Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.

org/
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

European Association of Geoscientists and Engineers

Co-sponsor of the DISC program at Eastern Hemisphere location

Corporate Sponsors
Supporting the DISC program through
directed gifts to the SEG Foundation
 Second Annual
1999 Distinguished Instructor Short Course
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Distinguished Instructor Series, No. 2

Society of Exploration Geophysicists

The Seismic Velocity Model

as an Interpretation Asset

presented by
Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Copyright ¹ 1998
Society of Exploration Geophysicists
Box 702740
Tulsa, OK USA 74170-2740
web site: www.seg.org
phone: 918-497-5500

ISBN ISBN 978-1-56080-086-6 (Series)

ISBN 978-1-56080-091-0 (Volume)

Published 1998
Reprinted 2009

All rights reserved. No part of this book may be reproduced,

stored in a retrieval system, or transcribed in any form or by any means,
electronic or mechanical, including photocopying and recording, without
prior written permission of the publisher.
 Phil Schultz

Table of Contents
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

About the Author........................................................................................................................ iv

SEG’s President Letter ............................................................................................................... v
The Seismic Velocity Model as an
Interpretation Asset – Course Description ................................................................. vi
Section 1. Course Introduction.............................................................................................. 1-1
Section 2. Creating Accurate Depth Maps ......................................................................... 2-1
Section 3. The Velocity Model-Building Process: A Tour ............................................ 3-1
Section 4. Clarifying the Image.............................................................................................. 4-1
Clarifying Complex Faulting ........................................................................................... 4-3
Enhancing Interpretability ............................................................................................... 4-11
Improving Volumetric Calculations.............................................................................. 4-21
Distinguishing a Salt Weld From a Salt Boundary................................................... 4-29
Improving the Structural Detail of a Subsalt Play .................................................... 4-37
Section 5. Seeing Below Gas Seepage................................................................................... 5-1
Detailing Structural Continuity ...................................................................................... 5-3
Enhancing the Estimate of Structural Closure Below Gas .................................... 5-11
Section 6. The Fault Shadow .................................................................................................. 6-1
Identifying a Possible Permeability Barrier as a False Image................................ 6-3
Section 7. Getting Well Placement Right ........................................................................... 7-1
Enhancing the Positioning of Complex Faulting ..................................................... 7-3
Clarifying the Termination of a Reservoir .................................................................. 7-19
Section 8. Evaluating Closure in Depth .............................................................................. 8-1
Uncovering Different Stories From Time and Depth Closure ............................. 8-3
Analyzing Conflicting Depth Conversions ................................................................. 8-15
Finding Closure in Depth, But None in Time........................................................... 8-27
Determining Depth Structure Under a Channel Fill ............................................... 8-29
Section 9. Getting Clues to the Lithology .......................................................................... 9-1
Investigating the Lithological Character
of the Faroe Shetland Escarpment ......................................................................... 9-3
Section 10. The Ultimate Lesson ........................................................................................... 10-1
Section 11. Glossary................................................................................................................... 11-1
Section 12. Acknowledgments ............................................................................................... 12-1

Distinguished Instructor Short Course s iii

 The Seismic Velocity Model as an Interpretation Asset

About the author

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Phil Schultz is a senior consulting geophysicist with Spirit Energy, a division of Unocal.
He was most recently Vice President for Development at Advanced Data Solutions in
Houston. He spent eleven years with Schlumberger, nine of which were overseas engi-
neering management assignments in Tokyo, London and finally Paris, where he headed
development of the company’s reservoir modeling workstation. Prior to Schlumberger,
he was a research geophysicist with Digicon in Houston, where he was involved in the
early development of depth migration technology and was the project manager for the
extension of the DISCO processing system to 3D.
An early member of the Stanford Exploration Project (SEP), directed by Jon
Claerbout, Dr. Schultz’s thesis on “slant stacks” has the distinction of having the last
single-digit SEP volume number. He is a recipient of the Outstanding Presentation
award of the SEG and has served as Associate Editor, Data Processing, for GEOPHYSICS.
Dr. Schultz currently resides in Houston with his wife, Sandy, and their three chil-
dren. He enjoys a good game of tennis.

iv s Society of Exploration Geophysicists

 Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Phil Schultz

Distinguished Instructor Short Course s v

 The Seismic Velocity Model as an Interpretation Asset

The Seismic Velocity Model as an Interpretation Asset

Course Description
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Purpose
A velocity model can have enduring and growing interpretive value, beyond its initial
creation as a means to optimize the seismic image. The 3D velocity model is often built
carefully with a combination of geophysical and geological input, because of the accura-
cy demands placed on it by the requirements of depth imaging. As such, this model
becomes an increasingly effective interpretive tool.
This course addresses the ways in which the interpreter should participate in the
development of the velocity model, and underscores its interpretive value with numer-
ous case study examples.

Who Should Attend?

The interpreter excited by the prospect of participating actively in the velocity model-
building process, and who wishes to pursue aggressively the additional advantages
offered by using the velocity model during interpretation, should attend this course.
Managers and supervisors seeking familiarity with velocity-related tools for developing
a geological model also will find this course useful.

Morning
After a brief introduction and preview of the day, the course begins with a comprehen-
sive case study in which interpretive input was key to the development of all phases of
the refined velocity model for depth imaging and depth conversion, and in which the
interpretation itself was refined in conjunction with the development of the velocity
model. Following this showcase study, and to complete the morning session, the course
embarks on a concise overview of the general model-building methodology. The types
of geological settings requiring varied approaches to velocity model building will be
reviewed, and model-building aspects normally requiring interpretive input will be
highlighted.

Afternoon
In the afternoon, the course focuses on over a dozen case studies supplied by major, inde-
pendent and national oil companies. Each of these will demonstrate some interpretation
or exploration problem that was solved by attention to the velocity model. The studies
follow the interpreter’s thought processes and approach to the problem. They run from
the simple to the complex, cover soft rock and hard rock environments, and touch on
imaging, depth conversion, fault location, well placement, lithology, anisotropy and other
velocity-related issues. All studies carry the common theme that the velocity model was a
key element in the development of the geological interpretation. Moreover, the interpreter
participated actively in the development of all the velocity models.

vi s Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Course Introduction
1
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Course Introduction
The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Seismic propagation velocities at one time were considered important only as parame-
ters for stacking seismic data or converting time structure maps to depth. Little else was
done with them. Since then, the importance of velocities has grown to the extent that
there is a new way to describe them—as a Velocity Model. When does velocity informa-
tion become a velocity model and why should we care?
Critically, the concept of the velocity model has risen in importance. Rather than
several applications (like stacking, migration, depth conversion, lithological estimation,
etc.), each having their own velocities or velocity-like parameterizations, there is an
increased tendency to unify the description of seismic velocities into a single 3D veloc-
ity model. This single model can be used, with some slight customizations, for applica-
tions such as stacking, time migration, depth migration, depth conversion, and litho-
logical or rock property estimation.
Processes like depth migration and depth conversion of 3D seismic data require a
geologically and geophysically competent velocity model. The constantly improving
quality of seismic data and increasing survey accuracy demand correspondingly high
quality and accuracy for velocity models. In depth conversion of time-based maps, for
example, use of a single regional velocity function is rapidly being replaced. Instead,
depth conversion of both the interpreted surfaces and the seismic data itself is taking
place by using a single, unified, high-quality 3D velocity model.
For 3D velocity models to achieve the required levels of accuracy, it is necessary to
use all available sources of velocity information. Seismic data provide information from
primary and residual moveout analysis on unmigrated and image gathers. Check shot
or vertical seismic profile (VSP) data give vertical traveltimes from the surface to depth
arrival time measurements. Well data provide depths to key formation boundaries from
drilling data or from well log correlation. Geological input quantifies and constrains
velocities for lithologically-based or depth-of-burial-based velocity environments.
How do we incorporate such diverse information into a unified 3D model? Where
do we start the process? Who builds the velocity model? The processor? The inter-
preter? Both? What role does each play? By the end of the day, the answers to these
questions will no longer be a mystery.
More important than how a velocity model is built, what value does a unified veloc-
ity model bring? How does it help the interpretation process? Can the velocity model
built for prestack depth migration be extended and used for interpretive processes, like
depth conversion?
This one-day course examines these questions and others in the context of case
studies. The course begins with an extensive case study in which the velocity model
was developed from the basic seismic analysis and geological inference stages through
depth imaging, and finally to depth conversion.

Distinguished Instructor Short Course ‡ 1-1

 The Seismic Velocity Model as an Interpretation Asset

What to Expect
This course focuses minimally on concepts of what the velocity model is and how it
should be built, and heavily on real-world examples. The examples relay how explora-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tion and production interpretation geophysicists solved a key problem by attention to

the velocity model. At the very least, this course will familiarize you with the manner in
which at least a dozen problems were solved by working geophysicists, like yourself.

Course Approach
Stated simply, we will minimize the exposition of concepts and focus on learning by
the examples of others. Accordingly, case histories will capture the vast majority of our
attention. Each case study will teach us perhaps one important thing, perhaps more
than one, depending on our individual interests/experience and on the case study itself.
We will look at over a dozen studies during the course of the day and our perspec-
tive on each will be, “What did attention to the velocity model bring to the interpreta-
tion process?” In order to help us focus on this aspect of each case study, they will be
described as “problems,” which in some way were solved by attention to the velocities.
Sometimes an extremely detailed and comprehensive velocity model was required, and
sometimes a very simple application of a general regional description of velocities was
all that was required. In all case studies, the interpreters in their exploration or produc-
tion departments solved some problem by attention to the velocity model. We will look
at the way they approached and solved it, make sure we understand what they did, and
hopefully ask ourselves penetrating questions about how we can solve some of the
problems on our desk right now by similar methods.
This course also takes the viewpoint that there is no one way to use velocities that is
always correct, to the exclusion of others. True, there are approaches to building the
velocity model that are sound, systematic and in general yield a high-quality product.
An example of such a process is the development of the seismic depth image through
iterative and model-based velocity updating. This process will be used in several of our
case studies. Where appropriate, it is a powerful tool to maximize what velocities can
bring.
In other case studies, different approaches are sometimes used to good effect, as
well. I take the view that attention to the velocities is the important issue, as opposed to
how the velocities are handled with regard to the methodology for velocity model-
building.

What Makes a 3D Velocity Model?

In this course we distinguish between a velocity model and a velocity field. 3D velocity
models incorporate structural and lithological information. They are more than just a
3D organized collection of velocity numbers, which one could argue is what a velocity
field is. Velocity models are usually constrained in such a way that velocity domains fol-
low defined geological layers. These layers can be defined to match structural infoma-
tion, or can be defined according to some other guideline such a depth of burial.

1-2 ‡ Society of Exploration Geophysicists

 Phil Schultz

The velocity model also can be described as a combination of the two types of con-
straints. There is no hard-and-fast differentiation between the terms in common use
today. As a “rule-of-thumb,” if you see geological structure of some kind in the velocity
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

field, you can call it a velocity model.

Hard Rock Environments

The layering in the velocity model normally follows the structural contours of impor-
tant velocity interfaces. Sometimes velocity models incorporate geological layer con-
straints that follow sedimentary or lithological boundaries. For example, Figs. 1 and 2
show a velocity model of an overthrust zone, in which the layers in the model corre-
sponding to the overthrust can readily be seen. The velocity contrasts that are important
are those that occur between the overthrust and the sedimentary basin. Defining the
layering in this manner predisposes the velocity model to accommodate this velocity
contrast. When the layering in the velocity model needs to follow geological structure,
as in a “hard rock” environment, lithology is the key driver of velocity.

Soft Rock Environments

On the other hand, sometimes the layers in the velocity model only remotely echo the
sedimentary layering, and follow more closely some other guideline for defining the
velocity behavior. This kind of behavior is expected in “soft rock” environments, whose
velocities are driven by compaction, rather than lithology, and accordingly express
themselves as a function of depth-of-burial. The Gulf of Mexico commonly exhibits this
characteristic.
An example of this type of velocity model is shown in Figs. 3 and 4. This is a model
built by BHP from copies of the water bottom structural surface. In this case, from the
Gulf of Mexico, it is known that the velocity is a stronger function of the depth-of-
burial than of the sedimentation. Accordingly, layers were defined anticipating iso-
velocity contours, as opposed to defining layering according to depositional geometry.
Fig. 5, which is a bit confusing to the eye, shows the actual sedimentary layering
superimposed on the velocity model. Also shown in color shading is the structure of
the interval velocities. The velocities in this example seem to behave somewhere
between following the rule of depth-of-burial and that of sedimentary layering. Where
depth-of-burial appears to reflect more accurately the iso-velocity trend than strict sedi-
mentary layering, it is an appropriate choice for the velocity model layering.
Notice also the salt body in this example, wherein the velocity is controlled by
lithology. Here we have a model with both soft rock and hard rock characteristics. The
salt is defined with precise model layering, and has been “inserted” (i.e., cut) into the
surrounding background, which itself had been defined according to depth-of-burial.

Velocity Variation Within Layering

Notice in the previous cases that the velocities are not required to be constant within
each layer. In soft rock environments, layering in the velocity model need only be a
guide to the variation of velocities within the model, and should follow closely the

Distinguished Instructor Short Course ‡ 1-3

 The Seismic Velocity Model as an Interpretation Asset

iso-velocity contours. It rarely, if ever, precisely does so, and so we should expect to
accommodate lateral velocity variations within the model’s layers. In hard rock environ-
ments, the lithology drives the velocity, but it is not the only factor. So again, we expect
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

and normally require velocity variation within layering in our velocity model.

The Role of and Benefit to the Interpreter

In the brave new world of seismic interpretation, the interpreter can no longer leave
processing solely to others. Particularly during depth imaging, velocity model-building
is critical. Increasingly, processing strives to enhance the most subtle features in the
data, and to do so requires a refined and geologically sensible velocity model. The previ-
ous examples illustrate that velocities can be driven by lithology or depth-of-burial.
Velocities also can be driven by age, tectonics and other factors. The interpreter will
take the lead in defining the velocity-controlling factors in each case and ensuring that
the model is built and refined within general guidelines. In addition to directing the
structure of the velocity model, the interpreter will be involved in controlling the actual
values of velocity imposed on the model.
Why should the interpreter take time to attend to the building of the velocity
model? Mainly for self-interest. A competent and thoroughly refined velocity model,
which adheres to the general geological characteristics of the area, will prove to be an
asset of enduring value to the interpreter. Velocities are needed for imaging, to be sure,
but because so much effort has gone into developing a velocity model suitable for depth
imaging (which requires a detailed interval velocity model), subsequent depth conver-
sion of the seismic image using a well-calibrated and anisotropy-corrected velocity
model is normally a few easy steps away, given a detailed imaging velocity model. As
more wells arrive, more detail can be added to the model and the seismic data recon-
verted to depth. Through the entire procedure, the velocity increases in value to the
interpreter because it delivers an increasingly accurate depth picture.
When traditional processing parameters are selected and used in seismic processing,
such as deconvolution gaps, filter bandwith parameters, and others, they can virtually
be forgotten after they have been applied, except to archive the processing sequence for
future reference. They are of little or no value in the interpretation process, except to
identify errors in the selections of the parameters that may have caused a less than opti-
mal image or created artifacts in the data. However, in a direct sense, they are useless
for interpretation.
Now, contrast this situation to the velocity model that was used for depth imaging.
Here is a velocity model that was built as accurately as possible for imaging and, with
small corrections, can be used as an excellent tool to render the seismic data in depth.
The corrections relate to anisotropy and other factors that make the imaging velocity
differ from the depth-conversion velocity. (These factors will be discussed in the context
of several of our examples that required anisotropy corrections.)
Calibrating the imaging velocity model with well data results in velocities for depth
conversion that carry the high accuracy of the well data. At the same time, this calibra-
tion provides for interpolation between wells using the imaging velocity model as a
guide. The result is a depth conversion with a unified velocity model, where seismic

1-4 ‡ Society of Exploration Geophysicists

 Phil Schultz

data, interpreted fault surfaces, and interpreted structural time surfaces are all converted
to depth together. That which tied in the time domain will now tie in the depth
domain.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

In addition to optimal imaging and the highest quality of depth conversion, litholo-
gy identification of large-scale units of uncertain genesis can sometimes be made with
the assistance of the velocity model, as we will see in one of the examples.

Nomenclature
As often as possible, existing nomenclature will be used in this course. We will resist
the compelling urge to define new terms, which is too frequently done. Many of the
velocity model terms used are defined or explained in the Glossary at the end of the
book, as well as in the text itself. Hopefully, I have covered all terms that you will find
sufficiently confusing or unclear. If not, I apologize and hope that the in-text descrip-
tions and explanations are sufficient.

A Brief Overview of the Day

The day will be broken into four sessions: two sessions in the morning, followed by two
in the afternoon, separated by breaks.
Morning Session 1: The first half of the morning will be dedicated to introducing the
idea of the velocity model and discussing how an interpreter’s input is critical. This will
be followed by a comprehensive case study from Geco-Prakla that showcases the devel-
opment of the velocity model for depth imaging of the pre-Zechstein Westphalian Sands
using geological input as constraints through well calibration and final depth rendering.
Morning Session 2: The second half of the morning will be spent reviewing the basic
concepts of velocity model-building, including the steps that normally involve inter-
preter input. We will discuss:
U The fundamental methodology for model-building (i.e., not specific to any one
company or vendor).
U The value of rapid visualization of the model in multiple domains, specifically time
and depth.
U The types of velocities, such as imaging and depth-conversion velocities and how
they differ.
U The image gather and how it is used to update velocities.
U A walk-through example to determine a residual velocity update on an image gath-
er, using only a ruler and calculator.
U How anisotropy causes the imaging velocity model to differ from the depth-
conversion velocity model.
U Another walk-through exercise and calibrate a one-dimensional velocity model to
well data, reported as depths to formation tops.

Distinguished Instructor Short Course ‡ 1-5

 The Seismic Velocity Model as an Interpretation Asset

Afternoon Sessions: The two afternoon sessions will comprise a series of case studies,
with a round-up and summary at the end of the day. We will finish the day with some
speculation about where the increasing commonality of a comprehensive velocity model
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

will take the interpretation process. The case studies addressed in the afternoon sessions
will include:

1) Clarifying the Image

UÊ 3DX Technologies: Clarifying complex faulting
UÊ ARCO: Enhancing interpretability
UÊ Elf Exploration: Improving volumetric calculations
UÊ Oryx Energy: Distinguishing a salt weld from a salt boundary
UÊ Phillips Petroleum: Improving the structural detail of a subsalt play

2) Seeing Below Gas Seepage

UÊ Mobil: Detailing structural continuity
UÊ Spirit Energy: Enhancing the estimate of structural closure below gas

3) The Fault Shadow

UÊ COX and Perkins/DDD Energy: Identifying a possible permeability barrier as a
false image

4) Getting Well Placement Right

UÊ British Petroleum: Enhancing the positioning of complex faulting
UÊ Enron Oil & Gas: Clarifying the termination of a reservoir

5) Evaluating Closure in Depth

U British Petroleum: Uncovering different stories from time and depth closure
U Amoco: Analyzing conflicting depth conversions
U Norcen: Finding closure in depth, but none in time
U Paradigm Geophysical: Determining depth structure under a channel fill

6) Getting Clues to the Lithology

U Veritas: Investigating the lithological character of the Faroe Shetland escarpment

1-6 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 1.
A velocity model of an overthrust
region in the Andes foothills illus-
trates that the best choice for structur-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

al layering of the model follows the

lithology, which dominates the con-
trol of the velocity contrasts.

Fig. 2.
This close-up view is of the shallow
overthrust region of Fig. 1.

Fig. 3.
The BHP water bottom copy model
is shown in the depth domain. In a
depth-of-burial environment, defining
layering as depth below water bottom
is a sensible way to geologically
constrain the velocity model. BHP
defined the layering for their Gulf of
Mexico velocity model accordingly.

Distinguished Instructor Short Course ‡ 1-7

 The Seismic Velocity Model as an Interpretation Asset

Fig. 4.
This close-up view of the shallow
region of BHP’s water bottom copy
model is still in the depth domain.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 5.
In this image, the Fig. 4 area has had
sedimentary layers and salt bodies
(dark blue) inserted. The shading fol-
lows interval velocity. The iso-velocity
contours follow neither the water bot-
tom copy (i.e., depth-of-burial) nor
the sedimentary layering (i.e., litholo-
gy) exactly, but are closer to being a
function of depth-of-burial, especially
on the left-hand side of the salt.
Because velocity models often do not
follow either guideline precisely, it is
necessary to allow spatial variation of
velocity within each layer of the
model.

1-8 ‡ Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Creating Accurate Depth Maps

2
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Creating Accurate Depth Maps

Geco-Prakla
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3D Poststack Depth Migration: A North Sea Case History*

Phil Schultz

The Objective and the Problem

A large 3D survey was shot and processed in Quad 43 of the North Sea in 1993 (Fig. 1).
In this portion of the southern North Sea the primary targets are the Carboniferous
Westphalian Sands. A structural cross section through the primary area of interest illus-
trates the basic stratigraphy and structure, and is shown in Fig. 2. The dominant struc-
tural trend is NW/SE, as seen in Fig. 3. This trend is apparent at the subsalt base
Zechstein level and in the alignment of the Zechstein salt swells and overburden.
Determination of an accurate depth structure map of the subsalt is therefore compli-
cated by large lateral velocity variations in the overburden. A folded, high-velocity
Platten Dolomite layer within the Zechstein salt results in additional raypath distortion.
Conventional time migration does not honor ray bending and thus interpretation below
the salt is made difficult by poor imaging and fault definition. True dip is also obscured
by velocity pull-up effects below the swells. Depth migration with an accurately defined
velocity model is necessary to provide optimum subsalt imaging in this area. In addi-
tion, accurate depth maps of the pre-Zechstein formations requires a high-quality veloc-
ity model, which can evolve from the model built during the depth migration. The
problem: To clarify the pre-Zechstein reflector image and create accurate depth maps of
the Carboniferous layers, in particular their fault locations.
The depth migration and velocity model-building study was motivated in the first
place by the need for image refinement for better structural interpretation in depth and
lateral fault placement. Accurate fault placement becomes increasingly important as
drilling costs go up, either because the target is deep (such as in the deepwater Gulf of
Mexico subsalt domain, as we shall see in the Oryx case study) or because the area is
remote (as in the Alaskan North Slope case study of BP). Lateral repositioning is one of
the benefits of depth migration with an accurate velocity model.
However, what does depth migration have to do with correct depth structure? After
all, shouldn’t it just be positioning events more correctly laterally, due to accommoda-
tions made to raypath distortions? Haven’t we been told that the depth-migration veloc-
ity model is not the same as the depth-conversion velocity model?
During this course, we will return frequently to the notion that velocity model-
building for the purpose of depth imaging generates an asset of enduring value. It will
serve, as we shall see in this case study, as an advanced jumping-off point for creating a
velocity model suitable for depth conversion.

*Adapted from “3D Poststack Depth Migration: A North Sea Case History,” by Michael O’Briain, Alastair Thomson
and Francois Daube, all of Geco-Prakla and Schlumberger.

Distinguished Instructor Short Course ‡ 2-1

 The Seismic Velocity Model as an Interpretation Asset

Survey, Data and Initial Model Description

The seismic survey covered 1,875 sq km, and was processed in 1994 by Geco-Prakla. In
1995, Geco-Prakla completed a poststack depth migration project over this area. Called
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the “Cavendish survey,” the outline forms a long rectangle approximately 71 = 25 km,
straddling two blocks of the Southern Gas Basin. The area included 11 wells, with
depth markers to key formation tops. Fig. 1 shows a map view of the area of the study.
The stack data volume comprised 11.4 million traces. The final time migration (before
the depth imaging study commenced) was done onto a grid of 12.5 m = 12.5 m using a
horizon-consistent velocity field derived from stacking velocities converted to interval
velocities using the Dix equation. That is, there was no accommodation made for ray-
path distortions in development of the migration velocity field. However, interval veloc-
ities were computed in a horizon-consistent manner.
Time migration tends to be a relatively forgiving operation, and the velocity field
(as opposed to a velocity model) required as input to the process normally requires only
salutory interpretation input. The interpreter can select the horizons to be used to guide
the interval velocities and review the velocity values to make sure that they are reason-
ably consistent with geological expectations and formation data from well logs. How-
ever, detailed input is not normally beneficial to either the processing or the interpreta-
tion tasks at hand, because the model is designed for a time migration and, therefore, is
not generally geologically consistent, except in a large-scale sense.
This tends to be true for two reasons. First, the time migration operation does not
require (and in general does not benefit from) a detailed velocity model. It just needs to
be accurate in a general sense. Second, the image has not been corrected for raypath
distortions beyond what the RMS approximation can itself accommodate. So, any effort
to interpret the time-processed image will fall victim to the imperfect image and to the
lack of depth representation of the seismic data.
In other words, we are looking for a velocity model to be developed to the point at
which its detail becomes sufficient to contribute to the interpretation process. Such
level of detail requires interpretation input for iterative refinement in the first place.
Here then, is a rule of thumb:
If there is no interpretation input in the development of the velocity model,
there is probably not much interpretive value to be derived from it.
The real interpretive value in a velocity model comes from its development as part of an
evolving understanding of the geological structure, tectonics and lithology.
Time migrations are normally the final stage done solely by the processing staff. The
velocity field created for this time migration step, as well as the velocities estimated
from DMO-corrected seismic data (another step solely in the domain of the processing
geophysicist), are two starting points for the involvement of the interpreter. This stage
is normally when, for the purposes of this course, interpretive input of any consequence
actually begins.

Let’s Build the Model

The standard processing ended with poststack 3D time migration, for which a reason-
ably mature processing velocity field was created. This information was used as an ini-

2-2 ‡ Society of Exploration Geophysicists

 Phil Schultz

tial model in the velocity updating required for depth migration. As is often the case, a
layer-by-layer approach was taken in developing the velocity model for depth migration.
Depth migration is one of the most powerful tools available to the interpreter to
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

develop an accurate velocity model. However, a properly depth migrated data set is also
one of the more difficult objectives to reach, in that it requires an effective handshake
between the geophysicist responsible for the processing and the geophysicist responsi-
ble for the interpretation. The process for developing the depth-migration velocity
model is iterative, requiring a series of successively more accurate trials.
In the next section on methodology, we will investigate the common steps in build-
ing the velocity model, in particular where interpretive input is required. For this study,
we will see an example of how this specific model was built.
In this case study, the size of the data set was so large that it was decided to update
the velocity model using a technique called coherency inversion, rather than with
prestack depth migration and the image gather, which is the most common way of
updating the velocity model. In coherency inversion, raypath distortions through the
most current model update are computed and nonhyperbolic moveout is modeled in
order to estimate the velocity in the layer currently under investigation.

A Layer-By-Layer Approach
The velocity model will eventually be defined by seven layers. Each layer was defined
with the following geophysical and geological considerations:
U Down to what next layer or formation boundary will the velocity behave in a rela-
tively uniform fashion? This information is normally available through local knowl-
edge and familiarity with the velocity profile from calibrated sonic log data, as was
the case in this study.
U What is the correct shape in depth for the layer boundary? Here the interpretation
must evolve together with the velocity model. In cases needing depth migration,
there normally is significant velocity time/depth distortion. Building a competent
velocity model requires defining depth surfaces that are either significant formation
boundaries or unconformities (since they often define velocity contrasts). The inter-
preter needs to remember that with seismic data alone there is not always a clear
indication of where are the significant velocity interfaces and formation boundaries,
which are important as the velocity model is defined by a series of layers. In some
cases, such as in the soft rock environment of the Gulf of Mexico, the significant
velocity changes are more a function of depth of burial than of facies changes. In
some cases, such as this one, there is a combination of the two: Depth of burial
velocity gradient behavior for the shallower layers and for deeper layers specific
formation boundaries (e.g., the Upper Cretaceous Chalk and the Zechstein)
defining natural layer boundaries.
U What is an appropriate gradient to apply to the velocity in this layer? Studying the
downhole data and general familiarity with the area are required here. In this case
study, logs are interpreted to estimate velocity gradient values on a layer-by-layer
basis, where the layers are defined as part of the velocity model. Note that the defi-

Distinguished Instructor Short Course ‡ 2-3

 The Seismic Velocity Model as an Interpretation Asset

nition of the layers themselves is an interpretive step, which could include consider-
ation of how the gradient behavior changes with depth. Are you beginning to see
the many ways that interpretive input increases the value and effectiveness of the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

velocity model?
U What thickness is appropriate for this layer for the upcoming model update (i.e.,
how deep shall this layer go)? This question is a combination of a processing and
interpretive issue. On the processing side, the layer needs to go deep enough so that
the model updating will develop the model as much as possible with each iteration.
On the interpretive side, the layer to be defined needs to include a sufficiently uni-
fied package of sedimentary layers so that the velocity behavior can normally be
described by a constant plus a linear gradient, which can be spatially variable.
U What is a reasonable range of interval velocities within which to constrain the
velocities in this layer of the model? Here the information comes from a combina-
tion of the processing velocity fields (stacking or time migration) and the knowl-
edge of the velocity profile as a function of depth and formation, gained from study
of logs, outcrops, cores, etc.

What Layers Would be Chosen?

For this survey, Geco-Prakla chose to define five major layers in the model overlaying
the Zechstein. The base of the layers chosen for the model corresponded to the Lower
Tertiary, the Cretaceous Chalk, the Lower Cretaceous-Lower Jurassic, the Upper
Triassic, and the Lower Triassic (the Bunter). Several deeper layers will be defined later
in the analysis.
This choice of layering has both processing and interpretive implications. That five
layers were chosen over the Zechstein was a judgment of the thickness of layering that
could be resolved in one iteration of the velocity model-building process. That the
boundaries were chosen at those formation boundaries reflected knowledge of the
velocity and gradient behavior in each of the grouped formations, as well as having the
velocity boundaries defined by visible reflectors on the seismic section, so that the
structure could be effectively interpreted. Remember that the velocity model is a combi-
nation of structured layers and velocities within them.

Getting That First Layer

Initially, it is necessary to interpret the base of the first layer in the time domain, since
the velocity model has yet to be defined in this layer. The next step in this study is to
estimate the interval velocities within this layer at specific locations using ray tracing
(remember that this study uses ray tracing and coherency inversion rather than prestack
depth migration and image gather analysis).
Accordingly, the Lower Tertiary was interpreted in the migrated time domain in the
data volume of this 3D example. Because ray tracing was being used, it was necessary to
inverse migrate this time interpretation to the unmigrated time (or stacked) domain,
using the migrated time volume from which the interpretation was made.

2-4 ‡ Society of Exploration Geophysicists

 Phil Schultz

After this Lower Tertiary structural surface was (1) chosen as the base of the first
velocity layer, and (2) interpreted structurally in the migrated time domain (and then
inverse migrated), the coherency inversion method was used to estimate velocities on a
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

grid of locations for the first layer. A total of 205 locations were analyzed with coheren-
cy inversion, resulting in an analysis grid of 2.5 km = 2.5 km spacing. Velocities esti-
mated via coherency inversion at these locations were mapped (i.e., spatially interpolat-
ed) throughout the layer.
Then, poststack migration was done to clarify the reflector in depth, at which point
the reflectory structure could be redefined in depth, if necessary. Before the first layer is
finished, the base of it must be defined in depth, and the velocities must be defined spa-
tially within the layer.

A Sample Iteration
We now go in detail through one of the iterations in the velocity model-building
process. At the start of any iteration, the velocity model down to a certain layer bound-
ary has been defined and is “known.” The task in this iteration is to estimate the depth
structure and the velocity map in the next layer, assuming that the previous layers are
correct and will not change. Fig. 4 shows the structural component of the “known”
velocity model, with the volume below yet to be determined.
In this iteration, we define a layer bounded above by the Lower Cretaceous/
Lower Jurassic, which is the base of the previous layer. Now, having decided to select
the Upper Triassic as the base of the current layer, we will define it.
The first step is to pick the time domain structural surface corresponding to the
Upper Triassic. This is typically done as any structural surface is interpreted, but here
we need to ensure that the surface is in some way continuous across the entire area of
the survey, since it needs to “contain” the velocities for this layer. So, the Upper Triassic
surface is interpreted in the migrated time domain normally. Again, because in this
study we are using coherency inversion, rather than image gather analysis from
prestack depth migration (which we will discuss later), we must “demigrate” this inter-
pretation to the unmigrated time domain. The result is shown in Fig. 3.
Next, we need to specify an initial velocity for this layer in order to start the coher-
ency inversion velocity analysis. Since we will scan many velocities, this choice is not
critical, but it is useful to be close. In this case, a constant velocity was used in the
unknown layer. Once the trial velocities are determined, the trial model layer can be
prepared for ray tracing. Fig. 5 shows the way in which Geco-Prakla prepared the
model for ray tracing, by a surface tesselation. This is not the final model for this layer;
it is simply a computational step to enable accurate ray tracing, which is part of the
coherency inversion operation.
Then, coherency inversion is performed to obtain the best estimate of interval veloc-
ity. Within the coherency inversion operation is ray tracing through the known and
unknown portions of the velocity model to estimate possible non-hyperbolic moveouts
for the event being analyzed (which in this case is the reflector corresponding to the
Upper Triassic), as shown in Fig. 5b.

