You are on page 1of 119

Seismic Data Processing

Seismic Data Processing


1. Introduction to Seismic surveys

¾ The seismic method is the most important geophysical technique. Its predominance
is due to high accuracy, high resolution, and great penetration.

¾ The widespread use of seismic methods is principally in exploring for petroleum.

¾ Seismic methods are also important in groundwater searches and in civil engineering,
especially to measure the depth to bedrock in connection with the construction of
large buildings, dams, highways, and harbor surveys.

¾ Exploration seismology is an offspring of earthquake seismology. When an


earthquake occurs, the earth is fractured and the rocks on opposite sides of the
fracture move relative to one another. Such a rupture generates seismic waves that
travel outward from the fracture

¾ Seismologists use the data to deduce information about the nature of the rocks
through which the earthquake waves traveled.

¾ Exploration seismic methods are different from the earthquake seismology in that the
energy sources are controlled and movable, and the distances between the source
and the recording points are relatively small.

¾ Source: explosives and other energy sources


¾ Receiver: arrays of seismometers or geophones or hydrophones

¾ The basic technique of seismic exploration consists of generating seismic waves and
measuring the time required for the waves to travel from the sources to a series of
geophones, usually disposed along a straight line directed toward the source.

¾ Structural information is derived principally from paths that fall into two main
categories:

¾ headwave or refracted paths, in which the principal portion of the path is along the
interface between two rock layers.
¾ Reflected paths, in which the wave travels downward initially and at some point is
reflected back to the surface, and the overall path is essentially vertical.

Topics in Seismic Prospecting SNU 1


Seismic Data Processing

¾ For both types of path, the traveltimes depend on the physical properties of the
rocks and the attitudes of the beds

¾ The objective of seismic exploration is to deduce information about the rocks,


especially about the attitudes of the beds, from the observed arrival times and from
variations in amplitude, frequency, phase, and wave shape.

2. Basic theory for seismic waves

(1) Stress-Strain Relation


The equation of wave propagation in elastic solids is derived by using Hooke’s law
and Newton’s second law of motion.

a. Stress

¾ Stress is defined as __________ per __________________.

¾ Imagine an infinitesimally small volume around a point within a solid body with
dimensions (dx, dy, dz). (see Figure L-1)

9 Pij: i indicates the direction of the normal vector to the plane,


and j denotes the direction of the stress component.

9 Normal stress: indicates stress which is normal to the surface


9 Shear stress: indicates stress which is tangential to the surface

Topics in Seismic Prospecting SNU 2


Seismic Data Processing

9 What is pressure? Normal stress


9 When the force is neither parallel nor perpendicular to the element of area, it
can be resolved into components parallel and perpendicular to the element.

Î Any stress can be decomposed into normal and shearing components.

9 When the medium is in static equilibrium, the forces must be balanced, which
means the three stresses acting on one face must be equal and opposite to
the corresponding stresses shown on the opposite face.
9 To retain the solid volume in equilibrium, nine stress components are required.
⎡ Pxx Pxy Pxz ⎤
{Pij } = ⎢⎢Pyx Pyy Pyz ⎥

⎢ Pzx Pzy Pzz ⎥⎦

Î The diagonal element are the normal stress components and the off-
diagonal elements are the shear stress components.
9 As the dimensions of the volume surrounding a point inside a solid body are
made infinitesimally smaller, the sum of the moments of all surface forces
about any axis must vanish. Æ Pxy = Pyx , Pxz = Pzx , Pyz = Pzy .
9 Fluids cannot support shear stress. In a fluid medium, only one independent
stress component remains, normal stress( Pxx = Pyy = Pzz = −P ,
Pxx = Pyy = Pzz = 0 ): the hydrostatic pressure (compressive).

b. Strain

¾ When an elastic body is subjected to stresses, changes in shape and dimensions


occur. These changes are called strains.
¾ Stress induces deformation of solid bodies. Consider two points P and Q within a
solid body in Figure L-2. Subject to a stress field, the solid is deformed in some
manner and the particles at points P and Q are displaced to new locations P’
and Q’. The displacement components can be expressed by the relationship.
⎛ ∂u ∂u ∂u ⎞
⎛δ u ⎞ ⎜ ⎟⎛δ x ⎞
⎜ ⎟ ⎜ ∂x ∂y ∂z ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ∂v ∂v ∂v ⎟ ⎜ ⎟
⎜δ v ⎟ = ⎜ ⎟⎜δ y ⎟
⎜ ⎟ ⎜ ∂x ∂y ∂z ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ∂w ∂w ∂w ⎟ ⎜⎜ ⎟⎟
⎜δ w⎟ ⎜ ⎟ δz
⎝ ⎠ ∂z ⎠ ⎝ ⎠
⎝ ∂x ∂y

Topics in Seismic Prospecting SNU 3


Seismic Data Processing

Topics in Seismic Prospecting SNU 4


Seismic Data Processing

¾ The classification of deformation (see Figure L-3)


9 Figure L-3a: an extension in one direction as a result of a tensional stress. The
δu
fractional change in length in the x-direction is . For the infinitesimal
δx
volume, it is defined as the normal strain component.

- There are three normal strain components


∂u ∂v ∂w
exx = , eyy = , ezz =
∂x ∂y ∂z
- Dimension Æ dimensionless
- Positive Æ extension, Negative Æ contraction
- The strain induced by seismic waves is very small, usually around 10-6.
9 Figure L-3b: shearing, Figure L-3c: rotation, Figure L-3d: the combination of
shearing and rotation.
9 Expression for ξ and ζ:
∂u ∂w 1 ⎛ ∂u ∂w ⎞
2ξ = + ξ= ⎜ + ⎟
∂z ∂x 2 ⎝ ∂z ∂x ⎠
∂u ∂w 1 ⎛ ∂u ∂w ⎞
2ζ = − ζ = ⎜ − ⎟
∂z ∂x 2 ⎝ ∂z ∂x ⎠

∂u
ξ +ζ =
∂z
∂w
ξ −ζ =
∂x

9 Strain tensor

ξ is defined as a strain in the x-z plane


1 ⎛ ∂u ∂w ⎞
exz = ⎜ + ⎟ = ezx
2 ⎝ ∂z ∂x ⎠
In the same way,
1 ⎛ ∂u ∂v ⎞
exy = ⎜ + ⎟ = e yx
2 ⎝ ∂y ∂x ⎠
1 ⎛ ∂v ∂w ⎞
eyz = ⎜ + ⎟ = ezy
2 ⎝ ∂z ∂y ⎠
⎛ exx exy exz ⎞
⎜ ⎟
{eij } = ⎜ eyx e yy e yz ⎟
⎜ ezx ezy ezz ⎟⎠

Topics in Seismic Prospecting SNU 5


Seismic Data Processing

9 Rotation
ζ indicates rotation in the x-z plane
1 ⎛ ∂w ∂u ⎞
ζ = θ xz = ⎜ − ⎟ = −θ zx
2 ⎝ ∂x ∂z ⎠
1 ⎛ ∂w ∂u ⎞
θ xy = ⎜ − ⎟ = −θ yx
2 ⎝ ∂x ∂z ⎠
1 ⎛ ∂w ∂u ⎞
θ yz = ⎜ − ⎟ = −θ zy
2 ⎝ ∂x ∂z ⎠
θ xx = θ yy = θ zz = 0
9 Displacement tensor
⎛ ∂u ∂u ∂u ⎞ ⎛ ∂u ∂u ∂u ∂u ∂u ∂u ⎞
⎜ ⎟ ⎜ + + + ⎟
⎜ ∂x ∂y ∂z ⎟ ⎜ ∂x ∂x ∂y ∂y ∂z ∂z ⎟
⎜ ∂v ∂v ∂v ⎟ 1 ⎜ ∂v ∂v ∂v ∂v ∂v ∂v ⎟
T =⎜ ⎟= ⎜ + + + ⎟
⎜ ∂x ∂y ∂z ⎟ 2 ⎜ ∂x ∂x ∂y ∂y ∂z ∂z ⎟
⎜ ∂w ∂w ∂w ⎟ ⎜ ∂w ∂w ∂w ∂w ∂w ∂w ⎟
⎜ ⎟ ⎜ + + + ⎟
⎝ ∂x ∂y ∂z ⎠ ⎝ ∂x ∂x ∂y ∂y ∂z ∂z ⎠
⎛ ∂v ∂u ∂w ∂u ⎞
T =S−A ⎜ 0 − − ⎟
⎜ ∂x ∂y ∂x ∂z ⎟
S=
2
(T + T )
1 t
1 ⎜ ∂u ∂v
− ⎜ − 0
∂w ∂v ⎟
− ⎟
2 ⎜ ∂y ∂x ∂y ∂z ⎟
A = (T t − T )
1
⎜ ∂u ∂w ∂v ∂w ⎟
2 ⎜ − − 0 ⎟
⎝ ∂z ∂x ∂z ∂y ⎠

⎡ exx eyx ezx ⎤ ⎡ 0 ξ xy ξ xz ⎤


⎢ ⎥ ⎢ ⎥
S = ⎢ exy eyy ezy ⎥ : actual deformation A = ⎢ −ξ xy 0 ξ yz ⎥
⎢ exz e yz ezz ⎥⎦ ⎢ −ξ xz −ξ yz 0 ⎥⎦
⎣ ⎣
c. Stress-strain relationship
¾ Hooke’s law
For an elastic solid, Hooke’s law states that the strain at any point is directly
proportional to the stresses applied at that point.
¾ Young’s modulus and Poisson’s ratio
When the tensional principle stress Pxx is acting on a cylinder, the strain will be
exx , eyy , and ezz
Pxx = Eexx (E = Proportionality constant ⇒ Young’s modulus)
Poisson’s ratio is the ratio of the lateral contraction defined as the strain
component −e yy to longitudinal extension defined by the strain component
exx

Topics in Seismic Prospecting SNU 6


Seismic Data Processing

E E
eyy e Pxx = − eyy = − ezz
σ =− = − zz σ σ
exx exx
−σ Pxx = Eeyy = Eezz
In the same way, when the tensional principle stress Pyy is acting on a cylinder
Pyy = Eeyy Pzz = Eezz
−σ Pyy = Eexx = Eezz −σ Pzz = Eexx = Ee yy

¾ Under the assumption that Pxx, Pyy, and Pzz exist


Pxx − σ Pyy − σ Pzz = Eexx
−σ Pxx + Pyy − σ Pzz = Ee yy
−σ Pxx − σ Pyy + Pzz = Eezz
Add +σ Pxx − σ Pxx to the above equation
(1 + σ )Pxx − σ (Pxx + Pyy + Pzz ) = Eexx
(1 + σ )Pyy − σ (Pxx + Pyy + Pzz ) = Eeyy
(1 + σ )Pzz − σ (Pxx + Pyy + Pzz ) = Eezz
(1 − 2σ )(Pxx + Pyy + Pzz ) = E(exx + e yy + ezz )
※ Unstrained volume is δ xδ yδ z
Strained volume is (δ x + δ u )(δ y + δ v)(δ z + δ w)
The fractional change in volume, Δ
ΔV (δ x + δ u )(δ y + δ v)(δ z + δ w) − δ xδ yδ z
Δ= =
V δ xδ yδ z
∂u ∂v ∂w
Neglecting high-order terms ⇒ + + = exx + e yy + ezz = Δ
∂x ∂y ∂z

Pxx + Pyy + Pzz =
1 − 2σ
위 식에 대입하면
EΔσ EΔσ E
(1 + σ ) Pxx = + Eexx Pxx = + e
1 − 2σ (1 + σ )(1 − 2σ ) 1 + σ xx
EΔσ EΔσ E
(1 + σ ) Pyy = + Eeyy Pyy = + e
1 − 2σ (1 + σ )(1 − 2σ ) 1 + σ yy
EΔσ EΔσ E
(1 + σ ) Pzz = + Eezz Pzz = + e
1 − 2σ (1 + σ )(1 − 2σ ) 1 + σ zz
Pxx = λΔ + 2μ exx
Eσ E
Pyy = λΔ + 2μ eyy λ= μ=
(1 + σ )(1 − 2σ ) 2 (1 + σ )
Pzz = λΔ + 2μ ezz

Topics in Seismic Prospecting SNU 7


Seismic Data Processing

Shear stress components Pxy , Pxz and Pyz will cause shear strain
⇒ No normal strains
Pxy = 2μ exy , Pxz = 2 μ exz , Pyz = 2μ eyz
⎛ Pxx Pxy Pxz ⎞ ⎛1 0 0⎞ ⎛ exx exy exz ⎞
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ Pxy Pyy Pyz ⎟ = λΔ ⎜ 0 1 0 ⎟ + 2 μ ⎜ exy eyy eyz ⎟
⎜ Pxz Pzz ⎟⎠ ⎜0 0 1⎟ ⎜ exz ezz ⎟⎠
⎝ Pyz ⎝ ⎠ ⎝ eyz
⇒ holds for homogeneous, isotropic elastic media
We can rewrite
⎛ Pxx ⎞ ⎛ λ + 2 μ λ λ 0 0 0 ⎞ ⎛ exx ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Pyy ⎟ ⎜ λ λ + 2μ λ 0 0 0 ⎟ ⎜ eyy ⎟
⎜ Pzz ⎟ ⎜ λ λ λ + 2μ 0 0 0 ⎟ ⎜ ezz ⎟
⎜ ⎟=⎜ ⎟⎜ ⎟
⎜ Pxy ⎟ ⎜ 0 0 0 2μ 0 0 ⎟ ⎜ exy ⎟
⎜ Pxz ⎟ ⎜ 0 0 0 0 2μ 0 ⎟ ⎜ exz ⎟
⎜ ⎟ ⎜⎜ ⎟⎜ ⎟
⎜P ⎟
⎝ yz ⎠ ⎝ 0 0 0 0 0 2μ ⎟⎠ ⎜⎝ e yz ⎟⎠
Only 2 constants λ and μ are required

¾ The generalized Hooke’s law


⎛ Pxx ⎞ ⎛ C11 C12 C13 C14 C15 C16 ⎞ ⎛ exx ⎞
⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ Pyy ⎟ ⎜ C21 C22 C23 C24 C25 C26 ⎟ ⎜ eyy ⎟
⎜ Pzz ⎟ ⎜ C31 C32 C33 C34 C35 C36 ⎟ ⎜ ezz ⎟
⎜ ⎟=⎜ ⎟⎜ ⎟
⎜ Pxy ⎟ ⎜ C41 C42 C43 C44 C45 C46 ⎟ ⎜ exy ⎟
⎜ Pxz ⎟ ⎜ C C52 C53 C54 C55 C56 ⎟ ⎜ exz ⎟
⎜ ⎟ ⎜⎜ 51 ⎟⎜ ⎟
⎜P ⎟ C
⎝ yz ⎠ ⎝ 61 C62 C63 C64 C65 C66 ⎟⎠ ⎜⎝ e yz ⎟⎠

d. Elastic wave equation in homogeneous, isotropic, and elastic solids

¾ Until now, we only discuss a medium in static equilibrium. Let’s consider what
happens when the stresses are not in equilibrium. Neglect body forces such as
gravity.
¾ Stress gradient will cause displacements
Total Force for the z axis
∂Pzz ∂P
ΔxΔy (Pzz + Δz ) − Pzz ΔxΔy = zz ΔxΔyΔz
∂z ∂z
∂P ∂Pyz ∂Pzz
( xz + + )ΔV
∂x ∂y ∂z

