Full Waveform Inversion Is FWI a Bust a Boom or Becoming a Commodity Gerard Schuster KAUST 00 10 Dow Jones Index Avg decade 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 20102016 ID: 779870
Download The PPT/PDF document "The Boom and Bust Cycles of" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
The Boom and Bust Cycles of Full Waveform Inversion: Is FWI a Bust, a Boom, or Becoming a Commodity?
Gerard SchusterKAUST
0.0
1.0
Dow Jones Index
Avg
/decade
1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
Normalized DJI
Slide2Outline2. Seismic Experiment:L m = d
L m = d1
1L m = d
22
...
NN
4. Summary
3
. FWI, History, Examples
L1
L2
d
1
d
2m =
1. Inversion Overview:
Slide3Medical vs Seismic Imaging
CAT Scan
MRI
Full Waveform Inversion Tomogram
Slide4Traveltime Tomogram & Migration ImagesVshallow = L/t
Vdeep = L/t
time
t
L
t
L
Slide5Vshallow = L/t
time
V
deep
= L/t
Intersection
of down & up rays
Traveltime
Tomogram & Migration Images
Slide6Vshallow = L/t
time
V
deep
= L/t
Traveltime
Tomogram & Migration Images
Slide7V
shallow = L/t
time
V
deep
= L/t
migration image
Shot
gather = d(
x,t
)
Traveltime
Tomogram & Migration Images
Problems: Hi-Freq. ray tracing, picking
traveltimes
, tedious,
l
ow resolution, fails in complex earth models
Slide8-
=
observed
predicted
residual
Time
Full Waveform Inversion
Given:
d(
x,t
) =
Find
:
v(
x,y,z
) minimizes
e
=
S
[d(
x,t
)-d(
x,t
)
obs
]
2
x,t
p
redicted traces
Problems: Hi-Freq. ray tracing, picking
traveltimes
, tedious,
l
ow resolution, fails in complex earth models
Slide9Outline2. Seismic Experiment:L m = dL
m = d11
L m = d2
2
...
NN
3. FWI, History, Examples
4. Summary
4.
Summary and Road AheadL
1
L2
d
1
d2
m =1. Inversion Overview:
Slide10Gulf of Mexico Seismic Survey
m
L
m
=
d
L
m
=
d
1
1
L
m
= d2
2.
.
.
N
N
Time (s)
6 X (km)
4
0
1
d
Goal:
Solve
overdetermined
System of equations for m
Predicted data
Observed data
Slide11Outline2. Seismic Experiment:L m = dL
m = d11
L m = d2
2.
..
NN
3. FWI, History, Examples
4. Summary
4.
Summary and Road AheadL
1
L2
d
1
d2
m =1. Inversion Overview:
Slide12Details of
L
d
m
=
d-
d
obs
Reflectivity
or velocity
model
Time (s)
6 X (km)
0
dobs
m
– 1 d
2
d(
g|s
) = F
c
2
dt
2
2
[ ]
Predicted data
Observed data
F
c
2
dt
2
1 d
2
[ ]
2
-
-1
d(
g|s
)
=
d
m
(k)
=
L
T
(d-
d
obs
)
(k)
Slide13L= & d = Given: Lm=d
Find: m s.t
. min||Lm-d||
2Solution
: m = [L L
] L d
T
T-1
m = m – a
L (L
m - d) T
(k+1)
(k)
(k)
(k)or if L is too big
Problem:L is too big for IO bound hardwareL
1L
2
d
1
d
2
= m –
a
L
(
L
m - d )
(k)
+ L
(
L
m - d )
1
1
2
2
2
1
T
T
[
]
In general, huge
dimension matrix
Conventional FWI Solution
Slide14Outline2. Seismic Experiment:L m = dL
m = d11
L m = d2
2
...
NN
3. FWI, History, Examples
4. Summary
4.
Summary and Road AheadL
1
L2
d
1
d2
m =1. Inversion Overview:
Slide15Dow Jones Index vs FWI Index
0.0
18.0
Dow Jones Industrial
Avg
/decade
1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
Bunks
Multiscale
Mora
Exxon+
BP+Pratt
Tarantola
+ French School
Slide160.018.01930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
Dow Jones Index
vs FWI Index
Slide17What Caused the 1st FWI Boom?
0.018.01930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
True v(
x,z
)
FWI v(
x,z
)
0
2
Z (km)
0 X (km) 40 X (km) 4
Slide18What Caused the 1st FWI Bust?
0.018.01930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
True v(
x,z
)
FWI v(
x,z
)
0
2
Z (km)
0 X (km) 240 X (km) 24
Slide19What Caused the 1st FWI Bust?
0.018.0
1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
Time (s)
5.0
0.0
x
-
=
observed
predicted
residual
Waveform Misfit
V
V
true
V
start
0
1
Slide20What Caused the 1st FWI Bust?