Distinguished Instructor Short Course ‡ 2-5

 The Seismic Velocity Model as an Interpretation Asset

Once the ray tracing is performed, statistics and semblances are presented to the
analyst, who picks velocities for the layer under study. Fig. 6 shows the analyst’s inter-
face for this procedure.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

After the interval velocities have been picked in each of the analyst’s locations for
this layer, these 205 “point velocity values” (which represent the velocity estimate in
the model for one layer at one x,y location) are mapped into an areal distribution of
velocities within the current layer, as seen in Fig. 7.
From the estimate of velocities in this layer, create a 3D velocity model combining
the velocities in all known layers, including the current one (Fig. 8). With these veloci-
ties, perform a 3D poststack Kirchhoff depth migration to a depth below the current
layer, to ensure that the base of the current layer (Upper Triassic) is completely imaged.
Then, repick the Upper Triassic surface in depth (Fig. 9a). The current layer is now
completely defined. Now perform a similar analysis for the next layer, and so on.
Why is it necessary to repick the Upper Triassic in depth? It was already picked in
the time domain. There are two reasons. The lesser reason is that the image is often
clearer, and the layer can be defined more accurately. A more important reason is that
“geology happens in depth,” and the geological structural surfaces, whose shapes are
important to refining the interpretation of the evolution of the basin, exist in the real
world in the depth domain.
Sometimes a “reasonable” surface interpreted in the time domain can easily become
an unreasonable surface in the depth domain. For example, let us digress briefly and
consider the model of Fig. 9b. It is a subsalt interpretation from a Gulf of Mexico exam-
ple, interpreted in the time domain. The reflector continuity below salt was not clearly
visible and had to be inferred.
Reflector “pull-up” was interpreted below salt to account for the velocity effect. Is
this model reasonable? A quick time-to-depth conversion, seen in Fig. 9c, shows what
the consequence of this interpretation looks like in the depth domain. Clearly, a non-
geological interpretation.
In the next figure, Fig. 9d, the interpretation of the subsalt reflectors is done in the
depth domain, where no mental gymnastics are required to infer a depth structure from
a time structure. The interpretation is done immediately in the depth domain.
The final figure in this sequence, Fig. 9e, shows what you would have had to inter-
pret in the time domain to create the same interpretation in depth as seen in Fig. 9d.
This interpretation requires significant mental gymnastics and shows that interpreting
in the time domain is fraught with traps and pitfalls.
While this example is severe to demonstrate the point (the salt creates a very large
velocity contrast), the same thing happens less dramatically whenever you interpret
seismic data in the time domain. This shows why it makes sense to reinterpret the
Upper Triassic horizon defining the base of the current velocity model layer in depth.

Post-Zechstein Overburden
As previously mentioned, post-Zechstein was defined in the velocity model with five
layers, with bases at the Lower Tertiary, the Cretaceous Chalk, the Lower Cretaceous/
Lower Jurassic, the Upper Triassic and the Lower Triassic (the Bunter). When the

2-6 ‡ Society of Exploration Geophysicists

 Phil Schultz

velocity model was estimated in each layer, gradients were imposed on the velocities,
according to what was seen on the sonic log data. Fig. 10 shows a sonic log that was
used to help infer gradients in the layers of the velocity model. (It also helped suggest
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

natural formation boundaries for the structural component of the velocity model). On
the log, linear velocity gradients were estimated by curve fitting, and the following val-
ues were used in the model: A gradient of 0.28 for the Lower Tertiary layer, 1.47 for the
Cretaceous Chalk, and 0.36 for the Jurassic, the Upper Triassic and the Bunter.
Fig. 11 shows the Upper Triassic velocity grid estimated from the model geometry
and the velocity estimates via coherency inversion in that layer. The repicked Upper
Triassic depth map is shown in Fig. 12. The actual repicking was done only on every
10th crossline and was interpolated into the intervening crosslines.

Zechstein and Substratum

Though coherency inversion honors ray bending, it becomes unreliable in areas of
severe ray path distortions. Poor data quality and a lack of moveout resolution, as seen
on a CMP at this depth, made this method unsuitable below the top Zechstein.
The Zechstein interval velocity is dependent on the varying proportions of halite,
dolomite, anhydrite and shale. The greater the thickness of the Zechstein, the greater is
the proportion of halite and, therefore, the interval velocity is lower. This effect is due
to the preferential lateral flowage of the halite into the salt swells. The salt velocity
within the Zechstein was therefore analytically determined using well-calibrated
isochron thickness compared against the picked velocities in crossplots.
Within the Zechstein interval, the highly folded Platten Dolomite unit was interpret-
ed with a constant thickness of 30 m, and assigned a velocity of 5,600 m/s (Fig. 13).
The Base Zechstein was the last true interpreted layer in the model (Fig. 14). An
isochron was used to subdivide the subsalt into two layers. The shallower Silverpit
Formation was assigned a constant velocity of 4,600 m/s, and the deeper Carboniferous
was assigned a velocity of 4,500 m/s. Fig. 15 shows a vertical cross section through the
final depth model, including the deeper velocity layers.

Imaging Results
A comparison of a time- and depth-migrated cross sections highlights the improvements
in imaging quality. Figs. 16 (time migration) and 17 (depth migration) show the compar-
ison sections. Note the Platten Dolomite and Base Zechstein where the time migration is
often characterized by under-migration effects. In the Carboniferous interval, imaging
and fault resolution is improved. The apparent dip of the Base Zechstein structure has
changed. It now reflects the true dip and is not distorted by velocity pull-up effects.

Well Calibration
Even though velocity model-building and depth imaging create a seismic depth volume,
their main contribution is an improved image. The depth rendering, via the velocity
model used for depth migration, is not sufficiently accurate to tie the wells. A major
reason for the misties is velocity anisotropy Fig. 18. The migration velocity analysis

Distinguished Instructor Short Course ‡ 2-7

 The Seismic Velocity Model as an Interpretation Asset

measures the horizontal component of velocity, and the depth conversion requires the
vertical component of velocity. The horizontal component is often faster, and commonly
makes the well depth markers come in at a shallower depth than the corresponding
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

seismic reflection event.

For this reason and others, it is almost always required to follow the depth imaging
operation with a depth calibration to the wells. Note that this is not a step requiring
further migration. The migration velocity model is the one that best images the seismic
data, and that velocity is used for all the migrations.
We start with the final depth-migrated image, but rendered in time. That is, the
depth-migrated volume is converted back to time using the depth migration velocity
model by doing a simple vertical time stretching. The migration velocity model is used
as a starting point. Well data, in the form of either time-depth curves or depths to for-
mation tops, are then used to calibrate the velocity model so that they tie at the well
locations, while the velocity model still retains its character between the wells. The
resulting velocity model is used to reconvert the depth-migrated seismic data, now in
the time domain, back to a corrected depth domain so it will tie the wells.
In this survey, there were 11 wells available to use for depth calibration of the final
imaging velocity model. As is often the case, these wells were largely confined to the
structural highs. The depth error between the formation depths, as recorded in the
wells and the depths of the corresponding seismic reflectors on the depth-imaged sec-
tion, were recorded for each layer (Table 1). Depth errors down to the Jurassic tended
to be random, or at least were not biased toward the positive or negative. Within the
Triassic, however, consistently positive depth errors of around 100 m were observed. A
crossplot of shale isopach with depth error (Fig. 19) indicated anisotropy of about 9.5%
within the unit.
Three different approaches to well calibration were considered:
U Method 1: A correction surface of 9.5% of the Triassic thickness was applied to the
Base Zechstein. Fig. 20 shows the correction surface in map view.
U Method 2: The errors induced by the Triassic section were contoured, and this sur-
face correction was applied to the Base Zechstein. This surface had a similar trend
and values to the correction surface in Method 1.
U Method 3: For each of the key layers, Triassic, Bunter and Base Zechstein, a single
average bulk error was calculated and applied to each of the depth surfaces.

Table 2 compares the remaining errors after each of the calibration techniques. The
average bulk error correction method was chosen. It was considered by Geco-Prakla as a
valid approach given the small variation of the Triassic isopach in the area. The resulting
average depth error at the Base Zechstein level was less than 0.79%, following calibra-
tion. The final calibrated Base Zechstein depth map from Method 2 is shown in Fig. 22.

Conclusions
The Cavendish project was a velocity model-building effort that started from the pro-
cessing result (i.e., stacking and time-migration velocities) and developed a comprehen-
sive velocity model for depth imaging using geological constraints together with geo-

2-8 ‡ Society of Exploration Geophysicists

 Phil Schultz

physical methodology. Finally, the velocity model for depth conversion was calculated
by calibration of the depth imaging velocity model to well data.
To the interpreter, this project shows that interpretive input is necessary in the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

process of velocity model-building for depth imaging, in order for the resulting depth
migration to yield the best seismic image. On the other side, for all the input of the
interpreter, the return is not only a high-quality seismic image with all lateral position-
ing correct, but a velocity model that can be used and updated on an ongoing basis
with additional well information. Thus, it becomes increasingly valuable as an enduring
asset in the interpretation process.
The velocities coming out of standard, time-based processing are normally not
sufficiently accurate for interpretation. However, the velocities developed by the itera-
tive updating of the velocity model in the depth imaging process, together with the
geological constraints imposed by interpretive input, create a model of sufficient
accuracy that it can contribute substantially to the interpretation process in its own
right. In this sense, the velocity model is an asset to be saved and used, and not simply
one more disposable set of parameters from the seismic processing routine.

Fig. 1.
The survey area, located in Quad
43/44 of the North Sea, covered
1,875 sq km. The data is called the
Cavendish survey, which is roughly
71 km = 25 km straddling two blocks
of the Southern Gas Basin. The sur-
vey objective was imaging the
Carboniferous Westphalian Sands.

Distinguished Instructor Short Course ‡ 2-9

 The Seismic Velocity Model as an Interpretation Asset

Fig. 2.
A stacked time cross section shows
the stratigraphy. The primary target is
the Carboniferous Westphalian Sands.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 3.
This time surface interpretation of the
Upper Triassic is based on a stacked
time volume of the seismic data. The
map shows the structure trending
northwest to southeast.

Fig. 4.
An intermediate stage of the iterations
to resolve the velocity model is shown
and includes the layers that already
comprise the velocity model. The
zones below the known layers are to
be estimated in the iterative process.
At this stage the deepest defined layer
in the model is the Lower Cretaceous/
Lower Jurassic. The layering in the
velocity model follows the interpreted
time surfaces. To this stage, three lay-
ers have been defined in the velocity
model.

2-10 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 5.
Within each layer in the model, the
velocity is allowed to vary both later-
ally and vertically, with gradients.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Thus, the layering serves as a guide

for the interpolation and mapping of
velocities in the volume and does not
force constant velocities within layers.

Fig. 5a.
In performing the velocity analysis
prior to depth migration, ray tracing
is done so that the best velocity esti-
mates are available. Using ray-traced
arrival times to estimate the velocity
in CDP gathers within defined layers
is usually achieved through coherency
inversion (see Fig. 6).

Fig. 6.
Coherency inversion lets the process-
ing analyst investigate details of the
nonlinear moveout associated with
specific events. A velocity estimate for
the target layer is obtained, accommo-
dating the layered velocity model.
Velocities are studied in detail nor-
mally at selected locations, where
velocity scans provide the best fit to
enable an accurate estimate of the
interval velocity in that layer, assum-
ing the layers above are already
defined and known. Because this
analysis is over interval velocity, the
analyst is ideally constrained by geo-
logical and downhole information
about the range of velocities and
expected values, typically provided
by the interpreter.

Distinguished Instructor Short Course ‡ 2-11

 The Seismic Velocity Model as an Interpretation Asset

Fig. 7.
After the detailed velocity analysis at
specific locations (nodes), it becomes
necessary to map the velocities into
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the layer. Here the areal behavior of

the distribution of velocities is con-
strained both by smoothness criteria
and geologically.

Fig. 8.
Once the velocities are mapped into
the layer, it can become part of the 3D
velocity volume for input into the
next iteration of prestack or poststack
depth migration.

Fig. 9a.
Here the poststack depth migration,
full-volume result was interpreted for
the structure of the next layer in the
velocity model, the Upper Triassic.
Ideally, the image down to the next
layer to be interpreted is clarified, as
well as displayed in depth. The inter-
pretation is then ideally done in the
depth domain, which allows interpre-
tations to proceed in a natural geolog-
ical context and eliminates, or signifi-
cantly reduces, time-based distortions
of the structure.

2-12 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 9b.
Interpretation in depth, rather than
time, is one way to avoid a problem
common in time interpretation: a
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

“reasonable” interpretation in the

time domain can quickly become an
unrealistic interpretation viewed in
depth. Here in the time domain, sub-
salt layers are inferred, with some
velocity pull-up included to account
for the velocity effect.

Fig. 9c.
The same model as defined in Fig. 9a
is shown here in the depth domain.
Clearly, the horizons picked are geo-
logically unreasonable.

Fig. 9d.
Now the horizons are picked in the
depth domain.

Distinguished Instructor Short Course ‡ 2-13

 The Seismic Velocity Model as an Interpretation Asset

Fig. 9e.
The time-domain effects of the depth-
domain picks of the previous figure
are shown. These are the horizons
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

that would have to have been picked

in order for the depth picture to look
“reasonable.”

Fig. 10.
Not only the definitions of the layers
themselves, but also the estimates of
the velocities within each layer
requires an implicit interpretation
step. Here the sonic well log is used
to estimate velocity gradient functions
in layers already selected for the
velocity modeling. Gradients were cal-
culated for the Lower Tertiary (0.28),
the Cretaceous Chalk (1.47), and for
the Jurassic, Upper Triassic and
Bunter (all 0.36). Note that the layer
boundaries in the velocity model can
be chosen to coincide with abrupt
velocity changes at geologic formation
boundaries.

2-14 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 11.
The Upper Triassic interval velocity
map was based on coherency inver-
sion results.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 12.
The Upper Triassic depth map was
picked as an interpretation step with-
in the iterative velocity model-build-
ing process. Picking structure in
depth gives the interpreter a chance to
apply geological intuition, and not be
required to mentally convert from
time to depth.

Fig. 13.
The Platten Dolomite unit within the
Zechstein interval is shown. The top
of this highly folded unit was picked
from the imaged seismic and inter-
preted (for velocity model-building
purposes) as having a uniform thick-
ness of 30 m and a velocity of 5,600
m/s.

Distinguished Instructor Short Course ‡ 2-15

 The Seismic Velocity Model as an Interpretation Asset

Fig. 14.
The Base Zechstein depth surface was
picked in depth, and it was the deep-
est interpreted surface in the velocity
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

model. Below this layer, an isochron

was used to subdivide the subsalt
region into two layers, the Silverpit
and the Carboniferous.

Fig. 15.
A cross section of the final velocity
model is shown. This model was
derived with repeated iterations of
imaging, structural surface picking
and estimation of layered velocities.
After the velocity has been defined in
a layered formulation, it can be con-
verted to gridded form, as seen here.

Fig. 16.
The time migration result shows sev-
eral deficiencies in image quality.
Also, there is no detailed velocity
model from which to build a final
depth rendering of the data.

2-16 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 17.
The depth migration result shows a
much higher overall image quality. In
addition to greater interpretability,
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

this result is accompanied by a

detailed velocity model, from which
further refinements can be made from
well calibration for depth rendering.

Fig. 18.
The depth surface of the Base
Zechstein is based on the velocity
model derived using the steps
described in the previous figures. This
is the velocity model used in the
depth migration. When the depth sur-
face (from the depth-migrated seismic
data) is compared to well data, they
do not tie. Here the average error
of the mistie is 3.2%. Anisotropy is
the usual culprit.

Fig. 19.
A crossplot shows the shale isopach
with depth error indicating anisotropy
of approximately 9.5% within the
Triassic.

Distinguished Instructor Short Course ‡ 2-17

 The Seismic Velocity Model as an Interpretation Asset

Fig. 20.
In the first correction approach, a cor-
rection surface of 9.5% of the Triassic
thickness was applied to the Base
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Zechstein.

Fig. 21.
In the second correction approach,
instead of an average error of 9.5%,
the errors were contoured, and this
correction to Triassic thickness was
applied to the Base Zechstein. The
resulting calibrated depth surface is
similar in appearance to that seen as a
result of applying the first correction
approach (Fig 20). For the third cor-
rection approach a single average bulk
error was calculated and applied to
each of the depth surfaces.

Fig. 22.
The final corrected Base Zechstein
depth map, as shown, was created
using approach 3—the depth surface
was picked from the seismic depth
migration result. The average error is
0.79% versus well depths.

2-18 ‡ Society of Exploration Geophysicists

 Phil Schultz

Table 1.
Depth errors of the interpreted sur-
faces used in the velocity model ver-
sus depth measurements from the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

wells. Anisotropy corrections may be

appropriate where depth errors seem
to be biased rather than random.

Table 2.
Depth errors of the Base Zechstein
layer at each well location after appli-
cation of each of the three approaches
to depth calibration.

Distinguished Instructor Short Course ‡ 2-19

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Process: A Tour
The Velocity Model-Building
3
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

The Velocity Model-Building Process: A Tour

Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Experience has taught that informed interpretive input, accessibility of the model, rapid
visualization and rapid response are at least as important to building an effective velocity
model in a timely manner as adherence to any specific updating scheme. Accordingly,
we now look at the way in which velocity models are built only in a general sense,
focusing on the interpreter’s role.
Velocity model-building is a relatively new concept in integrated interpretation and
processing. Traditionally, velocities have been described for stacking and poststack time
migration in such a way that formal modeling was not required. For these purposes, it
was (and still is) necessary only to describe the velocity as a processing parameter that
varies over space and time. In such a scheme, the geological environment over which
the seismic data have been collected provides little guidance to the quantitative selec-
tion of parametric velocity values. For stacking purposes, the distribution of the veloci-
ty parameter can be envisioned as a field of numbers: the velocity field.
For downstream work, such as depth migration and depth conversion, the accuracy
of the velocity estimates are required to be quite high. The velocity field evolves to a
velocity model when interpretive geological and well log input are necessary to achieve
the required accuracy. Fortunately, the velocity model evolves, starting from a very basic
understanding of stacking velocities. Later, as the model is refined, geologic constraints
become essential input.
Fig. 1 shows a diagram of the evolution of the velocity model. We can say that there
are three stages to the life of a velocity model. In the first stage, the velocities are built
up as a set of processing parameters. Here the input is typically purely from the proces-
sor, and velocities are often treated as processing parameters, which just happen to have
dimensions of velocity. In this stage, velocities are taken from brute stacking velocities,
through the DMO process and DMO stacking velocity analysis, all the way to poststack
time migration. There is normally no essential interpretive input in this first stage.
The second stage is marked by a joint effort between the interpreter and the proces-
sor. The depth migration velocity model is built up in a series of iterative steps, requir-
ing velocity updates based on image gather analysis and structural updates in the depth
domain. Both the interpreter and processor play essential roles in this stage, and nor-
mally the velocity model-building progresses much more effectively and accurately
when interpretive input is part of the updating process.
The third stage involves conditioning the imaging velocity model to prepare it for
depth conversion. The final imaging velocity model is taken as the starting point for
building the depth-conversion velocity model. The process involves calibration of the
velocities (not the maps) to downhole data. We will review later why the velocity model
for depth conversion differs from that of depth imaging, but the important point in this
stage is that the development of the depth-conversion velocity model is the last stage in
this process. The depth-imaging velocity model is used as a starting point. Velocity model
calibration can be accomplished purely with interpretive input. The processor can facili-
tate the execution, but his/her input is not required for an effective model calibration.

Distinguished Instructor Short Course ‡ 3-1

 The Seismic Velocity Model as an Interpretation Asset

Notice that the farther downstream the velocity model-building process proceeds, the
more valuable the product becomes to the interpretive process. The interpreter has an
interest in developing the best possible velocity model because the end product becomes
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

an asset with growing value. For example, when the next well is logged, the marker
depths can be included in a new calibration and update of the velocity model for a
refined depth conversion. In this vein, both the interpreter and the processor provide key
input to the velocity model-developing process and both benefit from it (Fig. 2).
Stage two, in which the depth-migration velocity model is developed and the princi-
ple collaboration between the interpreter and processor takes place, is depicted in more
detail in Fig. 3. The first step is to develop a strategy for building the initial model.
Choices have to be made on how thick the units are to be in the layering of the horizon
model, whether to use a gradient-based model and what the gradients are in each layer,
among others. We saw these issues addressed and resolved in the Geco-Prakla case
study, in which decisions were made about which geological units to consolidate into
velocity model units and how to utilize gradients measured in the well.
The initial migration model is then built, requiring a structural layer reflecting
either structure (in hard rock) or an expected iso-velocity contour surface (in soft
rock). Initial velocities are then imposed in that layer. Prestack depth migration gener-
ates image gathers and a depth-migrated section. From the first, velocity updates are
estimated. From the second, an updated structural surface is picked. The migration is
then run again.
If the seismic events on the image gathers are sufficiently flat, the current velocity
model is presumed accurate down to the current layer, and the next layer is defined
with initial structure and velocities. A prestack depth migration is then run. The process
is repeated until all the layers are resolved.
The procedure described in Fig. 3 is a generalized look that does not include details
of velocity updating, which involves tools such as global tomography or defining and
updating multiple layers simultaneously. Such details are beyond the scope of this
course and are addressed elsewhere.
A visual look at the procedure diagrammed in Fig. 3 for a real velocity model built
for 3D depth migration is shown in the next set of figures. In Fig. 4a, a set of horizons
interpreted on the time-migrated data volume, is input to the velocity model-building
process. Fig. 4b shows a map view of over 1,100 DMO-corrected (dip moveout) stacking
velocity control locations mapped over one horizon in the velocity model. The velocity
values from each of these control locations are sometimes referred to as “point velocity
estimates,” because they represent point values of velocity when seen in map view.
The velocity updates from these control locations are mapped onto the horizon to
form the initial velocity estimate in this layer of the model. These raw mapped values
need to be smoothed. This operation is shown in Figs. 4c and 4d. After all velocities are
mapped into each layer in this manner, the full initial model is ready in layered form
(Figs. 4e and 4f) and also can be expressed in gridded form (Fig. 4g).

3-2 ‡ Society of Exploration Geophysicists

 Phil Schultz

Interpretive Input Required to Build the Velocity Model

What kind of interpretive input is required in building the velocity model? Here are
some examples.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

U Lithological boundaries or iso-velocity surfaces for velocity model layer boundaries?

Velocity models are built from layered units, where the boundaries between layers
ideally involve velocity contrasts. The largest velocity contrasts between geological
layers are likely boundaries for layers in the velocity model. But, as we shall see in
the examples, it is not always necessary for the layer boundaries in the velocity
model to match the sedimentary layering. In cases where the velocity is controlled
more by depth-of-burial, the velocity boundaries in the layered velocity model
should follow closely the iso-velocity contour lines. In a hard rock setting, or in
dealing with monolithic intrusions, the lithological interfaces are also major velocity
boundaries and should also define layer boundaries in the velocity model.
U Consolidation of stratigraphic and structural packages into larger units representing
a single velocity model layer.
As we saw in the Geco-Prakla study, geological units need to be packaged into larger
layers in the velocity model, wherein the number of layers normally ranges between
6 and 30. Collecting geological layers into a single velocity package works best
when the velocity behavior of all the layers is similar, i.e., when the iso-velocity con-
tours are nearly parallel to the upper or lower boundary of the velocity model layer.
Why? Since the the velocity model is parameterizing all these layers as a unit, they
will be forced to have the same lateral velocity variation. If the geological layers all
share similar velocity behavior, then the approximation involved in lumping them
together is good and the error will be small.
U Initial velocities and velocity ranges for packages.
Knowing the local interval velocities for each geological unit or layer, at least in a
general sense, is normally valuable input to building the initial velocity model.
Iterations can be saved in the depth imaging sequence if a good initial model is
used. Sometimes, all that is required is to have a good idea of the ranges of veloci-
ties. This information comes from either downhole information, or knowledge of
the lithology or velocity gradient behavior.
U Local dip information.
Often, when building a velocity model to enable the imaging of a poor data zone,
identifying the primary reflection events is not always trivial. Yet, their identification
is key to refining the velocity model, which, in turn, allows a clearer image to be
seen. This bootstrap operation benefits from a little foresight, i.e., an anticipation of
what is expected to be seen.
U TVD (true vertical depth) to key markers from logs.
As we will see later in this section, and in several of the examples, the operation of
depth conversion requires a slightly different velocity model from that of depth
imaging, although many of the characteristics are shared. It becomes necessary, in
calibrating the velocity model for depth conversion, to specify the depths to key

Distinguished Instructor Short Course ‡ 3-3

 The Seismic Velocity Model as an Interpretation Asset

geological markers from log data. This step, of course, presupposes that the logs
have been interpreted and the key formation boundaries have been identified and
depth marked.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

U Identification of key markers on seismic sections.

Tying well data to seismic horizons is a standard interpretation task and one that is
normally necessary in velocity modeling, as well. To calibrate the well data to the
seismic, identifying the key formation boundaries on the log data and tying those
depth markers to seismic horizons enables the calibration of the velocity model for
depth conversion.
U Any anomalous behavior.
In several of the examples to follow, gas seepage into overlying sediments has dis-
guised the true structure and clear image of the prospective layer. Knowing that this
velocity anomaly exists, and its expected location and behavior, made the velocity
model-buiding proceed much faster than if no information of anomalous behavior
was present. Other cases of anomalous velocity behavior require the same attention.
U Identification of the target.
This is a relatively simple concept, so obvious that it hardly calls for discussion. In
seismic velocity model-building, attention to detail normally produces an improved
result. However, time is usually short. Focusing on the key zones at the expense of
some less critical parts of the model is usually a desirable trade-off. Target identifica-
tion clearly reserves certain zones in the velocity model as more important than oth-
ers.
U Salt genesis and evolution.
For velocity modeling in the presence of salt bodies, identification of the salt bound-
aries is normally the key interpretive issue. Base of salt can be highly interpretive,
because it often cannot be seen, even after prestack imaging. Knowing the genesis of
the salt, such as over how many episodes did it grow, can help to determine the
expected shape of the base salt and greatly assist the interpretation of its structure.
Once known, the subsalt reflectors and their structures become more clearly visible,
irrespective of the image quality of the base salt itself.

The Image Gather

In the Geco-Prakla case study, the technique of coherency inversion was used to esti-
mate interval velocities in the layers defined for the velocity model. Increasingly, the
image gather is used to estimate velocities. The image gather is nothing more than a
CMP gather after depth migration. It has the appearance of NMO-corrected (normal
moveout) gathers. When an event is flat, the correct velocity was used to migrate it.
When the event is over- or under-corrected, the velocity needs to be updated.
Image gather analysis is not essential to interpretive input in velocity model-building
or to the utilization of the velocity model afterward. However, it is useful for the inter-
preter to understand the basic concepts of image gather analysis and why it is useful. In
this section, we will look at the value of estimating velocities using image gather analy-
sis and walk through an example of estimating a velocity update.

3-4 ‡ Society of Exploration Geophysicists

 Phil Schultz

In appearance, an image gather looks like a normal CMP gather after NMO. The
“NMO correction” can be acceptable, in which case the events are flat. Or, the correc-
tion can be unacceptable, in which case there is sufficient moveout remaining on the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

gathers such that additional velocity work is required. In the case of an image gather,
the velocity correction involves changes to a 3D (or 2D) velocity model, rather than just
a local change to the stacking velocity values.
We will see several velocity models in the examples, but in this section we will look
at the image gather itself and ask two questions:
1) Why are image gathers useful?
2) How is the velocity update computed based on observed residual NMO curvature?

Why Are Image Gathers Useful?

To address the first question, we will look at an example from Amerada Hess, which
compares the value of a migration velocity analysis (i.e., image gather analysis) in 3D to
one in 2D. Image gathers, because they result from prestack depth migration, focus the
seismic energy from reflectors directly beneath the surface CMP onto a single CMP
gather (image gather) for analysis, rather than having it smear over many locations, as
seen in the schematic diagram of Fig. 5. Analyzing the image gather, therefore, address-
es velocity updates local to the CMP where the analysis is done.
Of course, this operation presumes that the migration itself is done with a velocity
model that is not too far from the final, correct model. If it is too far, then the assump-
tions about the velocity updates being local, in effect become increasingly inaccurate. At
this point, a purist will point out that to be even more accurate in updating the velocity
model, it is necessary to use a tomographic update procedure after the residual moveout
analysis is complete. The purist is correct, although in many cases tomographical
updating is not required. In any case, for our purposes, such detail within the method-
ology is beside the point and will not be discussed further.
Fig. 6 shows image gathers from three locations from 3D prestack depth migration.
The horizontal axis is offset and the flatness of the events on the gathers indicates that
the velocity model down to that event depth is sufficiently accurate to be considered of
high enough quality for the final model.
Consider the seismic section of Fig. 7. We saw from Fig. 5 that dipping events are
expected to behave in some kind of organized fashion on image gathers, so that we can
evaluate their moveout and residual velocity in the same manner as we do for flat
events on normal (unmigrated) CMP gathers. This figure shows a section with gentle
dip. We would not expect the 3D image gather to bring much benefit over its 2D
cousin. For the location marked V1, a velocity spectrum was computed based on the
image gathers. Fig. 8 shows that spectrum, comparing the 2D and 3D analysis of image
gathers.
Although Fig. 8 does not show a direct comparison between an image gather veloci-
ty analysis and an standard version on unmigrated data, the comparison between the
2D and 3D effect is similar. In a 2D image gather, we do not migrate the crossline direc-
tion. So, in the crossline, all dipping event problems are just as they would be on unmi-

Distinguished Instructor Short Course ‡ 3-5

 The Seismic Velocity Model as an Interpretation Asset

grated data. True, there are some differences between the 2D and 3D spectra (Fig. 8),
but not enough to provide convincing evidence that at this location a 3D image gather
analysis is required, or even desirable.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The most obvious benefit to analysis on image gathers comes from dipping reflec-
tors (and, of course, severe lateral velocity variations, which we will come to in most of
the examples). Fig. 9 shows a different section from the same data set. At location V2
the velocity analysis will encounter significant dip, both inline and crossline.
Fig. 10 shows the velocity spectra at location V2. Here, going from 2D to 3D clearly
brings significant improvements to velocity resolution on the image gathers. We can
interpret this difference as showing the added value of analyzing image gathers com-
pared to standard CMP gathers for velocity, since the difference between these two
spectra on Fig. 10 is whether the crossline direction was depth migrated or not.

How to Update Velocities From Image Gathers

When the event on an image gather is flat (and all the events above it as well), the veloc-
ity is correct. What do you do when it is not flat? Fig. 11 shows an example of an image
gather where three different velocities were used in the migration. The first velocity is
too low (at 1600 m/s), the second is correct and the third is too high (at 2400 m/s).
Imagine only the first and the third migration trials were done. How would it be possible
to estimate the correct velocity from the two trials with known, but incorrect velocities?
Fig. 12 shows a method to estimate the velocity. This method comes from Luis
Canales, who claims that he actually used it in production for a short time to estimate
velocity updates for depth migration before specialized software was available.
Pick an event. Calculate its radius of curvature for each of the two incorrect veloci-
ties. Plot the inverse radius of curvature against the velocities used for the migration.
Draw a straight line through the two points. The correct velocity is located where the
line intersects the zero axis of inverse curvature.
In Fig. 13 we calculate the radius of curvature for the first velocity, V1. We can also
do the same for the last velocity, V2. Fig. 14 shows the plot for the case in hand. The
correct velocity calculates to be 2100 m/s.
While this method is probably not used in this simple form any longer, the idea is
similar to all current methods for obtaining estimates of the velocity corrections. The
velocity that we just calculated is often called a point velocity update, because it is esti-
mated for one spatial (CMP) location and for one event. A collection of many of these
point velocity updates can be mapped onto the horizon model.

