Reflection Processing Using Matlab PDF

Series
Series
Series
ISSN:
ISSN:
ISSN:
1932-1236
1932-1236
1932-1236
MM
MOUSA • AL-SHUHAIL
MOUSA •• AL-SHUHAIL
MOUSA
SSSYNTHESIS
YNTHESISL
YNTHESIS LLECTURES
ECTURES ONSSS
ECTURESON
ON IGNALPP
IGNAL
IGNAL PROCESSING
ROCESSING
ROCESSING
&C
& C Mor
Mor
Morgan
gan
gan&
&&Cl
Cl
Claypool
aypool
aypoolPublishers
Publishers
Publishers
AL-SHUHAIL
Series
Series
Series
Editor:
Editor:
Editor:
José
José
José
Moura,
Moura,
Moura,
Carnegie
Carnegie
Carnegie
Mellon
Mellon
Mellon
University
University
University
Processing
Processing
Processingof
of
ofSeismic
Seismic
SeismicReflection
Reflection
ReflectionData
Data
Data Processing
Processingof
ofSeismic
Seismic
Using
Using
UsingMATLAB™
MATLAB™
MATLAB™
Ref
Reflection
lectionData
Data
PROCESSING OF SEISMIC REFLECTTION DATA USING MATLAB™

PROCESSING OF
PROCESSING
Wail
Wail
WailA.
A.
A.Mousa
Mousa
Mousaand
and
and
Abdullatif
Abdullatif
Abdullatif
A.
A.
A.
Al-Shuhail
Al-Shuhail
Al-Shuhail
Using
UsingMATLAB™
MATLAB™
King
King
King
Fahd
Fahd
Fahd
University
University
University
ofofof
Petroleum
Petroleum
Petroleum
&&&
Minerals
Minerals
Minerals
This
This
Thisshort
short
short
book
book
bookisisfor
isfor
for
students,
students,
students, professors
professors
professors and
and
andprofessionals
professionals
professionalsinterested
interested
interestedininin
signal
signal
signal
processing
processing
processing ofofof
seismic
seismic
seismicdata
data
data
OF SEISMIC
using
using
usingMATLAB.
MATLAB.
MATLAB. TheThe
The step-by-step
step-by-step
step-by-step demo
demo
demo ofofof
the
the
the
full
full
full
reflection
reflection
reflection seismic
seismic
seismicdata
data
dataprocessing
processing
processing workflow
workflow
workflow using
using
using
a aa
SEISMIC REFLECTTION
complete
complete
complete real
real
real
seismic
seismic
seismic data
data
data
setset
set
places
places
places
itself
itself
itself
asasas
a avery
avery
veryuseful
useful
useful
feature
feature
featureofofof
thethe
the
book.
book.
book.This
This
This
isisespecially
isespecially
especially true
true
true
when
when
when
students
students
studentsare
are
are
performing
performing
performing their
their
their
projects,
projects,
projects,
andand
andwhen
when
when professors
professors
professorsandand
and
researchers
researchers
researchers areare
are
testing
testing
testing their
their
their
new
new
new developed
developed
developed
REFLECTTION DATA
algorithms
algorithms
algorithms ininMATLAB
inMATLAB
MATLAB forfor
for
processing
processing
processingseismic
seismic
seismicdata.
data.
data.
The
The
The
book
book
book provides
provides
providesthe
the
the
basic
basic
basic
seismic
seismic
seismic and
and
andsignal
signal
signalprocessing
processing
processing
theory
theory
theoryrequired
required
required for
for
foreach
each
each
chapter
chapter
chapterand
and
and
shows
shows
showshowhow
how tototo
process
process
process
thethe
the
data
data
data
from
from
from raw
raw
rawfield
field
field
records
records
recordstototo
a afinal
afinal
final
image
image
image
ofofof
thethe
the
subsurface
subsurface
subsurface allall
all
using
using
usingMATLAB.
MATLAB.
MATLAB. TheThe
TheMATLAB
MATLAB
MATLAB codes
codes
codesandand
and
seismic
seismic
seismic data
data
data
can
can
can
bebe
be
downloaded
downloaded
downloaded atatat
http://www.morganclaypool.com/page/mousa.
Wail
Wail
WailA.
A.
A.Mousa
Mousa
Mousa
DATA USING
Abdullatif
Abdullatif
AbdullatifA.
A.
A.Al-Shuhail
Al-Shuhail
Al-Shuhail
USING MATLAB™
MATLAB™
About
About
About
SYNTHESIs
SYNTHESIs
SYNTHESIs
This
This
Thisvolume
volume
volumeisisaisaprinted
aprinted
printed
version
version
version
ofofof
a awork
awork
workthat
that
that
appears
appears
appears
ininin
the
the
the
Synthesis
Synthesis
Synthesis
MOR GAN &CL AYPOOL

MOR GAN
MOR
Digital
Digital
DigitalLibrary
Library
Library ofofEngineering
ofEngineering
Engineering and
and
and
Computer
Computer
Computer Science.
Science.
Science.
Synthesis
Synthesis
Synthesis Lectures
Lectures
Lectures
provide
provide
provideconcise,
concise,
concise,
original
original
original
presentations
presentations
presentationsofofimportant
ofimportant
importantresearch
research
research
and
and
and
development
development
development
GAN &CL
topics,
topics,
topics,
published
published
published quickly,
quickly,
quickly,
inindigital
indigital
digital
and
and
and
print
print
print
formats.
formats.
formats.
For
For
For
more
more
moreinformation
information
information
visit
visit
visit
www.morganclaypool.com
&CL AYPOOL
ISBN:
ISBN:
ISBN:
978-1-60845-792-2
978-1-60845-792-2
978-1-60845-792-2 SSSYNTHESIS
YNTHESISL
YNTHESIS LLECTURES
ECTURES ONSSS
ECTURESON
ON IGNALPP
IGNAL
IGNAL PROCESSING
ROCESSING
ROCESSING
&&
&
AYPOOL
Mor
Mor
Morgan
gan
gan Cl
Cl
Claypool
aypool
aypoolPublishers
Publishers
Publishers 90000
90000
90000
wwwwww.wm
. .mm
o orogr rgaganancncl cal layaypypopoooolo.l cl. .cocom
omm
9 9781608
9781608
781608
457922
457922
457922
José
José
JoséMoura,
Moura,
Moura,Series
Series
Series
Editor
Editor
Editor
Processing of
Seismic Reflection Data
Using MATLAB™
Synthesis Lectures on Signal
Processing
Editor
José Moura, Carnegie Mellon University
Synthesis Lectures in Signal Processing will publish 50- to 100-page books on topics of interest to
signal processing engineers and researchers. The Lectures exploit in detail a focused topic. They
can be at different levels of exposition—from a basic introductory tutorial to an advanced
monograph—depending on the subject and the goals of the author. Over time, the Lectures will
provide a comprehensive treatment of signal processing. Because of its format, the Lectures will
also provide current coverage of signal processing, and existing Lectures will be updated by authors
when justified.
Lectures in Signal Processing are open to all relevant areas in signal processing. They will cover
theory and theoretical methods, algorithms, performance analysis, and applications. Some Lectures
will provide a new look at a well established area or problem, while others will venture into a brand
new topic in signal processing. By careful reviewing the manuscripts we will strive for quality both
in the Lectures’ contents and exposition.
Processing of Seismic Reflection Data Using MATLAB™

Wail A. Mousa and Abdullatif A. Al-Shuhail
2011
Fixed-Point Signal Processing

Wayne T. Padgett and David V. Anderson
2009
Advanced Radar Detection Schemes Under Mismatched Signal Models

Francesco Bandiera, Danilo Orlando, and Giuseppe Ricci
2009
DSP for MATLAB™ and LabVIEW™ IV: LMS Adaptive Filtering

Forester W. Isen
2009
iii
DSP for MATLAB™ and LabVIEW™ III: Digital Filter Design
Forester W. Isen
2008
DSP for MATLAB™ and LabVIEW™ II: Discrete Frequency Transforms

Forester W. Isen
2008
DSP for MATLAB™ and LabVIEW™ I: Fundamentals of Discrete Signal Processing

Forester W. Isen
2008
The Theory of Linear Prediction

P. P. Vaidyanathan
2007
Nonlinear Source Separation

Luis B. Almeida
2006
Spectral Analysis of Signals: The Missing Data Case

Yanwei Wang, Jian Li, and Petre Stoica
2006
Copyright © 2011 by Morgan & Claypool
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in
any form or by any means—electronic, mechanical, photocopy, recording, or any other except for brief quotations in
printed reviews, without the prior permission of the publisher.
Processing of Seismic Reflection Data Using MATLAB™

ISBN: 9781608457922 paperback

ISBN: 9781608457939 ebook
DOI 10.2200/S00384ED1V01Y201109SPR010
A Publication in the Morgan & Claypool Publishers series

SYNTHESIS LECTURES ON SIGNAL PROCESSING
Lecture #10
Series Editor: José Moura, Carnegie Mellon University
Series ISSN
Synthesis Lectures on Signal Processing
Print 1932-1236 Electronic 1932-1694
Processing of
Seismic Reflection Data
Using MATLAB™

King Fahd University of Petroleoum & Minerals
SYNTHESIS LECTURES ON SIGNAL PROCESSING #10
M
&C Morgan & cLaypool publishers
ABSTRACT
This short book is for students, professors and professionals interested in signal processing of seismic
data using MATLAB. The step-by-step demo of the full reflection seismic data processing workflow
using a complete real seismic data set places itself as a very useful feature of the book.This is especially
true when students are performing their projects, and when professors and researchers are testing
their new developed algorithms in MATLAB for processing seismic data. The book provides the
basic seismic and signal processing theory required for each chapter and shows how to process the
data from raw field records to a final image of the subsurface all using MATLAB. The MATLAB
codes and seismic data can be downloaded at http://www.morganclaypool.com/page/mousa.
KEYWORDS
seismic data, amplitude correction, seismic noise attenuation, spiking deconvolution,
static correction, migration
To my dearly beloved father, mother, wife, and children, W.M.
To my dear wife and children, A.S.

ix
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
1 Seismic Data Processing: A Quick Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Seismic Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Examination of A Real Seismic Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Description of the Seismic Reflection Real Data Set . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Examining the Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.1 Header Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.2 Displaying Seismic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 The Proposed Processing Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Computer Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 Useful MATLAB Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3 Quality Control of Real Seismic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1 Quality Control of Reflection Seismic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Trace Editing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Amplitude Losses & Their Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4 Seismic Noise Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Seismic Signal & Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2.1 Random noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2.2 Coherent Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Spectrum Analysis and Filtering of Seismic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
x
5 Seismic Deconvolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 The Seismic Convolutional Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.3 Spiking Deconvolution as a Filtering Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.4 Spiking Deconvolution Using Wiener Optimum Filters . . . . . . . . . . . . . . . . . . . . . 32
5.5 The Trace-Wavelet Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.5.1 Amplitude Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.5.2 Autocorrelations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.6 Spiking Deconvolution in Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.6.1 Autocorrelation Window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.6.2 Filter Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.6.3 Percent Prewhitening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.6.4 Spiking Deconvolution of our Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.8 Useful MATLAB Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6 Carrying the Processing Forward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.2 Common Midpoint Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.3 Velocity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.4 Normal Moveout (NMO) Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.5 Stacking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.7 Useful MATLAB Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7 Static Corrections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.1 Elevation Static Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.2 Residual Static Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.2.1 Uphole Surveys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
7.2.2 Refraction Statics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
7.2.3 Surface-Consistent Residual Static Correction Method . . . . . . . . . . . . . . . 57
xi
8 Seismic Migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.2 Huygens’ Principle and Basic Migration Principles . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.3 Migration Kinds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.4 Migration Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8.4.1 Post-stack Time Migration Using the Stolt f − k Migration . . . . . . . . . . 67
9 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Authors’ Biographies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Preface
This book aims to provide students and interested experts from the DSP community with the
necessary industrial background of seismic data processing using MATLAB, starting from loading
the data, displaying, quality control of the data until obtaining a seismic image. We decided to
write it after offering and teaching various related courses, for several years, in both the Electrical
Engineering and Earth Sciences Departments at King Fahd University of Petroleum & Minerals
(KFUPM). We hope that the book will motivate students to conduct research in multidisciplinary
areas and assist DSP experts in further developing advanced algorithms for processing seismic data.
We have written most of the book MATLAB’s functions and m-files and used few MATLAB
functions from the available GNU seismic MATLAB codes provided by the Signal Analysis and
Imaging Group, Department of Physics, University of Alberta, Canada. Although this book is written
to be self contained, there exist many useful MATLAB tutorials on the internet in addition to the
Mathworks documentation for MATLAB, which will benefit the interested reader in understanding
the written m-files and M-functions. Nevertheless, as far as this book is concerned, it is sufficient
for the reader to know the basics of using and programming in MATLAB. The readers may enjoy
the seismic data and the book MATLAB codes by downloading them following the link http://
www.morganclaypool.com/page/mousa.
In Chapter 1, we provide the reader with an overview of seismic data processing. Then, in
Chapter 2, we examine the two-dimensional (2-D) real seismic data set and extract various useful
processing information and data for the proceeding chapters. Chapter 3, some commonly used
quality control techniques in order to correct for amplitude losses in real seismic data sets. Chapter 4
explains what the main signals are that we are looking for and how to analyze the data in the
spectrum domain. Frequency filtering to attenuate ground roll noise contaminating our real data is
also explained in the same chapter. In Chapter 5, we present the seismic convolution model and how
we can apply deconvolution to enhance the vertical resolution of the data. After that, in Chapter 6,
we carry the seismic data processing forward by sorting the data from shot gathers into common
mid-point gathers, picking seismic stacking velocities, applying normal moveout (NMO) correction
and, finally, stacking the data. To further enhance the seismic data, we then apply in Chapter 7
residual static correction to the NMO-corrected data and we stack the data again. We then apply
seismic migration to enhance the horizontal resolution of our data in Chapter 8. Finally, we close
with some concluding remarks in Chapter 9.

September 2011
Acknowledgments
We would like to thank KFUPM for its continuous support of our teaching and research efforts,
where special thanks are due to H.E. Dr. K. Al-Sultan, KFUPM Rector, for his constant encourage-
ments and support. Also, we would like to thank Prof. Lina Karam, School of Electrical, Computer,
and Energy Engineering, Arizona State University, for encouraging us to write this book. Special
thanks are due to Ayman Al-Lehyani, Lecturer, Earth Sciences Department, KFUPM, for his assis-
tance in fixing some problems with the provided reflection seismic data headers. As well, we thank
Professor Joel S. Watkins, Professor Emeritus of Texas A&M University for allowing us to use the
data set provided with this book. Finally, we thank Joel Claypool for his encouragement to complete
this work.

