METHODS OF EXPLORATION GEOPHYSICS

Lecture notes, AES1501

IDEA-League MSc program Applied Geophysics

Evert Slob, Guy Drijkoningen, and Ranajit Ghose

Delft University of Technology

Department of Geoscience & Eingineering / 2017
Contents

1 Introduction 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Potential fields in a homogeneous medium . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Diffusive fields in a homogeneous medium . . . . . . . . . . . . . . . . . . . . . . . . 2

1.4 Wave fields in a homogeneous medium . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Geo-electric method 5

2.1 The electric potential in a homogeneous space and a homogeneous half space . . . . 5

2.2 The electric potential difference for various measurement configurations on a homo-
geneous half space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.3 The electric current inside a homogeneous half space and 3D sensitivity . . . . . . . 15

3 Diffusive electromagnetic methods 21

3.1 Frequency and Transient EM methods . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 Interpretation of frequency EM data . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3.3 Interpretation of transient EM data . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4 Seismic imaging: From raw data to reflection images 27

4.1 Wave propagation, signal-to-noise ratio and resolution . . . . . . . . . . . . . . . . . 27

4.2 Sorting of seismic data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4.3 Noise suppression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

i
ii Contents

4.4 Normal move-out and velocity-model estimation . . . . . . . . . . . . . . . . . . . . 43

4.5 Noise suppression: marine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.6 Stacking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.7 Zero-offset migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5 To optimally recorded wave fields 63

5.1 The 5D character of 3D data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.2 Spatial sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5.3 Reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.4 3D acquisition geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.4.1 Single-fold subsets of 5D pre-stack wave field . . . . . . . . . . . . . . . . . . 67

5.4.2 Multi-fold subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5.4.3 Full 3D surveys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

6 Seismic characterisation: from wavefield to subsurface properties 75

6.1 Seismic characterization: targets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

6.2 P-waves, S-waves, Density and Poisson0 s Ratio . . . . . . . . . . . . . . . . . . . . . 76

6.3 Seismic velocities to properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

6.4 Seismic reflectivity to properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

6.5 Seismic intrinsic dispersion to properties . . . . . . . . . . . . . . . . . . . . . . . . . 84

6.6 Seismic velocity to strength of unconsolidated sandy soils . . . . . . . . . . . . . . . 85

6.7 Near-field seismic impedance to mapping changes in the near-surface . . . . . . . . . 86

6.8 Seismic monitoring to deduce changes in subsurface properties . . . . . . . . . . . . 86

6.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Chapter 1

Introduction

1.1 Introduction

In earth science and technology exploration geophysics employs potential field, diffusive field, and
wave field methods to obtain structural information about the subsurface, to characterize the sub-
surface layers in terms of physical and geological parameters, and to monitor subsurface processes
over time. Examples of potential methods are gravimetry, magnetometry and electric resistivity
methods, which employ fields that are assumed constant over the duration of the measurement and
for which a time-average value is obtained. An example of diffusive methods is the low frequency
electromagnetic method, where the space-time earth response is characterized by a spreading, or
diffusion, coefficient. Naturally occurring fields are called magneto-telluric fields and generated
fields are called controlled source fields. Examples of wave field methods are exploration seismol-
ogy, which uses sound waves that travel in the earth, and ground penetrating radar, which uses
electromagnetic waves that travel in the earth. For wave methods the space-time earth response is
characterized by a wave speed, or velocity. Three excellent text books on the methods of applied
geophysics are [1, 2, 3]. Textbooks that are of special interest for seismic and/or electromagnetic
waves and fields are [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21], for electro-
magnetic diffusive fields [22, 23, 24, 25] and for geo-electric and other potential methods [26, 27].
These references are but a small portion of the vast literature, but it provides a good start into the
subject.
These lecture notes contain brief introduction to the mathematics and physics underlying most
geophysical exploration methods. These techniques are used to obtain information about the sub-
surface. This information can take the form of an image of the subsurface, either structural or in
terms of detection and localization of objects. Through data analysis or inversion the information
can be given in terms of geophysical properties such as subsurface distributions of mass density, P-
en S-wave velocities, attenuation, and electric permittivity and conductivity. Combining geophys-
ical inversion with petrophysical models, the information van be given in terms of desired physical
properties such as lithology, porosity, water/oil/gas content, and hydraulic permeability.
The next section in this introductory chapter gives a description of potential fields, diffusive fields,
and wave fields in two and three dimensions. This chapter is followed by a chapter on the electrical

1
2 Introduction

resistivity method where several measurement techniques are discussed. Chapter 3 discusses diffu-
sive electric field methods in terms of application depended acquisition and data analysis and ends
with a brief discussion on data inversion. The following chapters all deal with wave methods and
seismic waves are discussed. Chapter 4 discusses the topic of seismic imaging and how to go from
raw data to reflection images. The steps commonly used to obtain a seismic image are described,
from data sorting, noise suppression via dip-filtering, velocity-model estimation and NMO correc-
tion, stacking and migration. Chapter 5 discusses the topic of acquiring seismic data to properly
acquire 3D wave fields. In this chapter, the basic steps are discussed of how to obtain optimally
recorded 3D seismic data, in order to get to 3D seismic images of the subsurface. The fundamental
issues of basic sampling intervals, reciprocity and the 5D character of 3D data are introduced. And
then the consequences of these issues are worked out for optimally recording 3D seismic wave-field
data. Finally, Chapter 6 treats subsurface characterization using seismic data. Considering both
deeper hydrocarbon exploration and shallow engineering investigations, the usual targets for char-
acterization are either static properties like bulk density, porosity, elastic stiffness/rigidity, strength,
compaction, or dynamic (generally time-varying) properties like in-situ stress, flow properties (e.g.,
permeability or hydraulic conductivity, viscosity), fluid saturation, etc. The layer-specific or spa-
tial variation of these properties is of interest. In some cases, local distributions are also crucial,
e.g., finding the distribution of contaminants, leachate flowpaths, overpressure zones at well loca-
tions or in hydrocarbon reservoirs. In this discourse, we exclude discussions on characterization of
discontinuities like fractures.

1.2 Potential fields in a homogeneous medium

Potential fields are regarded as fields that are independent of time and satisfy Poisson type equations

σ ∂x2 + ∂z2 G2D = −δ(x, z),

(1.1)

σ ∂x2 + ∂y2 + ∂z2 G3D = −δ(x),

(1.2)

where σ is the medium parameter controlling the distribution of the field, which in case of electric
resistivtiy method is the electric conductivity. The Green’s functions are given by
log(r)
G2Dφ (x) = , (1.3)
4πσ
1
G3Dφ (x) = − . (1.4)
4πσR
√
where r is two-dimensional distance r = x2 + z 2 and R is three dimensional distance R =
p
x2 + y 2 + z 2 and log denotes natural logarithm.

1.3 Diffusive fields in a homogeneous medium

Diffusive electromagnetic fields are characterized by the coupling of a second order space rate of
change to a first order time rate of change. Here the scalar diffusive field equation is given for the
1.4 wave fields 3

two- and three-dimensional Green’s functions in a homogeneous medium as,

∂x2 + ∂z2 − D∂t G2D = −δ(x, z, t),


(1.5)

∂x2 + ∂y2 + ∂z2 − D∂t G3D = −δ(x, t),

(1.6)
where D is the diffusion coefficient and x denotes the position vector in three-dimensional space.
The Green’s function is the impulse response and they are found as
exp(−Dr2 /(4t))
G2D (x, z, t) = . (1.7)
4πt
√
3D D
G (x, t) = exp[−DR2 /(4t)]H(t). (1.8)
(4πt)3/2
It can be seen that the diffusive field represented by the two- and three-dimensional Green’s func-
tions are Gaussian function for constant time, while for constant r and R and for large time values
they are inversely proportional to t in two dimensions and to t3/2 in three dimensions.

1.4 Wave fields in a homogeneous medium

Acoustic pressure and particle velocity interact when a mechanical disturbance occurs in a fluid,
electric and magnetic fields interact when an electromagnetic disturbance occurs, and elastic stress
and particle velocity interact when a mechanical disturbance occurs in an elastic solid material.
These interactions give rise to waves. Waves are characterized by the coupling of a space rate
of change in a field quantity and a time rate of change in the associated field together with the
reverse coupling of a space rate of change in the associated field and a time rate of change in the
field quantity. We assume all disturbances to have small magnitudes such that all interactions are
linear and we assume that the earth responds in a manner that a measurement can be repeated
without changing the outcome of the measurement. This implies that the earth can be regarded
a linear time-invariant system. In case the disturbances do not lead to the generation of heat, the
medium is called lossless, otherwise it is called dissipative, or lossy. In this section we give solutions
to wave equations in lossless and lossy media that have constant medium parameters. We look at
wave fields, which are generated by sources that are characterized in space and time by an impulse,
called delta or Dirac function, and the resulting wave fields are impulse response functions and
called Green’s functions. Because we have assumed the earth to behave as a linear time-invariant
system, responses to sources occupying a volume and being active over a longer duration of time
than spatial and temporal point the pressure or particle velocity is obtained by convolving the
Green’s function over the volume of the source and the time window in which it is active. This fact
is an expression of the principle of superposition.
In two dimensions the solutions to the two- and three-dimensional wave equations, given by
 
1
∂x2 + ∂z2 − 2 ∂t2 G2D = −δ(x, z, t), (1.9)
c
 
1 2
∂x + ∂y + ∂z − 2 ∂t G3D = −δ(x, y, z, t),
2 2 2
(1.10)
c
4 Introduction

are the two- and three-dimensional impulse response or Green’s functions G2D (x, z, t) and G3D (x, y, z, t)
given by

H(t − r/c)
G2D (x, z, t) = p , (1.11)
2π t2 − (r/c)2

δ(t − R/c)
G3D (x, y, z, t) = . (1.12)
4πR
It is observed that the two-dimensional Green’s function represents a wave field that has constant
amplitude on the circle in the (x, z)-plane, for which reason it is called a cylindrical wave field. It
is observed that the wave field represented by the three-dimensional Green’s function is constant
for fixed radial distance to the point source location and that the wave surface is a sphere. The
amplitude is proportional to the inverse of the radial distance to the source location with a pro-
portionality factor of 4π representing the solid angle of the full sphere. This is the surface area of
a sphere with unit radius over which the wave field spreads out as it travels away from the point
source. This amplitude decay behavior is known as the geometrical spreading factor.
Chapter 2

Geo-electric method

2.1 The electric potential in a homogeneous space and a homoge-

neous half space

The equation for the electric potentials can be derived from Maxwell’s equations when time varia-
tions are zero. Maxwell’s equations are given by

−∇ × H + (σ + ε∂t )E = 0, (2.1)

∇ × E + µ∂t H = 0, (2.2)

where E, H are the electric and magnetic field vectors, σ, ε, µ are the electric conductivity, per-
mittivity, and magnetic permeability, respectively, while ∂t represent differentiation with respect
to time. These equations reduce in the zero time variation limit when ∂t E = 0 and ∂t H = 0 to

−∇ × H + σE = 0, (2.3)

∇ × E = 0, (2.4)

We can take the divergence of equation (2.3) to remove the first part in the left-hand side of the
equation because the divergence of a curl is zero: ∇ · (∇ × H) = 0 for any (zero or non-zero)
magnetic field. This gives the equation

∇ · (σE) = 0. (2.5)

From equation (2.4) we can see that the electric field becomes curl-free and hence it can be written
in terms of the gradient of an electric potential V as

E(x, y, z) = −∇V (x, y, z). (2.6)

When we substitute this relation in equation (2.5) and replace conductivity σ by the inverse of
resistivity, ρ−1 , we find the equation for the electric potential V given by

∇ · (ρ−1 ∇V ) = 0, (2.7)

5
6 Geo-electric method

which written out in full gives

      
∂ ∂V ∂ ∂V ∂ ∂V
+ + = 0, (2.8)
∂x ρ∂x ∂y ρ∂y ∂z ρ∂z

when the resistivity is a function of position ρ = ρ(x, y, z). When the resistivity is a constant the
equation reduces to a Laplace equation
 2
∂2 ∂2

∂
+ + V = 0, (2.9)
∂x2 ∂y 2 ∂z 2

and the resistivity is no longer part of the equation. Such a medium is called a homogeneous
medium, whereas a medium where resistivity is a function of position is called a heterogeneous
medium. The solution to such equation where the potential originates from a point source in the
origin is well-known and can be written as

A
V (x, y, z) = + B, (2.10)
R
p
where the distance to the source is given by R = x2 + y 2 + z 2 and the constant B can be taken
zero by requiring that the potential goes to zero infinitely far way from the location of the source
of the potential field. To find the constant A we need to find out the influence of the source we use
to generate the electric potential. To do that we need the local version of Ohm’s law given by

J = ρ−1 E, (2.11)

in which J is the electric current density in A/m. Because a static electric field can be written as
the gradient of the electric potential this equation can be written as

J = −ρ−1 ∇V. (2.12)

To find the constant A we can think of a point at the origin where a current is injected and because
the spatial dependency of the potential is radial distance, the current will be distributed evenly
in all directions. Therefore, the surface of constant potential is a spherical surface as can be seen
from equation (2.10) as well when we find all the points where V is constant. Only the current
that is in the direction of the outward unit normal on any spherical surface can leave the spherical
surface and hence we can write the total electric current leaving that surface (and hence is going
into the earth) as I = 4πR2 n · J where n is the outward directed unit normal on any spherical
surface of constant potential and 4πR2 is the surface area of the sphere with radius R. We can
use equation (2.12) to express the electric current density in terms of the electric potential and
substitute the solution of the potential given in equation (2.10) to find

1 n·x 4πA
I = 4πR2 n · J = 4πR2 An · ∇ = 4πR2 A 3
= , (2.13)
R ρR ρ

because n = x/R and n · n = 1. Therefore

ρI
A= , (2.14)
4π
2.1 Potential due to stationary currents 7

surface potential distribution and current vectors

−2

−1.5
cross−line distance electrode spacings

−1

−0.5

0.5

1.5

2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
in−line distance in electrode spacings

a)

potential distribution and current vectors

0
depth in electrode spacings

0.5

1.5

2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
horizontal distance in electrode spacings

b)

Figure 2.1: Electric potential difference and electric current density vectors, a) on the surface and
b) in the vertical cross-section y = 0, z > 0, with two electrodes at x = −a/2 and x = a/2.
8 Geo-electric method

and the electric potential is given by

ρI
V (x, y, z) = , (2.15)
4πR
where the point of the current injection must be avoided so R > 0. The total current that is injected
in to the ground times the total resistance equals the electric potential, which is Ohm’s law. The
total resistance is given by the electric resistivity ρ divided by 4π times the radial distance from
the current injection point.
Let us look at the situation where there is a surface at z = 0 between a non-conductive air and the
conductive subsurface and the injection point is at the surface z = 0. In that case the current can
only go into the ground below the surface, hence for z > 0, and the surface area is 2πR2 , because
the current is now distributed over the surface area of half a sphere. Therefore the electric potential
anywhere in the half space z > 0 the electric potential is given by
ρI
V (x, y, z) = , (2.16)
2πR
where again R > 0. You see that the electric potential depends on the value of the electric resistivity
even though it does not occur in equation (2.10). This is because the electric potential depends
on the current strength and that in turn depends on the resistivity of the ground through which
this current must flow. When it is a constant it is merely a scaling parameter, but when it is a
function of position it can become a complicated relation that must be found numerically. Two
relatively simple examples are discussed in sections 2.4 and 2.5. Figure 2.1 shows a plot of the
electric potential V (x, y, z); 2.1a) for z=0 and 2.1b) for y = 0, z > 0. The arrows in the plot
indicate the vector directions of the electric current. For this configuration we have the electric
potential given by

∆V (x, y, z) = V (x − a/2, y, z) − V (x + a/2, y, z),

!
ρI 1 1
∆V (x, y, z) = p −p , (2.17)
2π (x − a/2)2 + y 2 + z 2 (x + a/2)2 + y 2 + z 2

where the point x = a/2 is the point of current injection (current goes into the ground) and the
point x = −a/2 is the point where the current is taken out of the ground for which reason the
potential related to that location is negative. To make the current run in the subsurface the two
points must be connected to a current source above the ground through an electronically controlled
connection with a battery of other charge storage/current producing device. This is because electric
current can only run in closed loops. If we take very small distances between the two injection
points we can see this equation as a central finite difference expression of V (x − x0 , y, z) at x0 = 0
with distance a
∂ V (x + a/2, y, z) − V (x − a/2, y, z)
0
V (x − x0 , y, z) ≈ − , (2.18)
∂x a
which is equivalent to a central finite difference of V (x, y, z) at x with distance a

∂ V (x + a/2, y, z) − V (x − a/2, y, z)
− V (x, y, z) ≈ , (2.19)
∂x a
2.1 Potential due to stationary currents 9

and hence we can approximate equation (2.17) for R  a as

ρIx
∆Vf f (x, y, z) ≈ , (2.20)
2πR3
where R is the distance from the point (x, y, z) to the midpoint between the current electrodes
and the subscript f f means this is the far field expression valid for very large distances. This
approximation is correct for all points below the midpoint between the electrodes and the worst
case scenario is for points at the surface where the approximation has an error of less than 1% for
distances R ≥ 5a. For this reason the two current electrodes are called an electric current dipole.
Most measurements are done within this distance in which case it is not really acting as an electric
dipole.
Even though the left-hand side of equation (2.17) is expressed as a potential difference, which
it is, it is not a measurable quantity because this difference is caused by two current electrodes
at different locations and not by actually measuring a potential difference between two points.
The two current electrodes are necessary to make an electric current run in the ground through
the battery or other source. The potential difference can be measured at two locations by using
electrodes, called potential electrodes, at those locations and connect them to the two terminals of
a measuring device. The two other electrodes are used to inject an electric current in the ground,
known as the current electrodes. The general expression of the measured potential difference is
obtained using equation (2.17) and can be written as
∆V = V (x1 − a/2, y1 , z1 ) − V (x1 + a/2, y1 , z1 ) − V (x2 − a/2, y2 , z2 ) + V (x2 + a/2, y2 , z2 ),

ρI 1 1
∆V = p + p
2π (x1 − a/2)2 + y12 + z12 (x2 + a/2)2 + y22 + z22
!
1 1
− p −p , (2.21)
(x1 + a/2)2 + y12 + z12 (x2 − a/2)2 + y22 + z22
where the potential at the location (x2 , y2 , z2 ) is subtracted from the potential at (x1 , y1 , z1 ).
For a measurement configuration at the surface let the four electrodes be located along the x-axis,
then the two current electrodes are at locations C1 being the current injection point and C2 being
the current extraction point, while the potential electrodes are located at P1 and P2 and we subtract
the potential at P2 from the potential at P1 . Then the voltage difference between the potential
electrodes is given by

  2 2
Iρ 1 1 1 1 Iρ X X
∆V = + − − = (−1)m+n |Pm − Cn |−1 .
2π |P1 − C1 | |P2 − C2 | |P2 − C1 | |P1 − C2 | 2π
m=1 n=1

(2.22)
A possible configuration is illustrated in Figure 2.2 for a two-layer subsurface with all electrodes
just touching the earth surface to have a good contact with the earth.

Exercises
10 Geo-electric method

2.1.1 Under which conditions can the electric field vector be written as the gradient of a scalar
potential function?

2.1.2 What is the difference between a homogeneous and a heterogeneous medium?

2.1.3 Explain from the property of the gradient operator what the electric field vector direction is
relative to the lines of equipotential. Show this by carrying out the gradient on the function
R−1 .

2.1.4 Use the result in equation (2.16) in equation (2.6) to derive the expression for the electric
field vector in the half space.

2.1.5 Use the result in equation (2.16) in equation (2.12) to derive the expression for the electric
current density in the half space.

2.1.6 Explain from Figure 2.1 in which direction the electric current flows relative to the value of
the electric potential.

2.1.7 Make a Matlab script that computes the electric potential given in equation (2.17) as a
function of x/a for y = 0 and z = 0 and avoid the points x = ±a/2 by taking −3a/2 ≤
x/a ≤ 3a/2 using 100 points and take ρI = 2π. Plot the resulting potential V as a function
of distance normalised by the electrode spacing, x/a, and limit the vertical axis between -10
and 10. Give a physical explanation for the curve you see in the plot.

2.1.8 Deduce from your plot in the previous exercise, your knowledge of the electric field lines
associated with static electric charges, and Figure 2.1, which of the two terminals connected
to the current locations is the positive electrode.

