field. For example, it is
because for a real time Series the Fourier transform coeffi common practice to reSample a Seismic trace to a larger
cients corresponding to positive and negative frequencies Sample rate to Save Storage when there is little useful
are directly related: they are complex conjugate pairs. In 1O information at the highest recorded frequencies. On the other
addition, there are two real values: the coefficient at Zero hand, on occasion a Seismic trace may be resampled to a
(“dc') hertz and the coefficient at the Nyquist frequency, Smaller Sampling rate when, for example, it is to be com
both of which could be Stored in a Single complex data value. bined with other-higher Sample rate-lines. In either case,
Finally, if L is an odd integer, the number of unique data the nominal Sample rate of the data may not accurately
values is (L+1)/2). If there are a total of J times K seismic 15 reflect its true spectral bandwidth. A simple modification of
traces within the Zone (cube) of interest, the total amount of the previous equation will accommodate that contingency:
auxiliary Storage required will be equal to, at minimum, the
product of L, J, and K measured in computer words. Let, the f Fray, n = - - - , . . . ,0,... 2--1,
array A(k,nt) represent an auxiliary storage area for the L
present embodiment.
AS a first computational Step, and as illustrated in FIG. 8, where F is the highest frequency contained in the data.
the data values within the Zone of interest are extracted from Since a seismic trace is a “real' function (i.e., non
an input trace XO.k,nt) taken from the Sub-volume: imaginary), its Fourier transform is symmetric and the
Fourier coefficients corresponding to the positive and nega
y(nl)=x(j, k, nil), ni=0, L-1
25 tive frequencies are related as follows:
REX(f)=REX(f),
and the weight function is optionally applied:
and
y(nl)=y(nl)w(nl), ni=0, L-1,
IMX(f)=-IMX(f),
where the array y(nl) is a temporary storage area. (Note that where REZ is a function that extracts the real portion of the
in this present embodiment, the length of the analysis complex value Z, and IMZ extracts the imaginary portion.
window is equal to the length of the Zone of interest.) The AS a consequence of this relationship, only L/2+1 unique
weight function w(t), or data window as it is referred to by 35 values are produced within each Fourier transform window.
Some, could take any number of forms. Some of the more Thus, for purposes of Specificity, only the positive frequen
popular data windows are the Hamming, Hanning, Parzen, cies will be considered in the discussion that follows,
Bartlett, and Blackman windows. Each window function has although those skilled in the art understand that the same
certain advantages and disadvantages. The present results could have been obtained by utilizing only the
inventors, however, have discovered that the use of a GauS 40 negative frequencies.
sian window is in many ways optimal for this application. A neXt Step in the process involves placing the calculated
The Gaussian weight function is defined by the following complex frequency values into the auxiliary Storage array.
expression:
These traces are filled with the calculated complex Fourier
coefficients as indicated below:
45

where,
wherein, “” and “k” match the indices corresponding to the
1. original data trace. In practice, the array AG,k,i) may not
50 ever actually be kept entirely in RAM (random access
memory) at one time, but may be located, in whole or in part,
In general, though, the weight function should be a real on tape, disk, optical disk, or other Storage means.
function and non-Zero over its range. Additionally, because the presently preferred thin bed dis
play requires the use of the frequency Spectrum rather than
After the weight function has been applied, the discrete 55 the complex values, it would be convenient at the same time
Fourier transform is then calculated according to the fol to calculate the complex magnitude as each coefficient is
lowing Standard expression: placed into the auxiliary Storage array:
L-1
X(n) = k=0
Sy(k)e-tint, n = -1,...,
2
0,... f2 -1,
60 However, there are many circumstances in which the
complex coefficients would be needed and useful, So, as
where X(n) represents the complex Fourier transform coef indicated in FIG. 8, the complex coefficients are preferen
ficient at the frequency, f, said frequency being dependent tially Stored in the auxiliary Storage area.