Topics in Seismic Prospecting SNU 8


Seismic Data Processing

¾ From Newton’s second law of motion, the displacement component u is


expressed by

F = ma
∂P ∂Pyz ∂Pzz ∂ 2W
( xz + + )ΔV = ρΔV( 2 )
∂x ∂y ∂z ∂t
∂ u ∂P
2
∂P ∂P ∂ 2 w ∂P ∂P ∂P
ρ 2 = xx + xy + xz ρ 2 = xz + yz + zz
∂t ∂x ∂y ∂z ∂t ∂x ∂y ∂z
∂ 2 v ∂P ∂P ∂P
ρ 2 = xy + yy + yz
∂t ∂x ∂y ∂z

¾ Substitute Hooke’s law into the above equation, then

Pxx = λΔ + 2 μ exx
Pxy = 2 μ exy Pxz = 2 μ exz

∂ 2u ∂Δ ∂e ∂eyy ∂e
ρ =λ + 2 μ xx + 2 μ + 2 μ zz
∂t 2
∂x ∂x ∂y ∂z
∂u ∂v ∂w ∂u 1 ∂v ∂u 1 ∂w ∂u
Δ= + + exx = exy = ( + ) exz = ( + )
∂x ∂y ∂z ∂x 2 ∂x ∂y 2 ∂x ∂z
∂ 2u ∂ ⎛ ∂u ∂v ∂w ⎞ ∂ 2u ⎛ ∂ 2 v ∂ 2u ⎞ ⎛ ∂ 2 w ∂ 2u ⎞
ρ 2 = λ ⎜ + + ⎟ + 2μ 2 + 2μ ⎜ + 2 ⎟ + 2μ ⎜ + 2⎟
∂t ∂x ⎝ ∂x ∂y ∂z ⎠ ∂x ⎝ ∂y∂x ∂y ⎠ ⎝ ∂z∂x ∂z ⎠
∂ ⎛ ∂u ∂v ∂w ⎞ ⎛ ∂ 2 u ∂ 2u ∂ 2u ⎞
= (λ + μ ) ⎜ + + ⎟ + μ ⎜ 2+ 2+ 2⎟
∂x ⎝ ∂x ∂y ∂z ⎠ ⎝ ∂x ∂y ∂z ⎠

∂ 2u ∂Δ
ρ = (λ + μ ) + μ∇ 2u
∂t 2
∂x
∂v2
∂Δ
ρ 2 = (λ + μ ) + μ∇ 2v
∂t ∂y
∂2w ∂Δ
ρ = (λ + μ ) + μ∇ 2 w
∂t 2
∂z
Displacement vector → u = (u , v, w)
∂ 2u
ρ = (λ + μ )∇Δ + μΔ 2u →elastic wave equations
∂t 2

Topics in Seismic Prospecting SNU 9


Seismic Data Processing

e. P- and S-waves

¾ From the above elastic wave equations, we can derive the equations for P and S
waves

¾ For P-waves
9 Let’s differentiate the above equations with respect to x, y, or z.
∂ ⎛ ∂ 2u ⎞ ∂ 2Δ ∂ ⎛ ∂ 2 u ∂ 2u ∂ 2u ⎞
ρ ⎜ 2 ⎟ = (λ + μ ) 2 + μ ⎜ 2 + 2 + 2 ⎟
∂x ⎝ ∂t ⎠ ∂x ∂x ⎝ ∂x ∂y ∂z ⎠
∂ ⎛ ∂ 2u ⎞ ∂ 2Δ ∂ ⎛ ∂ 2v ∂ 2v ∂ 2v ⎞
ρ ⎜ ⎟ = ( λ + μ ) + μ ⎜ + + ⎟
∂y ⎝ ∂t 2 ⎠ ∂y 2 ∂y ⎝ ∂x 2 ∂y 2 ∂z 2 ⎠
∂ ⎛ ∂ 2u ⎞ ∂ 2Δ ∂ ⎛ ∂2w ∂2w ∂2w ⎞
ρ ⎜ 2 ⎟ = (λ + μ ) 2 + μ ⎜ 2 + 2 + 2 ⎟
∂z ⎝ ∂t ⎠ ∂z ∂z ⎝ ∂x ∂y ∂z ⎠
∂ 2u ⎛ ∂u ∂v ∂w ⎞ ⎛ ∂2Δ ∂2Δ ∂2Δ ⎞ ∂ 2 ⎛ ∂u ∂v ∂w ⎞
ρ ⎜ + + ⎟ = ( λ + μ ) ⎜ 2 + + ⎟ + μ ⎜ + + ⎟
∂t 2 ⎝ ∂x ∂y ∂z ⎠ ⎝ ∂x ∂y 2 ∂z 2 ⎠ ∂x 2 ⎝ ∂x ∂y ∂z ⎠
∂ 2 ⎛ ∂u ∂v ∂w ⎞ ∂ 2 ⎛ ∂u ∂v ∂w ⎞
+μ 2 ⎜ + + ⎟+μ 2 ⎜ + + ⎟
∂y ⎝ ∂x ∂y ∂z ⎠ ∂z ⎝ ∂x ∂y ∂z ⎠
∂2Δ
ρ = ( λ + 2μ ) ∇ 2 Δ Δ ⇒ cubical dilation
∂t 2

= ∇ ⋅u
∂ 2

2 (
ρ ∇ ⋅ u ) = ( λ + μ ) ∇ 2 + μ∇ 2 ( ∇ ⋅ u ) : compressional waves, dilatational waves
∂t
λ + 2μ
α=
ρ

¾ For S-waves

∂ ⎛ ∂ 2v ⎞ ∂ ⎛ ∂2w ⎞
ρ ⎜ ⎟ , ρ ⎜ ⎟ 를 구하면
∂z ⎝ ∂t 2 ⎠ ∂y ⎝ ∂t 2 ⎠
∂ ⎛ ∂ 2v ⎞ ∂2Δ ∂
ρ ⎜ 2 ⎟ = (λ + μ ) + μ ( ∇2v )
∂z ⎝ ∂t ⎠ ∂z∂y ∂z
∂ ⎛ ∂2w ⎞ ∂2Δ ∂
ρ ⎜ 2 ⎟ = ( λ + μ ) + μ ( ∇2 w)
∂y ⎝ ∂t ⎠ ∂y∂z ∂y
∂ 2 w ⎛ ∂w ∂v ⎞ ⎛ ∂w ∂v ⎞ 1 ⎛ ∂v ∂w ⎞
ρ 2 ⎜
− ⎟ = μ∇ 2 ⎜ − ⎟ θ yz = ⎜ − ⎟
∂t ⎝ ∂y ∂z ⎠ ⎝ ∂y ∂z ⎠ 2 ⎝ ∂z ∂y ⎠
∂ 2θ yz
ρ = μ∇ 2θ yz
∂t 2

Topics in Seismic Prospecting SNU 10


Seismic Data Processing

In a similar way,
∂ 2θ xz
ρ = μ∇ 2θ xz
∂t 2

∂ 2θ xy
ρ 2 = μ∇ 2θ xy
∂t
θ = (θ yz ,θ xz , θ xy )
∂ 2θ μ
ρ = μ∇ 2θ : shear wave, rotational wave β=
∂t 2
ρ
1
θ = ∇×u
2
∂u2
ρ 2 = ( λ + μ ) ∇ × ∇Δ + μ∇ 2 ( ∇ × u )
∂t
∇ × ∇Δ = 0
E
β μ 2 (1 + σ ) 1 − 2σ
= = =
α λ + 2μ σ E + E (1 − 2σ ) 2(1 − σ )
(1 + σ )(1 − 2σ )
0 < σ ≤ 0.5 For normal rocks σ ≈ 0.25
For fluid σ = 0.5 : β = 0

f. Acoustic wave equation


In fluid, the acoustic wave equation is induced by hydrostatic stress
⇒ No shear stress
− P = Pxx = λΔ + 2μ exx
− P = Pyy = λΔ + 2μ e yy
− P = Pzz = λΔ + 2μ ezz
−3P = 3λΔ + 2μΔ
−3P = ( 3λ + 2 μ ) Δ
The ratio of the hydrostatic stress P to volumetric strain Δ is called
incompressibility K
⎛ 3λ + 2 μ ⎞ 2
K =⎜ ⎟=λ+ μ
⎝ 3 ⎠ 3
4
K+ μ
λ + 2μ 3
α= =
ρ ρ
For fluids μ = 0 K = λ
∂2P
ρ = K ∇2P
∂t 2

Topics in Seismic Prospecting SNU 11


Seismic Data Processing

K
Pressure wave field travels with a speed
ρ

We can also derive acoustic wave equation using Newton’s second law
F = ma force = − pressure * area
ma = − pressure * area
m = ρ V divide both sides by V
ρ a = − pressure gradient

Velocity vector V = (V , V , V )
x y z

∂Vx ∂P ∂V ∂P ∂V ∂P
ρ =− ρ y =− ρ z =−
∂t ∂x ∂t ∂y ∂t ∂z
* Velocity Vx +Δx is greater than Vx , flow will diverge, which leads to volume
expansion, pressure drop
* The amount of the pressure drop is in proportion to a property of the fluid
called its incompressibility

Pressure drop = (incompressibility) * (divergence of velocity)


∂P ∂V
− = K x for 1-D
∂t ∂x
∂P ⎛ ∂V ∂Vy ∂Vz ⎞
− =K⎜ x + + ⎟ for 3-D
∂t ⎝ ∂x ∂y ∂z ⎠
Take the derivative of both sides with respect to t
∂ 2 P ⎡ ⎛ ∂Vx ∂Vy ∂Vz ⎞⎤
− = ⎢K ⎜ + + ⎟⎥
∂t 2 ⎣ ⎝ ∂x ∂y ∂z ⎠⎦
⎡ ∂ 2Vx ∂ 2Vy ∂ 2Vz ⎤
=K⎢ + + ⎥ interchange the differential operator
⎢⎣ ∂t ∂x ∂t ∂y ∂t ∂z ⎥⎦
⎡ ∂ ⎛ ∂V ⎞ ∂ ⎛ ∂V ⎞ ∂ ⎛ ∂V ⎞ ⎤
= K ⎢ ⎜ x ⎟ + ⎜ y ⎟ + ⎜ z ⎟⎥
⎣ ∂x ⎝ ∂t ⎠ ∂y ⎝ ∂t ⎠ ∂z ⎝ ∂t ⎠ ⎦
⎡ ∂ ⎛ 1 ∂P ⎞ ∂ ⎛ 1 ∂P ⎞ ∂ ⎛ 1 ∂P ⎞ ⎤
= K ⎢ ⎜− ⎟+ ⎜− ⎟+ ⎜− ⎟⎥
⎣ ∂x ⎝ ρ ∂t ⎠ ∂y ⎝ ρ ∂t ⎠ ∂z ⎝ ρ ∂t ⎠ ⎦

1 ∂ 2 P ∂ ⎛ 1 ∂P ⎞ ∂ ⎛ 1 ∂P ⎞ ∂ ⎛ 1 ∂P ⎞
= ⎜ ⎟+ ⎜ ⎟+ ⎜ ⎟
K ∂t 2 ∂x ⎝ ρ ∂t ⎠ ∂y ⎝ ρ ∂t ⎠ ∂z ⎝ ρ ∂t ⎠

Topics in Seismic Prospecting SNU 12


Seismic Data Processing

3. What kinds of waves are there?

- Seismic waves travel away from any seismic source at speeds determined by elastic
moduli and the densities of the media through which they pass

¾ Body waves

9 Body waves pass through the bulk of a medium are known as body waves
9 In unbounded homogeneous isotropic media, only body waves exist.
9 Two types of body waves can travel through an elastic medium.

9 P-wave: They arrive at the stations first after earthquake occurs.


- Primary, longitudinal, dilatational, irrotational, or compressional waves (for
example, sound waves) sound waves are a representative of P-wave.
- Material particles oscillate about fixed points in the direction of wave
propagation by compressional and dilatational strain

It shows particle motion and wave propagation


- Velocity
4
K+ μ
λ + 2μ 3 2
In solid: α= = K =λ+ μ
ρ ρ 3

K
In fluid, α= λ=K
ρ

9 S-wave:
This wave arrives at the station after P-waves.
- Secondary, transverse, rotational, or shear waves.
- Particle motion is at right-angles to the direction of wave propagation and
occurs by pure shear strain

Topics in Seismic Prospecting SNU 13


Seismic Data Processing

- When particle motion is confined to one plane only, the S-wave is said to
plane-polarized
- SH and SV Shear wave polarized in horizontal plane
in vertical plane
- Velocity

μ
In solid: β=
ρ

In fluid: β =0

- All the frequencies contained within body waves travel through a given
material at the same velocity. Æ not dispersive.

¾ Surface waves
9 Surface waves: waves confined to the interfaces between media with contrasting
elastic properties, particularly the ground surface, are called surface waves

9 In an infinite homogeneous isotropic medium, only P and S wave exist. However,


when the medium does not extend to infinity in all directions, other types of
waves can be generated. They are called surface waves.
9 The waves do not penetrate deep into subsurface media. They are confined to
the interfaces.
They arrive after S-waves.
9 There are two types of surface waves: Rayleigh and Love waves: The waves are
named after the scientists who first discover the waves.
9 Large amplitude and low frequency waves are their main features.
9 Rayleigh waves
- Are generated by the combination of P and SV waves.
- Travel along the free surface of the Earth with amplitudes that decrease
exponentially with depth

Topics in Seismic Prospecting SNU 14


Seismic Data Processing

- Particle motion is in a retrograde elliptical sense in a vertical plane with


respect to the surface

particle motion and


wave propagation

- The velocity of Rayleigh waves depends upon the elastic constants near the
surface and is always less than the S wave velocity β. When the Poisson’s
ratio is 0.25, the Rayleigh wave velocity is 0.92β

- Because the elastic constants change with depth, the velocity of Rayleigh
waves varies with wavelength. Æ A variation of velocity with wavelength (or
frequency) is called dispersion.

- Rayleigh waves are dispersive in layered media, whereas they are not
dispersive in semi-infinite homogeneous media.

- Rayleigh waves manifest themselves normally as large-amplitude, low-


frequency waves called ground roll which can mask reflections on a seismic
record

- Rayleigh waves are used in research on assessment of stability of structure


such as dam and road (Spectral analysis of the surface waves: SASW). The
SASW employs the dispersive features of Rayleigh waves. MASW.

Topics in Seismic Prospecting SNU 15


Seismic Data Processing

9 Love waves

- Love waves occur only where a medium with a low S-wave velocity overlies
a half space with a higher S-wave velocity

- Velocities are intermediate between the S-wave velocity at the surface and
that in deeper layers. v1 < v2 < v2

- Particle motion is at right-angles to the direction of wave propagation but


parallel to the surface SH waves

- These are polarized like shear waves

- Energy sources used in seismic work do not generate Love waves to a


significant degree and they are not important in ordinary seismic
exploration

9 Stoneley waves

- They propagate along the interfaces.

9 Guided waves: are encountered in some applications, which are confined to


particular thin bands sandwiched between layers with higher seismic velocities by
total internal reflection.