I C2
C1 P2
P1
ΔV

Layer 1

Equipoten1al lines
Layer 2
Lines of electric current

Figure 2.2: Example four-electrode acquisition configuration in a two-layer model (modified from
eo-miners.eu).
 2.2 Expressions for potential differences in arrays 11

2.1.9 What is the decay rate of the electric field strength as a function of large distance to the
current injection point at the surface of a homogeneous half space? Use your result from
exercise 2.1.4. This is the electric field from a monopole source with constant current.

2.1.10 What is the decay rate of the electric field strength as a function of large distance to the
midpoint of two current injection points spaced a distance a apart at the surface of a homo-
geneous half space? Use the result of equation (2.20) in equation (2.6). This is the far field
electric field behaviour from a dipole source with constant current.

2.2 The electric potential difference for various measurement con-

figurations on a homogeneous half space

Several different configurations have been developed in the times when data was collected with
four electrodes which then had to be moved every measurement to either obtain a profile or a
so-called vertical sounding. A profile is obtained by taking a fixed distance between all electrodes
and moving them all four without changing their mutual distance. Then an average over a certain
depth range is obtained, while lateral changes are mapped. A vertical electric sounding (VES) is
obtained by keeping the midpoint between the electrodes fixed and changing the mutual distance
an integer multiple times the first electrode distance. Then no lateral resolution is obtained but
variations in depth are mapped. Both concepts are not correct. When there are lateral changes
in resistivity also the depth range that influences the measurement changes. When the electrode
spacing increases, the total volume of sensitivity increases and not only in the depth direction. For
this reason and the availability of modern multi-electrode arrays and measurement systems lateral
profiles are combined with VES profiles and a 2D subsurface image is obtained through a s-called
inversion process that will be briefly described in the next chapter. Here we describe the five most
commonly used arrays, the Pole-Pole, Pole-Dipole, Dipole-Dipole, Wenner and Schlumberger ar-
rays.

Pole-Pole Array
The name Pole-Pole already indicates that both the current injection electrodes and the potential
electrodes are modelled as single poles. In practise this is impossible, but from equation (2.22) we
see that when the position of one of the electrodes is very large compared to the others, two of
the distance terms become very small and can then be neglected. Let us assume that C1 → −∞
and P1 → ∞, and that the distance between C2 and P2 is |P2 − C2 | = a, then the Pole-Pole
measurement is described by
Iρ
∆V P P = . (2.23)
2πa
In practise the far away electrodes are put more than 20 times the largest electrode spacing away.
This array is reciprocal, which means that the current and potential electrodes can be interchanged
without changing the result of the measurement. This is clearly true for a homogeneous half space,
but it can be shown that it is always true. This is a general property for all four-electrode configu-
rations. The Pole-Pole array is also even, which means that while keeping the far away electrodes
the close potential and current electrodes can be interchanged without changing the result of the
12 Geo-electric method

measurement. Since the electrode distance a can be varied, this array is very suitable for modern
multi-electrode multi-channel systems.

Pole-Dipole Array
For this array, the name indicates that either one of the current or one of the potential electrodes
is the far away electrode. let us take one of the current electrodes to be the far away electrode,
while both the potential electrodes are used in the line of measurements. Let us further assume
that the second current electrode is the far away electrode, C2 → ∞. Now we have an extra degree
of freedom to put the other three electrodes in the line. Let us denote the distance between the
two potential electrodes a and the distance between the current electrode and the closest potential
electrode na then we have for the measured potential difference
 
PD Iρ 1 1 Iρ
∆V = − = . (2.24)
2π na (n + 1)a 2πn(n + 1)a

This array is of course also reciprocal but not even in principle, which means that the current elec-
trode cannot be put at the other side of the potential electrode pair without changing the result
of the measurement. It is an even array only when the medium is shift-invariant in the horizontal
directions. This implies that independent measurements can be done by putting the current elec-
trode to the other side of the potential electrode pair. Since both the potential electrode distance
a and the multiple electrode distance na can be varied, this array is very suitable for modern
multi-electrode multi-channel systems.

Dipole-Dipole Array
Now all four electrodes are put in the line at measurable distances of each other. The array has
received its name because both current electrodes are next to each other at distance a, while the
potential electrodes are also next to each other at distance a and separated from the current elec-
trode pair at a distance na. In principle, there is an extra degree of freedom if we would separate
either the current electrodes or the potential electrodes not at distance a, but at, say, ma. This
is not usually done practice, but it is possible with modern multi-electrode systems. The optimal
use of this extra degree of freedom is subject of ongoing research. Taking the above choices leads
to the measured potential difference as
 
Iρ 1 1 2 Iρ
∆V DD = + − = . (2.25)
2π na (n + 2)a (n + 1)a πn(n + 1)(n + 2)a

This array is clearly even, independent of the subsurface conductivity distribution. It is also ob-
served that the measured potential difference is proportional to n−3 and goes to zero an order of
magnitude faster than the measurement of the Pole-Dipole array. Consequently, this array is more
prone to noise at large n-factors than the Pole-Dipole array. Since the electrode distance na can
be varied, this array is very suitable for modern multi-electrode multi-channel systems.

Wenner Array
Wenner introduced this array and it was used a lot because of the large signal strength. Even
though all four electrodes are put at measurable distances from each other they are put all at
fixed distance a and the current electrodes are put on the outside of the array, while the potential
2.2 Expressions for potential differences in arrays 13

electrodes are put in the middle of the array. The measured voltage difference is
 
W Iρ 2 1 Iρ
∆V = − = . (2.26)
2π a a 2πa
This array is clearly even, independent of the subsurface conductivity distribution. It is also
observed that the measured potential difference is the same as for the Pole-Pole array in this homo-
geneous half space configuration. In more general heterogeneous half spaces, the two measurements
are very different. The Wenner array has lost most of its popularity because the electrode distances
are fixed and the potential electrodes must be at the same distance as the current electrodes. Mod-
ern multi-channel systems can measure potential differences between several (nine, for eight-channel
systems) different electrodes. The Wenner array is typically a single-channel measurement. It is
used still in situations where signal strength is the most important factor, like in situations where
the current electrodes are put in a very high resistivity top layer. Then it is difficult to put a
reasonable amount of current in the ground and the measured potential differences are small.

Schlumberger Array
This array was invented by the Conrad Schlumberger in his pioneering work in the 1920’s. He made
the geo-electric measurement a common and popular exploration tool when the Schlumberger com-
pany was very successful with the method. The current electrodes are put on the outside of the
array at distance (2n + 1)a from each other, while the potential electrode pair is put at distance
a in the middle of the array. Then the distance from each of the current electrodes to the closest
potential electrode is na. This array can be seen as a generalization of the Wenner array, since the
Wenner array is obtained for n = 1. The measured potential difference, for arbitrary n-values is
then
 
S Iρ 2 2 Iρ
∆V = − = . (2.27)
2π na (n + 1)a πn(n + 1)a
This array is also clearly even, independent of the subsurface conductivity distribution. In most
modern multi-channel systems this array is a compromise as it has good signal strength and can
be used in the so-called reversed mode, where the current electrodes are in the middle of the array
and the outer potential electrodes can be used in pairs. Only half the channels can be used because
the potential electrodes are not sequenced, but half of them are on the right of the array, while the
other half are on the left side of the current electrodes.
The concept of apparent resistivity
The homogeneous half space response of equation (2.22) can be used as a reference response for
any kind of measurement with four electrodes on the earth surface. Independent of the actual
subsurface resistivity distribution there will always be an equivalent homogeneous half space that
would represent the measurement as well. The value of the equivalent homogeneous half space
resistivity is called the apparent resistivity. When the geometry of the four electrodes changes, the
array becomes sensitive to a different part of the subsurface and when the value of the apparent
resistivity changes it means that the real subsurface is heterogeneous in its resistivity. The definition
of apparent resistivity is unambiguous and can be written for surface electrode geometries in a
general form given by
∆V
ρapp = k , (2.28)
I
14 Geo-electric method

where the factor k is the geometrical factor depending on the array type used. The geometrical
factor is given for each array as

Pole-Pole k = 2πa,

Pole-Dipole k = 2πn(n + 1)a,

Dipole-Dipole k = πn(n + 1)(n + 2)a,

Wenner k = 2πa,

Schlumberger k = πn(n + 1)a,

Exercises
2.2.1 If the factor 2πa∆V S /(Iρ) should be more than 1/30 to have a reasonable measurement,
which factors n can be used in the three arrays where n occurs?
2.2.2 If the contribution from a remote electrode in the field should be less than 3% of the smallest
contribution from the other three electrodes, how far away should it be to have a reasonable
approximation to the pole-dipole array?
2.2.3 Show that for any four-electrode array you can use in the field you will get the same mea-
surement result if you interchange the current and potential electrodes.
2.2.4 In reality the measurements that should theoretically be the same in Exercise 2.2.3 are not
the same. This can be due to effects that are not part of the model, such as currents and
potential distributions that do not come from your source but are naturally there. Would
interchanging source and receiver electrodes and averaging the results help to reduce this
noise? What about reversing the current direction for the same purpose, what will happen
and how can you use it to improve the average measured potential difference with minimal
effects from the noise?
2.2.5 Find the expression for the potential difference in a dipole-dipole array when the electrode
distances between the current and potential electrodes are l and L, respectively, and the
distance between the closest current and potential electrodes is A.
2.2.6 Find the expression for the potential difference in a Schlumberger array when the potential
electrodes are separated by a distance a and displaced from the midpoint between the current
electrodes by a distance x, while the distance between the current electrodes is A.
2.2.7 Simplify the result from the previous exercise for x = 0 when the potential electrodes are
centrally positioned around the midpoint between the current electrodes.
2.2.8 Use equation (2.21) with two current electrodes at distance a apart with a borehole in the
middle and the measurements taken between two potential electrodes at distances z1 and z2 .
What are the measured values for such set up?
 2.3 current distribution & 3D sensitivity 15

2.2.9 Let the borehole in the previous exercise be located a distance a away from the midpoint
between the current electrodes and along the line through them. One of the potential elec-
trodes is located at the surface and the other is at some depth z2 . If you use equation (2.21)
the potential at which electrode should you subtract from the potential at the other electrode
to have a positive value for the measured potential difference? What are the two values in
between which the measured potential difference will always lie?

2.2.10 How does the measured potential difference change as a function of increasing depth of the
point z2 ? For a depth z2 = 2a what is the ratio between is the voltage difference increase and
the measured value at that depth?

2.3 The electric current inside a homogeneous half space and 3D

sensitivity

From equation (2.11) it is clear that currents run along the electric field lines and the ratio of
the magnitude of their vectors is represented by a single number independent of the direction of
the vectors when the electric resistivity is a scalar function. Such a medium is called an isotropic
medium, whereas in an anisotropic medium the electric current vector and the electric field vector
are not necessarily aligned and the ratio of their magnitudes can not be expressed by a constant, but
will be direction dependent. In an anisotropic medium the resistivity is a 3 × 3 matrix to describe
the angle dependence and the angle-dependent length differences between the electric current and
electric field vectors.
Inside the ground the current gets distributed and it is useful to look at the expression for the
current density defined in equation (2.12). Substituting the expression for the potential given in
equation (2.17) and carrying out the derivatives gives

cos(φ− ) cos(θ) cos(φ+ ) cos(θ)

     
Jx (x, y, z)
 Jy (x, y, z)  = I  sin(φ− ) cos(θ)  − I  sin(φ+ ) cos(θ)  ,
− 2
(2.29)
2π(R ) 2π(R+ )2
Jz (x, y, z) sin(θ) sin(θ)

with

R± =
p
(x ± a/2)2 + y 2 + z 2 ,

x ± a/2 = R± cos(φ± ) cos(θ),

y = R± sin(φ± ) cos(θ),

z = R± sin(θ),

where φ± is the azimuth angle in the horizontal plane relative to the x-axis seen from x = ∓a/2,
and θ is the dip angle with respect to the horizontal plane. Equation (2.29) shows that the injected
current I gets distributed over three-dimensional space and is inversely proportional to distance
squared from each injection point with an angle dependence in the vector directions. It can also
be observed that the distribution of the total current in the homogeneous earth does not depend
16 Geo-electric method

1
0.9
0.8
0.7
0.6
Jx/Jx(z=0)

0.5
0.4
0.3
0.2
0.1
0
0 0.5 1 1.5
z/a

Figure 2.3: Normalised current density as a function depth divided by the constant electrode
distance a below the midpoint between the current electrodes on a homogeneous half space.

on the earth resistivity. If the earth is heterogeneous it will of course depend on the resistivity
distribution in the earth.
The point (x, y, z) with position vector x can be anywhere in the subsurface and if we take the
points on the vertical plane through the middle of the two current electrodes the point x lies in the
(y, z)-plane at x = 0. This results in an electric current density with only a non-zero x-component
given by

I a
Jx (0, y, z) = − , (2.30)
2π (a /4 + y 2 + z 2 )3/2
2

where the minus-sign indicates the current runs in negative x-direction. The normalised current
density as a function of depth at y = 0 in equation (2.30) is shown in Figure 2.3 in which the
normalisation is done through dividing by Jx (0, 0, 0). From this figure we can see that the current
strength has reduced just below 10% at one electrode spacing depth. On the other hand, if you
have a target depth of investigation you can find the electrode spacing at which the current strength
has a maximum at that depth.
Another important number in the analysis of sensitivity to subsurface changes in resistivity is the
fraction of current that flows through any particular area between two depth levels z1 and z2 in
the subsurface. If we define the rectangular domain somewhere in the (y, z)-plane bounded by the
depth levels z1 and z2 , the total current that flows through that plane is given by the integral of
2.3 current distribution & 3D sensitivity 17

the current density normal to that plane over the whole plane. For our particular plane it is given
by
Ia z2 ∞ 1
Z Z
I(z1 , z2 ) = − dydz,
2π z1 −∞ (a /4 + y 2 + z 2 )3/2
2

in which the integral over y is solved using the following rule

∂ χ 1 χ2 α
=p − 2 3/2
= 2 .
(χ + α)3/2
p
∂χ χ + α
2 χ + α (χ + α)
2

Hence the integral over y can be solved in this way with χ = y and α = a2 /4 + z 2 as
Z ∞ +∞
1 1 y 2
dy = = 2 , (2.31)
2 2
−∞ (a /4 + y + z )
2 3/2 (a /4 + z ) (a /4 + y + z ) −∞ a /4 + z 2
2 2 2 2 2 1/2

because the y-dependent fraction has limits of ±1 for y → ±∞. We can now evaluate the remaining
integral
Ia z2 1
Z
I(z1 , z2 ) = dz,
π z1 a /4 + z 2
2

which is an elementary integral and the result is given by

2I
I(z1 , z2 ) = (atan(2z2 /a) − atan(2z1 /a)) ,
π
in which the function atan is also known as arctangent or inverse tangent. The factor
I(z1 , z2 ) 2
= (atan(2z2 /a) − atan(2z1 /a)) , (2.32)
I π
is the fraction of the total injected current that runs through this plane region below the midpoint
between the electrodes and the depth levels z1 and z2 . Also this fraction can be maximised by
optimising the electrode distance.
The different arrays have different sensitivities to contrasts in the subsurface relative to the back-
ground conductivity. The sensitivity to a subsurface change in resistivity is determined by the
current strength at that location. Sensitivity is a scalar function and for a Pole-Pole array, with
electrode distance a, for a small contrasting object at point (x, y, z) it is given by

[∇V (x − a/2, y, z)] · ∇V (x + a/2, y, z)]

F3d (x, y, z) = , (2.33)
[ρ(x, y, z)I]2
which is a normalised quantity and the normalisation factor is the total current strength. The
points (x − a/2, 0, 0) and x + a/2, 0, 0) denote the two electrode locations at the surface. For a
generally heterogeneous subsurface the gradients have to be calculated numerically, which requires
a full forward model run. In a homogeneous half space we know the expression for the electric
potential and we can evaluate the sensitivity functions and it is given by

(x − a/2)(x + a/2) + y 2 + z 2
 
1
F3d (x, y, z) = 2 , (2.34)
4π [(x − a/2)2 + y 2 + z 2 ]1.5 [(x + a/2)2 + y 2 + z 2 ]1.5
18 Geo-electric method

2D slice of 3D Pole−Pole sensitivity function

−4
0
−3
normalized X−line distance to array midpoint

−0.5
−2

−1 −1

0
−1.5

1
−2
2

−2.5
3

4 −3
−4 −3 −2 −1 0 1 2 3 4
normalized in−line distance to array midpoint

Figure 2.4: A depth slice of the three-dimensional sensitivity function for the Pole-Pole array at a
depth of electrode distance a.

from which the sensitivity functions of all the other arrays can be easily constructed by superposi-
tion. As an example of a cross-section of the sensitivity at a fixed depth, the lateral sensitivity is
shown in Figure 2.4. The sensitivity functions for all standard arrays mentioned above are given
in Figure 2.5, where cross-sections are given in the line of the electrodes (inline sensitivity) and a
cross-section through the midpoint of the line (cross-line sensitivity).
Exercises

2.4.1 What are the lengths of the vectors in equation (2.29)?

2.4.2 Show mathematically that Jy (0, y, z) = 0 and Jz (0, y, z) = 0. Why should they be zero from
a physical point of view?

2.4.3 Use equation (2.30) at y = 0 and at a constant depth z to find the electrode spacing a that
maximises the current density strength at that depth.
2.3 current distribution & 3D sensitivity 19

2D slice of 3D Pole−Pole sensitivity function 2D slice of 3D Pole−Pole sensitivity function

0.5 0.5
1 1
normalized depth

normalized depth
1.5 1.5
2 2
2.5 2.5
3 3
3.5 3.5
4 4
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
normalized in−line distance to array midpoint normalized X−line distance to array midpoint

−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 −2.5 −2 −1.5 −1 −0.5 0 0.5 1

2D slice of 3D Pole−Dipole sensitivity function 2D slice of 3D Pole−Dipole sensitivity function

0.5 0.5
1 1
normalized depth

normalized depth
1.5 1.5
2 2
2.5 2.5
3 3
3.5 3.5
4 4
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
normalized in−line distance to array midpoint normalized X−line distance to array midpoint

−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 −2.5 −2 −1.5 −1 −0.5 0 0.5 1

2D slice of 3D Dipole−Dipole sensitivity function 2D slice of 3D Dipole−Dipole sensitivity function

0.5 0.5
1 1
normalized depth

normalized depth

1.5 1.5
2 2
2.5 2.5
3 3
3.5 3.5
4 4
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
normalized in−line distance to array midpoint normalized X−line distance to array midpoint

−2.5 −2 −1.5 −1 −0.5 0 0.5 1 −2.5 −2 −1.5 −1 −0.5 0 0.5 1

2D slice of 3D Schlumberger (Wenner) sensitivity function 2D slice of 3D Schlumberger (Wenner) sensitivity function

0.5 0.5
1 1
normalized depth

normalized depth

1.5 1.5
2 2
2.5 2.5
3 3
3.5 3.5
4 4
−4 −3 −2 −1 0 1 2 3 4 −4 −3 −2 −1 0 1 2 3 4
normalized in−line distance to array midpoint normalized X−line distance to array midpoint

−2.5 −2 −1.5 −1 −0.5 0 0.5 1 1.5 −2.5 −2 −1.5 −1 −0.5 0 0.5

Figure 2.5: Inline (left) and cross-line (right) three-dimensional sensitivity function for a line Pole-
Pole, Pole-Dipole, Dipole-Dipole and Wenner-Schlumberger array (top to bottom).
 20 Geo-electric method

2.4.4 Show that the y-dependent part in the middle term of equation (2.31) indeed goes to ±1
when y → ±∞.

2.4.5 Find the depth level for constant electrode spacing such that half the current runs between
the surface and that depth level and the other half between that depth level and infinity.

2.4.6 For given depth levels z1 and z2 find the electrode spacing that maximises the fraction of
current as expressed in equation (2.32) that runs through the entire planar region between
z1 and z2 .

2.4.7 Show that atan(a) + atan(1/a) = π/2 independent of a and find a such that atan(a) −
atan(1/a) = π4 . Hint: use the goniometric rule for atan.

2.4.8 How large is the current fraction that runs in the region atan(a) − atan(1/a) = π4 . defined in
exercise 2.4.7? Can you relate a to the result found in exercise 2.4.6?

2.4.9 How large is the current fraction that runs between the surface and the top of the region and
which fraction runs below the region defined in exercise 2.4.7?

