Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
University of British ColumbiaSLIM
Rajiv Kumar
Affordable omnidirectional subsurface extended image volumes
Released to public domain under Creative Commons license type BY (https://creativecommons.org/licenses/by/4.0).Copyright (c) 2015 SLIM group @ The University of British Columbia.
University of British ColumbiaSLIM
Tristan van Leeuwen and Felix J. Herrmann
Affordable omnidirectional subsurface extended image volumes
Why we need image gathers ?Full subsurface offset volumes allow us to conduct
• AVA analysis
• Geological dip estimation
• Velocity analysis in complex geological enviornment
• Given two-way wave equations, source and receiver wavefields are defined as
where discretization of the Helmoltz operator
source
data matrix
samples the wavefield at the source and receiver positions
squared-slowness
Extended images
H(m)U = PTs Q
H(m)⇤V = PTr D
H(m) :
Q :
D :
Ps, Pr :
m :
Forward propagation
Backward propagation
Extended images• Organize wavefields in monochromatic data matrices where each column
represents a common shot gather
• Express image volume tensor for single frequency as a matrix
E = UV ⇤
E = H�1PTs QD⇤PrH
�1
grid
poin
tssources
4D image volume as matrix
nx x nz
Extended images
example for one layer
Extended images
500 1000 1500 2000 2500
500
1000
1500
2000
2500
full matrixx [m]
z [m
]
800 900 1000 1100 1200
500
600
700
800
900
1000
one columnx [m]
z [m
]
800 900 1000 1100 1200
500
600
700
800
900
1000
diagonal
Extended images
Impediments
Prohibitively expensive to compute and store
Current Workflow• Compute all the source and receiver wavefields
• Severely subsample the image volume in the subsurface coordinates
• Allowed limited interactions in predefined directions• horizontal or vertical based upon geology of interest
Biondi and Symes, 2004, Sava and Biondi, 2004, Sava and Vasconcelos, 2011, Yang and Sava, 2015
Motivation
Can we avoid computing all of the wavefields if not forming the full image volumes?
Can we gleans information from the full image volume without requiring a-priori knowledge of the geology?
Motivation
Extended images• Probe volume with tall matrix
where represents single scattering points
van Leeuwen 2012eE = EW = H�1PTs QD⇤PrH
�1W
W = [w1, . . . ,wl]
wi = [0, . . . , 0, 1, 0, . . . , 0]
Computation
• mat-vec with extended image :
• (one subsurface source)
• (surface source function)
• (one surface source)
14
eE = EW = H�1PTs QD⇤PrH
�1w
ed = PrH�1w
ey = QD⇤ed
eE = H�1PTs ey
Computation computation of an image point gather
# of PDE solves “flops for correlations”
conventional 2Ns Ns x Nh
ours 2Nx Ns x Nr
Ns - # of sources Nr - # of receiversNh - # of subsurface offsetsNx - # of sample points
15
Computation computation of an image point gather
time (s) memory (MB)
conventional 23.6 103
ours 2.02 0.03
16
University of British ColumbiaSLIM
Application to WEMVA
WEMVA [ conventional method]
500 1000 1500 2000 2500
500
1000
1500
2000
2500
500 1000 1500 2000 2500
500
1000
1500
2000
25000
0.2
0.4
0.6
0.8
1
h E
.⇤
.⇤ stand for element-wise multiplication
Biondo & Symes, ’04 , Symes 2008, Sava & Vasconcelos, ’11
⇤matrix-matrix multiplication
Focusing [ propose method]
Ediag(x) ⇡ diag(x)E
500 1000 1500 2000 2500
500
1000
1500
2000
2500
500 1000 1500 2000 2500
500
1000
1500
2000
2500
E diag(x)
⇤
500 1000 1500 2000 2500
500
1000
1500
2000
2500500 1000 1500 2000 2500
500
1000
1500
2000
2500
diag(x) E
⇤⇡
Focusingwhere represents horizontal, vertical or all offset.x
horizontal offsethorizontal +vertical
offset
all offsets
Why do we need all offsets
x (m)
z (m
)
0 1000 2000 3000 4000 5000 6000
0
1000
2000
Yang and Sava, 2015, van Leeuwen et. al. 2015
Common image gathersHorizontal reflector
z − zo (km)
z (k
m)
−0.4 −0.2 0 0.2 0.4
1.3
1.4
1.5
1.6
1.7
1.8x − xo (km)
z (k
m)
−1 −0.5 0 0.5 1
1.4
1.5
1.6
1.7
1.8
x (m)
z (m
)
0 1000 2000 3000 4000 5000 6000
0
1000
2000
x (km)
z −
z o (km
)
0.5 1 1.5
−0.5
0
0.5x (km)
x −
x o (km
)0.5 1 1.5
−1
−0.5
0
0.5
1
x (m)
z (m
)
0 1000 2000 3000 4000 5000 6000
0
1000
2000
Common image gathersVertical reflector
Common image point gathers
x − xo (km)
z −
z o (km
)
−1 −0.5 0 0.5 1
−0.5
0
0.5x − xo (km)
z −
z o (km
)
−1 −0.5 0 0.5 1
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
x (m)
z (m
)
0 1000 2000 3000 4000 5000 6000
0
1000
2000
Fast WEMVA w/ randomized probing• Measure the error in some norm
• The Frobenius norm can be estimated via randomized trace estimation :
where
Avron and Toledo, 2011
||A||2F = trace(ATA)
KX
i=1
wiwTi ⇡ I
⇡KX
i=1
wTi A
TAwi =KX
i=1
||Awi||22
minm
||E(m)diag(x)� diag(x)E(m)||2?
