«наука – баба веселая и паучьей серьезности не терпит».

EAGE Education Tour 2007

Beyond conventional seismic imaging ( Part I)
Instructor: Dr. Evgeny Landa
In memory of my teacher and friend,
Sergey Goldin
 Summary
I. CMP-based imaging
1. Hyperbolic imaging
2. Shifted hyperbola
3. Non-hyperbolic imaging
4. Comparison of different CMP approximations

II. Non-CMP based imaging

1. Multifocusing method
2. Common Reflection Surface Stack
3. MF versus CRS
4. Advantages of the MF and CRS methods
5. Examples

III. Path integral imaging

1. Feynman path-integral idea
1. Path-summation zero-offset approximation
2. Path-summation time migration

IV. Diffraction imaging

V. Improved imaging based on recursive prestack datuming

VI. How correct is a velocity model

 Acknowledgements
This course contains the contribution of many people and it is often
inspired by the work of others.
Sergey Fomel, Sabrina Ernst, Vladimir Glogovsky, Boris Gurevich,
Shemer Keydar, Tijmen Jan Moser, Josef Pafenholz, Moshe Reshef,
Tury Taner, Tadeusz Ulrych, Oz Yilmaz.
To all of them I am greatly thankful.
I want to acknowledge several companies that provided me with field examples
illustrating applications of different non-conventional imaging methods: GII,
Geomage, TEEC, Petrologic, Landmark

I thank EAGE for proposing me to be an instructor in the EAGE Education Tour

and for their decision to publish this book.
I thank also the SEG for permission to use figures published in Geophysics and
The Leading Edge journals.
The acknowledgments are also extended to TOTAL for sponsoring this project and
for allowing me to work on this book and letting me to participate in the
educational tour.
 Introduction
Time versus depth
Myth 1: Depth imaging can be done without time processing
Demystification: While depth imaging plays an increasing role, imaging in
time constitutes a key step

Recorded seismic data Recorded seismic data

Wavefield parameter Velocity model

estimation estimation

Time Imaging Depth Imaging

 Introduction
Correlation and Stacking
Decomposition of total wavefield into parts corresponding to different body
waves is one of the fundamental principles of the seismic reflection method
ref
c tion lec
r efl e tio
n

ti on mul
fr ac tipl
e
d if
 Introduction
Correlation and Stacking
The central point of wavefield decomposition is a procedure for correlation of
seismic events on a set of recorded traces
 Introduction

Correlation and Stacking

Successful stacking of recorded seismic traces is possible only after
successful correlation procedure i.e. to correct all traces in a CMP gather in
such a way that events will be « in phase » => flattening
offset
TIME
 Introduction
Correlation and Stacking
Stack is a convenient and efficient way to represent « zero-offset » section.
TIME CMP
Why CMP method works ?

t t

x x

2
x
t = t02 + 2
V
 Stacking chart
Receiver coordinate

v
v
CMP v +
v ++
v + ++ +
v +
v +
v +
+
Shot coordinate
Time Imaging

Local time correction equation is a key to successful time

imaging.
This equation should satisfy two conditions:
3. It should be valid for arbitrary media (be model independent);
4. It should be valid for arbitrary observation geometry.
x2
t = t 2
0 + 2
V
 Why CMP method fails ?

• Short offset approximation

• Non-hyperbolicity
• Low signal to noise ratio
• Coherent noise and interference
• Statics problem
• Amplitude and phase distortions
• Stretch
 Shifted hyperbola
Convex interface: RMS velocities are computed and stacking velocities are
estimated using a velocity analysis. Interval velocities computed by Dix formula are
nowhere near the actual velocities.

After de Bazelaire, 1988

 Why CMP method fails ?
Concave interface: All the NMO curves have inverse curvature

After de Bazelaire, 1988

 Why CMP method fails ?
These CMP records show that upward curvatures can actually be observed.
The corresponding stacking velocities are imaginary.

After de Bazelaire, 1988

 Shifted hyperbola
The most general equation giving the distance between
a circle, the wavefront, and the plane tangent to it (the
ground surface) is of the form

(t + t r ) 2 = (t0 + t r ) 2 + x 2 / Va2
t r = AM ' / V1 − AM / V2 is the difference
between the actual and vertical traveltime

This is the equation of a hyperbola with three variables.

The variables are t0 , the time from the center of
coordinates to the apex of the hyperbola, t r the time
from the center of coordinates to the center of the
hyperbola, and Va , the average velocity.