2.4.10 Explain why the current fraction running between surface and the top at z1 of a region
between z1 and z2 , using any z1 and z2 > z1 values, is equal to the current fraction running
between the bottom at z2 and infinity, while maintaining the optimal electrode spacing found
in exercise 2.2.6.
Chapter 3

Diffusive electromagnetic methods

3.1 Frequency and Transient EM methods

Induction methods, whether in time domain or frequency domain are found at any scale, from
metal detectors to search for coins in beach sands, to a high-powered time-domain electromagnetic
sounding survey to resolve the structural conductivity distribution in the subsurface for mineral
exploration. In the early days, explorations were carried out to detect very conductive base metal
massive sulfide ores in a highly resistive host rock. Because of the high contrasts mineral exploration
surveys have been the prime application of the electromagnetic methods and the mineral exploration
industry has funded most of the developments. In those days, standard operation techniques, which
were especially designed for these applications, employed a steady state input signal, generated by
an alternating electric current, in a loop or a straight wire. It was soon found out that not all
massive sulfide ores are very conductive and not all their host rocks highly resistive. Modern
equipment operate well in the frequency and time domains, each with their specific purposes. Due
to the increased interest, many manufacturers have adapted the equipment to shallow geophysical
applications. To give an indication of the ranges in conductivities encountered in earth materials,
Fig. 3.1 is taken from [23, 24] and shows the typical conductivity ranges of earth materials, classified
to their geological background. As can be seen from the figure, the typical range of most rock
conductivities spans at least a decade and conductivity contrasts between rocks can be extremely
high. Diffusive EM methods mostly use thin-wire loops as source and receiver. Frequency and
transient EM methods exist. Frequency methods employ a sinusoidally varying voltage difference
that is impressed between two points of the coil. This generates a sinusoidally varying current in the
coil. The field resulting from this current induces a current in the receiver coil and in the earth, the
response of the earth also induces an electric current in the receiver coil, which results in a voltage
difference between the two points where the measurement is taken. Transient methods employ an
impressed constant voltage between two points of the source loop, that is suddenly switched on or
off. This generates an almost step switch-on or step switch-off current in the coil which then acts
like a transformer. The field directly couples in with the receiver coil and the earth response also
induces a current in the receiver loop.

21
 22 Diffusive electromagnetic methods

3.2 Interpretation of frequency EM data

Frequency soundings are carried out to get an estimate of the resistivity (inverse of conductivity)
distribution in the ground. Most commercially available systems use magnetic dipoles for both
source and receiver, operate at fixed frequencies and have fixed source-receiver configurations.
These systems do not allow any choice in designing an acquisition configuration for a survey. The
commercial value of these systems comes from their low cost, acquisition speed and the apparent
resistivity concept for data interpretation. The apparent resistivity concept relates the measured
potential, using the known input source current, to a value of resistivity assuming the earth is
a homogeneous half-space. This value is by no means an average value but is directly related
to the size of the loops, their horizontal distance and the frequency of operation. The apparent
resistivity is a parameter which reflects the actual changes of the resistivity with depth better than
the measured voltage does. Even in the case when there is enough data for more sophisticated
processing, apparent resistivities can be used as an initial model for such techniques. There is no
simple expression for the apparent resistivity. There are several definitions that we will describe
here.
Definition of apparent resistivity
A unique definition of apparent resistivity can be defined using the vertical magnetic magnetic
2 Introduction
and horizontal angular electric field measured at the earth surface and generated by a vertical

massive sulfides igneous and

unweathered
metamorfic rocks rocks
graphite
igneous rocks: mafic felsic mottled duricrust
zone weathered
rocks
metamorfic rocks
gravel and sand
clays glacial
tills sediments

shales sandstone conglomerate

sedimentary
rocks
lignite, coal dolomite, limestone
salt water fresh water permafrost
water

sea ice

102 10 1 10−1 10−2 10−3 10−4 10−5

conductivity (S/m)

Figure 1.1: Typical ranges of conductivities of earth materials.

Figure 3.1: Typical ranges of conductivities of earth materials.

ration surveys have been the prime application of the electromagnetic methods and the mineral
3.2 Interpretation of FDEM data 23

magnetic dipole also located at the earth surface. Apparent resistivity is then defined as the integral
over frequency, or equivalently the integral over source-receiver distance, of the weighted difference
between the two fields and thus the computation of the apparent resistivity involves multi-frequency
of multi-offset acquisition geometries and two field components to be measured. Here we only give
existing definitions that require only one field component to be measured at a single frequency
for a single offset. Any definition uses normalized quantities relative to the free space response,
the so-called mutual impedance ratios. The first definition definition uses the mutual impedance
ratio and does not give the apparent resistivity in closed form but as an optimization problem. The
second one does provide an explicit expression for apparent resistivity and is obtained by employing
a Taylor expansion on exp(−γd r) around γd r → 0, with γd = (iωσµ)1/2 , and substitute the result
in the expressions of the induced voltages normalized by the free-space response. The results are
called mutual impedance ratios and are given by,

2
Ẑ HCP = [9 − (9 + 9γd r + 4(γd r)2 + (γd r)3 ) exp(−γd r)], (3.1)
(γd r)2

2
Ẑ V CP = {(γd r)2 − 3 + [(γd r)2 + 3(γd r + 1)] exp(−γd r)}. (3.2)
(γd r)2

In both these expressions the difference between the measured data and the right-hand sides of
equations (3.1) or (3.2) can be minimized by finding the best possible conductivity according to
some norm, where usually a root mean square error norm is used. The second definition uses a
approximation to the above equations by assuming that the product of conductivity and frequency
is small. This is only accurate for low conductivities, or high resistivities. Carrying out the limiting
procedure we find,

Ẑ HCP = 1 + iωσ1 µ0 (r)2 /4 + O([γd r]3 ), (3.3)

|γd r|↓0

Ẑ V CP = 1 + iωσ1 µ0 (r)2 /4 + O([γd r]3 ). (3.4)

|γd r|↓0

This limit is also known as the low induction asymptote, where the induction number B is defined
as,
γd r p
B= = ωσµ0 /2r. (3.5)
1+i
It is observed that, apart from the apparent resistivity of the vertical co-axial system, all systems
have the same low induction number asymptote. Hence, they yield the same value for the apparent
resistivity, given by

ωµ0 r2
ρapp = . (3.6)
4|Im{Ẑ HCP ;V CP }|
Several commercial systems are based on this result, e.g., the EM31, EM34 and EM38 induction
tools from Geonics only give imaginary part of the mutual impedance ratio or the apparent re-
sistivity values based on this part of the measured field. The fact that one usually gets different
24 Diffusive electromagnetic methods

T R
V V

σ1

σ2

σ1

Figure 3.2: Simple three layer model with two different conductivity values

HCP ρ from low induction number HCP meausered potential

app
6000 0.015

5000
apparent resistivity [Ω.m]

0.01
4000
ℑ{ZHCP}

3000

2000 0.005
14.6 kHz 14.6 kHz
9.8 kHz 9.8 kHz
6.4 kHz 6.4 kHz
1000 1.6 kHz 1.6 kHz
400 Hz 400 Hz

0 0
0 5 10 15 20 25 30 35 40 1 1.0005 1.001 1.0015 1.002 1.0025 1.003 1.0035 1.004
loop distance [m] ℜ{ZHCP}

Figure 3.3: Apparent resistivity using low induction number approximation (left) and a close up of
the true response for the smaller offsets in a phasor plot (right).

values for the apparent resistivity using the two different methods is an indication that the volume
over which is averaged is not the same. Consider the configuration of Figure 3.2. The model is a
layered earth, consisting of a 20 cm thick low resistivity layer, σ2 = 0.1 S/m, in a homogeneous
half space of σ1 = 10−3 S/m. The top of the low conductivity layer is 60 cm below the surface. In
Figure 3.3, where the apparent resistivity is plotted per frequency as a function of loop separation
for the HCP-configuration, together with the true field response for this situation. The frequencies
used are used by the Geonics instruments and corresponding offsets. It is clear that the computed
apparent resistivities do not resemble the subsurface resistivity. Another derivation does not lead
to a parametric representation of the apparent resistivity in terms of the measured field. Since the
half-space response is non-linearly related to frequency of operation, coil separation and conductiv-
ity, the best single value one can obtain for apparent resistivity comes from projecting the data onto
the curve one would obtain when the earth would be a homogeneous half-space. In this sense, the
apparent resistivity is a first guess in a parametric inversion procedure. A simple method to achieve
this projection is by minimizing the squared error norm (least squares method), where the error is
given by the difference between the mutual impedance of the real measurement and the equivalent
3.3 Interpretation of TDEM data 25

4 HCP ρ app from low induction number HCP meausered potential

x 10 0.1
2 500 Hz
500 Hz
1000 Hz
1000 Hz 0
1.8 2000 Hz 2000 Hz
4000 Hz 4000 Hz
−0.1 8000 Hz
1.6 8000 Hz
16 kHz
16 kHz
32 kHz −0.2 32 kHz
1.4
apparent resistivity [Ω.m]

64 kHz
64 kHz
128 kHz 128 kHz
1.2 −0.3
256 kHz
256 kHz

ℑ{ZHCP}
−0.4
1

−0.5
0.8

0.6 −0.6

0.4 −0.7

−0.8
0.2

−0.9
0 −0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4
0 50 100 150 200 250
loop distance [m] ℜ{ZHCP}

Figure 3.4: Apparent resistivity using low induction number approximation (left) and a close up of
the true response for the smaller offsets in a phasor plot (right).

mutual impedance of a true homogeneous half-space computed from the known expressions given
in equations (3.1)-(3.2). This can be done for a single frequency and offset and the scheme searches
the resistivity that minimizes the squared error norm. This resistivity then is the desired value of
the apparent resistivity. This approach gives reasonable values for all induction numbers, since the
exact expressions for the half-space responses are used. Plotted as a function of frequency and/or
as a function of coil separation and position, the curves reflect the variation of the resistivity with
depth and thereby give a global insight in the subsurface variations. Systems, like the GEM-II
from Geophex, that do keep the total measured field and can operate at several frequencies exist
and the increased reliability of the final apparent resistivities using this last apparent resistivity
concept at different frequencies outweigh the extra amount of effort in data acquisition. That this
is indeed necessary in general is illustrated by Figure 3.4, where for the same model as used in Fig-
ure 3.3 the apparent resistivity is shown in the left graph and the true measured complex mutual
impedance ratio in a phasor plot in the right graph. The most used frequency range is used in the
plots together with reasonable offsets. The mutual impedance ratio is clearly in general a complex
number that is not well represented by the low induction approximation. This then also implies
that parametric inversion of the conductivity distribution is necessary for a high-resolution image
of the subsurface conductivity distribution. The apparent resistivities are clearly out of bound
relative to the actual resistivity values in the model.

3.3 Interpretation of transient EM data

Transient EM methods are used for the same reasons as frequency methods, but have the advan-
tage over frequency domain methods that the source is either constant or switched off during the
measurements. As is the case in the frequency domain, there is no expression for the apparent resis-
tivity that is valid for all times for loop responses. The mutual impedance ratio for the horizontal
26 Diffusive electromagnetic methods

coincident loop is given by

"  1/2  3/2 #
3 4τd π 4τd
e−τd /t − e−τd /t .
p
Z HCI = erf( τd /t) − (3.7)
2τd πt 6 πt

Now we can again define apparent resistivity by optimizing this expression relative to the measured
data finding the best conductivity value in some error-norm. Explicit expressions for apparent
resistivity can be defined in two ways, one is obtained from the early-time asymptote of the mutual
impedance ratios and the other is obtained from the late-time asymptote. Only the error function
contributes to the early-time limits of all the mutual impedance ratios and goes to unity. Hence
the early-time apparent resistivities is easily obtained as,

ρHCI;et
app = 61 µ0 a2 Z HCI (t ↓ 0). (3.8)

The apparent resistivities using the late-time asymptotes can be directly obtained from the late-
time expressions for the potential by normalizing it to the free-space response,
!2/3
3/2
3µ a3
ρHCI;lt
app = √ 5/2 0HCI . (3.9)
10 πt Z (t → ∞)
Chapter 4

Seismic imaging: From raw data to

reflection images

In this chapter, the basic steps are discussed of how to obtain a seismic reflectivity image from
seismic records. We will discuss the steps, commonly used to obtain a seismic image and common to
seismic data gathered on land (on-shore) as well as at sea (off-shore): sorting, noise suppression via
dip-filtering, velocity-model estimation and NMO correction, stacking and (zero-offset) migration.

4.1 Wave propagation, signal-to-noise ratio and resolution

A major goal of exploration seismics is obtaining 3D structural subsurface information from seismic
data. Seismic processing concerns itself with removing or compensating for the effects of waves that
propagate through the earth such that an image is obtained from the subsurface. Generally, ”noise”
is always present in raw seismic records. The task is to obtain an image of the subsurface from
noisy data.
We can consider two ways of introducing seismic processing to a newcomer. One is in terms of
wave theory. We have to understand the physical processes that are involved all the way from
the seismic source, through the subsurface, to the seismic recording instrument. We have to try
to obtain only those features which are due to the structure of the subsurface and not related to
other features. For instance, we want to know the source signal we put into the earth such that we
can compensate for it from our data later: the structure of the subsurface does not depend on the
source we use. In this way we can remove or suppress certain unwanted features in the image.
Another way of introducing seismic processing to a newcomer is more in terms of the image we
obtain: signal-to-noise ratio and resolution. In order to see the image we need to have at least
a moderate signal-to-noise ratio. We would like this ratio to be as large as possible by trying to
suppress unwanted features in the final image. The other aspect of the final seismic image is the
resolution: we would like the image to be as crisp as possible. As you may know, these two aspects
cannot be seen separately. Usually, given a certain data set, an increase in signal-to-noise ratio
decreases the resolution (as information is stacked together), and also an increase in resolution (by

27
28 Seismic Imaging

correctly incorporating wave theory) has normally the consequence that the signal-to-noise ratio
gets worse. In seismic processing we would like to obtain the optimum between the two: a good,
although not perfect, signal-to-noise ratio with a good resolution.
In these notes we take the view of trying to understand each process in the wave problem, and try
to find ways to cope with them. In this way we want to increase the signal-to-noise ratio, perhaps
at some cost with respect to resolution. This is a very important characteristic of raw seismic data:
it has a (sometimes very) poor signal-to-noise ratio, and it needs a lot of cleaning up before the
reflection image of the subsurface can be made visible. It is along this line that we will discuss
seismic processing: trying to understand the physical processes. Sometimes, we will refer to the
effect it can have on the total signal in terms of signal-to-noise ratio and resolution.
With seismic processing, we have many physical processes we have to take into account. Actually,
there are too many and this means that we must make simplifying assumptions. First, we only
look at reflected energy, not at critically refracted waves, direct body waves, surface waves, etc. Of
course, these types of waves contain much information of the subsurface (e.g. the surface waves
contain information of the upper layers) but these waves are treated as noise. Also critically
refracted waves contain useful information about the subsurface. That information is indeed used
indirectly in reflection seismics via determining static corrections, but in the seismic processing
itself, this information is thrown away and thus treated as noise. Another important assumption in
processing is that the earth is not elastic, but acoustic. In conventional processing, we mostly look
at P-wave arrivals, and neglect any mode-conversion to S-waves. Some elastic-wave processing is
done in research & development environments, but are still very rarely used in production. Money
is better spent on 3-D ”P-wave” seismics, rather than on 2-D ”elastic” seismics; 3-D seismics
with three-component sources and receivers is still prohibitively expensive in seismic exploration,
although in seismic monitoring it is getting more common.
As said previously, the conventional way of processing is to obtain an image of the primary P-
wave reflectivity, so the image could be called the ”primary P-wave reflectivity image”. All other
arrivals/signals are treated as noise. As the name ”primary P-wave reflectivity” suggests, multiples
are treated as noise (as opposed to ”primaries”); S-waves are treated as noise (as opposed to P-
waves); refractions are treated as noise (as opposed to reflectivity). Therefore, we can define the
signal-to-noise ratio as:
S Signal Primary P-wave Reflection Energy
= = (4.1)
N Noise All but Primary P-wave Reflection Energy
In this chapter we will consider one method of noise suppression, typically useful for 2D or 3D
seismic data: dipfiltering. It makes use of the difference in velocity of signal (desired events) and
noise (undesired events).
It can be seen from the definiton of S/N (equation (4.1)) that processing of seismic data is to cancel
out and/or remove all the energy which is not primary P-wave reflectivity energy, and ”map” the
reflectivity in depth from the time-recordings made at the surface. In terms of total impulse
response of the earth G(x, y, t), we want to obtain that part of the impulse response of the earth
which is due to primary P-wave reflections:

Processing
G(x, y, t) → Gprimary,P-wave,reflectivity (x, y, z) (4.2)
?? wave propagation 29

The most important information that must be added to the data, is the seismic velocity. This is
crucial for obtaining a proper image. In this chapter it is discussed how to obtain a first estimate
of the seismic velocities of the subsurface, and how to use this information to make the final image.
The problem can also be seen as being information we measure at the surface, which is a function
of time, is mapped to the correct position in depth. In other words, we want to convert ”time”-data
to ”depth”-data.
In this chapter, we will look at a basic processing sequence to obtain a seismic image from the raw
noisy seismic data. The steps that will be considered here are common to seismic processing of
data gathered on land (on-shore) as well as at sea (off-shore). They are: sorting, noise suppression,
velocity-model estimation and NMO correction (invoking velocities for imaging), stacking and mi-
gration (again using velocities for focussing the energy). This is a basic processing sequence; it does
not mean that this will always give a good image: on land topography effects and the first layer can
be the largest problem and then have to be dealt with separately (via so-called satic corrections);
at sea the source wavelet is not always a clean one and one has to compensate for such effects (via
so-called signature deconvolution).
30 Seismic Imaging

Figure 4.1: Shot gather measurement.

4.2 Sorting of seismic data

Common-shot and common-receiver gathers

When data are shot in the field, we record the shots sequentially. So by a record we mean all the
recordings from the sensors for a single shot experiment. Normally, the measurement for one source
at one receiver location is called a trace, which is a time series of reflections. It is obvious that for
each shot we will order these recordings (traces) by increasing (or decreasing) offset. The offset is
defined as the distance from source to receiver. A simple simulated example of such a shot is given
in figure 4.1. In this figure on the left hand side the ray paths from source to the receivers of the
seismic waves are shown. Note that due to the different velocities in the different layers, the ray
paths are bended according to Snell’s law. For this record, one shot consists of the explosion from
one charge of dynamite (supposing it is measured on land). The data are stored in the recording
instrument and then put onto a storage device, record by record.
When the next shot is fired, we do the same, record with the instrument and then write the data
onto the storage device. We say that the data are shot ordered. A section as shown in figure 4.1 is
commonly called a common-shot gather: we show the recorded wave fields for one shot.
It can be guessed that if we talk about shot ordered data, we could also have receiver ordered
data. This is indeed the case. One could get all the shots together, of course in an increasing
shot position, belonging to one receiver position. Such a gather is called a common-receiver gather.
However, this assumes that during acquisition the same receiver position is covered by different
shots. In practice, we often make use of reciprocity: interchanging source and receiver will give
exactly the same response (if the directional properties of the source and receiver can be considered
4.2 Data sorting 31

identical). In fact figure 4.1 can also be considered as a common-receiver gather, where all ray
paths from different shots come together at one receiver position.
Why should we need these distinctions? A nice feature about a common-shot gather is to see
whether a receiver position has a higher elevation than its neighbors and thus gives an extra time
shift in its record. This effect is called ”statics”. Therefore common-shot gathers are good for
detecting geophone statics. In the same way, we can see on common-receiver gathers whether a
shot was set deeper than the neighboring shot positions, and therefore common-receiver gathers
are good for detecting shot statics.

Figure 4.2: Midpoint definition in between sources and receivers.

Common-midpoint gathers
The way of organizing the data in common-shot gathers is just a consequence of the logistics in the
field, but for some processing steps it is not a convenient sorting of data. A commonly used way
of sorting the data is in common-midpoint gathers. A mid-point is here defined as the mid-point
between source and receiver position. An illustration of the mid-point is given in figure 4.2. We
gather those traces that have a certain midpoint in common, like in figure 4.2, the record from
receiver 3 due to shot 1, and the record from receiver 1 due to shot 2. Once we have gathered
all the traces with a common-midpoint (CMP) position, we have to decide how to order these
records for one CMP, and the logical choice is to order them by increasing (or decreasing) offset.
A gather for one mid-point position with the traces for increasing (or decreasing) offsets is called a
common-midpoint gather. Figure 4.3 shows a CMP gather for the same subsurface model as figure
4.1.
For what reason is the common-midpoint gather convenient? The most important one is for stack-
ing which we shall discuss in one of the next sections. Suppose the earth would consist of horizontal
layers as depicted in figure 4.4. Then the geometrical arrival from shot to receiver all reflect right
below the midpoint between the source and receiver, and thus the reflection points in the subsurface
then only differ in depths. With other words, all the reflections measured at the different offsets in a
CMP gather carry information on the same subsurface points (below the midpoint position). If we
would make a correction for the offset dependence of the travel time for each trace, the reflections
from the same place would arrive at the same time for all the traces, and thus we could add the
traces together to increase the signal-to-noise ratio. This process is called NMO correction and
stacking respectively, as will be discussed later. This argumentation is not valid for common-shot
gathers since the reflection points in the subsurface do not coincide for each trace (for a horizon-
32 Seismic Imaging

Figure 4.3: Common-midpoint gather.

