Spectral Decomposition
21
Course Outline
Geometric Attributes
Spectral Decomposition
Multiattribute Display
22
Spectral Decomposition
23
Fourier components of a broadband seismic pulse
Time (ms)
50
impedance 100
150
0
0.1
Time (ms)
50
reflectivity 100
150
Time (ms)
50
seismic
100
150 Env
2
0
Time (ms)
50
envelope 100
150
0
3/4
/16
/2
/4
/8
0
u
Time (ms)
100
Note!
u
iU ( )
We can DETECT an amplitude
difference between T=/8 and /16 !
But there is no change between peak t
trough thickness!
Rotate each Increase amplitude of
frequency by 900 high frequencies
28
Alternative Basis Functions
29
SWDFT wavelets SWDFT spectra
1
Tapered window
1
0.5 0.8
0.6
fc=10 Hz 0
0.4
0.2 0.1 0.0 0.1 0.2
0.5 0.2
0
1 20 Hz10 20 30 40 50 60 70 80 127
90 100 Hz
1
1
0.5 0.8
0.6
0
fc=20 Hz 0.4
0.2 0.1 0.5 0.0 0.1 0.2 0.2
0
1 0 10 20 30 40 50 60 70 80 90 100
1.0
1
0.5
0.8
0.6
fc=40 Hz 0.2 0.1
0.0
0.0 0.1 0.2 0.4
0.5 0.2
0
1.0 0 10 20 30 40 50 60 70 80 90 100
1.0
1
0.8
0.5
0.6
0.0 0.4
fc=20 Hz 0.2 0.1 0 0.1 0.2 0.2
0.5 0
0 10 20 30 40 50 60 70 80 90 10
1.0 0
1.0
1
0.8
0.5
0.6
0.0 0.4
fc=40 Hz 0.2 0.1 0 0.1 0.2 0.2
0.5 0
0 10 20 30 40 50 60 70 80 90 10
1.0 0
1.0
1
0.8
0.5
0.6
0.4
fc=20 Hz 0.2 0.1
0.0
0 0.1 0.2 0.2
0.5 0
0 10 20 30 40 50 60 70 80 90 10
1.0 0
1.0 1
0.8
0.5
0.6
0.0 0.4
fc=40 Hz 0.2
0.2 0.5 0 0.2
0
0 10 20 30 40 50 60 70 80 90 10
1.0 0
213
Source Seismic
Reflectivity wavelet Noise data
r(t) * s(t) + n(t) u(t)
Long window
Time
spectral
decomposition
and the
convolutional Fourier Transform
model
Amplitude Amplitude Amplitude Amplitude
Frequency
+ =
Bandlimited white
White spectrum
spectrum
214 (Partyka et al., 1999)
Spectral balancing
a(f)
Noise threshold=amax
Frequency, f
Noise threshold=0.5
Frequency, f
215
Source Seismic
Reflectivity wavelet Noise data
r(t) * s(t) + n(t) u(t)
Short window
spectral
Time
decomposition * + =
and the
convolutional Fourier Transform
model
Amplitude Amplitude Amplitude Amplitude
Frequency
+ =
Time (s)
Time (s)
Frequency (Hz)
Frequency (Hz)
218
Thin bed tuning variation with thickness
A A 30 Hz
Time (s) 15 Hz
A A
A A
15 Hz Map 30 Hz Map
219 (Laughlin et al., 2002)
Thin bed tuning variation with thickness
10 Hz 30 Hz 50 Hz
thalweg
moderate
channelthickness
edges
221 (Courtesy of Apache Corp.)