September 2011
1
CHAPTER 1
Seismic Data Processing: A

Quick Overview
1.1 INTRODUCTION
Among many geophysical surveying techniques, seismic reflection is the most widely used and
well-known geophysical technique. Seismic reflection data can be processed to reveal details of
geological structures on scales from the top tens of meters of the Earth’s crust to its inner core [1, 2].
Part of its success lies in the fact that the raw seismic data is processed to produce seismic sections
which are images of the subsurface structure. A geologist can then make an informed interpretation
by understanding how the reflection method is used and seismic sections are created. The analysis
of seismic data is performed for many applications such as petroleum exploration, determination of
the earth’s core structure, monitoring earthquakes, etc. [2, 3].
Seismic signals are generated by a source (transmitter), such as an explosion, which then
propagate through earth layers. Some of these signals will be reflected, refracted and lost due to
attenuation). At the surface, the reflected signals are then recorded by a receiver. The strength of
the reflected signal depends on the impedance contrast between adjacent layers. In summary, a
seismic survey analysis scenario involves collection of data by an array of receivers (geophones for
land and hydrophones for marine), transmission over a narrow band channel, and storage of data
for processing, and interpretation [1, 2, 4].
A seismic trace represents a combined response of a layered ground and a recording system
to a seismic source wavelet. Any display of a collection of one or more seismic traces is termed a
seismogram. Assuming that the pulse shape remains unchanged as it propagates through such a layered
ground, the resultant seismic trace may be regarded as the convolution of the input impulse with a
time series known as a reflectivity function, which is composed of spikes (delta functions). Each spike
has an amplitude related to the reflection coefficient at a layer boundary and a traveltime1 equivalent
to the two-way reflection time from the surface to that boundary. Furthermore, the reflection time
series represents the impulse response of the layered ground, which is basically the output for a
spike input. Since the source wavelet has a finite length, individual reflections from closely-spaced
1Traveltime is the time difference between zero time and the arrival time of a seismic event. It can be a one-way time such as for
direct waves or two-way time such as for reflected waves.
2 1. SEISMIC DATA PROCESSING: A QUICK OVERVIEW
Reflectivity Geological
function section
Noise
A noisy seismogram
Input source A seismogram
wavelet or pulse or
a seismic trace
Figure 1.1: Convolution seismic data model. A seismic pulse is convolved with the reflection coefficient
log (reflectivity function) to get a seismic trace. The reflection coefficient log is related to the geological
section of the subsurface through the reflection coefficient of each geological boundary and the two-way
travel time.
boundaries may overlap in time on the resultant seismogram (seismic section). Figure 1.1 represents
a typical seismic convolution model [1].
Due to many factors, unwanted waves such as surface waves, corrupt the seismic records
with noise (unwanted energy). This unwanted energy includes random or incoherent noise such
as instrument signals and coherent noise like the ground roll noise. A typical example for a noisy
seismic section is shown in Figure 1.2. As a consequence, of the above effects, seismic traces generally
have a complex appearance and reflection events are often not recognized without the application
of suitable processing techniques. The purpose of processing such data can, in general, be viewed
as the process of attenuating noise and then determining the input pulse and removing that to give
the reflectivity function, which ultimately allows the determination of the acoustic impedances (or
related properties) of the subsurface layers.
1.2 SEISMIC DATA PROCESSING

Seismic data processing can be considered as a sequence of cascaded operations that attenuate/re-
move noise accompanying seismic data as well as making geometrical corrections such that the final
image will truly show a map (seismic image) of the subsurface. Processing of seismic data includes,
but is not limited to: filtering, common mid-point (CMP) sorting, velocity analysis, normal
move-out (NMO) correction, and stacking. Each seismic trace has three primary geometrical
factors which determine its nature: shot position, receiver position, and the position of the subsurface
reflection point. The last factor is the most critical [1]. Before processing, this position is unknown
but a good approximation can be made by assuming that this reflection point lies vertically under
the position on the surface mid-way between the shot and the receiver for that particular trace.
This point is referred to as common mid-point or common depth-point (CDP). Traces reflected from
1.2. SEISMIC DATA PROCESSING 3
Figure 1.2: A typical example for a seismic section (http://cwp.mines.edu). The horizontal axis
represents the offset of each seismic receiver (recorder) from the source where each records a trace with
respect to the two-way travel time (vertical axis). Clearly, this section contains various noise types.
the same CMP define a CMP gather. The CMP gather is important for seismic data processing
because the subsurface velocity can be derived using it. In general, the reflection seismic energy
is very weak and it is essential to increase the signal-to-noise ratio (SNR) of most data. Once
the velocity is known, the traces in CMP gathers can be corrected for NMO, which is basically a
way of correcting for time differences which occur due to offset2 in a CMP gather, i.e., to get the
equivalent of a zero-offset trace. This implies that all traces will have the same reflected pulses at the
same time, but with different random and coherent noise. So combining all the traces in a CMP
gather will average out noise and the SNR increases. This process is known as stacking.
In general, the main objectives of seismic data processing are to improve the seismic resolution
and increase the SNR of the data. These objectives are achieved through three primary stages. In
their usual order of application, they are:
1. deconvolution, which increases the vertical resolution;.
2. stacking, which increases the SNR; and

2 An offset is the distance from the source-point to a geophone or the center of a geophone group.
4 1. SEISMIC DATA PROCESSING: A QUICK OVERVIEW
3. migration, which increases the horizontal resolution.
In addition to these primary stages, secondary processes may be implemented at certain stages
to condition the data and improve the performance of these three processes. Figure 1.3 shows a
conventional seismic data processing flow [2]. In the following section, we elaborate on the processing
steps in this figure.
Field Tapes (Amplitudes) Observer’s Log (Geometry)
Preprocessing
Demultiplexing
Reformating
Editing
Geometrical spreading correction
Setup of field geometry
Application of field statics
Filtering
Deconvolution
Common Midpoint Sorting
Velocity analysis
Loop
Residual statics
Normal move-out correction
Stacking
Post-stack processing
Migration
Figure 1.3: A conventional seismic data processing flow (after [2]).
1. Preprocessing: This process involves a series of steps to condition the data and prepare it for
further quality control and processing including:
• demultiplexing
• reformatting
• trace editing
• gain application
• setup of field geometry
1.2. SEISMIC DATA PROCESSING 5
• application of field statics
2. Filtering is used to attenuate components of the seismic signals based on some measurable
property. It is an important step in order to proceed further with the other seismic data
processing steps that will help geophysicists to better analyze and interpret the acquired data.
3. Deconvolution is performed along the time axis to increase vertical resolution by compressing
the source wavelet to approximately a spike and attenuating noise and unwanted coherent
energy such as multi-path signals.
4. CMP sorting transforms the data from shot-receiver (shot gather) to midpoint-offset (CMP
gather) coordinates using the field geometry information.
5. Velocity analysis is performed on selected CMP gathers to estimate the stacking, root-mean
squared (RMS), or NMO velocities to each reflector. Velocities are interpolated between the
analyzed CMPs.
6. Residual static correction is usually needed for most land data. It corrects for lateral variations
in the velocity and thickness of the weathering layer.
7. NMO correction and muting: The stacking velocities are used to flatten the reflections in each
CMP gather (NMO correction). Muting zeros out the parts of NMO-corrected traces that
have been excessively stretched due to NMO correction.
8. Stacking: The NMO-corrected and muted traces in each CMP gather are summed over the
offset (stacked)
√to produce a single trace. Stacking M traces in a CMP increases the SNR of
this CMP by M.
9. Poststack processing includes time-variant band-pass filtering, dip filtering, and other processes
to enhance the stacked section.
10. Migration: Dipping reflections are moved to their true subsurface positions and diffractions
are collapsed by migrating the stacked section.
In the remaining parts of this book, we are going to practically explore most of the above processing
steps along with application to a typical real seismic data set. Note that the above steps are general and
depending on the data type, land or marine, the accompanying noise types, and/or the acquisition
conditions, the above processing steps may vary as well.
7
CHAPTER 2
Examination of A Real Seismic

Data Set
2.1 INTRODUCTION
When geophysicists meet with their seismic data for the first time, they carefully must look into the
seismic data itself and its header, which contains information and tabulation of parameters used to
acquire the data. Usually, the data is stored in magnetic tapes or hard disks and is saved in various
standard data formats such as the well-known SEG-Y or Seismic Unix formats [2, 5]. We have
already read the seismic data (that we are going to process in this book) and its header information
and stored them in a MATLAB data format file called Book_Seismic_Data.mat. The aim of this
chapter is, therefore, to examine this data along with its header information using MATLAB. This
will assist us in further analysis and processing of this real reflection seismic data in the remaining
chapters of the book.
2.2 DESCRIPTION OF THE SEISMIC REFLECTION REAL

DATA SET
The real data set we will use to illustrate the processing codes consists of a two-dimensional (2-D)
land line from east Texas, USA. The following are some important parameters about the data:
• Number of shots = 18.
• Source type = dynamite in 80-100-ft depth holes.
• Number of channels per shot = 33.
• Receiver type = Vertical-component geophones.
• Array type = 12-element inline.
• Number of traces in line = 594.
• Receiver interval = 220 ft.
• Shot interval is variable.
• Time sampling interval = 2 milliseconds (ms)

8 2. EXAMINATION OF A REAL SEISMIC DATA SET
• Number of time samples per trace = 1501.
• Data format = SEG-Y.
• Byte swap type = Big endian.
• Data file name = data.sgy.
• Geometry has already been set up and recorded in the trace headers.
• Uphole times at shot locations have been recorded in the trace headers.
• An 8-64-Hz bandpass filter has been applied to the data in the field.
2.3 EXAMINING THE DATA SET

Let us now examine the 2-D seismic data set and its header information by loading the file
Book_Seismic_Data.m. By loading the data in the MATLAB workspace we are going to find
two variables as follows:
1 >> load B o o k _ S e i s m i c _ D a t a . m a t
2 >> whos
3 Name Size Bytes Class Attributes
4
5 D 1501 x594 7132752 double
6 H 1 x594 3303008 struct
2.3.1 HEADER INFORMATION

After loading the data, we will find that variable H is a 1 × 594 structure array with many fields such
as: the time sampling interval dt, the number of time samples per trace ns, the offset offset, the
trace numbers tracl, the shot gather numbers fldr, etc. To access any one of these fields, say for
example, the offset and store its values in a new vector called offset, then:
1 >> o f f s e t =[ H . o f f s e t ];
2 >> whos
3 Name Size Bytes Class Attributes
4
5 D 1501 x594 7132752 double
6 H 1 x594 3303008 struct
7 offset 1 x594 4752 double
The reader can explore more fields in a similar way. We have written a MATLAB function called
extracting_geometry.m which can be used to get useful geometrical information such as the
source coordinates sx and sy , the receiver coordinates gx and gy , the source elevations gz , etc. The
2.3. EXAMINING THE DATA SET 9
following MATLAB code shows an example of how one can extract certain geometrical acquisition
parameters:
1 load ( ' B o o k _ S e i s m i c _ D a t a . m a t ' , ' H ' )

2 [ sx , sy , gx , gy , s h o t _ g a t h e r s , n u m _ t r a c e _ p e r _ s g , sz , gz ]=
3 extracting_geometry (H);
These are different geometrical variables obtained from the seismic header structure H. We then can
use such variables in the following code:
1 figure , stem ( s h o t _ g a t h e r s , n u m _ t r a c e _ p e r _ s g )
2 x l a b e l ( ' S h o t g a t h e r n u m b e r s ' , ' F o n t S i z e ' ,14)
3 y l a b e l ( ' N u m b e r of t r a c e s / s h o t g a t h e r ' , ' F o n t S i z e ' ,14)
4 axis ([0 , max ( s h o t _ g a t h e r s ) +1 ,0 , max ( n u m _ t r a c e _ p e r _ s g ) +2])
5 set ( gca , ' Y M i n o r G r i d ' , ' on ' )
Also, here is another example:
1 figure , plot ( sx , ' . ' )

2 x l a b e l ( ' N u m b e r of t r a c e s ' , ' F o n t S i z e ' ,14)
3 y l a b e l ( ' S o u r c e s x−axis l o c a t i o n s ( ft ) ' , ' F o n t S i z e ' ,14)
4 axis tight
5 grid
Figure 2.1 shows different useful plots all of which are obtained from the seismic header
variable H. A final example is plotting the so-called seismic stacking chart (see Figure 2.2) using
the MATLAB function stack_chart.m where the vertical axis shows the shot number and the
horizontal axis shows the shot and receiver x-coordinate (we will discuss stacking charts in more
details in chapter 6):
1 num_shots = length ( shot_gathers );

2 s t a c k i n g _ c h a r t ( sx , gx , num_shots , n u m _ t r a c e _ p e r _ s g ) ;
The reader may enjoy more plots when running the m-files written for this chapter.
2.3.2 DISPLAYING SEISMIC DATA

There exist several ways of displaying seismic sections such as seismic shot gathers and CMP gathers.
The most commonly used displays are the following.
1. The wiggle display: plots seismic trace amplitudes as a function of time.

35
30
Number of traces/shot gather

25
20
15
10
0
0 2 4 6 8 10 12 14 16 18
Shot gather numbers
(a)
6 6
x 10 x 10
2.85 2.85
2.8 2.8
2.75
2.75
2.7
Receivers elevation (ft)
Sources elevation (ft)
2.7
2.65
2.65
2.6
2.6
2.55
2.55
2.5
2.5
2.45
2.45 2.4
2.4 2.35
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Number of traces Number of traces
(b) (c)
Figure 2.1: Various geometrical information plots: (a) the number of traces per shot, (b) the source
elevation profile for each trace and (c) the receiver elevation per seismic trace.
2. The variable area display: shades the area under the wiggle trace to make coherent seismic
events1 evident.
3. The variable density display: represents amplitude values by the intensity of shades of gray
(and sometimes in colors).
1 A seismic event is the arrival of a new seismic wave, usually indicated by a phase change and an increase in amplitude on a seismic
record. It may be a reflection, refraction, diffraction, surface wave, or random noise.
2.3. EXAMINING THE DATA SET 11

6RXUFHV
5HFHLYHUV

/LQHQXPEHU

9
8
7
6
5
4
3
2

ï
[ïD[LVORFDWLRQVIW
Figure 2.2: The stacking chart plot of the seismic data.
The following MATLAB script uses the written function extracting_shots.m to extract, for
example, shot gather number 8. Then this shot gather is displayed using, respectively, the variable
area display scheme and variable density display scheme in both gray and colors (see Figure 2.3):
1 load B o o k _ S e i s m i c _ D a t a . m a t
2 s h o t _ n u m =8;
3 p =0;
4 [ Dshot , dt , dx , t , o f f s e t ]= e x t r a c t i n g _ s h o t s ( D , H , shot_num , p ) ;
5 s c a l e =1;
6 m w i g b ( Dshot , scale , offset , t )
7 x l a b e l ( ' O f f s e t ( ft ) ' , ' F o n t S i z e ' ,14)
8 y l a b e l ( ' T i m e ( s ) ' , ' F o n t S i z e ' ,14)
9 figure , s i m a g e _ d i s p l a y ( Dshot , offset , t ,0)
12 figure , s i m a g e _ d i s p l a y ( Dshot , offset , t ,1)
Also, if we display shot gather number 16 (see Figure 2.4), we notice that trace number 31
amplitudes are increasing as time increases. This may require editing by muting this trace since it
will affect the subsequent processing steps. In chapter 3, we shed more light on this issue. The reader
may as well be interested in displaying a group of seismic shot gathers concatenated together using
the same function extracting_shots.m. For example, one can extract shot gathers number 4-6
(a) (b) (c)
Figure 2.3: Various displays for seismic data shot gather number 8: (a) variable area display, (b) gray-
scaled variable density display and (c) colored variable density display. The color bars in (b) and (c) refers
to the amplitude dynamic range of the data.
where he/she must provide shot_num=4:6; in line 6 of the above script. Figure 2.5 shows these
extracted shot gathers in their variable area display. The provided m-files illustrates more examples.
2.4 THE PROPOSED PROCESSING WORKFLOW

Based on the examination of the seismic data and its header, we are now ready to perform the
following processing steps on the data:
1. Preprocessing involving only the gain application using various methods
2. Ground roll removal via bandpass filtering
3. Spiking deconvolution
4. CMP sorting
5. Velocity analysis on several CMPs using the velocity spectrum method
6. Residual static correction using the surface-consistent method
7. NMO correction and muting
2.5. COMPUTER ASSIGNMENTS 13
(a) (b) (c)
Figure 2.4: Various displays for seismic data similar to those in Figure 2.3 but for shot gather number
16. Clearly, trace number 31 (at offset 0) requires muting (replacing it by zeros) as will be discussed later
in chapter 3.
8. Stacking
9. Migration using Stolt (F-K) post-stack time migration
2.5 COMPUTER ASSIGNMENTS

1. Load the seismic data Book_Seismic_Data.mat.
2. Examine the seismic data header information and obtain the following information: common-
depth-point (CDP) values, number of time samples nt, spatial sampling interval dx and time
sampling interval dt.
3. Display shot gathers 12–15 (as the case in Figure 2.5) using the wiggle plotting, gray-scaled
image plotting, colored image plotting, and wiggle on top of colored plotting. Comment on
the shot gather amplitudes. What do you notice from a trace to another?
Figure 2.5: Variable area display of seismic shot gathers 4-6.
2.6 USEFUL MATLAB FUNCTIONS

load, extracting_geometry.m, extracting_shots.m, plot, stack_chart.m, mwigb.m and
simage_display.m.
15
CHAPTER 3
Quality Control of Real Seismic