(For example)
Channel or seam waves, which propagate along coal seams
Tube waves, which travel up and down fluid-filled boreholes

Topics in Seismic Prospecting SNU 16


Seismic Data Processing

4. Seismic wave propagation

9 Waves classified according to propagation aspects


9 Direct waves
9 Reflected waves
9 Refracted waves (Head waves)
9 Scattered waves: When waves bump into a small object whose size is smaller
than the wavelength of the waves, some of the energies are diffracted.
9 Diffracted waves

9 Wavefront, Huygens’ principle, Raypath

All the points of the wavefront act as a new


secondary source for new wavefronts.

Ray path does not have any physical meaning.


But when we introduce raypath, we can explain
wave propagation more easily.

9 Acoustic wave propagation

1. direct wave

Topics in Seismic Prospecting SNU 17


Seismic Data Processing

1. direct wave

2. reflection

3. refraction at critical

4. refraction

wave front
and raypath

Topics in Seismic Prospecting SNU 18


Seismic Data Processing

Compute traveltime direct, refracted, and reflected waves

x
tdirect =
v1

AB BC CD
trefr = + +
v1 v2 v
2h x − 2h tan θ c
= +
v1 cos θ c v2
x 2h 2h sin θ c
= + −
v2 v1 cos θ c v2 cos θ c

SC + CR IR
trefl = =
v1 v1
x 2 + 4h 2
=
v1

Topics in Seismic Prospecting SNU 19


Seismic Data Processing

1. direct

2. reflection

3. Head waves

Topics in Seismic Prospecting SNU 20


Seismic Data Processing

¾ Diffraction

¾ For normal incidence


9 The total energy of the transmitted and reflected rays must equal the energy of
the incident ray.
9 The relative proportions of energy transmitted
and reflected are determined by the contrast in
acoustic impedance Z across the interface.
9 The acoustic impedance of a rock is defined as
ρv
9 Reflection coefficient from Zoeppritz equations is
ρ 2 v2 − ρ1v1
R=
ρ 2v2 + ρ1v1
ρ1v1
9 Transmission coefficient is T=
ρ 2v2 + ρ1v1

Topics in Seismic Prospecting SNU 21


Seismic Data Processing

¾ For oblique incidence

9 When a P-wave ray is obliquely incident on an interface of acoustic impedance


contrast, reflected and transmitted P-wave rays are generated as in the case of
normal incidence.
9 Additionally, some of the incident compressional energy is converted into
reflected and transmitted S-wave rays that are polarized in a vertical plane.
9 Zoeppritz’s equations show that the amplitudes of the four phases are a function
of the angle of incidence.
9 The converted rays may attain a significant magnitude at large angles of
incidence.

9 Elastic wave propagation

1. direct S-waves

2. P-direct

Topics in Seismic Prospecting SNU 22


Seismic Data Processing

1. Conical waves P-S coupling

2. direct S-waves

3. direct P-waves

1. reflected PP

2. reflected PS

3. Head waves

4. refracted PS

5. refracted PP

1. direct P

2. PP reflection

3. Rayleigh waves

4. PS reflection or SP

5. SS reflection

Topics in Seismic Prospecting SNU 23


Seismic Data Processing

9 Attenuation of seismic energy along ray paths

9 Geometrical spreading (spherical divergence)


- As a seismic pulse propagates in a homogeneous material, the original
energy E transmitted outwards from the source becomes distributed over a
spherical shell, the wavefront, of expanding radius.
- For the radius of the wavefront r, the amount of energy contained within a
E
unit area of the shell is . (I: is the intensity of energy)
4π r 2
- With increasing distance along a ray path, the energy contained in the ray
falls off as
r 2 due to the effect of the geometrical spreading of the energy.
- Wave amplitude falls off as r .

9 Intrinsic absorption
- As the wave passes through the medium, the elastic energy associated with
the wave motion is gradually absorbed by the medium and reappears
ultimately in the form of heat. Æ absorption
- Internal friction is involved.
- The measurement of absorption is very difficult. Absorption varies with
frequency
- The loss of energy by absorption appears to be exponential with distance
⎛r ⎞
for elastic waves in rocks A = A0 ⎜ 0 ⎟ e− q ( r − r0 )
⎝r⎠
q : absorption coefficient
A and A0 are the amplitudes at a and a0

- Losses by spreading are more important than losses by absorption for low
frequencies and short distances.
- As the frequency and distance increases, Absorption losses increase more
rapidly than spreading losses.
- The more rapid loss of higher frequencies results in change of wave shape
with distance.

Topics in Seismic Prospecting SNU 24


Seismic Data Processing

5. Seismic Survey
(1) Sources
Dynamite, Vibroseis, and Hammer for land survey
Air gun, Sleevegun, Aquapulse, Sparker, and Boomer for marine survey

(2) Receivers
Geophone for land survey
Hydrophone for marine survey

Topics in Seismic Prospecting SNU 25


Seismic Data Processing

Topics in Seismic Prospecting SNU 26


Seismic Data Processing

※ Vibroseis source passes energy into the ground for 7 s or more. A vibrator actuates a
steel plate pressed firmly against the ground. The output wavetrain consists of a sine
wave whose frequency increases continuously from 6 to 50 Hz during the 7 to 21 s.
Æ This is called sweep signal. By cross-correlating sweep signal with field data, we
can obtain signals. The frequency of signals varies gradually

sweep signal generated by vibroseis

When we use vibroseis for a seismic source


We may obtain the undistinguishable seismogram

Topics in Seismic Prospecting SNU 27


Seismic Data Processing

Topics in Seismic Prospecting SNU 28


Seismic Data Processing

6. Fundamentals of Signal Processing

- A seismic trace represents a seismic wavefield recorded at a receiver location.


- The digital form of a seismic trace is a time series which can be completely described
as a discrete sum of a number of sinusoids – each with a unique peak amplitude,
frequency, and a phase-lag.
- The analysis of a seismic trace into its sinusoidal components is achieved by the
forward Fourier transform. Conversely, the synthesis of a seismic trace from the
individual sinusoidal components is achieved by the inverse Fourier transform.

(1) Fourier Transform

a. Fourier Series

Topics in Seismic Prospecting SNU 29


Seismic Data Processing

b. The 1-D Fourier transform

9 Given a continuous function x ( t ) of a single variable t, its Fourier transform


is defined by

X ( w) = ∫ x(t )e − iωt dt
−∞

9 The Fourier transform is reversible; that is, given X (ω ) , the corresponding


time function is
1 ∞
x(t ) =
2π ∫
−∞
X (ω )eiωt dt
1 1 1
can be divided into two
2π 2π 2π
1
Sometimes is neglected for convenience.

9 Amplitude and Phase
X (ω ) is a complex
X (ω ) = R (ω ) + iI (ω ) = Aω eiφ (ω ) A(ω ) = R (ω ) 2 + I (ω ) 2
R (ω ) = A(ω ) cos φ (ω ) I (ω )
φ (ω ) = tan −1
I (ω ) = A(ω ) sin φ (ω ) R (ω )
9 Sign: In 2D Fourier transform, we should consider time and space
The sign of the exponent when the variable is time is opposite to that of the
exponent when the variable is space. For example, negative sign is used for
time, whereas positive sign is used for space.
c. Nyquist frequency (folding frequency)

9 Nyquist frequency is defined as the highest frequency that can be restored

Topics in Seismic Prospecting SNU 30


Seismic Data Processing

accurately once a continuous signal is digitized.


9 Nyquist frequency is the highest frequency that we obtain in Fourier
transform of the discrete data.
9 Nyquist frequency is determined by the sampling interval
1
fN =
2Δt
9 A seismic signal is a continuous time function, but in digital recording, the
continuous (analog) seismic signal is sampled with a sampling rate.
9 From sampled signal, we can reconstruct analog signal.
9 When digitized, a continuous analog signal loses frequencies above the
Nyquist frequency. Frequencies higher than the Nyquist frequency are folded.

Topics in Seismic Prospecting SNU 31


Seismic Data Processing

Topics in Seismic Prospecting SNU 32


Seismic Data Processing

d. The discrete Fourier transform

9 A couple of relationships
1
Δt & f N ( f max ) : Nyquist frequency is determined by f N =
2Δt
1
Δt =
2 f max
Δf & Tmax

9 The discrete Fourier transform


N −1 ⎛ n ⎞
1 ⎛ n ⎞ i 2π ( k Δt )⎜⎝ N Δt ⎟⎠
x(k Δt ) =
N

n =0
X ⎜ ⎟e
⎝ N Δt ⎠
N −1 i 2π kn
1 ⎛ n ⎞ N
=
N

n =0
X ⎜ ⎟
⎝ N Δt ⎠
e

9 For negative frequencies


X (ω ) = ∑ x(ti )eiωti
= ∑ x(ti ) [ cos(ωti ) + i sin(ωti ) ]
X (−ω ) = ∑ x(ti )e −iωti
= ∑ x(ti ) [ cos(ωti ) − i sin(ωti ) ] = X * (ω )

Real part is even function


Imaginary part is odd function

Topics in Seismic Prospecting SNU 33


Seismic Data Processing

e. Fourier Transform Pairs for several operators

¾ Cross-correlation (상호상관)

9 Seismic data processing often requires measurement of the similarity or time


alignment of two traces. Cross-correlation is used to measure the similarity
between two different functions.
9 Time-domain expression:
∞ ∞
φxz (τ ) = ∫ x(t ) z (t + τ )dt = ∫ x(t + τ ) z (t )dt = φzx (−τ )
−∞ −∞

9 Table 1-7 shows the crosscorrelation of wavelet 1 (2,1,-1,0,0) with wavelet 2


(0,0,2,1,-1)
9 The crosscorrelation of wavelet 2 (0,0,2,1,-1) with wavelet 1(2,1,-1,0,0) is
different from the crosscorrelation of wavelet 1 with wavelet 2. The output
depends on which array is fixed and which one is moved.

Topics in Seismic Prospecting SNU 34


Seismic Data Processing

9 Fourier transform of the crosscorrelation



φxz (τ ) = ∫ x(t ) z (t + τ )dt
−∞
∞ ∞
Φ xz (τ ) = ∫ ∫ x(t ) z (t + τ )dt e − iωt dt
−∞ −∞

Let’s interchange the integral variable

Φ xz (τ ) = ∫ x(t ) ⎡ ∫ z (t + τ )e − iωt dt ⎤ dt
∞ ∞
s = t +τ
−∞ ⎣⎢ −∞ ⎦⎥
⎛ X (ω ) = ∞ x(t )e− iωt dt ⎞
∞ ∞
= ∫ x(t ) ∫ z ( s )e − iω ( s −t ) dsdt ⎜ ∫−∞ ⎟
−∞ −∞ ⎜ * ∞ ⎟
∞ ∞ ⎜ X (ω ) = ∫ x(t )e dt ⎟
iωt
⎝ ⎠
= ∫ x(t )eiωt dt ∫ z ( s )e −iω s ds = X * (ω ) Z (ω ) −∞

−∞ −∞

¾ Autocorrelation (자기상관)

9 Autocorrelation is defined by the crosscorrelation of a time series with itself.



9 Time-domain expression: φxx (t ) = ∫ x(t ) x(t + τ )dt
−∞

Fourier transform of the autocorrelation: Φ xx (ω ) = X (ω ) X (ω ) = X (ω )


* 2
9

¾ Convolution: is a mathematical operation defining the change of shape of a wave


form resulting from its passage through a filter

9 The response of the reflectivity sequence to the source wavelet is obtained by


convolving the two series.
9 The convolution of two different functions is obtained by folding (reversing)

Topics in Seismic Prospecting SNU 35


Seismic Data Processing

one function and moving it



9 Time-domain expression: y (t ) = ∫−∞
x(t )h(τ − t )dt
9 Convolution is commutative – it does not matter which array is fixed and
which one is moved. The output is the same.

9 Fourier transform of the convolution



y (τ ) = ∫ x(t )h(τ − t )dt
−∞

t →τ + s

∫−∞
x(τ + s )h(− s )ds = H (ω ) X (ω )

x(t ) = ∫ X (ω )eiωt dt
−∞

x(t ) ⇔ X (ω )

x(−t ) = ∫ X (ω )e − iωt d ω = ∫ X * (ω )eiωt dω
−∞

x(−t ) ⇔ X * (ω )

Topics in Seismic Prospecting SNU 36


Seismic Data Processing

f. Frequency Filtering

9 Frequency-domain filtering is performed by multiplying the original function


with a filter operator. In the time domain, filtered data are obtained by the
convolution of the original function with a time-domain filter operator.

9 The classification of frequency filtering


Band-pass, band-reject, high-pass (low-cut), or low-pass (high-cut), and notch
filters.

9 When we design a filter operator, we should design a filter operator so that


Gibb’s phenomenon can be circumvented.

Topics in Seismic Prospecting SNU 37


Seismic Data Processing

Homework (need to be clarified)


1. Find the FFT results for boxcars. You can adjust the slope of the boxcars to
minimize the Gibb’s phenomenon.
2. Make the first derivative of gauss function whose maximum frequency is
100 Hz. Resample the curve with a sampling rate of 0.05 s. Perform FFT for
the resampled data, and then augment the data by zero padding for the
frequency ranges lower than 100 Hz. Perform IFT for the augmented data.
By doing so, you can check the relationship between the sampling rate
and the Nyquist frequency.
3. Provide some FFT and IFT examples for several functions such as sine,
cosine, and delta functions.

Topics in Seismic Prospecting SNU 38


Seismic Data Processing

Topics in Seismic Prospecting SNU 39


Seismic Data Processing

7. Seismic data acquisition

– The fundamental purpose of seismic surveys are accurately to record the ground motion
caused by a known source on a known location.
– The record of ground motion with time constitutes a seismogram.
– The essential instrumental requirements are to
∙ Generate a seismic pulse with a suitable source
∙ Detect the seismic waves in the ground with a suitable transducer
∙ Record and display the seismic waveforms on a suitable seismograph.

¾ Seismic transducers

9 Conversion of the ground motion to an electrical signal requires a transducer


which is sensitive to some component of the ground motion, and can record the
required range of frequencies and amplitudes without distortion.

9 The ground motion takes place in three dimensions: vertical, North-South, East-West

9 Historically, we have measured the vertical component of velocity only. The vertical
component has maximum sensitivity to P-waves relative to S-waves and surface
waves. Velocity is technically easier to detect and record than either displacement
or acceleration.

9 Recording only one component minimizes the technical problems


of data storage.

9 The most common geophone is the moving-coil geophone. (Fig.


3-19)

∙ A cylindrical coil is suspended from a spring support in the


field of a permanent magnet which is attached to the
instrument casing.
∙ The magnet has a cylindrical pole piece inside the coil. The
suspended coil represents an oscillatory system with a
resonant frequency determined by the mass of the coil and
the stiffness of tits spring suspension. generates
∙ Movement of the coil in the magnetic field Æ Voltages ⇒ flow

Î Magnetic field that interact with the field of the


magnet(magnetic field) to oppose the motion of the coil.

Topics in Seismic Prospecting SNU 40


Seismic Data Processing

∙ Geophones have a resonant frequency. Most commercial seismic reflection


surveys employ geophones with a resonant frequency between 4 and 15 Hz.
∙ To preserve the shape of the seismic waveform, geophones should have a flat
frequency response and minimal phase distortion within the frequency range of
interest.
∙ Above the resonant frequency, the output of a moving-coil geophone is
proportional to the velocity of coil

9 Hydrophones are composed of ceramic piezoelectric element which produces an


output voltage proportional to the pressure variations associated with the passage
of a compressional seismic wave through water.

Large numbers of individual hydrophones are made up into hydrophone streamers.