The thin bed tuning model
r1
r2=r1
r1=2r2 r1=+2r2
r1=r2 r1=+r2
r1=0.7r2 r1=+0.7r2
r1=0.5r2
r1=+0.5r2
1.1
2 km
Freq
Time
A B (Hz)
(ms)
0
30
15
60
30
Amplitude
Peak frequency
Temporal fromfrom
thickness synthetic
picks data
A B
Flatten Flatten
A
A
South North
Wharton #1 Pester #2 Griffin #1 Jameson #1
L. Skinner
Flattened
on easy
Novi
topick
reflector
320 ft
50 ms 1.6 km
100 m
L. Skinner
analysis
window
Novi
320 ft
50 ms 1.6 km
100 m
36 Hz spectral
component over
Red Fork
Shaded
illumination
map of
SkinnerNovi
isochron
Stage V shows
up due to
differential
compaction
Stage V
233
RedGreenBlue color blending of three discrete
spectral magnitude volumes
Z1<Z2>Z3 Z1<Z2<Z3
SWDFT SWDFT
CWT CWT
40 Hz Blue
50 Hz Green
60 Hz Red
amp
18 Hz Red
24 Hz Green
36 Hz Blue
f (Hz)
amp
f (Hz)
1. For a window (e.g. 100ms) about each time slice, assume that the
geology is random
2. Calculate average spectrum at each time slice
3. Rescale spectral components such that the new average value is 1.0
4. Plot relative spectral amplitudes on each time slice
237
Spectrum of a 100 ms analysis window
100
Mode
Amplitude
(peak) 50
Mean
(average)
0
20 40 60 80
Frequency (Hz)
Magnitude
Noise level
trough
Frequency
238
Definition of peak spectral magnitude
and peak frequency
High
Peak Magnitude
Magnitude
0
0 10 20 30 40 50
Frequency (Hz)
Peak freq
5 km 30
60
Coh
1.0
0.6
Peak Frequency
blended
Peak with Coherence
Frequency
Coherence
240 (Marfurt and Kirlin, 2001)
Spectral Decomposition using the
Continuous Wavelet Transform
(CWT)
241
1
0.8 Mother wavelet
0.6
Continuous Wavelet Transform
Amplitude
0.4
Amplitude
Amplitude
2 10
1 8
0 6
1 4
2 2
3 0
3 2 1 0 1 2 3 20 40 60 80 100 120
Time (s) Frequency (Hz)
Frequency
40 80
Scales
Escala
30 60
20 40
10 20
Time (ms)
100 200 300 400 500 600 700 800 900 1000
Amostras
Samples
0.4
Amplitude
0.2
0.2
100 200 300 400 500 600 700 800 900 1000
243 Amostras
Time (ms) (Matos and Marfurt, 2011)
Forward CWT
Reflectivity Synthetic CWT Magnitude Voices ICWT
CWT
magnitude
pos
Le Nozze di Figaro
245
Read seismic trace Matching pursuit flow
Precompute seismic chart
wavelets Generate complex seismic trace
and their spectra
Set residual=original complex trace
Set complex spectrum=(0.0,0.0)
no Residual
energy < threshold value?
yes
Sum spectra of
246 component wavelets (Liu and Marfurt, 2005)
Pennsylvanian Age Channels, CBP, Texas, USA
0.6 A A
Time (s)
1.060 s
Modeled
1.6 Modeled data
data after
after 2iterations
iterations
1 iterations
4
8
16 iteration
0
Time (s)
0
Time (s)
2 km
Original data
252 A (Liu and Marfurt, 2005)
Time Slice through Spectral Components
A
Pennsylvanian Channels
Time Slice t=1.060 s
2 km
20
10Hz
40
80
30
60
70
50
90 Hzcomponent
Hz component
component
253 A (Liu and Marfurt, 2005)
Peak frequency modulated 10
Peak frequency (Hz)
90
by peak magnitude high
low
2 km
254
A (Liu and Marfurt, 2005)
2 km
Peak frequency
modulated by
peak amplitude
(Time slices)
30
Peak Mag
010 20 40 60 80 100
Peak Freq (Hz)
Time (ms)
Amplitude
Amplitude
Amplitude
50 50 50
0 0 0
20 40 60 80 20 40 60 80 20 40 60 80
Frequency (Hz) Frequency (Hz) Frequency (Hz)
A Low High
Horizon slice
through peak
A frequency and
peak amplitude
above Atoka
unconformity
2 km
High
1.0
Atoka
Time (s)
1.2
Low
(a)
Amplitude
Amplitude
Amplitude
50 50 50
0 0 0
20 40 60 80 20 40 60 80 20 40 60 80
Frequency (Hz) Frequency (Hz) Frequency (Hz)
(b) (c) (d)
Low High
bandwidth
0.00
Jlow Jhigh Frequency index
1.00
Percentile
phigh
0.50
plow
0.00
Jlow J50 Jhigh Frequency index
mRTM mj
J high J low 1 j J low
Spectral slope
The slope that best fits the spectrum between Jlow and Jhigh
Spectral Roughness
A measure of how well the spectrum is fit by a linear
variation between Jlow and Jhigh .
0.9
1.0
Time
1.1
1.2
1.3
1 km
Negative
Coherence
Continuous Wavelet
Short Window DFT Transform (CWT)
Comparison of
constant
bandwidth, CWT,
and matching
pursuit spectral
decomposition
algorithms
SWDFT CWT Matching Pursuit
271
Average timefrequency spectrum (106 traces)
Mag
High
Mag
High
1.0
1.2
1.4
CDP no.
274 (Qi and Marfurt, 2014)
Balanced and blued seismic amplitude
A A
Amp
0.6 Top
Large scale
Positive
1.0
1.2
1.4
CDP no.