Data
3.1 QUALITY CONTROL OF REFLECTION SEISMIC DATA
When seismic data is acquired and recorded, various quality control (QC) steps (called preprocessing
in signal processing) are necessary to carry the remaining seismic data processing forward and
ultimately obtain an accurate seismic image of subsurface structures. The QC process involves a
series of steps to condition the data and prepare it for further quality control and processing. These
steps include:
1. Demultiplexing: The data is transposed from the recording mode, where each record contains
the same time sample from all receivers, to the trace mode where each record contains all time
samples from one receiver. This is usually done in the field.
2. Reformatting: The data is converted from one seismic digital format (e.g., SEG-Y ) to another
format that is convenient to the processing software and used throughout the processing flow
(e.g., MATLAB).
3. Setup of field geometry: The geometry of the field is written into the data (trace headers) in
order to associate each trace with its respective shot, offset, channel, and CMP.
4. Trace editing: During this step, bad traces, or parts of traces, are muted (zeroed) or killed
(deleted) from the data and polarity problems are fixed.
5. Gain application: Amplitude corrections are applied to account for amplitude losses due to
spherical divergence and absorption.
6. Application of field statics: In land surveys, elevation statics are applied to bring the sources
and receivers to a common datum level. This step can be delayed until the static correction
process where better near-surface velocities might be available.
Steps 1-3 have already been taken care of for our data set while we are going to work on step 6 later
on in Chapter 7. In the rest of this chapter, we will perform a simple trace editing step on the data
set and then focus the discussion on gain application to our seismic data.
16 3. QUALITY CONTROL OF REAL SEISMIC DATA
(a) (b)
Figure 3.1: Seismic data shot gather number 16: (a) before and (b) after trace editing by muting.
3.2 TRACE EDITING

We are going to use the east Texas seismic data set where we already have noticed some high noisy
amplitude traces from the previous chapter, particularly, in trace 31 of shot gather number 16. This
is due to the use of a bad geophone, in this case, since we are dealing with land data. Such a recorded
trace must be muted or, in other words, replaced by a zero trace. In MATLAB, there exist many
ways to mute such bad traces but we simply can use the following step to do so on the complete data
set (Figure 3.1):
1 [i , j ]= find ( D == max ( max ( D ) ) ) ;

2 D (: , j ) =0;
3.3 AMPLITUDE LOSSES & THEIR CORRECTION

Among the various QC steps necessary to ultimately obtain an accurate seismic image of the subsur-
face structures is to correct for amplitude losses. Usually, seismic data in its raw state show a noticeable
decrease in the amplitudes of its recorded traces with time (for example, see Figure 3.1(a)). It includes
3.3. AMPLITUDE LOSSES & THEIR CORRECTION 17
geometric divergence effects as waves spread out from a source as well as conversion of the seismic
energy into heat besides other factors such as transmission losses [2]:
• Transmission loss:This occurs at each geological reflector where part of the propagating seismic
incident waves will be reflected, refracted, diffracted, scattered, etc. There is no loss here in
terms of the mechanical energy since the lost energy merely travels somewhere else.
• Geometric divergence: As the seismic wave spreads out from its source, its amplitude decays
by an amount proportional to the reciprocal of the distance from the source to the location of
the propagating seismic wave.
• Absorption: This occurs where the seismic energy is converted into heat by friction. This loss
is proportional to the exponential of the distance from the source.
Now, amplitude correction or gain must be applied to seismic data sets at various stages. At the
preprocessing stage, we may want to correct amplitudes due to geometric divergence as well as
absorption losses. Also, whenever we want to display seismic data, we may want to boost weak
signals by adding more gain to the data. There exist data independent and dependent amplitude
correction schemes. The data-independent scheme corrects the amplitudes using a common scaling
function to all the traces such as the gain through multiplication by a power of time using:
fcorrected (t) = f (t)t α , (3.1)
where f (t) is the seismic trace amplitude to be corrected, t is the time independent variable, and α
is the power of time which controls the change in the amplitude of f (t). Another commonly used
function is the exponential gain function correction:
fcorrected (t) = f (t)eβt , (3.2)
where β is the parameter which controls the exponential function. The following MATLAB m-file
uses the written function iac.m to apply independent amplitude correction using Equation 3.1 (the
reader can also select to use the option T=1 for applying Equation 3.2) on shot gather 8:
1 s h o t _ n u m =8;
2 p =0;
3 [ Dshot , dt , dx , t , o f f s e t ]= e x t r a c t i n g _ s h o t s ( D , H , shot_num , p ) ;
4 pow =2;
5 T =0;
6 Dg = iac ( Dshot , t , pow , T ) ;
We can clearly see the amplitudes enhancements gained on this shot gather in Figure 3.2(b).
We can further analyze the gain increase after the corrections by plotting the average trace amplitudes
envelope in dB for the both shot gathers of Figure 3.2. Figure 3.2 shows such plots not only for
the average trace (a) but also for trace number 33, both are generated using our written MATLAB
function seis_env_dB.m as follows:
(a) (b)
Figure 3.2: Seismic data shot gather number 8: (a) before and (b) after applying amplitude correction
gain method of Equation 3.1.
1 tnum =33;
2 s e i s _ e n v _ d B ( Dshot , Dg , t , tnum )
3 s e i s _ e n v _ d B ( Dshot , Dg , t )
The data-dependent scheme, on the other hand, relies on multiplying each time sample by a
scalar derived from a window of data around the sample. Such a technique is known as automatic
gain control (AGC). Some of the famous AGC techniques include:
• RMS amplitude AGC: This method requires segmenting each trace into fixed time gates and
then:
1. Calculate the RMS value in each gate.

2. Divide the desired RMS scaler by the RMS value of step 1 and multiply it by the amplitude
of the sample at each gate center.
3. Interpolate between these gate centers and multiply the result by the amplitude of samples
corresponding in time.
• Instantaneous AGC: this is a bit different from the RMS AGC:

3.3. AMPLITUDE LOSSES & THEIR CORRECTION 19
20
20
0
0
Envelope (dB) of the average trace
Envelope (dB) trace number :33

ï
ï
ï
ï
ï
ï
ï
ï
ï
Before gain correction Before gain correction

After gain correction After gain correction
ï ï
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time (s) Time (s)
(a) (b)
Figure 3.3: The amplitude envelope gain in dB for (a) the average trace and (b) trace 33 both of Figure 3.2.
1. Calculate the absolute mean value for in a given gate of length w.

2. Divide the desired RMS scaler by the mean value of step 1 and multiply it by the
amplitudes of all the samples in the gate.
3. Slide the gate down by one sample and repeat steps 1-2 until we have calculated all the
amplitudes of all trace samples that have been corrected.
We use the MATLAB function AGCgain.m to apply AGC on shot gather 8 using the RMS AGC
(Figure 3.4(a)) and the instantaneous AGC (Figure 3.4(b)). We generated these results via the
following MATLAB scripts:
1 a g c _ g a t e =0 .5 ;
2 T =1;
3 Dg = A G C g a i n ( Dshot , dt , agc_gate , T ) ;
1 a g c _ g a t e =0 .5 ;
2 T =2;
3 Dg = A G C g a i n ( Dshot , dt , agc_gate , T ) ;
Note the amplitude envelope gain plots for the AGC’ed data in Figure 3.5. A question
that may come into the readers’ mind is: why do the AGC amplitude correction results show the
triangular shapes of high amplitudes and amplitude gaps at the top of the figures? The answer
is that these high-amplitude triangles are the result of noise that fills the whole AGC window
(0.5 s in this case) before the arrival of seismic signals and dividing by their RMS value or
absolute mean amplifies this low-amplitude noise. The amplitude gaps are due to the fact that the
(a) (b)
Figure 3.4: Seismic data shot gather number 8 after applying the AGC using the (a) RMS and (b) in-
stantaneous based methods.
AGC window at these time shifts contains both low-amplitude noise and high-amplitude seismic
signal and dividing by their RMS value or absolute mean amplifies the signal and attenuates the noise.
A final word of caution is that we must be careful when applying amplitude corrections
techniques since they may destroy the signals character. At this preprocessing stage, Equation 3.1
with α = 2, which is known as the t 2 -correction, is used for geometrical spreading correction [6].

1. Load the real seismic data file Book_Seismic_Data.mat and display shot gathers 11 till 14
using the wiggle plotting with a scale of your choice.
2. With α, β = 1.8, 2.2 and 3.4, use both the multiplication by a power of time and the expo-
nential gain function corrections on the selected shot gathers. Similarly, use the RMS AGC
and the instantaneous AGC methods on the same shot gathers. Display and compare your re-
sults with the data before applying the required amplitude corrections. In your opinion, which
method results in the best amplitude correction? Why?
3.5. USEFUL MATLAB FUNCTIONS 21
20 20
0
0

ï
ï
ï
ï
ï
ï
ï
ï
ï

ï ï
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time (s) Time (s)
(a) (b)
20 20
0
0

ï
ï
ï
ï
ï
ï
ï
ï
ï

ï ï
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
Time (s) Time (s)
(c) (d)
Figure 3.5: The amplitude envelope gain in dB for (a) the average trace and (b) trace 33 both of
Figure 3.4(a). Similarly, the amplitude envelope gain in dB for (a) the average trace and (b) trace 33 both
of Figure 3.4(b).
3. Mute the bad traces of shot gather 16 as in Section 3.2.Then apply the method of multiplication
by a power of time with α = 2.0 and all shot gathers and save the processed data with its header
information as Book_Seismic_Data_gain.mat to be used later on.

seis_env_dB.m, iac.m and AGC_gain.m.
23
CHAPTER 4
Seismic Noise Attenuation

4.1 INTRODUCTION
Seismic data are highly corrupted with noise or unwanted energy arising from different kinds of
sources. This unwanted energy can be classified into two main categories [2]: random noise (inco-
herent noise) and coherent noise. Seismic data processing, in general, can never eliminate all noise
contaminating the data. Hence, the objective of seismic data processing is to improve as much as
possible the signal-to-noise ratio (SNR). Using MATLAB examples, we will introduce some impor-
tant means for analyzing the frequency (wavenumber) content of seismic data1 and then attenuate
some of the noise contaminating our seismic data using linear filtering.
4.2 SEISMIC SIGNAL & NOISE

Noise in seismic records is variable in both time and space [7]. Poor seismic records usually have
SNR ratios less than one. One can define the signal of interest (coherent energy) as the energy
which is coherent from trace to trace. Random noise, on the other hand, is the energy that is
incoherent from trace to trace [8]. Furthermore, data from seismic events is correlated and its
energy is generally concentrated in a fairly narrow band of, while noise is more uncorrelated and
broadband [3]. However, this is only true for random noise. Spatially coherent noise is the most
troublesome noise and can be highly correlated and sometimes aliased with the signal [8] and [9]. In
general, noise can be considered as anything other than the desired signal. A more proper definition
of noise contaminating seismic signals can be stated by defining the type of signals we are interested
in. The authors in [8] define the signal of interest as the energy that is most coherent and desirable
for geophysical interpretation of primarily reflected arrivals (signals). Anything other than that is
considered to be unwanted energy, i.e., noise.
4.2.1 RANDOM NOISE

Disturbances in seismic data which lack phase coherency between adjacent traces are considered
to be random noise. Unlike coherent noise energy, such energy is usually not related to the source
that generates the seismic signals. In land seismic records, near-surface scatterers, wind, rain, and
instrument are examples of sources generating random noise. Based on the assumption that random
noise is an additive white Gaussian noise (AWGN) [2, 10], it can be attenuated easily in several
different ways such as frequency filtering, deconvolution, wavelet denoising [11, 12, 13], filtering
1 We are limiting the discussion here to 1 − D and 2 − D. The same concepts can be used for the case of 3 − D seismic data.
24 4. SEISMIC NOISE ATTENUATION
using Gabor representation [14], stacking [1, 2] and many other methods. As discussed before in
Chapter 1, stacking√usually suppresses most of the incoherent noise and, therefore, improves the
SNR by a factor of M, where M is equal to the number of stacked traces.
4.2.2 COHERENT NOISE

Spatially coherent noise is the energy which is generated by the source. It is an undesirable energy that
is added to the primary signals. Such energy shows consistent phase from trace to trace. Examples of
such a type in land seismic records are [2]: multiple reflections or multiples, surface waves like ground
roll and, air waves, coherent scattered waves, dynamite ghosts, etc. Improper removal of coherent
noise can affect nearly all the processing techniques and complicates interpretation of geological
structures (see [2], [1], and [15]). There exist loads of techniques which deal with the problem of
suppressing/attenuating coherent noise that contaminates seismic data [2, 15, 16, 17, 18, 19, 20, 21,
22]. Since our real seismic data contains mainly ground roll coherent noise, we are going to describe
its main characteristics in the following subsection.
4.2.2.1 Ground Roll Noise

They are surface (Rayleigh) waves traveling along the ground surface. They have generally low
velocity (< 1000 m/s), low frequency (< 15 Hz), and high amplitudes. Their time-distance t − x
curves2 are straight lines with low velocities and zero intercepts for an inline source. There might be
several modes of ground rolls in the record because of their dispersive nature (i.e., different frequency
components travel with different velocities). They are attenuated using source and receiver arrays in
the field and various processing methods (e.g., frequency filtering, fan or f − k filtering ). Figure 4.1
shows ground roll on shot gather number 8 from our real seismic data in the time-space (t − x) (a),
frequency-space (f − x) (b), and frequency-wavenumber (f − k) (c) domains. Clearly, we see the
low frequency, low velocity and high amplitudes around the zero-offset location (indicated by the
red lines in (a)).
4.3 SPECTRUM ANALYSIS AND FILTERING OF SEISMIC

DATA
The filtering process is an important step in order to proceed further with the other seismic data
processing steps (refer to typical seismic data processing workflow chart in Figure 1.3). The reader
can examine references [2] and [4] for more information about seismic data filtering and noise
suppression techniques. Since our land seismic data set contains ground roll noise, we simply
can apply frequency linear filters such as band-pass filters (BPF)’s to attenuate their effect. In
particular, BPF’s enhance the overall gain of each seismic shot gather, and increases the SNR ratio
by attenuating low and high frequency noise records, including the ground roll noise. However,
before doing so, let us learn how to analyze the seismic data in the spectrum domain using
2 A time-distance curve is a plot of wave arrival time against the source-to-geophone distance.
4.3. SPECTRUM ANALYSIS AND FILTERING OF SEISMIC DATA 25
(a) (b) (c)
Figure 4.1: (a) Seismic data shot gather number 8 containing ground roll noise (red lines indicate ground
roll noise). (b) Its f − x and (c) its f − k magnitude spectra. The color bars in (b) and (c) indicate the
magnitude values in dB.
MATLAB. There exist different means for such useful analysis: the frequency (or wavenumber)
content of one-dimensional (1 − D) time (or space) signals, two-dimensional (2 − D) like the
frequency-space (f − x) or the frequency-wavenumber (f − k) spectra; each of which can be used
to obtain meaningful interpretation and, therefore, apply a suitable filtering technique. These plots
and their corresponding analysis and interpretations are very useful, particularly, when applying
linear filtering to 1 − D and/or 2 − D seismic data sets.
Now, we can use the following MATLAB m-file containing the written M-functions fx.m
and fk.m to, respectively, obtain the f − x and f − k magnitude spectra in dB as seen Figure 4.1(b)
and (c):
1 clear , clc , c l o s e all

2 load B o o k _ S e i s m i c _ D a t a _ g a i n . m a t
3
4 s h o t _ n u m =8;
5 p =0;
6 [ Dshot , dt , dx , t , o f f s e t ]= e x t r a c t i n g _ s h o t s ( Dgz , H , shot_num , p ) ;
7
8 s c a l e =2;
12
13 [ Data_f , f ]= fx ( Dshot , dt ) ;
14 figure ,
15 p c o l o r ( offset , f , 2 0 ∗ l o g 1 0 ( f f t s h i f t ( abs ( D a t a _ f ) ,1) ) )
16 shading interp ;
17 axis ij ;
18 c o l o r m a p ( jet )
19 colorbar
21 y l a b e l ( ' F r e q u e n c y ( Hz ) ' , ' F o n t S i z e ' ,14)
22 set ( gca , ' x a x i s l o c a t i o n ' , ' top ' )
23 a x i s ([ min ( o f f s e t ) , max ( o f f s e t ) ,0 , max ( f ) ])
24
25 [ Data_fk , f , kx ]= fk ( Dshot , dt , dx ) ;
26 figure ,
27 p c o l o r ( kx , f , 2 0 ∗ l o g 1 0 ( f f t s h i f t ( abs ( D a t a _ f k ) ) ) )
28 shading interp ;
29 axis ij ;
30 c o l o r m a p ( jet )
31 colorbar
32 x l a b e l ( ' W a v e n u m b e r (1/ ft ) ' , ' F o n t S i z e ' ,14)
33 y l a b e l ( ' F r e q u e n c y ( Hz ) ' , ' F o n t S i z e ' ,14)
34 set ( gca , ' x a x i s l o c a t i o n ' , ' top ' )
35 axis ([ min ( kx ) , max ( kx ) ,0 , max ( f ) ])
Now, we design an Finite Impulse Response (FIR) BPF digital filter and apply it to each
seismic trace in the shot gather number 8 using the written M-function bpf_fir.m as follows (with
the given filter specifications):
1 N =100;
2 c u t _ o f f = [ 1 5 ,60];
3 [ Dbpf , D b p f _ f ]= b p f _ f i r ( Dshot , dt , N , c u t _ o f f ) ;
Figure 4.2 represents shot gather 8 before and after applying the above mentioned BPF on
each trace along with the difference between both of them where clearly a great amount of the ground
roll noise has been filtered. The f − x magnitude spectra of Figure 4.2 is also seen in Figure 4.3
where the spectrum was banded between 15 and 60 Hz and the remaining was left out. A final word
of caution is that although frequency filtering like BFP has improved the data SNR by attenuating
the ground roll noise, it has also lowered the vertical resolution of the seismic data. This is common
in seismic data processing and will be taken care of by applying deconvolution as we will explore in
the next chapter.
4.4. COMPUTER ASSIGNMENTS 27
(a) (b) (c)
Figure 4.2: Seismic data shot gather number 8 containing ground roll noise (a) before and (b) after BPF
filtering. (c) The difference between (a) and (b).