9 Arrays of geophones or hydrophones may be connected together into linear or


areal arrays containing tens or even hundreds of transducers whose individual
outputs are summed.

Such arrays provide detectors with a directional response that facilitates the
enhancement of signal and the suppression of certain types of noise.

¾ Multichannel reflection survey design

9 The basic requirement of a multichannel reflection survey is to obtain recordings


of reflected pulses at several offset distances from a shot pointer.

Topics in Seismic Prospecting SNU 41


Seismic Data Processing

∙ What do the “offset” and “channel” mean?


“offset” means distance between shot and receiver
∙ Each detector in a conventional reflection spread consists of an array (or group)
of several geophones or hydrophones arranged in a specific pattern and
connected together in series or parallel to produce a single channel of output.

∙ The effective offset of an array is taken to be the distance from the shot to the
center of the array.

9 In 2-D survey (reflection profiling), data are collected along survey lines that
nominally contain all shot points and receivers. Reflected ray paths are assumed to
lie in the vertical plane containing the survey line. Uniform geometry along strike
is assumed
9 3-D survey is needed in areas of structural complexity.

9 In the survey, we obtain common shot gather.

9 The two most common shot-detector configurations in multichannel reflection


profiling surveys are split spread (straddle spread) and the single-ended spread.

9 Surveys on land are commonly carried out with a split spread geometry, but in
marine reflection surveys the single-ended spread are the normal configuration.

9 Reflection surveys are normally designed to provide a specified depth of


penetration and a particular degree of resolution of the subsurface geology in
both the vertical and horizontal dimensions.

Topics in Seismic Prospecting SNU 42


Seismic Data Processing

Topics in Seismic Prospecting SNU 43


Seismic Data Processing

¾ Seismic resolution

9 Resolution relates to how close two points can be, yet still be distinguished.
9 Vertical resolution
∙ For two reflections, one from the top and one from the bottom of a thin layer,
there is a limit on how close they can be, yet still be separable.
∙ This limit depends on the thickness of the layer
∙ The dominant wavelength of seismic waves is given by

Topics in Seismic Prospecting SNU 44


Seismic Data Processing

∙ The acceptable threshold for vertical resolution generally is


a quarter of the dominant wavelength.
∙ This is subjective and depends on the noise level in the
data. Sometimes the quarter-wavelength criterion is too
generous (for small reflection coefficients with noise), and
sometimes it is too stringent for strong reflections.
∙ Resolution implies that reflections from the top and bottom
of a thin bed are seen as separate events or wavelet lobes.

9 Lateral resolution

∙ Lateral resolution refers to how close two reflecting points can be situated
horizontally, yet be recognized as two separate points rather than one.
∙ Consider the spherical wavefront that impinges on the horizontal planar
reflector.
∙ This reflector can be visualized as a continuum of point diffractors.

2z0
∙ The energy from the subsurface point (O) arrives at time t0 = .
v

∙ Let the incident wavefront advance in depth by the amount λ/4. Energy from
⎛ λ⎞
2 ⎜ z0 + ⎟
subsurface location A or A’ will reach the receiver at time t1 =
⎝ 4⎠
.
v

∙ Th en, the total energy arriving within the time interval (t1-
λ 1 T
t0), which equals = = . The reflecting disk AA’ is
2v 2f 2
called a half-wavelength Fresnel zone or the first Fresnel
λ 1
zone. fλ =v = (half the dominant period)
v 2f
∙ The Fresnel zone depends on wavelength, and it depends
on frequency. If the seismic signal is relatively high
frequency, then the Fresnel zone is relatively narrow. The
smaller the Fresnel zones, the easier it is to differentiate

Topics in Seismic Prospecting SNU 45


Seismic Data Processing

between two reflecting points.


∙ Lateral resolution depends on velocity and the depth of the reflecting interface
(the radius of the wavefront):

¾ Design of detector arrays

9 Each detector in a conventional reflection spread consists of an array (or group) of


several geophones or hydrophones arranged in a specific pattern and connected
together in series or parallel to produce a single channel of output.
9 Arrays of geophones provide a directional response and are used to enhance the
near-vertically travelling reflected pulses and to suppress several types of
horizontally travelling coherent noise.

9 To exemplify this, consider a Rayleigh surface


wave (a vertically polarized wave travelling
along the surface) and a vertically travelling
compressional wave reflected from a deep
interface to pass simultaneously through two
geophones connected in series and spaced at
half the wavelength of the Rayleigh wave.
At any given instant, ground motions
associated with the Rayleigh wave will be in
opposite directions at the two geophones and
the individual outputs of the geophones at any
instant will therefore be equal and opposite
and be cancelled by summing.

9 Ground motions associated with the reflected


compressional wave will be in phase at the two
geophones and the summed outputs of the
geophones will therefore be twice their
individual outputs.

Topics in Seismic Prospecting SNU 46


Seismic Data Processing

9 The directional response of any linear array is governed by the relationship


between the apparent wavelength λa of a wave in the direction of the array, the
number of elements n in the array and their spacing Δx . The response is given
by a response function R
sin nβ
R=
sin β
where
πΔx
β=
λa
R is a periodic function that is fully defined in the interval 0 ≤ Δx / λa ≤ 1 and is
symmetrical about Δx / λa = 0.5 .

The more the number of geophones gives the deeper troughs

9 The initial stage of a reflection survey involves field trials in the survey area to
determine the most suitable combination of source, offset recording range, array
geometry and detector spacing (the horizontal distance between the centers of
adjacent geophone arrays, often referred to as the group interval) to produce
good seismic data in the prevailing conditions.

9 Source trials involve tests of the effect of varying, for example, the shot depth and
charge size of an explosive source, or the number, chamber sizes and trigger delay
times of individual guns in an air gun array.

9 The detector array geometry needs to be designed to suppress the prevalent


coherent noise events (mostly source-generated). On land, the local noise is
investigated by means of a noise test in which shots are fired into a spread of
closely-spaced detectors (noise spread) consisting of individual geophones, or

Topics in Seismic Prospecting SNU 47


Seismic Data Processing

arrays of geophones clustered together to eliminate their directional response.


(Fig. 4.12)

A series of shots is fired with the noise spread being moved progressively out to
large offset distances. For this reason such a test is sometimes called a walk-away
spread.

Topics in Seismic Prospecting SNU 48


Seismic Data Processing

8. Seismic data processing

¾ The goal of seismic data processing to improve the usefulness and quality of seismic
data.

¾ Data Format
1967: SEG-A, B, X (multiplexed)
1972: SEG-C (multiplexed)
1975: SEG-Y (de-multiplexed)
1980: SEG-D (multi-purpose, multiplexed or demltiplexed)
?? SEG-2
All the data processing software have its own format.
A common format used in the seismic industry for data exchange is SEG-Y.

¾ Digital recording system

∙ Field data are recorded in a multiplexed mode using a certain type of format. A
multiplexer connects the different channels sequentially to the digitizer (or analog-
to-digital (A/D) converter).

∙ A multiplexer digitizes all the channels in order with an interval of 50 micro-


second. After the multiplexer samples data in channel 1, 2, … n in order, it returns
to the first channel. The period of multiplexer is called sample rate. Usually it is 2
ms or 4 ms.

¾ Gain control

– Gain is a time-variant scaling in which the scaling function is based on a desired


criterion.

Topics in Seismic Prospecting SNU 49


Seismic Data Processing

– Gain is applied to seismic data for display. An automatic gain control (AGC) is
applied to seismic data to bring up weak signals.
– An exponential gain may be applied to compensate for attenuation losses.
– Gain must be used with care, since it can destroy signal character
– An automatic gain control (AGC) is applied to seismic data to bring up weak
signals. An instantaneous AGC with a very small time gate makes strong
reflections indistinguishable from weak reflections.
– Trace balancing is a time-invariant scaling of amplitudes. Trace-balancing usually
is based on rms-amplitude criterion. Specifically, each trace in a group of traces
is scaled so that they all have the same desired rms amplitude level.

9 Geometrical spreading

– For a homogeneous medium without attenuation, the decay of wave


1
amplitude is described by
r

– For a layered earth, amplitude decay can be described approximately by


1/ ⎡⎣v 2 ( t ) t ⎤⎦ . t is the two-way traveltime and v(t) is the root-mean-squared
(rms) velocity.

– The gain function for geometrical spreading compensation is defined by


v2 (t ) t
g ( t ) = 2 , where v0 is the reference velocity at specified time t0 .
v0 t0

– While reflections have been brought up in strength, noise components in


the data also have been boosted. This is one undesirable aspect of any type
of gain application. Coherent noise and multiples have been overcorrected.
(FIG. 1.4-6)

– To prevent overcorrection of amplitudes of multiple reflections, a velocity-


α
independent scaling function, such as g (t ) = t , where α usually is set to 2,
can be used for geometrical spreading correction. (FIG 1.4-7)

Topics in Seismic Prospecting SNU 50


Seismic Data Processing

9 Programmed Gain Control (PGC)


– A gain function can be defined by interpolating between some scalar values
specified at particular time samples. Larger scalar values naturally would be
assigned at late times.
– Larger scalar values naturally would be assigned at late times. (Fig. 1.4-8, Fig.
1.4-9)

Topics in Seismic Prospecting SNU 51


Seismic Data Processing

9 RMS Amplitude AGC

– The rms amplitude AGC gain function is based on the rms amplitude within
a specified time gate on an input trace.

– The input trace is subdivided into fixed time gates. First, the amplitude of

Topics in Seismic Prospecting SNU 52


Seismic Data Processing

each sample in a gate is squared. Second, the mean of these values is


computed and its square root is taken. Æ rms amplitude over that gate.
Gain function at the gate center is given by
desiredrms
g (t ) = , where xi is the trace amplitude and N is the number

N 2
1
N i =1 i
x
of samples within the gate.

9 Instantaneous AGC

– Instantaneous AGC is one of the most common gain types used.


– First, the mean absolute value of trace amplitudes is computed within a
specified time gate. Second, the ratio of the desired rms level to this mean
value is assigned as the value of the gain function.
– Unlike the rms amplitude AGC, this value is assigned to any desired time
sample of the gain function within the time gate, say the nth sample of the
trace, rather than to the sample at the center of the gate.
desiredrms
g (t ) =

N
1
N i =1
xi
Figure 1.4-11: Very small time gates cause a significant can loss of signal
character by boosting zones that contain small amplitudes (64-ms). In
practice, AGC time gates commonly are specified between 200 and 500 ms.

Topics in Seismic Prospecting SNU 53


Seismic Data Processing

9 Relative Trace Balancing

– In true amplitude processing, it is necessary to display the data without


applying a time-varying data-dependent gain function.

– Trace balancing (trace equalization) schemes are used for this type of scaling.

– The balance factor is defined as the ratio of the desired rms to the rms
amplitude that is computed from a specified time window. A separate
balance factor is computed for and applied to each trace, individually.
– Alternatively, a single balance factor based on a selected trace within a
group of traces can be applied to the entire group. This is called relative
trace balancing.

Topics in Seismic Prospecting SNU 54


Seismic Data Processing

(1) Prestack processing

a. Demultiplex

9 If the data format is SEG-D, we don’t have to apply demultiplex. By reading data
from the reel tape, we can obtain demultiplexed data.
9 Mathematically, demultiplexing is seen as transposing a big matrix so that the
columns of the resulting matrix can be read as seismic trace recorded at different
offsets with a common shot point.

b. Trace editing (is almost the first step of seismic data processing)

9 Noisy traces, traces with transient glitches or monofrequency signals are deleted.
9 In case of very shallow marine data, guided waves are muted since they travel
horizontally within the water layer and do not contain reflections from the
substratum.

c. Filtering ⇒ in the frequency domain


9 Most marine data are contaminated by swell noise and cable noise. These types
of noise carry very low-frequency energy but can be high in amplitudes. They
can be recognized by their distinctive linear pattern and vertical streaks.
9 The swell noise and cable noise are removed from shot records by a low-cut
filtering.
9 Attenuation of coherent linear noise associated with side scatterers and ground
roll may require techniques based on dip filtering.

d. Dip filtering (f-k filtering): frequency, wavenumber

Topics in Seismic Prospecting SNU 55


Seismic Data Processing

9 In f-k spectra, we can discriminate signals from noises and ground roll based on
velocity information. (Fig. 9.36, Fig. 4.25)

Topics in Seismic Prospecting SNU 56


Seismic Data Processing

9 Coherent Linear noise can be removed by dip filtering


9 The pass or reject zones in the f-k spectra do not need to be constrained to fan
shape.

9 The left half of the spectra mostly contains spatially aliased data. By keeping the
fan shape in the right quadrant, while zeroing out most of the left quadrant, the
coherent noise trains, including the aliased energy, are eliminated.

9 Because aliasing is a serious concern in f-k filtering, it is advisable to apply it to


shot gathers rather than to CMP gathers, since CMP gathers can have twice the
trace spacing of shot gathers.
9 When we apply f-k filtering to CMP gathers, two neighboring CMP gathers can
be interleaved before applying f-k dip filtering, then split afterward, thus
alleviating aliasing.

Topics in Seismic Prospecting SNU 57


Seismic Data Processing

e. Gain recovery

9 A gain recovery function is applied to the data to correct for the amplitude
effects of spherical wavefront divergence.
9 This amounts to applying a geometric spreading function, which depends on
traveltime.
9 Optionally, this amplitude correction is made dependent on a spatially averaged
velocity function, which is associated with primary reflections in a particular
survey area
9 An exponential gain function may be used to compensate for attenuation losses.

f. Field geometry

9 Field geometry is merged with the seismic data. This precedes any gain
correction that is offset dependent.
9 Based on survey information for land data or navigation information for marine
data, coordinates of shot and receiver locations for all traces are stored on trace
headers.
9 Many types of processing problems arise from setting up the field geometry,
incorrectly. As a result, the quality of a stacked section can be degraded severely.

g. Static correction

9 In general, the source and detector are placed on the planar horizontal surface
of a uniform velocity layer. This is clearly not true for field data where the surface
elevations vary, primarily due to variable weathering of bedrock, drift deposits
and variable depth of the water table.

9 Near-surface irregularities have the effect of shifting reflection events on adjacent


traces out of their true time relationships.

9 For land data, elevation statics are applied to reduce traveltimes to a common
datum level. This level may be flat or vary along the line.

9 Reduction of traveltimes to a datum usually requires correction for the near-


surface weathering layer (weathering static corrections) in addition to differences
in elevation of source and receiver stations (elevation static corrections).

9 The precise heights of all survey stations can be determined by the global

Topics in Seismic Prospecting SNU 58


Seismic Data Processing

positioning system (GPS). Using differential GPS (DGPS), positions and height can
be determined in real time to an accuracy of better than 1 m.
9 The weathering static correction requires knowledge of the variable velocity and
thickness of the weathered layer.
9 Refraction method, uphole surveys, and downhole surveys can be used to
estimate the near-surface effects associated with the weathering layers.
9 For marine data, do we need static correction?

h. CMP sorting

¾ CMP gather & CDP gather (Fig. 4.9)

Topics in Seismic Prospecting SNU 59


Seismic Data Processing

¾ Fold

9 The fold refers to the number of traces in the CMP gather and may
conventionally be 24, 30, 60 or, exceptionally, over 1000. The fold is alternatively
expressed as a percentage.

9 Single fold: if the shot-receiver spread in a multichannel


reflection survey is moved forward in such a way that
no two reflected ray paths sample the same point on a
subsurface reflector, the survey coverage is said to be
single-fold.
Single fold = 100% coverage.

