The Common-Reflection-Surface Stack under Consideration ... - Sites
The Common-Reflection-Surface Stack under Consideration ... - Sites
The Common-Reflection-Surface Stack under Consideration ... - Sites
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
GEOPHYSIKALISCHES INSTITUT<br />
UNIVERSITÄT KARLSRUHE<br />
<strong>The</strong> <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> <strong>Stack</strong><br />
<strong>under</strong> <strong>Consideration</strong> of the Acquisition <strong>Surface</strong> Topography<br />
and of the Near-<strong>Surface</strong> Velocity Gradient<br />
Die <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> Stapelung<br />
unter Berücksichtigung der Topographie der Messoberfläche<br />
und des oberflächennahen Geschwindigkeitsgradienten<br />
Diplomarbeit<br />
von<br />
Zeno Heilmann<br />
Referent: Prof. Dr. Peter Hubral<br />
Korreferent: Prof. Dr. Friedemann Wenzel<br />
Karlsruhe, 31. 5. 2002
EHRENWÖRTLICHE ERKLÄRUNG:<br />
Hiermit versichere ich, dass ich die vorliegende Arbeit selbständig und nur mit den angegebenen<br />
Hilfsmitteln verfasst habe.<br />
Karlsruhe, den 31.5.2002.
Zusammenfassung<br />
Vorbemerkung<br />
Diese Diplomarbeit wurde bis auf die Zusammenfassung in Englisch verfasst. Da auch in der<br />
deutschen Sprache einige englische Fachausdrücke gebräuchlich sind, wurde bei diesen Ausdrücken<br />
auf eine Übersetzung verzichtet. Sie werden, mit Ausnahme ihrer groß geschriebenen<br />
Abkürzungen, kursiv dargestellt. Für die deutschsprachige Zusammenfassung wurde der gleiche<br />
Aufbau gewählt, wie er auch im Hauptteil der Arbeit verwendet wird.<br />
Einleitung<br />
Die <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> (CRS) Stapelung, hat sich in den vergangenen fünf Jahren<br />
als vielversprechende Alternative zu den bisher verwandten seismischen Abbildungsverfahren<br />
etabliert. Ein prinzipieller Vorteil gegenüber anderen Methoden, liegt in der rein<br />
datenorientierten Funktionsweise, die anders als z.B. die Pre-<strong>Stack</strong>-Depth-Migration keine<br />
Vorkenntnis über das zu ermittelnde Geschwindigkeitsmodell voraussetzt. Die Parametrisierung<br />
der CRS Summationsfläche basiert auf einem isotropen, lateral inhomogenen Modell mit<br />
gekrümmten Schichtgrenzen. Im Gegensatz dazu liegen beispielsweise der <strong>Common</strong>-Midpoint<br />
(CMP) Stapelung oder dem Normal-Moveout/Dip-Moveout (NMO/DMO) Prozess simple<br />
1D-Geschwindigkeitsmodellannahmen zugrunde. Folglich passt sich die CRS Stapelfläche<br />
der Reflexionsantwort im dreidimensionalen Datenvolumen weit besser an als herkömmliche<br />
Stapeloperatoren. Zudem tragen durch die flächenhafte Stapelung wesentlich mehr Signale zum<br />
Stapelergebnis bei, als beim CMP <strong>Stack</strong>, bei dem entlang von Linien summiert wird. Dies führt<br />
zu einer deutlichen Verbesserung des Signal-to-Noise Verhältnisses der erzeugten Stapelsektion.<br />
Statt einem einzigen Laufzeitparameter wie der NMO Geschwindigkeit, werden im Verlauf der<br />
CRS Stapelung drei bzw. fünf charakteristische Eigenschaften des Wellenfeldes, so genannte<br />
Wellenfeldattribute ermittelt. Diese können anschließend zur Lösung kinematischer und dynamischer<br />
Inversionsprobleme, wie der Amplitude-versus-Offset (AVO) oder Amplitude-versus-Angle<br />
(AVA) Analyse verwendet werden. Ursprünglich zur Erzeugung gestapelter 2D Zero-Offset<br />
(ZO) Sektionen konzipiert (Mann, 2002; Jäger, 1999; Müller, 1999; Höcht, 1998) wurde der<br />
CRS <strong>Stack</strong> in den letzten Jahren erfolgreich auf 3D erweitert (Höcht, 2002) und dahingehend<br />
ausgebaut, dass auch gestapelte Sektionen mit variablem Offset erzeugt werden können (Bergler,<br />
2001; Zhang et al., 2001a). Diese Art der Stapelung wird als Finite-Offset (FO) CRS <strong>Stack</strong><br />
bezeichnet. Der <strong>Common</strong>-Offset (CO) CRS <strong>Stack</strong> stellt einen wichtigen Spezialfall der FO CRS<br />