1. Load the QC’ed real seismic data called Book_Seismic_Data_gain.mat. Extract and display
shot gather 2 using the wiggle plotting with a scale of your choice. Using the MATLAB
functions fft and fftshift, plot the magnitude and phase spectra for traces 1, 10, and 15
of this shot gather. Write down your own observations.
2. For the same shot gather, plot the magnitude and phase spectra of its f − x and f − k data
representation. Use both linear and dB magnitudes. Comment on your results. In your opinion,
what are the main differences in terms of interpretation between both the f − x and the f − k?
3. Use the MATLAB’s signal processing toolbox to design and apply the following filters:
a. An FIR Low-pass filter (LPF) using any windowing method with a filter order of 50 and
a cut-off of 15 Hz. Select any trace and apply your designed filter to it and display both
the original trace and its LPF version. Comment on both traces by considering their time
and frequency domain representations.
b. Repeat (a) but with an Infinite Impulse Response (IIR) filter with an order of 5. Design
the IIR filter using the bilinear transformation method.
(a) (b) (c)
Figure 4.3: The f − x magnitude spectra of the seismic data shot gather number 8 containing ground
roll noise: (a) before and (b) after BPF filtering as well as for (c) the difference.
c. Apply the FIR and IIR LPFs from parts (a) and (b), respectively, on each trace of shot
gather 2. Display the original shot gather and the processed shot gathers. What can you
notice from your results?
Note: here and in the next parts of this assignment, make sure that you use the same wiggle
display scale of your filtered shot gathers in order to run fair comparisons.
4. Repeat Part 2 but with a high-pass filter (HPF) with a cut-off 60 Hz. In your opinion, which
is better the LP filtered or HP filtered data results?
5. Apply band-pass filtering with cut-off values of 15 and 60 Hz using the M-function
bpf_fir.m to all the shot gathers and save the resultant data set (with its header) as
Book_Seismic_Data_gain_bpf.mat to be used in the coming chapters.

fx.m, fk.m, bpf_fir.m, abs, pcolor and log10.
29
CHAPTER 5
Seismic Deconvolution
5.1 INTRODUCTION
After we have performed frequency filtering on seismic data via BPFs, the seismic data is smoothed
and, hence, its vertical resolution is affected due to the loss of some of its original wider frequency
band. One of the main aims of seismic deconvolution is to increase the vertical resolution of the
data by compressing the source wavelet (to a zero-phase spike, if possible). This is known as spiking
deconvolution. Seismic deconvolution, however, is not limited to this application where we can use it
as well for noise attenuation such as multiples. Since our land seismic data set requires at this stage
enhancement of its vertical resolution, we are going to elaborate here on spiking deconvolution and
its application to our data set. However, before we get to deconvolution, we must define the seismic
convolutional model.
5.2 THE SEISMIC CONVOLUTIONAL MODEL

The seismic convolutional model (recall Figure 1.1) is used to explain how the seismic trace is
formed. The seismic convolutional model approximates the earth by a linear system. A linear system
is one whose output o(t) is given by the convolution of its input i(t) with its response r(t). That is,
o(t) = i(t) ∗ r(t). In the seismic convolutional model:
• The system’s output is o(t) = s(t), where s(t) is the recorded seismic trace.
• The system’s input i(t) = w(t), where w(t) is the wavelet generated by the seismic source.
• The system’s response r(t) = e(t), where e(t) is a series of impulses corresponding, in time
and amplitude, to the reflection coefficients at layers boundaries. e(t) is also known as the
reflectivity (series). Therefore, according to this model, the seismic trace s(t) is given by:
s(t) = w(t) ∗ e(t). (5.1)
A random noise component γ (t), if present, is additive; hence, the noisy seismic trace becomes:
sn (t) = w(t) ∗ e(t) + γ (t). (5.2)
The forward seismic convolutional model is used to compute synthetic seismograms s(t) given the
source wavelet w(t) and earth’s reflectivity e(t) (Equations 5.1 and 5.2). The deconvolution (inverse
seismic convolutional model) is used to:
30 5. SEISMIC DECONVOLUTION
1. Compute the earth’s reflectivity e(t) given the seismic trace s(t) and the source wavelet w(t).
This is the most common objective of deconvolution. If the source wavelet is known, the
deconvolution becomes deterministic. Whereas, if the source wavelet is not known, the de-
convolution becomes statistical.
2. Compute the source wavelet w(t) given the seismic trace s(t) and the earth’s reflectivity e(t).
This is used if a seismic trace is recorded near a borehole.
The seismic convolutional model is widely accepted because it agrees well with the observed seismic
traces. The seismic convolutional model commonly assumes the following [2].
1. The earth is made up of horizontal layers of constant velocity.
2. The source generates only a primary wave, which is reflected on layer boundaries at normal
incidence.
3. The source wavelet is stationary. That is, it does not change its shape as it travels in the
subsurface.
4. The noise component γ (t) is zero.
5. The Earth’s reflectivity e(t) is a white random series of impulses.
6. The seismic wavelet is a minimum-phase wavelet, which means that the wavelet has its energy
concentrated at its start time.
Often, one or more of these assumptions might not be satisfied; in which case, advanced techniques
of deconvolution have to be implemented.
5.3 SPIKING DECONVOLUTION AS A FILTERING PROCESS

We will illustrate this concept using the spiking deconvolution as an example. The aim of spiking
deconvolution is to compress the source wavelet w(t) into a zero-phase spike of zero width, i.e.,
δ(t). This means that we are eliminating the effect of the source wavelet and leave only the effect
of the Earth’s reflectivity in the seismogram. We can achieve this by convolving the seismic trace by
the inverse filter, h(t), of the source wavelet defined as:
w(t) ∗ h(t) = δ(t). (5.3)
Taking the Fourier transform of Equation 5.3:
W (f )H (f ) = 1, (5.4)
where W (f ), H (f ), and 1 are the Fourier transforms of w(t), h(t), and δ(t), respectively. From
Equation 5.4, we can see that:
H (f ) = 1/W (f ) = [1/|W (f )|] exp[−φw (f )],

5.3. SPIKING DECONVOLUTION AS A FILTERING PROCESS 31
Figure 5.1: The Wiener optimum filter model.
which implies that |H (f )| = 1/|W (f )|, and φh (f ) = −φw (f ). Note that |W (f )| and φw (f ) are
the amplitude and phase spectra of w(t) and |H (f )| and φh (f ) are the amplitude and phase spectra
of h(t). Therefore, the amplitude spectrum of the inverse filter is the reciprocal of that of the source
wavelet whereas its phase spectrum is the negative of that of the wavelet. Taking the inverse Fourier
transform of the H (f ), we get the desired inverse filter h(t). Finally, spiking deconvolution is
accomplished then by convolving the inverse filter h(t) with the seismic trace s(t):
h(t) ∗ s(t) = h(t) ∗ [w(t) ∗ e(t)] = [h(t) ∗ w(t)] ∗ e(t) = δ(t) ∗ e(t) = e(t), (5.5)
which is the earth’s response that we want to extract from the seismic trace.
In practice, the data is uniformly sampled and, therefore, we can assume that all the signals
in Equation 5.2 are sampled and we can then rewrite it as:
sn (n) = w(n) ∗ e(n) + γ (n), (5.6)
where n represents the time index. Hence, we can also use the z-transform in Equation 5.4, which
results in H (z) = 1/W (z). H (z) is an infinite polynomial of z that is convergent only if w(n) is
a minimum-phase wavelet. For practical reasons, the infinite polynomial H (z) has to be truncated
to N terms (i.e., FN (z)). Truncation generates less error if w(n) is a minimum-phase wavelet and
we include more terms of HN (z). The truncated filter hN (n) is calculated by taking the inverse
z-transform of HN (z). Because of truncation, convolution of the truncated filter hN (n) with the
wavelet will not give the desired output d(n) = δ(n) ≡ (1, 0, 0, …). The actual output y(n) is given
by:
y(n) = hN (n) ∗ w(n). (5.7)
Apparently, y(n) = d(n) and there will be a truncation error E defined as:

E= [d(n) − y(n)]2 . (5.8)
n
5.4 SPIKING DECONVOLUTION USING WIENER
OPTIMUM FILTERS
Wiener optimum filtering involves designing a filter h(n) so that the error E in Equation 5.8 between
the desired output d(n) and the actual output y(n) is minimum (refer to Figure 5.1). The actual
output y(n) is given as:
y(n) = h(n) ∗ x(n), (5.9)
where x(n) is the input. Substituting equation 5.9 into 5.8:

E= [d(n) − {h(n) ∗ x(n)}]2 . (5.10)
n
The goal is to compute the filter coefficients (h(0), h(1), . . . , h(N − 1)) so that the error E
is minimum, where the filter length, N , must be predefined. This is a typical least-squares problem,
and the minimum error is attained by setting the partial derivative of E with respect to h(j ) to zero:
∂E/∂h(j ) = 0, (5.11)
where j = 0, 1, 2, . . . , (N − 1). Applying Equation 5.11 on 5.10 and simplifying, the result can be
expressed as:
N−1
r(j − n)h(n) = g(j ) (5.12)
n=0
for j = 0, 1, 2, ..., (N − 1). In matrix format, Equation 5.12 can be written as:
⎡ ⎤⎡ ⎤ ⎡ ⎤
r(0) r(1) · · · r(N − 1)
r(2) h(0) g(0)
⎢ r(1) r(0) · · · r(N − 2)
r(1) ⎥⎢ h(1) ⎥ ⎢ g(1) ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ .. .. ..
.. .. ⎥⎢ .. ⎥=⎢ .. ⎥
⎣ . . . . . ⎦⎣ . ⎦ ⎣ . ⎦
r(N − 1) r(N − 2) r(N − 3) · · · r(0) h(N − 1) g(N − 1)
(5.13)
In Equations 5.12 and 5.13, g(j ) is the j th term of the crosscorrelation between d(n) and x(n):
g(j ) = d(j ) ⊗ x(j ), j = 0, 1, 2, . . . , N − 1. (5.14)
Note that g(n) = d(n) ⊗ x(n) = x(n) ⊗ d(n) and that we only need lags 0, 1, . . . , (N − 1). rj −i
is the j th term of the autocorrelation of x(n):
r(j ) = x(j ) ⊗ x(j ), j = 0, 1, 2, . . . , N − 1. (5.15)
Note that r(n) = x(n) ⊗ x(n) = r(−n) and we only need lags 0, 1, . . . , (N − 1). If g(j ) and
r(j − τ ) are known, Equation 5.13 can be solved uniquely to find the filter h(n). Equations 5.12
and 5.13 are known as the normal equations. The autocorrelation (r(j )) matrix of Equation 5.13 is
5.5. THE TRACE-WAVELET RELATION 33
a Toeplitz matrix that can be inverted efficiently using the Levinson recursion procedure.
As an application of the normal equations, we will use them to design the inverse fil-
ter of the source wavelet such that the desired output is a zero-lag spike . The input x(n) is
the source wavelet w(n) = (w(0), w(1), . . . , w(N − 1)) of length N , the desired output d(n)
is a zero-lag impulse function δ(n) = (1, 0, ..., 0) of length 2N − 1, and the inverse filter is
h(n) = (h(0), h(1), . . . , h(N − 1)) of length N. Therefore, the input autocorrelations are:
r(0) = w2 (0) + w 2 (1) + . . . + w 2 (N − 1)

r(1) = w(0)w(1) + w(1)w(2) + · · · + w(N − 2)w(N − 1)
...
r(N − 1) = w(0)w(N − 1).
The crosscorrelations on the right-hand side of the normal equations are:
g(0) = w(0)
g(1) = 0
...
g(N − 1) = 0.
Therefore, the normal equations become:

⎡ ⎤⎡ ⎤ ⎡ ⎤
r(0) r(1) · · · r(N − 1)
r(2) h(0) w(0)
⎢ r(1) r(0) · · · r(N − 2)
r(1) ⎥⎢ h(1) ⎥ ⎢ 0 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ .. .. ..
.. .. ⎥⎢ .. ⎥=⎢ .. ⎥ . (5.16)
⎣ . . . . . ⎦⎣ . ⎦ ⎣ . ⎦
r(N − 1) r(N − 2) r(N − 3) · · · r(0) h(N − 1) 0
The actual application of this theory to the exploration of subterranean layers, such as spiking
and prediction deconvolution, began with the work of the Geophysical Analysis Group at the
Massachusetts Institute of Technology (MIT) between 1960 and 1965, where it was a significant
milestone in the history of seismic data processing [4, 23, 24].
5.5 THE TRACE-WAVELET RELATION

We just saw that we need to know the autocorrelation of the source wavelet in order to perform
spiking deconvolution. However, for most impulsive sources, the source wavelet is not known and
deconvolution or filtering cannot be performed directly. We need to either estimate the source wavelet
or find relations between the amplitude spectra or autocorrelations of the seismic trace and source
wavelet. Here, we investigate the latter case.
5.5.1 AMPLITUDE SPECTRA
The earth response e(n) can safely be approximated as a white random series of impulses [2, 23]. If
this is the case, the amplitude spectrum of e(n) will be constant. That is:
|E(f )| ⇐⇒ E0 . (5.17)
Using Equation 5.1 to find the trace amplitude spectrum and substituting Equation 5.17 yields:
|S(f )| = |W (f )||E(f )| = E0 |W (f )|. (5.18)
Equation 5.17 means that the amplitude spectrum of the seismic trace is a scaled version of the
amplitude spectrum of the source wavelet.
5.5.2 AUTOCORRELATIONS
Because of the random nature of e(n), its autocorrelation re (n) is generally zero everywhere except
at n = 0, where it is equal to the energy in e(n) given by:

re0 = e2 (n) = e2 (0) + e2 (1) + . . . (5.19)
n
Therefore, re (n) can be approximated as:
re (n) = re0 δ(n). (5.20)
Using this fact about re (n), we can derive a relation between the trace and wavelet autocorrelations
(rs (n) and rw (n)) as follows:
rs (n) = s(n) ⊗ s(n)