on the length of the window L. In general, it is well known The procedure described above is repeated for every trace
that the Fourier transform produces coefficients that provide 65 in the defined sub-volume, thereby filling the auxiliary
estimates of the Spectral amplitude at the following Fourier Storage array with transform coefficients in preparation for
frequencies: Viewing by the explorationist. Before Viewing the results,
 5,870,691
19 20
however, the data are preferentially Scaled in a novel frequency) amplitude slices, SG,k), corresponding to the
fashion, whereby the geological information within the “ith frequency may be extracted from Ao,k,i) for viewing
transform coefficients is emphasized relative to the contri and/or animation:
bution of the wavelet. This general method involved in this
frequency domain scaling is illustrated in FIG. 10. The
Scaling method disclosed herein is designed to equalize the When these slices are animated (i.e., viewed rapidly in
average spectral amplitude in each frequency Slice, thereby Succession) thin beds will be recognizable as those events
tending to produce a whitened wavelet spectrum. AS illus that Successively alternate between high and low amplitude
trated in some detail in FIG. 8, Let T(jk,i) represent a values. Further, for many sorts of thin beds, there will be a
temporary Storage array into which an entire tuning cube characteristic pattern of moving notches that clearly signal
will be Stored. For a given frequency Slice, i, calculate the that an event is generated by a thin bed. Note that it is
average spectral amplitude therein:
preferable for the method disclosed herein that the slices be
ordered in terms of frequency (either strictly increasing or
TAVG =
-- 15
decreasing) when they are animated and viewed.
FIGS. 9a–9a illustrates the source of this diagnostic
moving pattern. FIG. 9a contains a lens-type geologic thin
The Spectral magnitude has been calculated because the bed model and FIG.9b a stylized Fourier transform of said
T(j,k,i) are potentially complex valued. As a next step, the model, wherein only the notches have been drawn. AS
values in this particular frequency Slice are adjusted So that discussed previously, the notches are periodic with period
their average is equal to Some user Specified constant value, equal to the inverse of the temporal thickness of the model
represented by the variable AVG: at that point. Now, consider the model in FIG. 9a as
AWG
representing a 2-D cross Section of a 3-D (disk-type) radially
Tiki) -- symmetric model, and FIG.9b as a similarly radially sym
metric collection of one dimensional Fourier transforms of
25 said 3-D model. If the constant frequency plane labeled
where the primed notation has been used to indicate that the Plane 1 is passed through the Volume as indicated, the plan
T(jki) array has been modified. In practice, AVG will be set View display of Said plane will reveal a low amplitude
to Some specific numerical value, 100, for example. This circular region corresponding to the first notch. Plane 2
Scaling operation is repeated Separately for every frequency passes through two notches and exhibits two low amplitude
Slice (i=0, L/2) in the tuning cube volume. At the conclusion circular regions. Finally, Plane 3 contains three low ampli
of this operation, every Slice has the same average amplitude tude circular regions, corresponding to the three notches that
and a kind of Spectral balancing has been performed. Note it intersects. Now, when these Slices are viewed in rapid
that this form of Single-frequency Scaling is just one Scaling Succession in order of increasing frequency, there is a Visual
algorithm that could be applied to the tuning cube data and impression of a growing “bulls eye' pattern wherein the
the instant inventors have contemplated that other methods 35 rings move outward from the center. This pattern of moving
might also be used to advantage. By way of example, rather notches is diagnostic for thin beds.
than computing the arithmetic average of the items in the a When the thin bed is not circular, a related pattern is
Slice, another measure of central tendency or any other observed. Rather than concentric circles though, there will
Statistic could have been equalized instead (e.g., median, appear a Series of moving notches that progreSS from away
mode, geometric mean, variance, etc.). AS another example, 40 from the thicker areas and toward the thinner ones. For
rather Setting the average value in each frequency Slice equal example, consider the model of FIGS. 9a–9a as a cross
to the same constant, each Slice could be set equal to a Section of a lens-shaped Stream channel. When Viewed in
different constant average value, thereby enhancing Some Successive plan View frequency Slices, a pattern of outward
frequencies in the Spectrum and Suppressing others. moving notches-moving from the center of the channel
If the Scaled tuning cube data are now inverted back into 45 toward its periphery-will be observed all along its length.