9 Each seismic trace then represents a unique sampling of


some point on the reflector.

9 In common mid-point (CMP) profiling, which has


become the standard method of two-dimensional
multichannel seismic surveying, it is arranged that a set
of traces recorded at different offsets contains
reflections from a common depth point (CDP) on the
reflector (Fig. 4.14)

9 The fold of a CMP profile is determined by the quantity


N
, where N is the number of geophone arrays along a
2n
spread and n is the number of geophone array spacings
by which the spread is moved forward between shots
(the move-up rate).

With a 96-channel spread (N=96) and a move-up rate of


8 array spacings per shot interval (n=8), the coverage
would be _________________________.

¾ CMP sorting

9 Converts common shot gather (source-receivers) Æ common midpoint gather


(cdp –offset)

Topics in Seismic Prospecting SNU 60


Seismic Data Processing

i. NMO correction

9 It is also called dynamic correction, which is applied to reflection times to


remove the effect of normal moveout (NMO).

9 The correction has to be calculated separately for each time increment of a


seismic trace.

9 In common midpoint surveys the appropriate velocity is derived by computer


analysis of moveout in the groups of traces from a common mid-point (CMP
gathers).

9 Prior to this velocity analysis, static corrections must be applied to the individual
traces to remove the effect of the low velocity surface layer.
9 The stacking velocity Vst is defined as that velocity value which produces the
maximum amplitude of the reflection event in the stack of traces. This clearly
represents the condition of successful removal of NMO.

9 The travel-time curve for reflected rays in a multilayered ground is not a


hyperbola (see Fig. 4.3b). However, if the maximum offset value x is small
compared with reflector depth, the stacking velocity closely approximates the
1
⎡ n 2 ⎤2
⎢ ∑ vi τ i ⎥
root-mean-square velocity Vrms. Vrms .n = ⎢ i =1n ⎥
⎢ τi ⎥
⎢⎣ ∑ i =1
⎥⎦

Topics in Seismic Prospecting SNU 61


Seismic Data Processing

9 From Vrms, interval velocities can be derived using the Dix formula.

Dix formula
n

∑v τ i
2
i
Vrms.n 2 = i =1
n

∑τ
i =1
i

n −1

∑v τ i
2
i
Vrms.n −12 = i =1
n −1

∑τ
i =1
i

n n −1
tnVrms.n 2 − tn −1Vrms.n −12 = ∑ vi 2τ i − ∑ vi 2τ i
i =1 i =1

= vn τ n
2
τ n = tn − tn −1
tnVrms.n 2 − tn −1Vrms.n −12
= vn ⇒ Dix equation
tn − tn −1

Topics in Seismic Prospecting SNU 62


Seismic Data Processing

Topics in Seismic Prospecting SNU 63


Seismic Data Processing

j. Determination of Velocity

¾ Constant-velocity NMO corrected gather

Topics in Seismic Prospecting SNU 64


Seismic Data Processing

Topics in Seismic Prospecting SNU 65


Seismic Data Processing

¾ Constant-velocity stack

Topics in Seismic Prospecting SNU 66


Seismic Data Processing

¾ Velocity spectrum: based on cross correlation ⇒ we assume a number of


hyperbolic trajectories ⇒ a number of velocities

9 Fig. 3.2-9b is the velocity spectrum of Fig. 3.2.-9a.


9 Velocity spectrum is expressed as stacked traces on a plane of velocity
versus two-way zero-offset time. In other words, data are transformed from
the offset versus two-way time domain (Fig. 3.2-9a) to the stacking velocity
versus two-way zero-offset time domain (Fig. 3.-9b).

9 The velocity where the highest stacked amplitude occurs is used to stack the
event in the input CMP gather.

9 Figure 3.2-10 shows a CMP gather associated with a layered earth model
and its velocity spectrum.

9 The velocity spectrum not only can provide the stacking velocity function,
but it also allows one to distinguish between primary and multiple
reflections.
9 The aim in velocity analysis is to obtain picks that correspond to the best
coherency of the signal along a hyperbolic trajectory over the entire spread
length of the CMP gather.

Topics in Seismic Prospecting SNU 67


Seismic Data Processing

9 Measure of Coherency
∙ Stacked amplitude
M
S = ∑ f i ,t ( i ) M: the number of traces in the CMP gather
i =1

fi ,t ( i ) is the amplitude value on the ith trace at two-way


time t(i)
xi 2
t (i ) = t0 2 +
vs + k 2
∙ Normalized stacked amplitude:


M
f i ,t ( i )
NS = i =1


M
i =1
f i ,t ( i )

∙ Unnormalized crosscorrelation:
N M −1 M − K
1
CC =
N
∑∑ ∑
t =1 k =1 i =1
f i ,t ( i ) f i + k , t ( i + k )

1 ⎧⎪ ⎡ M ⎫⎪
2
⎤ M
= ∑ ⎨ ⎢ ∑ f i ,t ( i ) ⎥ − ∑ f 2 i ,t ( i ) ⎬
2 t ⎪⎩ ⎣ i =1 ⎦ i =1 ⎪⎭
∙ Normalized crosscorrelation:
M −1 M − K f i , t ( i ) f i + k ,t ( i + k )
2
NC = ∑∑ ∑
M ( M − 1) t k =1 i =1 ∑f t
2
i ,t ( i ) ∑f
t
2
i + k ,t ( i + k )

∙ Energy normalized crosscorrelation:


2 CC 1
EC = EC = ( M * NE − 1)
( M − 1) M
M −1
∑∑ ft i =1
2
i ,t ( i )

∙ Semblance
2

1 ∑ ⎣∑
⎡ M
f ⎤
i =1 i ,t ( i ) ⎦
NE =
t

M ∑∑ M
f
t i =1 i ,t ( i )

9 Factors affecting velocity estimates


∙ Spread length (Figure 3.2-13, 3.2-14)

Topics in Seismic Prospecting SNU 68


Seismic Data Processing

Lack of large-offset information means lack of the significant moveout


required for velocity discrimination.

The loss in sharpness of the peaks in the velocity spectra computed


from the small-spread portion of the CMP gather.

What if only the far offsets are included when computing the velocity
spectrum?

Adequate resolution in the velocity spectrum can only be obtained with


a sufficiently large spread that spans both near and far offsets.

∙ Stacking fold
For computational saving, high-fold data sometimes are reduced to a
low-fold equivalent gather by partial stacking.

Topics in Seismic Prospecting SNU 69


Seismic Data Processing

On the contrary, we may mix two CDP gathers to obtain one CDP
gather, when we have low-fold CDP gathers. Using low-fold CDP
gathers such as 8 fold may shift the peaks in the spectrum.

∙ Signal-to-noise ratio
Noise has a direct effect on the quality of a velocity spectrum.
Fig. 3.2-18 and Fig. 3.2-20

Topics in Seismic Prospecting SNU 70


Seismic Data Processing

∙ Muting
As a result of moveout correction, the waveform along a reflection
hyperbola is stretched. Muting is needed. But muting reduces fold in
the stacking process for shallow data, and weakens the peak amplitude
in velocity spectrum.

This is compensated by multiplying stacked amplitudes by a scale factor


equal to the ratio of the actual multiplicity to the number of live traces.
Fig. 3.2-21

∙ Time gate length


Fig. 3.2-23
If the gate length chosen is too coarse, the spectrum suffers especially

Topics in Seismic Prospecting SNU 71


Seismic Data Processing

from lack of temporal resolution.

In practice, the gate length is chosen between one-half and one times
the dominant period of the signal, typically 20 to 40 ms.
Since the dominant period can be time-variant (small in early and large
in late times), the gate length can be specified accordingly.

∙ Velocity sampling

The velocity range used in the analysis must be chosen carefully. It


should span the velocities that correspond to those of primary
reflections present in CMP gather.

The velocity increment must not be too coarse, for it can degrade the
resolution, especially for high-velocity events.

Several options are considered in constructing the velocity spectrum.


Partial stacking, band-pass filtering and automatic gain control (AGC).

Another way to improve the quality of a velocity spectrum is to use


several neighboring CMP gathers in the analysis.
Figure 3.2-25
One way is to sum the gathers and compute the velocity spectrum from
the sum. Figure 3.2-25b
The other is to compute the velocity spectrum from each individual
gather and sum the spectra as shown in Figure 3.2-25c

Topics in Seismic Prospecting SNU 72


Seismic Data Processing

∙ Choice of coherency measure

∙ True departures from hyperbolic moveout

∙ Bandwidth of data

Topics in Seismic Prospecting SNU 73


Seismic Data Processing

(2) Deconvolution

¾ Deconvolution compresses the basic wavelet in the recorded seismogram, attenuates


reverberations and short-period multiples, thus increases temporal resolution and
yields a representation of subsurface reflectivity.
¾ Deconvolution normally is applied before stack; however, it is also common to apply
deconvolution to stacked data.
¾ Deconvolution removes a considerable amount of ringyness, while it has compressed
the waveform at each of the prominent reflections.
¾ Recorded seismogram can be modeled as a convolution of the earth’s impulse
response with the seismic wavelet.
This wavelet has many components, including source signature, recording filter,
surface reflections, and geophone response.
¾ The earth’s impulse response is what would be recorded if the wavelet were just a
spike. The impulse response comprises primary reflections (reflectivity series) and all
possible multiples.
¾ Deconvolution should compress the wavelet components and eliminate multiples,
leaving only the earth’s reflectivity in the seismic trace.

Topics in Seismic Prospecting SNU 74


Seismic Data Processing

¾ Expression

When applied to a seismogram, the inverse filter


should yield the earth’s impulse response.
x(t ) = w(t ) * r (t )
r (t ) y (t ) = w(t ) * r (t ) + n(t )
Let’s assume that deconvolution filter i (t )
r (t ) = x(t ) * i (t )
= i (t ) * w(t ) * r (t )
s (t )
δ (t )
w(t ) * i (t ) = δ (t )
i(t): We can design a deconvolution filter.
Deconvolution tries to recover the reflectivity series
x(t )
(impulse response) from the recorded seismogram.

n(t )

Topics in Seismic Prospecting SNU 75


y (t )
Seismic Data Processing

¾ Assumptions ⇒ zero-offset recording


9 Assumption 1a: The earth is made up of horizontal layers of constant velocity.
9 Assumption 1b: The source generates a compressional plane wave that impinges
on layer boundaries at normal incidence. Under such circumstances, no shear
waves are generated.
9 Assumption 2: The source waveform does not change as it travels in the
subsurface; i.e., it is stationary
9 Assumption 3: The noise component is zero
9 Assumption 4: The source waveform is known
If the source waveform were known, then the solution to the deconvolution
problems is deterministic
If the source waveform were unknown, then the solution to the deconvolution
problem is statistical.
9 Assumption 5: Reflectivity is a random process. This implies that the seismogram
has the characteristics of the seismic wavelet in that their autocorrelations and
amplitude spectra are similar.
This assumption is the key in implementing the predictive deconvolution. It
allows the autocorrelation of the seismogram, which is known, to be substituted
for the autocorrelation of the seismic wavelet.
As a result of assumption 5, an inverse filter can be estimated directly from the
auto correlation of the seismogram.

a. Inverse filtering

9 The inverse filtering is accomplished mathematically by using the z-transform.

9 Let the basic wavelet be a two-point time series given by (1, -1/2). The z-
transform of this wavelet is defined by the following polynomial.
1
X ( z) = 1 − z
2
9 The inverse of the wavelet w’(t) is obtained by polynomial division of the z-
1 1 1
transform: F ( z ) = = 1+ z + z2 +"
1 2 4
1− z
2
9 As in any filtering process, in practice the operator is truncated.

Topics in Seismic Prospecting SNU 76


Seismic Data Processing

b. Least-squares inverse filtering

9 Find the filter such that the error between the actual output and the desired
output (1, 0, 0) is minimum in the least-squares sense.
9 Actual output is computed by convolving the filter (a, b) with the input wavelet
[1, -1/2]
9 The cumulative energy of the error E is defined as the sum of the squares of
differences between the coefficients of the actual and desired outputs.
9 The error made by the least-squares filter is much less than the error made by
the truncated inverse filter.

Topics in Seismic Prospecting SNU 77


Seismic Data Processing

2 2
⎛ a⎞ ⎛ b⎞
E = ( a − 1) + ⎜ b − ⎟ + ⎜ − ⎟
2

⎝ 2⎠ ⎝ 2⎠
Find coefficients (a, b) so that E takes its minimum value
⇒ Variation of E with respect to the coefficients (a, b) to vanish:
⇒ Take the partial derivatives of E with respect to a and b, and set the results to
zero
∂E ⎛ a ⎞⎛ 1 ⎞
= 2 ( a − 1) + 2 ⎜ b − ⎟ ⎜ − ⎟ = 0
∂a ⎝ 2 ⎠⎝ 2 ⎠
5
⇒ a −b = 2
2
∂E ⎛ a ⎞ ⎛ b ⎞⎛ 1 ⎞ 5
= 2 ⎜ b − ⎟ + 2 ⎜ − ⎟⎜ − ⎟ = −a + b = 0
∂b ⎝ 2 ⎠ ⎝ 2 ⎠⎝ 2 ⎠ 2

c. Minimum phase
9 Minimum phase: if its energy is maximally concentrated at its onset.
9 Maximum phase: If its energy is maximally concentrated at its end.
9 Mixed phase: in all in-between situations, the wavelet is mixed phase.
9 Realizable: A wavelet is defined as a transient waveform with a finite duration
9 Causal: A wavelet that is zero for t<0 is called causal.

Topics in Seismic Prospecting SNU 78


Seismic Data Processing

9 Input wavelet: minimum phase and Desired output is delayed spike

9 Input data: the maximum phase and output data: the delayed spike

Topics in Seismic Prospecting SNU 79


Seismic Data Processing

9 Four kinds of wavelets

9 Note that zero lag of the autocorrelation is equal to the total energy contained
in each wavelet.

9 The process by which the seismic wavelet is compressed into a zero-lag spike is
called spiking deconvolution.
Their performance depends not only on filter length but also on whether the
input wavelet is minimum phase.

9 Assumption 6: The seismic wavelet is minimum phase. It has a minimum phase


inverse. Æ This assumption is required by spiking deconvolution.

Topics in Seismic Prospecting SNU 80


Seismic Data Processing

d. Optimum Wiener Filter


9 Consider the desired output (1, 0, 0) and the input wavelet (1,-1/2)
9 Find the autocorrelation of the input wavelet (1, -1/2) and the crosscorrelation of
the desired output (1, 0, 0) with the input wavelet (1,-1/2)
9 We can rewrite the least-squares filter by the autocorrelation of the input wavelet
and the crosscorrelation of the desired output with the input wavelet.

* In the previous section.