Randomized probing [reflection]
x(m)z(
m)
0 2000 4000 6000
0
500
1000
1500
1800
2000
2200
2400
2600
2800
3000
3200
3400
x(m)
z(m
)
0 2000 4000 6000
0
500
1000
15002000
2500
3000
3500
4000
true model initial model
Randomized probing [reflection]
•Exact • Error bar of approximated objective function (different color represents different random realization)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 21.5
2
2.5
3
3.5
4
x 105
α
||E
dia
g(x
) −
dia
g(x
)E||
F2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 21.5
2
2.5
3
3.5
4
x 105
α
||E
dia
g(x
) −
dia
g(x
)E||
F2
K = 10 K = 80
Lens Model x (m)
z (
m)
0 500 1000 1500 2000 2500 3000
0
500
1000
1500
1500
2000
2500
3000
3500
4000
True model Initial model WEMVA
x (m)
z (m
)
1000 1500 2000
0
500
1000
1500
2000
2500
3000
3500
4000
x (m)
z (m
)
1000 1500 2000
0
500
1000
1500
2000
2500
3000
3500
4000
x (m)
z (m
)
1000 1500 2000
0
500
1000
1500
2000
2500
3000
3500
4000
Lens Model [image gathers]
offset(km)
z(k
m)
−1.5 −1 −0.5 0 0.5 1 1.5
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
offset(km)
z(k
m)
−1.5 −1 −0.5 0 0.5 1 1.5
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
offset(km)
z(km
)
−1.5 −1 −0.5 0 0.5 1 1.5
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
True model Initial model WEMVA
Least-squares RTM Images
True model Initial model WEMVA
offset(km)
z(k
m)
1 1.2 1.4 1.6 1.8 2
0
0.5
1
1.5
offset(km)
z(k
m)
1 1.2 1.4 1.6 1.8 2
0
0.5
1
1.5
offset(km)
z(k
m)
1 1.2 1.4 1.6 1.8 2
0
0.5
1
1.5
Conclusions• probings allows us to get offset information for all sub-surface
direction
• Prior knowledge of geology is not required
• randomized trace estimation allows us to compute WEMVA objective cheaply
University of British ColumbiaSLIM
Image gathers w/ surface-related multiplesNing Tu
Why need multiples ?
33
S R4R1 R2 R3
subsurface)reflector
water)surface
R1 R2 R3 R4S
subsurface)reflector
water)surface
illumination by primaries
illumination by multiples
34
Least-‐squares imaging: primary-‐only shot gather Least-‐squares imaging: multiple-‐only shot gather
Tu and Herrmann, 2015
• Leverage benefits of SRME
- highly accurate data-driven multiple prediction
• All in one go method
- we combine SRME within the extended imaging condition
Motivation
35
Verschuur et. al. ’92
Extended imaging with multiples
36
eE = EW = H�1PTs (Q� P )P ⇤PrH
�1W
: areal source
: total upgoing wavefieldP
(Q� P )
where
Least-square extended imaging
37
where
Kumar et. al. ’14
minimizeeE
1
2kF( eE)� Pk2F ,
F( eE) = PrH�1EWTH�1PT
s (Q� P ),
Velocity model
True model Initial model
38
x (m)
z (m
)
0 500 1000 1500 2000
0
200
400
600
800
1000 1700
1800
1900
2000
2100
2200
2300
x (m)
z (m
)
0 500 1000 1500 2000
0
200
400
600
800
1000 1700
1800
1900
2000
2100
2200
2300
39
Least-squares RTM images
z (
m)
x (m)0 500 1000 1500 2000
0
200
400
600
800
1000
z (
m)
x (m)0 500 1000 1500 2000
0
200
400
600
800
1000
z (
m)
x (m)0 500 1000 1500 2000
0
200
400
600
800
1000
Primary only Primary + multiplesw/o areal sources
Primary + multipleswith areal sources
Primary only40
z (m
)
x − xo (m)
−400 −200 0 200 400
0
200
400
600
800
1000
z (m
)
x − xo (m)
−400 −200 0 200 400
0
200
400
600
800
1000
Primary + multiplesw/o areal sources
z (
m)
x − xo (m)
−400 −200 0 200 400
0
200
400
600
800
1000
Primary + multipleswith areal sources
Conclusions
41
Multiples provide extra illumination that can complement primaries.
Multiples can be used with primaries to form subsurface image gathers via least-squares inversion.
Acknowledgements
This work was in part financially supported by the Natural Sciences and Engineering Research Council of Canada Discovery Grant (22R81254) and the Collaborative Research and Development Grant DNOISE II (375142-‐08). This research was carried out as part of the SINBAD II project with support from the following organizations: BG Group, BGP, BP, CGG, Chevron, ConocoPhillips, ION, Petrobras, PGS, Statoil, Total SA, WesternGeco, and Woodside.
Thank you for your attention ! https://www.slim.eos.ubc.ca/