Well Velocities and Seismic Velocities

Can we use seismic velocities derived from well data for migration? Not directly. For rea-
sons that we will not go into in this course, seismic imaging (migration) and processing
requires the horizontal component of velocity. Depth conversion requires the vertical
component of velocity. Well measurements always produce values for the vertical compo-
nent of velocity, and as such are suitable for depth conversion, but not for imaging.

3-6 ‡ Society of Exploration Geophysicists

 Phil Schultz

That being said, well velocity measurements are generally useful in building the
seismic velocity model because they provide ways to measure gradients (as we saw in
the Geco-Prakla study), and because the ratio of vertical to horizontal velocity is often
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

fairly predictable. In other words, although we normally should not use the well veloci-
ties for imaging without some kind of modification, knowing what they are will provide
valuable clues for the initial velocity model.
What is all this about horizontal and vertical velocities? Propagation velocities differ
in the horizontal and the vertical direction. Seismic waves travelling horizontally
through a geological unit will normally travel at a higher velocity than a similar wave
travelling vertically. This property is known, in general, as anisotropy, or in this specific
case, transverse isotropy. This type of anisotropy can be intrinsic, where the rock prop-
erties themselves dictate the velocity behavior. More often, the effect is a consequence
of the behavior of layered media. In a layer-cake earth model, where some layers are
high-velocity and some are low-velocity, the vertical propagation velocity will be lower
than the horizontal velocity. For this reason, velocities measured in wells will be too
low for migration.
Fig. 15 again shows an example from Amerada Hess. The spectra are computed
from image gathers and they show velocity picks from the seismic and velocities calcu-
lated from well measurements. Although the difference is not large, the well velocities
are significantly slower than those an analyst would have picked from the seismic veloc-
ity spectra.
To see that there is a significant difference, well velocities in this example were used
to migrate the seismic data. Then, velocities picked from the seismic data were used in
the migration. Fig. 16 shows the same seismic section from a 3D data set, where the
two different velocities drove the migration. The migration using well velocities looks to
be significantly under-migrated. The conclusion here is that velocity models built for
depth migration require updates and estimates directly from the seismic data itself. Well
velocities can be used for gradient measurements and to get a rough idea of the correct
imaging velocities, especially if we know the degree of anisotropy.
Often, well velocities are used in the initial model. Seismic velocities alone are then
used for all subsequent updates to the migration velocity model. Well data revisits again
when calibrating the velocity model for the final depth conversion, as we saw in the
Geco-Prakla study. This depth calibration does not require that the data be remigrated,
only that it be re-rendered into depth, using well-calibrated velocities.
Transverse isotropy is often the cause of the disparity between the best depth-
imaging velocities and the best depth-conversion velocities. Rather than digging into
all possible reasons and causes for the differences, a reasonable response it to plan to
recalibrate the seismic velocity model for depth conversion, after the imaging has been
completed. Note that the calibration of the velocity model from well velocity data
should take place after all imaging is complete. The reason, as Fig. 16 shows, is that
the well data will actually degrade the image if used for imaging.
The features of transverse isotropy are as follows:
U VH & VV.
U Usually VH > VV.

Distinguished Instructor Short Course ‡ 3-7

 The Seismic Velocity Model as an Interpretation Asset

U Stacking and imaging tend to require VH.

U Depth converison requires VV.
U It is normally caused by layered structure.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

U It causes the best imaging velocity often not to be the best depth conversion velocity.

A Word About Large Offsets

Offsets are being collected today in marine acquisition that would be impossible with-
out modern vessels and acquisition techniques. Some 6,000 m of offset is no longer
considered large. The reason for the large offsets is often the need for differential move-
out processing in subsalt domains to enhance primary reflections. One of the incidental
consequences of such large offsets is that standard normal moveout corrections are no
longer sufficiently accurate. They are based on a second order approximation to arrival
time versus offset for layered media, and assume that the moveout curve is very close to
hyperbolic. As offsets increase, this approximation is increasingly inaccurate.
Should the interpretation community care? Normally, no. This is purely a processing
problem. However, one consequence is that velocity model-building for depth imaging
may become increasingly common, because prestack depth migration has a built-in cor-
rection for the large offset problem. That is, it has no small offset approximation.
With model-building becoming, of necessity, a more and more standard procedure,
the interpreter can plan to 1) become a part of this process, directly influencing the evo-
lution of the model, and 2) benefit from having a detailed seismic velocity model as a
consequence of the processing, and usable in the interpretation stage.
Why not use it? Remember one of the themes of this course: The velocity model is
useful in several ways to the interpreter, but its value grows with the interpreter’s influ-
ence during the course of building it. In other words, if, as an interpreter, you plan to
use the velocity model for your own purposes, also plan to be a part of its creation.
Fig. 17 shows the processor’s view of why larger offset are desirable. Two velocity
spectra are shown, with the one from the larger offset range clearly showing more selec-
tivity in velocities to image the primary events, therefore giving more precise velocity
information in the model-building process.

Calibrating the Velocity Model for Depth Conversion

Normally, well velocity information comes in two forms: depth-time pairs from check
shots or VSPs (vertical seismic profiles), or TVDs (true vertical depths) to formation
boundaries. We will go through the exercise of calibrating the seismic velocity model at
one location from a set of formation top depths.
In order to use the depth information from wells, the velocity model should be lay-
ered (as opposed to gridded), so that a correspondence can be set up between the for-
mation top depths and the depths to the layer boundaries in the velocity model. Fig. 18
shows a schematic diagram of a six-layer seismic velocity model at some x,y location.
The base of layers 3, 5 and 6 represent horizons on the seismic data that correspond to
formation boundaries as interpreted in well logs.

3-8 ‡ Society of Exploration Geophysicists

 Phil Schultz

Horizon A in the seismic velocity model (base of layer 3) corresponds to the top of
the A formation. Horizon B in the seismic velocity model (base of layer 5) corresponds
to the top of the B formation. Horizon C in the seismic velocity model (base of layer 6)
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

corresponds to the top of the C formation.

Because the seismic velocity model contains both structural and velocity infomation,
there is an inferred depth to each of the layer boundaries in the model. We say
“inferred” because the observations on the seismic data are in the time domain. Depths
on well logs are actually measured and the only interpretive step is the conversion from
measured depth to true vertical depth.
The depth to horizon A in the seismic model is 6,012 ft, with a corresponding time
of 1.670 s of one-way time. As measured from the well data, the depth to the formation
boundary corresponding to horizon A in the seismic model is 6,240 ft. There is in this
case no time measurement in the well (as there would be with check shots or VSPs).
The two true measurements are the depth from the well data and the time from the
seismic data. Combining these two together gives a value for average velocity, which is
the total depth divided by the vertical one-way time. So, Vavg down to the base of the
third velocity model layer is:
Vavg = Z3/T3 = 6240/(1.670/2) ft/s = 7473 ft/s.
Now, the Vavg from the velocity model down to this layer was different, because it
used the depth from the seismic velocity model (which, as we said, was “inferred” and
therefore subject to calibration from actual measurements). So, the average velocity
from the seismic model had a value:
Vavg = Z3/T3 = 6012/(1.670/2) ft/s = 7200 ft/s.
Now, the “correct” average velocity is 7,473 ft/s and the “current” average velocity is
7,200 ft/s. If we multiply the current value by the ratio, 7,473/7,200 = 1.038, then we
get the correct value. We will use this ratio to multiply all the interval velocities in the
model down through the first three layers. Then,
V1 = 5000.0 ==> V1 = 5190
V2 = 6341.7 ==> V2 = 6582
V3 = 8610.7 ==> V3 = 8937.
But, the depths to the layer boundaries in the seismic velocity model still need to be
calibrated, as well. Because average velocity is total depth divided by vertical one-way
time, we can scale the depths directly with the scale factor of 1.0379 that we computed
earlier.
Z1 = 500 = 1.0379 = 519
Z2 = 2783 = 1.0379 = 2889
Z3 = 6012 = 1.0379 = 6240.
We now have three new depths and three new interval velocities for the first three
layers. The calibration for these layers is finished. Now, we move to layers 4 and 5.
The depth interval in these two layers as measured in the well is:
6Z = 10130 – 6240 = 3890.

Distinguished Instructor Short Course ‡ 3-9

 The Seismic Velocity Model as an Interpretation Asset

The one-way time interval between the two events on the seismic identified as these
two depth points in the well is:
6T = (2.500 – 1.670)/2 = 0.83/2 = 0.415.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

So, the new average velocity in this interval from the well data is:
Vavg = 6Z45/6T45 = 3890/0.415 = 9373.
The average velocity in this interval from the seismic velocity model is:
Vavg = 6Z45/6T45 = (10837.5 – 6012)/0.415 = 11628.
And, the correction ratio is:
Ratio = Vavg (well)/ Vavg (seismic) = 9373/11628 = 0.8061.
Multiply this ratio by the interval velocities in layers 4 and 5 to get the new interval
velocities in this layer:
V4 = 10531 = 0.8061 = 8489,
V5 = 12554 = 0.8061 = 10120.
The new depths are calibrated as before, but on an interval basis. The depths are com-
puted from the new interval velocities and the measured times.
Z4 = Z3 + 6Z4 = Z3 + (8012.8 – 6012) = .8061 = 6240 + 1613 = 7853,
Z5 = Z4 + 6Z5 = Z4 + (10837.5 – 8012.8) = .8061= 7853 + 2277 = 10130.
Now, the new depth for the base of layer 6 is given directly by the well measurement
as:
Z6 = 13050,
and the new interval velocity is simply,
V6 = (Z6 – Z5)/(T6 – T5)/2 = (13050 – 10130)/(3.0 – 2.5)/2 = 2920/0.250 = 11680.
Although not needed because it comprises only a single layer, the ratio for this last
layer is:
Ratio = [Z6(well) – Z5(well)]/[Z6(seismic) – Z6(seismic)] = 0.9661.
I leave it to the reader to calculate the new average velocities down to each layer bound-
ary in the new calibrated velocity model.
There is an advantange of a model calibrated in this way, as compared to the meth-
ods involving converting time maps directly to depth. Here we have a unified velocity
model that can be used to depth convert both seismic data and interpreted time maps
(i.e., interpreted time horizons in the seismic volume).
Because all depth conversions are done with a single velocity model, rather than a
collection of average velocity maps (one for each horizon), depth conversions for all
layers are internally consistent. In addition, depth conversion of the seismic itself
retains the same internal consistency. So, if horizons are interpreted in the time volume,
and horizons and seismic are all converted to depth using this unified model, the new
depth horizons will overlay the new depth-converted seismic in the same way when
viewing the depth volume of the seismic data together with the depth horizons.

3-10 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 1
In the evolution of the seismic veloci-
ty model, we can say that there are
three stages to the life of the seismic The Evolution of the Seismic Velocity Model
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

velocity model. Initially, it is purely a

processing tool and the seismic data Seismic
Brute velocity
[100% processor] Data Processing
processor normally exercises 100% (time processing) functions

control over its development. The sec-

ond stage involves the development of Stacking velocity DMO Time migration
functions velocities velocity field
the model for depth migration and
the interpreter ideally works together
with the processor to develop a geo- Seismic
[Joint processor/ processor] Data Processing Depth imaging
logically and geophysically sensible (depth processing) velocity model
velocity model. Finally, in calibrating
the velocity model to a tool for depth
conversion, the interpreter can nor- [Normally, 100% interpreter] Depth conversion Depth conversion
mally exercise 100% control. velocity model

Fig. 2
This cartoon shows that both the seis-
mic data processor and the interpreter Input Value
must provide input to the develop-
ment of the velocity model. At the
end, a well-developed velocity model
has value for both the processor and
the interpreter.
Processor Processor
Velocity
model

Interpreter Interpreter

Fig. 3
Devise strategy
This figure is an expansion of stage for initial
two. It shows the basic steps in devel- velocity model
oping a velocity model for prestack
depth migration. Important elements Identify unit boundary
are the interpretation of key horizons on seismic
(base layer)
to be used as velocity interfaces and
the updating of velocities using
Impose velocities
depth-migrated (image) gathers.

Prestack depth
migration

Depth Image
section gathers

Update Velocity
base layer estimate

Next layer?

Distinguished Instructor Short Course ‡ 3-11

 The Seismic Velocity Model as an Interpretation Asset

Fig. 4a
Building the velocity model normally
starts with interpreted time surfaces.
Here surfaces are imported from an
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

interpretation system and contain no

velocity information.

Fig. 4b.
The initial velocity values from the
data processing stage are merged with
the horizon data to build the initial
velocity model. Here, over 1,100 loca-
tions are shown in map view.

Fig. 4c.
Normally, after merging the velocity
values onto the horizons the veloci-
ties need to be smoothed. Here, the
smoothing is being done over interval
velocity.

3-12 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 4d.
A 3D view of Fig. 4c.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 4e.
The layered velocity model in 3D
results from the merging of point
velocity values (as seen in map view)
onto all of the horizons.

Fig. 4f.
A cutaway view of Fig. 4e.

Distinguished Instructor Short Course ‡ 3-13

 The Seismic Velocity Model as an Interpretation Asset

Fig. 4g.
A gridded version of the 3D model of
Figs. 4f and 4e. A gridded model is
one in which there is a velocity value
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

present for every 3D cell of equal size.

Many migration programs require
gridded models.

Fig. 5.
A geometry diagram for the behavior
of image gathers. These nonphysical
rays show that on an image gather
(depth-migrated gather) energy
reflected from points below the loca-
tion of the gather are present on the
image gather irrespective of the
event’s dip. On an unmigrated gather,
the energy reflected from the dipping
event at that location would show up
on a different gather, in the down-dip
direction.

Fig. 6.
Image gathers from three locations in
a 3D data set. Flat events indicate a
correct velocity model.

3-14 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 7.
A seismic section from the 3D data set
on which a velocity analysis will be
run. The dip is gentle, and we do not
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

expect much difference between 2D

and 3D, or between velocity analyses
run on unmigrated or migrated data.
The velocity analysis location is
marked as V1. (Figures 7–10 are
courtesy Amerada Hess.)

Fig. 8.
The velocity spectra from the location
marked in Fig. 7. There is little differ-
ence of any consequence between the
analysis being done in 2D or 3D, or
since the 2D case means unmigrated
data in the crossline direction,
between unmigrated gathers and
image gathers. In a gentle dip case,
image gather analysis brings little,
provided there is not much lateral
velocity variation.

Fig. 9.
A different section from the same 3D
data set. Here the dip is significant
and we expect migration to make a
difference in the velocity analysis. The
second velocity location is indicated
as V2.

Distinguished Instructor Short Course ‡ 3-15

 The Seismic Velocity Model as an Interpretation Asset

Fig. 10.
The velocity spectra for location V2,
where there is significant dip. This
comparison, which is between 2D and
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3D image gather analysis, also shows

that image gather analysis should
improve over unmigrated gather
analysis, since the 2D image gather is
unmigrated in the crossline direction.

Fig. 11.
Three versions of the same image
gather, where three different trial
velocities were used for prestack
depth migration. If only the first and
last were run, could we compute the
correct velocity of the second panel?

Fig. 12.
This method of estimating the
unknown velocity of the second panel
was developed by Luis Canales. Plot
the velocity as a function of the
inverse radius of curvature and look
for the intercept with the axis where
the inverse curvature is zero.

3-16 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 13.
This figure shows the calculation of
the radius of curvature for the first
panel. The calculation for the last
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

panel proceeds in a similar fashion.

This radius of curvature and the
velocity of the migration will be used
in the calculation in Fig. 14.

Fig. 14.
We use the calculated parameters
from the previous figure to estimate
the velocity needed to migrate the
event correctly (as seen in the second
panel of Fig. 13).

Fig. 15.
Seismic velocities as measured from
well data differ from those measured
directly from the seismic data. Well
velocities tend to be lower.

Distinguished Instructor Short Course ‡ 3-17

 The Seismic Velocity Model as an Interpretation Asset

Fig. 16.
Migration using the well velocities
from Fig. 15 results in undermigra-
tion. Anisotropy is the common cul-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

prit, whereby vertical velocities differ

from horizontal velocities.

Fig. 17.
Larger offsets provide more velocity
resolution. The semblance peaks are
sharper on the spectrum, which
included the farthest offsets out to
6,000 m. Offsets this large will not
move out correctly with the usual
hyperbolic moveout. Prestack depth
migration is one way to solve the
problem, but it requires a velocity
model.

Fig. 18. Before Well #1 After

A schematic view of a velocity model V1 = 5000 V1 = 5000

V´1 =
V´1 = Z´1 =
at a location where well data are pre- Z1 = 500
t1 = 0.200
sent. The well data are in the form of V2 = 6050 V´2 =

depths to formation tops. These tops V2 = 6341.7 V´2 = Z´2 =

Z2 = 2783
need to correspond to horizons in the t2 = 0.920
velocity model and can be used to cal- V3 = 7200 V´3 =

ibrate the seismic velocity model to V3 = 8610.7 V´3 = Z´3 =

use for depth conversion. The calcula-
Z3 = 6012 Horizon A
tion is done in the text. t3 = 1.670 V4 = 7825
Top A
D1 = 6240 V´4 =

V4 = 10,531 V´4 = Z´4 =

Z4 = 8012.8
t4 = 2.050
V5 = 8670 V´5 =

V5 = 12,554 V´5 = Z´5 =

Top B
D2 = 10,130
Z5 = 10,837.5 Horizon B
t5 = 2.500 V6 = 9240 V´6 =

V6 = 12,090 Top C
V´3 = Z´6 =
D3 = 13,050
Z6 = 13,860 Horizon C
t6 = 3.000

3-18 ‡ Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Clarifying the Image

4
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Clarifying the Image

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The original impetus for refining velocity was for imaging when depth migration is
required. In such cases, seismic depth imaging techniques assemble the image through
a complex velocity model, where normal hyperbolic summation techniques do not ade-
quately focus the image. While most of the recent applications have been for subsalt
imaging, these techniques have found more general application for clarifying the seis-
mic image when there are complex velocities in the overburden. These case studies
highlight some interesting situations in which modeling the velocity in detail con-
tributed in large part to the enhancement of the image. In all of these cases, as with all
studies in this course, interpretability was enhanced.
U 3DX Technologies: Clarifying complex faulting

U ARCO: Enhancing interpretability

U Elf Exploration: Improving volumetric calculations

U Oryx Energy: Distinguishing a salt weld from a salt boundary

U Phillips Petroleum: Improving the structural detail of a subsalt play

Distinguished Instructor Short Course ‡ 4-1

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Clarifying Complex Faulting

3DX Technologies
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Phil Schultz

3DX was pursuing prospects in the Frio trend, onshore Texas, where they had shot 3D
seismic. Associated with several major regional listric faults, at about 2.4 to 3.0 s into
the section, were multiple rotated fault blocks causing structurally complex bedding.
This severely distorted zone provides the setting for trap opportunities. However, these
traps were sufficiently difficult to image and were small enough so that clear imaging
and correct placement of events, both spatially and in depth, were critical to maximiz-
ing the effectiveness of each exploratory well (Figs. 1 and 2).
The target of the exploration activity was the second Anomalina Formation within
the lower Frio. This formation was known to be productive in the Frio trend (Fig. 3),
and heavy faulting in the area generated isolated fault blocks with virgin pressure. The
problem was to confidently map the multiple fault blocks and productive zones, since
the expected reservoir compartments would be relatively small and hydraulically isolat-
ed (Figs. 4 and 5).
Complex imaging problems in faulted areas are not always a candidate for depth
imaging. In this case, depth migration was chosen as an approach because the associat-
ed methodology of velocity model-building would ensure a high-quality velocity model.
This situation is seen often. The imaging problem does not necessarily call for prestack
depth migration. Yet, the methodology of the velocity model-building associated with
prestack depth migration is so thorough, that going through it brings value just on this
strength alone.
With this reasoning in mind, 3DX decided that a high-quality velocity model was
their best tool for obtaining the clearest seismic images. 3DX chose to build a 3D veloci-
ty model using the model-building methodology of prestack depth migration, so that
the best velocity model would be available to subsequent imaging.
The velocity model was developed from a layered, horizon-based model, where
three iterations of prestack depth migration into image gather locations allowed succes-
sive updates of the velocity. After monitoring closely the ongoing development of the
velocity model, 3DX decided that three iterations were sufficient to test the image
improvements.
Ten target prestack depth migration lines were produced using the derived velocity
model, and showed a significant improvement in clarity of the fault block images (Figs.
6–9). The same velocity model was also used to perform a poststack depth migration.
The improvement seen in the prestack depth migration was for the most part present in
the poststack migration, as well. Time and cost considerations, coupled with the superi-
or results from the poststack depth migration using the new velocity model, led to a
decision to proceed with full-volume, poststack depth imaging as the primary imaging
approach.

Distinguished Instructor Short Course ‡ 4-3

 The Seismic Velocity Model as an Interpretation Asset

The new image obtained through poststack depth migration showed the fault blocks
with more clarity. While the interpreter’s maps did not change substantially based on
the improved image, he reports being able to convince partners of the validity of his
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

interpretation when shown on the improved image sections, because both the fault
blocks and the reflector continuity within them were noticeably improved.
A well was approved, based on the fault block interpretation and formation map.
The well struck pay, confirming the interpretation.
For depth conversion, a problem arose. The seismic would not tie the well data
below the Anahuac shale. It was determined from analyzing the well data and the final
imaging velocity model, that the Anahuac was a transverse isotropic layer, with differing
vertical and horizontal velocities, causing the depth conversion velocity model to be dif-
ferent from the imaging velocity model. The transverse isotropy exists to a lesser extent
below the Anahuac shale, as well (Fig. 10 and 11).
So, there is a problem using the depth imaging velocity model unaltered for depth
conversion. But, as we saw in the Geco-Prakla study, its ability to provide detail
between wells makes it worth calibrating the model to the well velocities. Recognizing
that calibration is necessary is not normally difficult. Displays of time-depth curves for
the seismic model and the well data at the same map location, such as those shown in
the figures of this study, diverge when anisotropy is present. Such behavior, which is the
rule rather than the exception, indicates the presence of anisotropy and signals the need
for a calibration step.

Fig. 1.
A map view of one of the Anomalina
Formations in the lower Frio trend,
onshore Texas, shows some of the
mapped fault blocks. Many of these
blocks are in hydraulic isolation from
neighbors, and each represents a
potential separate target reservoir.
The objective of this study was to
map one of these fault blocks.

4-4 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 2.
A dip line through the
complex rotated fault
blocks delimits the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

prospective reservoir com-

partments. These images
were obtained from the
poststack depth migration
described in the text.

Fig. 3.
The same dip line as in
Fig 2. is shown without
the fault interpretations
and with the well location
drilled on the prospect.
Commercial hydrocarbons
were found in the Anom-
alina formations of the
lower Frio. However, the
ties between the well
markers and the seismic
were increasingly in error
with depth. The time-
depth curves shown in
the later figures explain
the reason.

Distinguished Instructor Short Course ‡ 4-5

 The Seismic Velocity Model as an Interpretation Asset

Fig. 4.
The same dip line as in
Figs. 2 and 3 is shown
with both fault interpreta-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tions and well markers,

indicating the target fault
block for the exploratory
well.

Fig. 5.
A close-up view of the
fault blocks and the target
reservoir zone.

4-6 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig 6a.
Comparing the same dip line in 6a (the
3D poststack depth migration) and 6b
(the 3D poststack time migration, with-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

out interpretations) shows that the main

differences are the improved clarity of
the definition of the fault blocks and the
improved continuity of the events with-
in the fault blocks. The 3D poststack
depth migration was done with the
velocity model derived in this study,
while the time migration was done
with processing velocities.

Fig. 6b.

Distinguished Instructor Short Course ‡ 4-7

 The Seismic Velocity Model as an Interpretation Asset

Fig. 7a.
A comparison in a strike line of 7a
(the 3D poststack depth migration) and
7b (the 3D poststack time migration)
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

without interpretations shows that the

differences are subtle. The fault struc-
ture is difficult to identify on these
strike lines without the benefit of
multiple dip lines.

Fig. 7b.

4-8 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 8.
An adjacent strike line is
shown with fault interpre-
tations.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 9.
A phase attribute time
slice at 2 496 ms shows
the complexity of the
structure.

Distinguished Instructor Short Course ‡ 4-9

 The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 10.
The interval velocity and time-depth curves for the well data are shown along with the seismic velocity model depict-
ing anisotropy. Horizontal velocities are measured in seismic imaging analysis, and vertical velocities are recorded in
well measurements. The maximum divergence of the slope of the time-depth curves indicates the maximum
anisotropy, which in this case occurs at the Anahuac shale layer.

Fig. 11.
A dip line is displayed on the poststack depth migration in the depth domain with a log curve and the depth markers
from the well. The yellow marker corresponds to the top of the Anahuac shale, and it corresponds reasonably well to
the seismic horizon interpreted as the Anahuac. Just below, the green marker of the Top Frio comes in at a shallower
depth than the seismic horizon corresponding to that formation boundary. This behavior is common, and consistent
with the well velocities being slower than the seismic velocities. The depth of the Frio is too large, based on the seis-
mic velocities, which are too high for depth conversion. As seen from the relative slopes of the time-depth curves in
Fig. 10, the degree of anisotropy depends on the geological layer.

4-10 ‡ Society of Exploration Geophysicists

 Phil Schultz

Enhancing Interpretability
ARCO
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Geologically Constrained 3D Velocity Model Building and Depth

Migration—North Sea Example*
Charles J. Sicking, Bill Kilsdonk and James Sun

Editors Note: A total of 21 figures follow this article and carry informative captions. While
they are not cited in the text, the reader is encouraged to review them carefully to further
illuminate the lessons learned herein.

Summary
Three-dimensional (3D) poststack depth migration of a North Sea survey both
improved the subsurface image and enhanced the structural interpretability of the 3D
data volume. Prior poststack time migration of the 3D volume had produced unsatisfac-
tory results in the vicinity of a graben that runs north-south through the center of the
survey area. The complex structural geometry of the graben made building an accurate
velocity model the most critical component of the depth migration. We constrained and
revised the velocity model by iterating structural analysis, kinematic restoration, and
seismic interpretation with successive intermediate migrations. In addition to constrain-
ing both the present-day structural geometry and the resulting velocity field, the struc-
tural analysis revealed the structural and geometric evolution of the graben. The result-
ing depth-migrated images are superior to existing time-migrated images. Depth migra-
tion increased the resolution, focus, and clarity of seismic reflectors throughout the vol-
ume. Over-migration “smiles” in the time-migrated data are minimized. Most impor-
tantly, subsalt faults are crisper and subsalt structure is better defined. Structures imaged
by the depth-migrated data more accurately reflect true geologic structures and improve
interpretation accuracy and reliability.

Introduction
Structural traps in the area are commonly defined by highs at the Base Zechstein Salt
(Top Rotliegendes) horizon, which overlies the Base Permian Unconformity. Because
traps in this area are usually low-amplitude structures, prospect generation and evalua-
tion depend heavily on the quality of the seismic image, the seismic interpretation, and
the accuracy of time-to-depth conversion. The interpretation of the overlying seismic
velocity field is critical to both imaging and depth conversion. Interpretation of the
velocity field is straightforward throughout most of the volume. However, the structure
of the velocity field is locally complicated by normal faulting and salt diapirism in the
north-south trending graben that runs through the center of the survey. In order to
derive the most accurate velocity model possible, we repeatedly iterated depth migration
with seismic interpretation, structural modeling and restoration, and velocity modeling.
* Published in the Expanded Abstracts, SEG Annual Meeting, Paper INT2.4, Vol. 1, p. 336, 1996.

Distinguished Instructor Short Course ‡ 4-11

 The Seismic Velocity Model as an Interpretation Asset

Velocity Model
Five sets of data were available for this project: (1) seismic processing velocities, (2) the
stacked time volume, (3) a time interpretation of six horizons that was incomplete in
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the area of the graben, (4) horizon depths in wells, and (5) layer velocities from well
and vertical seismic profile (VSP) data. The final model comprised nine velocity layers.
Four depth-migration iterations were required to achieve the final velocity model. After
each migration iteration, the model was revised based on the improved image. The first
iteration migrated down to the Top Chalk horizon using only the velocity of the overly-
ing Tertiary clastics. The Top Chalk horizon is an important velocity boundary. The
velocity above Top Chalk is roughly 2,100 m/s while the velocity below it varies from
2,800 to 3,600 m/s.
Locally, buried bodies of extruded salt exist along the graben axis and the first itera-
tion defined the top of these salt bodies. Additionally, the Top Chalk was picked in this
first depth volume. The velocity of the Chalk layer was then added to the model as
input to the second migration iteration. After the second iteration, the remainder of the
input model for the third iteration was built by first defining the velocity of each of the
remaining layers and then map migrating the time horizons to depth. The complex
structure within the graben could not be defined by this process because of the poor
spatial resolution of map migration.
Using this crude model as input, the third iteration of depth migration was carried
out. A substantially improved image was produced within the graben, especially in areas
of extensive salt diapirism. We interpreted this improved image using constraints im-
posed by structural analysis and restoration. The structural analysis both produced a
working structural interpretation model and locally defined the structure of the graben
and the salt diapir. The velocity model within the graben was updated accordingly and
used as input to the fourth iteration of depth migration. After the fourth iteration, struc-
tural analyses and restorations of 30 cross sections were used to define and constrain
the final velocity volume.

Structural Model
The graben and an underlying salt ridge of laterally varying relief dominate the struc-
ture of the Cenozoic-Mesozoic section. Extension above the Zechstein was localized in
the graben by a process in which conjugate normal faults intersect and offset one anoth-
er. Displacement on intersecting conjugate faults produced both a Mesozoic graben and
a fault bounded salt ridge below it. The graben/salt ridge pair initiated as a necking
instability in reaction to East-West extension. The reactive growth continued until the
diapir locally reached a critical height. At this critical point, the head differential gener-
ated by relief on the diapir was sufficient to drive both isostatic folding and piercement
of the overburden. There are at least three places along the graben where the salt has
pierced its overburden and flowed into the graben.
Serial cross sections were drawn to constrain the structural model through the
Mesozoic graben. The interpretations were based on seismic data and guided by struc-
tural models and restoration. Well-imaged structures in the shallow section were used
to constrain the structure of the poorly-imaged deeper section. Fault offsets clearly

4-12 ‡ Society of Exploration Geophysicists

 Phil Schultz

imaged in the shallow section were carried down to the Top Zechstein. Because the Top
Zechstein must have the same amount of displacement as the shallow section, the height
and location of the salt ridge can be interpreted. Cross section restorations show that the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

total extension along the graben varies from 1 km in the south to 1.5 km in the north.

Conclusions
The final 3D depth migration produced a seismic structural image that is much im-
proved over the 3D time migration. While the primary structural differences are due to
time-to-depth stretch, there are significant differences in the images within and under
the graben. Successful depth imaging of a complex structure depends critically on the
accuracy of the velocity model and should not be treated as conventional “black-box”
processing.
An integrated approach to velocity model building yielded an accurate and inter-
pretable depth image of the 3D seismic data volume and a structural model for the evo-
lution of the Mesozoic graben. Both improve the interpretations for the area. In addition
the structural model contains interpretation concepts that may be applied by explo-
rationists throughout the southern North Sea gas basin.

Acknowledgments
Data and consulting for this project were provided by Marilyn Merryweather and Rob
O’Conner of ARCO British Limited. Thanks are due to ARCO British Limited for per-
mission to publish this project.

Fig. 1.
Outline of the 3D survey is shown in
cyan. Well locations are shown with a
red asterisk. The graben is shown as
a yellow-filled polygon. The two
crosslines indicated are the two
example lines used in the study.
Notice the offsets in the graben. Prior
poststack time migration of the 3D
volume had produced unsatisfactory
results in the vicinity of a graben that
runs north-south through the center
of the survey area.

Distinguished Instructor Short Course ‡ 4-13

 The Seismic Velocity Model as an Interpretation Asset

Fig. 2.
Crossline 680 is indicated on the Top
Chalk structural surface. The Top
Chalk was autopicked. The Chalk
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

deposition became karst from subse-

quent surface exposure. Notice the
karst characteristic along the fracture
zones. The colored surfaces are picked
faults. The graben is apparent and
very pronounced down the center of
the horizon. Note also the fractures in
the southeast corner.

Fig. 3.
The velocity of the Chalk layer is
shown in map view, overlain with the
graben outline and two crosslines.
Note the large range of velocities
within this layer. This extreme lateral
change in velocity suggests using
depth migration to enhance the
imaging.