the time domain using a Standard Fourier transform inverse, It should be noted that if the thin bed is not homogeneous,
a spectrally balanced version of the original input Seismic for example if it contains a gradational Velocity increase or
traces are thereby obtained. Let X(k) represent a Scaled decrease, it may not exhibit the characteristic "notch' pat
collection of transform coefficients obtained by the previ tern of the homogeneous thin bed, but rather have Some
ously disclosed process and taken from location (j,k) within 50 different frequency domain expression. In these cases, the
the Scaled tuning cube array. preferred method of identifying the characteristic response is
Then, a spectrally whitened version of the input data may to create a model of the event and calculate its Fourier
be obtained by means of the following equation: transform, as was illustrated previously in FIG. 4b. Armed
with this information, an explorationist may then examine
55 an animated tuning cube for instances of the predicted
response.
Not only is the pattern of notches a qualitative indication
where x(k,nl) represents the now modified (spectrally of a homogeneous thin bed, but it is also yields a quantitative
balanced) version of the input data X(k,nl). The divisor measure of the extent of the thin bed. Returning to FIGS.
w(nl) is there to remove the effects of the weight function 60 9a–9a, note that notches are limited in lateral extent by the
that was applied prior to transformation. That term may be outer most edges of the model. Thus, by panning through a
omitted if no weight was applied in the forward transform Stack of frequency Slices and noting the Outermost limits of
direction. movement by the notches, a quantitative estimate of the
However, rather than inverting the Scaled tuning cube, the extent of the bed may be obtained.
presently preferred use for it is as an exploration tool for 65 The foregoing is a Striking visual effect that can be readily
detecting thin beds. After all of the traces have been pro observed in actual Seismic data Volumes. Since the typical
cessed and placed in auxiliary storage, horizontal (constant non-thin bed event will have a Somewhat consistent and
 5,870,691
21 22
Slowly changing amplitude spectrum, the thin bed response produced from the tuning cube calculations represent local
is distinctive and easily identified. Note that in the present ized analyses of reflector properties (being calculated, as
embodiment where a Single window is calculated for the they are, from a short window) and, as Such, are potentially
entire Zone of interest, the actual time position (i.e., depth) of considerable importance to the advancement of the inter
of the thin bed within the Zone of interest is not particularly pretation arts.
important. If the thin bed is located anywhere within the According to a Second aspect of the present invention,
temporal Zone of interest, the Spectrum for that window will there has been provided a method of enhancing thin bed
exhibit the characteristic moving notch pattern. Those effects using a discrete Fourier transform wherein a Series of
skilled in the art will understand that moving the location of Sliding short-window Fourier transforms are calculated over
an event in time does not change its amplitude Spectrum. a window Spanning the Zone of interest and thereafter
Rather, it only introduces a change in the phase which will displayed in a novel manner. This method is illustrated
not be apparent if the amplitude spectrum is calculated and generally in FIG. 6 and in more detail in FIG. 8.
viewed. Conceptually, the present embodiment may be thought of as
AS an alternative to displaying the amplitude Spectra in producing a Series of tuning cubes of the Sort disclosed
animated plan view, the present embodiment may also be 15 previously, one tuning cube for each Fourier transform
used with any number of other attributes calculated from the window position Specified by the user.