Stapelung dar und dient zur Erzeugung gestapelter CO Sektionen.<br />
Ziel dieser Diplomarbeit ist es den bestehenden 2D CRS <strong>Stack</strong> Formalismus, der, auf eine<br />
ebene Messoberfläche bezogene, seismische Daten voraussetzt, dahingehend zu erweitern, dass<br />
auch Daten die auf einer gekrümmten Messoberfläche gemessen wurden, ohne vorhergehende<br />
Bearbeitung, verwendet werden können. Zusätzlich soll der Einfluss eines oberflächennahen<br />
V
VI<br />
Gradienten der Wellenausbreitungsgeschwindigkeit, wie er in verwitterten Deckschichten<br />
häufig vorzufinden ist, berücksichtigt werden. Bisher war es notwendig diese Faktoren vor<br />
dem Stapeln mit Hilfe statischer Korrekturen zu beseitigt, was jedoch im Allgemeinen nur<br />
näherungsweise möglich ist und oft zu systematischen Fehlern führt. Diese haben meist<br />
nur geringen Einfluss auf das Stapelergebnis, können jedoch, wie in dieser Arbeit gezeigt<br />
wird, zu einer Verfälschung der beim Stapeln ermittelten Wellenfeldattribute führen. Obwohl<br />
die vorgestellte Herleitung der CRS Laufzeitformeln für gekrümmte Messoberflächen mit<br />
oberflächennahem Geschwindigkeitsgradienten die Finite-Offset CRS Stapelung einbezieht,<br />
konzentrieren sich die anschließenden Betrachtungen auf den spezielleren Fall der Zero-Offset<br />
CRS Stapelung. Einige der erzielten Ergebnisse lassen sich auf den FO CRS <strong>Stack</strong> übertragen,<br />
allerdings besteht für Finite-Offset momentan das Problem noch darin, die einzelnen Parameter<br />
der Laufzeitformel, unter Berücksichtigung der Messoberflächeneigenschaften, durch Wellenfeldattribute<br />
auszudrücken.<br />
Als theoretische Grundlage dieser Arbeit dienten, neben den oben genannten Veröffentlichungen,<br />
die Arbeiten Chira et al. (2001) und Chira and Hubral (2001), sowie die darin enthaltenen Referenzen.<br />
Die dort dargestellten Ergebnisse wurden mit Hinblick auf ihre praktische Anwendung<br />
weiterentwickelt und anhand eines synthetischen Datenbeispiels erprobt. Leider konnte die<br />
Erweiterung der bestehenden 2D ZO CRS <strong>Stack</strong> Software, auf gekrümmte Messoberflächen<br />
mit oberflächennahem Geschwindigkeitsgradienten, im Rahmen dieser Diplomarbeit nicht<br />
vollständig abgeschlossen werden. Jedoch wurden alle hierbei zu beachtenden Aspekte<br />
gründlich untersucht, sowie ein schneller und allgemein anwendbarer Code zur Bestimmung der<br />
Messoberflächeneigenschaften entwickelt und in die bestehende Software integriert.<br />
<strong>The</strong>orie<br />
Herleitung einer CRS Laufzeitformel für gekrümmte Messoberflächen mit oberflächennahem<br />
Geschwindigkeitsgradienten<br />
Nach einer kurzen Einführung in die Paraxiale Strahlentheorie, die sich aus der Zero-Order Ray-<br />
<strong>The</strong>ory ableitet, wird die so genannte <strong>Surface</strong>-to-<strong>Surface</strong> Propagator Matrix für 2D hergeleitet.<br />
Diese 2 × 2 Matrix, die nur von Größen, die sich auf einen so genannten Zentralstrahl beziehen,<br />
abhängt, ermöglicht es näherungsweise jeden beliebigen Strahl zu beschreiben, der in der Umgebung<br />
des Zentralstrahls verläuft. Hierfür wird angenommen, dass das selbe Raytracing-System<br />
welches den Zentralstrahl beschreibt auch näherungsweise für alle benachbarten (paraxialen)<br />
Strahlen gilt (paraxial assumption). Mit Hilfe der <strong>Surface</strong>-to-<strong>Surface</strong> Propagator Matrix wird<br />
dann, über die so genannte Hamilton Gleichung, eine Laufzeitformel abgeleitet, die die Laufzeit<br />
paraxialer Strahlen bis zur zweiten Ordnung exakt repräsentiert. Als Variablen, dienen die<br />
Lokation von Quelle und Empfänger entlang der Seismischen Linie, wobei später zu Offset und<br />
Midpoint Koordinaten übergegangen wird. Aus der ursprünglich parabolischen Darstellung der<br />
Laufzeitformel, lasst sich eine hyperbolische Repräsentation ableiten, welche in den meisten
Fällen eine bessere Näherung der wahren Laufzeit darstellt (Höcht, 1998; Müller, 1999; Jäger,<br />