De Bazelaire, E., 1988, Normal moveout revised – Ingomogeneous media

and curved interfaces: Geophysics, 53, 143-157
 Shifted hyperbola

Making the variable change t p = t 0 − t r and after some algebraic

transformation, we obtain a shifted hyperbola moveout correction:

For small offsets the parameter V may be replaced by the near surface
velocity, resulting in a robust single-parameter correlation procedure. For a
given t p , the correction is a function of a trace distance to the source only.
Whatever the time t 0 , all the seismic events which can be stacked into
focus at this time are corrected simultaneously.
 Shifted hyperbola
If CMP gather is centered, there is stigmatism of the object and the image, and the
shifted hyperbola is perfect. Instead, NMO gives a very poor fit.

After de Bazelaire, 1988

 Shifted hyperbola
If CMP gather is not centered, the image point is smeared along the reflection
horizon. Shifted hyperbola does not handle this case but it is still better than NMO
up to 3000 m offset. Three term gives satisfactory results up to 4000 m.

After de Bazelaire, 1988

 Shifted hyperbola
For limited apertures, the shifted hyperbola generally yields equivalent
and often better results than the conventional equation.

After de Bazelaire, 1988

 Shifted hyperbola
All the events that assume the shape of a hyperbola receive equal treatment.
However the multiples have very different velocities. Clear where the reflections
stop and where the diffractions start.

After de Bazelaire, 1988

 Non-hyperbolic moveout
Tsvankin and Thompsen anisotropic moveout

After introducing we obtain:

 An offset dependent NMO velocity model
Non-hyperbolic behaviors can be modeled using the horizontal
velocity acceleration model
t ( x) = t0 + x 2 V 2 ( x) V (x) = V0 + ax 2
V0 where is the velocity at zero offset, and a is a horizontal velocity
acceleration parameter. Both parameters can be estimated from the data
For horizontally stratified model:

Taner, T., S. Treitel, M. Al-Chalabi, and S. Fomel, 2007, An offset dependent

NMO velocity model, EAGE meeting.
 Comparison of different CMP approximations
The Taylor series of the standard hyperbola, Alkalifa & Tsvankin,
Taner’s and de Bazelair’s. Each formula is expanded up to the 4th order:

Standard, Alkalifa’s and Taner contain the same second-order coefficients.

Shifted hyperbola depends on . Relationship between :

After Sabrina Ernst, 2006

Comparison of different CMP approximations
Synthetic model

Synthetic seismogram

After Sabrina Ernst, 2006

 Comparison of different CMP approximations

Traveltime could be adequately described by two-parameter equation

After Sabrina Ernst, 2006
Comparison of different CMP approximations
Synthetic model

Synthetic seismogram

After Sabrina Ernst, 2006

 Comparison of different CMP approximations

Anisotrophy can reduce the deviation of the reflection event from a standard hyperbola.
Compared to the standard hyperbola the two-parameter equations slightly improve the moveout
correction
After Sabrina Ernst, 2006
Comparison of different CMP approximations
Synthetic model

Synthetic seismogram

After Sabrina Ernst, 2006

 Comparison of different CMP approximations

Reflection events can split into two parts due to the complexity of the model

After Sabrina Ernst, 2006

Conclusion of Comparison of different CMP approximations

2>1
 Non-CMP oriented methods
The critical operation in the stacking procedure is time correction.
« Good » time correction formula should satisfy two conditions:
2) validity for arbitrary media and
3) validity for arbitrary observation geometry
In practice it implies that:
e) the time correction formula should be model independent
f) the formula is valid for arbitrary distribution of source-receiver pairs

• Multifocusing (MF) and Common Reflection Surface (CRS) present a

different approach to traveltime correction approximation
• These correction formulas are valid for arbitrary subsurface models and for
arbitrary source-receiver configurations
 Multifocusing method

Incidence angle Focusing point

∆X − ∆X +

Paraxial ray

Central (normal) ray

Gelchinsky, B., Berkovitch, A., and Keydar, S., 1999: Multifocusing homeomorphic
imaging - Part 1. Basic concepts and formulas. J. Appl. Geoph., 42(3,4):229-242.
 Multifocusing method
∆X − ∆X +

Traveltime correction

are the source and receiver offsets for a given central ray,

are the radii of curvature of the fictitious wavefront

 Multifocusing method
∆X − ∆X +

- are the radii of curvature of two fundamental wavefronts corresponding

to the normal (N) and NIP waves respectively.
Multifocusing method
Radius NIP wave
Multifocusing method
Radius N wave
 Multifocusing method
-is the focusing parameter, which defines the position of the focusing
point P on the normal ray
= 0 means that which implies that point P coincides with the centre
of curvature of the normal wave and corresponds to the case of coinciding source
and receiver (zero-offset configuration).