⎛ 1⎞
The least-squares filter of input wavelet ⎜ 1 − ⎟
⎝ 2⎠
and the desired output (1, 0, 0)
⎡5 ⎤ ⎡a ⎤ ⎡ 2⎤
⎢ 2 −1⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥⎢ ⎥ = ⎢ ⎥
⎢ −1 5 ⎥ ⎢ b ⎥ ⎢ 0 ⎥
⎢⎣ 2 ⎥⎦ ⎣ ⎦ ⎣ ⎦

9 These observations were generalized by Wiener to derive filters that convert the
input to any desired output. The general form of the matrix equation for a filter
of length n is

⎡ r0 r1 r2 " rn −1 ⎤ ⎡ a0 ⎤ ⎡ g 0 ⎤
⎢r r0 r1 " rn − 2 ⎥⎥ ⎢⎢ a1 ⎥⎥ ⎢⎢ g1 ⎥⎥
⎢ 1
⎢ r2 r1 r0 " rn −3 ⎥ ⎢ a2 ⎥ = ⎢ g 2 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ # ⎥⎢ # ⎥ ⎢ # ⎥
⎢⎣ rn −1 rn − 2 rn −3 " r0 ⎥⎦ ⎢⎣ an −1 ⎥⎦ ⎢⎣ g n −1 ⎥⎦
⇒ Symmetric Toeplitz matrix └→ desired filter coefficients
⇒ Levinson-recursion에 의해 풀 수 있음

9 The optimum Wiener filter ( a0 , a1 , a2 ," , an −1 ) is


optimum in that the least-squares error between the
actual and desired outputs is minimal.
9 When the desired output is the zero-lag spike (1, 0, 0,
…, 0), then the Wiener filter is identical to the least-
squares inverse filter.

Topics in Seismic Prospecting SNU 81


Seismic Data Processing

9 The Wiener filter applies to a large class of problems in which any desired output
can be considered:
Type 1: zero-lag spike Type 5: Any desired arbitrary shape
Type 2: Spike at arbitrary lag
Type 3: Time-advanced form of input series
Type 4: Zero-phase wavelet
¾ Spiking Deconvolution

9 When the desired output is zero-lag spike, it is called spiking deconvolution.


9 Crosscorrelation of the desired spike (1, 0, 0, …, 0) with input wavelet
( x0 , x1 , x2 ," , xn −1 ) yields the series ( x0 , 0, 0," , 0 ). The generalized form of the
normal equations is
⎛ r0 r1 " rn −1 ⎞ ⎛ a0 ⎞ ⎛ 1 ⎞
r2
⎜ ⎟⎜ ⎟
⎜ r1 r0 " rn − 2 ⎟ ⎜ a1 ⎟ ⎜⎜ 0 ⎟⎟
r1
⎜ r2 r1 r0 " rn −3 ⎟ ⎜ a2 ⎟ = ⎜ 0 ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ # ⎟⎜ # ⎟ ⎜ # ⎟
⎜r
⎝ n −1rn − 2 rn −3 " r0 ⎟⎠ ⎜⎝ an −1 ⎟⎠ ⎜⎝ 0 ⎟⎠
1
→ Autocorrelation/ x0 ⇒ scaled by
x0
9 Spiking deconvolution is mathematically identical to least-squares inverse
filtering.
9 A distinction is made in practice between the two types of filtering. The
autocorrelation matrix on the left side of the above equation is computed from
the input seismogram in spiking deconvolution (statistical deconvolution; based
on assumption 5), whereas it is computed directly from the known source
wavelet in the case of least-squares inverse filtering (deterministic deconvolution).
9 Assumption 5 is valid based on Fig. 2.1-6. From Fig. 2.1-6, we can note that
autocorrelograms and amplitude spectra of the seismogram are similar to those
of input wavelet.
(fig. 2.1-6)

Topics in Seismic Prospecting SNU 82


Seismic Data Processing

9 The goal of spiking deconvolution is to yield spiked wavelet as the desired


output
9 In the frequency domain, the amplitude spectrum of the spike wavelet is flat. We
should design spiking deconvolution operator so that the amplitude of the
desired output is flat.
9 The spiking deconvolution operator is the inverse of the minimum-phase
equivalent of the input wavelet. When the phase spectrum of the output is not
minimum phase, the spiking deconvolution cannot convert it to a perfect zero-
lag spike, even though the amplitude spectrum is virtually flat.
¾ Prewhitening
9 The amplitude spectrum of the spiking deconvolution operator is the inverse of
the amplitude spectrum of the input wavelet. What if there are zeroes in the
amplitude spectrum of the input wavelet?
9 Poor results can occurs because the deconvolution operator tries to boost the
absent frequencies.
9 To ensure numerical stability, an artificial level of white noise is introduced before
deconvolution. This is called prewhitening. Prewhitening is achieved by adding a
constant to the zero lag of the autocorrelation function (Figure 2-21).
⎡ (1 + ε )r0 r1 r2 " rn −1 ⎤
⎢ r (1 + ε )r0 r1 " rn − 2 ⎥⎥
⎢ 1

⎢ r2 # ⎥
⎢ ⎥
⎢ # # ⎥
⎢⎣ rn −1 " " " (1 + ε )r0 ⎥⎦

Topics in Seismic Prospecting SNU 83


Seismic Data Processing

¾ Wiener Shaping Filters ⇒ They are designed to be applied to any desired arbitrary shape

9 Spiking deconvolution had trouble compressing wavelet, when the input wavelet
is not minimum phase
9 For any given input wavelet, a series of desired outputs can be defined as
delayed spikes. The least-squares errors then can be plotted as a function of
delay.
9 The delay (lag) that corresponds to the least error is chosen to define the
desired delayed spike output.
9 The process that has a desired output (any desired arbitrary shape) is called
wavelet shaping Æ Wiener shaping filter

¾ Predictive Deconvolution
9 Given the input x(t), we want to predict its value at some future time (t+α),
where α is prediction lag.
9 Input data is x(t), and output data is x(t+α)
9 [
When α=2 and x(t ) = x0 , x1 , x2 , x3 , x4 ]

Topics in Seismic Prospecting SNU 84


Seismic Data Processing

⎛ r0 r1 r2 r3 r4 ⎞ ⎛ a0 ⎞ ⎛ r2 ⎞
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ r1 r0 r1 r2 r3 ⎟ ⎜ a1 ⎟ ⎜ r3 ⎟
⎜ r2 r1 r0 r1 r2 ⎟ ⎜ a2 ⎟ = ⎜ r4 ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ r3 r2 r1 r0 r1 ⎟ ⎜ a3 ⎟ ⎜ r5 ⎟
⎜r r0 ⎟⎠ ⎜⎝ a4 ⎟⎠ ⎜⎝ r6 ⎟⎠
⎝ 4 r3 r2 r4

Prediction filter (a0 , a1 ," , a4 )


Prediction error filter (1, 0, − a0 , −a1 , −a2 , −a3, − a4 )
Generalized form of prediction filter

Topics in Seismic Prospecting SNU 85


Seismic Data Processing

⎡ r0 r1 " rn −1 ⎤ ⎡ a0 ⎤ ⎡ rα ⎤
⎢ # # ⎥⎥ ⎢⎢ # ⎥⎥ ⎢⎢ rα +1 ⎥⎥
⎢ =
⎢ # # ⎥⎢ # ⎥ ⎢ # ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣ rn −1 " " r0 ⎦ ⎣ an −1 ⎦ ⎣ rα + n −1 ⎦

9 Does a seismic trace have a predictable component with a periodic rate of


occurrence? If yes, what is it?
9 Consider the special case of unit prediction lag, α=1
⎡ r0 r1 " r4 ⎤ ⎡ a0 ⎤ ⎡ r1 ⎤
⎢# # ⎥⎥ ⎢ # ⎥ ⎢⎢ r2 ⎥⎥
⎢ ⎢ ⎥=
⎢# # ⎥⎢ # ⎥ ⎢# ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣ r4 r3 " r1 ⎦ ⎣ a4 ⎦ ⎣ r5 ⎦
⎡−r1 r0 r1 " r4 ⎤ ⎡ a0 ⎤ ⎡0 ⎤
⎢ ⎥
⎢ −r2 # # ⎥ ⎢⎢ # ⎥⎥ ⎢0 ⎥
=⎢ ⎥
⎢# # # ⎥ ⎢ # ⎥ ⎢# ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢⎣ −r5 r4 r3 " r0 ⎦⎥ ⎣ a4 ⎦ ⎣0 ⎦
⎡ r0 r1 r2 r3 r4 r5 ⎤ ⎡L⎤
⎢r r r r r r ⎥ ⎡ 1 ⎤ ⎢0⎥
⎢ 1 0 1 2 3 4 ⎥ ⎢ −a ⎥ ⎢ ⎥
⎢ r2 r1 r0 " ⎥ ⎢ 0 ⎥ ⎢0⎥
⎢ ⎥ ⎢ −a1 ⎥ = ⎢ ⎥
⎢3 r " ⎥ ⎢ # ⎥ ⎢0⎥
⎢ r4 " ⎥⎢ ⎥ ⎢0⎥
⎢ ⎥ −a ⎥ ⎢ ⎥

⎢⎣ r5 " r0 ⎥⎦ ⎣ 4 ⎦ ⎣ 0 ⎦
L = r0 − r1a0 − r2 a1 − r3 a2 − r4 a3 − r5 a4
Six eq. six unknowns ( a0 a1 a2 a3 a4 L)

e. Predictive deconvolution in practice


¾ Operator Length
9 Consider operator length of the prediction filter (n), prediction lag (α), and
percent prewhitening (ε). Then, the length of the prediction error filter is ( ).
9 See Fig. 2.4-7: Prediction lag is unity (i.e., 2-ms sampling rate), prewhitening is
0 %, and operator length varies. We assume that we know the exact source
wavelet, and use the autocorrelogram of the source wavelet for deconvolution.
9 Short operators yield spikes with small-amplitude and relatively high-frequency
tails. The 128-ms-long operator gives an almost perfect spike output. Longer
operators whiten the spectrum further, bringing it closer to the spectrum of the
impulse response.

Topics in Seismic Prospecting SNU 86


Seismic Data Processing

9 To select an operator length, ideally we want to use the autocorrelation of the


unknown seismic wavelet. Fortunately, the autocorrelation of the input
seismogram has the characteristics of the wavelet autocorrelation.
9 Fig. 2.4-8 shows results obtained by using the autocorrelogram of seismogram
rather than that of the wavelet.
9 Which operator length gives the best results?
9 In the autocorrelogram of seismogram, can you discriminate the autocorrelogram
of source wavelet?

Topics in Seismic Prospecting SNU 87


Seismic Data Processing

9 Fig. 2.4-9 shows the results when source wavelet is unknown, and the
autocorrelogram of seismogram is used for deconvolution.
9 Which operator length gives the best results?

¾ Prediction lag
9 Fig. 2.4-15 shows the results obtained when operator length is 128 ms,
prewhitening is 0%, and prediction lag varies. The source wavelet is known
9 In Fig. 2.4-15, unit prediction lag implies the highest resolution. If prediction lag
is increased, the output from predictive deconvolution becomes less spiky.
9 At a long prediction lag, predictive deconvolution does nothing to the input
wavelet because almost all the lags of its autocorrelation have been left
untouched.
9 Fig. 2.4-16 shows the results when the source wavelet is unknown.
9 The application of spiking deconvolution to field data is not always desirable,
since it boosts high-frequency noise in the data.
9 The most prominent effect of the nonunity prediction lag is suppression of the
high-frequency end of the spectrum and preservation of the overall spectral
shape of the input data.

Topics in Seismic Prospecting SNU 88


Seismic Data Processing

¾ Percent Prewhitening
9 Prewhitening narrows the spectrum without changing much of the fitness
character, while larger prediction lag narrows the spectrum and alters its shape.
9 Prewhitening preserves the spiky character of the output, although it adds a low-
amplitude, high-frequency tail.
9 Prewhitening yields a band-limited output. However, the effect is less controllable
when compared to varying the prediction lag.
9 In practice, 0.1 to 1 percent prewhitening is standard.

Topics in Seismic Prospecting SNU 89


Seismic Data Processing

¾ Multiple suppression

9 Note the two distinct goals for predictive deconvolution: (a) spiking the seismic
wavelet, and (b) predicting and suppressing multiples.
9 The first goal is achieved using an operator with unit prediction lag, while the
second is achieved using an operator with a prediction lag greater than unity.
9 The autocorrelation of the input trace can be used to determine the appropriate
prediction lag for multiple suppression.

9 Prediction lag should be chosen to bypass the first part of the autocorrelogram
that represent the seismic wavelet.
9 Operator length should be chosen to include the first isolated energy packet in
the autocorrelogram.

9 It is only with vertical incidence and zero-offset recording that periodicity of the
multiples is preserved. Therefore, predictive deconvolution aimed at multiple
suppression may not be entirely effective when applied to nonzero-offset data
such as common shot or common midpoint data.
9 Predictive deconvolution sometimes is pplied to CMP stacked data in an effort to
suppress multiples. The performance of such an approach can be unsatisfactory.
9 Which one yields the best results?

Topics in Seismic Prospecting SNU 90


Seismic Data Processing

Topics in Seismic Prospecting SNU 91


Seismic Data Processing

(3) Migration

a. Introduction

¾ On seismic sections, each reflection event is mapped directly beneath the mid-
point of the appropriate CMP gather.
¾ However, the reflection point is located beneath the mid-point only if the reflector
is horizontal.
¾ In the presence of a component of dip along the survey line the actual reflection
point is displaced in the up-dip direction.

¾ Migration is the process of reconstructing a seismic section so that reflection


events are repositioned under their correct surface location and at a corrected
vertical reflection time.
¾ Migration moves dipping reflections to their true subsurface positions and
collapses diffractions, thus increasing spatial resolution and yield a seismic image
of the subsurface.
¾ Migration removes the diffracted arrivals resulting from point sources since every
diffracted arrival is migrated back to the position of the point source.
¾ The goal of migration is to make the stacked section appear similar to the
geologic cross-section in depth along a seismic traverse.

Topics in Seismic Prospecting SNU 92


Seismic Data Processing

¾ Migration classification I: Time migration & Depth migration. In time migration, the
migrated seismic sections still have time as the vertical dimension. In depth
migration, the migrated reflection times are converted into reflector depths using
appropriate velocity information.
¾ Time migration is appropriate as long as lateral velocity variations are mild to
moderate. When the lateral velocity gradients are significant, time migration does
not produce the true subsurface image. Instead, we need to use depth migration.
¾ The migrated section is commonly displayed in time. One reason for this is that
velocity estimation based on seismic and other data is limited in accuracy. Another
reason is that interpreters prefer to evaluate the validity of migrated sections by
comparing them to the unmigrated data.
¾ Time migration example (Fig.4.0-4).

9 Migration classification II: Prestack & poststack migration

Topics in Seismic Prospecting SNU 93


Seismic Data Processing

¾ Exploding reflectors

9 When a stacked section is migrated, we use the migration theory applicable to


data recorded with a coincident source and receiver (zero-offset).
9 Imagine exploding sources that are located along the reflecting interfaces.
Consider one receiver located on the surface at each CMP location along the line.
The sources explode in unison and send out waves that propagate upward. The
waves are recorded by the receivers at the surface. Æ exploding reflectors model.

9 The seismic section that results from the exploding reflectors model is largely
equivalent to the zero-offset section. One important distinction is that the zero-
offset section is recorded as ( ) traveltime, while the exploding reflectors
model is recorded as ( ) traveltime. To make the sections compatible, we
can imagine that the velocity of propagation is half the true medium velocity for
the exploding reflectors model.
9 The assumption that a stacked section is equivalent to a zero-offset section is not
appropriate for some reflections when the strong lateral velocity variations are
present.

¾ Migration strategies

9 Choice of an appropriate migration strategy requires input from the interpreter as


to the structural geology and stratigraphy in an area.
9 Dipping event on a stacked section call for time migration.
9 Conflicting dips with different stacking velocities is one case in which a
conventional stacked section differs from a zero-offset section.
9 Conflicting dips often are associated with salt flanks and fault planes, which can
have 3-D characteristics. This requires 3-D prestack time migration.