complex values Stored in the tuning cube. By way of Once again X(k,n) represents a 3-D seismic data volume
example, the phase of the complex transform coefficients and “L” the length of the chosen sliding-window Fourier
provides another means of identifying thin bed events and, transform. In this present embodiment “L” will generally be
more generally, lateral discontinuities in the rock mass. The Substantially shorter than the length of the Zone of interest,
phase tuning cube is calculated as follows: N. As before, the length of the Fourier transform window is
to be Selected, not on the basis of computational efficiency,
P(iik, i) = tan-1 ( IMA(i.ki) ) but rather with the intent of imaging particular classes of thin
bed events in the SubSurface. By way of example, a reason
25 able Starting point for the transform length is one that is just
where, P(j,k,i) contains the phase portion of the complex long enough to span the “thickest' thin bed within the Zone
Fourier transform coefficients for every point in the original of interest. Note that it may be necessary to increase this
tuning cube. Phase Sections have long been used by those minimum length in circumstances where, for instance, the
skilled in the art to assist in picking indistinct reflectors, a waveform is not particularly compact. In this later case, the
phase Section tending to emphasize continuities in the Seis minimum window length might be increased by as much the
mic data. In the present embodiment however, it is lateral length of the wavelet measured in Samples.
discontinuities in the Spectral phase response that are indica The integer variable, NS, will be used to represent the
tive of lateral variability in the local rock mass, of which increment in Samples that is applied to Successive windows.
truncation of thin beds is a prime example. When viewed in For example, if NS is equal to 1, a short window Fourier
animated plan View, the phase values in the vicinity of a 35 transform will be calculated at every possible starting posi
lateral edge will tend to be relatively"unstable”: tending to tion within the Zone of interest, with Successive sliding
have an ill-behaved first derivative. Thus, the edges of thin windows differing by only a single sample. If NS is equal to
beds and, more generally, lateral discontinuities in the rock 2, Successive windows will share all but two of the same data
mass (e.g., faults, fractures, non-conformities, values and transforms will be calculated at every other Star
unconformities, etc.) will tend to have a phase that contrasts 40 The Fourin within the Zone of interest.
with Surrounding phase values and will be, therefore, rela The Fourier transform coefficients in the present embodi
tively easy to identify. This behavior may used either by ment are calculated as follows. Beginning at the top of the
itself to identify lateral boundaries or in tandem with the Zone of interest for a particular Seismic trace, a Series of
amplitude spectrum tuning cube as a confirmation of the Sliding window Fourier transforms of length L are computed
presence of local rock mass variability. 45 for each feasible position within Said Zone of interest. AS
Finally, it is anticipated by the instant inventors that the illustrated in FIG. 8, let the integer variable “M” be a counter
tuning cube technology disclosed herein might yield addi variable that represents the current sliding window number.
tional insights into Seismic reflection data. The tuning cube M is Set initially equal to unity to Signify the first sliding
(either containing phase or amplitude data) might be dis window position.
played and examined for empirical correlations with Sub 50 Now, for the trace at location (jk) within the Seismic data
Surface rock contents, rock properties, SubSurface Structure sub-volume X(j,k,i), the data for the Mth sliding window
or layer Stratigraphy. Alternatively, the Fourier transform may be extracted and moved into Short-term Storage, Said
values Stored in the tuning cube may be further manipulated sliding window starting at sample number (M-1)*NS:
to generate new Seismic attributes that can be useful in
exploration Settings. By way of example only, attributes that 55
could be calculated from the tuning cube values include the and thereafter transformed via a Fourier transform. As
average Spectral magnitude or phase, and any number of disclosed previously, a weight function may optionally be
other attributes. The importance of this aspect of the present applied to the data before transformation. For a fixed value
invention is best described as follows. It is well known in the of M, applying the previous calculation to every trace in the
Seismic interpretation arts that spatial variations in a Seismic 60 Sub-volume will produce a tuning cube for this particular
reflector's character may often be empirically correlated window position. Similarly, incrementing M and passing the
with changes in reservoir lithology or fluid content. Since entire data Volume through the algorithm again results in
the precise physical mechanism which gives rise to this another complete tuning cube, this one calculated for a
variation in reflection character may not be well understood, window location that begins NS Samples below the previous
it is common practice for interpreters to calculate a variety 65 window.