= 1 or –1 imply and correspond to the common shot or

common receiver configuration

= leads to and corresponds to the situation where the focusing

point P coincides with NIP. CRP configuration
 Multifocusing method

Single CMP

∆τ =
(R ) + 2
+ 2 R h sin β + h − R
+ 2 +
R± =
1
1±σ
σ
V0 ±
R N R NIP

+
(R )
− 2
−2 R −h sin β + h 2 − R −
σ =−
RNIP
V0 h sin β

RN= (plane reflector)

8

L2 cos 2 β
t = t0 − t p + t + 2
p V02

Shifted hyperbola of de Bazelaire, tp=2RNIP/V0

 Common Reflection Surface method

The CRS stacking operator is based on three wavefront attributes of two so-called
eigenwaves: one is obtained by placing a point source at R (NIP wave), the second is
connected to an exploding reflector (N wave)

Jäger, R., Mann, J., Höcht, G., and Hubral, P. (2001). Common-Reection-
Surface stack: image and attributes. Geophysics, 66(1):97.109.
 Common Reflection Surface method

This equation describes the reflection time along a ray reflected in the vicinity of
normal reflection point in the vicinity of surface position Xo. The offset between
source and receiver is denoted as X and the CMP between source and receiver Xm.
The angle is the emergence angle of the normal ray.
Single CMP Xm=Xo
 Common Reflection Surface method

The CRS operator approximates true reflection surface in the vicinity of Po.

CRS operator for known parameters Kirchhoff PSTM operator for known
velocity

After Jager et al., Geophysics, 2001

MF/CRS stacking chart
 MF-Gather MF-Gather
before moveout after moveout correction
correction

20,31
Stacked Section
Multifocusing Section
Emergence angle Section
NIP-wave Section
N-wave Section
Sigsbee model
 Sigsbee model

Zero Offset Section

 Sigsbee model

Multifocusing Stack
 MF/CRS advantages
•MF/CRS moveout correction can be applied to any trace if its source and receiver
are in the vicinity of the central point
•Stacking a large number of traces increases the stacking power.
•MF/CRS processing preserves dipping events and incorporates the key property of
the DMO transform
•MF/CRS processing makes it possible to recover dip-independent RMS velocities:

•Multifocusing moveout correction is « NMO stretch » free: it does not depend on

zero-time but only on three wavefield attributes
•Estimated MF/CRS parameters can be considered as a basis for wavefield
decomposition
 MF versus CRS
MF for one CMP

∆τ =
(R ) + 2
+ 2 R h sin β + h − R
+ 2 +
R± =
1
1±σ
σ
V0 ±
R N R NIP

+
(R )
− 2
h sin β + h − R −
− for one CMP 2
−2 RCRS
σ =−
RNIP
V0 h sin β
 MF versus CRS
Circular reflector - one CMP
CMP

RNIP=500m
V=2000m/s

R=500m
 MF versus CRS
Circular reflector - one CMP
1
Exact
MF
Hyperbolic(CRS)
0.9
Traveltime, sec

0.8

0.7

0.6

0.5

0.4
0 100 200 300 400 500 600 700 800
Half offset, m
 MF versus CRS
Offset (m)
-1000 0 1000 2000
Xo

CMP aperture
1
V1 = 2000 m / s
Depth (km)

3
R = 1000 m

4
 MF versus CRS

Time difference (s)

)
k m
et(
s f f
fo
ture (km)
CMP aper
a l
H
 MF versus CRS
Offset (m)
-1000 0 1000 2000
Xo

CMP aperture
1
V1 = 2000 m / s
Depth (km)

2
R = 100 m
3

4
 MF versus CRS

Time difference (s)

)
km(
set
f f
o
(km)
lf
a
ap e r t u r e
CMP
H
 Multifocusing and CFP

RNIP = RN ⇒ R + = R − = R

R 2 + 2 R∆ X s sin β + ∆ X s2 − R R 2 + 2 R∆ X r sin β + ∆ X r2 − R
∆τ = +
V0 V0
 Multifocusing and CFP

