Appendix A

Seismic data processing - overview

Processing of seismic reflection data involves numerous mathematical operations.

Those can be however grouped into three main categories:

I. Time corrections
II. Signal to noise improvement
III. Repositioning or inverse modelling

Time corrections involve:

a) Static corrections which aim to equalise travel times through typically low
velocity near surface weathered layer or so-called regolith in hard-rock
environment. For this computation we utilise waves which totally refract at the
weathering/fresh rock interface. This wave is a “diving” wave (Figure A1).
For land seismic data this step is of crucial importance since time delays
through the regolith layer could completely misalign reflected signals and
result in destructive summation during the process of stacking. Computed
refraction statics cannot however fully compensate the trave time differences
through the weathered layer (regolith) due to the nature of head waves which
are “diving” through highly rugose intervals (Figure A1). Consequently
additional computations are required to fine-tune reflection events.

b) Residual statics. Those are conventionally computed using reflection events,

again in a surface consistent manner. We can also use here a modified
procedure, which we developed recently for the case of a low fold data and
very week signal. The procedure consists of the following steps:
- Time correct refracted events using linear equation (linear moveout or
LMO) with a velocity of the fresh rock derived under step a).
- Sort and stack refraction events.
- Pick horizon along refractions and select time window that
encompasses only refracted events.
- Compute residual statics using the algorithm devised for reflection
data.

28
c) Dynamic corrections involve flattening reflection hyperbolae with appropriate
velocity (NMO velocity for short offset) in preparation for data summation
(Figure A2 and A3). The correction is based on the travel time equation
(Figure A2) which is derived for a single layer, constant velocity case. The
same equation is valid for a multi layer case as long as they are horizontal. The
actual correction is implemented on a sample by sample basis using the
normal-move-out (NMO) equation (Figure 3A). Before this step can be carried
out we need to define the velocity of the layers. Clearly the “best” velocity
will be the one which flattens reflection hyperbolae. This “flattening” or
search for the best velocity involves interactive velocity analysis (Figure A4).
Velocities are typically analysed in regular intervals along seismic line.

Signal to noise improvement:

This sequence of processes involves a large number of diverse mathematical

operations. They are however all aimed to improve the signal-to-noise ratio. Initial
steps here are compensation of reflection amplitudes. Two main type of amplitude
corrections aim to compensate losses due to natural wavefront spreading and
losses at each interface (reflection, refraction and mode conversion processes).
While the first correction is readily specified, the latter one is of an exponential
type but hard to determine explicitly (Figures A5 and A6). Consequently the total
amplitude correction is estimated by testing and visual inspection.

A large group of operations is aimed at removing all but primary reflection energy
from seismic data. These techniques are either single-channel operations such are
band-pass filtering and deconvolution or multi-channel operations such are F-K
filtering, tau-pi filtering, etc. The basic principles of these techniques with data
examples are shown in Figures A7-A12.

Finally the most powerful and the simplest technique we have for enhancing the
primaries is stacking, that is summation of a number of traces which contain
signals, which originates from the same depth point for various source-receiver
separations, into one trace. Hence in general larger is the fold the better the signal

29
to noise ratio. The main problem with the stacking method is that it assumes that
all of the traces, which go into summation, carry exactly the same information that
is contained in the signal recorded from the same depth point. Strictly speaking
this is only true for a layer-cake earth model. The summation process is effective
even for the dipping structures but it becomes discriminative if a range of different
dips, particularly conflicting dips, are present in the data.

Repositioning or inverse modelling

If dipping events are recorded in the data they need to be repositioned in their true
location since seismic traces are plotted vertically. This provides geometrical
explanation of the migration process (Figure A13a). The actual process after stack
is accomplished by using wave equation (inverse modelling) and commonly the
exploding reflector principle for the imaging condition. All of the migration
schemes naturally produce a depth section (true inverse modelling). In practice
they are modified to produce time migrated images which contain no immediately
recognisable errors, until the interpretation results (maps, horizons or time-
section) are converted to depth.

30
 STATIC corrections – travel time equalisation
through weathered layer

Source Surface elevation

Datum R239 E
R1 R240
R2
D
Base
regolith t1 t240 H
t2

Diving wave
Fresh rock
H + (D − E) H H + (D − E)
S = −t regolith + =− +
V fresh rock Vregolith V fresh rock

t1-t2 = receiver static difference, reciprocity for shots

Average (in various ways) 120/240 differences for each shot and receiver
Various methods: ABS, Delay Time, GRM, DRM

Figure A1. Computation of travel times and weathering depth from totally refracted (head)
waves. This is accomplished in a surface consistent manner, using reciprocity principle.

Dynamic corrections (NMO)

S R
x

NMO hyperbola
H V

M Reflector
SMR = S|R
H

x2 4 ⋅ H 2 x2
S| Travel time S R = 2
+ |
2
= 2
+ T02 = T
V V V

Figure A2. Travel time equation – reflection hyperbolae, derived for a single layer case.
Note that S’R is square time as well as T.