of Seismic attributes and then plot or map them, looking for The Fourier coefficients may now be placed in auxiliary
an attribute that has some predictive value. The attributes Storage until they are to be viewed. The notation developed
 5,870,691
23 24
above must be modified slightly to accommodate the fact the top of the Zone of interest and thin beds near the bottom
that Several windows might possibly be applied to each produce the Same characteristic amplitude spectra. This is
individual trace. Let A(i.k,i) represent the Volume of advantageous from the Standpoint of identifying thin beds,
collected of Fourier transform coefficients taken from all but it is a disadvantage in terms of determining their
traces in the Zone of interest for the “M”th calculated potential for hydrocarbon accumulation-higher bed eleva
window position. Note that the amount of Storage that must tions being generally preferable.
be allocated to this array has increased markedly. Now, the However, in the present embodiment the volume of same
total amount of Storage depends on the number of Sliding frequency Slices has "time' as its vertical axis: the variable
windows calculated for each trace, say NW, and must be at M being a counter that roughly corresponds to distance
least as may words of storage as the product of NWL, J, and down the Seismic trace. This organization provides addi
K: tional utility in that an approximate time duration of a thin
bed event can be established.
Storage=(NW)(L)(J)(K). For purposes of illustration, assume that a given thin bed
event has a frequency domain notch as 10 hertz. Then, every
AS was mentioned previously, it is entirely possible that short window Fourier transform that includes that bed will
15 exhibit the same notch. If a 10 hertz volume of slices is
A(j,k,i) may never be kept completely in RAM, but instead
kept partially in RAM and the remainder on disk. examined, there will be a range of Slices that contain the
Using the array notation introduced above and again notch. Thus, by Viewing Successive Slices in the constant
assuming that the Fourier transform of the weighted data is frequency Volume, it is possible to localize in time the
stored in X(i), the transform coefficients for the Mth window reflector of interest. More importantly, if it is known that a
of trace (i,j) are stored in array location: particular notch occurs at, Say, 10 hertz, the 10 hertz tuning
cube can be animated and Viewed as an aid in determining
the lateral extent of the thin bed: the limits of the notch as
observed in this frequency tuning cube defining the terminus
Once again, the individual frequency Slices within the of the bed.
numerous tuning cubes Stored in A(j,k,i) are preferably 25 In the previous discussion, the language has been
scaled by the procedure disclosed in FIG. 8 prior to their expressed in terms of operations performed on conventional
examination for thin bed artifacts. In each case, the hori seismic data. But, it is understood by those skilled in the art
Zontal frequency Slices are individually Scaled So that their that the invention herein described could be applied advan
average value is Set to Some particular constant, thereby tageously in other Subject matter areas, and used to locate
whitening the Spectra. other subsurface minerals besides hydrocarbons. By way of
After processing the Seismic traces within the Zone of example only, the same approach described herein could be
interest, each tuning cube may be individually examined for used to process and/or analyze multi-component Seismic
evidence of thin bed effects. As before, thin bed effects may data, shear wave data, magneto-telluric data, croSS well
be identified in the amplitude Spectra by viewing a Series of Survey data, full waveform Sonic logs, or model-based
horizontal Slices corresponding to different frequencies. 35 digital Simulations of any of the foregoing. In Short, the
Furthermore, this may be separately done for the tuning cube process disclosed herein can potentially be applied to any
corresponding to each window position, thereby obtaining Single geophysical time Series, but it is preferably applied to
Some general indication as to the temporal and Spatial extent a collection of Spatially related time Series containing thin
of a particular thin bed event. bed events. Thus, in the text that follows those skilled in the
According to a third aspect of the present invention, there 40 art will understand that "Seismic trace' is used herein in a
has been provided a method of enhancing thin bed effects generic Sense to apply to geophysical time Series in general.