Fig. 4.
This velocity model was used for for-
ward modeling to generate a zero-off-
set synthetic. The horizons are based
on the cross section of the 3D volume
at crossline 680. The velocity of each
layer is constant, and approximately
the same as the velocities in the 3D
volume. The Tertiary has a velocity of
2000 m/s, the Chalk is 3200 m/s, the
Jurassic is 3150 m/s, the Upper
Triassic is 3750 m/s, the Lower
Triassic is 4200 m/s, and the Salt is
4500 m/s. The synthetic generated
using this model was depth migrated
twice—first using this model and
then using the model of Fig. 5.

4-14 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 5.
The second velocity model is shown
with the Top Chalk horizon
smoothed. This model was used to
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

migrate the zero-offset synthetic,

which was generated using the veloci-
ty model in the previous figure. The
velocity of each layer is the same as in
the previous figure. This illustrates
the imaging problems encountered
when the model used for imaging
does not adequately represent the
earth.

Fig. 6.
The 2D, zero-offset synthetic data
generated from the model is shown.
There has been no smoothing on the
Top Chalk and the data have been
migrated using the correct model.
Note the crisp focusing on the hori-
zons below the graben. The migration
result is good because the velocity
model used for migration exactly
matches the model used in forward
modeling.

Fig. 7.
This is the same image as shown in
Fig. 6 but it has been migrated using
the smoothed model. Note that the
horizons below the graben are not
focused. The smearing is caused by
the inaccurate representation of the
Top Chalk horizon at the graben. This
illustrates the importance of having
an accurate model for horizons that
have a high velocity contrast. It also
shows that imaging through salt bod-
ies is difficult because one must have
a very accurate depth model of Top
Salt and Bottom Salt.

Distinguished Instructor Short Course ‡ 4-15

 The Seismic Velocity Model as an Interpretation Asset

Fig. 8.
The Top Salt structural surface over
the entire survey is shown, along with
the locations of crosslines 680 and
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

2160. Note the large ridge to the

north, the medium ridge toward the
center, and the small ridge to the
south. There are three locations along
the ridges where salt flow into the
graben above the Chalk can be
observed in the 3D volume.

Fig. 9.
This structural model shows five hori-
zons in the northern part of the vol-
ume. Note the Salt flow into the
graben above the Chalk. The salt
ridge has flowed up into the separa-
tion in the Triassic layers.

Fig. 10.
The velocity model-building process
used linked iterative poststack depth
migration with basin model recon-
struction methods. A layer stripping
approach was used. The two Tertiary
layers were built in the first pass. The
Chalk layer was built in the second
pass. The Triassic layers were built in
the third pass. Structural restoration
was used in the third and fourth pass-
es only.

4-16 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 11.
The velocity model evolution of
crossline 680 is shown. Note that the
color scale changes from one iteration
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

to the next. For the first iteration,

there are two Tertiary layers. The rest
of the model is flooded with the lower
Tertiary velocity. For iteration two, the
Top Chalk horizon is defined and the
lower part of the model is flooded
with Chalk velocities. For iteration
three, the Base of Chalk, the Bunter
and the Top Salt horizons are defined.

Fig. 12.
The velocity model evolution of
crossline 2160 is shown. The process
was the same as in Fig. 11.

Fig. 13.
This is a kinematic restoration of the
cross section for crossline 680 down to
the Upper Triassic at the inception of
tectonic deformation. The faulting in
the graben below the Chalk layer is not
focused in the seismic image. In order
to better focus the faulting, it is neces-
sary to have a good velocity model. The
approach to defining the velocity model
below the Chalk layer was first to define
the structure of the layers and then to
carry the velocity of those layers from
outside the graben to inside. The kine-
matic restoration proceeded by moving
the cross section polygons along the
faults, taking the cross section from pre-
sent day to the time of the Top Triassic
and back. By iterating through this
process, a model of the faulting and the
present day configuration of the layers
was obtained.
Distinguished Instructor Short Course ‡ 4-17
 The Seismic Velocity Model as an Interpretation Asset

Fig. 14.
This is a kinematic restoration of the
cross section for crossline 2160. Note
that there is a great deal of horizontal
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

expansion from the time of the Top

Triassic to present. The section exten-
sion varies from 800 to 1,000 m along
the graben from north to south. In
crossline 2160, there is a bright reflec-
tor in the graben above the Chalk,
which can be resolved only by mak-
ing it a salt flow. Trying to make this
reflector fit as a part of the Top Chalk
horizon does not work. This salt flow
can be mapped along the graben in
the 3D volume. Rules for projecting
the amount of throw on the deeper
faults were developed as a result of
these restorations. The rules were
used to help build the Top Salt surface
along the ridge between the restored
cross sections.

Fig. 15.
This is a 3D time poststack migration
for crossline 680. Note the pull-down
in the Base Salt under the graben. The
pull-down is due to the slow veloci-
ties in the Jurassic layer compared to
the faster Triassic velocities adjacent
to the Jurassic graben. There is much
distortion in the reflectors below the
chalk graben. The arrows point to
reflectors that are better focused in
the 3D depth migration section
shown in Fig. 16.

Fig. 16.
This is a 3D poststack depth migra-
tion for crossline 680. Note that the
pull-down in the time section in the
Base Salt under the graben has been
corrected by depth migration. The
focusing of the reflectors in the
graben is improved. There is still
structural distortion just under the
chalk graben.

4-18 ‡ Society of Exploration Geophysicists

 Phil Schultz

Fig. 17.
This is a 3D poststack time migration
for crossline 2160. The salt ridge is
very large in this cross section. All of
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the data in and below the graben are

extremely distorted. The arrows point
to reflectors that are better focused in
the 3D depth migration section
shown in Fig. 18.

Fig. 18.
This is a 3D poststack depth migra-
tion for crossline 2160. The focusing
of the reflectors in the graben is
improved. There is still structural dis-
tortion below the graben. The area of
distortion is significantly decreased
after depth migration.

Fig. 19.
Shown are the differences in the Top
Chalk surface when interpreted from
the time-migrated volume, the map
migration to depth, and from the
depth-migrated volume. The Top
Chalk horizon as picked in the time
volume is well defined in time, but
after map migration the depth hori-
zon is very smooth. If the horizon is
picked in the depth-migrated volume,
it is well defined in depth. In order to
adequately represent this horizon in
depth and to obtain good focusing
below the chalk within the graben,
the Top Chalk must be picked in the
depth-migrated volume after wave
equation depth migration.

Distinguished Instructor Short Course ‡ 4-19

 The Seismic Velocity Model as an Interpretation Asset

Fig. 20.
Shown are the unmigrated
stack and the 3D poststack
time migration compari-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

son for crossline 680. This

figure and the next com-
pare the migration results
for the same line among
the following processes:
stack, 3D poststack time
migration, 3D poststack
depth migration, and 3D
prestack depth migration.
After the model-building
process was completed
and the poststack depth
migration was delivered,
the prestack depth data
were obtained. Using full
aperture input, line 680
was migrated using 3D
prestack depth migration.
As we progress from stack
through to prestack depth
migration, the area of dis-
tortion is decreased.

Fig. 21.
This is a comparison of
the 3D poststack and the
3D prestack depth migra-
tions for crossline 680. In
the northern portion of
the survey, where the salt
ridge is large, modeling
results determined that
the data acquisition geom-
etry would prevent imag-
ing under the ridge. The
data were acquired in a
north-south direction.
Therefore, prestack depth
migration of line 2160 was
not attempted.

4-20 ‡ Society of Exploration Geophysicists

 Phil Schultz

Improving Volumetric Calculations

Elf Exploration UK plc
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Elgin-Franklin Case History: UK North Sea

Michel Goupillot, Claude LaFond, Paul Sexton and Thang Nguyen

Background
Elgin-Franklin is a deep, 5,000 to 6000 m (16,400 to 19,690 ft) Upper Jurassic, high-
temperature, high-pressure gas condensate accumulation, located within the UK Central
Graben of the North Sea (Blocks 22/30c and 29/5b). Elgin field is a complex faulted
anticlinal structure, while Franklin is a complex tilted fault block structure (Fig. 1).
The structures were first interpreted on a 1989 3D seismic survey that was of generally
poor quality, especially in the northwest and southeast portions of Elgin, and the north
and northeast portions of Franklin. Complicating seismic imaging were the deteriora-
tion of data quality below the Base Cretaceous Unconformity, strong anisotropy in the
overburden and a high degree of faulting, especially in the Elgin area.
In order to estimate volumes as well as possible, accurate and detailed structural
imaging of the fields became a main issue: on Franklin this constituted the improved
definition of the main northeastern bounding fault and on Elgin the delineation of the
boundary and intrafield faults. It was decided, therefore, to acquire a new 3D seismic
data set in 1996, which was processed including prestack and poststack time migration
and reinterpreted.
The new seismic brought considerable improvement of data quality and inter-
pretability. The issue of depth positioning of horizons and fault delineation was
addressed by building a 3D depth velocity model followed by 3D poststack depth
migration.

Theoretical Framework
We used in-house software to build the depth velocity model (Fig. 2). The stacking
velocity inversion technique is designed to replace the standard Dix-type inversion
schemes that currently dominate industrial practice. The approximations used in these
standard analytical methods of velocity estimation are well known. Velocity heterogene-
ity, finite offsets and anisotropy all contribute to a systematic mismatch between stack-
ing velocities and the vertical root mean square (RMS) velocity, which are assumed to
be identical. Instead, we use a model-based technique that allows many of these funda-
mental problems to be overcome successfully.
Our inversion is based around two stages of ray tracing (Fig. 3). The first of these
takes the picked interpretations in migrated time, demigrates them and then maps them
into depth, using a hypothetical interval velocity field. The second step checks the
validity of this field by simulating, as fully as possible, the kinematics of seismic acqui-
sition and processing. Inconsistencies between the actual and the modelled data can
then be used to update the interval velocity model using a conventional linear scheme.

Distinguished Instructor Short Course s 4-21

 The Seismic Velocity Model as an Interpretation Asset

The quality of the model can be assessed on four potential criteria:

UÊ / iÊVœ˜ÃˆÃÌi˜VÞʜvÊÌ iʓœ`ii`Ê>˜`Ê>VÌÕ>ÊÃÌ>VŽˆ˜}ÊÛiœVˆÌÞÊvˆi`ð
UÊ / iÊVœ˜ÃˆÃÌi˜VÞʜvÊÌ iʓœ`ii`ʈ˜ÌiÀv>ViÃÊ>˜`ÊÌ iÊ`i«Ì ʓ>ÀŽiÀÃÊ`ivˆ˜i`Ê>ÌÊÌ iÊ
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

wells.
UÊ / iÊVœ˜ÃˆÃÌi˜VÞʜvÊÌ iʓœ`ii`Ê>˜`Ê«ˆVŽi`ÊÜiÊÃiˆÃ“ˆVÊÌÀ>ÛiÌˆ“ið
UÊ / iÊVœ˜ÃˆÃÌi˜VÞʜvÊÌ iʓœ`iÊ«>À>“iÌiÀÃÊÜˆÌ Ê>Ê«ÀˆœÀˆÊŽ˜œÜi`}iÊL>Ãi`ʜ˜ÊÌ iÊ
geology and other information from, for example, the well logs.

The actual stacking velocity field is provided, by the processor, as a set of picks of
stacking velocity against zero-offset traveltime. During conventional processing, the
final stacking velocity analysis is often performed after dip moveout (DMO) and, in
some cases like Elgin-Franklin, after prestack time migration. As these operators can
drastically change the stacking velocity field, their action must be accurately accounted
for during the modelling.
This approach is a powerful aid to interpretation and can significantly reduce the
uncertainty involved in the interpretation process. It has the speed of an interpretation-
based approach, but does not make strong assumptions about the validity of the time-
migration velocity model. This makes interaction between the interpreter and the pro-
cessing geophysicist essential.

Isotropic Depth Imaging

We first performed isotropic stacking velocity inversion to build a 3D depth interval
velocity model. The area under study is 13 by 25 sq km. Six seismic horizons were
interpreted from the time-migrated block: Top Balder, Top Maureen, Top Hod, Top
Hidra, Base Cretaceous Unconformity (BCU) and Top Pentland (Fig. 4). The data quali-
ty for the shallow horizons allowed automatic 3D horizon picking followed by manual
checking. The Top Pentland is heavily faulted and required manual picking. Data quali-
ty below BCU deteriorates significantly due to multiples generated from the Top Hidra
and BCU events. Each time-migrated interpretation was sampled onto a regular grid
with a spacing of 50 by 50 m (164 by 164 ft).
A Kirchhoff demigration operator was applied to the picked interpretations.
Stacking velocity analysis locations were spaced at 500 m (1,640 ft), and were taken
after DMO and prestack time migration where a constant velocity of 2,000 m/s was
used (this is not the final migration velocity). Maximum offset was 4,700 m (15,421 ft)
for the deepest horizons. To build the isotropic model, each layer required three to five
iterations to converge, starting from constant initial velocity guesses. The final layer
took about 30 minutes for five iterations on a workstation. In the end, the modelled
stacking velocities fit the actual ones very well (Fig. 5).
We had six wells available on the area. All the wells have markers defined for each
layer with the exception of well 22/5b-4, which is not deep enough to penetrate the
Pentland. Since we were not modelling anisotropy yet, for this run, we did not use
these wells during the inversion, but later we checked the consistency of horizon
depths. Of course, they did not tie the wells, with misties of up to 600 m (1,969 ft) in

4-22 s Society of Exploration Geophysicists

 Phil Schultz

some areas (10% of the average depth). We did not have walkaway vertical seismic pro-
file (VSP) data available at the time in order to invert first arrivals to further constrain
the velocity model.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The model was used to perform a 3D poststack depth migration that was then verti-
cally stretched back into time using the same velocity model used for time interpreta-
tion. A major problem in these computations was the exact location of the major faults,
which have very large throws. The depth conversion process had not explicitly consid-
ered fault positions, as we were using smoothed maps. In the following study, we decid-
ed to incorporate the faults boundaries (picked in time) into the migration, to obtain
accurate positioning in depth. We obtained more meaningful depth maps with this last
method (Fig. 6).

Anisotropic Depth Imaging

We then built an anisotropic depth velocity model for comparison. Anisotropic inver-
sion is five times slower than the isotropic one, due to the ray tracing. In the resulting
model, depth surfaces tied the well markers within 10 m (33 ft). The inverted
anisotropy parameters (Thomsen model) were generally very large except in the Hod-
Hidra interval. Interestingly, this interval is a homogeneous chalk layer with the gamma
rays logs showing low shale content. Conversely, the other layers with large inverted
anisotropy parameters contain thick shale beds.
It should be noted that in this case, the trade-off between epsilon and delta cannot
be resolved. We would need more structured horizons or well seismic information with
large aperture to resolve this trade-off. Techniques were developed to quantify the reso-
lution between the inverted parameters, and different realisations were obtained using
the various models that explained the input data equally well (Fig. 7).
We did not perform anisotropic depth migration at that time, but gross rock volume
(GRV) calculations from the depth maps indicated volume changes compared to the
volumes derived from the isotropic study. It is important to note that this is a matter
not just of vertical differences, but also of differences in lateral positioning: the maps of
vertical differences (Fig. 8) and of displacement vectors (Figs. 9 and 10) show that lat-
eral displacement can be up to 500 m (1,640 ft) in dipping areas. The discrepancy
results in a gross rock volume difference of almost 8% in Franklin.
When it was decided that anisotropy played a major part in volumetric calculations,
by better imaging of dipping reflectors, we then had the available technology to perform
anisotropic 3D poststack depth migration. We had two more wells available then, and
we used that extra information to build a revised, anisotropic depth velocity model
down to the BCU, followed by a 3D poststack anisotropic depth migration. The BCU
horizon was repicked on the depth volume and inserted into the velocity model instead
of the map-migrated horizon. The Top Pentland was inverted, followed by another iter-
ation of 3D anisotropic poststack depth migration.
The improvement on the result of the final image is significant because now the
faults are better imaged and the structural image has changed the gross rock volume
estimations, all of which allow reliable computation of field reserves.

Distinguished Instructor Short Course s 4-23

 The Seismic Velocity Model as an Interpretation Asset

Fig. 1.
The Elgin-Franklin base map shows
major fault patterns.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 2.
This screenshot from the Superdix
(velocity model-builder) software
interface indicates the layer-stripping
flow and shows the forward model-
ling rays.

Fig. 3.
This is an outline of the inversion
method.

4-24 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 4.
These are the Elgin-Franklin horizons
used for the inversion. The depth of
the Pentland is between 5,000 m and
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

8,000 m.

Fig. 5.
This is a comparison of the modelled
and actual stacking velocities, with
the corresponding depth maps, for the
BCU and Pentland, for the isotropic
model. Note that, despite strong varia-
tions in the velocities, we are able to
recover most of the details.

Fig. 6.
This depth map of the Pentland is
from the anisotropic model. Note the
definition of the faults, which were
depth-migrated from their time-
migration position.

Distinguished Instructor Short Course s 4-25

 The Seismic Velocity Model as an Interpretation Asset

Fig. 7.
This is a comparison of the modelled
and actual stacking velocities, with
the corresponding depth maps, for the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

BCU and Pentland, for the anisotropic

model.

Fig. 8.
This illustrates the vertical difference
(meters) between the Pentland depth
maps obtained by isotropic and
anisotropic inversion. This difference
can reach more than 200 m on dip-
ping reflectors.

Fig. 9.
The point migration vectors from
time to depth, in map view, show the
lateral difference of the locations of
depth points when migrated with the
isotropic and anisotropic models. The
change of well locations in the differ-
ent domains (Pentland map over
Elgin) are also illustrated.

4-26 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 10.
The point migration vectors from
time to depth, in map view, show the
lateral difference of the locations of
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

depth points when migrated with the

isotropic and anisotropic models.
The change of well locations in the
different domains (Pentland map over
Franklin) are also illustrated.

Distinguished Instructor Short Course s 4-27

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Distinguishing a Salt Weld From a Salt Boundary

Oryx Energy
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Velocity Modeling For Subsalt Prospecting*

Phil Schultz

Oryx was prospecting in the Gulf of Mexico subsalt, where deep targets (about 5 km)
require costly drilling programs (Fig. 1). To minimize risk team members decided to
process a 3D seismic data set through prestack depth migration with a velocity model
built from iterative updating of image gathers. As usual in such cases, both geological
and geophysical information was required to build a competent velocity model (Fig. 2).
The interest of this case study to the course is the redefinition of the geological
interpretation, based on the results of the velocity modeling and updating. What was
originally interpreted as a monolithic salt body based on the best 3D imaging data avail-
able at the time was altered to a pair of salt welds, with sediment lying between them.
This altered interpretation was due in large part to the quantitative velocity analysis
involved in updating the velocity model during the iterative velocity model-building
process.

Velocity Analysis
Velocity analysis proceeded in two phases. For the initial model, a conventional velocity
analysis, modified to include velocity gradients and combined with constraints from
well logs and check shot data, was used on a grid of common midpoint (CMP) gathers.
The gathers were spaced every 60 CMPs on nine, equally spaced shot lines that spanned
the survey. Points on four dominant interfaces (including the water bottom) were then
map migrated to produce a 3D, four-layer model for the shallow section above 2.0 s.
With the velocities in the shallow layers fixed, velocities at greater depths were esti-
mated iteratively by attempting to flatten reflections on image gathers obtained by
prestack depth migration. Three iterations were performed. The final 3D migration used
a smoothed version of the updated model predicted by the third iteration.
The data were acquired over an 8.75 km = 15.21 km area in a dual-cable, dual-
source marine survey. Source and receiver spacings were 26.67 m in the inline direction,
with minimum and maximum offsets of 121 m and 3,284 m. A total of 43 dual lines
were shot with approximately 540 sources per line. The total recording time was 7.5 s
at an interval of 4 ms, producing 47 GB of data with more than 6 million traces and fill-
ing 28 8-mm tapes.
The typical common source gather from the dual cable, Fig. 3, shows the character-
istics of the data. The signal-to-noise ratio is sufficiently high that reflections from sedi-
ments and dipping salt features are clearly visible.
The data were binned into 174 lines, each with 1,140 midpoints. Data preproces-
sing included only gain recovery, deconvolution and first break muting. Thus, the true
* Adapted from Chang, et al, GEOPHYSICS, v. 63, no. 2, p 546-556.

Distinguished Instructor Short Course s 4-29

 The Seismic Velocity Model as an Interpretation Asset

amplitudes were well preserved. Well GB260#1, located about 1.6 km south of the cen-
ter of the survey, was used to identify and constrain depths of shallow events (using
check shot data) as well as deeper events (from log data).
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The intial horizon-based velocity model was constructed by stacking data across the
dual cables for each shot. Four shallow horizons were then identified and coherency
inversion was performed to obtain interval velocities for those layers.
After the initial velocity model building, iterative refinements were produced by
three cycles in which prestack depth migration alternated with image gather velocity
analysis. For the first two iterations, the common image data used were at points on a
998 = 805 m grid. This corresponded to every 20th bin line and every 16th midpoint in
the inline direction.
For the third and final iteration, the spacing in the center half of the survey was
decreased to every 10th bin line and every 30th midpoint. For all iterations, 30 offsets
were used for velocity analysis.
Fig. 4 shows representative results for one common image gather after three itera-
tions. The reflections have been flattened and the semblance plot has most of its energy
down the center line. This indicates that the process has nearly converged to the super-
imposed velocity profile, shown as the white line. Results from the other gathers were
similar.
Often a preconceived geologic framework drives the interpretation of the velocity
field, possibly leading to geologically incorrect conclusions. Fig. 4 shows two main
advantages of image-based velocity estimation. The two strong reflections at 4.27 km
and 5.48 km depth were originally believed to be the top and base of a salt body, but
the interval velocity between the reflections (Fig. 4, right panel) is too low to be salt.
Therefore, the interpretation is not supported by the data. The structure is apparently
two thin salt welds with sediment between them.
If a salt velocity were inserted in this interval, the data would be over-migrated,
because the velocity would be too high and the horizons would be distorted in depth.
The second advantage of image-based processing, which also applies to other sedimen-
tary basins similar to the Gulf of Mexico, is that the estimated velocity functions do not
have restrictions of constant velocity or constant gradients within the layers.
The final migration was performed using the updated velocity model predicted by
the third iteration. The corresponding 3D velocity volume was constructed by linear
interpolation between the estimated velocity-depth functions. Fig. 5 shows the inline
slice through the migration velocity model at the same location as the images in Figs. 4
and 6.
Comparing the results of 3D prestack depth migration with 3D poststack finite dif-
ference time migration (Fig. 6–9), the depth-migrated image shows a significant
improvement in clarity of salt weld reflections as well as all steeply dipping events. The
depth section was converted to time (using the 3D interval velocity field) to allow this
comparison and to facilitate use of interpretation workstation systems, which at the
time of the study did not readily support depth-domain seismic.
The GA5 gas sand, which is a classic Gulf Coast bright spot, shows a constant AVO
response (Fig. 7), which is consistent with amplitudes calculated from petrophysical
data. Poisson’s ratio decreases as gas saturation increases, which in turn increases the

4-30 s Society of Exploration Geophysicists

 Phil Schultz

magnitude of the negative polarity reflection as a function of incident angle and gives
the net flat AVO response.
One of the main questions that prompted 3D prestack depth migration for the data
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

was the lateral extent of the GA5 sand. Fig. 7 suggests that the GA5 gas sand truncates
cleanly at a salt weld.
Another important result is the discovery, in the prestack depth migration image, of
a second, shallower gas sand (Fig. 7) that was not previously known. This one has a
normal AVO response for a gas sand, with amplitude increasing with offset. This discov-
ery had an impact on platform location.

Fig. 1.
This stacked section is from a subsalt
play in the Gulf of Mexico. Velocity
model-building began with the inter-
pretation of horizons in the sedimen-
tary overlying sediments.

Fig. 2.
The 3D velocity model was com-
prised of structural surfaces and
velocities defined within the associat-
ed layering.

Distinguished Instructor Short Course s 4-31

 The Seismic Velocity Model as an Interpretation Asset

Fig. 3.
In this sample common
source gather the two
groups of traces corre-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

spond to the two halves of

the dual recording cable.
There are 120 receiver
groups in each half. The
labeled reflections are
from the shallowest salt
weld, as seen in the profile
of Fig. 4.

Fig. 4.
Shown are the representative velocity estimates at iteration 3. The left panel shows the depth-migrated, inline slice,
where the vertical line identifies the location of the velocity analysis of the right panel. In the semblance plot, the
brown line shows the depth error predicted by the pick curve. The white and pick lines are the current and updated
velocity profiles, respectively. Convergence is indicated both by the focusing of energy along the vertical (zero devia-
tion) center line and by the small differences between the two velocity models. The scale along the bottom indicates
the depth error in meters; the interval velocity is in km/s. The velocity distribution on this section is shown in Fig. 5.

4-32 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 5.
This representative vertical slice
through the 3D interval velocity
model used for prestack depth migra-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tion has the same location as those in

Figs. 4 and 6.

Fig. 6.
Shown are representative vertical slices through the output migrated volumes. The left panel is from the 3D prestack
depth migration, and the right panel is from a poststack 3D time migration. For comparison purposes, both sections
are portrayed in the time domain.

Distinguished Instructor Short Course s 4-33

 The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 7.
Shown are post-3D depth migration AVO data for the GA5 gas sand. The left panel is the final stacked image for one
slice through the 3D prestack depth migration volume. The five right panels are the images obtained by preststack
depth migration from data for different offsets for that part of the section marked by vertical lines in the left-hand
panel. The constant offset images are 5 of the 30 that are stacked to obtain the image at the left.

Fig. 8.
Shown are representative constant time slices through the 3D prestack depth migration volume. The number above
each panel is the corresponding time in ms after depth-to-time conversion.

4-34 s Society of Exploration Geophysicists

 Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 9.
Shown are representative intersecting constant time and vertical slices through the output prestack depth migration
volume after depth-to-time conversion. The heavy, near-vertical yellow line in each panel is an existing well to the GA5
gas sand.

Distinguished Instructor Short Course s 4-35

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Improving the Structural Detail of a Subsalt Play

Phillips Petroleum Co.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Advancing Technology and the Ship Shoal 32 Field: Surface Seeps to

3D Prestack Depth Migration*
Craig S. Hartline, Kay Dautenhahn Wyatt, Paul A. Valasek,
Scott A. Irvine, David A. Walker, Robert R. Sartain

Summary
“Because much of the more accessible geological territory in presently known producing
regions has been intensively explored, the search for oil is being extended to more
remote areas and deeper regions.” This statement, which could just as easily be made
today, was extracted from Phillips Petroleum Company’s 1947 Annual Report. This
report’s date of publication coincides with the offshore discovery of the Ship Shoal 32
Field. The 1947 discovery well, 10.5 miles off the coast of Terrebone Parish, Louisiana,
was the first well drilled out of site of land in the world. This paper will review the histo-
ry of the field, with an emphasis on the advancing technology which has been utilized in
developing the salt-related hydrocarbon accumulations. This study will conclude with a
discussion of 3D velocity model building and 3D prestack depth migration imaging
recently completed for this area of complex structure and highly variable seismic velocity.

Introduction
The Ship Shoal 32 Field in the U.S. Gulf of Mexico has been producing hydrocarbons
since 1947, just two years after the end of World War II. The technologies involved in
the exploration and development of hydrocarbons have changed greatly over the past
51 years. This paper discusses the history of the Ship Shoal 32 field, focusing on the
major advances which have helped to more clearly define drilling targets.

History
In 1947, most of the prospective U.S. acreage onshore and nearshore had been acquired
by major oil companies. This forced a committment to financially and geologically risky
offshore exploration programs in the U.S. Gulf of Mexico. The result of this in the fol-
lowing years was an industry 25% wildcat success rate offshore, compared with 10%
onshore. Many geologists thought that hydrocarbon accumulations “pinched out” with
increasing water depth. This has proved to be far from the truth. The placement of the
Ship Shoal 32 Field’s discovery well was based on two sources of information. First, U.S.
Navy depth soundings detected a sea floor high, suggesting the potential for faulting/
trapping. Also, oil seeps were noted on the sea surface above the presently defined salt
structure (Fig. 1).

* Presented at the September 1998 SEG Annual Meeting.

Distinguished Instructor Short Course s 4-37

 The Seismic Velocity Model as an Interpretation Asset

Nearly 90 wells have been drilled within Ship Shoal blocks 30, 31, 32 and 33, with
much success prior to the availability of seismic data. Well placement at that time was
simply step-out drilling based on mapping of the existing well information. Due to shal-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

low water depths (averaging 18 ft) and surface obstructions, 2D seismic data was not
acquired over this field until 1973. However, 2D seismic data of 13 different vintages
was available by 1985, assisting in the drilling programs.
Acquisition of 3D seismic data over the Ship Shoal 32 salt structure began in 1988
with a 3D bay cable survey. A high-resolution survey was needed to explore for traps
underneath salt overhangs surrounding the structure. A 1988 3D time-migrated volume
and a 1991 3D poststack depth migration (PoSDM) were produced from this data.
Drilling targets located using the PoSDM imaging resulted in four successful gas wells,
adding significantly to reserves.
In 1996, Phillips purchased 14 blocks of ocean bottom cable (OBC) 3D seismic
data. This purchase was made with the hope for better illumination of reservoir targets
beneath salt overhangs with the OBC acquisition technique. A recent 3D prestack depth
migration (PSDM) has utilized the 3D OBC seismic data.
And what of the use of gravity methods for salt body definition? Subsea sediments
above the “cross-over zone” contain enough water to actually be less dense than salt.
This increases the problems with nonuniqueness inherent in gravity modeling.
Although some recent work has been done to confirm the 3D PSDM depth/velocity
model with a density conversion using Gardner’s equation, gravity methods were of lim-
ited use in well placement for this field.

3D Velocity Model Building/3D Seismic Imaging

The velocity model used in the 1990 poststack depth migration was derived mainly
from sonic logs, checkshot data and converted stacking velocities. The imaging
improvements over the 3D time-migrated data were significant and assisted in the
placement of four successful gas wells.
A technique developed by Phillips’ geoscientists, horizon-based prestack depth
migration (HPSDM), was used to provide an accurate depth/velocity model for a recent
3D prestack depth migration.1,2 This layer-stripping technique prestack depth migrates a
limited window of data centered about an interpreted seismic horizon. This allows rapid
generation of common image gathers on a fine grid. Residual velocity corrections are
made by comparing the imaged offset volumes at each analysis location. In theory, a
common image gather should image a reflective interface at the same depth for all off-
sets when the velocity model is correct. The basic steps of the HPSDM modeling tech-
nique are:
1) Determine an initial depth/velocity model with N layers. The velocity for the first N-
1 layers are assumed to be accurate for this layer-stripping method.
2) Create the depth surface that will form the “imaging horizon.” This can be derived
from a poststack depth migration completed with the initial depth/velocity model,
or a time-to-depth conversion of a surface interpreted on time-migrated data.
3) Calculate traveltimes for a finely spaced grid coincident with the imaging horizon.

4-38 s Society of Exploration Geophysicists

 Phil Schultz

4) Perform 3D HPSDM for a window of data surrounding the imaging horizon.

Traveltimes are calculated only for the imaging horizon, and a technique known as
“block moveout” allows computationally efficient imaging with sufficient accuracy.1,2
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

5) Analyze the PSDM offset volumes using common image gathers, flattened represen-
tations of the offset volumes, residual curvature grids, coherence grid, and sem-
blance grids to aid in the estimation of residual velocity corrections. Interpretation
of a composite event (summed along the calculated residual curvature) allows for a
“structural correction” to be applied.
6) Update the depth/velocity model. Adjustments are made to the velocity and depth
values of the layer under analysis. Iterations are carried out on steps 2 through 6
until velocity errors are within an acceptable limit.

After determination of the velocity model down to the level of the top salt, orthogo-
nal 3D PSDM “saltflood” swaths were used to define the base salt structure. Advanced
3D visualization utilities assisted greatly in interpreting the base salt event.3 With the
insertion of the salt structure into the depth/velocity model, the subsalt velocities were
then determined with the same 3D HPSDM layer-stripping approach (Fig. 2). Of
course, the velocity model is compared with other existing velocity information (sonic
logs, checkshots, etc.) throughout the model-building process. The Kirchhoff imaging
of the full 3D PSDM volume was completed using a massively parallel supercomputer.
This recently completed 3D prestack depth migration has produced a significant
improvement in the imaging of the base salt and surrounding sediments (Fig. 3).
As always, there is room for improvement. Some inconsistencies between the depths
obtained from the PSDM imaging and the available well control are thought to be
related to anisotropy in the clastic sedimentary units.
Studies are under way to compensate for the effects in both the velocity model
building and 3D PSDM imaging.