31
 offset

x2
∆t = −T0 + 2
+ T02
V
∆t1
V = VSTK – non physical
time

i.e. velocity that produces

∆t2
“the best looking” stack section

Dor dipping events (dip angle Q)

V=VSTK/cosQ
∆t3 (eg.5000m/s can go to 7000m/s)

∆t1> ∆t2> ∆t3

Figure A3. The actual dynamic correction ∆t is known as normal moveout (NMO)
correction.

Figure A4. An example showing interactive velocity analysis display.

32
 1
Amplitude decay = ;
V 2
RMS •t

Figure A5. Various mechanisms causes seismic waves to gradually lose their energy.
Spherical divergence is a consequence of the wavefront spreading, resulting in natural
reduction of energy density with depth or time. The spreading rate is proportional to wave
velocity. Correction is shown in blue box.

Energy partitioning

Sr
Pr

Pt
St

P
S

Figure A6. Reflection energy can be greatly decreased if the sedimentary section contains
units which posses strong contrast in elastic properties. Then the amount of energy
dissipated via refractions, reflections and mode conversions can be exceptionally large. In
such case more powerful seismic sources may be required. This type of correction that is
compensation is usually empirical. Parameters are determined by visual estimates.

33
 Signal and noise
Filtering (band-pass)
Amplitude
Direct P
Head Filter
(F)

Amplitude
S S spectrum

Reflections

Frequency (Hz)
Air blast
Filtered trace = FFT-1(S xF)
Raleigh
waves Figure A8. Real part of the forward Fourier
transform of a seismic trace produces amplitude
Figure A7. Various coherent noise spectrum. Multiplication is the frequency domain
trains generated by the source or by with a trapezoid and inverse Fourier transform
conversion at geological interfaces. yields band-pass filtered seismic trace. In time
We aim to remove or attenuate all domain this is equivalent to convolving seismic
events but reflected primary waves. trace with a Sinc–function.

Figure A9. Estimation of the frequency content of the primary signal is usually
accomplished by visual inspection of a narrow-band filtered record. Both the low-cut and the
high-cut filter points are determined with such tests.

34
 F-K filtering
Filter design Filter

Aliased

K – wave-number; Kmax=1/2*geophone separation

Figure A10. Two-dimensional Fourier transform is the basis for multi-channel filtering. In
this domain events separate by the apparent velocity and frequency content. The reason is
that an even with apparent slope α maps into frequency-wavenumber domain with slope 90-
α. Piece-slice, velocity or polygonal (shown here) type of filtering is often used to attenuate
coherent noise which has low apparent velocity. This is typically not possible for air blast
which has a wide frequency content and velocity too low to allow for adequate spatial
sampling.

35
 Deconvolution
A seismogram is the convolution of the source wavelet w, with the earth’s
impulse e (defined by the reflection coefficient R=z2-z1/z2+z1), plus noise. That
can be represented as:

x = w * e + n(t)

Deconvolution is simply inverse filtering aiming to recover earth impulse

response e. Due to band-limited nature of seismic signal the best we can hope to
recover is a sharp wavelet of a simple shape (like bell-shape). Deconvolution is
commonly implemented via Wiener filtering.

Optimum Wiener Filtering

⎡ r0 r1 .... rn−1 ⎤ ⎡ a0 ⎤ ⎡ g0 ⎤
⎢ . . . r ⎥⎢ a ⎥ ⎢ g ⎥
⎢ n−2 ⎥ ⎢ 1 ⎥
= ⎢ 1 ⎥

⎢ . . . . ⎥⎢ . ⎥ ⎢ . ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣rn−1 rn−2 .... r0 ⎦ ⎣an−1 ⎦ ⎣ g n−1 ⎦
Desired Cross-correlation
Autocorrelation of the input wavelet filter coeff. of the desired
output with the
input wavelet
There are five choices of desired outputs:

Type 1. Zero-lag spike (Spiking Deconvolution)

Type 2. Spike at arbitrary lag
Type 3 Time advanced form of input series (Predictive
Deconvolution)
Type 4 Zero phase wavelet
Type 5 Any desired arbitrary shape

Figure A12. Principles of deconvolution and computations required.

36
 Migration – geometrical consideration
O x A B
β
α Vt1
reflector Vt2 tanα=sinβ
C’ α- dip on CMP stack
z β - true geological dip

D’ C Plotted
a)

C’D’ – true position CD – CMP stack section

Wave equation migration (after stack)

surface
V1 b)
CMP Data

reflector

surface
V1/2 Exploding
reflector Model
reflector
Exploding Reflector (imaging principle)

The CMP stack as being equivalent to

a wavefield P(x,y,t) observed at the earth
Figure A13. Repositioning of seismic reflection energy into it’s “true” spatial
position: a) geometrical aspect of migration and b) imaging principle (exploding
reflector) commonly used for wave equation based migration techniques after
stack.

37