using a discrete Fourier transform in the manner described While the inventive device has been described and illus
above for the Second embodiment, but containing the addi trated herein by reference to certain preferred embodiments
tional Step of organizing the Fourier transform coefficients in relation to the drawings attached hereto, various changes
into Single-frequency Volumes prior to display and analysis. 45 and further modifications, apart from those shown or Sug
This method is illustrated generally in FIG. 7. As disclosed gested herein, may be made therein by those skilled in the
Supra in connection with the Second embodiment, the aux art, without departing from the Spirit of the inventive
iliary Storage array A(j,k,i) will be filled with Fourier concept, the Scope of which is to be determined by the
transform coefficients and will be preferably scaled. following claims.
Let F(j,k,m) represent a single-frequency volume 50 What is claimed is:
extracted from A.G.k.i). There will be L/2+1 different 1. A method for the exploration of hydrocarbons, com
volumes ((L+1)/2 values if L is odd), one for each Fourier prising the Steps of
frequency produced by a transform of length “L”. A volume (a) accessing a set of spatially related Seismic traces, said
corresponding to the “i'th Fourier frequency is extracted Spatially related Seismic traces containing digital
from A(j,k,i) as follows: 55 Samples, Said digital Samples being characterized by at
least a time, a position, and an amplitude;
(b) selecting a part of Said set of spatially related Seismic
In effect, the array F(j,k,m) may be viewed conceptually traces to define a Zone of interest;
as being constructed by taking horizontal Slices from each of (c) transforming at least a portion of Said Seismic traces
the Sliding window Volumes and Stacking them in order of 60 within Said Zone of interest using a Fourier
increasing short window counter, M. transformation, Said Fourier transformation
The advantage of this present data organization for pur (i) being characterized by a plurality of orthonormal
poses of thin bed recognition is that it provides a means by basis functions, and
which the location of the thin bed event in time and space (ii) being applied to a window containing said digital
may be determined. By way of explanation, as was indicated 65 Samples to produce a plurality of transform coeffi
previously the temporal location of the thin bed within the cients associated with Said orthonormal basis func
Zone of interest does not affect its response: thin beds near tions,
 5,870,691
25 26
(d) organizing said transform coefficients into a tuning 6. A method according to claim 2, wherein a weight
cube, function is applied within Said window containing digital
(e) multiplying said transform coefficients by a Scaling Samples prior to transformation by Said discrete orthonormal
value to form a Scaled tuning cube, Said Scaling value transform.
being determined by 7. A method according to claim 6, wherein Said weight
(i) selecting at least two transform coefficients corre function is a Gaussian weight function.
sponding to a Same Said basis function, 8. A method according to claim 2, wherein the step (g)
(ii) calculating a complex magnitude of all transform includes the further Step of recording Visually perceptible
coefficients So Selected, images representative of one or more of Said tuning cubes on
(iii) calculating an average value from all transform a generally flat medium.
coefficient magnitudes So calculated, and, 9. A method according to claim 8, further including the
(iv) calculating a Scaling value from said average value; Step of using Said Visually perceptible images to identify
and, SubSurface Structural and Sedimentological features com
(f) displaying Said Scaled tuning cube. monly associated with the entrapment and Storage of hydro
15 carbons.