Gravity Modeling
Three-dimensional gravity modeling was completed after determination of the 3D
depth/velocity model using the HPSDM techniques mentioned above. Due to limita-
tions related to the density cross-over zone, this was done merely to confirm that good
agreement existed between observed and calculated gravity values. A poor agreement
could suggest an error in the interpretation of the salt structure (there were some areas
of uncertainty, mainly for the base salt event). Modeling was done using LCT’s 3MOD
potential field modeling software.
The velocity model used for the 3D PSDM imaging consisted of a series of depth
and velocity grids. The depth grids were imported directly to the modeling software.
The velocity grids for the clastic section were converted to density grids using Gardner’s
equation {density = 0.23(velocity)**0.25}. Velocity gradients within layers were con-
verted to density gradients by determining the best fit to calculated densities, since the
relationship is not linear. The resulting depth/density model’s calculated gravity values
agreed well with the observed gravity values (Fig. 4).

Distinguished Instructor Short Course s 4-39

 The Seismic Velocity Model as an Interpretation Asset

Conclusions
The advancing technology utilized throughout the Ship Shoal 32 Field’s existence paral-
lels the achievements made by the petroleum industry as a whole. In 51 years, explo-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ration and development for this field has advanced from detecting seafloor anomalies
and oil seeps to 3D prestack depth migration.
Recent advances in velocity model building and 3D seismic depth imaging tech-
niques have produced well-focused imaging in areas of complex structure and highly
variable seismic velocity. This aids in the definition of drilling targets within these diffi-
cult imaging areas, allowing for new field discoveries and continued production from
older fields.
The 3D gravity modeling work, as well as sonic logs, checkshots and converted
stacking velocities, give additional confidence in the correctness of the depth/velocity
model. Success in the search for hydrocarbons (or anything else) has always required
that risks be taken. The knowledge gained from those risks can then be applied to
future work. We should remember that the level of technology we utilize today has
been built on the efforts of many people, and over a long span of time.

Acknowledgements
The authors would like to acknowledge the pioneering efforts of those at Phillips,
AMOCO and Kerr-McGee who were responsible for the discovery, development, and
technological advances associated with this field over its 51-year existence. We also
appreciate the contributions of Tim McGinty, Susan Towe and Brian Little, regarding
their early 3D interpretive work on this field. The graphical applications used to inter-
pret, analyze and display the PSDM data were developed by Jackie Pope, Jesse Layton,
Cherie Skillern, Steve Wyatt, Ross Snider, Manita Forney and Ed Butler. Processing
assistance was provided by Dan Wieder, Bob Olson, Keith Branham, Bob Heaton and
Brackin Smith. Carl Cook, Jerold Barnett and Warren Neff provided assistance with the
3D PSDM imaging. Gravity modeling advice was given by Reve Hammons. We also
wish to express our appreciation to Phillips Petroleum Co. for the release of this materi-
al for publication.

References
1
Wyatt, Kay D., Valasek, P. A., Wyatt, S. B., 1995, Rapid Velocity Estimation Using
Horizon-Based 3D PreStack Depth Migration, presented at the Research Workshop.
2
Wyatt, Kay D., Valasek, P. A., Wyatt, S. B., R. M. Heaton, 1997, Velocity and
Illumination Studies from Horizon-based PSDM, 67th Annual Meeting, SEG,
Expanded Abstracts, pp. 1801-1804.
3
Wyatt, K. D., Wyatt, S. B., Towe, S. K., Layton, J. E., von Seggern, D. H. and
Brockmeier, C. A., 1994, Building velocity-depth models for 3D depth migration: The
Leading Edge, no. 8, pp. 862-866.

4-40 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 1.
These photographs are of early off-
shore exploration scenes in the Gulf
of Mexico, taken from Phillips
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Petroleum Company’s 1947 Annual

Report.

Fig. 2.
Three-dimensional visual-
ization of the Ship Shoal
32 Field’s interpreted salt
structure. Translucent
PSDM seismic data and a
deeper velocity boundary
are also shown.

Distinguished Instructor Short Course s 4-41

 The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 3.
Progressing from left to right, the advances in 3D seismic imaging are shown. Left: 3D time migration (1988). Center:
3D poststack depth migration (1991). Right: 3D prestack depth migration (1996).

Fig. 4.
Slices extracted from 3D gravity modeling. Display units are in ft. Gravity values are in milligals.

4-42 s Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Seeing Below Gas Seepage

5
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Seeing Below Gas Seepage

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Gas seepage zones cause a major velocity distortion effect because of their anomalously
low velocities. Artificial structures are created under them. While this effect is not a new
problem to interpreters, it has always been a challenge to define the extent of the veloci-
ty distortion.
Since it is generally not possible to image the zone of gas seepage directly, it
becomes necessary to infer its structure via indirect means, by estimating the velocity
from focusing analysis of layers above, below and within the gas seepage zone. Why
bother? In one of Nature’s more perverse tricks, hydrocarbon deposits of commercial
interest are commonly found in the vicinity of gas seepages, often below them.
One associate of mine claims, only partially tongue-in-cheek, that companies would
find more oil simply by drilling in bad data areas. He has a point. Bad data areas are
often associated with gas seepage. Here are two examples in which that problem was
tackled head-on.

U Mobil: Detailing structural continuity

U Spirit Energy: Enhancing the estimate of structural closure below gas

Distinguished Instructor Short Course s 5-1

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Detailing Structural Continuity

Mobil Producing Nigeria
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

J. Reilly and C. E. Edoziem

While exploring a prospect on the Nigerian shelf (Fig. 1), Mobil faced a problem—the
seismic sections were poorly imaged where normal faults penetrated to the near surface.
This behavior had been seen before, in fact, it was common in the area, but Mobil want-
ed to know why.
After accumulating data and experience from several development drilling programs
on the Nigerian shelf, numerous causes for these effects have been proposed, including:
U Velocity contrasts across faults due to changes in lithology or pressure regime,
U Shallow near-surface, low-velocity anomalies related to gas seepage, or
U Gas migration along fault planes.
Imaging within these features is often very sensitive to streamer geometry and
shooting direction. Quite often, a priori knowledge of the cause of the fault shadow is
required to significantly improve time or depth imaging and structural mapping.
Unfortunately, unless the cause of each specific imaging problem has been penetrated
by a wellbore and adequately logged, this knowledge is not available.
So, solving Mobil’s problem required an interpretation of the cause of the velocity
anomaly in order to set the constraints properly for an effective model. In this case,
interpretation input is pivotal to the effective construction of the velocity model.
Prospect mapping was initially based upon a 2D seismic grid (Fig. 2). The prospect
was identified within a rotated normal fault block. The objective section was the Intra
Biafra sands. Quality of the 2D seismic data was generally good immediately adjacent to
the fault plane and outside of the fault zone. However, a poor-quality imaging “shadow”
projected downward under the fault plane. This effect gave a characteristic triangular
bad data zone within a setting of basically good data.
Based on 2D data, exploration well W01 was drilled in 1991. This well encountered
310 ft of oil and 194 ft of gas pay from four reservoir intervals in the target sands. W01
was drilled directly into the shadow zone of the fault plane, which was imaged reason-
ably well on the 2D data.
In 1992, appraisal well W02 was drilled in a more down-dip position to test the
extent of the field. This well avoided the shadow zone by drilling outside (down dip) of
the affected area and encountered 54 ft of oil pay in the target sands.
In 1993, following successful completion of these discoveries, a 3D seismic program
was conducted to delineate the structure and reservoir for production drilling purposes.
One of the initial expectations was that the imaging problem would be resolved during
3D processing.
Due to operational constraints, the survey was shot in a NE-SW orientation, essen-
tially parallel to the main fault plane. This acquisition program contrasted with the 2D
data set, where the best images in the vicinity of the faults were obtained from lines
shot perpendicular to the local strike of the fault (i.e., NW-SE).

Distinguished Instructor Short Course s 5-3

 The Seismic Velocity Model as an Interpretation Asset

The images generated from the 3D program showed significantly improved imaging
over the majority of the area. However, in the vicinity of the faults the data appeared
more poorly imaged than in the 2D data (Fig. 3). The 3D data displayed what appeared
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

to be extensive overmigration and time sag effects in the fault shadow zone (Fig. 4). In
addition, the seismic traveltimes did not appear to coincide with well-based, time-depth
information.
An immediate problem in the 3D data was the interpretation of the time discontinu-
ity observed to the southeast of the main fault. Was it a true structural feature, or was it
an artifact of the behavior of the velocities above? If truly a fault, this feature could
define multiple reservoir compartments and affect the drilling program.
Sonic and checkshot well data were examined in order to determine if gas seepage
was indeed the actual cause of the fault shadow imaging problem. Unfortunately, as a
result of the deviated well path and difficulties in obtaining wireline and check shot
data in the very shallow section, the well data did not provide a unique answer. In addi-
tion, there was a complete lack of information about the down-thrown section.
Based on experience gained in 2D prestack depth migration projects conducted in
the area, two possible velocity model types were tested on crossline extracted gathers,
using 2D prestack depth migration. The more conventional model assumed that there
was a velocity contrast across the fault, due either to a pressure differential or lithology.
The alternate model consisted of a narrow, low-velocity chimney, centered on the fault,
due to gas seepage in the vicinity of the fault (Figs. 5 and 6).
Either model design proved effective when used in a subsequent model-building
process to flatten the image gathers, by iterative update of the model within the con-
straints imposed by either model. That is, the model-building process inherent in the
depth-imaging methodology was not able, on its own merit, to discriminate between
the two likely model styles.
The original root mean square (RMS) velocity field from the original time process-
ing did not “see” any indication of anomalous behavior in the vicinity of the faults,
which could possibly have been used to discriminate between the model options. The
prospective velocity model options themselves came about as an inference from local
knowledge of the area’s geology, rather than anything directly observed in the surface or
downhole data.
The result of the depth analysis was a depth image with subspatial false depth pull-
up beneath the fault plane, as confirmed by the well data. Furthermore, the best images
were obtained when only the far offset data were included in the depth imaging. The
alternate velocity, which assumed a very narrow (less than 0.5 km), low-velocity zone
on the up-thrown side of the fault, produced the most consistent depth model, especial-
ly when only the near offset information was used in the image.
Re-examination of the well check shot data revealed low-velocity anomalies in the
range of 200–400 m/s in association with the penetrated gas zones. Based on this infor-
mation, it was assumed that the fault shadow in this case was caused by a gas chimney,
rather than the standard velocity contrast across the fault (Fig. 7).
A problem common to both time- and depth-domain velocity analysis is that the
velocity anomaly was, at 0.5 km, much smaller than the full 4.0 km range of offsets in
the data. Even when the mute zone of the data was considered, the anomaly still

5-4 s Society of Exploration Geophysicists

 Phil Schultz

extended only over less than 25% of the effective spread. Because it is extremely diffi-
cult and error-prone to infer the details of such a fine lateral velocity structure without
some interpretive input for initial constraints, it was appreciated that a highly interpre-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tive velocity model-building process is an essential part of both the depth imaging and
subsequent depth conversion.
As it turned out, prestack time migration generated an image that matched the
coherency of an image generated by depth migration, but it cost much less and was gen-
erated more quickly. Accordingly, it was the chosen approach for the bulk imaging stage,
followed by depth conversion using the velocity model, as derived from key line depth
imaging with the interpretive constraints. The prestack time migration processing strate-
gy included constant velocity migration, velocity analysis, stack, demigration and remi-
gration with a model-based velocity field. The final remigrated image in the time domain
was then converted to depth using the velocity field based on the chimney model.
Interestingly, the different behavior of the far and near offset volumes on the
prestack time migration, as seen clearly on key line image sections, reinforced the inter-
pretation of a narrow zone of low velocity. The far offset images behaved more like an
undershoot (Figs. 8–12).
An important lesson learned from this study is that with localized velocity anom-
alies, interpretive input is sometimes needed when choosing from among the various
velocity model styles. Any one of these candidates, when used as constraints for model
updating, could result in flattened image gathers and achieve similar image clarity. The
resulting depth structure, however, would be essentially accurate only for that velocity
model that best matched the geology.

Fig. 1.
This map shows the reservoir area on
the up-thrown side of the major fault.
The surface projection of the fault
plane is traced on the map. It reveals
that the secondary break is between
the two and potentially caused by a
shadow effect of the velocity contrast
above it. As it turned out, this sec-
ondary feature was a phantom fault.

Distinguished Instructor Short Course s 5-5

 The Seismic Velocity Model as an Interpretation Asset

Fig. 2.
The original 2D data shows the imag-
ing of the reservoir event near 2.0 s
just under the fault. This imaging was
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

sufficient to warrant drilling discovery

well W01 and subsequently acquiring
a 3D survey.

Fig. 3.
This 3D image is of the same location
as shown in Fig. 2. It is a crossline
and a strike line (see Fig. 1). A signifi-
cant degradation of the reflector con-
tinuity is seen under the fault,
although data quality elsewhere is
improved over the 2D data. In addi-
tion, artificial anticlinal or rollover
features begin to appear near the fault
below 2.0 s.

Fig. 4.
Image gathers from prestack time
migration marching up to and across
the fault shows that events are flat
until they reach the fault, and then
show anomalous behavior. The appar-
ent velocity correction is to increase
the velocity locally to flatten the
event, but that is exactly the opposite
of the local behavior, which occurs
because the local velocity anomaly is
much smaller than the spread length.

5-6 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 5.
The velocity model shows a common
behavior across faults, under which
imaging problems exist. The velocity
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

exhibits a transition across the fault.

This was the first model tried in the
model-building exercise.

Fig. 6.
The chimney model was tried in a
second attempt to improve the poorly
imaged data under the fault. While
both models, by suitable adjustment
of velocities within them, could be
made to flatten the image gathers,
only this model showed images con-
sistent with the well data.

Fig. 7a.
The VSP data from well W01 was
examined for evidence of low-velocity
zones above the zone of the fault. The
blue curve is from well W01, the well
nearer the fault, and the magenta
curve is from W02. Although the
VSP examines only a localized zone
near the well, there are portions of the
curve that appear to behave as if it is
sensing localized low velocities, such
as at 1200 ms and 1700 ms.

Distinguished Instructor Short Course s 5-7

 The Seismic Velocity Model as an Interpretation Asset

Fig. 7b.
The VSP data from well W01 was
examined for evidence of low-velocity
zones above the zone of the fault. The
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

blue curve is from well W01, the well

nearer the fault, and the magenta
curve is from W02. Although the
VSP examines only a localized zone
near the well, there are portions of the
curve that appear to behave as if it is
sensing localized low velocities, such
as at 1200 ms and 1700 ms.

Fig. 8.
The 3D crossline section, in which
only the far offsets were used, and the
chimney model were used for the
prestack migration. The reflector con-
tinuity is acceptable and interpretable.

Fig. 9.
This is the same 3D crossline section
as shown in Fig. 8, but only the near
offsets were used in the image. The
degradation is clear by comparison
and suggests that the velocity anom-
aly is indeed narrowly limited because
it appears that the far offsets are par-
tially undershooting the low-velocity
zone, in spite of the acquisition direc-
tion being along strike.

5-8 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 10.
A close-up version of Fig. 8 is shown.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 11.
A close-up version of Fig. 9 is shown.

Fig. 12.
This poststack depth migration of the
3D crossline was created using the
alternate chimney velocity model.
Continuity of the event at about 2.0 s
(red line) is reasonably monoclinal,
which was expected. Artificial rollover
is still seen on the deeper horizon just
under the fault.

Distinguished Instructor Short Course s 5-9

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Enhancing the Estimate of Structural Closure Below Gas

Spirit Energy
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Phil Schultz

Spirit Energy was evaluating a mature zone in the Gulf of Mexico for placement of addi-
tional development wells. Specifically, should a deeper, higher-risk well be drilled into a
known gas producing area (Fig. 1)? The problem was that the depth structure was dis-
torted on the 3D seismic, due to an overlying low-velocity zone caused by gas seepage.
The true depth structure of the prospective layer could only be inferred from depth-
converted maps and velocity functions derived from shallower wells into the gas seep-
age zone and the surrounding unaffected sedimentary layers. Such mapping showed a
definite, but modest, structural closure leading to a drillable prospect.
However, independent confirmation was required before the commitment of funds
for the deep, high-risk drilling program. Spirit chose to analyze the problem by building
a velocity model using image gathers generated from iterative prestack depth migration
(Fig. 2). Because independent confirmation was required, wells were not included in
the model-building from image gathers, but were to be considered later for calibration
and comparison.
Gas seepage zones often present a difficult problem because data below them are of
poor quality. Fortunately, the overall data quality here was good (Fig. 3), although the
reflection quality was lower below the gas zone. Spirit was able to resolve the seismic-
derived velocity model after several iterations of prestack depth migration and model
updating using image gather analysis (Fig. 4). The seismic velocity model indeed
showed the low-velocity zone, but it was a bit broader than the low-velocity zone in the
model derived from well data.
Subsequent to the velocity modeling from seismic alone, Spirit first compared the
quantitative aspect of the seismic velocity model against the time-depth relationship
from the shallow wells. The degree of accuracy of the velocity pull-down predicted by
the seismic velocity model was then known. The seismic velocity model was, in fact, in
good agreement with the well data and gave confidence to the overall methodology for
deriving the velocity model from the seismic, and to the suitability of using the seismic
data in this manner (Figs. 5–7).
At this point, with confidence in the seismic velocity model established, it was time
to use both models in concert. The seismic velocity model was calibrated to the well
data and the resulting velocity model was used for all subsequent time-depth work.
This included building a final depth-structure map for the deep prospect with newly
established confidence in its accuracy.

Distinguished Instructor Short Course s 5-11

 The Seismic Velocity Model as an Interpretation Asset

Fig. 1.
At a depth of about 24,000 ft, a
rollover structure on the up-thrown
side of a large listric fault was identi-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

fied as a potentially attractive deep

prospect in a mature zone. Known gas
seepage was corrupting attempts to
map this structure accurately in
depth. Existing wells were too shal-
low to penetrate the prospect,
although they did provide a quality
check for the shallow part of the
velocity model. The objective was to
map the depth structure as accurately
as possible.

Fig. 2.
A map shows the distribution of velocity control locations for the seismic velocity analysis. Approximately 600 loca-
tions were analyzed through four iterations to determine the velocity model. The semblance plots in the figure were
computed from image gathers and show the data quality.

5-12 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 3.
A set of image gathers shows the data
quality.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 4.
This time structure map is of the horizon in the velocity model that corresponds to the target horizon. Also shown is
the 3D velocity model.

Distinguished Instructor Short Course s 5-13

 The Seismic Velocity Model as an Interpretation Asset

Fig. 5.
Shown is a depth structure map of the
prospective horizon in the velocity
model. This horizon was defined as
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

part of the structural constraints of

the velocity model and is shown in
final form rendered in depth, accord-
ing to the latest update of the velocity
model. This rendering in depth is
based on the 3D velocity model, not
on a conversion based on a map of
average velocity.

Fig. 6.
Shown is a map of interval velocity
over the prospect layer in the model.
It describes the seismic resolution of
the lateral extent of the gas zone, but
does not give an accurate reading of
the time-depth relationship because it
describes the velocity in only one of
the layers.

Fig. 7.
Shown is a time surface from a layer
of constant depth in the model. This
layer lies just under the prospect and
forms an accurate portrayal of the
time-depth relationship in the model
at the approximate depth of the
prospect. In other words, the velocity
pull-down of the anticlinal feature in
the time domain can be inferred from
the time structure of this layer.

5-14 s Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The Fault Shadow

6
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

The Fault Shadow

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Hydrocarbon deposits are often found trapped against an impermeable fault plane.
Under such circumstances, there are often distinct fluid saturation characteristics across
the fault, which affect the velocity model. At other times, such as with the next case
study, a high-velocity layer is itself faulted, and the prospect lies underneath this faulted
layer. The step of the high-velocity layer causes a strong lateral change to velocity,
which usually causes only a subtle image distortion. However, when this distortion is
precisely where a prospect sits, it can become a large problem, and it can create difficul-
ties in the well placement decision. Our next case study looks at just such a problem.

U Cox and Perkins/DDD Energy: Identifying a possible permeability barrier as a false

image

Distinguished Instructor Short Course s 6-1

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Identifying a Possible Permeability Barrier

as a False Image
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Cox and Perkins/DDD Energy

Phil Schultz

Introduction
Cox and Perkins and DDD Energy were partners in an exploration project in the
Wilcox trend, onshore Texas, and had difficulty deciding where to locate an exploratory
well. The play was in a heavily faulted zone below a high-velocity layer associated with
the Wilcox, where two high-amplitude events truncated into a major fault, as evident
on the 3D data set. The events were prospects, since no corresponding high-amplitude
events were visible on the other side of the fault.
Underneath the major fault was what appeared to be another fault, although a
minor one. It seemed to break the two events representing the prospective layers. While
the amplitudes were equally strong on either side of the apparent minor fault, it was
possibly a permeability barrier creating two reservoir compartments not in hydraulic
communication. If so, the partners wanted the first well to be placed in the larger com-
partment, to drain the greater amount of oil, before a second or sidetrack well was
drilled. On the other hand, if the minor fault was not real, the initial well should be
placed in another location.
If the minor fault is not real, why the apparent break of the horizon on the seismic
section? The minor fault was nearly parallel to the major fault and could be mapped
over a fairly extensive region. However, in cross section it was almost directly under the
break of the high-velocity layer from the major fault.
The partners suspected a potential fault shadow distortion and elected to examine
the problem first with poststack, and later with prestack, velocity model-building and
depth migration. The poststack depth migration showed some differences and improve-
ments, but still showed the apparent minor fault. On prestack depth migration, the
minor fault showed a significant healing. Although the healing was not 100% complete,
the partners, together with the depth-imaging contractor, concluded that any residual
break in continuity was due to a less than perfect velocity model.
The following figures and captions guide the reader through the case study, as it
unfolded.

Distinguished Instructor Short Course s 6-3

 The Seismic Velocity Model as an Interpretation Asset

Fig. 1.
On this schematic view of an interpre-
tation problem related to a possible
secondary fault beneath a major
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

regional listric fault, the major fault is

shown, and like a shadow, a smaller
secondary fault seems partially to
echo its position along the first. Is the
fault real, acting as part of a fault sys-
tem, or an imaging artifact caused by
the faulted high-velocity layer?

Fig. 2.
In this 3D velocity model, the high-
velocity layer was modeled as an
interpreted layer, since it was too nar-
row to be seen clearly in the seismic
velocity analysis.

6-4 s Society of Exploration Geophysicists

 Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 3.
Results of an image gather velocity analysis are shown, whereby the structural surfaces of the velocity model overlay
the semblance plots and the image gathers. The high-velocity layer in the Wilcox trend is too narrow to detect from
the seismic analysis alone.

Fig. 4.
This cross section of the velocity
model for line 230 in the depth
domain shows interval velocities in
color.

Distinguished Instructor Short Course s 6-5

 The Seismic Velocity Model as an Interpretation Asset

Fig. 5.
This figure shows the interval velocity
depth model overlaying the prestack
depth-migrated seismic data for
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

crossline 230.

Fig. 6.
This cross section of the velocity
model for line 230 shows the data and
model in the time domain.

6-6 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 7.
Note on this 3D poststack time-
migrated section at line 240 that at
CDP 290 and two-way time of 2 800
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ms, two horizons truncate into the

regional fault, as shown in Fig. 1. The
fault is sloping downward to the left.
There appears to be an indication of a
minor break in both of these events,
which were the subject of exploration
interest by the partners. Is the minor
break real? If so, does it hydraulically
compartmentalize the reservoir?

Fig. 8.
This close-up of the two horizons
shown in Fig. 7 reveals more detail of
the events at 2 700 ms and 2 800 ms.

Distinguished Instructor Short Course s 6-7

 The Seismic Velocity Model as an Interpretation Asset

Fig. 9.
Note in this view of Line 240 after 3D
poststack depth migration that the
two events appear again at 12,000 ft
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

and 13,000 ft. They are better imaged

here, and it still appears that they
have suffered minor faulting at CDP
290, although suspicions have arisen
because the minor fault appears verti-
cal. It may still be an artifact. Notice
also the clearer image of the rotated
fault blocks from 13,000 ft to 15,000
ft, and between CDPs 170 and 260, as
well as the deeper listric fault trending
downward to the left and flattening
near 24,000 ft.

Fig. 10.
A close-up view of Fig. 9 reveals more
clearly both the problematic faulted
events at CDP 290 and the rotated
fault blocks.

6-8 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 11.
Looking again at line 240, but after
full 3D prestack depth migration, it
now appears that the mystery faults
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

have healed, at least for the most part.

The partners and the seismic contrac-
tor felt that the imperfect healing was
due to some remaining imperfections
in the velocity model, and that indeed
the apparent fault was an artifact of
the faulted, high-velocity layer.

Fig. 12.
A close-up of Fig. 11 brings even
more clarity to the problematic fault-
ed events and the rotated fault blocks.

Distinguished Instructor Short Course s 6-9

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Getting Well Placement Right

7
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Getting Well Placement Right

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

When velocity overburden has lateral velocity variations, it can cause a mispositioning
of imaged features, such as fault planes at depth, through the mechanism of ray bend-
ing, which is not accommodated correctly by the velocity model. On refining the veloci-
ty model, ray-bending corrections are made more accurate, and lateral positioning of
features, such as faults, becomes more accurate. This problem becomes increasingly
more severe as the depth to the target increases. So, with today’s deeper targets, well
mispositioning can become a costly mistake. The following two case studies address the
issue of where to place a well, when the velocities are suspected of causing a misposi-
tioning on the seismic image.

U British Petroleum: Enhancing the positioning of complex faulting

U Enron Oil & Gas: Clarifying the termination of a reservoir

Distinguished Instructor Short Course s 7-1

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Enhancing the Positioning of Complex Faulting

British Petroleum
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The Application of 3D Depth Migration to the Development

of an Alaskan Offshore Oil Field*
David N. Whitcombe, Eugene H. Murray, Jr.,
Laurie A. St. Aubin and Randall J. Carroll

Editor’s Note: A total of 20 figures accompany this case study, although they are not specifi-
cally cited. The reader is encouraged to refer to these while reviewing this work.

Abstract
Inconsistencies in fault positioning between overlapping 3D seismic surveys over the
northwestern part of the Endicott Field highlighted lateral positioning errors on the
order of 1,000 ft (330 m) in the seismic images. This large uncertainty in fault position-
ing placed a high and often unacceptable risk on the placement of wells. To quantify
and correct for the seismic positioning error, 3D velocity models were developed for
ray-trace modeling. The lateral positioning error maps produced revealed significant
variation in the mispositioning within the Endicott Field that were mainly caused by
lateral variations in permafrost thickness. These maps have been used to correct the
positions of mapped features and have enabled several wells to be successfully placed
close to major faults.
Prior to this analysis, these wells were considered too risky to place optimally. The
seismic data were 3D poststack depth migrated with the final velocity model, producing
a repositioned image that was consistent with the ray-trace predictions. Additionally, a
general enhancement of data imaging improved the interpretability and enabled the
remapping and subsequent successful development of the peripheral Sag Delta North
accumulation.
Endicott Field was the first Arctic offshore development. Reserve estimates are
approximately 1.1 billion bbls OIP (oil in place) with 480 million STB (stock tank bar-
rels) recoverable. The reservoir is in the Mississippian Kekiktuk Formation and is struc-
turally and stratigraphically complex. All the development wells have been drilled from
two man-made islands.
It was evident early in the development of Endicott Field that significant inconsis-
tencies in fault location [on the order of 1,000 ft (330 m)] existed between the Endicott
3D seismic data set and the overlapping Niakuk 3D data set in the northwestern part of
the field. In addition, there were fault location inconsistencies [on the order of 750 ft
(240 m)] between a Kirchhoff1 and a F-K Stolt stretch2 time migration of the Endicott
data set itself. These lateral positional discrepancies of major faults were accommodated
by simply positioning wells off both fault locations by a minimum of 500 ft (150 m).
* Published in GEOPHYSICS, Vol. 59, No. 10, pp 1551–1560, 1994.

Distinguished Instructor Short Course s 7-3

 The Seismic Velocity Model as an Interpretation Asset

Initial development wells were placed as far as possible from faults. Subsequently, as
field development continued, it became necessary to place infill wells closer to the
faults. In some cases, to optimize reservoir depletion, it was necessary to drill wells as
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

close as possible to faults (without, of course, crossing them).

An example of the mispositioning inherent in the time-migrated seismic data is pro-
vided at well Sag Delta 3. On the seismic data, it appears that the well is entirely north
of the Mikkelsen Bay Fault, completely in the main fault block of the Endicott Field.
The well data, however, are interpreted differently. It appears that the well penetrates
the Mikkelsen Bay Fault, drilling from the hanging-wall block south of Endicott, into
the foot-wall block of the field itself. This is evidenced by approximately 175 ft (50 m)
of missing section in the Upper Kekiktuk Formation. In fact, if the top Kekiktuk depth
from Sag Delta 3 is used in correcting the structure maps, the Kekiktuk structure
appears to be too deep. It seems as though the Mikkelsen Bay Fault is positioned on the
seismic approximately 450 ft (135 m) south of where the well actually encounters it.
This problem was initially corrected by arbitrarily moving the fault north to tie the
well and locally not honoring a strict interpretation of the seismic image. After the 3D
depth migration of the data, this well ties the seismic data correctly.

Geophysical Concepts Behind Lateral Mispositioning, Depth Migration

Standard seismic time-migration data generally do not position events correctly at their
true lateral position, even if the migration is 3D. There are two reasons for this. First,
the migration velocities may be incorrect by a small amount. This velocity-induced mis-
positioning will get worse the steeper the dip of the event. Second, time migration does
not account correctly for refraction.3 If there are strong lateral changes in velocity, such
as those caused by a permafrost wedge in Alaska, or a salt structure, or a thrust belt set-
ting, then depth migration is necessary to take into account the lateral refraction.
Delta 3 well does not appear to intersect the prominent Mikkelsen Bay Fault plane.
Instead, the well appears to be entirely within the up-thrown fault block. In (b), the
equivalent section from the 3D depth-migrated data set (using model C), the well does
appear to intersect the Mikkelsen Bay Fault plane. This is consistent with the geologist’s
interpretation of the section in the well. The trace spacing is 110 ft.
While time migration requires standard root mean square (RMS) velocity functions,
depth migration requires an interval velocity depth model. Depth migration can be car-
ried out using ray-trace depth migration (also known as map migration) of picks from
the unmigrated seismic,4 or by seismic depth migration of the unmigrated seismic
image.5 These two methods should provide the same event positioning, given the same
velocity field.
A way of understanding why time migration fails to position events at their correct
lateral position is provided by “image rays.”3 Time migration can be thought of as an
operation that collapses a diffraction hyperbola, and a reflection is made up of the
summed effects of a set of diffraction hyperbolas. If there is significant lateral velocity
variation, the apex of a hyperbola observed at the surface no longer lies directly above
the subsurface point on the reflector. When time migration is carried out, the hyperbola
will collapse to its apex, and the information about the subsurface reflection point con-

7-4 s Society of Exploration Geophysicists

 Phil Schultz

sequently will be mispositioned. The amount of mispositioning is defined by an “image

ray,” which is the ray that starts off vertically at the surface and refracts down to the
reflector. The image ray approximation holds well for the low structural dips encoun-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tered on the North Slope. Note that if there were no wedging of the permafrost, the
image ray would not bend, and consequently there would be no mispositioning induced
by lateral refraction.
If there is significant lateral velocity variation, the hyperbola becomes skewed in
shape and no longer collapses to a point when a hyperbolic time-migration operator is
used.6 Both correct positioning and improved imaging can be achieved using a depth-
migration algorithm.