Topics in Seismic Prospecting SNU 94


Seismic Data Processing

¾ Migration algorithms

9 The one-way-in-depth scalar wave equation is the basis for common migration
algorithms. These algorithms do not explicitly model multiple reflections,
converted waves, surface waves or noise. Any such energy present in data input to
migration is treated as primary reflections.
9 Migration algorithms can be classified under three main categories:
∙ Integral solution to the scalar wave equation
∙ Finite-difference solutions
∙ Frequency-wavenumber implementations
9 Three principle migration techniques
∙ The semicircle superposition method: was used before the age of computers

∙ The diffraction-summation technique is based on summing the seismic


amplitudes along a diffraction hyperbola whose curvature is governed by the
medium velocity.

The Kirchhoff summation technique basically is the same as the diffraction


summation technique with added amplitude and phase corrections applied to
the data before summation.

Topics in Seismic Prospecting SNU 95


Seismic Data Processing

∙ Another migration technique is based on the idea that a stacked section can be
modeled as an upcoming zero-offset wavefield generated by exploding reflectors.

The migration can be conceptualized as consisting of wavefield extrapolation in


the form of downward continuation followed by imaging.

The fact that the wavefront shape at t=0 corresponds to the reflector shape is
called the imaging principle.

To define the reflector geometry from a wavefield recorded at the surface, we


only need to extrapolate the wavefield back in depth then monitor the energy
arriving at t=0.
The reflector shape at any particular extrapolation depth directly corresponds to
the wavefront shape at t=0.

Downward continuation of wavefields can be implemented conveniently using


finite-difference solutions to the scalar wave equation. Æ finite-difference
migration

∙ Migration by Fourier transform


This method involves a coordinate transformation from frequency to vertical
wavenumber axis, while keeping the horizontal wavenumber unchanged.

Another frequency-wavenumber migration is the phase-shift method. This


method is based on the equivalence of downward continuation to a phase shift
in the frequency-wavenumber domain. The imaging principle is invoked by
summing over the frequency components of the extrapolated wavefield at each
depth step to obtain the image at t=0.

∙ Finite-difference algorithms can handle all types of velocity variations, but they
have different degrees of dip approximations. Furthermore, differencing schemes,
if carelessly designed, can severely degrade the intended dip approximation.
∙ Frequency-wavenumber algorithms have limited ability in handling velocity
variations, particularly in the lateral direction.

¾ Migration parameters
9 Migration aperture width is the critical parameter in Kirchhoff migration. A small
aperture causes removal of steep dips; it generates spurious horizontal events and
organizes the random noise uncorrelated from trace to trace.

Topics in Seismic Prospecting SNU 96


Seismic Data Processing

9 Depth step size in downward continuation is the critical parameter in finite-


difference methods.
9 An optimum depth step size is the largest depth step with the minimum tolerable
phase errors. It dependes on temporal and spatial samplings, dip, velocity and
frequency.

¾ Migration velocities
9 Since velocities generally increase with depth, errors in migration are usually larger
for deep events than shallow events.
9 The steeper the dip, the more accurate the migration velocities need to be, since
displacement is proportional to dip
9 In the presence of large vertical velocity gradients, a 3-D migration can also cause
overmigration of steep dips even with the correct migration velocities.

b. Migration Principle

¾ Semicircle superposition and diffraction migration

9 The dip angle of the reflector in the geologic section is greater than in the time
section; thus, migration steepens reflectors
9 The length of the reflector, as seen in the geologic section, is shorter than in the
time section; thus, migration shortens reflectors.
9 Migration moves reflectors in the updip direction.

Topics in Seismic Prospecting SNU 97


Seismic Data Processing

Topics in Seismic Prospecting SNU 98


Seismic Data Processing

c. Kirchhoff migration
¾ Diffraction summation, which was the first computer implementation of migration, is
based on the summation of amplitudes along hyperbolic paths.
¾ The curvature of the hyperbolic trajectory for amplitude summation is governed by
the velocity function.
¾ The velocity function used to compute the traveltime trajectory is the rms velocity
at the apex of the hyperbola at time τ

Topics in Seismic Prospecting SNU 99


Seismic Data Processing

Topics in Seismic Prospecting SNU 100


Seismic Data Processing

¾ Amplitude and phase factor

9 Obliquity factor: indicates the angle dependence of amplitudes


9 Spherical divergence: Wave energy decays as (1/r2), where r is the distance from
the source to the wavefront, while amplitudes decay as (1/r). Amplitudes must be
scaled by factor (1/r) before summation for wave propagation in three
dimensions.
The spherical spreading factor is proportional to 1/ vr for 2D wave
propagation, and (1/vr) for 3-D wave propagation.

Newman, P., 1973, Divergence effects in a layered earth: Geophysics, 38, 481-488.

9 The wavelet shaping factor: Huygens’ secondary sources must respond as a


wavelet along the hyperbolic paths with a unique phase and frequency
characteristic. Otherwise, there would be no amplitude cancelation when they are
close to one another.

Topics in Seismic Prospecting SNU 101


Seismic Data Processing

¾ Kirchhoff summation
9 The diffraction summation that incorporates the obliquity, spherical spreading
and wavelet shaping factors is called the Kirchhoff summation, and the migration
method based on this summation is called the Kirchhoff migration.
9 The output image Pout ( x0 , z = vτ / 2, t = 0 ) at a subsurface location ( x0 ,z) is

computed from the 2-D zero-offset wavefield Pin ( x0 , z = 0, t − r / v ) :

Δx ⎡ cos θ ⎤
Pout = ∑ ⎢
2π x ⎢⎣ vrms r
ρ ( t ) * Pin ⎥
⎥⎦
Schneider, W., 1978, Integral formulation for migration in two and three
dimensions: Geophysics, 43, 49-76.
R
u ( x, y, 0, )
1 ∂
u ( x, y, z , 0) = −
2π ∂z ∫ ∫ dxdy
R
C (3D)

1 ∂ u ( x, 0,τ )
π ∂z ∫ ∫
u ( x, z , 0) = − dx dτ
r2
τ2 − 2
c
d. Finite-difference migration
9 By recording data at another distance closer to the source of the diffraction
hyperbola, we can perform migration. This process is equivalent to downward
continuation. The process is stopped when the hyperbola collapses to its apex.
This is called imaging principle.
9 For the downward continuation, finite-difference algorithms can be used. In
other words, finite-difference migration algorithms are based on differential
solutions to the scalar wave equation that are used to downward continue the
input wavefield recorded at the surface.
9 The scalar wave equation is a two-way wave equation in depth that describes
propagation of both upcoming and downgoing waves.
9 If we consider the resulting wavefield from the exploding reflectors model as the
upcoming waves, then we are really interested in a one-way wave equation to
downward continue the upcoming waves.
9 We normally use some rational approximation to the one-way wave equation in
finite-difference implementations.

Topics in Seismic Prospecting SNU 102


Seismic Data Processing

9 Derivation
* Scalar wave equation
∂2 P ∂2 P 1 ∂2 P
+ − =0
∂x 2 ∂z 2 v 2 ( x, z ) ∂t 2
* Assume constant velocity and perform 3-D Fourier Transform of the pressure wavefield.
⇒ This is equivalent to substituting the plane-wave solution
ω2
exp(ik x x + ik z z − iωt ) − k x 2 P − k z 2 P + P = 0 ⇒ dispersion relation
v2
v
v←
2
4ω 2
kz 2 = 2 − kx2
v
⎛ 2ω ⎞
2
⎛ vk x ⎞
kz = ⎜ ⎟ 1− ⎜ ⎟
⎝ v ⎠ ⎝ 2ω ⎠
⇒ 15˚ degree approximation

∂P(k x , z , ω ) ⎛ 2ω ⎞ ⎡ 1 ⎛ vk x ⎞ ⎤
2

= −i ⎜ ⎟ ⎢1 − ⎜ ⎟ ⎥ P(k x , z, ω )
∂z ⎝ v ⎠ ⎣⎢ 2 ⎝ 2ω ⎠ ⎦⎥

Topics in Seismic Prospecting SNU 103


Seismic Data Processing

After each downward continuation step, we retard the wavefield by translating it in time.
Time retardation can be expressed by applying a linear phase shift to the pressure wavefield P
Q = P exp(−iωτ )

¾ Wavefield extrapolation and migration

9 A fundamental equation of reflection seismology is the double square-root (DSR)


equation.
9 Before discussing the DSR equation, we review the basic theory for wavefield
extrapolation.
9 2-D scalar wave equation in a medium with constant material density and
compressional wave velocity v(x, z):
⎛ ∂2 ∂2 1 ∂2 ⎞
⎜ 2 + − 2 ⎟
P ( x, z , t ) = 0 (D-1)
⎝ ∂x ∂z v ∂t ⎠
2 2

9 Given the upcoming seismic wavefield P(x, 0, t), which is recorded at the surface,

Topics in Seismic Prospecting SNU 104


Seismic Data Processing

we want to determine reflectivity P(x, z, 0). Determining reflectivity requires


extrapolating the surface wavefield to depth z, then collecting it at t=0.
9 The process of obtaining the earth’s reflectivity P(x, z, t=0) from the observed
wavefield P(x,z=0,t) at the surface z=0 is called migration, and the reverse
process is called modeling.

Fig. D-1. Relationship between migration


and wavefield modeling

¾ Derivation of finite-difference algorithm

∂2 P ∂2 P 1 ∂2 P
+ − =0 (D-2)
∂x 2 ∂z 2 v 2 ( x, z ) ∂t 2
Assume constant velocity and perform 3-D Fourier transform of the pressure
wavefield.
Î This is equivalent to substituting the plane-wave solution
exp ( ik x x + ik z z − iωt )
ω2 ω2 4ω 2
− k x2 P − k z2 P + 2
P = 0 , k z2 =2
− k x2 , k z2 =
2
− k x2 (v=v/2) (D-3)
v v v
⎛ 2ω ⎞ ⎛ 2ω ⎞ ⎡ 1 ⎛ vk x ⎞ ⎤
2 2
⎛ vk x ⎞
−ik z = −i ⎜ ⎟ 1− ⎜ ⎟ = −i ⎜ ⎟ ⎢1 − ⎜ ⎟ ⎥ (D-4)
⎝ v ⎠ ⎝ 2ω ⎠ ⎝ v ⎠ ⎣⎢ 2 ⎝ 2ω ⎠ ⎦⎥
Î 15 degree approximation
∂P ( k x , z , ω ) ⎛ 2ω ⎞ ⎡ 1 ⎛ vk x ⎞ ⎤
2

= −i ⎜ ⎟⎢ 1 − ⎜ ⎟ ⎥ P ( kx , z, ω ) (D-5)
∂z ⎝ v ⎠ ⎣⎢ 2 ⎝ 2ω ⎠ ⎦⎥
∂P ⎛ 2ω vk x2 ⎞
= −i ⎜ − ⎟P
∂z ⎝ v 4ω ⎠
After each downward-continuation step, we retard the wavefield by translating
it in time. Time retardation can be expressed by applying a linear phase shift to
the pressure wavefield P
P = Q exp(−iωτ ) , (D-6)
Consider vertically varying velocity function v(z)
z dz
τ = 2∫ , (D-7)
0 v ( z)

Topics in Seismic Prospecting SNU 105


Seismic Data Processing

∂P ∂Q −iωτ ∂e − iωτ ∂τ
= e +Q
∂z ∂z ∂τ ∂z
∂Q −iωτ 2
= e + Q ( −iω ) e − iωτ (D-8)
∂z v ( z)
⎡ ∂ 2iω ⎤ − iωτ
=⎢ − ⎥ Qe
⎣ ∂z v ( z ) ⎦
Where v ( z ) is the horizontally averaged velocity.

∂Q − iωτ 2iω − iωτ 2ω − iωτ ivk x2 − iωτ


e = Qe −i Qe + Qe (D-9)
∂z v ( z) v 4ω
∂Q ivk x2 ⎡ 1 1⎤
= Q + 2iω ⎢ − ⎥Q (D-10)
∂z 4ω ⎣v ( z) v ⎦

∂Q ivk x2 ∂ 2Q v ∂ 2Q
= Q for v = v ( z ) = , (D-11)
∂z 4ω ∂z∂t 4 ∂x 2
∂Q ⎡ 1 1 ⎤ ∂Q
= 2⎢ − ⎥ (D-12)
∂z ⎣ v ( z ) v ⎦ ∂t
∙ Equation (D-11) accounts for collapsing diffraction energy to the apex of the
traveltime curve only. Hence, it is referred to as the diffraction term.
∙ When lateral velocity variations are significant, the diffraction curve is somewhat
like a skewed hyperbola with its apex shifted laterally away from the diffraction
source. This lateral shift is accounted for by the thin-lens term given by
equation (D-12).
∙ If the lateral velocity variations are significant, then the thin-lens term is not
negligible.
∙ Migration algorithms that implement both the diffraction and thin-lens terms
generally are two-step schemes that alternately solve these two terms.
∙ To propagate one depth step, first apply the diffraction term on wavefield Q.
The thin-lens term then is applied to the output from the diffraction calculation.
∙ A migration method that includes the effects of the thin-lens term is called
depth migration, since the output section is in depth.

In some practical implementations, the downward continuation is performed in


τ rather than z
2
vk z ⎛ vk ⎞ v 2 k x2
ωτ = , ωτ = ω 1 − ⎜ x ⎟ , ωτ = ω − , (D-13)
2 ⎝ 2ω ⎠ 8ω
The first term is associated with a vertical time shift that can be removed by
retardation. To obtain the differential equation in terms of τ, apply the second
term to the retarded wavefield Q and inverse Fourier transform

Topics in Seismic Prospecting SNU 106


Seismic Data Processing

∂ 2Q v 2 ∂ 2Q
= (D-14)
∂τ∂t 8 ∂x 2

e. Frequency-wavenumber migration

¾ Mathematical background

9 It is advantageous to decompose the wavefield into monochromatic plane


waves with different angles of propagation from the vertical:

P ( k x , z , ω ) = ∫∫ P ( x, z , t ) exp ( ik x − iωt )dxdt (D-15)

P ( x, z , t ) = ∫∫ P ( k x , z , ω ) exp ( −ik x + iωt )dxdt (D-16)

∂2 ⎛ ω2 2⎞
P ( k , z , ω ) + ⎜ 2 − k x ⎟ P ( k x , z, ω ) = 0 (D-17)
∂z 2 x
⎝v ⎠
This equation has two solutions: one for upcoming waves, and the other for
downgoing waves.
The upcoming wave solutions is
⎡ ω2 ⎤
P ( k x , z , ω ) = P ( k x , z = 0, ω ) exp ⎢ −i 2 − k x2 z ⎥ (D-18)
⎢⎣ v ⎥⎦
This is the solution of the following one-way wave equation:
∂ ω2
P ( k x , z, ω ) = −i 2 − k x2 P ( k x , z , ω ) (D-19)
∂z v

9 Dispersion relation of the one-way scalar wave equation.


By defining the vertical wavenumber as

ω
2
⎛ vk ⎞
kz = 1 − ⎜ x ⎟ Æ Dispersion relation (D-20)
v ⎝ ω ⎠
Then
P ( k x , z , ω ) = P ( k x , z = 0, ω ) exp [ −ik z z ] (D-21)
Î Downward continuation
9 To determine the reflectivity P(x,z,0) from the wavefield recorded at the earth’s
surface P(x,0,t), proceed as follows:
∙ Perform a 2-D Fourier transform over x and t to get P(kx, 0, ω).