275 (Qi and Marfurt, 2014)
Bandwidth Extension
276
Inverse CWT
CWT
magnitude
pos
CWT
magnitude
pos
X 0
0
Upper Barnett Lm
Upper Barnett Sh
60
Forestburg
Lower Barnett Sh
Viola
282 (Matos and Marfurt, 2011)
Sparsespike frequency domain inversion
Marble Falls
Amplitude
10
0
Upper Barnett Lm
Upper Barnett Sh
10
Forestburg
Lower Barnett Sh
Viola
283 (Matos and Marfurt, 2011)
Geotraces bandwidth extension using the
inverse CWT
before after
Geotraces
bandwidth
Vertical extension
slice AA using inverse
CWT
Phantom horizon slice
20 ms below top
carbonate
A A A A
before
after
289
Thickness estimate for an even impulse function
calculation
Impulse Function
Center at t=0
Fourier Transform
G(f) = r1 ()exp(i2f [t + T/2]) + r2 ()exp(i2f [t  T/2])
293
Seismic Estimation of Q
Reference
(1, )
(1 0) = ln( )
(0, )
Target
ln[B(t1,f)/B(t0,f)]
0 Frequency Slope~1/Q
Frequency
1/Q
1/Q
Reference
(1 0)1 20
=
2(20 21
Target
wavelet.
(1 0)20
=
0 1
0 Centroid Centroid Frequency
Frequency Frequency
Lowfrequency
Attenuation Slope
Highfrequency
Attenuation Slope
0 Frequency
High
frequency Energy Difference Attribute
Magnitude
Energy Decay
Lowfrequency
Energy
Augmentation
0 Frequency
Depth High
Lithology
Well A Well B Well C
H2
Low
H1
Well A
Highfrequency
Attenuation Slope
High
Well C
Low
Well A
Highfrequency
Attenuation Slope
High
Well C
Low
Highfrequency
Energy Decay
Depth Lithology
Well A Well B Well C
High
H2
H1
Low
Lowfrequency
Energy
Augmentation
Depth
Lithology Well C
Well A Well B
High
H2
H1
Low
2106
Attenuation from Natural and Induced Fractures
Attributes for fracture delineation
Barnett Shale
2108 (Courtesy of Fangyu Li and Huailai Zhou. Data courtesy of Devon Energy.)
Attenuation from Natural and Induced Fractures
Most negative curvature
2109 (Courtesy of Fangyu Li and Huailai Zhou. Data courtesy of Devon Energy.)
Attenuation from Natural and Induced Fractures
(TimeLapse Seismic)
amp T
Unstimulated zone
Stimulated zone
2115
Stabilized InverseQ filter
Low Q
2119 (Courtesy of eSeis)
QBased PorePressure Procedure
1. Compute Q 2. Correlate mud
from spectral weights to Q
3. Predict pore pressure
components Mud weights Q
Weight
2 lbs/gal
Time (s)
20
3
15
4
10
5
2 km
15
1
10
2
Time (s)
15
1
10
2
Time (s)
Shale
2
Sandy
Depth (km)
3
Shale
4
Carbonate
5
2124
90
80
70
60 Time
Spectra 50
Domain
Before 40
Wavelet
Filtering 30
20
Before
10 Entering
0 Reservoir
0 20 40 60 80 100
90.00%
80.00%
Compressible 70.00%
Fluid 60.00%
Energy Loss
Energy 50.00%
Loss Function
40.00% Compressible Fluid
30.00%

20.00%
10.00%
Gas Reservoir
0.00% Filter
0 10 20 30 40 50 60 70 80 90
Hz
90
80
70
Spectra 60
After 50
Time
Filtering 40 Domain
30 Wavelet
20
After
10
0
Exiting
0 10 20 30 40 50 60 70 80 90
Filter
2125 (courtesy of Apex Spectral Technology, Inc.)
Dry Gas Attenuation Signature
90
80
70
Dominant
60 Frequency or
DF drops due
50 to the filtering
effect of gas as
40 the traveling
wavelet passes
30 through the
reservoir
20
10
0
0 10 20 30 40 50 60 70 80 90
70
60
Delta
50
Frequency
shortens due to
40 the filtering
effect of gas as
30 the traveling
wavelet passes
20 through the
f reservoir
10
f
0
0 10 20 30 40 50 60 70 80 90
f curve
US Patent
Application No.
11/903,602
Wilcox reservoir
exhibits strong f
signature
2129
Estimating fluid mobility (permeability and viscosity)
R r()
R0 Permeable
tn Impermeable
t
t t t n = i f n
fluid
mobility reservoir geometry dominant
dominant r
f
Impermeable
Permeable
0 10 20 30 40 50 60 70
Frequency (Hz)
1.5
Time (s)
sandstone
2.0 reservoir
fractured
shale
2.5 reservoir
Oil
Oil and Water
Water
OilWater contact predicted by
low frequency analysis
101
63
77 86
Blind Test 73
6
4 Calibration wells 93
3 Oil 74
3
1 Water 75 78
2133
(Goloshubin et al, 2002)
Estimating fluid mobility
1200
Production rate (m3/day)
800
400
0
0 4 8 12 16 20
fluid r
2
mobility f
f
2134
(Goloshubin et al, 2002)
FrequencyDependent AVO
(Gulf of Mexico, U.S.A.)