Figure 4.4: Common-shot and common-midpoint gather for horizontally layered earth.

tally layered earth). For a laterally varying medium also for a CMP the reflection points do not
coincide, but they are coming still coming from a small region, as shown in figure 4.3, and the
stacking procedure will often still give acceptable results.

Common-offset gathers
4.2 Data sorting 33

Figure 4.5: Zero offset gather.

Figure 4.6: Relation between different sortings of seismic data.

As can be expected, we can also form a common-offset gather, a gather in which we collect all those
source-receiver pairs that have a certain offset in common. Usually, we shoot with fixed distances
between source and receivers, and so we will have as many traces in our common-offset gather as
there are shots, thus often quite a large amount. For the model of figure 4.1 and figure 4.3 the
zero-offset configuration (i.e. source and receivers at the same positions) is shown in figure 4.5.
Note that in the zero-offset section the general structures can already be recognized. Common-
offset gathers are used in pre-stack migration algorithms since it can give a check on velocities.
Migrating a common-offset gather for a small offset should give the same image as a migration of
such a gather for a large offset, otherwise the velocity used in the migration is not the right one.
A graph combining all this information is given in figure 4.6. Here we assumed we have recorded
along a line in the field, which we call the x-direction. Also, we have assumed that we have 10
34 Seismic Imaging

receiver positions with the first receiver at the source location (i.e. at zero offset). On the horizontal
axis we have plotted the x–coordinate of the source (xs ), while on the vertical axis we have put the
x–coordinate of the receiver (xr ). Then, each grid point determines where a recording has taken
place. In this graph a column represents a common-shot gather, and a horizontal line a common-
receiver gather. A common-midpoint gather is given by the line xs + xr = constant, which is a line
at 45 degrees with a negative slope. A common-offset gather is given by the line xs −xr = constant,
which is a line of 45 degrees but now with a positive slope.
What can be noticed in the graph, is that we started out with 10 receiver positions for each shot,
while the CMP gather contains only 5 traces. Why that so? This can be seen in figure 4.2. When
we shift one source position to the next, we actually shift two CMP’s because the distance between
each CMP is half the source spacing. So a factor two is involved. On the other hand there are
twice as many CMP gathers, as the total of traces in the survey is of course constant.
In figure 4.6 we assumed the spacing between the shot positions and the receiver positions were
the same but this does not need to be so. This also influences the number of traces in a CMP. The
number of traces in a CMP is called the multiplicity or the fold. It can be shown easily that the
multiplicity M is:
Nr Lr
M= = (4.3)
2∆xs /∆xr 2∆xs

in which Nr is the number of receivers per shot, ∆xs is the spacing between the shot positions,
∆xr is the spacing between the receivers, and Lr is the length of the receiver line (=Nr ∆xr ).
In the above argumentation there is still one assumption made, and that is that the earth is
horizontally layered. When the earth is not like that, the reflection points do not coincide any
more, see figure 4.3. Still, the results obtained with this assumption are very good, it only gets
worse results when the dips of the layers of the earth are becoming steep. We will come to that
later on when discussing stacking.
4.3 Noise suppression 35

4.3 Noise suppression

In the first section 4.1 it was stated that one of the goals of seismic processing is to remove ”noise”:
an important characteristic of seismic data is that they have a (sometimes very) poor signal-to-noise
ratio, and they need to be cleaned up significantly before the reflection image of the subsurface
can be made visible at all. ”Noise” in this case is most often source-generated, meaning that we
have generated it ourselves; one could think of critically refracted waves, direct body waves, shear
waves, multiply reflected waves, surface waves, etc. Of course, these types of waves contain much
information of the subsurface (e.g. the surface waves contain information of the upper layers) but
for the imaging of reflectors that part of the data is not used. In this chapter we will focus on
how we could remove, or at least suppress significantly, such noise elements via a technique which
depends on the 2D/3D character of seismic-reflection data, the so-called f-k or dip filtering.

The two-dimensional Fourier transform (in relation to waves)

Previously, the temporal and spatial Fourier transforms were defined. Commonly for reflection
imaging for exploration purposes, we deal with two- and three-dimensional data. At one sensor
position, we record the signal in time, and we record at many sensor positions, so we sample the
time and the horizontal space coordinate(s). In chapter 1, the spatial Fourier transformation was
defined for full space, so including the vertical coordinate, but here we transform only the horizontal
coordinates.
In obtaining the seismic image of the subsurface, we heavily depend on the multi-dimensional nature
of our data. In this section, we shall only focus on two-dimensional data or analysis, i.e., in time t
and the one horizontal coordinate x. Some operations take place directly in the (t, x) domain, but
others are done in the so-called (f, kx ) domain where the f and kx are the transform variables with
respect to time t and space x, respectively. As well known, f is the temporal frequency, and kx
is the (angular) spatial frequency in the horizontal direction where we now specifically denote the
subscript x to denote that we are sampling in one horizontal Cartesian coordinate direction, namely
x. In the analysis we use the wavenumber in the same definition as the frequency, so without the
factor 2π, and denote this by Kx , so:

kx = 2πKx (4.4)

The two-dimensional Fourier transform of a signal v(x, t) is then defined as:

Z ∞ Z ∞
ṽ(Kx , f ) = v(x, t) exp[2πi(Kx x − f t)] dt dx. (4.5)
−∞ −∞

For seismic applications, the two-dimensional transformation is not a simple representation since
the quantities on the axes are different: space and time. But, space and time are related, due to
the fact that we deal with a wave phenomenon: a wavefront is propagating through the earth, and
we record it at the earth’s surface. Therefore, the quantity Kx should be interpreted also in terms
of waves. For a sinusoidal signal, the frequency f is the number of cycles per unit time; the same
is valid for the spatial frequency Kx . In the same way, Kx is the number of wavelengths per space
36 Seismic Imaging

unit, so Kx is related to the horizontal wavelength λx by:

1
Kx = . (4.6)
λx
When we substitute the relation:

cx = f λx , (4.7)

in which cx denotes the wave speed in the x–direction, we obtain for the spatial wavenumber Kx :
f
Kx = . (4.8)
cx

Let us look at the following example to illustrate some of the above. Suppose we have a linear
event in x − t as depicted in figure 4.8. In general, the event can be defined as:

v(x, t) = s(t − x/cx ), (4.9)

in which s(t) is some characteristic function for the source signal. Applying a two-dimensional
Fourier transform, i.e. a Fourier transformation with respect to time as well as the space-coordinate
x:
Z ∞Z ∞
ṽ(Kx , f ) = s(t − x/cx ) exp[2πi(Kx x − f t)] dt dx. (4.10)
−∞ −∞

Putting τ = (t − x/cx ), we have that dt = dτ , at t = −∞, τ = −∞, and at t = +∞, τ = +∞.

From this it follows that:
Z ∞Z ∞
ṽ(Kx , f ) = s(τ ) exp[−2πif τ ] exp[−2πix(f /cx − Kx )] dτ dx. (4.11)
−∞ −∞

The latter expression can be considered as the product of two integrals, the first one being a Fourier
transform with respect to the variable τ only, i.e.:
Z ∞
ŝ(f ) = s(τ ) exp[−2πif τ ] dτ, (4.12)
τ =−∞

and the second integral being a Fourier transform with respect to the variable x only, so that the
total expression (4.11) can be written as:
Z ∞
ṽ(Kx , f ) = ŝ(f ) exp[2πix(Kx − f /cx )] dx. (4.13)
x=−∞

Now, because the δ–function is the inverse transform of the constant value 1, we have:
Z ∞
δ(Kx ) = exp[2πixKx ] dx. (4.14)
x=−∞

The integral transform in (4.13) can be recognized as a delayed δ–function of Kx . Therefore it can
be expressed as:

ṽ(Kx , f ) = ŝ(f )δ(Kx − f /cx ) (4.15)

4.3 Noise suppression 37

This expression can be recognized as the product of a complex frequency spectrum ŝ(f ) (which is
a function of the frequency f only!) and a ”razor-blade”–like two-dimensional δ–function: δ(Kx −
f /cx ), which is zero everywhere in the (Kx , f ) plane, except on the line Kx − f /cx = 0, where it
assumes infinite amplitude. Note also, that the dip of the event in the (f, Kx ) domain is reciprocal
to the one in the time domain. For the linear event in the (t, x) domain, we obtain:

∆t 1
= , (4.16)
∆x cx
while in the (f, Kx ) domain we obtain:

∆f
= cx . (4.17)
∆Kx

In the above example we saw that the wave s(t − x/cx ) mapped onto the function ŝ(f )δ(Kx − f /cx )
in the (f, Kx ) domain. But as we have seen in the definition of the spatial wavenumber (eq.(4.8)),
a frequency component f is included. This is the reason why often a substitution of Kx = f p is
done, in which p is called the horizontal slowness or ray parameter. The latter name is often used
because it relates to the parameter that is constant across an interface when using Snell’s law :
1/cx = sin θ/c. When using Kx = f p, the forward transformation (eq.(4.10)) reads:
Z ∞Z ∞
ṽ(f p, f ) = s(t − x/cx ) exp[2πif (px − t)] dt dx. (4.18)
−∞ −∞

where we now recognize the linear event s(t − x/cx ) from above in the exponent. The linear event
maps in the (p, f ) domain as:

ṽ(f p, f ) = ŝ(f )δ(p − 1/cx ) (f 6= 0) (4.19)

In this domain, the wave s(t−x/cx ) becomes only non-zero for the constant ray parameter p = 1/cx .
The type of transformation described here is often used in the so-called Radon transformation. The
Radon transformation exploits the wave character of our seismic data.
In figure 4.7 three plane waves are displayed with increasing ray parameter, the time and space
sampling being 4 ms and 10 m respectively. Clearly, the effect of aliasing can be observed both
in the time domain (positive dip looks as negative dip) as well as the wavenumber domain (wrap
around along the wavenumber axis).
In figure 4.8 we have created three linear events with different velocities. When we transform
the time-axis to the frequency domain, we obtain an interference pattern from which we cannot
recognize very much. the last picture gives the (f, Kx ) spectrum by Fourier transforming with
respect to the horizontal coordinate.
In the following, we make use of this domain, where we can make some nice classification based on
its characteristics in this domain. We have seen in the previous that a dipping broadband event
in the (t, x) domain gives a dipping event in the (f, Kx ) domain. The dip is not the reciprocal
of the velocity as in the (t, x) domain, but the velocity itself. We shall discuss two applications
which make use of these characteristics of different slopes in the (f, Kx ) domain, together with
some other properties in this domain. These applications are the removal of ground roll in land
38 Seismic Imaging

(a) p = 0 s/m (b) p = 0.3 · 10−3 s/m (c) p = 2.0 · 10−3 s/m

(d) p = 0 s/m (e) p = 0.3 · 10−3 s/m (f ) p = 2.0 · 10−3 s/m

Figure 4.7: Three discretised plane waves of frequency 30 Hz with ray parameter p = 0, p = 0.3·10−3
and p = 2.0·10−3 s/m respectively in the time domain (upper part) and in the wavenumber domain
(lower part). The time sampling is 4 ms and the spatial sampling is 10 m.

data, and removal of multiples in marine data. The latter will be treated in a later section, after
the so-called NMO has been treated in the next section.

Noise suppression in land data: Ground roll filtering

Let us first look at ground roll in land data. Ground roll is the general term for surface waves which
travel along the surface, and only if the earth would be a homogeneous half space, the ground roll
would be a perfect Rayleigh wave, giving one event in the measurements. Since this is not the case
for the real earth, it is a combination of resonances which show a very dispersive character due to
interferences of the different modes.
At the same times that we receive the ground roll, we get the reflections back from the subsurface,
so they overlap. A difference between these two types of arrivals, i.e. the ground roll and the
4.3 Noise suppression 39

Figure 4.8: A linear event in (t, x) (left), (f, x) (middle) and (f, Kx ) (right).

Figure 4.9: Ground roll and a reflection in the (f, Kx ) domain.

reflections, is that they travel with a different speed along the surface. The ground roll travels
along the surface with a relatively low speed, while the reflections from below arrive nearly at the
same time for a close group of receivers with a slight hyperbolic move-out because in reflection
seismology we are aiming at reflections from small angles of incidence. In the (t, x) domain, we
have a response:
x
vGR (t, x) = sGR (t − ) (4.20)
cGR
where sGR is the wavelet for the ground roll, where the subscript GR stands for ground roll. For
the reflected events, we have the response in the (t, x) domain:

x sin(θ)
vRef l (t, x) = sRef l (t − ) (4.21)
cRef l
40 Seismic Imaging

(a)
(b)

Figure 4.10: Simulated shot record with dispersive ground roll. a) Shot record with ground roll
and two reflections in (t, x) domain. b) Shot record of a) in (f, Kx ) domain.

where we made use of the fact that the apparent speed along the surface for the reflected events
is cRef l / sin(θ), where θ denotes the angle the normal on the wave front makes with the vertical.
Since θ is small, the apparent velocity is large. In principle the reflection events are located within
a pie-slice in the (f, Kx ) domain for apparent velocities between cRef l and ∞. We can now make
a sketch of these two arrivals in the (f, Kx ) domain, as is given in figure 4.9. A nice feature of this
diagram is that the ground roll and reflections are now well separated.
A shot record with dispersive ground roll, simulated in an earth model with a few thin layers on
top, together with two reflections is displayed in figure 4.10. Note the separation of the ground roll
and reflections in the (f, Kx ) domain.
The use of the two-dimensional Fourier transform enables us to carry out filtering in the (f, Kx )
domain. In the case of one-dimensional filtering we are used to perform multiplication of the
complex spectrum of an input signal with the complex spectrum of a filter function, being equivalent
to convolution in the time domain:

Ft [v(t) ∗ h(t)] = v̂(f )ĥ(f ) (4.22)

in which v(t) is the input time response (seismic data), and h(t) the filter response. Similarly, in
the (f, Kx ) domain, filtering may be carried out through multiplication with some window function
or a more general function of Kx and f , i.e.,:

Ftx [v(t, x) ◦ h(t, x)] = ṽ(f, Kx )h̃(f, Kx ), (4.23)

4.3 Noise suppression 41

(a) (b)

Figure 4.11: ”Pie-slice” filters in the (t, x) and (f, Kx ) domain.

where Ftx stands for the two-dimensional Fourier transform, and ◦ stands for two-dimensional
convolution. The two-dimensionality of the (f, Kx ) domain provides the extra freedom to design
2D (or 3D when the data are also a function of y) filters. In particular, filters with boundaries being
linear functions of Kx and f , i.e. filters with sector-shaped pass-band or rejection-band boundaries
(see figure 4.11).
These filters, called ”pie-slice” filters of which the (linear) sector boundaries are specified by ap-
parent velocities, may be designed in such a way that, say, reflection energy is passed in a sector
covering the f –axis (apparent velocity is ∞) and a range of relatively high apparent velocities on
both sides of the f –axis, while suppressing energy from the regions further away from the f –axis,
e.g. in the region between the +Kx –axis and a line through the origin with velocity parameter
cGR = f /kx,GR somewhat larger than the highest apparent velocity of the low-velocity ground roll
events. This is shown in figure 4.12 in which the example of figure 4.10 has been filtered such that
the ground roll is removed (except for some edge effects).
A field example of a section with ground roll and without ground roll, removed by (f, Kx ) filtering,
is shown in figure 4.13. Note that the reflection events (e.g. at 0.8 and 1.5 seconds) become better
visible after the filtering. However, in the lower part of the data, some of the ground-roll appears
to be spatially aliased, and the (f, Kx ) filtering procedure will smear this aliased ground roll. It
will be difficult to make a distinction between smeared ground roll and true reflection events in
that region.
So far, we established a separation of the arrivals in the (f, Kx ) domain by means of different
apparent velocities, but the process is even more accentuated by the fact that ground roll usually
contains much lower frequencies than the reflection events; it can sometimes even happen that
42 Seismic Imaging

(a) (b)

Figure 4.12: Simulated shot record with dispersive ground roll after (f, Kx ) domain pie-slice filter-
ing. a) Filtered shot record in (t, x) domain. b) Filtered shot record of a) in (f, Kx ) domain.

this separation can be achieved only on frequency considerations, but even in that case (f, Kx )
”pie-slice” filtering has the preference.
4.4 NMO & velocity analysis 43

(a) (b)

Figure 4.13: Field example of a shot gather (after statics correction) with ground roll and after
removal of ground roll by dip-filtering.

4.4 Normal move-out and velocity-model estimation

NMO curve for single interface

The most important physical constant needed for obtaining an accurate image of the subsurface,
is the velocity of the medium. We record our data at the surface in time, and what we wish to
obtain is an image of the subsurface in depth. The link between time and depth is of course the
wave velocity. Unfortunately, it is not so easy to obtain a good velocity model and this is often an
iterative process. In this section we will discuss the effect of the velocity on the obtained data. We
will first discuss some simple models in order to understand the features we can encounter in real
data. As a consequence of this, we will discuss which operations have to be applied to the data in
order to obtain the desired information. We assume here that we deal with a CMP gather.
Let us first consider the reflection from a single interface as depicted in figure 4.14. The time for
the geometrical ray from source to receiver is given by:

R (4d2 + x2 )1/2
T = = (4.24)
c c
in which x is the source-receiver distance, R is the total distance traveled by the ray, d is the
thickness of the layer and c is the wave speed. When we write 2d/c as T0 , then we can write this
44 Seismic Imaging

Figure 4.14: a) Distances in a subsurface model with one flat reflector. b) NMO curve for geometry
of a) with depth 300 m and velocity of 1500 m/s. The dashed line is the parabolic approximation
of the hyperbola.

as:

1/2
x2

T = T0 1 + 2 2 (4.25)
c T0

note that this function describes a hyperbola. We can see that we have an extra time delay due to
the factor x2 /(c2 T02 ). The extra time delay is called the Normal Move Out, abbreviated to NMO.
This extra term is solely due to the extra offset of the receiver with respect to the source; at
coincident source-receiver position this term is zero. Often, the square-root term in this equation
is approximated by its one-term Taylor series expansion, i.e.:

x2
T ' T0 + . (4.26)
2c2 T0

Figure 4.14b shows the travel time curve for a layer of 300 meter depth and a velocity of 1500 m/s.
The dashed line in this figure shows the parabolic approximation according to equation (4.26).
In seismic processing we are not interested in the extra time delay due to the receiver position: the
image of the subsurface should be independent of it. The removal of the extra time delay due to
NMO is called the NMO correction.