= s(n) ∗ s(−n)
= [w(n) ∗ e(n)] ∗ [w(−n) ∗ e(−n)]
= [w(n) ∗ w(−n)] ∗ [e(n) ∗ e(−n)]
= rw (n) ∗ re (n)
= re0 (n) ∗ rw (n)
= re0 rw (n). (5.21)
Equation 5.21 means that the autocorrelation of the seismic trace is a scaled version of the autocor-
relation of the source wavelet. The benefit of Equations 5.19 and 5.21 is that we can use |S(f )| or
rs (n) whenever |W (f )| or rw (n) are needed in deconvolution or filtering. Provided the randomness
and whiteness assumptions about e(n) are satisfied, all we miss by using |S(f )| or rs (n) instead of
|W (f )| or rw (n) is a scaling factor.
5.6 SPIKING DECONVOLUTION IN PRACTICE

To perform spiking deconvolution in practice, we need to set the following parameters:
5.6. SPIKING DECONVOLUTION IN PRACTICE 35
• Autocorrelation window (w): This sets up the part of seismic trace from which we will select
the elements of the autocorrelation matrix in the normal equations.
• Filter length (N ): This sets up the length of the spiking filter h(n).
• Percent prewhitening (ε): This sets up the amount of white random noise we want to include
into our autocorrelation matrix to stabilize the solution of the normal equations.
5.6.1 AUTOCORRELATION WINDOW

The choice of deconvolution parameters depends largely on the characteristics of the autocorrelation
of the seismic trace. Therefore, it is important to choose a suitable autocorrelation window (gate)
that will be used to calculate the deconvolution parameters. The autocorrelation window should
include the part of the record that contains useful reflection signal and should exclude coherent (e.
g., ground roll) or incoherent noise (e.g., later parts of the record). The length of the autocorrelation
window should be greater than eight times the largest operator length that will be used for that data
set [2]. Here, we use the whole time window length of the data for spiking deconvolution.
5.6.2 FILTER LENGTH

The filter (operator) length should be equal to the wavelet length. The first transient zone of the
trace autocorrelation is the part that mostly represents the autocorrelation of the source wavelet. The
first transient zone is the first part of the autocorrelation that contains high amplitudes. The operator
length should be selected so that it is approximately equal to the length of the first transient zone
of the trace autocorrelation. The optimum operator length should not leave considerable amount of
energy in the trace autocorrelogram.
5.6.3 PERCENT PREWHITENING

The amplitude spectrum of the spiking deconvolution operator (inverse filter) is the reciprocal
of that of the source wavelet. If the amplitude spectrum of the source wavelet is zero at some
frequencies, then the amplitude spectrum of inverse filter will be unstable (i. e., it will be infinite at
these frequencies). A similar numerical instability might be encountered when inverting the trace
autocorrelation matrix if the determinant of the autocorrelation matrix is zero. To ensure numerical
stability, we introduce an artificial level of white random noise into the trace amplitude spectrum
and autocorrelation before deconvolution. This process is called prewhitening.
Prewhitening is achieved by adding a white random noise, with a very small variance (ε), to
the trace amplitudes at every time sample. This is equivalent to adding a very small positive constant
to the zero-lag autocorrelation (r(0) in Equation 5.17) of the trace [2]. It is also equivalent to adding
a very small positive constant to the amplitude spectrum of the trace at every frequency component.
The magnitude of prewhitening is measured as a percentage of the zero-lag autocorrelation value
r(0). In practice, typically 0.1% to 0.3% (i.e., 0.001r(0) to 0.003r(0)) prewhitening is standard in
processing.
5.6.4 SPIKING DECONVOLUTION OF OUR DATA SET

We use here our written M-functions spiking_decon.m and auto_correlation_map.m to per-
form spiking deconvolution filtering based on 5.17 and calculate the autocorrelogram, respectively.
The m-file is as follows:
1 load B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f . m a t
2
3 shot_num =4:6;
4 p =1;
5 [ Dshot , dt , dx , t , o f f s e t ]= e x t r a c t i n g _ s h o t s ( Dbpf , H , shot_num , p ) ;
6 [ nt , nx ]= size ( Dshot ) ;
7
8 s c a l e =5;
10 x l a b e l ( ' T r a c e n u m b e r ' , ' F o n t S i z e ' ,14)
12
13 m a x _ l a g =0 .2 ;
14 [ Dauto , lags ]= a u t o _ c o r r e l a t i o n _ m a p ( Dshot , max_lag , dt ) ;
15 s c a l e =5;
16 m w i g b ( Dauto , scale , offset , lags )
18 y l a b e l ( ' T i m e lag ( s ) ' , ' F o n t S i z e ' ,14)
19
20 mu =0 .1 ;
21 Ds = s p i k i n g _ d e c o n ( Dshot , max_lag , mu , dt ) ;
22 s c a l e =5;
23 m w i g b ( Dt , scale , offset , t )
Figure 5.2 shows the first three shot gathers of the data before applying spiking deconvolution.
Their autocorrelograms are shown in Figure 5.3, while Figure 5.4 shows them after applying spiking
deconvolution. We notice that the data became more spiky and we can further analyze this in the
spectrum domain via the power spectral density of the average traces as we see in Figure 5.5. This
figure was generated using the following commands:
1 D a v g _ b e f o r e = mean ( Dshot ' ) ;

2 fs =1/ dt ;
3 [ D a v g _ b e f o r e , f ] = p e r i o d o g r a m ( D a v g _ b e f o r e ,[] ,2∗ nt , fs ) ;
4 D a v g _ b e f o r e = D a v g _ b e f o r e / max ( D a v g _ b e f o r e ) ;
5 D a v g _ b e f o r e = 2 0 ∗ l o g 1 0 ( abs ( D a v g _ b e f o r e ) ) ;
6
7 D a v g _ a f t e r = mean ( Ds ' ) ;
8 [ D a v g _ a f t e r , f ] = p e r i o d o g r a m ( D a v g _ a f t e r ,[] ,2 ∗ nt , fs ) ;
5.6. SPIKING DECONVOLUTION IN PRACTICE 37
Trace number
0 10 20 30 40 50 60 70 80 90 100
0
0.5
Time (s) 1.5
2.5
Figure 5.2: Shot gathers: 4, 5 and 6 before applying spiking deconvolution.
Trace number
0 10 20 30 40 50 60 70 80 90 100
0
0.02
0.04
0.06
Time lag (s)
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Figure 5.3: Autocorrelograms of shot gathers: 4, 5 and 6.
9 D a v g _ a f t e r = D a v g _ a f t e r / max ( D a v g _ a f t e r ) ;
10 D a v g _ a f t e r = 2 0 ∗ l o g 1 0 ( abs ( D a v g _ a f t e r ) ) ;
11
12 f = l i n s p a c e (−0.5 ,0 .5 , 2 ∗ nt ) / dt ;
13 figure , plot ( f , D a v g _ b e f o r e , f , D a v g _ a f t e r , ' r−−' )
14 x l a b e l ( ' F r e q u e n c y ( Hz ) ' , ' F o n t S i z e ' ,14)
15 y l a b e l ( ' N o r m a l i z e d PSD ' , ' F o n t S i z e ' ,14)
16 grid
17 legend (' Before decon ' ,' After decon ')
Trace number
0 10 20 30 40 50 60 70 80 90 100
0
0.5
Time (s)
1.5
2.5
Figure 5.4: Shot gathers: 4, 5 and 6 after applying spiking deconvolution.
0
Before decon
After decon
ï
ï
Normalized PSD
ï
ï
ï
ï
ï
ï
0 50 100 150 200 250
Frequency (Hz)
Figure 5.5: PSD of the average trace of shot gathers: 4, 5 and 6 before and after spiking deconvolution.
Finally, to further enhance the results before we sort the data in the next chapter, we have
applied instantaneous AGC with window length of 0.5 s to the deconvolved data as seen in Figure 5.6.
We applied AGC to compensate for the lost amplitudes after deconvolution.
1. Apply spiking deconvolution using the M-function spiking_decon.m followed by an in-

stantaneous AGC to all the shot gathers and save the resultant data set (with its header) as
Book_Seismic_Data_gain_bpf_sdecon_gain.mat to be used in the coming chapters.
5.8. USEFUL MATLAB CODES 39
Trace number
0 10 20 30 40 50 60 70 80 90 100
0
0.5
Time (s)
1.5
2.5
Figure 5.6: Shot gathers: 4, 5 and 6 after applying spiking deconvolution and instantaneous AGC.
2. Use the formulation of Equation 5.13 to generalize the M-function spiking_decon.m so it

can be used for predication deconvolution. For information, we recommend the reader to look
into [2].
5.8 USEFUL MATLAB CODES

spiking_decon.m, auto_correlation_map.m, and periodogram.
41
CHAPTER 6
Carrying the Processing Forward

6.1 INTRODUCTION
After we have performed various seismic data processing steps such as: data quality control, frequency
filtering and deconvolution on the east Texas real seismic data, we are ready now to prepare the data
for revealing a true image of the subsurface. This requires compressing the seismic data to infer a
first approximation for such image and may include:
• sorting the shot gathered data into common mid-point gathers (CMP);
• picking appropriate stacking velocities and applying accordingly normal move-out corrections
(NMO); and
• stacking all CMPs, i.e., average each CMP traces to form one stacked CMP trace and then
concatenate all stacked CMP traces together. This will be our first approximation for the
subsurface image of interest called the stacked section.
Therefore, we are going in this chapter to briefly explain and apply to our data each process alone in
order to obtain a stacked section.
6.2 COMMON MIDPOINT SORTING

Surface seismic reflection surveys are commonly acquired using the common midpoint (CMP)
method. In this method, points in the subsurface are covered more than once by primary reflections
from different source-receiver pairs. See Figures 6.1 and 6.2 for examples of shot gather and CMP
configurations. Other names for this method include common reflection point (CRP) and common
depth point (CDP). Traces reflected from the same midpoint form a CMP gather while the number
of traces in a gather is called the fold of that gather. Common CMP folds are 80, 120, 240, and 480.
We note here that the CMP spacing is half of the trace (receiver) spacing in a survey.
Seismic data is acquired in the shot gather mode while most seismic data processing is per-
formed in the CMP-offset mode. Therefore, we need to sort the traces between these modes.
Generally, we need a way to sort the traces into other modes as well. For this purpose, we use stack-
ing charts. A stacking chart is a chart in which the x-axis indicates the geophone location and the
y-axis indicates the source location. It is used to sort the traces into various modes or gathers such as
shot, receiver, offset, or CMP. On a stacking chart (refer to Figure 6.3 which shows the data stacking
chart):
42 6. CARRYING THE PROCESSING FORWARD
Figure 6.1: Two-sided common shot gather configuration where five receivers from each side of the shot
record.
Figure 6.2: Common mid-point gather (CMP) configuration. We see that each of the shot-receiver
combinations is centered around the same mid-point.
• Points along one diagonal have a common midpoint (the gather is called a common midpoint
(CMP) gather).
• Points along the other diagonal have a common offset (the gather is called a common offset
gather (COG)).
• Points along a horizontal line have a common source (the gather is called a common source
gather).
• Points along a vertical line have a common receiver (the gather is called a common receiver
gather (CRG)).
In order to sort the our seismic data set, we can use the M-function ssort.m as follows:
1 load B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n . m a t
6.2. COMMON MIDPOINT SORTING 43
Figure 6.3: The stacking chart plot of the seismic data.
2 [ Dsort , H s o r t ] = s s o r t ( Ds_gain , H ) ;
3 save B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d . m a t D s o r t H s o r t
Note that we have saved the sorted file so it will be easily accessible later on. Furthermore, if we run
the following script:
1 [ cmps , f o l d _ c m p ]= e x t r a c t i n g _ c m p _ f o l d _ n u m ( Dsort , H s o r t ) ;
2 figure , stem ( cmps , fold_cmp , '−' )
3 x l a b e l ( ' CMP n u m b e r s ' , ' F o n t S i z e ' ,14)
4 y l a b e l ( ' F o l d ' ,' F o n t S i z e ' ,14)
5 set ( gca , ' Y M i n o r G r i d ' , ' on ' )
we obtain the fold or number of traces per CMP as shown in Figure 6.4 based on the M-function
extracting_cmp_fold_num.m. From this figure, we see that the data starts at CMP 203 with
a single trace then it increases until it reaches the maximum fold of 18 at CMP number 235 and
then ends at CMP 266 with one trace. This is also confirmed by the stacking chart in Figure 6.3.
Figure 6.5 shows several CMP gathers extracted from the sorted data and its header displayed using
the following script:
18
16
14
12
10
Fold
8
0
200 210 220 230 240 250 260 270
CMP numbers
Figure 6.4: The fold (number of traces per CMP) versus the CMP numbers.
1 cmp_num =208;
2 [ Dcmp , t , cdp ]= e x t r a c t i n g _ c m p ( Dsort , Hsort , c m p _ n u m ) ;
3 s c a l e =2;
4 m w i g b ( Dcmp , scale , cdp , t )
5 x l a b e l ([ ' CMP : ' , n u m 2 s t r ( c m p _ n u m ) , ' ' ] , ' F o n t S i z e ' ,14)
Note the number of traces per CMP gather varies from a CMP gather to another and is consistent
with Figure 6.4.
6.3 VELOCITY ANALYSIS

The objective of velocity analysis is to determine the seismic velocities of layers in the subsurface.
Seismic velocities are used in many processing and interpretation stages such as: spherical divergence
correction, NMO correction and stacking, interval velocity determination, migration, and time to
depth conversion. There are different types of seismic velocities such as: the NMO, stacking, RMS,
average, interval (Dix), phase, group, and migration velocities. The velocities that can be derived
reliably from the time-space (t − x) data (such as those in Figure 6.5) are the NMO, RMS, and
stacking velocities. We are going to work with stacking velocities which will be used afterwards for
NMO correction of our data.
We can obtain or, to be more precise, pick stacking velocities by running a velocity analysis
on a given CMP gather, where we aim to obtain velocities that flatten out hyperbolas (which mainly
6.3. VELOCITY ANALYSIS 45
(a) (b) (c)
Figure 6.5: CMP gathers number (a) 208, (b) 234, and (c) 250 of our CMP-sorted seismic data.
represent reflections) on CMP gathers. This will increase the energy contained in a seismic event
after adding all the traces of that particular CMP together, i.e., we increase the power of the overall
stack.
There exist many means that we can use to compute the stacking velocities such as the the
velocity spectrum and the constant velocity stack techniques. Both are commonly used and we will select
the velocity spectrum technique to obtain the stacking velocities of our data.
6.3.0.1 The Velocity Spectrum

This method attempts to find the stacking velocity to each reflector. It maps the time-space data of
a single CMP gather onto a velocity-spectrum plane. In the velocity-spectrum plane, the vertical
axis is t0 and the horizontal axis is vs . The method consists of the following steps.
1. Select a CMP gather that has a relatively high SNR ratio. The CMP gather should be sorted
in offset.
2. Determine the minimum (usually equals 0) and maximum (usually equals to the record length)
t0 that you want to analyze.
3. Determine the minimum, maximum, and increment of vs to be attempted.

4. Determine the gate width, w, around the reference time t0 .This is usually equal to the dominant
period of the data.
5. Start with the minimum t0 and vs .

6. Compute t (x) = t02 + x 2 /vs2 , where t0 and vs are set to the minimum t0 and minimum vs
of step (5) and x is the offset of the traces in that CMP gather.
7. The amplitudes in a gate of width w centered about t (x) calculated from step (6) are selected
from all the traces in the gather. We will call the amplitude of a time sample within this gate
Aij , where 1 ≤ i ≤ N, 1 ≤ j ≤ M, N is the number of samples in the gate w, and M is the
number of traces in the CMP gather.
8. The sum of the amplitudes corresponding to the first time sample of the gate on all traces in

the gather is computed and squared ( M 2

j =1 Aij ) .
9. Step 8 is repeated for all the time samples in the gate w.

N
M
10. The squared sums are added together to give the stack energy Es = i=1 ( j =1 Aij )
2.
11. Now, sum up the squared amplitudes of the first sample on every trace .
12. Repeat step (11) for all the other samples in the gate.