0.0
18.0
1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
Time (s)
5.0
0.0
x
-
=
Waveform Misfit
V
V
true
V
start
observed
predicted
residual
0
1
=
Slide21What Caused the 1st FWI Bust?
0.018.0
1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
Time (s)
5.0
0.0
x
-
=
Waveform Misfit
V
V
true
V
start
Gradient opt. gets
stuck local minima
observed
predicted
residual
0
1
=
Slide22How to Cure the 1st FWI Bust?
0.0
18.0
1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
Time (s)
5.0
0.0
x
-
=
Waveform Misfit
V
V
true
V
start
Low-pass filter
=
observed
predicted
residual
0
1
Gradient opt
global
minima
Slide23How to Cure the 1st FWI Bust?
0.0
18.0
1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
Time (s)
5.0
0.0
x
-
=
Waveform Misfit
V
V
true
V
start
=
Window early events
observed
predicted
residual
0
1
Multiscale FWI v(
x,z
)
Slide240.018.01930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
True v(
x,z
)
FWI v(
x,z
)
0
2
Z (km)
0 X (km) 24
0 X (km) 24
How to Cure the 1st FWI Bust?
Multiscale FWI v(x,z)
2004 EAGE Meeting Started New Boom
Slide25Outline2. Seismic Experiment:L m = dL
m = d11
L m = d2
2
...
NN
3. FWI, History, Examples:
Transmission FWI Norway
4. Summary4
. Summary and Road Ahead
L1
L2
d
1
d2
m =1. Inversion Overview:
Slide268 km
Transmission 3D FWI Norway Marine Data
16 km1.5 km/s
3.5 km/s
2300 buried hydrophones, 50,000 shots, sea bottom 70 m
S.
Operto
, A.
Miniussi
, R.
Brossier
, L. Combe, L. Metivier, V.
Monteiller,
Ribodetti A., and J.
Virieux, 2015, GJI. Small dimensions of structures such as sandy outwash channels to 175m depth (Figure 1c) and the scars left on the sea paleo-bottom by drifting icebergs to 500m depth (Figure 1d). A wide low speed region defines the geometry of the gas cloud (Figure 1a, e) the periphery of which a fracture network is identified (Figure 1b, e). The image of a deep reflector, defining the base of the Cretaceous chalk under the tank (Figure 1a, b, white arrows), is uniquely identifiable despite the screen formed by the overlying gas cloud that opposes penetration seismic wave.
4.5
Gas cloud
gas
Slide273.5 km/s8 km2300 buried hydrophones, 50,000 shots, sea bottom 70 mgas
16 km
1.5 km/s
4.5
Transmission 3D FWI Norway Marine Data
Slide28Transmission 3D FWI Norway Marine DataVdeep ??????
2300 buried hydrophones, 50,000 shots, sea bottom 70 m
V
shallow
V
shallow
V
shallow
Transmissions 10x
stronger than
reflections
Therefore gradients will spend greater
effort updating shallow v(
x,y,z)
2D FWI R+T Gulf of Mexico Marine Data
Slide29Outline2. Seismic Experiment:L m = dL
m = d11
L m = d2
2
...
NN
3. FWI, History, Examples:
R+T FWI Gulf of Mexico
4. Summary4
. Summary and Road Ahead
L1
L2
d
1
d2
m =1. Inversion Overview:
Slide30observedpredictedTransmission Cigars
Reflection Rabbit Earspredicted
2
D FWI R+T Gulf of Mexico Marine Data
Abdullah
AlTheyab
(2015)
Slide31observedTransmission Cigarspredicted
2
D FWI R+T Gulf of Mexico Marine Data
Abdullah
AlTheyab
(2015)
3.6
0.0
Z (km)
19
0.0 X (km)
Migration Image with Initial V(
x,z
)
Migration Image with FWI V(
x,z
)
Migration Image with FWI V(
x,z
)
R+T
Slide32Outline2. Seismic Experiment:L m = dL
m = d11
L m = d2
2
...
NN
3. FWI, History, Examples:
Phase FWI Surface Waves
4. Summary4
. Summary and Road Ahead
L1
L2
d
1
d2
m =1. Inversion Overview:
Slide332D KSA Potash Model Test
Start
Model m/s0 60 120 x(m
) 0
30
800
600
400z (m)
WD Vs Tomogram
m/s
0 60
120 x(m
)
0
30 800
600400z (m)
1D Vs Tomogram m/s
0 60 120
x(m)
0
30
800
600
400
z
(m)
True
Model
m/s
0
60 120
x(m)
0
30
800
600
400
z
(m)
s
urface waves
10 m
Slide34Problem: 1D Dispersion inversion assumes layered medium (Xia et al. 1999).