Topics in Seismic Prospecting SNU 107


Seismic Data Processing

∙ Multiply by the all-pass filter exp(-ikzz) to obtain the wavefield P(kx, z, ω) at


depth z.
∙ Subsequent summation over ω and inverse Fourier transform over kz yields
the earth’s image P(x, z, 0) at that depth.
- Perform summation over ω to obtain P(kx, z, 0)
- Finally, inverse Fourier transform over kx to obtain the earth’s image P(x,
z, 0) at that depth.

9 Proof of downward continuation


∙ Given the upcoming wavefield recorded at surface P(x,0,t), we can
decompose this wavefield into monochromatic plane waves, each traveling at
a different angle from the vertical. We identify these plane waves by
attaching them to a unique (kx,ω) pair of numbers. This plane-wave
decomposition is equivalent to Fourier transforming the wavefield to yield
P(kx,0,ω).

∙ Imagine that this plane wave passed the point P at t=0, traveled upward, and
was recorded by a receiver at surface point G at time t.

∙ Suppose that the wavefront is moved to a depth Δz = GG ′ beneath the


receiver at G so that the wavefront at G now is at G’’. If a receiver were
buried at G’’, it would have recorded the plane wave at t-Δt, where Δt is the
traveltime between G and G’’. In other words, moving the receiver at G
vertically down a distance Δz to a new location G’ changes the traveltime
along the raypath from G to P by –Δt.
Δz
Δt = cos θ (D-22)
v
∙ Suppose the distance between G and G’’ is one wavelength λ. At time t-Δt,
the wavefront intersects the x-axis at distance λx from G:
λ
= sin θ , λ = 2π / (ω / v) , λx = 2π / k x (D-23)
λx
2
vk x ⎛ vk ⎞
sin θ = and cos θ = 1 − ⎜ x ⎟ (D-24)
ω ⎝ ω ⎠

2
1 ⎛ vk ⎞
Δt = 1 − ⎜ x ⎟ Δz (D-25)
v ⎝ ω ⎠

∙ Given the change in traveltime Δt, the corresponding phase shift is -ω Δt. At
each Δz step of descent, we may propagate the plane wave with a different

Topics in Seismic Prospecting SNU 108


Seismic Data Processing

velocity v(z). The total phase shift to which the waveform was subjected with
arrival at P is − ∫ ωdt .
⎡ z ⎤
( ) ( )
2
⎢ ω ⎛ v z k ⎞
⎟ dz ⎥
P
exp −i ∫ ω dt = exp −i ∫ 1− ⎜ x
(D-26)
G ⎢ 0 v( z) ⎝ ω ⎠ ⎥⎦

∙ In the beginning, we consider the stratified earth with a velocity v(z).
∂ ω2
P ( k x , z , ω ) = −i − k x2 P ( k x , z , ω ) (D-27)
∂z v( z)
2

⎛ v ( z ) kx ⎞
2
ω
kz = 1− ⎜ ⎟ (D-28)
v( z) ⎝ ω ⎠
P ( k x , z , ω ) = P ( k x , z = 0, ω ) exp ⎡ −i ∫ k z ( z )dz ⎤
z
(D-29)
⎣⎢ 0 ⎦⎥

We need to check whether this solution satisfies the two-way scalar wave
equation.

∂2 dk ( z ) dv ( z ) ∂
P = −i z P − ik z ( z ) P
∂z 2
dv dz ∂z
(D-30)
dk ( z ) dv ( z )
= −i z P − k z2 ( z ) P
dv dz

If the velocity gradient dv(z)/dz is ignored, then the first term on the right
side drops out. The final expression then is
∂2 ∂2 ⎡ ω2 ⎤
P + k z2 ( z ) P = 0 , P + ⎢ 2 − k x2 ⎥ P = 0 (D-31)
∂z 2 ∂z ⎣v ( z)
2

¾ Phase-shift method

P ( x, z , t ) = ∫∫ P ( k x , 0, ω ) exp ( −ik z z ) exp ( −ik x x + iωt ) dk x dω


P ( x, z , t = 0 ) = ∫∫ P ( k x , 0, ω ) exp ( −ik x x − ik z z ) dk x d ω
(a) Start with the stacked section – an approximation to the zero-offset section
P(x, z=0, t), and perform 2-D Fourier transform to get P(kx, z=0, ω).
(b) By using equation (4-22), for each frequency w, extrapolate the transformed
wavefield P(kx, z, ω) at depth z with a phase-shift operator exp(-ikzΔz) to
get the wavefield P(kx, z+Δz, ω) at depth z+dz. At each z step, a new
extrapolation operator with the velocity defined for that z value is computed.
(c) As for any other migration, invoke the imaging principle t=0 at each

Topics in Seismic Prospecting SNU 109


Seismic Data Processing

extrapolation step to obtain the migrated section P(kx, z, t=0) in the


transform domain. The imaging condition t=0 is met by summing over all
frequency components of the extrapolated wavefield at each depth step.
This is easily shown from the integral representing the inverse Fourier
transform of the extrapolated wavefield
(d) Repeat steps (b) and (c) for downward continuation and imaging,
respectively, for all depth steps to get the migrated section in the transform
domain P(kx, z, t=0).
(e) Final step involves inverse transforming in the x direction to get the
migrated section P(x, z, t=0).

¾ Stolt migration

v 2
ω=k x + k z2
2
v kz
dω = dk z
2 k x + k z2
2

⎡v kz ⎤ ⎡ v 2 ⎤
P ( x, z , t = 0 ) = ∫∫ ⎢ ⎥P ⎢ k x , 0, k x + k z2 ⎥ exp(−ik x x − ik z z )dk x dk z
⎢⎣ 2 k x2 + k z2 ⎥⎦ ⎣ 2 ⎦
If the medium velocity is constant, migration can be expressed as a direct
mapping from temporal frequency w to vertical wavenumber kz.

Stolt’s algorithm for constant velocity thus involves the following steps:
(a) Start with the input wavefield P(x, z=0, t) approximated by the CMP stack,
and apply 2-D Fourier transform to get P(kx, z=0,ω).
(b) Map the wavefield from w to kz using the dispersion relation.
(c) Apply the scaling factor S as part of the mapping procedure.
(d) Invoke the imaging principle by setting t=0 and obtain P(kx, kz, t=0).
(e) Finally, apply 2-D inverse transform to get the migrated section P(x, z, t=0).

Topics in Seismic Prospecting SNU 110


Seismic Data Processing

f. Migration in practice

¾ Kirchhoff migration in practice


9 Aperture width
∙ Kirchhoff migration involves a summation of amplitudes along
diffraction hyperbolas.
∙ The spatial extent that the actual summation path spans, called the
migration aperture, is measured in terms of the number of traces the
hyperbolic path spans.
∙ The curvature of the diffraction hyperbola is governed by the velocity
function.
∙ Because of the vertical variation in velocity, aperture width generally is
time variant. (Fig. 4.2-2c)
∙ Migration of the shallow part of the section requires a narrow aperture,
while migration of the deep portion requires a wider aperture (Fig.4.2-
2c)
∙ The smaller the aperture, the less capable the migration is in collapsing
the diffraction hyperbola.
∙ The small-aperture migration causes smearing in the deeper part of the
section. This smearing effect destroys the dipping events and produces
spurious horizontally dominant events.
∙ Excessively large aperture means more computer time. More importantly,
large apertures can degrade the migration quality in poor signal-to-
noise ratio conditions.

FIG. 4.2-2. Summation paths for Kirchhoff migration in a medium with (a) low velocity (2000 m/s), (b) high
velocity (4000 m/s), and (c) vertically varying velocity. Migration aperture is small for low velocities and large
for high velocities.

Topics in Seismic Prospecting SNU 111


Seismic Data Processing

FIG. 4.2-3 (좌). Tests for aperture width in Kirchhoff migration: (a) a zero-offset section that contains a
diffraction hyperbola with 2500-m/s velocity, (b) desired migration using the phase-shift method; Kirchhoff
migration using (c) 35-trace, (d) 70-trace, (e) 150-trace, and (f) 256-trace half-aperture width.

FIG. 4.2-4 (우). Tests for aperture width in Kirchhoff migration: (a) a zero-offset section that contains dipping
events with 3500-m/s velocity, (b) desired migration using the phase-shift method; Kirchhoff migration using
(c) 35-trace, (d) 70-trace, (e) 150-trace, and (f) 256-trace half-aperture width.

FIG. 4.2-5. Tests for aperture


width in Kirchhoff migration:
Insufficient aperture width
causes removal of steeply
dipping
events.

Topics in Seismic Prospecting SNU 112


Seismic Data Processing

FIG. 4.2-6. Tests for aperture


width in Kirchhoff migration:
Insufficient aperture width
causes spurious horizontal
events in
deep, noisy part of a stacked
section.

9 Maximum dip to migrate


∙ Limiting the dip parameter is a way to reduce computational cost, since it
is related to aperture width, which determines the cost.
∙ The smaller the maximum allowable dip, the smaller the aperture. This
combination of maximum aperture width and maximum dip limit
determines the actual effective aperture width used in migration.

Topics in Seismic Prospecting SNU 113


Seismic Data Processing

FIG. 4.2-9. Tests for maximum dip to migrate in Kirchhoff migration: A low value for
maximum dip to migrate can be hazardous. All dips of interest must be preserved
during migration.

9 Velocity errors
∙ With increasingly lower velocities, the diffraction hyperbola is collapsed
less and less taking the shape of a frown. Æ it is undermigrated.
∙ With increasingly higher velocities, the diffraction hyperbola is inverted
more and more taking the shape of a smile Æ it is overmigrated.

Topics in Seismic Prospecting SNU 114


Seismic Data Processing

FIG. 4.2-10 (좌). Tests for velocity errors


in Kirchhoff migration: (a) a zero-offset
section that contains a diffraction
hyperbola with 2500-m/s velocity, (b)
desired migration; Kirchhoff migration
using (c) the medium velocity of 2500
m/s, (d) 5 percent lower, (e) 10 percent
lower, and (f) 20 percent lower velocity.

FIG. 4.2-11 (우). Tests for velocity errors


in Kirchhoff migration: (a) a zero-offset
section that contains a diffraction
hyperbola with 2500-m/s velocity, (b)
desired migration; Kirchhoff migration
using (c) the medium velocity of 2500
m/s, (d) 5 percent higher, (e) 10 percent
higher, and (f) 20 percent higher
velocity.

FIG. 4.2-12 (좌). Tests for velocity errors


in Kirchhoff migration: (a) a zero-offset
section that contains dipping events
with 3500-m/s velocity, (b) desired
migration; Kirchhoff migration using (c)
the medium velocity of 3500 m/s, (d) 5
percent lower, (e) 10 percent lower, and
(f) 20 percent lower velocity.

FIG. 4.2-13 (우). Tests for velocity errors


in Kirchhoff migration: (a) a zero-offset
section that contains dipping events
with 3500-m/s velocity, (b) desired
migration; Kirchhoff migration using (c)
the medium velocity of 3500 m/s, (d) 5
percent higher, (e) 10 percent higher,
and (f) 20 percent higher velocity.

Topics in Seismic Prospecting SNU 115


Seismic Data Processing

¾ Finite-difference migration in practice

9 Parabolic approximation method


∙ The parabolic approximation theoretically limits the algorithm to handling
dips up to 15 degrees. It can handle dips up to 35 degrees with sufficient
accuracy.
∙ 45 degree scheme can hanble steeper dips up to 80 degrees with
reasonable accuracy.

FIG. 4.3-2. (a) CMP stack, (b) desired migration by phase-shift method, (c) 15-
degree finite-difference migration. The finite-difference migration based on the
parabolic equation has the inherent property of undermigrating the steep flank
of the diffraction and the steeply dipping event.

9 Depth step size


∙ Increasing depth step size causes more and more undermigration at
increasingly steep dips. (FIG. 4.3-6, 7)
9 Velocity errors
9 Cascaded migration
∙ Larner and Beasley (1990) proposed performing migration using the 15-
degree equation, repeatedly – the input to the next migration stage being
the output from the previous stage.
∙ The more the number of stages in the cascaded migration the more the
improvement in focusing the energy. (FIG. 4.3-17)
∙ With a sufficient number of cascades, the 15-degree algorithm can actually
position the events as accurately as a 90-degree algorithm applied only
once.

Topics in Seismic Prospecting SNU 116


Seismic Data Processing

FIG. 4.3-6 (좌). Tests for extrapolation depth step size in 15-degree finite-difference migration: (a) a zero-
offset section that contains dipping events with 3500-m/s velocity, (b) desired migration; 15-degree finite-
difference migration using (c) 20-ms (d) 40-ms, (e) 60-ms, and (f) 80-ms depth step.

FIG. 4.3-7 (우). Tests for extrapolation depth step size in 15-degree finite-difference migration: (a) a zero-
offset section that contains dipping events with 3500-m/s velocity, (b) desired migration; 15-degree finite-
difference migration using (c) 4-ms (d) 8-ms, (e) 12-ms, and (f) 16-ms depth step.

Topics in Seismic Prospecting SNU 117


Seismic Data Processing

FIG. 4.3-16. The undermigration and overmigration effects


from Figures 4.3-14 and 4.3-15. B = dipping event before
and A = dipping event after desired migration, D
=diffraction before and D = diffraction after 15-degree
finitedifference migrations, and L = percent lower velocities,
and H = percent higher velocities.

FIG. 4.3-17. (a) A zero-offset section that contains a diffraction hyperbola with 2500-m/s velocity, (b)
desired migration using the phase-shift method, (c) 15-degree finite-difference migration using a depth
step of 20 ms; cascaded application of the 15-degree migration using (d) 4 cascades with 80-ms depth
step, (e) 10 cascades with 200-ms depth step, and (f) 20 cascades with 400-ms depth step.

Topics in Seismic Prospecting SNU 118


Seismic Data Processing

¾ Frequency-wavenumber migration in practice

9 Maximum dip to migrate


∙ As with the Kirchhoff summation method, migration with the phase-shift
can be limited to smaller dips by truncating the semicircular wavefront.
∙ This dip filtering capability is useful in rejecting coherent noise from the
stacked section while migrating the data.
∙ If migration is constrained to small dip values, then the steeply dipping
reflectors may be filtered out unintentionally.
9 Depth step size
9 Velocity Errors
9 Wraparound

∙ Wraparound is the effect of finite data length in time and space on a


migration algorithm implemented in the Fourier transform domain.
∙ A migration algorithm implemented in the time-space domain does not
suffer from wraparound effect.
∙ A migration algorithm implemented in the frequency-space domain
suffers from wraparound along the time axis.
∙ Similarly, a migration algorithm implemented in the frequency-
wavenumber domain suffers from wraparound effects both along the
time and space axes.

FIG. 4.5-23 (좌). Wraparound effect in frequency-wavenumber migration: The input stacked section is shown
in Figure 4.2-15a and the desired migration using the phase-shift method is shown in Figure 4.2-15b.

FIG. 4.5-24 (우). Wraparound effect in frequency-space migration: The input stacked section is shown in
Figure 4.2-15a and the desired migration using the phase-shift method is shown in Figure 4.2-15b.

Topics in Seismic Prospecting SNU 119

You might also like