Building Model A From the Time-Migration Velocity Field

Available data. Initially, the time-migration velocities derived in the original processing
were used to provide the interval velocity field. To convert these velocities to a migra-
tion velocity field, the velocities were resampled to constant time increments, inverted
into 100-ms thick interval velocity layers using the Dix formula, and then smoothed lat-
erally. Our inspection of the derived interval velocities showed them to be well behaved
and plausible. They show a strong lateral change with the higher velocities being land-
ward, as expected.
Analysis using model A. The interval velocity/time model was depth converted into a
velocity/depth model using 3D image rays. The image rays start vertically at the surface
and are propagated downward, refracting through the variable velocity layers for a one-
way traveltime equal to that of the target horizon.
The difference between the start position and the end position of the image ray
gives the “image ray shift,” which defines the lateral positioning error inherent in the
time-migrated image. The greatest shifts are in the inline direction (i.e., perpendicular
to the shoreline) and are as much as 900 ft (270 m) at the P4 level. Smaller, but still sig-
nificant errors of up to ±300 ft (90 m) occur in the crossline direction.
The mapped vertical faults were corrected for this image ray effect by shifting each
point making up the fault trace by an amount obtained from this lateral-positioning
error map.
Sensitivity study for the base permafrost. When we embarked on this project, we were
aware of the sparse data pertaining to the base of the permafrost. Only in a few of the
Endicott wells had this boundary been logged with any confidence. Also, because field
wells radiate from two islands where shallow wellbores are only 10 ft apart, log data at
the shallower levels were not areally representative. These logs indicated a gradational
base, rather than an abrupt transition. Additionally, the logs indicated variability in both
the velocity of the permafrost and its detailed velocity structure. It concerned us that we
had difficulty defining a good permafrost velocity model at the well locations. To correct
the 3D image, we would additionally need a good velocity model between the wells.
The processing velocities were attractive as they provided information between
wells. Additionally, they showed a gradational variation, broadly consistent with the
well logs. The question arose, though, as to how sensitive the image ray shifts were to
the details of this velocity field.

Distinguished Instructor Short Course s 7-5

 The Seismic Velocity Model as an Interpretation Asset

A sensitivity study was carried out, combining the first eight of the 100-ms thick
layers in the model into a single layer with the same average velocity. The base of this
new layer was at 800 ms, approximately the average time of the T2 marker, a mapable
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

coal marker across the survey. This single layer, therefore, included permafrost and the
immediate subpermafrost strata. The average velocity of this layer varied from 6,900 to
8,900 ft/s (2,100 to 2,700 m/s).
Three-dimensional image ray tracing was repeated using this model. The predicted
mispositioning was within ±30 ft (9 m) of that predicted from model A. This surprising
result indicated that we did not need to be able to accurately define the vertical velocity
structure of the permafrost, only the average velocity to a level below the permafrost.
We now had a much more tractable problem. Regional mapping has indicated the
almost planar depth nature of shallow markers below the permafrost. These depths,
coupled with the seismic times, could provide an average velocity that included the per-
mafrost layer.
To investigate this issue further, the first fourteen of the 100-ms thick time layers in
the model were combined into a single layer with the same average velocity. The base of
this layer is located approximately at the time of the K7 horizon, a strong Cretaceous
coal event throughout the region. The 3D image ray displacements for this model were
within ±80 ft (24 m) of the estimates obtained for the more detailed model A.

Building Model B From Well Data

Methodology. Velocity model A was used to predict the depth to the K7, M7 and P4 by
using the mapped traveltimes to these events. For the K7 level, the average depth error
was 25 ft (8 m), with an RMS error of 100 ft (30 m). For the M7 level, however, the
average depth error was significantly larger at 600 ft (180 m), with an RMS error of 300
ft (90 m). Because the misties to the wells were unacceptably large, model B was gener-
ated, which more directly used the well data.
We decided to keep the velocity model to the T2 horizon and to use the well data to
derive a velocity for the T2 to K7 interval. Velocity crossplots that used the K7 well data
were studied. It was observed that a crossplot of the traveltime to K7 showed an almost
linear relationship to the average velocity to K7. By using this function and backing out
the velocity model above T2, an interval velocity field was produced for the T2 to K7
interval.
Sixty well penetrations, coupled with the 3D time surface, provided a good depth
surface to the M7 level. Using this, together with the K7 depth and time surfaces, a K7
to M7 interval velocity field was generated. This model closely tied the well depth data,
which are most widely distributed at the deepest levels.
Beyond image rays. Having derived a velocity model different from that used for the
original time migration of the data, there is an implication that the original migration
velocity field was in error. It follows that the original time migration was to a certain
degree mismigrated. In these circumstances, image rays should not be used for the
depth conversion because image rays account for only the refraction element of mispo-
sitioning, not for the mismigration of the time-migrated image that results from using
an incorrect migration velocity field.7

7-6 s Society of Exploration Geophysicists

 Phil Schultz

To account for both refraction and mismigration, ray-trace depth migration must be
used. This procedure involves taking interpreted events from an unmigrated time sec-
tion and starting the depth conversion rays at angles calculated using the observed
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

unmigrated time dip of the event. This approach, however, has the inherent problem
that the interpretation must be done from an unmigrated data set. Such an interpreta-
tion was not available for the Endicott data set. A solution to this problem is to use the
existing migrated time interpretation, and then demigrate the interpretation using the
processing migration velocities.7
Analysis using model B. Velocity depth model B was generated by this time demigra-
tion and ray-trace depth migration procedure, and the lateral positioning error maps
associated with the time-migrated image were derived. These maps show the same
broad features seen in the image ray derived maps, but certain features are significantly
different. For model B, the mispositioning looks more “2-D-like” than for model A.
Note also that for model B, discontinuities in the mispositioning can be seen associated
with the major fault planes. This is to be expected if there is a dip discontinuity across
the fault plane. The lateral positioning error maps for model B were used to generate
shifted fault maps.

Building Model C: Accounting for Anisotropy

The affect of anisotropy on migration. There is considerable evidence for the existence
of significant rock anisotropy8, 9 and for the need to consider anisotropy when carrying
out migrations.10,11 Anisotropy in this context means the variation of instantaneous
velocity with angle to the bedding, which is distinct from lateral velocity variation.
Anisotropy can manifest itself in the seismic-derived interval velocities, which are effec-
tively measured in the horizontal direction and are generally larger than vertically mea-
sured well velocities. If the lower, well-derived velocities are used in a migration, the
structure will be under-migrated. The structure is better imaged by migrating with the
higher, horizontal velocities.11
Analysis using velocity model C. A comparison of the well- and seismic-derived veloci-
ties over the entire extent of the model indicated the seismic velocities averaged 6.8 per-
cent higher than the well velocities. Model C was therefore generated by scaling the
well velocities by this value and by repeating the ray-trace depth migration.
The lateral positioning error estimates were obtained from the ray-trace depth
migration of model C and are similar to the results for model B. However, differences of
up to 150 ft (45 m) in the inline and 80 ft (25 m) in the crossline direction occur.
Model C gave the best tie of the Mikkelsen Bay Fault to the Sag Delta 3 well. Several
other wells in the field that are faulted also tie very closely, if not exactly, using fault
positions predicted from model C. In addition, the repositioned midfield fault, which is
a major transmissibility barrier, is consistent with the pressure data from the wells on
either side of this fault. The results of model C are the most consistent with the existing
well and surveillance data.

Distinguished Instructor Short Course s 7-7

 The Seismic Velocity Model as an Interpretation Asset

Impact Of The Ray-Trace Depth Migration Results On The Niakuk Fault Block Area
Because of the uncertainties in fault positioning before this work was carried out, a rule
was imposed that no well should be drilled within 500 ft (150 m) of any major fault.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Development was limited to a narrow corridor down the middle of the fault block.
Wells L-15 and L-16 were required to optimally drain this fault block, but were consid-
ered too risky to drill.
Prior to depth migration of the seismic volume, lateral positioning error maps gen-
erated from the ray-trace depth migration were used to correct the fault locations
mapped from the time-migrated data. This enabled wells L-15 and L-16 to be success-
fully drilled close to the bounding fault. Neither of these wells would have been drilled
prior to the reduction in fault-location risk obtained by quantifying the magnitude and
direction of the lateral positioning errors. At an average well cost of U.S.$3 million, the
additional reserves associated with these two wells alone paid for the entire Endicott
depth-migration project many times over.

3D Depth Migration of the Seismic Data

Test swaths and comparisons with the 3D time-migrated image. A subset of the data
was initially chosen for testing prior to the migration of the entire survey. Depth migra-
tions were carried out using models B and C (with its 6.8 percent higher velocities and
depths). An omega-X algorithm was used with extended steep dip logic. The depth
images were vertically converted back to time so that they could be compared to the
original Kirchhoff time migration. These test swaths showed that the repositioning of
the horizons and faults was as predicted by the 3D ray-trace depth migration to within
the lateral resolution of the data [about 100 ft (30 m)]. In particular, the seismic image
of the Mikkelsen Bay Fault was now correctly positioned in relation to the Sag Delta 3
well by model C, the model that accounts for anisotropy.
An examination of the two data sets for interpretability and for signs of under- or
over-migration indicated that model C was preferred, although the advantages of model
C in this regard were subtle. Both migrations showed better horizontal resolution than
the original Kirchhoff time migration. The complete data set was migrated with model C.
Impact of the seismic depth migration on the Sag Delta North area. A peripheral
accumulation in the Triassic Ivishak Formation to the northwest of the Endicott Field
was indicated by the Sag Delta 9 exploration well. Approval to develop this region from
the Endicott Main Production Island was obtained in early 1990. Initial appraisal, volu-
metrics, and well targeting were completed using the Kirchhoff time-migrated Endicott
3D seismic data set. A few weeks prior to the spud of the first Sag Delta 9 offset, the Sag
Delta North 3 well, the depth-migrated data volume became available. Reinterpretation
of the area forced a significant last minute change in the well’s target location.
The initial interpretation of the time-migrated data set indicated that the Sag Delta 9
area was a faulted, anticlinal structure truncated by M7, the Lower Cretaceous
Unconformity (LCU), and overlain by the Cretaceous HRZ shales. (In this area, the
HRZ shales are both source and seal for the Ivishak reservoir.) Several faults were inter-
preted and mapped in the accumulation. A crestal position had been chosen as the first
offset, which maximized the thickness of the oil column.

7-8 s Society of Exploration Geophysicists

 Phil Schultz

Interpretation of the depth-migrated data produced a dramatically different structur-

al picture because of its improved lateral resolution and better reflector definition. The
initial location was directly on a crestal fault of a complex and steeply dipping trans-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

pressional “flower structure.” Several additional faults were also interpreted. Only days
before the Sag Delta North 3 well was to spud, a new development plan was conceived,
and the well location was moved to prevent a probable complete or partially faulted-out
reservoir section. The repositioned well encountered a reservoir at the predicted depth
and in a correct relationship to the faulting. The depth-migrated data have been the key
to reducing structural risk in the later stages of Endicott Field development.

Conclusions
The major technical conclusions from this work are:
1) In the presence of laterally varying velocity fields, such as are associated with per-
mafrost on Alaska’s North Slope, mispositioning of the seismic image can be signifi-
cant to the extent that it complicates field development planning. Lateral position-
ing errors of up to 1,000 ft (300 m) existed in the 3D time-migrated data that had
been used for field development mapping.
2) The mispositioning was quantified by 3D ray tracing as lateral positioning error
maps. These error maps are an immediately useful, initial product that can indicate
if the amount of mispositioning is enough to justify proceeding with the depth
migration of the seismic volume. Moreover, corrections can be made to fault loca-
tions from the error maps, and this information can be quickly input into the well
planning process without waiting for the result of the seismic depth migration.
3) The amount of mispositioning is not sensitive to the details of the vertical velocity
structure of the permafrost. A simple, single-variable velocity top layer that includes
the permafrost and the immediate subpermafrost structure provides a refraction
equivalent to a more detailed and realistic velocity model. Thus, the refractive
effects of the permafrost can be removed without detailed knowledge of the vertical
velocity structure of the permafrost.
4) A 3D depth migration of the seismic image resulted in an image that was positioned
consistently with the 3D ray-trace depth migration to within the interpreter’s ability
to precisely pick a fault.
5) The quality of the 3D depth-migrated image was significantly superior to the origi-
nal time migration and has enabled better remapping of the field. Specifically, the
ability of the interpreter to pick faults has been greatly enhanced. A much more
complex fault picture has emerged, which is consistent with the well and surveil-
lance data.
6) While it is desirable to quantify and correct for mispositioning as early in a develop-
ment project as possible, if the effect is indeed significant, even fields in the mature
stages of development can benefit from this type of analysis. Often, later in a field’s
life, only the difficult, more constrained wells remain to be drilled. Late-stage infill
drilling and analysis of surveillance results may both benefit from a lateral-position-
ing/depth-migration study.

Distinguished Instructor Short Course s 7-9

 The Seismic Velocity Model as an Interpretation Asset

Finally, but most importantly, the results of the study have enhanced the commercial
success of the Endicott and Sag Delta North development programs.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Acknowledgments
We acknowledge the significant contributions made to this work by the following BP
staff: Nick Burke, Janet George, Rich Randall, Samuel R. Johnson, Kent Greenes, and
Jim Keggin. We thank BP Exploration, the field operator, and the other Endicott owners
for permission to publish this work.

References
1
Schneider, W. A., Integral formulation for migration in two and three dimensions:
GEOPHYSICS, 43, pp 49–76, 1978.
2
Stolt, R. H., Migration by Fourier transform: Geophysics, 43, pp 23–48, 1978.
3
Hubral, P., Time migration—Some ray theoretical aspects: Geophys. Prosp., 25, pp
738–745, 1977.
4
Fagin, S. W., Defining a salt sill using three-dimensional ray-trace modeling and
inversion, in Fagin S. W., Ed., Seismic modeling of geologic structures: Geophysical
Development Series: v. 2, SEG, pp 209–248, 1991.
5
Judson, D. R., Lin, J., Schultz, P. S., and Sherwood, J. W. C., Depth migration after
stack: Geophysics, 45, pp 376–393, 1980.
6
Robinson, E. A., Migration of geophysical data: International Human Resources
Development Corporation, 1983.
7
Whitcombe, D. N., Fast model building using demigration and single-step ray migra-
tion: GEOPHYSICS, 59, pp 439–449, 1994.
8
Banik, N. C., Velocity anisotropy in shales and depth estimation in the North Sea
Basin: GEOPHYSICS, 49, pp 1411–1419, 1984.
9
Thomsen, L., Weak elastic anisotropy: GEOPHYSICS, 51, pp 1954–1966, 1986.
10
Whitcombe, D. N., and Collyer, V. A., Evidence for the existence of anisotropy
beneath the Southern North Sea: Presented at the 50th Meeting of the European
Association of Exploration Geophysicists. 1986, Weak elastic anisotropy: GEOPHYSICS,
51, pp 1954–1966, 1988.
11
VerWest, B., Seismic migration in elliptically anisotropic media: Geophys. Prosp., 37,
pp 149–166, 1989.

7-10 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 1
The location of Endicott Field is
shown in relationship to other North
Slope fields.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 2.
This Endicott Field generalized type
log defines the reservoir zonation and
major seismic reflectors. Coals pro-
duce the P3 and P4 reflectors; where-
as, an angular unconformity (LCU)
creates the M7 reflector, defining the
top of the reservoir.

Fig. 3.
The Endicott 3D seismic survey is
shown with respect to the Endicott
Field.

Distinguished Instructor Short Course s 7-11

 The Seismic Velocity Model as an Interpretation Asset

Fig. 4a.
A representative inline section
through the 3D seismic data is shown
with key horizons identified.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 4b.
Detail of the reservoir, which is locat-
ed between M7 and P4, is shown.

Fig. 5.
Well penetrations are shown at the
reservoir level in the Endicott Field
and Sag Delta North area. All wells
have been drilled directionally from
two man-made islands.

7-12 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 6.
In the time migration (a), the Sag
Delta 3 well does not appear to inter-
sect the prominent Mikkelsen Bay
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fault plane. Instead, the well appears

to be entirely within the up-thrown
fault block. In (b), the equivalent sec-
tion from the 3D depth-migrated data
set (using model C), the well does
appear to intersect the Mikkelsen Bay
Fault plane. This behavior is consis-
tent with the interpretation of the sec-
tion in the well. Trace spacing is
1,100 ft.

Fig. 7.
This figure addresses Niakuk fault
block well targeting issues and shows
the mapped locations of the bounding
faults prior to depth migration,
together with the 500-ft stand-off cor-
ridor around these faults. Note that
the wells have been placed away from
these major faults (150 ft to 700 ft
throw).

Fig. 8.
This figure shows the mapped position
of the faults using the Kirchhoff time
migration and ray trace predicted,
depth-migrated positions. The optimal
locations of the L15 and L16 wells are
shown by polygons, and the success-
fully drilled locations by circles.

Distinguished Instructor Short Course s 7-13

 The Seismic Velocity Model as an Interpretation Asset

Fig. 9.
Structural mapping of the Sag Delta
North Area from the 3D Kirchhoff
time-migrated data. Also shown is the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

proposed location of well SDN3.

Fig. 10.
After depth migration, the Sag Delta
North Area (Top Lower Ivishak Sand)
was remapped, as shown. Note that
the proposed well lay directly on top
of a fault position. The well was
moved several hundred feet and was
successfully drilled. Subsequent
drilling in this area has yielded results
consistent with this new map.

7-14 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 11.
Crossline 243 is shown through the
Sag Delta North area from (a) the
Kirchhoff time-migrated data and (b)
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

from the 3D depth-migrated data.

Note the improved resolution and dif-
ferent structural interpretation.

Fig. 12.
A discontinuity at the target generates
a diffraction hyperbola. Time migra-
tion collapses this hyperbola to its
apex. The surface position of the
image ray predicts the apex position.
Image rays can, therefore, be used to
estimate the mispositioning that is
caused by a high-velocity permafrost
wedge.

Distinguished Instructor Short Course s 7-15

 The Seismic Velocity Model as an Interpretation Asset

Fig. 13.
The seismic processing velocities
derived from the Endicott 3D survey
are shown. A velocity inversion (a)
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

below the high-velocity permafrost is

evident in the stacking velocities. The
migration velocities (b) were derived
by inverting the stacking velocities
into interval velocities, smoothing and
converting to Vrms.

Fig. 14.
Seismic-based interval velocities are
derived by making the stacking velo-
cities horizon consistent with the time
horizons. The horizon-consistent
velocities are inverted to interval
velocities using the Dix equation.

Fig. 15.
These lateral positioning error maps
were determined from model A,
which is based on seismic velocities
only. The upper diagram shows the
errors in the crossline direction, while
the lower map shows the larger errors
in the inline direction.

7-16 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 16.
This figure depicts the mapped posi-
tions of the faults at the P4 level as
taken from the Kirchhoff time migra-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tion, together with the moved posi-

tions of the faults predicted by 3D ray
trace depth migration using velocity
model A.

Fig. 17.
Lateral positioning error maps were
developed from model B, which
includes all the well velocity control.
The upper diagram shows the errors
in the crossline direction, while the
lower map shows the larger errors in
the inline direction.

Fig. 18.
This figure depicts the mapped posi-
tions of the faults at the P4 level from
the Kirchhoff time migration, together
with the moved positions of the faults
predicted by 3D ray trace depth
migration using velocity model B.

Distinguished Instructor Short Course s 7-17

 The Seismic Velocity Model as an Interpretation Asset

Fig. 19.
This figure depicts the lateral posi-
tioning error maps determined from
model C, which increase the well
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

velocities by 6.8% to tie the process-

ing velocities. The upper diagram
shows the errors in the crossline
direction, while the lower map shows
the larger errors in the inline direc-
tion.

Fig. 20.
This figure depicts the mapped posi-
tions of the faults at the P4 level from
the Kirchhoff time migration, together
with the moved positions of the faults
predicted by 3D ray trace depth
migration using velocity model C.

7-18 s Society of Exploration Geophysicists

 Phil Schultz

Clarifying the Termination of a Reservoir

Enron
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Phil Schultz, from a conversation with Paul Garber

Editor’s Note: The 20 uncited figures that follow this narrative visually complete the story.

Enron located a new well in the vicinity of a salt dome, where the hydrocarbon layers
sloped up-dip to trap against the salt. Seismic showed bright spots and therefore hydro-
carbon prospects in the Bul1 and TexX layers. When the well was drilled, both hydro-
carbon layers came in about 30 and 70 ms shallower than expected, based on the origi-
nal time-migrated section. The time/depth correspondence was determined from check
shot analysis in the well. Enron knew that the 70-ms error was more than just an
annoyance; it was important for the placement of the next well. These water-drive reser-
voirs needed to be tapped near the structural high.
Enron identified one possible source of error as channeling of the seismic waves
through the salt, like in a salt proximity survey, causing an erroneous arrival. Such an
arrival would be earlier than along a vertical raypath, and thus possibly causing the
inconsistency between the times on the seismic and in the well. However, for several
reasons, Enron eventually concluded that the waves were not channeling. First, the
downhole measurement used a 3-component phone, which did not see any steeply dip-
ping arrivals coming near-horizontal from the direction of the salt, rather than near-ver-
tical. Also, in calculating an interval velocity function down the well, there were no
clearly anomalous readings that would indicate a transition to a different arrival. Finally,
check shots in a nearby well away from the salt were consistent with these downhole
measurements. Based on these observations and analyses, Enron concluded that the
check shot was recording a near-vertical path, and was therefore valid to use in convert-
ing between time and depth for the well data.
In addition to the misties in the time domain, there was the problem with the mea-
sured dips. The dips measured in the well were high compared to dips measured on the
seismic data. Datum errors were not likely. Why the mistie and the inconsistent dips?
Enron speculated that the time-migrated data may have been over-migrated, sliding
the up-sloping hydrocarbon layer down and away from the well. (Can over-migration
cause the seismic event to be pushed down? Doesn’t migration always move energy up?)
Enron decided to apply poststack depth-migration velocity analysis to the problem.
While keeping the character of the overall velocity model unchanged, all velocities were
uniformly scaled with several scale factors. Enron found that 95% velocities tied the well
to the seismic (still using the same check shot data). However, now the data looked
under-migrated in the other wells, although by less than the original mistie. Did the new
depth migration move the event back too far, relative to the time-migrated image?
Drilling decisions needed to be made. With no time for further analysis, Enron
decided that the depth-migrated data were indeed under-migrated, but closer to the true
depth. The answer lay somewhere between the time and depth migrations. This conclu-

Distinguished Instructor Short Course s 7-19

 The Seismic Velocity Model as an Interpretation Asset

sion was good enough to plan their next well, called well 2 in this report, which was
successful. Their strategy was to drill close to the edge of the bright spot on the depth-
migration image. Since the depth migration was thought to be under-migrated, this
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

approach would ensure not going too far up-dip and running into salt, which would be
a costlier mistake than drilling slightly down-dip. It was important to get as far up-dip
as possible because these are water-drive reservoirs, and any up-dip reserves would not
be captured.
Had the well been located based on the original time-migrated data volume, it
would have been placed too far to the south. It would have missed the formation and
hit salt.

Fig. 1.
On this horizon image of the Bul1
event from the original 3D poststack,
time-migrated data set, north is up.
Data were acquired with N–S lines.
The producing zones are shown as
high-amplitude events at the edge of
the salt dome. Wells 1 and 2 are
shown. However, well 2 was not
drilled from this image, but from the
poststack depth-migrated image. The
correction of the lateral positioning of
the hydrocarbon accumulation, far-
ther toward or away from the salt
dome, was the objective of the
velocity analysis.

Fig. 2.
This oblique NW-SE line, coincident
with well 1, from the original time-
migrated volume, is in the time
domain. The Bul1 and TexX events
are shown with their high-amplitude
signatures, but the well events come
in at 30 ms and 70 ms, too early for
the Bul1 and TexX events, respective-
ly. The nature of this mistie was why
the velocity analysis was undertaken.

7-20 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 3.
A close-up of the Bul1 event from the
oblique line of Fig. 2 shows the mist-
ie. The well marker is about 30 ms
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

too high versus the seismic.

Fig. 4.
A close-up of the TexX event from the
oblique line of Fig. 2 shows the mistie
at about 70 ms.

Fig. 5.
This Fig. shows a N-S line through
the time-migrated volume. This line
intersects well 1 near the Bul1 zone.

Distinguished Instructor Short Course s 7-21

 The Seismic Velocity Model as an Interpretation Asset

Fig. 6.
On this close-up view of Fig. 5, the
mistie is evident, but should be mea-
sured on the oblique line (Fig. 2).
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 7.
A N-S line that intersects well 1 near
the TexX event also shows the mistie,
but should be measured on the
oblique line (Fig. 2).

Fig. 8.
A close-up of the TexX event on
Fig. 7.

7-22 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 9.
This is the same horizon
map view as in Fig. 1,
except from the poststack
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

depth-migrated volume,
using 95% velocities. The
coarse appearance is due
to the migration being
done on every other line.
The high-amplitude zones
have moved slightly to the
north, away from the salt
dome.

Fig. 10.
A NW-SE oblique line from the post-
stack depth-migrated data, along the
same track as that in Fig. 2, shows
that the seismic events are much clos-
er to tying the well markers. However,
this line was interpreted as being
slightly under-migrated. The irregular,
coarse appearance is due to a subset
of the lines being output from 3D
depth migration, coupled with the
oblique line direction.

Distinguished Instructor Short Course s 7-23

 The Seismic Velocity Model as an Interpretation Asset

Fig. 11.
On the close-up view of the Bul1
event on the oblique line of Fig. 10,
the tie was interpreted as showing
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

that the seismic was slightly under-

migrated.

Fig. 12.
On the close-up view of the TexX
event on the oblique line of Fig. 10,
the tie was interpreted as showing
that the seismic was slightly under-
migrated.

7-24 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 13.
This is a N-S line from the depth
migration output, corresponding to
Fig. 5, which intersects well 1 at the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Bul1 event.

Fig. 14.
On the close-up of the Bul1 event
from Fig. 13, the tie is now very close.
However, it was interpreted as being
slightly under-migrated.

Distinguished Instructor Short Course s 7-25

 The Seismic Velocity Model as an Interpretation Asset

Fig. 15.
This is a N-S line from the depth
migration output, corresponding to
Fig. 7, which intersects well 1 at the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

TexX event.

Fig. 16.
On the close-up of the TexX event
from Fig. 15, the tie with the seismic
is very close.

7-26 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 17.
A N-S line from the depth-migration
output, shows the actual location of
well 2 at the Bul1 event. The location
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

of well 2 was chosen as a result of the

velocity analysis in this study. As can
be seen on the depth section, it was
positioned very aggressively. This
choice of positioning was due to the
interpretation of the depth migration
being slightly under-migrated. This
well was successful.

Fig. 18.
A close-up of the Bul1 event from
Fig. 17.

Distinguished Instructor Short Course s 7-27

 The Seismic Velocity Model as an Interpretation Asset

Fig. 19.
On a N-S line corresponding to the
line of Fig. 17, but from the original
time migrated volume, the well loca-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tion appears to be extremely conserv-

ative. However, this is misleading
since the event is mispositioned
because of the velocity error. If the
well were located using this section,
it would probably have come in south
of the target zone and hit salt.

Fig. 20.
A close-up of the Bul1 event from
Fig. 19.

7-28 s Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Evaluating Closure in Depth

8
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Evaluating Closure in Depth

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Lateral variations to velocity can cause the time structure of an interpreted event to
exhibit significantly differing structural behavior to the true depth structure.
Sometimes, closures do not exist in the time structure, but do exist in depth, and vice
versa. At other times, the apparent size of a prospect, based on a simple depth conver-
sion from its time structure, is incorrect. As with the lateral mispositioning problem,
deeper targets cause this problem to become more prevalent and more costly when it
occurs. A legitimate prospect can turn out to be much smaller than anticipated because
of velocity complexity that was not modeled. These next case studies all recognized that
pitfall, and the interpreters put effort into developing a sufficiently detailed velocity
model so that depth closure distortions were minimized.

UÊ British Petroleum: Uncovering different stories from time and depth closure

UÊ Amoco: Analyzing conflicting depth conversions

UÊ Norcen: Finding closure in depth, but none in time

UÊ Paradigm Geophysical: Determining depth structure under a channel fill

Distinguished Instructor Short Course s 8-1

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Uncovering Different Stories From Time

and Depth Closure
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

British Petroleum

Analysis of the Velocity Model for the K Prospect: Gulf Of Mexico

Tim Summers

Introduction
The “K” prospect lies in the footwall of a major growth fault and relies on depth con-
version to generate a structure that is significantly larger in depth than indicated by the
time mapping. The velocity gradient generating the structure results from the large
expanded sections in the hanging wall, which are overpressured and have velocities
10% to 15% slower than their equivalent intervals in the footwall. Such velocity gradi-
ents across the major growth faults are observed throughout the area.
The depth map on the primary target, the M20, is shown in Fig. 1. The outline of
the time structure superimposed on this map demonstrates the importance of depth
conversion in generating a prospect. The regional velocity grid from a commercial enti-
ty, with an accuracy quoted at 2%, was used for this depth conversion. The velocity
database was built by Dix inversion of a one-mile grid of “spec” stacking velocities,
which are subsequently smoothed, converted to average velocities, and finally calibrated
by check shot data.
Quality assurance of the stacking velocity was done by reviewing and examining the
velocity analyses over the prospect. In the vicinity of the fault, however, the raypaths
are nonhyperbolic due to the varying velocity inversion. Therefore, further analysis by
ray trace modeling of proprietary data was carried out to confirm the validity of the
velocity field for depth conversion.

Data Set
A proprietary data line, a dip line across the major fault, was used for the ray tracing
analysis. The stack of this line is shown in Fig. 2, with the stratigraphy annotated to
highlight the major growth intervals. The analysis described below was directed at the
M20 target horizon in the footwall of the fault.
Note that there is an area of shallow gas and associated multiples between shot-
points 400 and 440. An iso-velocity plot of the stacking velocities shown in Fig. 3 clear-
ly displays the slower velocities in the hanging wall section when compared to equiva-
lent times in the footwall. In addition, the stacking velocities vary rapidly between shot-
points 330 and 480 because of the nonhyperbolic moveout of the raypaths in the gath-
ers across the fault.

Distinguished Instructor Short Course s 8-3

 The Seismic Velocity Model as an Interpretation Asset

Data Analysis
The migration of this line, displayed in Fig. 4, shows the M20 event dipping down-
wards into the fault and no evident trap. Note also that there is a severe “bust” in the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

data at 4.2 s and shotpoint 400, severely degrading reflector continuity and interpreta-
tion of the M20 in the immediate vicinity of the fault. This discontinuity can be
explained from examination of the near offset traces.
Fig. 5 is a display of the stack of the near 3,000 ft, that is, the near one-fifth of the
cable. A sag is clearly evident in the data below the shallow gas anomaly. It is slightly
offset from the velocity gradient across the fault. When all offsets are used in the stack,
this sag is mis-stacked, resulting in a discontinuity in the reflector (Fig. 2).
Subsequent migration further complicates the target imaging, as observed in Fig. 4.
In order to get a good image of the M20 horizon, the stacking error had to be removed
prior to poststack migration. This requirement was achieved by deriving surface-consis-
tent residual statics for the shallow gas anomaly and applying them to the data.
The migration of the full offset stack with residual statics applied is shown in Fig. 6.
Note that the discontinuity along the M20 has now been removed, resulting in signifi-
cantly improved imaging.
The effect of the residual statics can also be observed on the seismic velocities. Fig.
7 is a display of horizon-based velocity analysis (HVA) along the M40 reflector between
shotpoints 210 to 510. The location of the shallow gas anomaly and the fault are anno-
tated. The major change from slower to faster velocities across the fault is evident.
However, the velocities also decrease to the right of the shallow gas between shotpoints
420-500, due to the far offsets in the receiver spread passing through slower velocity
than the near offsets, resulting in slower than expected observed stacking velocities. A
corresponding increase should occur when the near offsets pass through the gas and the
far offsets do not, but it is superimposed on the effect of the common depth point
(CDP) spread crossing the lateral velocity change across the fault, and cannot be identi-
fied on the HVA profile.
Fig. 8 is the HVA on the data with residual statics applied. Note that the decrease in
velocity previously observed in the hanging wall has now been removed, and the major
change in velocities from footwall to hanging wall is now clearly seen. There is still
some undulation along the velocity profile, which is interpreted as the effect on the
velocities as the spread crosses the lateral velocity change over the fault. This feature
has been left in the data, as it is not a near-surface effect, and should not be removed
using a near-surface tool (residual statics).