NMO curve for more than one interface

Let us now move to a model with two interfaces, as depicted in figure 4.15. We call the source-
receiver distance x, the horizontal distance the ray has traveled in the second layer x2 , the wave
speed in the first layer c1 , and in the second c2 , the thickness of the first layer d1 , and of the second
4.4 NMO & velocity analysis 45

Figure 4.15: a) Distances in a subsurface model with two flat reflectors. b) NMO curve for second
reflector with depth 300 m of each layer and velocities of 1500 m/s and 2500 m/s in the first and
second layer respectively. The dashed line is the hyperbolic approximation of the travel-time curve.

d2 . Then the travel time from source to receiver is given by:

1/2
1/2
(x − x2 )2 + 4d21 x2 + 4d22
T = + 2 (4.27)
c1 c2
1/2 1/2
(x − x2 )2 x22
 
2d1 2d2
= 1+ + 1+ 2 (4.28)
c1 4d21 c2 4d2
1/2 1/2
x21 x22
 
= T1 1 + 2 2 + T2 1 + 2 2 , (4.29)
T1 c1 T2 c2

in which T1 and T2 are the zero-offset travel times through the first and second layer respectively,
and x1 = x − x2 . The problem with this formula is that, if we assume that c1 and c2 , are known,
we do not know x2 . Therefore we cannot directly use this expression to describe the move-out
behaviour of this two-reflector model.
In order to tackle this, we first expand the square-root terms in equation (4.29) in a Taylor series
expansion as we did for the one-interface case:
x21 x22
T ' T1 + + T2 + . (4.30)
2T1 c21 2T2 c22
and we square this equation in order to obtain:
 2
x22

2 2 x1
T = (T1 + T2 ) + (T1 + T2 ) + + O(x4 ). (4.31)
T1 c21 T2 c22

In this equation, we still have the distances x1 and x2 present. A relation between x1 and x2 can
46 Seismic Imaging

be found using Snell’s law at the interface, being:

sin α sin β
= , (4.32)
c1 c2
where α and β are the angles of the ray with the normal in layer 1 and 2 respectively, when
crossing the first interface (see also figure 4.15). We make an approximation for small angles for
which sin α ≈ tan α and sin β ≈ tan β, such that equation (4.32) becomes:
x1 x2
≈ , (4.33)
2d1 c1 2d2 c2
or
x1 x2
2 ≈ . (4.34)
T1 c1 T2 c22

Writing this as x2 = (T2 c22 )/(T1 c21 )x1 and substituting this in x1 + x2 = x, we have:

T1 c21
x1 = x , (4.35)
T1 c21 + T2 c22

Similarly for x2 , we obtain:

T2 c22
x2 = x . (4.36)
T1 c21 + T2 c22

We can use equations (4.35) and (4.36) in the quadratic form of eq.(4.31) to obtain:

T1 c21 + T2 c22
 
2 2 2
T ≈ (T1 + T2 ) + (T1 + T2 )x (4.37)
(T1 c21 + T2 c22 )2

(T1 + T2 ) 2
≈ (T1 + T2 )2 + x . (4.38)
T1 c21 + T2 c22

This equation is of the form:

x2
T 2 = Ttot (0)2 + . (4.39)
c2rms

with crms is what is called the root-mean-square velocity:

N
1 X 2
c2rms = ci Ti (0), (4.40)
Ttot (0)
i=1

in which Ti (0) denotes the zero-offset travel time through the i–th layer; Ttot (0) denotes the total
zero-offset time:
N
X
Ttot (0) = Ti (0). (4.41)
i=1
4.4 NMO & velocity analysis 47

Figure 4.16: CMP gather with one reflection before and after NMO correction.

We see here that with the assumptions made, a hyperbolic move-out for the interfaces below the
first one is obtained. The approximation however is a very good one at small and intermediate
offsets (for horizontal layers) but becomes worse when the offset becomes large. This effect can be
observed in figure 4.15b, where the hyperbolic approximation of the second interface reflection is
plotted with a dashed line.
Applying NMO correction
Then, how do we apply this NMO correction? First we have to determine the stacking (i.e. root-
mean-square) velocities for each zero-offset time T0 (see next section). Then, for each sample of the
zero-offset trace will remain in its position. For a trace with offset x, we calculate the position of
the reflection according to equation (4.39) and find the sample nearest to this time T . This sample
is then time-shifted back with the time difference between T and T0 (in fact it is mapped from
time T to time T0 ). In this simple scheme we have taken the sample nearest to the time T , but in
general we can be much more accurate by using a better interpolation scheme. It is important to
realize that with NMO we interpolate the data.
An artifact of the NMO correction is the NMO stretch. An example of this effect is shown in figure
4.16. How does this occur? We can see that the correction factor not only depends on the offset x
and the velocity crms , but also on the time T0 . So given a certain stacking velocity and offset, the
correction T − T0 becomes smaller when T0 becomes larger. Thus, the correction is not constant
along a trace, even if we have a constant offset and constant velocity. Also, we can see from this
correction that the effect will become more prominent when the offset becomes larger as well. This
effect is called NMO stretching.
48 Seismic Imaging

vnmo < vrms vnmo = vrms vnmo > vrms

Figure 4.17: CMP gather with one reflection after NMO correction with too low, correct and too
high stacking velocities.

Velocity estimation
In the application of the NMO correction, there is of course one big question: which velocity do
we use? Indeed, we do not know the velocity on beforehand. Actually, we use the alignment of a
reflection in a CMP gather as a measure for the velocity. Since, if the velocity is right, the reflection
will align perfectly. However, when the velocity is taken too small, the correction is too large and
the reflection will not align well; in the same way, when the velocity is chosen too big, the correction
is too small, and again the reflection will not align. An example of these cases is given in figure
4.17.
As the earth is consisting of more than one interface, we need to determine the velocities, although
they may just be root-mean square velocities for each layer. The goal is the same as in the case of
just one interface: we would like all the reflections to be horizontally aligned. A systematic way of
determining these velocities is to make common-midpoint gathers which are each NMO corrected
for a constant velocity. Then we can see for those velocities the reflector will align or not; usually
the deeper the interface the higher the (root-mean-square) velocity. An example of such an analysis
is given for a four reflector model (see figure 4.4) in figure 4.18.
The most commonly used way of determining velocities is via the velocity spectrum, which has
some relation to the aligning of reflectors. What we do with a velocity spectrum is that for a
certain velocity, we correct the CMP gather and apply a coherency measure to the data. This gives
us one output trace. Then, for a next velocity, we do the same. For a complete set of velocities, we
plot these results next to each other, which is then called the velocity spectrum. On the vertical
axis we then have the time, while on the horizontal axis we have the velocity. As an example we
consider again the synthetic CMP gather in the model of figure 4.4, for which we calculate the
semblance for velocities between 1000 m/s and 3000 m/s, see figure 4.19. The result we obtain is
often displayed in contour mode or color mode.
As a coherency measure, the semblance is most often used. The semblance S(t, c) at a time t for a
4.4 NMO & velocity analysis 49

Figure 4.18: CMP gather NMO corrected with a range of constant NMO velocities from 1300 to
2700 m/s with steps of 200 m/s.

Figure 4.19: CMP gather with its velocity spectrum, using a semblance calculation with window
length of 20 ms.

velocity c is defined as:

P 2
M
1 m=1 A(x m , t, c)
S(t, c) = PM , (4.42)
M 2
m=1 A (xm , t, c)

in which M is the number of traces in a CMP and A is the amplitude of the seismogram at offset xm
and time t after NMO correction with velocity c. For the definition of other coherency measures,
50 Seismic Imaging

the reader is referred to Yilmaz (1987, page 169, 173). Note that if an event is perfectly aligned
with constant amplitude for all offsets, the semblance has value 1. Therefore, the semblance has
always values between 0 and 1.
For a more extensive discussion on the velocity analysis we would like to refer to the book of Yilmaz
(1987, pp.166—182).
4.5 noise suppression: marine 51

Figure 4.20: The ray path of a multiple and a reflection from deeper down.

4.5 Noise suppression: marine

Although this section links up to the noise suppression as discussed in section 4.3, at that stage
the NMO was not discussed yet while this knowledge is necessary for the approach taken in this
section. Here we discuss the removal of the main ”noise” in marine data: multiple reflections, or
just multiples, due to the water layer on top of the solid layers.
We use again filters defined in the (f, Kx )-domain. As for the removal of ground roll in land data, we
make use of the fact that there is a difference in apparent velocity of the multiple, compared to the
apparent velocities of the reflections coming from deeper down. The reflections from deeper down
have generally encountered a higher wave speed, and so these arrivals will arrive more vertically
at the surface (see also figure 4.20), compared to multiples that arrive at the same travel time
(and therefore have propagated longer in shallower layers). Due to structure in the subsurface,
reflections do not appear so hyperbolic in shot-sorted data, and therefore, for multiple elimination,
data are sorted into CMP gathers in order to make the hyperbolic assumption of the reflection
events better valid.
The way the multiple-elimination filtering is applied is as follows:

• Sort data into CMP gathers;

• Apply NMO correction with velocity in between primary and multiple events, which will map
the upward curved primaries to the negative Kx values and the downward curving multiples
to the positive Kx plane;

• Apply (f, Kx ) filtering by removing one half of the Kx plane, containing the multiple events;

• Resort data into shot gathers or continue processing in CMP gathers.

A field example for a CMP gather from a marine line in the North Sea is given in figure 4.21. Here
we see an enormous amount of multiple reflections, which can be identified after NMO correction,
with velocities in between primary and multiple velocities, as shown in figure 4.21b. The upward
curving events are identified as primaries, whereas the downward curving events are the multiples.
In the (f, Kx ) domain, the positive Kx plane is zeroed, and after inverse NMO correction, figure
4.21d is the result. Note that due to the mute in the NMO correction, some reflection information
is lost. Note also that in the lower part of the region, we are not sure that all the remaining
52 Seismic Imaging

events are primaries: generally the difference in move-out velocity is not present for all multiples,
especially in the deeper part of the data.
4.5 noise suppression: marine 53

(a) (b)

(c) (d)

Figure 4.21: Field example of multiple elimination on a marine CMP gather, using NMO correction.
a) Input CMP gather. b) NMO corrected CMP gather with NMO velocities in between primary
and multiple velocities. c) Result of b) after removing positive dipping events. d) Filtered CMP
gather after inverse NMO correction.
54 Seismic Imaging

(a) (b) (c)

Figure 4.22: CMP gather with 2 primaries and 1 multiple before (a) and after (b) NMO correction
and after stacking (c).

4.6 Stacking

As stated before, an important goal in seismic processing is to increase the signal-to-noise ratio, and
a very important, if not the most important, step towards this goal, is stacking NMO-corrected
CMP gathers. With stacking we add the NMO-corrected traces in a CMP gather to give one
output trace: the word ”stacking” comes from the old days when papers were stacked. A better
nomenclature is perhaps horizontal stacking because we stack in the horizontal direction. This is
in contrast to vertical stacking, which is recording the data at the same place from the same shot
position several times and adding (i.e. averaging) these results. With (horizontal) stacking, we
average over different angles of incidence of the waves, even in horizontally layered media. This
means that we lose some information on the reflection coefficient since, as the reader may know, the
reflection coefficient of an interface is angle-dependent. Therefore, the stacked section will contain
averaged angle-dependent reflection information.
In figure 4.22 a CMP gather with two primaries and one multiple is shown before and after NMO
correction in figures (a) and (b), respectively. What can be seen in the figure that after NMO,
which is correct for the primary reflections, the multiple has not been corrected properly. For the
multiple, the velocity that is chosen at that time, is too large compared to the velocity necessary
for the multiple. Since the velocity is too large for the multiple, the correction is too small so the
wrongly corrected multiple still shows some hyperbolic behaviour. The stacked result is obtained
by adding all traces for a particular time, so we add along the horizontal direction. The result of
this adding, or stacking, is given in figure (c). The resulting stacked trace shows a reduced multiple
energy, which is a desired feature of the stack.
Although the signal-to-noise ratio is increased by stacking, we will also have introduced some
4.6 Stacking 55

correct stacking velocities stacking velocities 7% too high

Figure 4.23: Stacked sections with correct and too high stacking velocities.

distortions. We have already discussed the NMO stretch and the approximation with the root-
mean-square velocity. Therefore, when we add traces, we do not do a perfect job so we lose
resolution. The effect of an erroneous velocity for the NMO is shown in figure 4.23, which shows
a stacked section with the correct stacking velocities and with 7% too high stacking velocities for
the data generated in the model of figure 4.1. One can see that the stacked trace is getting a
lower frequency content and that the amplitudes are decreasing in some parts with the erroneous
velocities. Note that a stacked section simulates a zero-offset section, but with much better signal
to noise ratio. Compare therefore the stacked result to the zero-offset section of figure 4.5, which
shows exactly the same region (1000 - 4000 m) in the model. Note the resemblance of the stack with
the zero-offset section. Note also that the stack is twice as dense sampled in the trace direction,
due to the fact that there are twice as many CMP positions as there are shot positions.
Finally, it should be emphasized that, with stacking, we reduce the data volume. The amount of
data reduction is the number of added traces in a CMP gather. There are certain algorithms which
are expensive to compute and are therefore applied to stacked data rather than on pre-stack data.
An example of this is migration as shall be discussed in the next section.
56 Seismic Imaging

4.7 Zero-offset migration

Although we removed some timing effects with the NMO correction, this does not mean that we
removed the wave effects: it is just one of many. We still need to ”focus” the energy further.
Migration deals with a further removal of wave phenomena via focussing in order to arrive at a
section which is a true representation of the subsurface. After the NMO correction and stacking, we
have only synthesized a zero-offset section, since we removed the offset dependence of the receiver
position with respect to the source. That means, we have a section as if we did a seismic survey
with source and receiver at the same place, thus zero-offset. Migration could be defined as :
Migration is the focussing process which results in a true image of the subsurface from primary-
reflection data, assuming the velocity model is correct.
Equivalently, migration obtains the true image in (x, y, z) from data that are obtained in (x, y, t),
where x, y and z stand for the two horizontal and vertical coordinate, respectively; again, under
the assumption the velocity model is correct.

Three simple configurations and their zero-offset time sections

Let us consider the simple example of a point diffractor in the subsurface. A point diffractor is
like a ”ball” in the subsurface: when wave energy impinges on it, it scatters (reflects) energy back
in all directions. When the source and receiver are at the same point at the surface, the receiver
will only receive the ray that is scattered back as drawn in picture 4.24. So notice that even not
right above the diffractor, we will receive energy. The (zero-offset) time section for a diffractor at
position (xd , zd ) is described by:
2
4(xs − xd )2

2R
T2 = = Td2 + , (4.43)
c c2

where R being the distance in a homogeneous medium with velocity c, Td being the time 2zd /c and
xs being the surface position of source/receiver. This time section is (again) a hyperbola. As may
be clear now, a zero-offset section is not a good representation of the subsurface, since that should
be the left picture in figure 4.24. The process that converts the right picture (hyperbola) into the
left picture (ball) is called seismic migration.
So far, we considered one point diffractor, but we can build up any reflector by putting all point
diffractors on that reflector. When the spacing between the point diffractors become infinitely
small, the responses become identical. This concept agrees with Huygens’ principle. As example,
consider four point diffractors, as depicted on the top left of figure (4.25). Each diffractor has the
behaviour as discussed above, as can be seen on the right of figure (4.25), but the combination of
the time responses shows an apparent dip. The actual dip goes, of course, through the apexes of
the hyperbolae. In the next figures, the number of point diffractors is increased to 8, 16 and 32.
Note that for 32, the separate diffractors are hardly observable any more, and the response also
looks more like a dipping reflector (with some end-point effects).
Let us now look at a full dipping reflector. Of course, it has some of the characteristics as we
saw with the point diffractors, only with a full reflector we no longer see the separate hyperbolae.
Actually, we will only see the apparent dip. As we saw with the point diffractors, we need to bring
4.7 Zero-offset migration 57

Figure 4.24: A diffractor (left) and its seismic response (right), a hyperbola in the zero-offset
section.

the reflection energy back to where they came from, namely the apex of each hyperbola. When
connecting all the apexes of the hyperbolae, we get the real dip. This is depicted in figure (4.26).
The next figure (4.27) quantifies the effect of migrating the energy to its actual location. In
particular, compare the figures in the middle and and on the right: the difference is a factor cos θ,
where θ is the angle of the reflector with the horizontal. The zero-offset travel time at a certain
x-value can be specified by tZO = (2/c)x sin θ, assuming that x = 0 corresponds to the point
where the reflector hits the surface in figure 4.27a. The slope in the zero-offset section is therefore
dt/dx = (2/c) sin θ, see figure 4.27b. If this zero-offset section is migrated and the result is displayed
in vertical time τ = z/c, the resulting slope of the reflector is dτ /dx = (2/c) tan θ (figure 4.27c).
Thus, migration increases the time dip in the section by cos θ and thus reflectors in the unmigrated
section are increased in their up-dip direction in the migrated section. At the same time, migration
decreases the apparent signal frequency by the factor cos θ. The reason that the dip is increased
by cos θ and the frequency decreased by cos θ lies in the fact that the horizontal wavenumber is
preserved.
Another commonly observed phenomena is the so-called ”bow-tie” shaped zero-offset response, due
to synclinal structures in the earth. This is shown in figure 4.28, where it can be observed how in
the middle above the syncline multi-valued arrivals are present. This behaviour can be predicted
by considering small portions of the reflected signal, and increasing the dip of each portion of the
reflected signal. Note that in figure 4.29 such structures are also visible.
So far, we discussed three typical cases: point diffractor, dipping reflector and a syncline. From field
observation, we usually obtain much more complicated structures. An example is given in figure
4.29. When we consider our typical configurations, we can well (in a qualitative sense) understand
the effect of migration of the real data set, as shown in figure 4.29. We can observe that all the
diffractions in the stacked section are well collapsed after the migration.

Diffraction stack
58 Seismic Imaging

Figure 4.25: Point diffractors (left) and their seismic responses (right). From top to bottom: 4, 8,
16 and 32 points. Note the apparent dip from the hyperbolae.
4.7 Zero-offset migration 59

Figure 4.26: Relation between the reflection points in depth (a) and the traveltimes in the zero-offset
section (b) for a dipping reflector (from Yilmaz, 1987, fig. 4-14).

depth section zero-offset section migrated zero-offset section

Figure 4.27: Migration increases the dip in the zero-offset section.

So far, we haven’t described how to migrate a full dataset like the one shown in figure 4.29. The
simplest case of a migration is adding (stacking) the data along hyperbolae. In that case, each point
of the section (each time and space point!) is seen as a diffractor. As we saw in the 4-point-diffractor
case compared with the dipping-reflector case, any reflector can be synthesized by point diffractors
60 Seismic Imaging

(although infinite). So if each point of the zero-offset time section is seen as a point diffractor (and
the velocity is known), we can add data along the particular hyperbola for that point. In case a
real hyperbola is present in the observed time section, due to a real point diffractor, energy will
be added up constructively to give a relatively large output signal (=migrated) at that point. In
case no real diffractor is present, the energy along the hyperbola at that point does not add up
constructively and therefore the output signal (=migrated) will be small.
This procedure is called a diffraction stack. In the early days of computers the diffraction stack
was used to apply the migration. In formula form, the diffraction stack is given by (assumed to
have a discrete number of x’s, being the traces in a zero-offset section):
r !
X
2 (xs − xd )2
pzo (xd , td ) = pzo xs , t = td + 4 , (4.44)
x
c2
s

where pzo stand for zero-offset data and c is the stacking velocity. From the formula it may be
obvious that data are added along hyperbolae for each output point (xd , td ), being the apex of the
hyperbola for point (xd , td ). (Note that we used the same notation as in equation 4.43.) What we
do when stacking along hyperbolae, is actually removing the wave propagation effect from a point
diffractor to the source/receiver positions. A very nice feature about the diffraction stack is that
it visualizes our intuitive idea of migration, and is very useful in a conceptual sense. Of course, for
this procedure to be effective we need to know the stacking velocity.
What is lacking in the approach of the diffraction stack is the basis on deeper physical principles
than (kinematic) ray theory alone. The final migrated result may be correct in position (if the
diffraction responses can be assumed to have a hyperbolic shape, i.e. if the subsurface exhibits
moderate variations in velocity), but not in amplitude. We leave that as a subject of later courses.

Further Reading
Yilmaz, O., 2001. Seismic Data Analysis, (2 Volumes), Society of Exploration Geophysics, Tulsa,
OK, USA, 2027 pages. ISBN 9781560800941, 1-56080-094-1.

depth section with zero-offset ray paths zero-offset section

Figure 4.28: A syncline reflector (left) yields ”bow-tie” shape in zero-offset section (right).
4.7 Zero-offset migration 61

Figure 4.29: Stacked section (a) and its time migrated version (b) (from Yilmaz, 1987, fig. 4-20)
62 Seismic Imaging
Chapter 5

Seismic acquisition: To optimally

recorded 3D wave fields

In this chapter, the basic steps are discussed of how to obtain optimally recorded 3D seismic data,
in order to get to 3D seismic images of the subsurface. The fundamental issues of basic sampling
intervals, reciprocity and the 5D character of 3D data are introduced. And then the consequences
of these issues and of noise and fold are worked out for optimally recording 3D seismic wave-field
data.

5.1 The 5D character of 3D data

A main characteristic of seismic data is that they contain a lot of arrivals that are not desired
and therefore considered as ”noise”. To repeat what was said in earlier chapters: seismic data is
characterized by a bad signal-to-noise ratio. Therefore noise needs to be suppressed. One of the
possibilities was by direct filtering of the noise, as discussed in section 4.3. Another possibility,
which in practice is always applied, is (CMP-)stacking, as discussed in chapter 4. The parameter
controlling the success of the stack, is the fold. And fold implies that we have to sample the wave
field both with sources and receivers at the surface. As we have seen in chapter 4, even a seismic
survey along a line, called 2D, is a 3D dataset, namely a wave field W as a function of time t,
source position xs and receiver position xr , i.e.,:

W2D = W (t, xs , xr ) (5.1)

For data from an area rather than along a line, also called 3D data, the positions are in (x, y), so
for 3D data we have a 5-dimensional dataset:

W3D = W (t, xs , ys , xr , ys ) (5.2)

What we will see is that it is practically impossible to fully sample this 5D dataset. Fortunately
it is not really necessary but we have to make proper choicesfor optimal results. That will be the
main aim of this chapter. But before we do that, we have to discuss two fundamental items, namely
spatial sampling and reciprocity.