2. A method for the exploration of hydrocarbons, com 10. In the exploration for hydrocarbons, a seismic
prising the Steps of attribute map prepared by a process, Said process comprising
(a) obtaining a representation of a set of spatially related the Steps of:
Seismic traces distributed over a pre-determined Vol (a) accessing by means of a computer, a data set com
ume of the earth, Said Seismic traces containing digital prising Seismic traces distributed over a pre-determined
Samples, Said digital Samples being characterized by at Volume of the earth, Said Seismic traces containing
least a time, a position, and an amplitude; digital Samples, Said digital Samples being character
(b) selecting a part of said volume and the spatially related ized by at least a time, a position, and an amplitude;
Seismic traces contained therein to define a Zone of
interest within Said Volume;
(b) selecting a plurality of spatially related Seismic traces
25 from Said Seismic trace data Set,
(c) defining a window within said Zone of interest, said (c) Selecting a Zone of interest within said Selected plu
window having a Starting Sample number and encom rality of Spatially related Seismic traces,
passing digital Samples, (d) transforming at least a portion of Said spatially related
(d) transforming at least a portion of Said spatially related Seismic traces within Said Zone of interest using a
Seismic traces within Said Zone of interest using a discrete orthonormal transformation, Said discrete
discrete orthonormal transformation, Said discrete orthonormal transformation producing transform coef
orthonormal transformation ficients from Said Spatially related Seismic traces. So
(i) being characterized by a plurality of orthonormal transformed, and,
basis functions, and (e) organizing Said transform coefficients into a tuning
(ii) being applied to a window containing said digital 35
cube,
Samples to produce a plurality of transform coeffi (f) calculating a plurality of Seismic attribute values from
cients associated with Said orthonormal basis func
tions, Said transform coefficients organized into Said tuning
cube, and,
(e) organizing said transform coefficients into a tuning (g) displaying said Seismic attribute values at locations
cube, Said tuning cube and the transform coefficients 40
representative of Said positions.
therein being associated with Said Starting Sample num 11. In the exploration for hydrocarbons, wherein seismic
ber; data comprising reflected Seismic energy are recorded as a
(f) repeating Steps (c) and (d) for a plurality of window function of time over a pre-determined volume of the earth
definitions, thereby producing a plurality of tuning to produce a plurality of Spatially related Seismic traces, said
cubes, and, 45
Spatially related Seismic traces containing Samples, Said
(g) displaying one or more of Said plurality of tuning Samples being characterized by at least a time, a position,
cubes. and an amplitude, a map for the exploration of oil and gas
3. A method according to claim 2, wherein step (g) produced by the process of claim 10, comprising:
comprises the Steps of: 50 (a) a generally flat medium for recording visually percep
(i) Selecting an orthonormal basis function; tible images thereon, and,
(ii) Selecting a tuning cube from Said plurality of tuning (b) at least one visually perceptible image on said gener
cubes, ally flat medium, Said Visually perceptible image rep
(iii) extracting from said Selected tuning cube a plurality resentative of Said calculated Seismic attribute values.
of the transform coefficients associated with Said 55 12. A method for the generation of seismic attributes for
Selected orthonormal basis function; use in the exploration of hydrocarbons, comprising the Steps
(iv) repeating Steps (ii) and (iii) for at least one other of:
Selected tuning cube, (a) obtaining a representation of a set of Seismic traces
(v) organizing said extracted transform coefficients into a distributed over a pre-determined volume of the earth,
Single frequency tuning cube, and, 60 Said Seismic traces containing Samples, Said Samples
(vi) displaying said Single frequency tuning cube. being characterized by at least a time, a position, and an
4. A method according to claim 3, wherein Said organi amplitude;
zation in Step (v) includes ordering Said extracted transform (b) Selecting a part of said volume and the Seismic traces
coefficients by Said Starting Sample number associated there contained therein to define a Zone of interest within Said
with. 65 Volume;
5. A method according to claim 2, wherein Said discrete (c) transforming at least a portion of Said Seismic traces
orthonormal transform is a Fourier transform. within Said Zone of interest using a discrete orthonor
 5,870,691
27 28
mal transformation, Said very short time discrete (iii) calculating a statistical value from all transform
orthonormal transformation producing transform coef coefficient magnitudes So calculated;
ficients, (iv) calculating a Scaling value from Said Statistical value;
(d) organizing said transform coefficients into a tuning and,
cube, and, (v) applying said Scaling value to a plurality of transform
(e) calculating, from Said tuning cube, a plurality of coefficients corresponding to Said Same basis function.