Velocity Model Building

Having removed the effect of the shallow gas anomaly to improve the time-migration
imaging, the velocity field was analyzed for depth conversion. Interval velocities were
derived from stacking velocities at least one spread length away from each side of the
fault. This ensured that the velocities used were not corrupted by raypath distortion
passing through laterally varying overburden. The interval velocities within each strati-
graphic interval were then projected into the fault plane and vertically time-to-depth
converted. (In subsequent analysis, the interval velocities were smoothed and mapped

8-4 s Society of Exploration Geophysicists

 Phil Schultz

areally and interpolated vertically to improve the model further.) The resulting depth
model is shown in Fig. 9. Note that the M20 horizon in the footwall is dipping away
from the fault plane, showing that the lateral velocity change across the fault can
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

generate structures in depth not observed in the time section.

This depth model was used for forward modeling of the normal incidence and
image ray section to validate the model against the seismic data. Note that the shallow
gas anomaly is present in the depth model. It was used in the forward modeling to
match the effect seen on the near offset stack time section. It was not included in the
model for the depth conversion, as its effect had previously been removed by the resid-
ual statics. Figs. 10 and 11 are the modeled zero-offset normal and image ray responses,
which show very good agreement with the stack and migration, respectively.
To increase confidence in the model, the prestack data was compared to ray traced
gathers. Fig. 12 shows the observed and computed CDP gathers at three locations
across the fault with normal moveout applied using the picked stacking velocities. The
locations of the gathers used are annotated on the depth model in Fig. 9. The match
between the calculated and observed data is good at the M40 and M20 levels.
Fig. 13 is the equivalent display for the data with residual statics applied and ray
traced through the model without the shallow gas anomaly. There is good correspondence
between the observed and calculated gathers, which shows that although the residual stat-
ics have removed the nonhyperbolic moveout due to the gas, the effect of the velocity
change across the fault is still present. The similarity between the observed and modeled
gathers and the normal moveout (NMO) correction to the synthetic gathers using the
observed stacking velocities is very good at M40 and M20, generating confidence in the
velocity/depth model across the fault and its validity for use in depth conversion.

Depth Conversion and Depth Migration

The time migration vertically converted to depth using the derived model is displayed
in Fig. 14, and shows the structure at the M20 target in the footwall not present on the
time section (Fig. 6). The time migration of the stack without the shallow gas effect
removed was also depth converted (Fig. 15). Note the difference in reflector continuity
and the improvement in ability to interpret the M20 horizon in the shadow of the fault
on the data with the residual static correction in Fig. 14.
Further improvement was achieved by applying poststack depth migration, using
the derived velocity/depth model (Fig. 16). Comparison with Fig. 14 shows improved
imaging of the M20 in the vicinity of the fault, and also shows significant enhancement
in imaging below M20. This demonstrates the improvement in lateral positioning pro-
vided by the depth migration. Although nonhyperbolic events were still evident after
removing the shallow gas effect, and the result of mis-stacking can be seen in the broad
wavelet character in the footwall events between M40 and M20. The poststack depth
migration does provide much improved imaging in the fault shadow. To further improve
the imaging, prestack depth migration would be necessary.
Finally, the depth-migrated data converted to time, using a time-variant, spatially
invariant velocity function, is shown in Fig. 17. When this result is compared to the
conventional time migration in Fig. 4, the critical role played by the velocity variations
over this prospect can be appreciated.

Distinguished Instructor Short Course s 8-5

 The Seismic Velocity Model as an Interpretation Asset

Conclusions
Detailed evaluation of the velocity field on proprietary data over the K prospect verified
that the velocity gradient across the major growth fault can generate a depth structure
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

not evident in time. Ray tracing was effectively used to build confidence in the velocity
field by analysis of both prestack and poststack data. In addition, problems caused by
shallow gas anomalies were identified and corrected. The use of residual statics
removed the main problem in stacking the data correctly prior to applying poststack
depth migration to verify the difference in structure between time and depth.

Fig. 1.
On this depth map of the primary tar-
get, the M20 horizon, the outline of
the superimposed time structure
demonstrates the importance of depth
converison in generating a prospect.

Fig. 2.
This is a full offset stack section of a
proprietary data line, which is a dip
line across a major fault. The stratig-
raphy is annotated to highlight the
major growth intervals. Note the area
of shallow gas and associated multi-
ples between shotpoints 400 and 440.

8-6 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 3.

An iso-velocity plot of the

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

stacking velocities clearly

displays the slower veloci-
ties in the hanging wall
section, when compared
to the equivalent times in
the footwall. Also, the
stacking velocities vary
rapidly between shot-
points 330 and 480
because of the nonhyper-
bolic moveout of the ray-
paths in the gathers across
the fault.

Fig. 4.
This is the time migration
of the stack section shown
in Fig. 2. Note the M20
event dipping down into
the fault and no evident
trap. Note also the severe
“bust” in the data at 4.2 s
and shotpoint 400, which
severely degrades the
reflector continuity and
interpretation of M20 in
the immediate vicinity of
the fault.

Distinguished Instructor Short Course s 8-7

 The Seismic Velocity Model as an Interpretation Asset

Fig. 5.
This near trace stack is of the near
3,000 ft of the cable, and represents
about the near one-fifth. A sag is
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

clearly evident in the data below the

shallow gas anomaly. Compare this to
Fig. 2, which contains all offsets in
the area of the sag. The full offset
range mis-stacks the event, resulting
in a discontinuity of the reflector.

Fig. 6.
This migration of the full offset stack
has residual statics applied. Note that
the discontinuity along M20 has now
been removed, resulting in significant-
ly improved imaging.

8-8 s Society of Exploration Geophysicists

 Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 7.
This horizon-based velocity analysis (HVA) is along the M40 reflector between shotpoints 210 and 510. The location
of the shallow gas anomaly and the fault are annotated.

Fig. 8.
This image shows the HVA on the data with residual statics applied. The decrease in velocity associated with the hang-
ing wall has now been removed, and the major change in velocities from footwall to hanging wall is now clearly seen.

Distinguished Instructor Short Course s 8-9

 The Seismic Velocity Model as an Interpretation Asset

Fig. 9.
This is the depth-based structural
model derived from the time-based
interpretation and the velocity analy-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

sis described in the text. Note that the

M20 horizon in the footwall is dip-
ping away from the fault plane, show-
ing that lateral velocity changes across
the fault can generate structures in
depth that are not observed in the
time domain.

Fig. 10.
This is the modeled zero-offset nor-
mal ray response. This modeling
analysis shows good agreement with
the time section (Fig. 2).

8-10 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 11.
This modeled, zero-offset image ray
response shows good agreement with
the migrated section (Fig. 4).
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 12.
This is a comparison of the observed and computed CDP gathers at three locations across the fault with normal move-
out applied using the picked stacking velocities. The locations of these gathers are annotated on Fig. 9.

Distinguished Instructor Short Course s 8-11

 The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 13.
This is an equivalent display to Fig. 12, but residual statics have been applied and ray traced through the model with-
out the shallow gas anomaly.

Fig. 14.
This is the time migration after it has
been vertically converted to depth
using the derived model. Note the
structure at the M20 target in the
footwall, which is not present on the
time section (Fig. 6).

8-12 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 15.
This depth conversion of the time
migration of the stack has had the
shallow gas effect removed. Note the
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

difference in reflector continuity and

the improvement in the ability to
interpret the M20 horizon.

Fig. 16.
This is the poststack depth migration
generated using the derived velocity
model (Fig. 16). A comparison with
Fig. 14 shows improved imaging of
the M20 in the vicinity of the fault,
and significant enhancement in imag-
ing below M20.

Distinguished Instructor Short Course s 8-13

 The Seismic Velocity Model as an Interpretation Asset

Fig. 17.
This is the depth-migrated data,
which has been converted to time
using a time-variant, spatially-invari-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ant velocity function. Compare this

image with Fig. 4, which is also in the
time domain. The velocity variations
critically affect both the apparent
structures in the time domain and the
interpretability of the events.

8-14 s Society of Exploration Geophysicists

 Phil Schultz

Analyzing Conflicting Depth Conversions

Amoco
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Instructor’s Notes: Amoco was exploring offshore Gabon in presalt sediments and facing the
problem of accurate structural mapping in depth. Depth mapping in this area is complicated
not only by the usual problem of high salt-sediment velocity contrasts, but also by the pres-
ence of high-velocity carbonates in the shallow section.
How to map in depth accurately? Presalt imaging with depth migration was not
required, as a reasonably accurate time-structure map was available. The issue in this study
is choosing the best way to model the velocities for depth conversion. Amoco tried three
alternate approaches: 1) A Faust single-layer velocity model, 2) a multilayered model based
on well data, and 3) a multilayered model based primarily on seismic velocities. This study
reports on the results of the comparison.

A Case Study of Time-To-Depth Conversion from the Gryphon-Marin

Permit Area, Gabon
T. C. Stiteler, Amoco Gabon B. V., Houston, C. Crabbe,
Amoco Production Company, Houston,
R. M. Priem and P. W. Johnson, Priemere GeoTechnology LLC, Houston

Abstract
Past efforts at regional time-to-depth conversion of presalt horizons offshore Gabon
have been hindered by complicated geology and the large volume of data. In this study,
three different methods of depth conversion were employed in an effort to identify a
preferred method for the area.
The three methods can be described as follows:
U Method 1: a layer-cake method using curve fitting of interval velocities from well
data,
U Method 2: a layer-cake method using curve fitting of interval velocities from seismic
stacking velocity data, and
U Method 3: a single-layer method that uses a Faust curve that is fit to seismic stack-
ing velocities.

Both methods 2 and 3 use the same input stacking velocities, which are composited
and calibrated to available well control. Also, methods 1 and 2 both use the same four-
layer velocity model.
The data set consists of 15,000 line-km of 2D seismic, 26 well velocity surveys, 16
sonic logs and 11,600 2D stacking velocity functions. Interpretation of the seismic and
well data suggests a velocity regime consisting of four distinct layers: surface to top

Distinguished Instructor Short Course s 8-15

 The Seismic Velocity Model as an Interpretation Asset

carbonate, carbonate, salt and presalt. Time structure maps from each of these horizons
were converted to depth using each of the three methods.
Evaluation of combined results from the three methods yielded unexpected benefits
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

that were not readily apparent from a single method. The primary benefits of the collec-
tive evaluation were providing a qualitative and quantitative assessment of time-to-depth
conversion for the block and helping to focus work in areas where risk could be mitigat-
ed. Comparison tools such as difference maps, cross sections and mistie reports were
useful in identifying problems with the input data and interpretation, and evaluating the
risks associated with trap definition. Also, cycle time was greatly reduced by computer
programs that were created to handle repetitive gridding and mapping operations.

Introduction
Careful depth conversion is necessary for presalt exploration offshore Gabon. Past
efforts have shown that the complicated structure and the varying rock properties of the
section cause large errors in depth prediction. This is particularly true where large dif-
ferences in thickness and velocity of overlying material distort the trapping geometry of
the presalt target on seismic sections, making time-to-depth conversion difficult.
Careful evaluation was needed to determine the presence and volume of potential
prospects and the associated risk.
In this study, we found that a combination of time-to-depth conversion methods
was preferable over any single method. Evaluation of the different results provided a
qualitative and quantitative assessment of time-to-depth conversion and helped to focus
work in areas where risk could be mitigated.

Physical Problems with Depth Conversion

A series of simple models illustrates the effects of velocity on time-to-depth conversion
(Fig. 1). A flat horizon on a depth section that is overlain by material with various
thicknesses and velocity develops a seismic event that is distorted by the overlying
material. This is analogous to the 2D seismic section after processing. Depth is deter-
mined from the time interpretation and interval velocity that is derived from well and
seismic control. It is important that the assumed velocities be well defined for proper
time-to-depth conversion. If the assumed velocity of the overlying section is too fast,
then underlying horizons of the converted depth section will appear to be pulled up
under the faster material, and if too slow, will appear pushed down. In either case, a
false break in reflection will appear beneath a sharp change in velocity of overlying
material and may be misinterpreted as a trap.

Velocity Model and Basic Depth Conversion

The velocity problem was handled in the study area by careful analysis of well and seis-
mic-derived velocities. We determined that the velocity in the area was best described
by four velocity layers as described in Fig. 2: 1) surface to top Cretaceous carbonate,
2) Cretaceous carbonate, 3) salt and 4) presalt. The bounding horizons for these layers
were interpreted as top Madiela carbonate, top salt and base salt. It was also recognized

8-16 s Society of Exploration Geophysicists

 Phil Schultz

that, with the exception of the salt, the interval velocity of each lithology increased with
depth as described by Pennebaker and Sattlegger.1,2 Many authors attribute the increase
in velocity with depth of a given lithology as an effect of compaction.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Salt is the exception. According to Jackson, et al.,3 salt develops a polycrystalline

microstructure of low permeability during accumulation and remains almost incompress-
ible throughout subsequent burial. This agrees with our observation of a constant salt
interval velocity of 4500 m/s from well check shot surveys through thick salt intervals.
To take advantage of these observations, well and seismic velocity data were fit to
functions of depth and time, and key parameters were gridded and mapped, forming a
three-dimensional description of the velocity over the area. Depth conversion was done
by grid operations on interpreted time grids.

Limited Data Quality

Analysis of the area was complicated by coarse 5–10 km spacing of 2D seismic data and
the limits of 2D time migration. Interpretation of the complicated structure makes it obvi-
ous that much of the reflection data are out of the plane and are not migrated properly
with two-dimensional migration. Nonetheless, areas with structural potential can be iden-
tified and analyzed for economic assessment and determination of a future work program.
To manage these problems, time-to-depth conversion was limited to vertical depth conver-
sion methods that were suitable for scoping analysis of leads from 2D seismic.
Also, the limited frequency content and resolution of the data makes interpretation
of horizons, such as the top and base of salt, difficult. In particular, the base salt reflec-
tion is prone to time distortions due to changes in overlying structure. The interpreter
is often challenged to identify a true presalt structure in areas where a combination of
resolution and velocity problems exist. To combat these problems, an iterative approach
to interpretation was used to minimize time, velocity and depth errors—both at well
control and at leads below complicated structures. Also, different methods of depth
conversion were used to identify uncertainty. The methods employed were also
designed to be accurate enough to assess the risk of leads and previously drilled areas.

Data Preparation
The data set gathered for this study was large, consisting of 15,000 line-km of 2D seis-
mic, 26 well velocity surveys, 16 sonic logs and 11,600 2D stacking velocity functions.
Sources include corporate relational databases (Oracle/PPDM), application databases
(Landmark, LogM), hard copy, 8mm tape and floppy disk files. All data were imported
to OmniVel, a velocity processing system that was used for editing and compositing of
the velocity data. Well velocities were also determined from integration of sonic log data
in LogM, a synthetic seismogram program.
Preparation of the seismic stacking velocity functions was done using the OmniVel
application.4 Preparation included: 1) an initial editing step to remove gross keypunch
and line position errors, 2) conversion of seismic two-way time and stacking velocity to
interval velocity, depth and other significant parameters using Dix equations,5 3) dip
correction using dip and azimuth computed from initial seismic interpretations,

Distinguished Instructor Short Course s 8-17

 The Seismic Velocity Model as an Interpretation Asset

4) regression filtering to remove extraneous points, 5) composite averaging to improve

signal-to-noise ratio, and 6) calibration to well velocities.
Well-related velocity data consisted of check shot surveys, vertical seismic profiles
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

(VSPs) and sonic logs. All time-depth functions were referenced to sea level and some
functions from shallow wells were extrapolated to reach the deeper horizons.Time-
depth pairs for each key velocity horizon were then interpolated in Microsoft Excel.
For wells with a sonic log and no velocity survey, a time-depth curve was generated
from integration of a sonic log using LogM.

Time-to-Depth Conversion
Three time-to-depth conversion methods were compared: WellT2D, Sas3dM and the
Faust method.
WellT2D is a layer-cake method based on well data. Time, depth and interval veloci-
ty data from wells were analyzed in velocity layers. For the first layer, a binomial curve
was fit to each well (Fig. 3) and “A” and “B” coefficients from these curves were grid-
ded between well locations over the study area (Fig. 4). The “A” and “B” parameters
were used to define a 3D relationship between time and depth over the area. In cases
where well control was sparse, or absent, pseudo-well parameters were created so the
study area would be covered. Zmap macros were created to perform the repetitive grid
operations.
For the carbonate, salt and presalt intervals, thicknesses were calculated from time
between the top and bottom of the intervals and compaction curves calculated from well
data. This method used a gross linear fit of interval velocity versus depth to determine
the slope, or general compaction term, “m” for the study area (Fig. 5). Then, a con-
strained linear fit is made through each well to determine the velocity intercept parame-
ter while keeping the compaction rate, “m” constant. Then map grids were made from
intercept parameters, describing the velocity of the intervals over the study area (Fig. 6,
map of carbonate intercept velocity parameter “Vb”). Finally, depth maps were created,
adding layer thicknesses together to form deeper maps in a “layer-cake” fashion. A time
map and depth map of the presalt intervals are shown in Figs. 7 and 9, respectively.
Sas3dM is a layer-cake method based on a seismic velocity data that has been com-
posited and calibrated to well data and interpreted time grids of key velocity horizons.6
First, seismic velocity data is processed in the OmniVel program to create a uniform
grid of velocity data that covers the study area. Then, time and interval velocity data are
extracted between interpreted time horizons corresponding to velocity boundaries. In a
similar fashion to the WellT2D method, a linear fit is made to time versus interval
velocity data to determine a compaction slope, and a constrained fit is made to the data
at each point on the velocity grid over the area. Likewise, intercept parameters are grid-
ded and used to convert interpreted time grids to depth.
The Sas3dM program also provides many interactive statistical analysis tools to
refine seismic velocity data and exclude poor, or erroneous, data. Simple quality control
plots were found to be useful, such as the depth cross section plot, Fig. 8. The primary

8-18 s Society of Exploration Geophysicists

 Phil Schultz

benefit of the program is that it is able to provide reasonable time-to-depth conversion

in areas where there is little, or no, well velocity control, and incorporate well data
when it is present.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The Faust method is a single-layer method in which interpreted seismic horizons

are converted to depth by fitting a Faust equation7,8 to seismic velocities that have been
composited and calibrated to well data. Time maps were converted to depth using the
Sattlegger program.2

Evaluation
After three iterations of time interpretation and depth conversion, time misties aver-
aged 10 ms with a standard deviation of 20 ms. The average depth misties to the top
Salt were 40 m with a standard deviation of 200 m. Early runs favored the WellT2D at
well locations, but heavier weighting of the well data in the Sas3dM and Faust methods
made the quantitative assessment from well ties essentially the same.
A qualitative assessment of time-to-depth conversion was based on comparison of
depth maps and difference maps from the three methods. Initial inspection noted a sim-
ilar roughness to the WellT2D and Sas3dM depth maps (Figs. 9 and 10, respectively),
and a smoother appearance to the Faust method map (Fig. 11). This was attributable to
the ability of the layer-cake methods to model sharp changes in overlying material and
the Faust method’s tendency to smoothly vary velocity across these zones. For the pur-
pose of this study, the layered-velocity methods were preferred, because it was impor-
tant to recognize faulting in the base salt section and to distinguish that from false
breaks due to the complexity of the overlying material.
Also, inspection of seismic, depth-converted maps and cross sections led to classify-
ing false effects in the presalt section due to 1) the presence or absence of salt, 2) fault-
ing of the carbonate section, 3) synclines between salt-motivated features that are filled
with slow clastic material, and 4) uncertainty in the interpretation of the top and base
salt. Identification of these pitfalls allowed the interpreter to focus on areas where fault
interpretations were less risky and decide which areas were worth reinterpreting to
attempt to lower the trapping risk.
Difference maps were found to be very useful in evaluating the large volume of data
and identifying subtle problem areas (Fig. 12). These maps were created by subtracting
a depth map created with one method from that of another. A good qualitative assess-
ment of the time-to-depth conversion methods was evident from the small differences
in the maps over much of the area. In addition, the difference maps clarified areas
where the results varied significantly, and allowed a quick focus of problem-solving to
these areas.
For example, a WellT2D method problem area was identified and attributed to poor
extrapolation of velocity parameters outside of well control. To fix this problem, pseu-
do-well values were added to constrain the grid extrapolation error. Likewise, an area
was identified where Sas3dM and Faust results were suspect due to poor seismic data
quality and sparse control at the end of seismic lines, where seismic-derived velocities
are often questionable. As a result, seismic data from this area were recalibrated to well

Distinguished Instructor Short Course s 8-19

 The Seismic Velocity Model as an Interpretation Asset

data and reinterpreted. Also, a subtle effect due to a grid node shift of 1 km was readily
identified and the error in the grid translation program was corrected.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Conclusions
1. A four-layer velocity model was developed for the area. We found that layer-cake
methods were necessary to sufficiently describe complexities in the velocity field
and properly convert time to depth.
2. A preferred time-to-depth conversion process was developed that involved compari-
son and contrast of three different methods of time-to-depth conversion involving
differing focus on well and seismic velocity data.
3. Accuracy of time-to-depth conversion was determined from misties at well loca-
tions. These misties yielded a quantitative assessment of structural risk of prospec-
tive areas. Consistent misties from various methods increased confidence in the
results and minimized uncertainty.
4. A qualitative assessment of the depth prediction was determined from difference
maps between well- and seismic-based depth conversion methods. The difference
maps showed consistency over most of the study area and allowed future work to be
focused on areas where the differences were large.
5. The depth-conversion process was streamlined through development and applica-
tion of the OmniVel, Sas3dM and WellT2D products. These tools and work process-
es will be reused for future projects.

References
1
Pennebaker, E. S., “An Engineering Interpretation of Seismic Data,” SPE of AIME,
1968, No. 2165.
2
Sattlegger, ISP003 Interpretive Seismic Processing System, Terminal Site Manual,
1993, pp. 172–197.
3
Jackson, M. P. A. and Talbot, C. J., Geological Society of America Bulletin, 1986, Vol.
97, No. 3, pp. 305–323.
4
Johnson, P. W. and Priem, R. M., OmniVel: A Velocity Data System for GeoScience
Exploration & Production, Users Guide, 1996, Version 1.5.
5
Dix, C. H., “Seismic velocities from surface measurements,” GEOPHYSICS, 1955, Vol.
20, pp.68–86.
6
Priem, R. M., Sas3dM: An Automated Process for Creating 3-D Velocity Models and
Depth Structure Maps, Users Guide, 1996, Version 0.3.
7
Faust, L. Y., “Seismic velocity as a function of depth and geologic time,” GEOPHYSICS,
1951, Vol. 16, pp.192–206.
8
Faust, L. Y., “A velocity function including lithologic variation,” GEOPHYSICS, 1953,
Vol.18, pp. 271–288.

8-20 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 1
This schematic illustrates the effects
of velocity errors when converting
seismic data from time to depth.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Velocities are more critical in depth

conversion than in any other process-
ing operation involving velocities.

Fig. 2.
A general velocity model is shown for the Gryphon-Marin Area, Gabon.

Distinguished Instructor Short Course s 8-21

 The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 3.
Surface to top carbonate two-way time (TWT) versus depth.

Fig. 4.
Map of well velocity parameter A—
from surface to top carbonate.

8-22 s Society of Exploration Geophysicists

 Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 5.
Carbonate depth versus interval velocity.

Fig. 6.
Map of well intercept velocity para-
meter Vb, Madiela carbonate interval.

Distinguished Instructor Short Course s 8-23

 The Seismic Velocity Model as an Interpretation Asset

Fig. 7.
Presalt time-structure map.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 8.
First iteration cross section from the process Sas3dM.

8-24 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 9.
Presalt depth-structure map, WellT2D
method.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 10.
Presalt depth structure map, Sas3dM
method.

Distinguished Instructor Short Course s 8-25

 The Seismic Velocity Model as an Interpretation Asset

Fig. 11.
Presalt depth structure map, Faust
method.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 12.
Presalt difference map, Sas3dM-
WellT2D methods.

8-26 s Society of Exploration Geophysicists

 Phil Schultz

Finding Closure in Depth, But None in Time

Norcen Explorer
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Phil Schultz

In the Gulf of Mexico, Norcen located a high-amplitude anomaly on the up-thrown side
of a normal fault at 3.1 s on the seismic time section (Fig. 1). The high amplitude
occurred against the fault, strongly suggesting trapped hydrocarbons. But there was a
problem. When the structure was mapped in time there was no closure (Fig.2). The
structure map showed extremely gentle structural relief, so there could have been closure
with only a small lateral variation in the velocity model. However, without independent
confirmation of the velocity model, the prospect was too deep to risk drilling a well.
Sensibly, Norcen decided to make use of existing well data in the neighborhood of
the prospect, where time-depth curves were available. Norcen knew that they would
need more than one well to resolve the structural closure issue. Three wells near to the
prospect had time-depth information and were used to calibrate the existing velocity
model derived from the seismic time domain processing.
After the well data were incorporated into the velocity model, the structure could be
mapped in depth (actually, the map was converted to depth). Now, the Norcen team
had the confirmation of structural closure required to complete the geological picture
(Fig. 3).
Were three wells sufficient or were four actually needed? After all, three points map
to a plane, and at least four points are needed to map curvature. Norcen knew that
three wells may be sufficient. Why?

Fig. 1.
The seismic time section shows the
prospective high-amplitude horizon at
3100 ms on the up-thrown side of the
interpreted normal fault, sloping
downward to the west. A well at loca-
tion V1 is shown on the section. This
well is one of three used to calibrate
the velocity model for depth conver-
sion.

Distinguished Instructor Short Course s 8-27

 The Seismic Velocity Model as an Interpretation Asset

Fig. 2.
The time structure map of the
prospective horizon is shown with the
zone of high-amplitude shaded. The
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

seismic line meanders from SE to NW.

The high-amplitude behavior of the
horizon against the fault strongly sug-
gests closure and hydrocarbon accu-
mulation, but there is actually no
closure seen on the time structure
map. CI = 10 msec; V1, V2, V3 are
velocity surveys used to make correct-
ed time maps.

Fig. 3.
After using three wells to calibrate the
velocity model, the time structure map
was converted to depth. The well loca-
tions are at V1, V2 and V3. This depth
structure map now shows a very gen-
tle, but definitive, structural closure.
The contour intervals are 10 m. How
can only three calibrating wells turn
no closure into closure? Three points
define only a plane; they cannot create
additional curvature. Was there an
error in the depth conversion proce-
dure? CI = 10 msec.

8-28 s Society of Exploration Geophysicists

 Phil Schultz

Determining Depth Structure Under a Channel Fill

Paradigm Geophysical
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Using Velocity Modeling to Assist Time-to-Depth Conversion

David Chao

Explorationists using seismic data always face the time-to-depth conversion problem.
Finding useful data for this conversion process can be difficult, and choosing the appro-
priate conversion technique involves more than just personal preference or settling on a
preferred method. Rather, identifying the optimum time-to-depth conversion technique
for a given area requires first building an accurate velocity/depth model. This entails
understanding local geologic control on velocity variation, which is a function of rock
lithology, age and depth of burial.
A time-to-depth conversion system encompasses more than simply converting time
data to depth. Key are good data quality control (QC) and analysis. Data QC helps to
identify and correct any data problems. Data analysis extracts information from the data
on hand that are useful for building a geologically correct velocity/depth model. Such
useful information can include seismic velocity data, well tops, check shots, sonic logs,
VSP (vertical seismic profile) surveys and time interpretations. How to use this type of
information to build a geologically valid velocity/depth model is the challenge.
The following text uses data from an offshore Australia project to reveal how to use
velocity modeling to build an accurate velocity/depth model.

Background
At the time of this exercise, the chosen offshore Australia prospect, which lies in 80 m
(263 ft) of water, had both seismic and well data available from 14 exploration wells—
all had check shot data and five had sonic logs (Fig 1a–b). To simplify this exercise, we
considered a two-layer model (Fig 2).
The first layer is a channel fill whose velocity is depth dependent (Fig 3). The sec-
ond layer lies from the base of the channel to the top of the reservoir (Fig. 4). Its base is
an unconformity surface sealed by fine-grained marine clastics. Fig. 5 shows an interval
velocity profile through the two layers.
A large lateral velocity variation at the top of the reservoir has complicated seismic
mapping of this field. Most of the velocity variation is due to sedimentary fill in the
channel. At the head of the channel the fill is a mixture of coarse skeletal fragments.
However, in the middle and distal portions of the channel the infilling sediment is
dense micritic limestone. The interval velocity of this micritic limestone may be as high
as 15,000 ft/sec, while the surrounding shales measure 8,000 to 9,000 ft/sec, which cre-
ates a velocity pull-up.
The migrated time maps of these two surfaces show that the structure time-high of
the reservoir corresponds to the deepest portion of the channel (Figs. 3 and 4). Is this

Distinguished Instructor Short Course s 8-29

 The Seismic Velocity Model as an Interpretation Asset

structural high real or is it caused by the velocity pull-up? Solving this problem requires
estimation of the velocity variation due to lithology changes in the overlying section.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Data Q/C
Scatter plots and crossplots of different data types revealed that data quality was a prob-
lem. For example, the Fig. 7 crossplot of well depths against migrated time maps shows
several aberrant points. These could be caused by mispicked well tops, errors in seismic
interpretation or a true velocity variation at different well locations. We returned to the
interpretation and velocity analysis to evaluate the causes of these anomalies. In this
case, they were erroneous correlations between the logs and the seismic.
In this example, understanding the causes and estimating the size of time anomalies
is essential for successful depth conversion. The time structure map of the target surface
shows a total relief of about 400 ms. The time pull-up map is computed from the cur-
rent velocity model, and predicts the time pull-up caused by the high-velocity channel
(Fig. 8). It represents the part of the time structure map that is not real. The total pull-
up across the channel is about 150 ms.
We compared sonic logs to interval velocity functions derived from seismic veloci-
ties to see how well the two types of functions correlate with each other. In Fig. 9, the
sonic log from well 50 at the deeper part of the channel is plotted along with the inter-
val velocities from the nearby shot point locations. In this instance, these two functions
match well.

Data Analysis
A very important aspect of time/depth conversion is analyzing the information content
of the available data, especially the seismic velocity information and the relationships
between velocity and other factors such as depth, isochron, etc. In Figs 10a and 10b
interval velocities from well check shots are plotted against interval velocities derived
from seismic for layers 1 and 2.
For the first layer there is fairly good correlation, which indicates that the velocities
for this layer can be used to build a velocity model. For the second layer, the correlation
between check shot and seismic-derived interval velocity data is very poor. This may be
caused by anisotropy within the velocities of the second layer. In this case we chose to
de-emphasize the seismic data in layer 2.

Geostatistics
Geostatistical tools provide a way to integrate well control with seismic data to build a
velocity model and perform depth conversion. They provide better estimates of value
away from well control. A geostatistical approach analyzes data characteristics to under-
stand spatial continuity and provides a better way to integrate different data types via
kriging with external drift and co-kriging.
One very important aspect of geostatistics is that these techniques provide a better
understanding of uncertainty and risk prediction. Uncertainty analysis is useful in
choosing the velocity model for depth conversion.

8-30 s Society of Exploration Geophysicists

 Phil Schultz

To illustrate use of a geostatiscal tool to select a proper velocity model, we used 14

exploration wells from the designated area offshore Australia. Seven were used as seed
wells (i.e., they were used for data input) to develop an appropriate velocity model for
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the field. This model was then used to predict velocities for the other seven wells.

Mapping Velocities
There are two possible ways to map the velocity of the first layer. The first, conventional
approach divides the isopach with the isochron and multiplies by two. The result is an
interval velocity map, or Vint. The other uses the linear characteristics of the velocity
increase with depth by mapping out the instantaneous velocity (Vo) + kz, where k is the
velocity gradient and z is depth. We chose a value of the constant, k, based on the well
data, and determined Vo areally, based on the seismic velocity analysis.
Velocity maps for the seven seed wells were created using these two approaches.
Results are shown in Fig. 11. Both types of maps show a generally increasing trend from
south to north; however, the Vo map has a much more restrictive range than the Vint map,
indicating that it is probably a more natural choice for the velocity model in this case.
Fig. 12 shows the corresponding uncertainties for the two types of maps, Vint and Vo,
each derived using geostatistics. Uncertainty ranges from zero right at well control to
200 m/s away from a well. The range of uncertainty on the Vo map is much less than that
of the Vint map, again suggesting that the gradient model is a better choice in this case.
In Fig. 13, the upper interval velocity (Vint) cross section shows a slow velocity (red)
ranging to high velocity (green) in the deeper part of the channel. The lower instanta-
neous velocity (Vo) cross section shows a linearly increasing velocity from shallow to
deep. The velocity is relatively constant at a reference depth. In essence, the gradient
model profile is depth-normalized; velocity interpolation is constrained by the depth of
burial. This appears to be a more reasonable velocity model in this case.