N
M
13. The sums of the squares are added together to give the prestack energy Eu = i=1
2
j =1 Aij .
14. Calculate the semblance NE = (1/M)(Es/Eu). Note that: 0 ≤ NE ≤ 1, and that it is larger
if the amplitudes in the gate are aligned following a hyperbola whose t0 and vs are equal to
those of the hyperbola you are currently fitting.
15. Now you have one point on the velocity-spectrum plane, namely (minimum t0 , minimum vs ,
N E).
16. While fixing t0 , increment V s and repeat steps (6)-(14) until you reach the maximum vs .
17. Increment t0 by L = N/2 samples and repeat steps (5)-(16) until you reach the maximum t0 .
18. For a reflection that has a zero-offset two-way traveltime1 equal to t0 , its correct vs is the one
that is associated with the maximum semblance occurring at that t0 .
19. Select another CMP gather and repeat steps (5)-(17).

1Two-way travel time is the total time it takes a seismic wave to travel down from the source to a reflection point and back up to
the receiver.
6.4. NORMAL MOVEOUT (NMO) CORRECTION 47
Applying these steps should end up with a set of picks (t0 , vs ) for every selected CMP. We do not
pick for each CMP. Hence, to find the (t0 , vs ) sets for the other, unprocessed CMPs, we interpolate
them.
The parameters Es, Eu, and NE are measures of coherency (similarity) of the signal along
a hyperbolic curve [25]. The coherency measure is usually displayed as a contour plot. Other
important parameters to consider when using the velocity spectrum method are the minimum,
maximum, and increment vs . The velocity spectrum method is more suited for noise-contaminated
data sets.
The following script uses the M-function vel_picking.m that will interact with its user
through the semblance plot and the computer mouse. The user will pick possible velocities by
pointing and clicking the pointer on the semblance plot. It will also display the selected CMP
gathers along with their semblance as shown, for example, in Figure 6.6(a):
1 load B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d . m a t
2
3 c m p _ s t e p =5;
4 cmp_start =205;
5 cmp_end =255;
6 vmin = 5000;
7 dv = 2 0 0 ;
8 nv = 51;
9 n _ p t s =8;
10
11 [ v_stack , t _ s t a c k ]= v e l _ p i c k i n g ( Dsort , Hsort , vmin , dv , nv , cmp_start , ...
12 cmp_end , cmp_step , n_pts ) ;
13
14 save B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d _ v e l o c i t i e s . m a t
15 v_stack t_stack cmp_step cmp_start cmp_end
The output of the M-function vel_picking.m is composed of two vectors: one containing the
stacking times and the other containing the stacking velocities, where we are going to use them for
NMO correction in the coming section.
6.4 NORMAL MOVEOUT (NMO) CORRECTION

NMO correction will apply the picked stacking velocities to CMP traces and corrects with the non-
zero offset traveltimes for their additional traveltime from source to receiver. The goals of NMO
correction are:
• Preparation of the data for stacking.
• Estimating the NMO velocity function.

The normal moveout
tNMO (x) is defined as the time difference (on the seismic section) between
the two-way traveltime, t (x), at an offset x = 0 and the two-way traveltime at zero-offset (i.e.,
x = 0) t0 :

tN MO (x) = t (x) − t0 . (6.1)
In the case of a single horizontal layer with constant velocity, the time-distance (t − x) curve is
exactly a hyperbola given by:
t 2 (x) = t02 + x 2 /v 2 , (6.2)
where v is the layer velocity. We can show that the NMO-correction, in this case, is given by:
x2
tN MO (x) ≈ , (6.3)
2t0 v 2
where the approximation is better for small offsets or where the maximum offset over the reflector
depth is less than 2. For NMO correction,
tN MO (x) is subtracted from t (x) such that the two-way
traveltime at offset x after NMO correction, tN MO (x), is approximately equal to t0 :
tN MO (x) = t (x) −
tN MO (x) ≈ t0 . (6.4)
Equation 6.3 implies that

tN MO (x) increases with offset and decreases with depth (or t0 ) and
velocity. We can see from Equation 6.4 the following.
• If the correct NMO velocity is used for NMO correction, the event will be horizontally aligned
at t = t0 .
• If a higher velocity is used for NMO correction, then the event will be undercorrected (i.e.,
concave down).
• If a lower velocity is used for NMO correction, then the event will be overcorrected (i.e.,
concave up).
For multiple horizontal constant-velocity layers, the t − x curve is not a hyperbola, but we approx-
imate it by a hyperbola. This approximation is better for small offsets. Therefore, we replace the
multiple layers with one layer with an average velocity that will produce the closest t − x curve to a
hyperbola. This average velocity is called the stacking or NMO velocity. For a dipping layer, or a set
of dipping layers, the NMO velocity to any dipping reflector increases with its dip angle. However,
for small dip angles (< 15), the effect of dip on the NMO velocity is negligible. Therefore, for mul-
tiple horizontal and gently dipping layers (as the case of our data), we use equation 6.4 for NMO
correction. For highly dipping layers, we must correct for the effect of dip on velocity. We do this by
first Dip Moveout (DMO) processing then we perform the NMO correction. We recommend the
reader to look into [2] for more DMO details.
6.4. NORMAL MOVEOUT (NMO) CORRECTION 49
(a) (b) (c)
Figure 6.6: (a) CMP gather 220 of our sorted seismic before NMO correction. (b) Its semblance and
(c) after correction.
6.4.0.2 NMO Stretching

The NMO correction maps the data in a CMP gather from the x − t (x) plane to the x − tN MO (x)
plane using equation 6.4. However, it does not shift the traces up in time. Because both the velocity
and t0 increase downward, the NMO correction applied (using equation 6.3) for a later t0 will be
smaller than that applied at an earlier t0 . Therefore, two points on the same event with a time
separation equal to
tb before NMO correction will have a separation of
ta >
tb after NMO
correction because the upper point will experience larger NMO correction than the lower one. We
call this NMO stretching and it causes a frequency distortion to lower frequencies. Stretching is
quantified as:

tN MO (x) 1 x
SN MO (x) = ≈ ( )2 . (6.5)
t0 2 t0 v
NMO stretching is mainly confined to large offsets and shallow times as we can see from equation
6.5. Stacking NMO-corrected and stretched traces will severely damage the shallow seismic events.
Therefore, we have to remove the stretching before we stack. We do this by muting the considerably
stretched zones from the gather. The muted zone is usually set to a threshold in terms of SN MO so
that zones with SNMO greater than the threshold will be muted. Typical values for mute threshold
are between 0.5–1.50. Selecting a small threshold value avoids stretching but might cause excessive
(a) (b) (c)
loss of data, while selecting a large threshold value avoids loss of data but might introduce excessive
stretching of the data.
The following MATLAB script uses the M-function called nmo_correction.m to perform
NMO correction on seismic data CMP gathers based on the estimated stacking velocities from the
previous section:
1 load B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d . m a t
2 load B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d _ v e l o c i t i e s . m a t
3 m a x _ s t r e t c h =10;
4 [ Dsort , H s o r t ]= n m o _ c o r r e c t i o n ( Dsort , Hsort , v_stack , t_stack , ...
5 cmp_start , cmp_end , cmp_step , m a x _ s t r e t c h ) ;
6 save B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d _ n m o _ c o r r e c t e d
7 Dsort Hsort
Figures 6.6(c), 6.7(c), and 6.8(c) show various NMO-corrected CMP gathers. We can notice that
the hyperbolic seismic events have become more flat for different layers. Now, we are ready to obtain
our first approximation of the subsurface image using stacking. Note that the maximum stretch value
6.5. STACKING 51
(a) (b) (c)
we used in our NMO M-function is equal to 50. A very good exercise for the user is to vary the
maximum stretch value and see how it will affect the results.
6.5 STACKING
The purpose of stacking is to enhance the SNR ratio by eliminating coherent and incoherent noise
in the data and to reveal a first subsurface image approximation. The traces in the NMO-corrected
CMP gather are stacked (summed up) to produce one stacked trace that represents that CMP.
The amplitude of the stacked trace can be the sum or average of the amplitudes of traces √ in the
CMP gather. As we mentioned before, stacking M traces enhances the SNR ratio by M. The
stacked section is displayed with the CMP number along the horizontal direction, and t0 along the
vertical direction as we see in Figure 6.9 based on the following script which stack the data using
the M-function sstack.m:
1 load B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d _ n m o _ c o r r e c t e d
2 [ Dstacked ,t , c m p _ n u m ]= s s t a c k ( Dsort , H s o r t ) ;
3 save
4 Book_Seismic_Data_gain_bpf_sdecon_gain_sorted_nmo_corrected_stacked
(a) (b) (c)
Figure 6.9: The stacked section of our seismic data in various displays: (a) variable area-wiggle, (b)
variable density in gray and (c) variable density in colors. Recall that the color bars in (b) and (c) refers
to the amplitude dynamic range of the data.
5 cmp_num t Dstacked
Clearly, we can see the continuity of "geological" layers but yet this stacked section requires us to
improve its horizontal (spatial) resolution. We are going to perform one more seismic data processing
step, namely, migration, to do that in the next chapter.

1. Display using the wiggle variable area mean the 18 shot gathers and the 65 CMP gathers, one
at a time and analyze this carefully with Figure 6.3. You can use the MATLAB capabilities to
make a movie for each.
2. Select CMP gather number 250 along with its stacking time and velocities and then:
• Add 1500 ft/s to the velocity values and perform NMO correction.
• Subtract 1500 ft/s to the velocity values and perform NMO correction.
What can you realize from both NMO corrected CMP gathers? Explain.
6.7. USEFUL MATLAB CODES 53
6.7 USEFUL MATLAB CODES
ssort.m, extracting_cmp.m, extracting_cmp_fold_num.m, vel_picking.m,
nmo_correction.m, and sstack.m.
55
CHAPTER 7
Static Corrections
Static corrections are applied to seismic data in order to compensate for various effects on the data
such as those related to near surface, variations in elevations, weathering and reference to a datum [2].
By applying static corrections, we aim to determine the reflection time assuming the existence of a
flat surface plane with no weathering layer. These corrections include:
• Elevation static correction, which accounts for variable elevations of the sources and receivers.
• Residual static correction, which accounts for lateral variations in the velocity and thickness
of the weathering layer (WL).
The WL is the shallowest low-velocity layer. It is composed of unconsolidated and loosely consoli-
dated sediments. Common WLs in arid areas include: sand dunes, sabkhas, gravel plains, karsts, and
valley fills. The WL velocity is usually much less than those of the underlying sub-WL (bedrock)
and deeper layers. Therefore, the WL produces a large contribution to the overall traveltime of rays.
7.1 ELEVATION STATIC CORRECTION

Elevation static correction involves the computation and removal of the effect of different source and
receiver elevations. It is also known as field, datum, and topographic correction. It involves bringing
the sources and receivers to a common datum, usually below the elevation of the lowest source or
receiver. For this, we need a replacement velocity (Vr ) for the material between the datum and the
source or receiver. The replacement velocity is either assumed from prior knowledge of the area or
can be estimated from uphole times or direct arrivals. The elevation static correction (TD ) is given
by [2]:
(ES − ZS − ED ) + (ER − ZR − ED )
TD = , (7.1)
Vr
where ES is the ground elevation at shot location (from mean sea level), ZS is the depth of shot (and
is equal to 0 for a surface source), ER is the ground elevation at receiver location (from mean sea
level), ZR is the depth of receiver (and is equal to 0 for a surface geophone), and ED is the datum
elevation (from mean sea level). TD is always subtracted from the two-way traveltime of the trace
belonging to that particular source-receiver pair.
7.2 RESIDUAL STATIC CORRECTION

After the elevation static correction, it is important to correct for the effect of variable thickness and
lateral velocity variation of the WL. The main methods used to correct for these effects are [2]:
56 7. STATIC CORRECTIONS
1. uphole surveys
2. refraction statics
3. surface-consistent statics
7.2.1 UPHOLE SURVEYS

In this method, a fairly shallow hole (100-200 m) that penetrates the WL and the upper part of the
sub-WL is drilled for this purpose. Several geophones are placed at various (known) depths in the
hole. The geophone locations must span the weathering and sub-WLs. A shot is fired at the surface
near the hole and the direct traveltimes to the geophones are recorded. A plot of the direct traveltimes
versus the geophone depths can be used to compute the velocities of the weathering and sub-WLs as
well as the thickness of the WL at the uphole location. This method attempts to construct a model
of the WL by estimating the velocity and thickness of the WL at several locations and interpolating
between these locations. This method has the advantage of providing highly accurate near-surface
velocities and thicknesses. However, it is very costly if high lateral resolution is required. Therefore,
this method is used in estimating long-wavelength statics. Wavelength of statics refers to the width
of the lateral (velocity or thickness) change in the WL relative to the spread length (maximum
offset).
7.2.2 REFRACTION STATICS

This method is especially effective in estimating long-wavelength statics and is used to construct a
model of the WL by estimating the velocity and thickness of the WL. The following are standard
methods used for refraction statics calculation:
1. Delay-time Method: It uses the slopes of the direct and head waves as well as the head wave’s
intercept time of many shot records along the profile to estimate the WL velocity and thickness
under each receiver.This method requires picking first breaks, which is difficult. It also requires
reversed raypath1 geometries, which might not be available.
2. Generalized reciprocal Method (GRM): It uses reversed refraction profiles to estimate the
optimum WL thickness under each receiver. This method requires picking first breaks, which
is difficult. It also requires reversed raypath geometries, which might not be available.
3. Least-squares Method: It uses least square analysis to find the best-fit WL velocity-thickness
model to the first arrivals (i.e., direct and head waves). This method employs similar concepts
to those used for the surface-consistent method (that we will discuss in the coming subsection),
but uses refraction rather than reflection data.
1 A raypath is a line that is perpendicular to the wavefront in isotropic media.
7.2. RESIDUAL STATIC CORRECTION 57
7.2.3 SURFACE-CONSISTENT RESIDUAL STATIC CORRECTION METHOD
This method is especially effective in estimating short-wavelength statics. The basic assumption
of this method is that the static shifts are time delays that depend only on the source and
receiver locations on the surface, not on raypaths in the subsurface. This assumption is valid
only if all raypaths, regardless of source-receiver offset, are vertical in the near surface. The
surface-consistent assumption is generally good because the WL usually has a much lower ve-
locity than the sub-WL and refraction towards the normal at its base tends to make raypaths vertical.
The total residual static time shift on any trace can be expressed as:
Tij k = Ri + Sj + Gk + Mk Xij
2
, (7.2)
where Ri is the residual static time shift associated with the i th receiver position (i = 1, . . . , I , where
I is the number of receivers used in the survey), Sj is the residual static time shift associated with
the j th source position (j = 1, . . . , J , where J is the number of sources used in the survey), Gk is
the difference in two-way traveltime (due to structure) at a reference CMP and the traveltime at the
k th CMP (k = 1, . . . , K, where K is number of CMPs covered in the survey), and Mk Xij 2 is the
residual NMO associated with the trace generated by the j th source and recorded by the i th receiver
and it accounts for possible imperfect NMO correction due to using imperfect NMO velocities for
the k th CMP. Generally, we have more equations than unknowns for typical seismic surveys. This is
a typical least-squares problem. Our objective is to find those Ri , Sj , Gk , and Mk that will minimize
the error energy between the observed and calculated Tij k using model parameters in Equation 7.2:

I
J
K
E= [Ri + Sj + Gk + Mk Xij
2
− Tij k ]2 . (7.3)
i=1 j =1 k=1
7.2.3.1 Surface-Consistent Residual Static Correction in Practice

The most widely used method is the pilot trace method [2], which consists of the following steps.
1. A CMP with good SNR ratio is gained and NMO-corrected using a preliminary velocity
function.
2. The CMP gather is stacked to produce the first pilot trace.
3. Each individual trace in this CMP gather is crosscorrelated with the first pilot trace.