Seismic attributes. 15. A method according to claim 14, wherein Said Statis
13. A computer based method of filtering geophysical tical value is an arithmetic average of all transform coeffi
time Series, comprising the Steps of: 1O
cient magnitudes So calculated.
(a) obtaining a representation of a set of spatially related 16. In a digital computer wherein Seismic traces obtained
Seismic traces distributed over a pre-determined Vol over a pre-determined Volume of the earth are read into
ume of the earth, Said Spatially related Seismic traces memory, wherein a plurality of Spatially related Seismic
containing digital Samples, Said digital Samples being traces are Selected from Said Seismic traces, and wherein a
characterized by at least a time, a position, and an 15
Zone of interest within Said spatially related Seismic traces
amplitude; has been defined, a digital computer programmed to perform
(b) selecting a part of said volume and the spatially related a process comprising the Steps of
Seismic traces contained therein to define a Zone of (a) transforming at least a portion of Said spatially related
interest within Said Volume; Seismic traces within Said Zone of interest using a
(c) transforming at least a portion of Said spatially related discrete orthonormal transformation, Said discrete
Seismic traces within Said Zone of interest using a orthonormal transformation producing transform coef
discrete orthonormal transformation, Said discrete ficients from Said Spatially related Seismic traces. So
orthonormal transformation producing transform coef transformed;
ficients from Said Spatially related Seismic traces. So (b) organizing said transform coefficients into a tuning
transformed; and, 25 cube,
(d) organizing said transform coefficients into a tuning (c) Scaling said transform coefficients within said tuning
cube, cube, and,
(e) scaling Said transform coefficients within said tuning (d) inverting said tuning cube using a discrete orthonor
cube, and, mal transformation inverse, thereby producing a fil
(f) inverting Said tuning cube using a discrete orthonormal tered version of Said transformed portion of Said Spa
transformation inverse, thereby producing a filtered tially related Seismic traces.
17. A device adapted for use by a digital computer
version of Said transformed portion of Said Spatially wherein a plurality of computer instructions defining the
related Seismic traces.
14. A method according to claim 13, wherein the step of process of claim 16 are encoded, said device being readable
35 by Said digital computer, and Said computer instructions
Scaling Said transform coefficients within Said tuning cube programming Said computer to perform Said process.
comprises the Steps of: 18. The device of claim 17, wherein said device is
(i) selecting at least two transform coefficients corre Selected from the group consisting of a magnetic tape, a
sponding to a same basis function; magnetic disk, an optical disk and a CD-ROM.
(ii) calculating a complex magnitude of all transform 40
coefficients So Selected; k k k k k
 UNITED STATES PATENT AND TRADEMARK OFFICE
CERTIFICATE OF CORRECTION
PATENT No. : 5, 870 691 Page l of 2
DAED Feb. 9, 1999
INVENTOR(S) : Gregory A. Partyka; James M. Gridley
it is certified that error appears in the above-identified patent and that said Letters Patent is hereby
Corrected as shown below:

Col. Line

3 30 "data transformations that could used"

should read:
data transformations that could be used.--

7 23 "an input trace xo, k, nt) "

should read:
an input trace x (j, k, nt) --
20 2 "extracted from Aok, i)"
should read:
--extracted from A ( j, k, i) --
 UNITED STATES PATENT ANDTRADEMARK OFFICE
CERTIFICATE OF CORRECTION
PATENT NO. : 5, 870, 691 Page 2 of 2
DATED ; Feb. 9, 1999
INVENTOR(S) : Gregory A. Partyka; James M. Gridley
it is certified that error appears in the above-identified patent and that said Letters Patent is hereby
Corrected as shown below:
Coil. Line
22 39-40 "calculated at every other star
The Four in within the zone of interest."
should read:
-- calculated at every other starting
position within the zone of interest.

Signed and Sealed this

Thirteenth Day of July, 1999

Q. TODD DICKINSON
Attesting Officer Acting Commissioner of Parents and Tradenarks