∆τ = ∆τ s + ∆τ r
∆τ s = ∆τ s ( R, β , ∆X + )−" focusing in emission"
∆τ r = ∆τ r ( R, β , ∆X − )−" focusing in detection"
 Benefits of MF/CRS
CURVED
REFLECTOR

S X0 R

C* Σ ∗

C Σ
Benefits of MF/CRS

CURVED REFLECTOR
Rugged Topography
Gurevich et al., 2002
 Benefits of MF/CRS
RUGGED TOPOGRAPHY
Benefits of MF/CRS
RUGGED TOPOGRAPHY
 Benefits of MF/CRS

AVO story

S CMP1 CMP2 R1 R2

σ1
σ2

Gurevich and Landa, 2002

 Benefits of MF/CRS

AVO story

CMP 170 CMP 190

β

σ=0.15

A Vp=2000m/s
σ=0.25
σ=0.35
Vp=2300m/s
B

σ=0.15
 Benefits of MF/CRS
AVO story
0.2
Within anomaly
Outside anomaly
0.15
R

0.1

0.05

0
0 5 10 15 20 25 30
Incidence Angle, degrees
Benefits of MF/CRS
AVO story
Benefits of MF/CRS
AVO Gradient Stack
 Implementation

•Requires for each time sample determination of three wavefield

• parameters:
•Interactive picking procedure similar to the velocity stacking analysis is
out of question: 1) it is computationally heavy and 2) requests 3-
dimensional display for each time sample
•Automation is required!
•Global optimization is required!
Conventional stack
MF stack
Angle

Data courtesy GII

NIP Radius

Data courtesy
GII
RMS Velocity

Data courtesy
GII
Migrated stack
Migrated MF stack
 LAND
DATA

Low fold (12)

 Co
St n
ac ve
k , n ti
fo on
ld a l
12

Data courtesy GII

 M
Fo F S
ld ta
12 ck,
x5

Data courtesy GII

 e
gl
An

Data courtesy
GII
 us
di
Ra
P
NI

Data courtesy GII

MARINE DATA

Fold 60
DMO stack

Data courtesy GII

MF stack

Data courtesy
GII
Angle

Data courtesy GII

RMS Velocity

Data courtesy GII

DMO stack

Data courtesy GII

MF stack

Data courtesy GII

Data courtesy Geomage
Data courtesy Geomage
Data courtesy Geomage
Data courtesy Geomage
Data courtesy Geomage
DMO stack

Data courtesy TEEC

CRS stack

Data courtesy TEEC

DMO stack (3D)

Data courtesy Petrologic

CRS stack (3D)

Data courtesy Petrologic

AVO Gradient Stack

Data courtesy GII

MF-AVO Gradient Stack

Data courtesy GII

AVO Gathers

DMO MF

Data courtesy GII

AVA Gathers

DMO MF

Data courtesy GII

3D CRS implementation (Landmark)
Angle estimation

Data courtesy Landmark

 Path integral imaging
The Red Queen shook her head,- You may call it
« nonsense » if you like, - she said, - but I’ve head
nonsenses, compared with which that would be as sensible
as a dictionary!

-Разве это чепуха? - сказала Королева и

затрясла головой. - Слыхала я такую
чепуху, рядом с которой эта разумна, как
толковый словарь!

Lewis Carroll. Through the looking

glass
 Path integral imaging

What do we need to achieve accurate subsurface image?

• An accurate (true, adequate) velocity model

• A fundamentally new imaging method that
does not require precise velocity information
• Imaging methods that are tested using
synthetic data with precise model are not
addressing the real-world problem
 Path integral imaging
STACK
Q(t 0 , x0 ; α ) = ∫ dh ∫ dtU (t , h)δ (t − τ ( x0 , t 0 , h; α ))

where U(t, h) is the recorded CDP gather for location x 0

h is the offset to be summed over the measurement aperture. The
quantity τ =τ( x 0 , t 0 , h; α) represents the time-integration
path/trajectory, which is parametrized by a parameter α

Offset
(i)

U (x,t)

P(x,t)

Optimal time curve

 Path integral imaging
STACK
The conventional zero-offset stack is obtained by optimizing for α, i.e.
QO (t 0 , x0 ) = Q(t 0 , x0 ; α 0 )

Offset (i)

U (x,t)

P(x,t
Fermats time curve
)