4. Time shifts Tij k , which correspond to maximum crosscorrelations, are picked.

5. Shift each original trace by its corresponding time shift Tij k .
6. A second pilot trace is constructed by stacking the shifted traces in the gather.
58 7. STATIC CORRECTIONS
7. The second pilot trace is, in turn, crosscorrelated with the original traces in the gather and
new time shifts Tij k are computed.
8. Shift each original trace by its corresponding new time shift Tij k .
9. The process is performed this way on all CMP gathers moving to left and/or right from the
starting (reference) CMP gather.
The following parameters are important when picking the time shifts in practice:
(a) Maximum allowable shift: which is the maximum shift allowed for crosscorrelations where a
value between 30 and 40 ms is reasonable.
(b) Correlation window: where it should be chosen in an interval with the highest possible SNR
ratio.
The following MATLAB script uses the M-function scr_static.m that implements the above
steps and apply them to our CMP NMO corrected data, followed by stacking those statically
corrected CMP gathers:
1 cmp_start =205;
2 cmp_end =255;
3 lags =20;
4 D s o r t _ s t a t i c = s c r _ s t a t i c ( Dsort , Hsort , cmp_start , cmp_end , lags ) ;
5 [ D s t a c k e d _ s t a t i c , t , c m p _ n u m ]= s s t a c k ( D s o r t _ s t a t i c , H s o r t ) ;
6 save
7 B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d _ n m o _ c o r r e c t e d _ s t a t i c ...
8 Dsort Dsort_static Hsort Dstacked_static cmp_num t
Note that we selected 40 ms (equivalent to 20 samples) for our data set as seen in the above MATLAB
script. Also, we selected the whole trace (0-3s), which is the default in our code, for our data set.
Figure 7.1 shows both the stacked data before and after applying the surface-consistent residual
static correction method. Clearly, the data quality has improved after applying the correction where
we notice the extension of the continuity to many of the layers.

scr_static.m
(a)
(b)
Figure 7.1: The stacked section of the east Texas seismic data set: (a) before applying residual static
correction and (b) after applying residual static correction.
61
CHAPTER 8
Seismic Migration
8.1 INTRODUCTION
After performing the necessary processing on the data and obtaining the stacked section, it can be very
misleading to think of such data as a true image of the subsurface.This is because the actual reflection
points are unknown. We need to take wave propagation effects into account to correctly determine the
reflection points of the subsurface structure [26]. Improper correction of such undesirable geometrical
effects leads to false interpretation and, therefore, oil/gas wells may be damaged or even misplaced.
This is done using a process known as Seismic Migration. Hence, we can define migration as the
process of reconstructing a seismic section so that the reflection events are repositioned under their
correct surface locations at their correct vertical reflection time or depth location [1, 2]. Basically,
migration removes the distorting effects of dipping reflectors from seismic sections. It also removes
the diffracted arrivals resulting from sharp lateral discontinuities [27, 28, 29, 30]. There exist various
migration techniques which are dependent upon the type of migration that suites our data or that
we want to perform. In this chapter, we are going to perform post-stack time migration since our
data mainly contains flat to gently-dipping layers as seen in its stacked section (Figure 7.1).
8.2 HUYGENS’ PRINCIPLE AND BASIC MIGRATION

PRINCIPLES
For this section, we will describe the physical principles of migration and we will follow [2, 31, 32].
Huygens’ principle is the basis of migration. Consider the harbor example shown in Figure 8.1.
Let us assume that a storm barrier exists at some distance z3 from the beach and that there is a gap
in the barrier. Now, also imagine that a calm afternoon breeze that comes from the ocean causes
a plane incident water wave to hit the barrier. Its wavefront is parallel to the storm barrier. As we
walk along the beach line, we notice a different wavefront from the incoming plane wave. The gap
on the storm barrier has acted as a secondary source and generated the semicircular wavefront that
is propagating towards the beach. Now, assume that we did not know about the storm barrier and
the gap. We may want to lay out a receiver cable along the beach to record in time the approaching
waves. Figure 8.2 illustrates this idea of recording in time the approaching waves with semicircular
wavefronts. So the gap in the storm barrier acts as a Huygens’ secondary source.
We can apply this principle to reflection seismology by imagining that each point on a reflector
(geological interface) generates a secondary source in response to the incident wavefield. This is
known as the exploding reflector model [31]. Consider a single point scatterer in a medium as shown
62 8. SEISMIC MIGRATION
Figure 8.1: The beach example for illustrating Huygens’ principle. The ocean causes a plane wave to hit
the storm barrier where a different wavefront moving towards the beach is noticed after the storm barrier
due to the barrier gap (modified after [2]).
Figure 8.2: This figure shows how the approaching waves with semi-circular wavefronts generated by
Huygens’ secondary source from the beach example (Figure 8.1) are recorded as hyperbolas in time.
in Figure 8.3(a). The minimum travel time is given by:

2z
t0 = , (8.1)
c
where z is the depth of the scatter and c is the velocity of the propagating wave (which we assume
to be constant). Also, assume that the source and the receiver are co-located (zero-offset) as in
8.2. HUYGENS’ PRINCIPLE AND BASIC MIGRATION PRINCIPLES 63
(a)
(b)
Figure 8.3: (a) A point scatterer (acting as Huygens’ secondary source), and (b) a curved reflector which
is produced based on the point scatterer.
Figure 8.3(a). The travel time as a function of distance, x, is given by [32]:

√
2 x 2 + z2
t (x) = . (8.2)
c
By squaring, rearranging, and using Equation (8.1) then, Equation (8.2) can be expressed as:
t (x)2 4x 2
− = 1. (8.3)
t02 c2 t02
This shows us that the travel time curve for the scattered arrival has the form of a hyperbola with
the apex directly above the scattering point (our secondary source) as seen in Figure 8.3(b).
Now, consider a horizontal reflector (Figure 8.4(a)) that is composed of a series of point
scatterers (gaps), each one of which generates a diffraction hyperbola in a zero-offset section as
in Figure 8.4(b). Following Huygens’ principle, these hyperbolas sum coherently only at the time
of the reflection while the later contributions cancel out (Figure 8.4(b)). However, if the reflector
vanishes at some point, then there will be a diffracted arrival from the endpoint that will show up in
zero-offset data. This creates an artifact in the section that might be falsely interpreted as structure.
We need to migrate such sections in order to remove such artifacts.
Another principle goal of migration is to map the apparent dip that is seen on the zero-offset
sections into true dip [2, 26, 31, 33]. True dip angle is always greater than apparent dip angle.
Consider a reflector dipping at an angle of θ in the true earth as in Figure 8.5. The zero-offset travel
time for a wavefield propagating from distance x down to the reflector and back up again is given
by t = 2r/c where r is the wavefield path length and is equal to r = x sin θ. Now, to compare the
apparent and true dip angles, the travel time must be converted to depth via Equation (8.1) and,
therefore, in the unmigrated depth section z = x sin θ. By definition, the slope of this event is the
tangent of the apparent dip angle, say β. Therefore, we have:
tan β = sin θ. (8.4)
Equation (8.4) shows clearly that the apparent dip angle is always less than true dip angle. Again in
Figure 8.5 the events associated with the two zero-offset wavefields drawn from the dipping reflector
to the two receivers will appear on the unmigrated section at the position locations associated with
the two receivers [2]. Therefore, migration moves the energy up dip. In addition, from the same
figure, the length of the reflector in the geological section is shorter than in the time section. Thus,
migration also shortens reflectors. In summary, migration focuses energy by collapsing diffractions
as well as it correctly steepens, shortens, and moves reflectors up-dip.
8.3 MIGRATION KINDS

Migration can be implemented based on different techniques. It is an important and expensive
process that is applied to reflection seismic data before it is interpreted. Since it is the last major
process applied to the data, it is likely to be blamed for all sorts of things like inconsistent
amplitudes and lack of structural details even though these problems may arise from acquisi-
tion or earlier processing steps [34]. Therefore, it is important to know what type of migration to use.
Migration can be classified as Pre-stack migration or Post-stack migration. For the former,
migration is performed on pre-stacked data either on shot gathers or on CMPs. For large surveys,
it will require massive computer storage and days, maybe even weeks, of CPU time on a super
computer [34]. For the latter case, the migration is applied on the stacked CMP data. It is much less
expensive than pre-stack migration but it is also less accurate in complicated areas of the subsurface.
Finally, it can be accomplished on workstation class machines.
8.3. MIGRATION KINDS 65
Model =HURïRIIVHWVHFWLRQ

3
Depth (m)
Time (s)

5

7

ï ï ï ï ï ï ï ï ï ï
Lateral position (m) Lateral position (m)
(a) (b)
Figure 8.4: (a) An array of point scatterers positioned at different locations, and (b) its resultant curved
reflectors interfering with each other.
Figure 8.5: Migration principles. The apparent reflector with a dip angle β in the time section when
migrated is moved up-dip, steepened (to an angle θ), shortened, and mapped on its true subsurface
location.
Moreover, migration can be classified in terms of how much physics we put into the algorithm,
specifically for handling velocity variations. This type of classification is based on time migration
versus depth migration (see Figure 8.6). Note that any migration (pre-stack or post-stack) can be
output in time or depth. In areas of strong velocity variations, depth migration is used and the output
is given as a depth section. Geological examples of strong lateral velocity variations include salt over-
hangs, sub-salt areas, or combinations of such features [34]. The main difference between time and
depth migration is mostly for ease of interpretation afterwards where one can make more simplifica-
tions in time migration than for depth migration. Time migration is valid only when lateral velocity
variations are mild (10%) to moderate (30%). When this assumption fails, we have to use depth
migration. Figure 8.7 illustrates schematically were pre/post stack time/depth migration is employed.
Finally, we also can classify migration into two-dimensional (2-D) and three-dimensional
(3-D). In 2-D migration, we migrate the data once along the profile. This might generate miss-ties
on intersecting profiles. In addition, 2-D migration is prone to sideswipe effects. Sideswipes are
reflections from out of the plane of the profile. In contrast, 3-D migration first migrates the data in
the inline direction then takes that migrated data and migrates it again in the crossline direction.
This is the two-pass 3-D migration. A one-pass 3-D migration can also be done using a downward
continuation approach. Therefore, considering 2-D versus 3-D, pre-stack versus post-stack, and
time versus depth, we can have the following types of migrations (ordered from fastest but least
accurate to slowest but most accurate) [35]:
1. 2-D poststack time migration (fastest, least-accurate)
2. 2-D poststack depth migration
3. 2-D prestack time migration
4. 2-D prestack depth migration
5. 3-D poststack time migration
6. 3-D poststack depth migration
7. 3-D prestack time migration
8. 3-D prestack depth migration (slowest, most accurate)
Since our data set is 2-D and contains mainly flat layers, we, therefore, choose to perform post-stack
time migration.
8.4 MIGRATION ALGORITHMS

There exist many migration algorithms but mostly belong to the following three main post-stack
migration algorithms:
1. Kirchhoff migration.
8.4. MIGRATION ALGORITHMS 67
Figure 8.6: This is another migration classification which depends on how much physics one puts into
the migration algorithm.
2. Frequency-wavenumber-(f − k) migration.
3. Finite-difference (downward continuation) migration.
8.4.1 POST-STACK TIME MIGRATION USING THE STOLT f − k

MIGRATION
The zero-offset stack section requires moving the reflectors to their true positions and we can obtain
this via post-stack time migration. We select to use the popular method of Stolt [2, 36] to do so. This
method basically depends on one forward and one backward 2-D Fast Fourier Transform (FFT)
computations as will as one mapping and a multiplication. Assuming that our stacked data is denoted
by u(t, x) (where t represents time and x is the distance), the algorithm works as follows [2]:
1. Take the 2-D Fourier transform of u(t, x) to obtain U (ω, kx ), where ω is the angular frequency
and kx is spatial wavenumber.
Figure 8.7: Migration types as a function of computational complexity and lateral velocity variations
(after [34]). The most accurate and most expensive migration kind is the pre-stack depth migration.
2. Apply the following mapping to the ω variable to become the depth wavenumber kz where
we map U (ω, kx ) to become U (kz , kx ):

2ω 2
kz = ( ) − kx2 .
v
3. Then we compute a scale value:
v kz
S=
2 k2 + k2
z x
and multiply it by U (kz , kx ).
4. Finally, take the 2-D inverse Fourier transform of the result and this will be our migrated
section.
The following script uses the M-function stolt_mig.m to perform post-stack time migration based
on Stolt’s f − k migration algorithm [36]:
1
2 load
3 B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d _ n m o _ c o r r e c t e d _ s t a t i c ...
4 load B o o k _ S e i s m i c _ D a t a _ g a i n _ b p f _ s d e c o n _ g a i n _ s o r t e d _ v e l o c i t i e s . m a t
5
6 dx = 1 1 0 ;
7 dt =0 .002 ;
8 v = mean ( mean ( v _ s t a c k ) ) ;
9 D s t a c k e d _ s t a t i c = D s t a c k e d _ s t a t i c (: ,1:60) ;
10 cmp_num = cmp_num (1:60)
11
12 D m i g r a t e d = s t o l t _ m i g ( Dstacked ,v , dt , dx ) ;
13
14 figure , s i m a g e _ d i s p l a y ( D s t a c k e d _ s t a t i c , cmp_num , t ,0) ;
15 x l a b e l ( ' CMP ' ,' F o n t S i z e ' ,14)
17 t i t l e ( ' S t a c k e d s e c t i o n ' , ' F o n t S i z e ' ,14)
18
19 figure , s i m a g e _ d i s p l a y ( Dmigrated , cmp_num , t ,0) ;
20 x l a b e l ( ' CMP ' ,' F o n t S i z e ' ,14)
22 t i t l e ( ' M i g r a t e d s e c t i o n ' , ' F o n t S i z e ' ,14)
Note that when migrating the data we have used the first 60 traces (out of 65) of our stacked data
since trace 61 is equal to zero (refer to Figure 7.1. Figure 8.8 shows the comparison between the
stacked and post-stack time migrated section based on Stolt’s f − k algorithm. The horizontal
resolution was improved after we applied the post-stack time migration algorithm to our stacked
section as we can see, for example, at time equal to 2, 2.5 and 2.75 s. That is, we have improved
the continuity of the layers. Of course, the migration result will be improved if we apply post-stack
depth migration but this requires estimating the velocity model for this geological section and then
performing the post-stack depth migration in the f − k domain as described in [2].