2136
Earlier recognition of waveform singularities
Original data
component)
Envelope
(real
component)
Quadrature
(imaginary
+180
Phase
180
Weighted average frequency
Frequency
Singularity
2137 (Taner et al, 1979)
Instantaneous Frequency
0.0
1.0
Time (s)
2.0
3.0
2138
(Courtesy Rock Solid Images)
Envelope weighted Instantaneous Frequency
0.0
1.0
Time (s)
2.0
3.0
2139
(Courtesy Rock Solid Images)
Thin bed indicator (finst<finst>)
0.0
1.0
Time (s)
2.0
3.0
2140
(Courtesy Rock Solid Images)
SPectral Imaging of Correlative Events (SPICE)
Fit the spectral amplitudes r(s,t=t0) in a loglog plot with a
straight line. The slope of this line is the Hlder exponent, h:
h=0.684
log2(r(s,t=t0))
log2(s=scale)
2141 (Li & Liner, 2004)
The Hlder exponent, h, measures the strength of a
singularity
h = 1 (a Dirac distribution)
+1.0
Hlder exponent from AI
0.0
+0.1
0.0 Reflection Coefficient (RC)
0.1
0.0
Time (samples)
2143 (Smyth et al., 2004)
1 km Mapping unconformities
1.0
Time (s)
Seismic amplitude
1.5
1.0
Time (s)
Hlder exponent
1.5
2144 (Liner et al, 2004)
Seismic through a Miocene channel
Gulf of Mexico, USA
A A
1.0
Time (s)
1.5
2.0
A A
Zero
Negative
Thickness (ms)
a) b) c) d)
0 50
0
Time (s)
0.125
0.250
Positive +

Zero
Time (s)
Time (s)
0.25 0.25
4 Frequency (Hz) 84 4 Frequency (Hz) 84
b) 23 ms
0 0 Phase
Amplitude +
Positive
0

Zero
Time (s)
Time (s)
0.25 0.25
2150
4 Frequency (Hz) 84 4 Frequency (Hz) 84
c ) 25 ms CWT Magnitude CWT Phase
0 0 Phase
Amplitude +
Positive
0

Zero
Time (s)
Time (s)
0.25 0.25
4 Frequency (Hz) 84 4 Frequency (Hz) 84
d ) 48 ms
0 0 Phase
Amplitude +
Positive
0

Zero
Time (s)
Time (s)
0.25 0.25
2151
4 Frequency (Hz) 84 4 Frequency (Hz) 84
Computation of phase residues in timefrequency domain
Phase
0 (radians)
+
0 r dr r0

Time (s)
Frequency (Hz)
35 40 45
0.100
0.25 4
4 Frequency (Hz) 84
1 A 3
Time (s)
2 8
0.125
A 5 B 7
B
0.150
2152
Computation of phase residues in timefrequency domain
Frequency (f) Frequency (f)
Time (t)
Time (t)
4 8
0.2 0.4 0.6 0.2 0.4 0.6
1 3 5 7
2 6
0.2 0.4 0.8 0.2 0.4 0.8
1=0.4/2 5=0.8/2
2=0.2/2 6=0.4/2
3=0.8/2 7=W(1.4)=0.6/2
4=0.2/2 8=0.2/2
Amplitude
0
Positive
Zero
Negative
Time (s)
0.25
0 50 0
Thickness (ms)
2154 (Matos et al., 2010)
Mapping clinoforms associated with
ProgradationRegression
0.0
Time (s)
0.5
1.0
0.7
0.9
0.7
0.9
2157
Incised Valleys (Anadarko Basin)
Components of these spectra can be used to detect lateral changes in layer thickness and
heterogeneity, well below the limits of classic /4 seismic resolution
Peak spectral frequency (the mode of the spectrum) is a good zero order representation of the
seismic spectral response. Other statistical measures may help delineate upwardfining and
upwardcoarsening sequences.
Matching pursuit spectral decomposition provides less vertical mixing of stratigraphy than that
based on fixedlength window discrete Fourier transforms.
Low frequency anomalies are often associated with low GOR some interpret this as a
viscosity effect, others as complex waveform healing
Complex trace attributes degenerate when multiple reflectors interfere with each other the
SPICE algorithm and phase residues use this interference as signal
2164
Patternrecognition Math
sinx
six
?
?
n
2165