63
64 Seismic Imaging

5.2 Spatial sampling

As soon as you stick a geophone in the ground, you have decided about the position of one spatial
sample of the wave field. So, inherently, deciding about this position is ”digitizing” continuous
space. The situation here is different from sampling in time since a recording sensor initially senses
in continuous time which is subsequently digitized. But before the digitization of this continuous
signal, the signal is filtered with an analog circuit in order to suppress the signal below a certain
threshold for frequencies larger than a certain frequency, called fmax . The signal is still continuous
at this stage. The maximum frequency then determines the sampling time ∆t with which we can
digitally sample the signal properly, via Nyquist theorem:
1
∆t ≤ (5.3)
2fmax
In space, we cannot filter the data in an analog sense, so therefore we cannot control the maximum
spatial frequency as we can for the temporal frequency. In practice, this means that you are actually
always aliased in space. However, it is not that bad because we can put some knowledge of wave
propagation into this, in order to get spatially well sampled data, and that is what we will do next.
Let us recall from section 4.3 that for wave fields, the spatial frequency can be interpreted as
the number of wavelengths per period, also called the wavenumber. And the wavenumber can be
written in terms of frequency and wave velocity, via:
f
Kx = (5.4)
cx
In order to sample space according to the Nyquist criterion, we need at least 2 samples per minimum
wavelength or maximum wavenumber, which is determined by the maximum frequency fmax and
the minimum velocity cx,min in the x−direction:
1 cx,min
∆x ≤ = (5.5)
2Kmax 2fmax

In order to appreciate the distance that will come out of this formula, let us specify some specific
cases. First, let us take the case of a record on land where surface waves are present. They are
usually the lowest-velocity events. When the top layers consist of unconsolidated soil then a velocity
of 200 m/s is a typically encountered value. When we record data with a sampling interval of 2
ms (often used in oil-exploration practice), we have a maximum frequency of 250 Hz. Putting this
in formula 5.5, gives a value ∆x = 0.4 m. This is a very very small distance, especially when we
think of a 3D survey that needs to cover an area of 10 × 10 km.
Therefore, as a next case, we can take the maximum frequency that is contained in the surface wave
rather than the sampling frequency. When analyzing the surface wave, one can estimate that the
maximum frequency contained is some 50 Hz. Taking this value, we then get to a spatial interval
of 2 meters. That is more reasonable, although still small.
Then, as a next case, consider the desired signal, the P-wave reflections. First of all, a reflection
is not an event which travels solely in the horizontal direction but in all directions. In the formula
(eq.(5.5)) above, the minimum velocity in the x-direction is needed. A P-wave reflection has
4.7 Zero-offset migration 65

different angles at which the wave arrives at the surface since in the x-direction it is not a linear
event but a hyperbolic one. Commonly, a maximum angle for spatial calculations of some 30 degrees
with the vertical is taken. A typical velocity for P-waves is 1600 m/s. The minimum apparent
velocity in the x-direction is then VP / sin(30o ) = 3200 m/s. A note to make is that a reflection
from below the source has an infinite apparent velocity in the x-direction, so this 3200 m/s is really
a minimum. We assume that the reflections contain energy up to a frequency of 100 Hz. Then we
get to a horizontal distance of 16 meters.
The last two numbers for the spatial sampling intervals give the setting of spatially sampling the
wave field. The ”noise” (surface waves) require a denser sampling than the (desired) signal, i.e.,
P-wave reflections. We cannot just sample with the spatial sampling of the (desired) signal, since
then the noise would fold back in an aliased fashion in our record. So we would like to record the
total wave field, and then subsample the data after a spatial anti-alias filter. Since space is only
recorded ”digitally”, the spatial anti-alias filter is a digital filter. The standard way to implement
this is via hard-wired arrays, where the distance between the elements of the array have the spatial
sampling of the noise (surface waves), while the whole array has a length consistent with the spatial
sampling of the P-wave reflections (desired signal).
A new development in the hydrocarbon-exploration industry is to do this filtering in the processing
centre from the signals of all the separate elements within the array, instead of just one signal for
the total array. This is much more costly, but these single-sensor recordings look to become the
future (”massive-sensor systems”).

5.3 Reciprocity

Reciprocity is an important concept for wave fields. The main idea is that source and receiver can
be interchanged to give the same result. A condition here is that the characteristics of the source
and receiver need to be the same. An example of this reciprocity was already given in chapter 4
on processing: the ray paths as drawn in figure 4.1. In that figure, at position x = 3000, it can
be a source or a receiver, and the others can then be receivers or sources, respectively. So the
ray paths are identical when source and receiver are interchanged. This is known as Helmholtz
reciprocity from optics. And also the seismograms are the same — as long as the source and
receiver characteristics are the same —:

W (t, xs , xr ) = W (t, xr , xs ) (5.6)

This expresses reciprocity.

Although this is a simple relation, it has some consequences for acquiring seismic wave-field data.
A first consequence is that one should always avoid shot points that coincide with receiver stations.
If they do, then that would lead to reciprocity traces where exactly the same path in the subsurface
has been travelled, and therefore these reciprocity traces do not add any subsurface information to
the image.
Another consequence is that, under the condition of the same source and receiver characteristics,
the properties in common-shot gathers are the same as in common-receiver gathers. When the
source and receiver characteristics do differ, then differences start to show up. One could think of
66 Seismic Imaging

Figure 5.1: The simplest sampling of 3D wavefield data.

bad source coupling which would show up in common-receiver gathers, while bad geophone coupling
would show up in common-source gathers.
Related to the previous consequence is that sampling on the source side should be carried out in
the same way as on the receiver side. This is called symmetric sampling (Vermeer, 2001). This
would mean, ideally, that the shot interval (∆xs ) should be identical to the receiver interval (∆xr ).
And also that the shot array should be identical to the receiver array. In practice, the latter is often
not the case due to feasibility or costs, leading to asymmetries in the data, and thereby decreasing
the quality obtainable in the final image, due to acquisition limitations.

5.4 3D acquisition geometries

In this section we will discuss some 3D acquisition geometries, making use of the issues above. The
main challenge was stated in the beginning, namely that we need to sample a 5-dimensional data
set. If we would do that in the simplest way, we would get a geometry like given in figure 5.1, where
the sources and receivers are given on a grid. And when using a basic sampling interval needed for
sampling the total wave field, we need to use distances like 2 meters when using the values of the
second case in the section on spatial sampling intervals (section 5.2). When an area of 10 × 10 km
needs to be imaged, then this would lead to immense acquisition times and costs. But it would also
lead to an immense fold which is technically not needed. So optimal ways to reduce this simplest
geometry to a practically feasible one is desired.
4.7 Zero-offset migration 67

Figure 5.2: Single-fold data: 3D shot (left) or ... a crossed array of shots and receivers, the
cross-spread (right). (from: Vermeer, 2001)

5.4.1 Single-fold subsets of 5D pre-stack wave field

Let us first reduce all the shots to one 3D shot. We then get the situation as depicted in Figure
5.2. In this geometry we would record a 3D continuous shot gather, with no sudden jumps due to
the geometry. The associated CMP’s are given by the gray area. The CMP-fold is 1.
When looking at the properties of this 3D shot, we could obtain the same properties of fold and
CMP coverage with the situation as given on the right-hand side of figure 5.2: a line of receivers,
and perpendicular to it a line of shots. This is called a cross-spread. When one considers that for
a 3D shot, the receivers need to be on an areal grid, which often is practically not feasible, while
for a cross-spread, the receivers and shots ”only” need to be on a line, which often is practically
feasible. Also for a cross-spread, the data are spatially continuous, no sudden jumps due to the
geometry occur, so this is a good approach to get 3D data.
That a cross-spread is practically feasible is shown in the next figures 5.3, 5.5 and 5.7. These are
data that have been shot with a vibratory source, and geophones as receivers. As spatial sampling
interval, 2 meters have been taken, in accordance with sampling the total wave field. The data
were gathered in a polder in the Netherlands, called the ”Noord-Oost polder”. In figure 5.3 the
data are shown from a shot with a cross-line offset of 1 meter, so the shot is nearly on the line with
receivers, so it looks like a 2D shot. This shot can be Fourier-transformed to the (f, Kx )–domain,
giving the image as given in 5.4. Here we can discern the direct events, as were discussed in 2D in
section 4.3.
68 Seismic Imaging

Figure 5.3: 3D microspread: shot with cross-line offset of 1 meter.

When looking at a shot with larger cross-line offset, we get the data as shown in figure 5.5. Because
of the cross-line offset, the direct waves do not look like linear events any more but become hyper-
bolic events. In the (f, Kx ) domain, this can also be seen since the direct waves like the surface
waves are not linear events any more so they will form a triangle in the (f, Kx ) domain (see Figure
5.6). (If we would like to suppress the direct events as discussed in chapter 4.3, a 2D (f, Kx )-filter
would hardly work without affecting the reflected waves. A 3D (f, Kx , Ky )-filter is then required
for suppressing those direct waves.)
When one combines all the shots, the direct waves would be observable as a cone. Then, for showing
the last dimension, we can make a time slice through the data, as shown in figure 5.7. Here we can
see all concentric events due to the cones and hyperboloids of the direct waves and reflected waves,
respectively.
We briefly discussed the 3D shot and the cross-spread. Actually, they are part of a class of geome-
tries which are called minimal data sets (Gesbert, 2002). A minimal data set is defined as:

• It is a continuous 3D subset of a 5D pre-stack wave field;

• It is single-fold;

• It has a wide areal coverage.

These items have the consequences that a minimal data set has a wide illumination area, and is
suitable for pre-stack migration.
4.7 Zero-offset migration 69

Figure 5.4: 3D microspread: (f, Kx )-domain representation of shot with cross-line offset of 1 meter.
Note the low-cut filter applied to the data below 10 Hz.

Figure 5.5: 3D microspread: shot with cross-line offset of 759 meter.

70 Seismic Imaging

Figure 5.6: 3D microspread: (f, Kx )–domain representation of shot with cross-line offset of 759
meter. Note the low-cut filter applied to the data below 10 Hz.

5.4.2 Multi-fold subsets

In the previous subsection, we introduced the minimal data sets that have the characteristic of
a fold of 1. But we need a fold of more than 1 in order to suppress undesired events or noise,
by stacking. Equivalent to the 2D case, we want to fill a CMP bin with the desired number of
source-receiver pairs; in 3D the CMP bin is an area rather than a length (or strip with infinite
cross-line length). The fold M for 3D can be determined by:

M = Minline · Mcrossline (5.7)

where Minline and Mcrossline are the fold in the inline and crossline directions, respectively. Minline
can be just the 2D case, as shown in the previous chapter (eq. 4.3).
Each minimal data set can be (pre-stack) migrated, and the migrated sets can then be stacked
to suppress the noise, similar to the stacking that we saw in the previous chapter. As for the
number of the fold: usually in a new area, one has to test how much fold is needed to acquire the
desired quality, i.e. signal-to-noise ratio, of the reflectivity image. However, in certain cases, like
in surveying off-shore, the required fold is pretty well established from past experiences.
Let us first consider the so-called orthogonal geometry as used on land; the orthogonal geometry
can be considered as a succession of cross-spreads. With a geometry we assume that we do not have
one receiver line laid out but many, and we have source positions that go across all the receiver
lines. Then, one can take one receiver line and one shot line that makes one cross-spread. When one
takes the next receiver line with shots along the same source line, we obtain the next cross-spread,
and so on for for the shots on the same shot line. Next, one takes the next source line, and does the
4.7 Zero-offset migration 71

Figure 5.7: 3D microspread: time slice at 600 ms.

same for each receiver line. And so on. An important requirement to fullfill is the fold, necessary
to suppress the unavoidable noise. One achieves that by realizing that the CMP area that belongs
to one cross-spread overlaps with the one from a next cross-spread. And that in both source and
receiver directions. An example of that is given in figure 5.8.
Based on this approach, one can specify the fold for such a survey geometry. If we call the source-
line length Ls , the source-line (in-line) interval ∆Ls , the receiver-line length Lr and the receiver-line
(cross-line) interval ∆Lr , then the fold can be calculated as:

Lr Ls
M = Minline · Mcrossline = · (5.8)
2∆Ls 2∆Lr

or, in terms of maximum offsets (see Figure 5.9):

Xoff,max,i Xoff,max,c
M= · , (5.9)
∆Ls ∆Lr
where Xoff,max,i is the maximum in-line offset and Xoff,max,c the maximum cross-line offset.
To fully characterize the orthogonal geometry, 3 ”attributes” are necessary, namely the so-called
bin, unit cell and cross-spread. The bin is defined by the size of the source interval ∆xs and the
receiver interval ∆xr . The unit cell is defined by the source-line interval ∆Ls and the receiver-line
72 Seismic Imaging

Figure 5.8: Orthogonal geometry: Building areal coverage with cross-spreads. (from: Vermeer,
2001)

interval ∆Lr . And the cross-spread is defined by the source- and receiver-line lenghts Ls and Lr ,
or by the maximum in-line offset Xoff,max,i and the maximum cross-line offset Xoff,max,c . All these
sizes are shown in Figures 5.9. What is important are the ratio’s R of these sizes, the so-called
aspect ratio’s. So for the bin, the aspect ratio is:

∆xs
Rbin = . (5.10)
∆xr
For the unit cell, the aspect ratio is:

∆Lr
Runitcell = . (5.11)
∆Ls
And for the cross-spread the aspect ratio can be defined as:

Ls Xoff,max,c
Rcrossspread = = . (5.12)
Lr Xoff,max,i

When one designs a survey on land, one should aim for the situation that all these three aspect
ratio’s Rbin , Runitcell and Rcrossspread are as near as 1 as possible.
4.7 Zero-offset migration 73

Figure 5.9: Orthogonal geometry: Characteristic distances. (from: Vermeer, 2001)

Let us now consider the geometry, as usually deployed offshore: a boat pulls the source and receiver
lines along a certain track, so this is called a parallel geometry. The basic subset of the parallel
geometry is one shot line and one receiver line, which defines one mid-point line in the middle.
Since this resembles a 2D configuration that we discussed before, one should have equal shot and
receiver intervals, and equal shot and receiver arrays (if possible). A 3D aspect is that the mid-point
line spacing should preferably be chosen as half the cross-line shot interval (which may be difficult
to achieve in practice). These choices would lead to data that have equal numbers of shots in a
common-receiver gather as there are receivers in a common-shot gather. And it would also lead to
square bins, which is a desired aspect ratio of 1.
In the above, we have discussed two commonly used geometries. What they have in common is
that both the source and receiver positions are along lines, and therefore called line geometries.
But there are other types of geometries too. An often-used classification for the types of geometries
is:

• Line geometry: the orthogonal and parallel geometries as discussed above are of this type.
Perhaps one other type to mention is the so-called zig-zag geometry where the source positions
are at an angle of 45 degrees with the receiver lines, and the source positions are going zig-zag
between two receiver lines (hence the name). This one is also commonly used on land;

• Areal geometry: in this geometry, the sources and/or receivers are placed regularly over an
area rather than along lines;

• Random geometry. In this geometry the source and receiver positions are not regularly but
randomly positioned.

The latter two types are not much used, although random geometries are sometimes used in (pro-
tected) areas where sources and/or receivers cannot be placed on a regular grid.
74 Seismic Imaging

5.4.3 Full 3D surveys

One final aspect to discuss is that the lengths of the source and receiver lines are commonly shorter
than the sizes of the target to be imaged. So one has to, e.g., drag the receiver lines to next
positions in order to cover an area adjacent to the previous one(s). And the source positions also
have to be taken to that adjacent area. This replacing of receivers and sources makes full 3D
surveys expensive too. So the idea for the design of a 3D survey is to ”fill” the whole survey area
with minimal data sets and make sure it has the required fold.
One last remark about the building up of a full survey is properly taking the edge effects into
account: at the edges, the CMP bins have not been filled with the required number of traces yet
(see Figure 5.8 ) so one has to start surveying at positions which lie well outside the area to be
imaged. And this outer area is even increased when realizing that migration operators also need
full-fold data outside the area to be imaged.

Further Reading

Vermeer, G.J.O., 2012. 3-D seismic survey design (2nd edition), Society of Exploration Geophysi-
cists, Tulsa, OK, USA., 342 pages. ISBN 1560803037, 9781560803034.
Chapter 6

Seismic characterisation: from

wavefield to subsurface properties

Of the various methods in applied geophysics, seismic methods are one of the most commonly used
ones for characterizing the subsurface. Seismic data see the mechanical properties of the subsurface
as opposed to the ionic/chemical properties that are sensed by the electrical and electromagnetic
methods.
Seismic trace data is primarily used to obtain the structural information of the subsurface layers
or heterogeneities. This is known as seismic imaging. Migration is an important step in seismic
imaging. On the other hand, from seismic data obtaining the properties of the subsurface is
generally called characterization. Characterization is important for both hydrocarbon reservoirs
and engineering/geotechnical or hydrogeophysical site investigations. When one tries to obtain the
temporal change of one or more properties in the subsurface, then it is called monitoring. Usually
the target properties for monitoring are dynamic properties, i.e., properties that change with time.
From seismic data, through processing, one can extract the seismic properties like velocity, am-
plitude, dispersion, attenuation, etc. In seismic characterization, the challenge is to retrieve the
subsurface properties of interest from the seismic properties. In this chapter, we shall discuss
some prevalent and emerging approaches for seismic characterization and monitoring, taking into
consideration both hydrocarbon exploration and shallow engineering fields.

6.1 Seismic characterization: targets

Considering both deeper hydrocarbon exploration and shallow engineering investigations, the usual
targets for characterization are either static properties like bulk density, porosity, elastic stiff-
ness/rigidity, strength, compaction, or dynamic (generally time-varying) properties like in-situ
stress, flow properties (e.g., permeability or hydraulic conductivity, viscosity), fluid saturation, etc.
The layer-specific or spatial variation of these properties is of interest. In some cases, local distribu-
tions are also crucial, e.g., finding the distribution of contaminants, leachate flowpaths, overpressure
zones at well locations or in hydrocarbon reservoirs. In this discourse, we exclude discussions on

75
76 Seismic wavefield to subsurface properties

characterization of discontinuities like fractures.