stolt_mig.m.
(a)
(b)
Figure 8.8: The east Texas seismic data set: (a) its stacked section and (b) its post-stack migrated section
using Stolt’s f − k migration algorithm.
71
CHAPTER 9
Concluding Remarks
In this book, we introduced the basic concepts encountered in a typical seismic data processing flow.
The concepts were then applied on a real 2-D data set from east Texas, USA. The processes applied
to the data set included:
1. Obtaining geometry information
2. Quality control of the data
3. Ground-roll attenuation using band-pass frequency filtering
4. Spiking deconvolution using Wiener optimum deconvolution
5. CMP sorting
6. Velocity analysis using the velocity spectrum method
7. NMO correction and stretch muting
8. Residual static correction using the surface-consistent method
9. Stacking
10. Migration using the f − k (Stolt) migration
MATLAB codes were used to perform each of these processes. Most of these codes were written
by the authors while few others were borrowed from public-domain resources. We made genuine
attempts to make these codes as simple as possible with plenty of comments to encourage interested
readers to modify them to fit their own data sets. Nevertheless, we greatly appreciate feedback from
our readers.
73
Bibliography
[1] P. Kearey, M. Brooks, and I. Hill. An Introduction to Geophysical Exploration. Blackwell Science,
3rd edition, 2002. Cited on page(s) 1, 2, 24, 61
[2] Öz. Yilmaz, editor. Seismic Data Analysis: Processing, Inversion, and Interpretation of Seismic
Data. Society of Exploration Geophysicists, 2nd edition, 2001. Cited on page(s) 1, 4, 7, 17,
23, 24, 30, 34, 35, 39, 48, 55, 57, 61, 62, 64, 67, 69
[3] A. S. Spanias, S. B. Jonsson, and S. D. Stearns. Transform methods for seismic data compression.
IEEE Tran. on Geosciences and Remote Sensing, 29(3):407–416, May 1991.
DOI: 10.1109/36.79431 Cited on page(s) 1, 23
[4] B. Buttkus. Spectral Analysis and Filter Theory in Applied Geophysics. Springer, 2000. Cited on
page(s) 1, 24, 33
[5] D. Forel, T. Benz, and W. D. Pennington. Seismic Data Processing with Seismic Un*x: A 2-D
Seismic Data Processing Primer. Society of Exploration Geophysicists (SEG), 2005. Cited on
page(s) 7
[6] J. F. Clearbout. Imaging the Earth’s interior. Blackwell Scientific Publications, 1985. Cited on
page(s) 20
[7] X. Miao and S. Cheadle. Noise attenuation with wavelet transform. SEG 1998 Expanded
Abstracts, 1998. DOI: 10.1190/1.1820071 Cited on page(s) 23
[8] T. J. Ulrych, M. D. Sacchi, and J. M Graul. Signal and noise separation: Art and science.
Geophysics, 64:1648 – 1656, Sept.- Oct. 1999. DOI: 10.1190/1.1444670 Cited on page(s) 23
[9] A. Özbek. Adaptive beamforming with generalized linear constrains. Geophysics Extended
Abstracts, 2000. DOI: 10.1190/1.1815855 Cited on page(s) 23
[10] L. Duval and T. Rosten. Filter bank decomposition of seismic data with application to com-
pression and denoising. SEG 2000 Expanded Abstracts, 2000. DOI: 10.1190/1.1815847 Cited
on page(s) 23
[11] Qiansheng Cheng, Rong Chen, and Ta-Hsin Li. Simultaneous wavelet estimation and de-
convolution of reflection seismic signals. IEEE Transactions on Geoscience and Remote Sensing,
34(2):377 – 384, March 2001. DOI: 10.1109/36.485115 Cited on page(s) 23
74 9. CONCLUDING REMARKS
[12] K. Berkner and Jr. Wells, R.O. Wavelet transforms and denoising algorithms. Conference Record
of the Thirty-Second Asilomar Conference on Signals, Systems and Computers, 2:1639 – 1643, Nov.
1998. Cited on page(s) 23
[13] Rongfeng Zhang and Tadeusz J. Ulrych. Physical wavelet frame denoising. Geophysics,
68(1):225–231, Jan 2003. DOI: 10.1190/1.1543209 Cited on page(s) 23
[14] J. E. Womack and J. R. Cruz. Seismic data filtering using a gabor representation. IEEE Tran. on
Geosciences and Remote Sensing, 32(2):467–472, March 1994. DOI: 10.1109/36.295061 Cited
on page(s) 24
[15] A. F. Linville and R. A. Meek. A procedure for optimally removing localized coherent noise.
Geophysics, 60(1):191 – 203, Jan.-Feb. 1995. DOI: 10.1190/1.1443746 Cited on page(s) 24
[16] B. Duquet and K. J. Marfurt. Filtering coherent noise during prestack depth migration.
Geophysics, 64(4):1054 – 1066, July-Aug. 1999. DOI: 10.1190/1.1444613 Cited on page(s) 24
[17] S. Treitel, J. L. Shanks, and C. W. Fraster. Some aspects of fan filtering. Geophysics, XXXII:789
– 800, 1967. DOI: 10.1190/1.1439889 Cited on page(s) 24
[18] M. Z. Mulk, K. Obata, and K. Hirano. Design of fan filters. IEEE Trans. on Acoustics, Speech,
and Signal Processing, 31(6):1427 – 1434, Dec. 1983. Cited on page(s) 24
[19] A. H. Kayran and R. A. King. Design of recursive and nonrecursive fan filters with com-
plex transformation. IEEE Trans. on Circuits and Systems, 30(12):849 – 857, Dec. 1983.
DOI: 10.1109/TCS.1983.1085321 Cited on page(s) 24
[20] D. W. McCowan, P. L. Stoffa, and J. B. Diebold. Fan filters for data with variable spatial
sampling. IEEE Trans. on Acoustics, Speech, and Signal Processing, 32(6):1154 – 1159, Dec.
1984. DOI: 10.1109/TASSP.1984.1164466 Cited on page(s) 24
[21] R. Ansari. Efficient IIR and FIR fan filters. IEEE Trans. on Circuits and Systems, 34:941 –
945, August 1987. DOI: 10.1109/TCS.1987.1086224 Cited on page(s) 24
[22] R. H. Bamberger and M. J. T Smith. A filter bank for the directional decomposition of
images: Theory and design. IEEE Tran. on Signal Processing, 40(4):882–893, December 1992.
DOI: 10.1109/78.127960 Cited on page(s) 24
[23] E. A. Robinson and S. Treitel. Geophysical Signal Analysis. SEG, 2000. Cited on page(s) 33,
34
[24] E. A. Robinson and S. Treitel. Digital Imaging & Deconvolution: The ABCs of seismic exploration
and processing. SEG, 2008. Cited on page(s) 33
[25] N. Neidell and M. T. Tanner. Semblance and other co-herency measures for multichannel
data. Geophysics, 36:482–497, 1971. DOI: 10.1190/1.1440186 Cited on page(s) 47
75
[26] E. A. Robinson. Migration of geophysical data. Intr. Human Resources Development Corpo-
ration, 1983. Cited on page(s) 61, 64
[27] J. W. Thorbeck and A. J. Berkhout. 3-D recursive extrapolation operators: an overview. Geo-
physics Extended Abstracts, 1994. Cited on page(s) 61
[28] V. K. Madisetti and D. B. Williams, editors. The Digital Signal Processing Handbook. CRC
Press and IEEE Press, 1998. Cited on page(s) 61
[29] D. E. Dudgeon and R. M. Mersereau. Multidimensional Digital Signal Processing. Prentice-

Hall, 1984. Cited on page(s) 61
[30] L. J. Karam and J. H. McClellan. Efficient design of digital filters for 2-D and 3-D
depth migration. Signal Processing, IEEE Transactions on, 45(4):1036–1044, April 1997.
DOI: 10.1109/78.564191 Cited on page(s) 61
[31] J. F. Claerbout. Imaging the Earth’s Interior. Blackwell, 1984. Cited on page(s) 61, 64
[32] P. M. Shearer. Introduction to Seismology. Cambridge Uni. Press, 1999. Cited on page(s) 61, 63
[33] J. A. Scales. Theory of Seismic Imaging. Samizdat Press, 1997. Cited on page(s) 64
[34] C. L. Liner. Elements of 3-D Seismology. PennWell, 1999. Cited on page(s) 64, 66, 68
[35] B. L. Biondi. 3D Seismic Imaging. SEG, 2006. Cited on page(s) 66
[36] R. H. Stolt. Migration by Fourier transform. Geophysics, 43(1):23–48, 1978.

DOI: 10.1190/1.1440826 Cited on page(s) 67, 68
77
Authors’ Biographies
WAIL A. MOUSA
Wail A. Mousa holds two B.S. with honors in electrical engineering and mathematical sciences
and an M.S. in electrical engineering from KFUPM. Wail was the first Saudi to be sponsored by
Schlumberger Dhahran Carbonate Research (SDCR) for a Ph.D., which he obtained at the School
of Electronics & Electrical Engineering at the University of Leeds in 2006. He was a graduate
assistant at KFUPM and then promoted to lecturer between 2000-2003. Wail then worked between
2003-2009 as a research scientist on applied signal processing for geology and geophysics at the
SDCR. He is currently working as an assistant professor in the Electrical Engineering Department,
King Fahd University of Petroleum & Minerals (KFUPM), teaching various signal and image
processing courses and supervising students. Wail is an active researcher and publishes regularly
where his research interests include: digital signal, image, and video processing and their applications
in geophysics, geology as well as petro-physics, design & implementation of digital filters including
wavefield extrapolation filters, image segmentation, pattern recognition and classification and their
applications related to geophysical and geological data, signal and image compression. More recently,
he published his first patent in the US patent and trade office. He was additionally appointed as
the director of business development at KFUPM as of February 2010. He is a member of many
international and local professional organizations. In late 2006, he represented both Saudi Arabia in
the inaugural World Petroleum Council-Youth Committee (WPC-YC) and in early 2007, he became
the vice chair of the WPC-YC, where he has been chairman since June 2008. Dr. Mousa has received
many awards of special honor including a letter of distinction from the CEO of Schlumberger in
2008 due to his technical contributions to the company since he joined them after his PhD.
ABDULLATIF A. AL-SHUHAIL
Abdullatif A. Al-Shuhail is an associate professor of Geophysics at King Fahd University of
Petroleum & Minerals (KFUPM). He received his B.S. degree in Geophysics from KFUPM in
1987. He got his M.Sc. and Ph.D. in Geophysics from Texas A&M University, College Station.
He founded the Near Surface Seismic Investigation Consortium at KFUPM in 2006. His fields of
interests include seismic characterization of fractured reservoirs, near-surface effects on petroleum
seismic data, and ground penetrating radar. He is member of the Society of Exploration Geo-
physicists (SEG), European Association of Geoscientists and Engineers (EAGE), and Dhahran
Geoscience Society (DGS).
79
Index
absorption, 15, 17 display

AGC, 18, 38 variable area, 10
amplitude variable density, 10
correction, 15, 17 wiggle, 9
data dependent correction, 17 displaying seismic, 9
data independent correction, 17
losses, 15, 16 exploding reflector model, 61
autocorrelation, 33 exponential gain, 17
matrix, 32, 35
Fast Fourier Transform, 67
window, 35
FFT, 67
autocorrelograms, 36
filter
automatic gain control, 18
inverse, 33
BPF, 24, 26, 29 length, 35
filtering, 2
CMP, 2, 41, 47, 57 f − k, 24
coherency, 47 bandpass, 12
measure, 47 fan, 24
common depth point, 41 frequency, 23, 24, 29, 41
common mid-point, 2 linear, 23, 25
common reflection point, 41 Wiener, 32
constant velocity stack, 45 filters
convolution, 29 band pass, 24
crosscorrelation, 32 Finite Impulse Response, 26
FIR, 26
deconvolution, 3, 23, 26, 41 fold, 43
deterministic, 30 frequency, 23–25
parameters, 35 frequency-space, 24, 25
predication, 33 frequency-wavenumber, 24, 25
spiking, 12, 29, 30
statistical, 30 gain application, 4, 15
demultiplexing, 15 geometric divergence, 17
80 INDEX
geophone, 1, 16, 56 normal equations, 32
normal move-out, 2, 41
Huygens’ principle, 61
hyperbola, 46, 48 offset, 3, 8, 15, 48
hyperbolas, 44, 64 maximum, 56
minimum, 48
instrument, 23 non-zero, 47
zero, 3, 24, 46, 62, 64, 67
least-squares, 32, 56
overcorrected, 48
linear system, 29
power of time, 17
magnitude spectra, 25
prewhitening, 35
f − x, 26
primary reflections, 41
migration, 4, 13, 44, 61
f − k, 67 QC, 15, 16
2-D, 66 quality control, 4, 15, 41
3-D, 66
depth, 66 rain, 23
downward continuation, 67 Rayleigh waves, 24
Finite-difference, 67 raypath, 56, 57
Kirchhoff, 66 reflection
post-stack time, 13, 61 coefficient, 1
pre-stack, 64 coefficients, 29
Stolt, 67 seismology, 61
time, 65 time, 1, 55
muting, 5 reflectivity, 29
function, 1
near surface, 55 reformatting, 15
near-surface scatterers, 23 residual static correction, 12
NMO, 2, 5, 12, 41, 44, 47, 57 resolution
correction, 49 horizontal, 4
stretching, 49 vertical, 3, 29
noise, 2, 23
air waves, 24 seismic
coherent, 2, 3, 23, 24 convolution model, 2
ground roll, 2, 12, 24 convolutional model, 29, 30
incoherent, 2, 23 deconvolution, 29
multiples, 24 event, 10, 23
random, 3, 23, 29 header, 7
surface waves, 24 image, 2, 16
INDEX 81
reflection, 1 editing, 4, 15
shot gather, 24 transmission losses, 17
trace, 1, 9 traveltime, 1, 47
velocities, 44 two-way, 46, 48
seismogram, 1 traveltimes, 47, 56
seismograms, 29
shot gather, 41 undercorrected, 48
signal-to-noise ratio, 3, 23
SNR, 3, 23, 24, 26, 51, 57 velocity
sorting, 2, 12, 41 picking, 47
spherical divergence, 15, 44 velocity analysis, 2, 12, 44
stack section, 67 velocity spectrum, 45, 47
stacked section, 41, 51
wave propagation effects, 61
stacking, 2, 3, 13, 24, 41, 44, 51
wavefield, 61
velocities, 41, 45
static correction wavelet, 29, 30
elevation, 55 denoising, 23
residual, 55 length, 35
statics, 5, 15 minimum-phase, 30, 31
long-wavelength, 56 source, 1, 29, 30
short-wavelength, 57 wavenumber, 23, 25
subsurface image, 41 wind, 23
time-distance, 48 zero-lag
curves, 24 autocorrelation, 35
time-space, 44, 45 impulse function, 33
Toeplitz matrix, 33 spike, 33
trace zero-phase spike, 29

Reflection Processing Using Matlab PDF

Uploaded by

Reflection Processing Using Matlab PDF

Uploaded by

Series

PROCESSING OF SEISMIC REFLECTTION DATA USING MATLAB™

MOR GAN &CL AYPOOL

Processing of Seismic Reflection Data Using MATLAB™

Fixed-Point Signal Processing

Advanced Radar Detection Schemes Under Mismatched Signal Models

DSP for MATLAB™ and LabVIEW™ IV: LMS Adaptive Filtering

DSP for MATLAB™ and LabVIEW™ II: Discrete Frequency Transforms

DSP for MATLAB™ and LabVIEW™ I: Fundamentals of Discrete Signal Processing

The Theory of Linear Prediction

Nonlinear Source Separation

Spectral Analysis of Signals: The Missing Data Case

Processing of Seismic Reflection Data Using MATLAB™

ISBN: 9781608457922 paperback

A Publication in the Morgan & Claypool Publishers series

Wail A. Mousa and Abdullatif A. Al-Shuhail

SYNTHESIS LECTURES ON SIGNAL PROCESSING #10

To my dear wife and children, A.S.

1 Seismic Data Processing: A Quick Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Examination of A Real Seismic Data Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Quality Control of Real Seismic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Seismic Noise Attenuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6 Carrying the Processing Forward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

Wail A. Mousa and Abdullatif A. Al-Shuhail

Wail A. Mousa and Abdullatif A. Al-Shuhail

Seismic Data Processing: A

1.2 SEISMIC DATA PROCESSING

1. deconvolution, which increases the vertical resolution;.

2. stacking, which increases the SNR; and

Field Tapes (Amplitudes) Observer’s Log (Geometry)

Figure 1.3: A conventional seismic data processing flow (after [2]).

Examination of A Real Seismic

2.2 DESCRIPTION OF THE SEISMIC REFLECTION REAL

• Number of shots = 18.

• Source type = dynamite in 80-100-ft depth holes.

• Number of channels per shot = 33.

• Receiver type = Vertical-component geophones.

• Array type = 12-element inline.

• Number of traces in line = 594.

• Receiver interval = 220 ft.

• Shot interval is variable.

• Time sampling interval = 2 milliseconds (ms)

• Data format = SEG-Y.

• Byte swap type = Big endian.

• Data file name = data.sgy.

2.3 EXAMINING THE DATA SET

2.3.1 HEADER INFORMATION

1 load ( ' B o o k _ S e i s m i c _ D a t a . m a t ' , ' H ' )

Also, here is another example:

1 figure , plot ( sx , ' . ' )

1 num_shots = length ( shot_gathers );

2.3.2 DISPLAYING SEISMIC DATA

1. The wiggle display: plots seismic trace amplitudes as a function of time.

Number of traces/shot gather

Figure 2.2: The stacking chart plot of the seismic data.

(a) (b) (c)

2.4 THE PROPOSED PROCESSING WORKFLOW

(a) (b) (c)

9. Migration using Stolt (F-K) post-stack time migration

2.5 COMPUTER ASSIGNMENTS

Figure 2.5: Variable area display of seismic shot gathers 4-6.

2.6 USEFUL MATLAB FUNCTIONS

Quality Control of Real Seismic

3.2 TRACE EDITING

1 [i , j ]= find ( D == max ( max ( D ) ) ) ;

3.3 AMPLITUDE LOSSES & THEIR CORRECTION

fcorrected (t) = f (t)t α , (3.1)

fcorrected (t) = f (t)eβt , (3.2)

1. Calculate the RMS value in each gate.

• Instantaneous AGC: this is a bit different from the RMS AGC:

Envelope (dB) trace number :33

Before gain correction Before gain correction

1. Calculate the absolute mean value for in a given gate of length w.

3.4 COMPUTER ASSIGNMENTS

Envelope (dB) trace number :33

Before gain correction Before gain correction