Non-Fermats time curves

 Path integral imaging
Path-summation stack

QW = ∫ dαw(α )Q(α )
Instead of stacking seismic data along only one time trajectory
corresponding to the Fermat path our construction involves
summation over all possible time trajectories.
 Path integral imaging
Classical mechanics
(t b , xb )

x(t )
(t a , x a )


x (t )
The classical path is singled out of all possible
paths as the one having the least action S

tb

S = ∫ dt L( x, x . , t )
ta

where L – is the Lagrangian

 Path integral imaging
Quantum mechanics

(t b , xb )

x(t )
(t a , x a )
Instead of only considering the classical
trajectory, consider every possible path between
a and b. Each path contributes to the total
amplitude. This
K (bamplitude
, a ) = ∑A(is
x (t ))
all paths
from a to b

where A is the contribution of each individual

path
and A( x(t )) = const * exp((i / h) / S ( x(t )))
 Path integral imaging

Stack
There are several choices for the weighting function for
QW = ∫ dαw(α )Q(α )
If we choose an oscillatory weighting function
Q F = ∫ dα exp(iβS (α ))Q (α )
S(α) is the signal semblance, then the operator can be considered as a
form of the Feynman’s path-integral.
For an exponential weighting function
QE = ∫ dα exp( βS (α ))Q(α )

we have the Einstein–Smoluchovsky path integral, which was first

introduced in the theory of Brownian motion.
A trivial choice is the Dirac-delta weighting function
QD = ∫ dαδ (α − α 0 )Q (α ) = Q (α 0 ) = Q0
 Path integral imaging

The path-integral stacks Q approach the classical limit Q for β → ∞. For

F O

the Feynman’s path-integral, this can be done by a stationary-phase

approximation under the assumptions
Q (α ) → 0 for α − α 0 → ∞, S (α 0 ) = 0, S (α 0 ) ≠ 0 , i.e.
' ''
 Path integral imaging
Path-integral weighting functions, exponential (thick) and oscillating
(thin). The dashed line corresponds to the optimal value α 0 used in
the classical limit.
 Path integral imaging

The path-summation imaging consists of

weighted summation along a representative
sample of all possible travel time curves (paths)
between the source and observation points
Path integral imaging

Zero-offset section
Path integral imaging

Stacked section
Path integral imaging

Path-summation section
Path integral imaging

Stacked section
Path integral imaging

Path-summation section
 Path integral imaging

Time migration

VO ( x) ≈ ∫ dξ ∫ dtU (t , ξ )δ (t − t d (ξ , x; α 0 ))
ξ is source-receiver configuration, x is (t0 , x)
U (t , ξ ) is the recorded data

The path-integral time migration

VW ( x) ≈ ∫ dαw(α , x) ∫ dξ ∫ dtU (t , ξ )δ (t − t d (ξ , x; α ))

t
where d represents all possible time trajectories w(α, x) denotes the
weighting factor

where p can be taken as a flatness index

 Path integral imaging
Path-integral time migration
Path integral imaging
Path integral imaging

PSTM
 Path integral imaging

Path-summation time migration

 Marine vs Land – A Heuristic Analogy

Marine Land
Classical Mechanics Quantum Mechanics

Offset Offset

Time
Time

Panos G. Kelamis et al., 2006, SEG

 New subtraction scheme
• Localized parabolic path-summation
Offset

L
N L
T ij M = ∑ Wi (∑ Tij / L)
i j

Calculate weighting
factor Wi using least-
squares
M
Ni
Time

Panos G. Kelamis et al., 2006, SEG

 Synthetic - Multiples
Input New method HR-Radon Radon FK

Panos G. Kelamis et al., 2006, SEG

 Synthetic - Primaries
Input New method HR-Radon Radon FK

Panos G. Kelamis et al., 2006, SEG

 Path summation imaging

Conventional stack Path summation stack

Keydar, Shtivelman, 2005, First Break

 Path summation imaging

Y
er
e cei v source
r
X

diffractor

Keydar, Shtivelman, 2005, First Break

 Path summation imaging

Vertical slices for one diffractor Vertical slices for 3 diffractors

Keydar, Shtivelman, 2005, First Break

 Path integral imaging

Conclusions
•Path-integral method allows subsurface structural
imaging without precise knowledge or selection of a
velocity model.
•Path-integral seismic imaging does not involve any
optimization or estimation of parameters.
•Instead, the image is constructed by summation over
many (ideally all) possible travel time trajectories.