Problem & Solution
True model
0 60 120
x(m)
0
30 Z (m)
v
(m/s)
Z (m)
1D
Inversion
w
(Hz
)Dispersion Curves C (m/s)
Radon
w
(Hz)
C (m/s)
Dispersion Curves
Radon
(Radon Transform)
x
(m)
t(s)
CSGs
t(s)
x (m)
CSGs
Z (m)
v
(m/s)
1D
Inversion
1D Vs Tomogram
0 60 120
x(m)
0
30
Z (m)
WD Vs Tomogram
0 60 120
x(m
)
0
30
Z (m)
2D
WD
Solution:
v(
x,y,z
) minimizes
e
=
S
[
c
(
k
,
w
)-
c
(
k
,
w
)
obs
]
2
(Li & Schuster, 2016)
FWI
: e
=
S
[
d
(
x,t
)-
d
(
x,t
)
obs
]
2
Slide35Seismic Imaging of Olduvai Basin
Kai Lu, Sherif
Hanafy
, Ian
Stanistreet, Jackson
Njau, Kathy Schick ,Nicholas Toth
and Gerard Schuster
Slide36Olduvai, Tanzania Seismic DataCOG
0 500 1000 1500 2000 2500 3000 3500
0 0.6
z (m)
P-wave Velocity Tomogram Tomogram
0 500 1000 1500 2000 2500 3000 3500
0
0.6
3500
20001500
m/sz
(m)
0 500 1000 1500 2000 2500 3000 3500
0 0.6
S-wave Velocity Tomogram (WD)1000800600m/s
z (m)
The Fifth Fault
The Fifth Fault
The Fifth Fault
Slide37Summary 3. Is FWI a commodity? Almost according to 2 industry expertsMultiscale+Skeletonized FWI:
e =
S |di –di
pred |2
v(x,y,z
), r(
x,y,z)
2. History
1930s 1980 1990 2010-2016
Is FWI a black box?
Not yet, works ~80% time (2 experts)
Challenges?
Deeper imaging, CPU cost,
multiparameter
i
Slide38Summary4. Road Ahead 3D Elastic Inversion & Adaptive Grid. Worth it?3D Viscoelastic Inversion. Worth it?Clever Skeletonized FWIAnisotropic InversionMultiples????Inversion Deeper than src-rec offset/depth<2
Slide39Summary(We need faster migration algorithms & better velocity models)IO 1 vs 1/20 or betterCost 1 vs 1/20 or betterResolution dx 1 vs 1
Sig/MultsSig ?
Stnd. FWI Multsrc
. FWI
Slide40Qademah Fault, Saudi Arabia Field Data
P-wave T
omogram
2D WD S-wave Tomogram
1D
S-wave Tomogram
Slide41Dispersion curve
f (Hz)
v (m/s
)
Comparison of 2D WD Inversion with FWI
x
(m)
t (s)
x
(m)
z (m
)
Vs True Model
x
(m)
z (m
)
Vs Tomogram
FWI
x
(m)
t (s)
A shot gather
FD
x
(m)
z (m
)
Vs Tomogram
WD
FWI
of
Surface Waves
Easy to get stuck in a local
minimum (Solano, et al., 2014)
.
2D WD of
S
urface Waves
Avoid
local
minimum and apply in 2D/3D model.
Start Model
Slide420.018.01930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s 2010-2016
FWI Index
Avg
/decade
True v(
x,z
)
FWI v(
x,z
)
0
2
Z (km)
0 X (km) 24
0 X (km) 24
How to Cure the 1st FWI Bust?
Multiscale FWI v(x,z)
2004 EAGE Meeting Started New Boom
Slide43V
shallow = L/t
time
V
deep
= L/t
migration image
Given: d(
x,t
) =
Find: v(
x,y,z
) minimizes
e
=
S
[d(
x,t
)-d(
x,t
)
obs
]
2
Full Waveform Inversion
x,t
Slide448 km
Transmission 3D FWI Norway Marine Data
16 km1.5 km/s
3.5 km/s
2300 buried hydrophones, 50,000 shots, sea bottom 70 m
S.
Operto
, A.
Miniussi
, R.
Brossier
, L. Combe, L. Metivier, V.
Monteiller,
Ribodetti A., and J.
Virieux, 2015, GJI. Small dimensions of structures such as sandy outwash channels to 175m depth (Figure 1c) and the scars left on the sea paleo-bottom by drifting icebergs to 500m depth (Figure 1d). A wide low speed region defines the geometry of the gas cloud (Figure 1a, e) the periphery of which a fracture network is identified (Figure 1b, e). The image of a deep reflector, defining the base of the Cretaceous chalk under the tank (Figure 1a, b, white arrows), is uniquely identifiable despite the screen formed by the overlying gas cloud that opposes penetration seismic wave.
4.5
Gas cloud
gas