6.2 P-waves, S-waves, Density and Poisson0 s Ratio

The degree of elasticity/plasticity of real Earth material depends mainly on the strain rate, i.e., on
the length of time it takes to achieve a certain amount of distortion. At very low strain rates, such
as movements in the order of mm or cm per year, it may behave as a ductile material. Examples
are the formation of geologic folds or the slow plastic convective currents of the hot material in the
Earths mantle with velocity on the order of several cm per year. On the other hand, the Earth
reacts elastically to the small but rapid deformations caused by a transient seismic pulse. Only for
very large amplitude seismic deformations in soft soil (e.g., from earthquake strong-motions in the
order of 40% or more of the acceleration due to gravity of the Earth) or for extremely long-period
free-oscillation modes does the inelastic behavior of seismic waves have to be taken into account
(Bormann et al., 2009).
Within its elastic range the behavior of the Earth material can be described by Hooke0 s Law that
states that the amount of strain is linearly proportional to the amount of stress. Beyond its elastic
limit the material may either respond with brittle fracturing (e.g., earthquake faulting) or ductile
behavior/plastic flow.
Elastic material resists differently to stress depending on the type of deformation. It can be
quantified by various elastic moduli:
(a) the bulk modulus κ is defined as the ratio of the hydrostatic (homogeneous on all-sides) pressure
change to the resulting relative volume change, i.e., κ = ∆P/(∆V/V), which is a measure of the
incompressibility of the material;
(b) the shear modulus µ (or rigidity) is a measure of the resistance of the material to shearing, i.e.,
to changing the shape and not the volume of the material. Its value is given by half of the ratio
between the applied shear stress τxy (or tangential force ∆F divided by the area A over which the
force is applied) and the resulting shear strain xy (or the shear displacement ∆L divided by the
length L of the area acted upon by ∆F), that is µ = τxy /2xy or µ = (∆F/A)/(∆L/L). For fluids
µ = 0, and for material of very strong resistance (i.e. ∆L → 0) µ → ∞;
(c) the Young0 s modulus E describes the behavior of a cylinder of length L that is pulled on both
ends. Its value is given by the ratio between the extensional stress to the resulting extensional
strain of the cylinder, i.e., E = (F/A)/(∆L/L);
(d) the ”static” Poisson0 s ratio σ is the ratio between the lateral contraction (relative change of
width W) of a cylinder being pulled on its ends to its relative longitudinal extension, i.e., σ =
(∆W/W)/(∆L/L).
The Young0 s modulus, the bulk modulus and the shear modulus all have the same physical units
as pressure and stress, namely in SI units: 1Pa=1N/m2 .
The most general linear relationship between stress and strain of an elastic medium is governed in
the generalized Hooke0 s law by a fourth order parameter tensor. It contains 21 independent moduli.
The properties of such a solid may vary with direction. Then the medium is called anisotropic.
6.2 Seismic wavefield to subsurface properties 77

Otherwise, if the properties are the same in all directions, a medium is termed isotropic.
In the simplest case of isotropy the number of independent parameters in the elastic tensor reduces
to just two. They are called after the French physicist Lamé (1795-1870) the Lamé parameters λ
and µ. The latter is identical with the shear modulus. λ does not have a straightforward physical
explanation but it can be expressed in terms of the above mentioned elastic moduli and Poissons
ratio:
2µ σE
λ=κ− = . (6.1)
3 (1 + σ)(1 − 2σ)

The other elastic parameters can also be expressed as functions of µ, λ and/or κ:

(3λ + 2µ)µ
E= (6.2)
λ+µ

and
λ 3κ − 2µ
σ= = . (6.3)
2(λ + µ) 2(3κ + µ)

For a Poisson solid λ = µ and thus, according to equation (6.3), σ = 0.25. Most crustal rocks have
Poisson0 s ratio between about 0.2 and 0.3. But σ may reach values of almost 0.5 for unconsolidated,
water-saturated sediments.
The elastic parameters govern the velocity with which seismic waves propagate. The equation of
motion for a continuum can be written as

∂ 2 ui
ρ = ∂j τij + fi , (6.4)
∂t2
where ρ is bulk density, ui is displacement, τij is the stress tensor, and fi is the body force or the
source term. In homogeneous media and for small deformations, the equation of motion for seismic
waves outside the source region (i.e., without the source term) takes a simple form:

∂ 2 ui
ρ = (λ + 2µ)∇∇ · u − µ∇×∇×u (6.5)
∂t2
where the first term of the right hand side contains the scalar product ∇·u=div u, which describes a
volume change (compression or dilation) containing always some (rotation-free) shearing too, unless
the medium is deformed hydrostatically. The second term in equation (6.5) is a vector product
∇×u corresponding to a curl (rotation) and describes a change in shape without volume change
(pure shearing). Generally, every vector field, such as the displacement field u, can be decomposed
into a rotation-free (ur ) and a divergence-free (ud ) part: u = ur + ud . Since the divergence of a
curl and the rotation of a divergence are zero, we get accordingly two independent solutions for
equation (6.5) when forming its scalar product ∇ · u and vector product ∇×u, respectively:

∂ 2 (∇ · u) (λ + 2µ) 2
= ∇ (∇ · ur (6.6)
∂t2 ρ
78 Seismic wavefield to subsurface properties

and
∂ 2 (∇×u) µ
= ∇2 (∇×ud . (6.7)
∂t2 ρ

Equations (6.6) and (6.7) are solutions of the wave equation for the propagation of two indepen-
dent types of seismic body waves, namely longitudinal (compressional - dilatational) P waves and
transverse (shear - rotational) S waves. Their velocities are
s s
λ + 2µ κ + 4µ/3
cP = = (6.8)
ρ ρ

and
r
µ
cS = . (6.9)
ρ
√
For a Poisson solid (i.e., λ=µ) cP /cS is 3. This value is close to the cP /cS ratio in consolidated
sedimentary and igeneous rocks. As κ ≥ 0, P waves travel faster than S waves. The higher the
rigidity of the material, the higher the P- and S-wave velocities. The rigidity usually increases
with density ρ, but more rapidly than ρ. This explains why denser rocks have normally faster
wave propagation velocities although we see density in the denominator in Equations (6.8) and
(6.9). Fluids (liquids or gasses) have no shear strength (µ = 0) and thus do not allow shear wave
propagation.
Using equations (6.3), (6.8) and (6.9), dynamic Poisson0 s ratio can be obtained from cP /cS :

c2P /c2S − 2
σ= (6.10)
2(c2P /c2S − 1)

Although in geophysics this dynamic Poisson0 s ratio is normally used, there are substantial differ-
ences between static and dynamic measurements of Poisson0 s ratio for the same rock reported in the
literature. The dynamic Poisson0 s ratio is generally higher. This is explained by the fact that in the
dynamic measurements the propagating waves have strain amplitudes smaller than 10−6 , while the
static measurements (squeezing the sample) have strain amplitudes greater than 10−3 . Therefore,
static measurements squeeze the earth material more than dynamic methods. Most sedimentary
rocks have pores or cracks that are elongated (not spherical) and the propagating seismic waves
in the dynamic method do not squeeze these cracks/pores closed. Thus the rock appears to be
stronger (lees compressible) than the situation where the cracks close. In the static method, the
cracks close and the rock will appear to have a lower strength. If the experiment was conducted at
higher confining pressures so that the cracks are closed at the beginning of the experiment, then
the static and dynamic measurements of Poisson0 s ratio approach one another. For unconsolidated
formations and soils, the dynamic Poisson’s ratio generally corresponds to the undrained state,
which is of great interest to the engineers.
cP /cS or Poisson0 s ratio is significant for differentiating different lithologies and hydrocarbon sat-
urations. They decrease substantially when, for instance, the gas saturation in a reservoir reaches
only a few percent. At a particular depth, shales tend to have a Poisson0 s ratio that is larger
6.3 Seismic wavefield to subsurface properties 79

than that for sands, especially gas-saturated sands. As the depth of investigation becomes shallow,
Poisson0 s ratio for sand and shale move toward 0.5. Also the sand and shale trends tend to over-
lap more. Conversely, as the depth of investigation increases, the sand and shale trends tend to
separate and have lower Poisson0 s ratio values - with sand still having a lower Poisson0 s ratio than
shale. With changes of depth, Poisson0 s ratio for limestone and dolomite, however, does not vary
as much as it does for sandstone and shale.
An applicable relationship between cP and cS is often used to predict cS where only cP is known.
cP /cS is also used to identify pore fluids from seismic data and in amplitude variation with offset
analysis.

6.3 Seismic velocities to properties

Seismic surveys yield, among others, maps of the distribution of seismic velocities. A review
of the relationships between the intrinsic rock or soil properties and the measured velocities or
reflectivities are needed before seismic survey results can be interpreted quantitatively in terms of
lithology. Many of these relationships are empirical; velocities are found to be related to certain
rock units in a given locale by actual laboratory measurements on core samples of the rock or soil.
The relationship between intrinsic properties such as porosity, fracture content, fluid content and
density, and velocity underlie these empirical relationships. The most important medium properties
that affect seismic velocities are porosity, cementation, effective stress and fluid saturation.
1) Porosity: A very rough rule due to Wyllie is the so called time-average relationship:

1 φ 1−φ
= + , (6.11)
cbulk cf luid cmatrix

where φ is the porosity and solid grains constitute the matrix (Wyllie et al., 1956). This relationship
is not based on any convincing theory but is approximately correct when the effective pressure is
high and the rock is fully saturated.
2) Cementation or lithification: The degree to which grains in a sedimentary rock are cemented
together by post depositional, usually chemical, processes, has a strong effect on the effective elastic
moduli. By filling pore space with minerals of higher density than the fluid it replaces, the bulk
density is increased. The combination of porosity reduction and cementation causes the observed
increase of velocity with depth of burial and age.
3) Effective pressure or stress: Seismic velocities are strongly dependent on effective pressure. [For
a rock buried in the earth the confining pressure is the pressure due to the overlying rock column,
the pore water pressure may be the hydrostatic pressure if there is connected porosity to the surface
or it may be greater or less than hydrostatic. The effective pressure is the difference between the
confining and pore pressure.] In general, seismic velocity rises with increasing confining pressure
and then levels off to a terminal velocity when the effective pressure is high. The effect is probably
due to crack closure. At low effective pressure cracks are open and easily closed with an increase
in stress (large strain for low increase in stresssmall κ and low velocity). As the effective pressure
increases the cracks are all closed, κ goes up and the velocity increases. The effects are similar
80 Seismic wavefield to subsurface properties

for µ, however the grain-boundary contacts play an important role there to determine the stress-
sensitivity. Finally even at depth, as the pore pressure increases above hydrostatic, the effective
pressure decreases as does the velocity. Overpressured zones can be detected in a sedimentary
sequence by their anomalously low velocities.
4) Fluid saturation: From theoretical and empirical studies it is found that the compressional wave
velocity decreases with decreasing fluid saturation. As the fraction of gas in the pores increases,
κ and hence cP decreases. Less intuitive is the fact that cS also decreases with an increase in gas
content. The reflection coefficient is strongly affected if one of the contacting media is gas saturated
because the impedance is lowered by both the density and velocity decreases.
For soft soils, stress and strain history, grain sorting and angularity, state of packing, overconsol-
idation ratio, clay content and cohesiveness are additional important factors affecting the seismic
velocities. Because seismic velocity changes affect the reflection amplitude, we shall consider these
effects in more details when discussing seismic reflectivity, next.

6.4 Seismic reflectivity to properties

Anomaly in reflection amplitude, either in poststack data (normal moveout reflectivity) or in

prestack data (angle-dependent reflectivity or AVO anomaly), can be useful for property estima-
tion. We shall first discuss briefly on characterization using normal moveout reflectivity. Normal
moveout reflectivity is given by:
ρ2 c2 − ρ1 c1
R= , (6.12)
ρ2 c2 + ρ1 c1
where ρ and c represents, respectively, bulk density and seismic velocity and the suffices 1 and 2
correspond to upper and lower medium for a downward propagating wave. Whenever an amplitude
anomaly is recognized on a seismic section, the immediate challenge is to distinguish which rock
or soil property variation could have caused the amplitude change. In order to assist this decision,
rules-of-thumb are desirable as to how seismic velocity is affected by changes in (a) elastic moduli,
(b) densities, and (c) various environmental conditions (Figure ??). For porous materials (which are
usually the rocks and soils of exploration interest), the bulk and shear moduli need to be separated
into its components. i.e., κ=f (κf , κma , κdry , φ, ...) and µ=f (µf , µma , µdry , φ, ...), where the
suffices f , ma and dry indicate, respectively, fluid, matrix and dry properties, and φ is porosity.
The bulk density is given by:

ρ = (1 − φ)ρma + φρn , (6.13)

where ρma and ρn are density of the grains (matrix) and density of the pore fluid, respectively.
When the pore fluid is made of two phases, e.g., brine and hydrocarbon, then ρn can be expressed
as,

ρn = (1 − SW )ρhyd + SW ρbr , (6.14)

where ρhyd and ρbr are densities of hydrocarbon and brine, respectively, and SW is the fraction of
pore space filled with brine subtracted from water saturation.
6.4 Seismic wavefield to subsurface properties 81

Figure 6.1: A schematic overview of factors affecting seismic velocity (after Hilterman, 2001). Here
V denotes velocity (which is marked as c elsewhere in the text).

Gassmann (1951) derived an equation that relates effective bulk modulus (κ) of a fluid-saturated
rock to that of the bulk moduli of the matrix (κma ), the frame (κdry ), pore fluid (κf ) and porosity
(φ):

4 (1 − κdry /κma )2
ρc2P = κdry + µdry + , (6.15)
3 (1 − φ − κdry /κma )1/κma + φ/κf

and

ρc2S = µdry . (6.16)

Of the 8 unknowns in the above two equations, usually cP , ρ and φ are known from the well logs.
κma is estimated using the mixing rules for macroscopic effective media proposed by Voigt (1928),
Reuss (1929) and Hill (1952), collectively known as V-R-H model. Note that Hill model cannot
be used to estimate shear moduli, and V-R-H model is not applicable for gas-saturated rocks. In
order to estimate the effective bulk modulus with the V-R-H model, the volumetric percentage of
each mineral and the porosity must be known. In addition, the bulk moduli of the mineral and
pore fluid are required. Next, κf and pore-fluid density are estimated using the detailed approach
of Batzle and Wang (1992), which takes into consideration pore temperature, pressure, salinity,
gas-oil-ratio (GOR) and specific gas gravity, and which uses the equation of Wood (1955) to obtain
the effective bulk modulus. The 3 unknown parameters are κdry , µdry and cS . Because there are
82 Seismic wavefield to subsurface properties

two equations and 3 unknowns, extra equations are required. For this purpose, several possibilities
exist: (a) using the cP -to-cS transform cS,wet = f (cP,wet ) as proposed by Greenberg and Castagna
(1992), (b) using the Biot coefficient B to estimate frame bulk modulus κdry = (1 − B)κma , or (c)
2(σdry +1)
using the known dry rock Poisson0 s ratio σdry in the equation κdry /µdry = 3−6σ dry
.
Numerous assumptions are involved in the derivation and application of Gassmann’s equation: (1)
the porous material is isotropic, elastic, monomineralic, and homogeneous; (2) the pore space is well
connected and in pressure equilibrium (zero-frequency limit); (3) the medium is a closed system
with no pore-fluid movement across boundaries; (4) there is no chemical interaction between fluids
and rock frame (shear modulus remains constant). Many of these assumptions may not be valid
for hydrocarbon reservoirs and depend on rock and fluid properties and the in-situ conditions. For
example, most rocks are anisotropic to some degree, invalidating assumption (1). The work of
Brown and Korringa (1975) provides an explicit form for an anisotropic fluid substitution. Also,
in seismic applications, it is normally assumed that Gassmann’s equation works best for seismic
data at frequencies less than 100 Hz (Mavko et al., 1998). Recently published laboratory data
(Batzle et al., 2001) show that acoustic waves may be dispersive in rocks within the typical seismic
band, invalidating assumption (2). In such cases, seismic frequencies may still be too high for the
application of Gassmann’s equation. Pore pressures may not have enough time to reach equilibrium,
and the rock remains unrelaxed or only partially relaxed. A Gassmann-type calculation provides
an estimate of relaxed velocity at zero frequency, which is a lower bound of the fluid-saturation
effect. Other similar violations of the assumptions often lead to the misapplication of Gassmann’s
equation.
Several modifications for use of the Gassmann’s equation have subsequently been proposed and are
used. This simplicity remains a primary reason for the wide application of this fluid-substitution
equation to calculate the cP , cS and ρ of a fluid-saturated formation, where the fluid composition
and saturation is considered to have changed from a known value corresponding to known seismic
velocities and density. These velocities are then used to calculate synthetic poststack or prestack
seismic traces. These equations predominate in the analysis of direct hydrocarbon indicators (DHI)
such as amplitude brightspots, amplitude variation with offset (AVO), and time-lapse reservoir
monitoring.
So far we have discussed normal moveout or poststack reflectivity. The use of prestack reflectivity,
on the other hand, utilizes the angle-dependent variation of the reflection amplitude. The amplitude
variation with offset or amplitude variation with angle(AVA) anomalies are used to predict local
changes in reservoir properties, like saturation, fluid type. A detailed discussion on AVO is beyond
the scope of this course, However, in relation to subsurface property estimation, we briefly cover
the basic idea.
Seismic reflection coefficient as a function of angle-of-incidence at an interface separating two media
is strongly affected by the relative values in Poisson’s ratio in the two media. At angles other than
normal-incidence, some portion of the incident P wave is converted to an S wave, and vice-versa.
As explained earlier, S waves differ from P waves in that their velocity is not significantly affected
by changes in the fluid content of a rock. Gas within the pore space of a rock dramatically
lowers P-wave velocity. Because cP /cS , and hence Poisson0 s ratio, is significantly different for gas-
charged rocks as opposed to water-bearing rocks, gas-sand/shale or gas-sand/wet-sand reflectors
6.4 Seismic wavefield to subsurface properties 83

are different when compared to most other reflectors. Reflections associated with high porosity
gas-bearing rocks usually exhibit an increase in amplitude with offset when compared to gas-free
reflectors. Because most reflections decrease with offset, amplitude increases in AVO analysis are
anomalous. Physical rock parameters measured in the laboratory should be used to optimize AVO
modeling. Note, however, that many seismic amplitude anomalies are not caused by economic
hydrocarbon (gas) accumulations and false positive AVO response is problematic.
The partitioning of incident wave energy at a reflecting interface can be expressed by the ratio of
incident to transmitted or reflected displacement amplitudes, displacement potentials, or energy.
Zoeppritz (1919) developed equations to describe particle displacements and calculate reflection
coefficients at the boundary between two layers. Zoeppritz’s equations estimate the expected change
in amplitude at any incident angle for any combination of rock properties. Several approximations
(simplifications using more intuitive variables) to Zoeppritz’s equations have been developed, e.g.
Bortfeld (1961), Richards and Frasier (1976), Aki and Richards (1980), Shuey (1985), Smith and
Gidlow(1987), Fatti et al. (1994), Verm and Hilterman (1995), Goodway et al.(1997). Angle-
dependent reflectivity can be modeled based on evaluating the sensitivity of rock properties using
these approximations. These different approximations focused on different physical quantities. For
instance, Shuey’s equations, which is one of the most successful and widely used ones, considers
Poisson0 s ratio as the elastic property most directly related to the angular dependence of the
reflection coefficient (inspired by the work of Koefoed, 1955). The 3-term Shuey’s equation is given
by:

R(θ) = R(0) + Gsin2 θ + F (tan2 θ − sin2 θ), (6.17)

where
 
1 ∆cP ∆ρ
R(0) = + , (6.18)
2 cP ρ

c2
 
1 ∆cP ∆ρ ∆cS
G= + 2 2S +2 , (6.19)
2 cP cP ρ cS

and
1 ∆cP
F = . (6.20)
2 cP
In these equations θ is the angle of incidence; ∆cP , ∆cS and ∆ρ are, respectively, the P-wave
velocity, S-wave velocity and density contrast across the boundary. R(0) is the reflection coefficient
at normal incidence and is controlled by the contrast in acoustic impedances. G, referred to as the
AVO gradient, describes the variation of reflection amplitudes at intermediate offsets and the third
term, F, describes the behaviour at large angles/far offsets that are close to the critical angle. This
equation can be further simplified by assuming that the angle of incidence is less than 30 degrees
(i.e. the offset is relatively small), so the third term will tend to zero. This is the case in many
seismic surveys and gives the Shuey approximation:

R(θ) = R(0) + Gsin2 θ. (6.21)

84 Seismic wavefield to subsurface properties

This is an equation of a straight line when R(θ) is plotted against sin2 θ (known as AVO cross-plot).
The intercept describes the normal-incidence reflectivity, while the slope or the gradient shows how
the amplitude changes with angle. The presence of hydrocarbon or a change in lithology or Poisson’s
ratio will show up as a deviation from the linear trend in the AVO cross-plot. The interpretation
of an AVO anomaly requires rock-physical constraints and modelling (often using Gassmann’s
equation and its modifications). AVO analysis incorporating anisotropy, use of shear waves, and
joint inversion of both P and converted wave data have proven to be useful. As mentioned earlier,
AVO is not fail-safe, as not all exploration targets (e.g., presence of oil and gas) are associated with
an obvious AVO anomaly.