Extrapolation
The two velocity maps created were then used to predict velocity at the seven other
wells. Results are shown in Table 1. From this study, we found that using geostatistics
to create uncertainty maps helped with error prediction. Error is minimized by identify-
ing the method that has less geostatistical uncertainty. As illustrated, this requires
knowledge of the geological controls on velocity variation. The method with least
uncertainty is the method that models the geological control on velocity variation in a
manner most consistent with both the well and seismic velocities.

Table 1.
V0 + kz Vint
(Instantaneous)

Average Uncertainty from Uncertainty Map 86 232

Average Error in Velocity Prediction 81 175

Distinguished Instructor Short Course s 8-31

 The Seismic Velocity Model as an Interpretation Asset

From the well and seismic data available, we identified these characteristics in the
two model layers:
U Layer 1 (surface to base of channel)—velocity increases with depth for all wells,
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

lateral velocity change is related to facies change.

U Layer 2 (base of channel to top of the reservoir)—velocity is relatively constant in
the interval and not depth dependent, lateral changes are related to facies.

With this understanding we used the following strategy for velocity building:
U Layer 1:
1. Use checkshots to determine average vertical velocity gradient and spatially
varying Vo (reference velocity).
2. Analyze the spatial relationship between well velocity (sparse) and seismic data
(dense continuous coverage).
3. Use a geostatistical tool to integrate well and seismic information to produce a
velocity model that accounts for both strong vertical and lateral velocity gradients.
U Layer 2:
1. Use checkshot data to determine layer interval velocity.
2. Analyze the spatial relationship between well and seismic data.
3. Use a geostatistical tool to integrate well and seismic information to produce a
velocity model that characterizes the lateral velocity gradient.
Figs. 14 and 15 are the resulting interval velocity maps of these two layers.
Fig. 16 is the final depth map of the top of reservoir surface (Layer 2) overlaid with
contours from the time migration of the same surface. It shows that the depth crest is
considerably southwest of the time crest. The time crest is about 50 m (164 ft) down-
dip and 5 m (16 ft) east of the true crest of the structure.
In addition to the 14 wells used for this time-to-depth conversion exercise, there are
another 41 production wells with depth information. The root mean square (RMS)
error of depth prediction at these production wells is 8 m (54 ft) with a spread of 18 to
24 m (59 to 79 ft). If only well velocities from 14 exploration wells are used in depth
prediction of the top of reservoir, it has an RMS error of 13 m (43 ft) with a spread of
29 to 64 m (95 to 210 ft ).
We can use the velocity/depth model to convert the seismic section to depth by
either scaling or migration. Fig. 17 is the scaled depth section from the same line as
shown in Fig. 6.

8-32 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 1a.
A base map of the study area shows
23 seismic lines and exploration
wells. The wells provide velocity and
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

depth interpretations, from check

shot, tops and sonic logs.

Fig 1b.
The seismic lines and well names that
accompany Fig. 1a are given.

Fig. 2.
In this perspective view of the time
surfaces, the upper surface is the base
of the channel and the lower surface
is the top of the reservoir. In the time
view, the reservoir peak is directly
below the deepest portion of the
channel base.

Distinguished Instructor Short Course s 8-33

 The Seismic Velocity Model as an Interpretation Asset

Fig. 3.
A migrated time map of the base of
channel (Layer 1). Well spost are
shown.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 4.
A migrated time map of the top of
reservoir (Layer 2).

Fig. 5.
This sonic log is from well 50, which
is located in the deeper portion of the
channel. Layer 1 is the channel; Layer
2 represents the area between the base
of channel and top of reservoir. The
velocity for Layer 1 is depth depen-
dent, while the velocity for Layer 2 is
relatively constant.

8-34 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 6.
A migrated time section
for Line 40. Well 50, as
spotted on Fig. 1, is
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

shown in yellow.

Fig. 7.
This crossplot of well
depth and migrated time
from the seismic of the
same surface/top clearly
shows potential misties at
several well locations.

Fig. 8.
The time pull-up effect at
the top of reservoir, as
shown on this time struc-
ture map, is caused by the
high velocities in the
channel. This map is
based on the current
velocity model.

Distinguished Instructor Short Course s 8-35

 The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 9.
The seismic and downhole velocity functions are shown for well 50. The downhole data are purple.
Vertical axis is depth.

Fig. 10a.
Shown is a crossplot of check shot interval velocity data against seismic interval velocity data for the two layers. There
is good correlation between the two data types in Layer 1 (a) which suggests seismic velocities in this case are appro-
priate for velocity model-building.

8-36 s Society of Exploration Geophysicists

 Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 10b.
This image is the same as Fig. 10a, except it is for Layer 2. There is less consistency between the two data types in this
image.

Fig. 11.
These velocity maps for Layer 1 were generated from the data taken from the seven control wells. The same color scale
is used for each. The upper map is of instantaneous velocity (Vo); the lower is of interval velocity (Vint). To the right of
each map is its distribution histogram. Note that the Vo map shows a narrower scatter than the Vint map.

Distinguished Instructor Short Course s 8-37

 The Seismic Velocity Model as an Interpretation Asset
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 12.
These geostatistically derived uncertainty maps are displayed in the same color scale. The upper is from the Vo data
while the lower is from the Vint data. The Vo has a much smaller uncertainty than the Vint.

Fig. 13.
Velocity cross sections were built using two different approaches.

8-38 s Society of Exploration Geophysicists

 Phil Schultz
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 14.
The Vo map for Layer 1. Vo is the instantaneous velocity at the top of the layer in the model with a non-zero gradient.

Fig. 15.
The Vint map for Layer 2. Vint is the interval velocity for the model built with a gradient equal to zero.

Distinguished Instructor Short Course s 8-39

 The Seismic Velocity Model as an Interpretation Asset

Fig. 16.
The final depth map for
the top of reservoir (base
of Layer 2) includes an
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

overlay of the contours

from the migrated time
section of the same sur-
face. It clearly shows that
the structural peak, in
depth, is to the southwest
of the time peak.

Fig. 17.

This depth section of Line 40 was created by depth conversion from the time-migrated section using the velocity
model created. Well 50 is again spotted.

8-40 s Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Getting Clues to the Lithology

9
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Getting Clues to the Lithology

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Finally, velocities can give clues to lithology. While velocity modeling from seismic
moveout analysis is normally not of sufficient detail to predict lithology at a scale of
seismic events, it can give indications of lithology in packages or groups of formations.
In the following case study, no wells were present in the vicinity, and the lithological
picture of the area was based on geological inference. In such a case, the velocity trend
from a detailed model analysis was used to eliminate some lithological possibilities and
qualify others.

U Veritas: Investigating the lithological character of the Faroe Shetland escarpment

Distinguished Instructor Short Course s 9-1

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Investigating the Lithological Character of the

Faroe Shetland Escarpment
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Veritas DGC
Phil Schultz

Instructor’s Notes: The following figures and captions were supplied by Richard Morgan of
Veritas. They guide the reader through a lithological identification problem.
Veritas acquired nonproprietary marine seismic data near the Faroe Islands, where there
was no well control. Their primary objective was, of course, to image the data in the optimal
manner. However, an important associated problem was to determine the geological nature
of a set of clinoform structures. Conventional wisdom held that they were volcanic in origin.
Were they basalt? Were they sedimentary? Where they possibly a combination of episodic
volcanic and sedimentary processes? If so, what was their depositional history?
This study presents an investigation into the lithological nature of a suspected volcanic
depositional system with the help of velocities derived from prestack depth migration velocity
analysis and modeling. Although spatial resolution of velocity analysis from seismic data
lacks the fine resolution required for lithological identification of individual formations, an
entire depositional system can sometimes be determined when present geological knowledge
limits the choices to several alternatives. In this case, a “package” of layers was analyzed for
confirmation of its lithological nature.
What was the result of this study? As is often the case when seismic alone is involved in
the analysis, irrefutable confirmation of a geological premise is rare. In this case, the seismic
model-building helped Veritas to form the most likely geological interpretation. Have a look
at the study, follow the chain of reasoning and conclusions, and decide how you would
interpret the evidence.

Fig. 1.
Geological review: Late
Paleocene volcanism, cen-
tered in the west,
occurred shortly before
the onset of seafloor
spreading between the
Faroe Platform and
Greenland. Uplift of the
Shetland Platform accom-
panied the volcanism.
The uplifted platform
shed copious amounts of
sediments to the east, as
the Andrew and Forties
fans, and to the west as
time-equivalent fan
sequences. Whether simi-
lar sedimentary systems
existed off the Faroe
Platform, synchronous
with the volcanic events,
remains undetermined.

Distinguished Instructor Short Course s 9-3

 The Seismic Velocity Model as an Interpretation Asset

Fig. 2.
The Faroe Shetland
escarpment is a well-
known seismic landmark.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Clinoform structures are

clearly visible within the
body of the escarpment.
Although assumed to be
of volcanic origin, there
are no well data to support
this contention. The clino-
forms can be compared
with both sedimentary or
volcanic analogues,
depending upon the veloc-
ities used for depth con-
version.

Fig. 3.
This is a different view of
Fig. 2.

Fig. 4.
This is an example of the
Late Paleocene fan
sequences that prograde
eastwards off the Shetland
Platform into the East
Shetland Basin. These sed-
iments are broadly time
equivalent with the vol-
canics to the west and are
developed on a compara-
ble scale to the Faroe
Shetland escarpment
structure. The importance
of imaging beneath these
fans is self-evident.

9-4 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 5.
Field names have been
added to this image.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Fig. 6.
Details of the strategy for
using prestack depth
migration (PSDM) to
investigate the Faroe
Shetland escarpment
(FSE) are given.

Fig. 7.

Distinguished Instructor Short Course s 9-5

 The Seismic Velocity Model as an Interpretation Asset

Fig. 8.
The Faroe Shetland
escarpment was imaged by
conventional time process-
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ing.

Fig. 9.
The Faroe Shetland
escarpment was also
imaged by PSDM data dis-
played in the time
domain. Distortion of
reflectors beneath the
escarpment when dis-
played in time is to be
expected if the structure
includes high-velocity vol-
canic rocks.

Fig. 10.
Depth stretching of the
conventional time-
processed data introduces
distortion of reflectors
beneath the escarpment.
This effect is caused as
time processing has limit-
ed ability to deal with a
complex velocity struc-
ture.

9-6 s Society of Exploration Geophysicists

 Phil Schultz

Fig. 11.
The Faroe Shetland
escarpment imaged via
PSDM and displayed in
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the depth domain shows

greater coherence of
reflectors beneath the
escarpment without the
distortion observed in the
time-processed data. This
feature in itself is a mea-
sure of the improvement
in data migration using a
more complex but appli-
cable velocity field.

Fig. 12.
This is the velocity model
built from velocity analy-
ses using semblance plots
and gather flattening.
There are no wells within
200 km of this area. The
velocities derived show
complexity within the
escarpment and imply fast
rock composition.

Fig. 13.
This is a velocity field
built using a geological
model that interprets the
clinoforms to be sedimen-
tary. This velocity model
proved to be inapplicable.

Distinguished Instructor Short Course s 9-7

 The Seismic Velocity Model as an Interpretation Asset

Fig. 14.
This is a geological ana-
logue for the Faroe
Shetland escarpment. This
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

analogue includes inter-

layered, welded and
reworked volcanic flows
as well as other volcanical-
ly derived sediments.

Fig. 15.

9-8 s Society of Exploration Geophysicists

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The Ultimate Lesson

10
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

The Ultimate Lesson

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

A Final Word
Velocity is no longer just a processing parameter. It can provide essential interpretive
input. Accordingly, it benefits both the processing and interpreting geophysicists. Not
surprisingly, velocity models of sufficient detail to provide value to the interpretive
process normally require some interpretive input in their development. At the risk of
overgeneralizing, I’ll make the following statement:
Those velocity models built with the greatest amount of interpretive input generate the
greatest amount of interpretive benefit.
We saw examples in this volume of cases where the velocity model provided key
interpretive input, apart from acting as a processing parameter. In fault location, in well
positioning for production, in clarification of subtle features in the seismic image, in
resolving an accurate depth structure, and in other examples, the velocity model played
a key role in the interpretive process, and it required interpretive input. In all the exam-
ples we reviewed, the interpreter had direct input to the model, whether just for guid-
ance or for more substantial input.
Geophysicsts today are using a detailed velocity model to solve interpretation prob-
lems, some of which were previously addressable only with the drillbit. Obtaining a
high-quality velocity model does not require additional acquisition of data, but it does
require attentiveness to detail and diligence to applying systematic methodologies. The
benefits speak for themselves.

Distinguished Instructor Short Course s 10-1

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Glossary
11
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Glossary
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The definitions given here are meant to assist you in this course. The list is deliberately
brief and largely refers to terms specific to velocity modeling, especially those that
might be confusing when reading these notes. Similarly in presenting the definitions
here I have deliberately limited the discussion to the rather narrow scope of this text.
For a more detailed and thorough discussion of terms in exploration geophysics I
strongly recommend Bob Sheriff’s Encyclopedic Dictionary of Exploration Geophysics,
published by the SEG.

Anisotropy. A property of earth materials said to exist when seismic propagation vel-
ocities differ in the same medium (earth layer) depending on their direction of travel.
More generally this refers to the variation of any physical property depending upon the
direction in which it is measured.

Aperture. (esp. migration aperture) The horizontal distance over which a migration
operator gathers data in order to assemble an image of the reflector. The aperture gets
larger with depth. It is important to the quality of an image that sufficient data be
acquired to allow for data over the entire aperture.

Average velocity. Depth divided by vertical time. This velocity is commonly used for
time-to-depth conversion of structure maps. It is not directly measurable on seismic
data, but it is the velocity that a vertically-incident check shot survey tends to measure.

Check shot. Similar to a VSP (q.v.), in which signals from sources at the surface near a
wellhead are recorded downhole. The purpose is to derive an accurate time-to-depth
relationship at the wellbore location. The practical differences between a check shot and
a VSP are that the measurement is only made of the first arrival (no waveforms are
recorded, as they are with VSP) and the measurements are made at widely spaced
depths (a VSP is recorded at closely spaced depths).

Coherency inversion. A method of estimating propagation or interval velocities from

unmigrated prestack seismic data. It uses ray tracing over finite depth intervals and
finite azimuth slices to simulate the actual travel path of sound waves in the presence
of lateral velocity variations. With these simplifications, the effect on travel time of
many variants on a 3D velocity model can be estimated rather quickly, leading to some
indication of which model best suits the seismic data. This technique is not considered
to be as accurate as the velocity updating process in iterative depth migration, but is
often useful and cost effective as a first step in the model-building process.

Co-kriging. A variation on kriging in which the estimates of the unknown earth prop-
erty are based not only on a sparse sampling of that property but also on the distribu-
tion of some other, related observable.

Distinguished Instructor Short Course s 11-1

 The Seismic Velocity Model as an Interpretation Asset

Crossline. In a processed 3D survey, a seismic section perpendicular to the predomi-

nant receiver line orientation.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Demigration. An inverse operation to migration, applied to seismic data or horizon

interpretations made on migrated data. If demigration follows migration, and if the
same velocity field is used for both, then the output of demigration is the same (except
for numerical noise effects) as the original input to migration. As a concept, it is equally
valid prestack or poststack. It is most often applied to interpreted horizons, rather than
seismic data, to identify where those same events are on the unmigrated data.

Depth conversion. The conversion of seismic data from the time domain (where it was
recorded) to the depth domain. In correctly migrated data this conversion is simply a
matter of multiplying the one-way seismic time by the average velocity to that time. In
practice, proper depth conversion is much more complicated.

Depth migration. The process of repositioning (migrating) reflections from a point

directly below the source-receiver midpoint (CMP) to the subsurface location at which
the reflection actually occurred. Depth migration, whether done on prestack (PSDM) or
poststack (PoSDM) data, uses a more complete model of sound propagation in the earth
than does time migration. In particular, this approach can accommodate refractions of
the sound wave at nonvertical incidence angles. This allows for more accurate imaging
in the presence of dipping horizons and other types of lateral velocity variation. Depth
migration algorithms usually run longer than time migrations and have the further
requirement of a detailed velocity model, the creation of which can be time consuming
and expensive. The seismic data output from depth migration is often rendered in the
depth domain, although it can equally well be output in the time domain.

Diffraction. A phenomenon common to all waves, in the realm of seismic this is most
simply conceived of as the reflection from a single point reflector at depth (imagine a
large buried boulder reflecting seismic waves). The curve on a seismic section describes
a near-hyperbola with the apex at the point of reflection. Migration collapses this curve
to a point.

Dix formula. The formula used to calculate the interval velocity of a layer from known
values of RMS velocity at the top and bottom of that layer. Strictly speaking this formu-
la holds only for flat lying, parallel reflectors.

DMO. An acronym for dip moveout, which is a partial prestack migration (not a full
prestack migration). This process is carried out prestack and partially corrects for the
smearing of the image caused by lateral velocity variations.

Faust single-layer velocity model. A velocity model containing a single layer whose
velocity varies with depth according to a relationship attributed to Faust. Specifically,
the velocity increases with depth proportional to compaction (depth of burial) and age,
raised to a power (often 1/6).

11-2 s Society of Exploration Geophysicists

 Phil Schultz

FK or Stolt stretch migration. A very fast migration algorithm performed in the FK

(frequency-wavenumber) domain. The migration is exact for a constant velocity earth.
An approximation is built into the FK migration, allowing it to accommodate a slow
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

velocity variation in the vertical direction. The approximation amounts to a stretch of

the time axis of the input seismic data, so that it looks like it was recorded over a con-
stant velocity earth; hence the name, Stolt stretch (after its originator, Bob Stolt of
Conoco) or FK stretch migration.

Geostatistics. Generally a name for any statistical method applied to estimating the dis-
tribution of real properties of the subsurface (i.e., porosity, net pay, ore grade, seismic
velocity) from sparse samples of that property.

Global tomography. (more commonly, tomography) The generalized name given to

the process of estimating the properties of a medium (in this case the velocity) from
measurements of some energy that has passed through the medium (in this case sound
or more correctly seismic arrival time). In the case of velocity modeling, this becomes
an automated technique for amending and correcting the input or starting velocity
model using arrival times at specific events measured on the seismic data.

Gradient-based model. A layered velocity model in which vertical variation of velocity is

allowed within the layers, given by the formula V(z) = V0 + kz, where k is the gradient.

Grid model. A way to describe the velocity model whereby the velocities are given in
a 3D grid, rather than with layers defined by picked horizons. A horizon velocity model
can be converted to a grid model, but not vice versa, unless horizons are also input.

Hale McClellan transform. Refers to a mathematical operation that converts a one-

dimensional filter to two dimensions using a rotation transformation. It is normally
used to calculate high-fidelity spatial filters in 3D omega-x migration.

Horizon model. A short expression for a “horizon-based velocity model.” This is a

model wherein a set of interpreted horizons are an integral part of the velocity model.
The horizons mark significant velocity contrasts and delimit layers within which the
velocities vary only slowly.

Hyperbola. A mathematical curve that approximately describes the behavior of a recorded

seismic signal. On an unmigrated seismic section, it describes the diffraction curve (q.v.) of
a point reflector. On a CDP gather, it describes the curve of a reflection event from a flat
lying reflector, which is described by the NMO equation (see stacking velocity).

Image gather. A prestack depth migrated CDP gather. These are often displayed and
analyzed in the iterative process of velocity model-building through successive remigra-
tions of updated velocity fields.

Image ray. A conceptual ray that departs from the earth’s surface in a vertically down-
ward direction and travels to a point on the reflector (in general, not normally

Distinguished Instructor Short Course s 11-3

 The Seismic Velocity Model as an Interpretation Asset

incident). The concept is useful for predicting and sometimes correcting the lateral
misplacement of events after time migration. Time migration assumes that the correct
image ray is a vertical line, while in reality, in the presence of lateral velocity variations,
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the ray will refract and deflect from the vertical.

Imaging. The art of generating accurate representations of the subsurface from recorded
seismic data. It usually implies depth migration and rendering the seismic data in depth,
rather than time. In difficult areas, it requires a systematic approach to the estimation of
the velocity model.

Inline. In a processed 3D survey, a seismic section parallel to the predominant receiver

line orientation.

Interval velocity. The velocity at which sound propagates in a specific time or geologic
interval. Normally calculated using the Dix equation.

Isochron. The time difference between two interpreted geological horizons.

Isopach. The depth difference between two interpreted geological horizons.

Kinematic restoration. A geometrical process wherein the tectonic deformations (fault-

ing, erosion and sedimentation) that created the present-day geologic cross section are
modeled in reverse chronological order, for the purpose of confirming a working theory
of the geological history of a basin. The qualifier, “kinematic,” implies that compaction
is not modeled.

Kirchhoff migration. A method for time or depth migration that is based on the
Kirchhoff integral solution to the wave equation, hence the name. It is often the migra-
tion method of choice for prestack depth migration, because it can accept prestack
input traces from irregular acquisition geometries. For other migration methods, such
as finite difference, input data must be on a regular grid.

Krigging. A least squares based geostatistical method of estimating the distribution

of an earth property from a sparse sampling of that property. Named after the South
African mining engineer who first proposed the method.

Listric fault. Often called a growth fault, this is a normal fault whose fault plane be-
comes increasingly more horizontal with depth. During those geologic times in which
the fault has been active, one will see increased bed thickness near the fault plane on its
down-dip or hanging wall side.

MMstb. Million stock tank barrels of oil, an alternative to MMBOE or million barrels of
oil equivalent.

Migration. The process of repositioning (migrating) reflections from a point directly

below the source-receiver midpoint (CMP) to the subsurface location at which the

11-4 s Society of Exploration Geophysicists

 Phil Schultz

reflection actually occurred. The process may be performed on prestack or poststack

seismic data. The process can be thought of as similar to “triangulation,” whereby one
pinpoints an energy source by taking several observations of the distance to that source
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

from differing locations and plotting the locus of points that can possibly satisfy all the
difference observations. The result is a seismic section that accurately portrays an earth
cross section, either in the depth domain or in two-way vertical travel time.

Migration velocity. The velocity field or velocity model used to migrate the data. It is
usually not the same as the stacking velocity.

Monolithic intrusion. A geological intrusion of a single material or lithology, such as

halite, and more importantly here, exhibiting a single propagation velocity. In velocity
modeling, the problem to be solved with these bodies is to map their precise boundaries
since the interior velocity is well established.

MPP. An acronym for massively parallel processors, a distributed processor computer

design offering high-capacity computation. This architecture implies a distributed mem-
ory system. An alternative parallel machine architecture to SMP.

NMO. An acronym for normal moveout, which is the tendency for arrival times to
increase with increasing source-receiver offset, describing a near-hyperbolic curve. The
normal moveout correction removes this effect before the traces from a common mid-
point are stacked.

OIP. An acronym for “oil in place,” used in reference to the volume of hydrocarbons to
be found in a reservoir. Sometimes the word original is added to this acronym, or “origi-
nal oil in place” or OOIP. This amount is always greater than the volume that will be
ultimately recovered from the reservoir (called the reserves) owing to the inefficiencies
and economics of production.

Ray trace migration. Another name for Kirchhoff migration, which requires that trav-
eltimes are computed in order to accommodate ray bending, hence the name, ray trace
migration.

Residual migration. When the seismic data has been migrated with incorrect veloci-
ties, the output can be passed through a partial or residual migration to correct for the
errors caused by the incorrect velocities. This type of migration is recognized as valid
only for time migration, and is used because it is a fast and inexpensive alternative to a
full remigration with the correct velocity field.

Residual moveout. The remaining curvature of a seismic event, as a function of offset,

after normal moveout has been applied. The residual moveout may indicate an incorrect
velocity model or that the event moveout is nonhyperbolic. Measurement of the resid-
ual moveout enables an estimate of a correction to the velocity field (velocity update)
to be made.

Distinguished Instructor Short Course s 11-5

 The Seismic Velocity Model as an Interpretation Asset

RMS velocity. The root mean square average of the interval velocities down to a given
interface or event. It is usually assumed to be equivalent to the velocity estimated by nor-
mal moveout velocity analysis, but is so only to the extent of the RMS approximation.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

RMS approximation. The approximation that the stacking or NMO velocity (q.v.) is
equal to the RMS velocity (q.v.). This approximation is valid only in a horizontally-
layered earth, and for small offsets (i.e., offset << depth to reflector).

Salt proximity survey. A type of VSP survey performed near a vertical salt interface
in order to determine the location of the salt. The channeling of the seismic waves
through the high-velocity salt body results in a characteristic event arrival at the down-
hole geophone that can be processed to determine the geometry of the near-vertical
salt-sediment interface. This geometry is often very difficult to determine from the
surface seismic reflection data alone.

Semblance. A mathematical calculation on seismic data in CDP gathers measuring sim-

ilarity between traces in the gather as a function of time. The purpose is to assist the
detection of the correct moveout trajectory of reflected events. It is presented as an
image or contour plot of stacking velocity (q.v.) versus time (or occasionally, depth).
Peaks in the semblance plot represent likely velocity-time pairs.

SMP. An acronym for symmetrical multiprocessor. A computer architecture offering

tightly integrated design for parallel computation. It is this architecture to which the
term “shared-memory” normally refers. An alternative parallel design to MPP.

Stacking velocity. Also known as NMO velocity, it is the velocity parameter v used in
the equation, t2 = t20 + x2/v2, which is applied to CDP gathers prior to stack. This is the
velocity picked by the data processor during stacking or normal moveout velocity
analysis.

Stolt migration. See FK Stolt migration.

Tessellated surface. A term describing a series of scattered points in x,y space that
have been connected as vertices of triangles. It is used as a method of gridding and
mapping.

Time migration. A name given to what has been known for years as, simply, migration.
The expression time migration is used solely to distinguish it from depth migration.
Time migration does not as accurately handle rapidly varying velocities in the earth.
A common but incorrectly held belief is that time migration implies a straight-line
raypath assumption. Although time migration does not honor Snell’s Law ray bending
exactly, it does in fact perform an approximation ray bending.

Transverse isotropy. A common type of anisotropy, where vertical propagation veloci-

ties differ from lateral propagation velocities but the lateral propagation velocities are
independent of compass direction (e.g., E-W waves travel at the same speed as N-S

11-6 s Society of Exploration Geophysicists

 Phil Schultz

waves). This effect is often cited as the reason that depth-migrated data does not tie
well depths, since moveout velocities are closer to the lateral than the vertical velocity.
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Traveltime. The transit time of a seismic pulse from a source to a receiver, at the sur-
face or downhole, either as a direct arrival, a refracted arrival or via a reflection.

Traveltime trajectory. The traveltime of a single subsurface reflector as a function of

lateral position on the surface (for the case of a seismic section) or as a function of off-
set (for the case of a CDP gather).

TVD. An acronym for true vertical depth, as applied to downhole measurements. When
downhole measurements are made (logs, for example), downhole depth is measured
along the well trajectory. In the case of a deviated well, this measured depth (MD) must
be converted to the equivalent true vertical depth (TVD).

Velocity field. A two- or three-dimensional field of numbers describing the propagation

velocity. It can be given in either the time or the depth domain and is used to describe
the spatial variation of velocity for programs such as time migration. The velocity field
in this text is distinguished from the velocity model.

Velocity model. Normally understood to be a more systematic and detailed version of a

velocity field. Like a velocity field, it describes the spatial variation of propagation
velocity. It is often distinguished from the velocity field because it commonly includes
geological surfaces (sedimentary layers, erosional surfaces, or lithological boundaries) to
define layers within which the velocity varies slowly, and across which the velocity con-
trasts tend to be high, such as at a salt-sediment boundary.

Velocity pull-up. An apparent structural high on a seismic time section or map caused
by a shallower, high-velocity interval that is limited in areal extent.

VSP. An acronym for vertical seismic profile. One or more receivers are placed down-
hole to record a seismic pulse from the surface. The vertical traveltime and the full seis-
mic waveform are recorded.

Water-drive reservoir. A hydrocarbon reservoir in which the gas or liquid reserves flow
into the wellbore driven by water pressure in the porous rocks from below. This con-
trasts to a pressure-drive or pressure-depletion reservoir, which is driven by its internal
pressure. In a water-drive reservoir as production progresses the hydrocarbon/water
interface moves to shallower depths, implying that well placement needs to be at the
highest elevation of the reservoir.

Distinguished Instructor Short Course s 11-7

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Acknowledgments
12
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank

 Phil Schultz

Acknowledgments
Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

The work involved preparing a course such as this one cannot possibly be the work of
one person. From conception, this course was to be heavily dependent on case study
material. Working geophysicists are comfortable with learning “on the job,” and this
means learning by watching the successes and mistakes of others, and by observing the
chain of reasoning, whereby clues are pieced together and new methodologies some-
times have to be invented to fit the need.
This course will share with you some of the new approaches taken by your associ-
ates to build an interpretation from a relatively new source of evidence: the refined seis-
mic velocity model.
I had a lot of help. Primarily my support came from individual professionals who
took time from their principle activities to collect, organize and submit a case study for
this course. Their generous commitment of time made this course possible. The materi-
al they supplied will allow you to learn from their experiences and see how they used
velocities as an essential piece of the puzzle.
Because professional affiliations change so frequently, I have chosen to acknowledge
companies and individual professionals separately.
First, I thank the companies and their management whose policies allowed their
professionals to provide both published and unpublished material to this course, and to
their partners who gave permission and support to use the material. They are listed in
alphabetical order.

U 3DX Technologies

U Advanced Data Solutions

U Amerada Hess

U Amoco

U ARCO

U BHP

U British Petroleum

U Cox and Perkins

U DDD Energy

U Elf Exploration

U Enron

U Fairfield

Distinguished Instructor Short Course s 12-1

 The Seismic Velocity Model as an Interpretation Asset

U ForceEnergy (partner to ARCO)

U Kerr-McGee (partner to Phillips)

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

U Mobil Producing Nigeria

U Norcen Explorer

U Oryx Energy

U Paradigm Geophysical

U Phillips Petroleum

U Priemere GeoTechnology

U Sandia National Laboratories (partner to Oryx)

U Schlumberger (Geco-Prakla)

U Spirit Energy (Unocal)

U TGS

U University of Texas—Dallas (partner to Oryx)

U Veritas DGC

U Western Geophysical

Second, I give my sincere appreciation to those professionals who committed their

time and experience to share case studies with you. They are listed in alphabetical order.
The names in bold face type provided extensive support in preparation of this course.

U Craig A. Barker U Gene Colgan U Paul Garber

U Nick Bernitsas U C. Crabbe U Bob Godfrey

U Luis Canales U Francois Daube U Michel Goupillot

U Randall J. Carroll U Barry M. Davis U Claudia Groesbeck

U Herman Chang U Peter Duncan U Craig S. Hartline

U David Chao U C. E. Edoziem U Tim Hartnett

U Bob Choate U Duryodhan Epili U Scott A. Irvine

12-2 s Society of Exploration Geophysicists

 Phil Schultz

U P. W. Johnson U Thang Nguyen U James Sun

U Bill Kilsdonk U Michael O’Briain U T. C. Stiteler

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

U Eric Kubera U James O’Connell U Laurie A. St. Aubin

U Claude LaFond U Rob O’Conner U Jaime Stein

U James A. LeGate U Al Poprik U Tim Summers

U Mike LeNoble U R. M. Priem U Alastair Thomson

U Jacques Leveille U Joe M. Reilly U Paul A. Valasek

U Mike May U Robert R. Sartain U John P. VanDyke

U Mark McCuen U Mike Saunders U David A. Walker

U George A. McMechan U Paul Sexton U Kay D. Wyatt

U Marilyn Merryweather U Charles Sicking U David N. Whitcombe

U Richard Morgan U Paul Singer U Gary Yu

U Eugene H. Murray U Marcelo Solano

The SEG deserves mention for its foresight and vision in creating a program meant
to benefit all its members worldwide. Of particular note, Peter Duncan was the vision-
ary in initiating the move to create what eventually became this program, the Executive
Committee of the SEG lent its enthusiastic support and helped to shape its final form,
Ruth Ives of SEG Continuing Education provided essential support to make the vision
a reality, and Laura Pankonien and Martha Dutton ensured the professional appearance
and readability of this volume.
Finally, my family had the patience to tolerate my disappearance into my study
many evenings and weekends to prepare this course. Thanks and much love to my wife,
Sandy, and to my three girls, Natalie, Ashley and Kelly.

Distinguished Instructor Short Course s 12-3

Downloaded 09/25/15 to 129.78.139.28. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

This page has been intentionally left blank