6.5 Seismic intrinsic dispersion to properties

In oil and gas exploration and production, aquifer management and other applications such as the
underground storage of CO2, seismic characterization techniques are used to capture as much infor-
mation as possible on fluid-filled reservoir rocks. Biot theory (Biot, 1956) and its extensions provide
a convenient framework to connect the various parameters characterizing a porous medium to the
frequency-dependent seismic wave properties, namely, their amplitudes and velocities. Poroelastic-
ity theory takes into account the differential motion between the solid grain and the pore fluid in the
porous medium during seismic wave propagation. The poroelastic model involves more parameters
than the elastodynamic theory, but on the other hand, the intrinsic dispersion (frequency-dependent
amplitude decay/attenuation and velocity dispersion) characteristics at the macroscopic scale are
determined by the intrinsic properties of the medium without having to resort to empirical rela-
tionships. Attenuation mechanisms at microscopic and mesoscopic scales, which are not considered
in the original Biot theory, can be introduced into alternative poroelastic theories (e.g. Pride et
al., 2004).
Porosity, permeability and fluid saturation are the most important properties for reservoir engineers
and hydrogeophysicists. The estimation of poroelastic properties of rocks or soils from seismic waves
is challenging. In the poroelastic case, eight model parameters are required to describe the medium,
compared with only one or two in the acoustic case, and three in the elastic case if wave attenuation
is not considered. The advantages of using a poroelastic theory in an inversion (e.g., De Barros
et al., 2012) are (1) to directly relate seismic wave characteristics to porous media properties; (2)
to use information that cannot be described by viscoelasticity or elasticity with the Gassmann
equation, and (3) to open the possibility to use seismic dispersion to determine permeability and
fluid/porous medium properties.
The main assumption behind the governing equations for the poroelastodynamic theory first derived
by Biot (1956) are that seismic wavelengths are longer than the pore size and the medium can then
be considered homogenoeus. Poroelastic theories have since been derived and improved by many
authors for both soft soils and hard rocks (e.g. Geertsma and Smith, 1961; Stoll and Bryan, 1970;
Stoll 1974, 1977, 1989; Auriault et al., 1985; Johnson et al., 1987; Berryman and Wang, 2000). The
Biot’s equations in the frequency domain can be written as:

[(κU + µ/3)∇∇ + (µ∇2 + ω 2 ρ)I]u + [C∇∇ + ω 2 ρn I]w = 0 (6.22)

6.6 Seismic wavefield to subsurface properties 85

[C∇∇ + ω 2 ρn I]u + [M ∇∇ + ω 2 ρI]w = 0, (6.23)

where u and w respectively denote the average solid displacement and the relative fluid-to-solid
displacement, ω is the angular frequency, I the identity tensor, ∇∇ the gradient of the divergence
operator and ∇2 the Laplacian operator. The other quantities appearing in the above equations
are properties of the medium.
The bulk density ρ is obtained from fluid density ρn , solid grain or matrix density ρma , and porosity
φ following equation (6.13). κU is the undrained bulk modulus and µ is the shear modulus. M
(fluid storage coefficient) and C (C-modulus) are mechanical parameters. In the quasistatic limit,
at low frequencies, these parameters are real, frequency-independent and can be expressed in terms
of the drained bulk modulus κdry , porosity φ, mineral bulk modulus κs and fluid bulk modulus κn
using Gassmann’s equation.
Solution of equations (6.22) and (6.23) leads to the expressions giving the dispersive (frequency-
dependent) velocity of a fast and a slow P wave and an S wave. The fast P-wave has fluid and
solid motion in phase, while the slow P-wave has out-of-phase motions. At low frequency, the slow
P-wave wave has a diffusive pattern and can be seen as a fluid pressure diffusion wave. At high
frequency, the inertial effects are predominant. A relaxation frequency ωc , which is dependent on
the viscosity η of the medium, separates the low frequency regime where viscous losses are dominant
from the high frequency regime where inertial effects prevail.
The intrinsic velocity dispersion, thus, depends on material properties like φ, κdry , κs and κn .
Finally, the wave attenuation is explained by the generalized Darcys law which uses a complex,
frequency-dependent dynamic permeability k(ω). Using these expressions, the observed velocity
dispersion and frequency-dependent attenuation of seismic waves in fluid-saturated porous media
can be inverted for estimation of the properties of the porous media. In particular, the permeability
k has been of great interest in poroelastic inversion. Lately, for soft saturated soils it has been
possible to obtain reliable estimates of both φ and k from S-wave dispersion data. These properties
are, otherwise, nearly impossible to estimate in in-situ condition.

6.6 Seismic velocity to strength of unconsolidated sandy soils

S waves are especially relevant in geotechnical engineering because from S-wave velocity and bulk
density one obtains directly the small-strain (10−5 −10−6 ) stiffness or rigidity µ of the soil layers (see
equation 6.9). In fact, this is the only means to get the in-situ stiffness of the soil. Laboratory esti-
mation on samples collected in the field generally suffers from the large uncertainties resulting from
sample disturbance. Several in-situ methods, including analysis of surface wave extrinsic (layering
induced) dispersion (SASW or MASW methods), refraction or reflection methods, tomography, or
seismic cone-penetration tests (CPT) are common approaches for obtaining S-wave velocity infor-
mation of the shallow soil layers. Small-strain dynamic stiffness or rigidity is then calculated, using
some idea of the bulk density ρ. This dynamic stiffness is important in all dynamic loading prob-
lems in civil/geotechnical engineering, e.g., in earthquake engineering site investigations, to study
the effect of vibration due to heavy machineries, sea waves, or moving trains, and in liquefaction
86 Seismic wavefield to subsurface properties

potential evaluations.
On the other hand, the large-strain (>10) failure-properties or strength of the soil is necessary for
all large excavations, tunnelings, building projects and in foundation engineering. Elastic strain
being very small, seismic waves do not address soil strength (maximum shear stress where the soil
fails). One very common field approach to measure the strength or the pile-bearing capacity of
the soil is cone-penetration testing or CPT. In CPT a metal cone is pushed into the soil using a
hydraulic ram and the resistance encountered by the cone tip is measured continuously in depth.
From this cone tip resistance qc the bearing capacity of the soil layers and the strength properties
are estimated. Interestingly, it has been found that CPT qc and shear-wave velocity cS has a clear
correlation with each other, although they correspond to very different properties (strength versus
stiffness) and very different strain levels. Recent research has found the cause to be the common
dependence of both qc and cS on the state of compaction or relative density DR of sand and the
in-situ stress state. For clay, the reason for the observed correlation is still under investigation.
This understanding has led to the possibility of localizing the more compact zones in sands us-
ing cS /qc ratio and translating spatially continuous cS field obtained from high-resolution S-wave
reflection data to spatially continuous strength section using site-specific seismic cone-penetration
test (SCPT) calibrations at several points.

6.7 Near-field seismic impedance to mapping changes in the near-

surface

It is difficult to measure accurately the force applied on the ground by a surface seismic source.
However, for well-controlled small vibratory sources developed for near-surface applications, where
much efforts have gone to measure, over the entire frequency band of interest, the source motion
accurately and hence the groundforce Fg , it is possible to derive the complex near-field radiation
impedance ZN F :

Fg
ZN F = , (6.24)
ü
where ü is the particle velocity below the source. Vibrator baseplate velocity is taken as an ap-
proximation for ü.
The real and imaginary parts of ZN F can be calculated based on measurements made on the source.
No other seismic receiver is needed. It has been found that the complex ZN F can map changes in
the near-surface ground properties, e.g., presence of shallow cavities, sinkholes, or buried levees.

6.8 Seismic monitoring to deduce changes in subsurface properties

Time-lapse or 4D seismic monitoring has been a powerful approach to capture changes in the
subsurface properties. The application areas and the challenges involved are diverse. Three areas
where time-lapse seismic has proven its strength are (a) hydrocarbon reservoirs, (b) sequestrated
CO2 reservoirs, and (c) geotechnical engineering and hydrogeophysical investigation sites.
6.8 Seismic wavefield to subsurface properties 87

The basic premise of 4D is a simple one. The method involves the acquisition, processing and
interpretation of repeated seismic surveys (from the surface or in a borehole) over a producing
field with the aim of understanding the changes in the reservoir over time, particularly its behavior
during production. Fluid movement and the effects of secondary recovery are important targets.
This understanding has very real budgetary consequences as increasing the recovery factor of a
reservoir, even by a few percent, has significant revenue implications. 4D seismic has evolved
from a qualitative tool to identify producing zones and bypassed oil, to become an integral part of
quantitative reservoir management. By its nature, 4D is interpretation-driven and requires accurate
acquisition and processing to ensure that the bridge between a static and a dynamic reservoir model
is robustly built.
Applications of 4D seismic technology now span the life of the reservoir, from initial production to
identify pressure cells through mid-field life monitoring of waterflood fronts to late-field life where
the primary driver is identifying bypassed oil to extend economic recovery. The differences between
the surveys may be attributed to changes in saturation, pressure and, in many fields, overburden
stress due to reservoir compaction. Time-lapse surveys may indicate the presence of barriers to
reservoir connectivity, changes in reservoir saturation and pressure, and changes in overburden rock
strength.
The technical benefits include the ability to better place wells in the reservoir to maximize produc-
tivity and to avoid losses in the overburden. Many factors influence whether or not the time-lapse
signal in the reservoir can be detected as well as the frequency with which 4D surveys may be
repeated. These include the reservoir rocks themselves and the nature and rate of change of fluids
being produced from or injected into the reservoir. Together, these influence the relative strength of
the time-lapse seismic signal. The ability to detect the seismic signal is also affected by the ability
to exactly duplicate the previous survey. The first step in any time-lapse survey is a feasibility
study, which will assess the ability to detect a signal, repeat the earlier survey and determine the
optimum time interval between surveys.
Three main challenges are (a) resolving the (weak) time-lapse signal in noisy situations, (b) ac-
quisition and processing repeatability, and (c) separation of the effect multiple parameters (e.g.,
pressure and fluid saturation changes) that has given rise to the observed time-lapse difference in
the seismic property e.g., reflectivity or velocity. The development of time-lapse seismic technology
has focused on addressing these challenges.
In the future, many of the world’s hydrocarbon reserves are to be found onshore, and equally
important, many of these reserves are located in carbonate reservoirs. Two factors inhibit the
application of 4D seismic methods in these geographic areas and geologic environments. First, the
bulk strength of carbonate rocks is higher than most sandstones, resulting in a relatively weak
time-lapse signal. Second, high-amplitude near surface-generated noise can be challenging, new
technologies are expected to alleviate this problem.
Time-lapse seismic has also been used to successfully monitor the CO2 plume movement in a number
of sequestrated reservoirs. Both surface and downhole 4D studies are made. The changes in seismic
wave velocities are investigated using Gassmann fluid substitution model linking multiphase fluid
flow.
Finally, to monitor changes in the shallow subsurface. time-lapse seismic survey has been useful.
88 Seismic wavefield to subsurface properties

The change in stiffness of a sand body due to biogrouting (strengthening of sand due to biochemical
reaction) has been monitored through repeated S-wave velocity measurements. The estimated
change in stiffness is in good agreement with the value obtained by laboratory tests on collected
sand samples. It has also been found that changes in the in-situ stress in near-surface soil layers
causes changes in S-wave velocity. This change in velocity is measurable in case an array of sensor
is deployed and kept fixed between different measurements. The inverted stress and porosity
estimates are quite accurate. Such in-situ noninvasive monitoring of underground properties like
stress, stiffness or permeability is otherwise not possible.

6.9 References

[1] P. Bormann, B. Engdahl and R. Kind, Seismic Wave Propagation and Earth models. In P. Bor-
mann (Ed.), New Manual of Seismological Observatory Practice (NMSOP) (pp. 1-70). Potsdam:
Deutsches GeoForschungsZentrum GFZ. doi:10.2312, 2009.
[2] M. R. J., Wyllie, A. R. Gregory and L. W. Gardner, Elastic wave velocities in heterogeneous
and porous media. Geophysics, 21, 1, pp. 41-70, 1956.
[3] F. Gassmann, Uber die elastizitat poroser medien: Vierteljahrsschr. Der Naturforsch. Gesellschaft
Zurich. 96, pp. 1-21, 1951.
[4] F. J. Hilterman, Seismic amplitude interpretation. EAGE Distinguished Instructor Series, No.
4, 2001.
[5] W. Voigt, Lehrbuch der Kristallphysik. Teubner, Leipzig, Germany, 1928.
[6] A. Reuss, Berechnung der Fliessgrenzen von Mischkristallen auf Grund der Plastiziätsbedinggung
für Einkristalle: Zeitschrift für Angewandte Matehematik und Mechanik, 9, pp. 49-58, 1929.
[7] R. W. Hill, The elastic behavior of crystalline aggregate. Proc. Physical Soc., London, A65, pp.
349-354, 1952.
[8] M. Batzle and Z. Wang, Seismic properties of fluids. Geophysics, 57, pp. 1396-1408, 1992.
[9] A. W. Wood, A textbook of sound. The MacMillan Co., 1955.
[10] M. L. Greenberg and J. P. Castagna, Shear-wave velocity estimation in porous rocks: Theo-
retical formulation, preliminary verification and applications. Geophys. Prosp., 40, pp. 195-210,
1992.
[11] J. S. Brown and J. Korringa, On the dependence of the elastic properties of a porous rock on
the compressibility of the pore fluid. Geophysics, 40, pp. 608-616, 1975.
[12] M. Batzle, R. Hafmann, D. Han and J. Castagna, Fluids and frequency dependent seismic
velocities of rocks. The Leading Edge, pp. 168-171, 2001.
[13] K. Zoeppritz, Erdbebenwellen VIIIB, On the reflection and propagation of seismic waves.
Gottinger Nachrichten, 1, pp. 66-84, 1919.
[14] R. Bortfeld, Approximation to the reflection and transmission coefficients of plane longitudinal
and transverse waves. Geophys. Prosp., 9, pp. 485-503, 1961.
Exercises 89

[15] P. G. Richards and C. W. Frasier, Scattering of elastic waves from depth-dependent inhomo-
geneities. Geophysics, 41, pp. 441-458, 1976.
[16] K. Aki and P. G. Richards, Quantitative Seismology - Theory and Methods. 1: W. H. Freeman
and Co., San Francisco, 1980.
[17] R. T. Shuey, A simplification of the Zoeppritz equations. Geophysics, 50, pp. 609-614, 1985.
[18] G. C. Smith and P. M. Gidlow, Weighted stacking for rock property estimation and detection
of gas. Geophys. Prosp., 35, pp. 993-1014, 1987.
[19] J. L. Fatti, G. C. Smith, P. J. Vail, P. J. Strauss and P. R. Levitt, Detection of gas in sandstone
reservoirs using AVO analysis. Geophysics, 59, pp. 1362-1376, 1994.
[20] R. Verm and F. Hilterman, Lithology color-coded seismic sections, The calibration of AVO
crossplotting to rock properties. The Leading Edge, 17, pp. 227-234, 1998.
[21] W. Goodway, T. Chen and J. Downton, Improved AVO fluid detection and lithology discrim-
ination using Lam petrophysical parameters; lr, mr & l/m fluid stack”, from P and S inversions.
SEG Expanded Abstracts, pp. 183-186, 1997.
[22] O. Koefoed, On the effect of Poisson0 s ratios of rock strata on the reflection coefficients of plane
waves. Geophys. Prosp., 3, pp. 381-387, 1955.
[23] M. A. Biot, Theory of propagation of elastic waves in fluid-saturated porous solid, I and II. J.
Acoust. Soc Am., 168-191, 1956.
[24] S. Pride, J. Berryman and J. Harris, Seismic attenuation due to wave-induced flow. J. Geophys.
Res., 109, B01201, 2004.
[25] L. De Barros, B. Dupuy, G. S. O’Brien, J. Virieux and S. Garambois, Using a poroelastic
theory to reconstruct subsurface properties: numerical invetigation. Seismic Wave - Research and
Analysis (M. Kanao Ed.), ISBN: 978-953-307-944-8, InTech, 2012.
[26] J. Geertsma and D. Smith, Some aspects of elastic wave propagation in fluid-saturated porous
solid. Soc. Pet. Eng. J., 26, pp. 235248, 1961.
[27] R. D. Stoll and G. M. Bryan, Wave attenuation in saturated sediments. J. Acous. Soc. Am.,
47, pp. 1440-1447, 1970.
[28] R. D. Stoll, Acoustic waves in saturated sediments. in Physics of Sound in merine Sediments,
pp. 19-39, 1974.
[29] R. D. Stoll, Acoustic waves in ocean sediments. Geophysics, 42, pp. 715-725, 1977.
[30] R. D. Stoll, Sediment Acoustics, Lecture Notes in Earth Sciences. Springer-Verlag, 1989.
[31] J. -L., Auriault, L. Borne and R. Chambon, Dynamics of porous saturated media, checking of
the generalized law of Darcy, J. Acous. Soc. Am., 77, 5, pp. 1641-1650, 1985.
[32] D. Hohnson, J. Koplik and R. Dashen, Theory of dynamic permeability and tortuosity in
fluid-saturated porous media. Journal of Fluid Mechanics, 176, pp. 379402, 1987.
[33] J. Berryman and H. Wang, Elastic wave propagation and attenuation in a double-porosity dual
permeability medium. Int. J. Rock Mech., 37, pp. 6378, 2000.
90 Seismic wavefield to subsurface properties

Exercises

Exercise 1.
Discuss the differences between static and dynamic elastic moduli in relation to their measurement
conditions and underlying physical processes. Why is the dynamic Poisson’s ratio sensitive to fluid
(especially gas) saturation?
Exercise 2.
What are the merits and challenges in using the Gassmann’s fluid substitution equation?
Exercise 3.
How does the Shuey approximation help in recognizing an AVO anomaly?
Exercise 4.
What are the main challenges in 4D seismic monitoring and why? How can such monitoring be
useful in near-surface engineering investigations?
Bibliography

[1] W. Telford, L. Geldart, and R. Sherrif, Applied Geophysics. Cambridge: Cambridge University
Press, 1991. 1.1

[2] D. Parasnis, Principles of Applied Geophysics. London: Chapman and Hall, 1997. 1.1

[3] J. Milsom, Field Geophysics. Chichester: Wiley, 2003. 1.1

[4] J. A. Stratton, Electromagnetic theory. New York: McGraw-Hill Book Company Inc., 1941.
1.1

[5] R. Harrington, Time-harmonic electromagnetic field. New York: McGraw-Hill, 1961. 1.1

[6] K. Aki and P. G. Richards, Quantitative Seismology, Theory and Methods, Vol. 1. New York:
W. H. Freeman and Company, 1980. 1.1

[7] B. L. N. Kennett, Seismic wave propagation in layered media. Cambridge: Cambridge Uni-
versity Press, 1981. 1.1

[8] A. J. Berkhout, Seismic migration: imaging of acoustic energy by wavefield extrapolation A.

Theoretical aspects. Amsterdam: Elsevier Science Publishers, 1985. 1.1

[9] J. A. Kong, Electromagnetic Wave Theory. New York: John Wiley & Sons, 1986. 1.1

[10] C. P. A. Wapenaar and A. J. Berkhout, Elastic wave field extrapolation: Redatuming of single-
and multi- component seismic data, ser. Advances in Exploration Geophysics. Elsevier Science
Publishers, 1989. 1.1

[11] J. G. van Bladel, Singular electromagnetic fields and sources, ser. The Oxford engineering
science series 28. Oxford: Clarendon, 1991. 1.1

[12] R. W. P. King, M. Owens, and T. T. Wu, Lateral electromagnetic waves; theory and applica-
tions to communications, geophysical exploration, and remote sensing. New York: Springer,
1992. 1.1

[13] I. V. Lindell, A. H. Sihvola, S. A. Tretyakov, and A. J. Viitanen, Methods for electromagnetic

field analysis. Oxford: Clarendon, 1992. 1.1

[14] D. S. Jones, Methods in electromagnetic wave propagation. Oxford: Clarendon, 1994. 1.1

91
92 References

[15] A. T. de Hoop, Handbook of radiation and scattering of waves. Amsterdam: Academic Press,
1995. 1.1

[16] D. Daniels, Surface Penetrating Radar. London: The Inst. Electrical Eng., 1996. 1.1

[17] A. Kovetz, Electromagnetic Theory. Oxford: Oxford University Press, 2000. 1.1

[18] C. S. Bristow and H. M. Jol, Ground penetrating radar in sediments, ser. Geological Society
Special publication. London: Geological Society, 2003. 1.1

[19] U. Mukherji, Electromagnetic field theory and wave propagation. Oxford: Alpha Science
International, 2006. 1.1

[20] J. G. van Bladel, Electromagnetic fields, ser. IEEE Press series on electromagnetic wave theory.
Hoboken: Wiley-Interscience, 2007. 1.1

[21] W. C. Chew, M. S. Tong, and B. Hu, Integral equation methods for electromagnetic and elastic
waves. Morgan and Claypool, 2009. 1.1

[22] A. Kaufman and G. Keller, Frequency and transient soundings, ser. Methods Geochem. Geo-
phys., vol. 16. New York: Elsevier Sci., 1983. 1.1

[23] M. N. Nabighian and J. D. Corbett, Electromagnetic methods in applied geophsyics - Vol. 2

Applications Part A and B, ser. investigations in geophysics. Tulsa, OK: Society of Exploration
Geophysicists, 1993. 1.1, 3.1

[24] ——, Electromagnetic methods in applied geophsyics - Vol. 1 Theory, ser. investigations in
geophysics. Tulsa, OK: Society of Exploration Geophysicists, 1994. 1.1, 3.1

[25] A. A. Kaufman and P. A. Eaton, The theory of inductive prospecting. Amsterdam: Elsevier,
2001. 1.1

[26] O. Koefoed, Geosounding Principles 1: Resistivtiy Sounding Measurements, ser. Methods in

geochemistry. Amsterdam: Elsevier, 1979. 1.1

[27] R. J. Blakely, Potential Theory in Gravity and Magnetic Applications. Cambridge: Cambridge
University Press, 1996. 1.1