1999; Bergler, 2001).<br />
Spezielle seismische Konfigurationen<br />
Die fünf Parameter, die im Falle eines Zentralstrahls mit finitem Offset die parabolische und die<br />
hyperbolische Laufzeitgleichung bestimmen, lassen sich durch dessen Ab- bzw. Auftauchwinkel<br />
an Quelle und Empfänger und die drei unabhängigen Elemente der, sich auf diesen Zentralstrahl<br />
beziehenden, <strong>Surface</strong>-to-<strong>Surface</strong> Propagator Matrix ausdrücken. Um diese Größen für<br />
eine spätere Stapelung zu bestimmen, eignen sich besonders solche Sektionen (oder Gather) des<br />
drei dimensionalen Datenraums, in denen die Laufzeit nicht von allen fünf Parametern gleichzeitig<br />
abhängt. Diese sind, im Falle einer gekrümmten Messoberfläche, das <strong>Common</strong>-Shot (CS)<br />
Gather, das <strong>Common</strong>-Receiver (CR) Gather sowie das Odd-Dislocation (OD) Gather und das<br />
Even-Dislocation (ED) Gather. Die letzteren beiden Sektionen sind Verallgemeinerungen, der<br />
für eine ebene Messoberfläche definierten, <strong>Common</strong>-Midpoint (CMP) und <strong>Common</strong>-Offset (CO)<br />
Messkonfigurationen.<br />
Einführung der Wellenfeldattribute<br />
Die drei unabhängigen Elemente der <strong>Surface</strong>-to-<strong>Surface</strong> Propagator Matrix lassen sich im Fall<br />
einer ebenen Messoberfläche mit drei Wellenfrontkrümmungen verknüpfen, und werden dadurch<br />
geometrisch interpretierbar (Bergler et al., 2001; Zhang et al., 2001a). Wie diese Verknüpfung für<br />
eine gekrümmte Messoberfläche und unter Einbeziehung des Gradienten der oberflächennahen<br />
Geschwindigkeit aussieht, lässt sich anhand der vielfach bestätigten Ergebnisse von Červen´y<br />
(2001) bestimmen, indem man den dort hergestellten Zusammenhang zwischen der <strong>Surface</strong>to-<strong>Surface</strong><br />
Propagator Matrix und der strahlzentrierten Propagations Matrize Π nutzt. Die im<br />
Anschluss benötigte Relation zwischen den Elementen der Π-Matrix und geeigneten Wellenfeldattributen<br />
ist allerdings nur für den Fall eines Zentralstrahls mit koincidentem Quell- und<br />
Empfängerpunkt bekannt (Hubral, 1983). Folglich lässt sich auch nur für diesen Fall eine Laufzeitformel<br />
aufstellen, mit der sich alle Wellenfrontattribute, unter Berücksichtigung der Messoberflächenkrümmung<br />
und des oberflächennahen Geschwindigkeitsgradienten, bestimmen lassen.<br />
Auf diese Zero-Offset CRS <strong>Stack</strong> Laufzeitformel (in parabolischer und hyperbolischer Darstellung)<br />
konzentrieren sich die weiteren Betrachtungen. Durch die vereinfachte Geometrie des ZO<br />
Zentralstrahls verringert sich die Zahl der Parameter von fünf auf drei. Die mit diesen drei Parametern<br />
verknüpften Wellenfeldattribute sind der Abtauchwinkel β 0 des Zentralstrahls am koinzidenten<br />
Quell- und Empfängerpunkt, und die dort theoretisch messbaren Krümmungen K N und<br />
K NIP der so genannten NIP- und N-Welle (Hubral, 1983).<br />
Verhältnis zwischen Messoberflächeneigenschaften und Wellenfeldattributen<br />
Natürlich stellt auch die erweiterte CRS Laufzeitformel, für gekrümmte Messoberflächen mit<br />
oberflächennahem Geschwindigkeitsgradienten, nur eine Näherung zweiter Ordnung der wahren<br />
Laufzeit dar. Aus diesem Grunde ist es nahe liegend zu untersuchen in welcher Beziehung<br />
VII
VIII<br />
die wahren Wellenfeldattribute zu, unter Vernachlässigung der Acquisitionstopographie und des<br />
oberflächennahen Geschwindigkeitsgradienten, mit konventioneller ZO CRS <strong>Stack</strong> Software bestimmten<br />
scheinbaren Wellenfeldattributen stehen. Es zeigt sich, dass durchaus die Möglichkeit<br />
besteht solche “pseudo” Attribute im Nachhinein zu korrigieren - allerdings nur wenn sie richtig<br />
bestimmt wurden. Hierin liegt das Problem dieses, ansonsten sehr pragmatischen, Ansatzes.<br />
Bestimmung geeigneter Suchparametergrenzen<br />
Durch den Einfluss des Geschwindigkeitsgradienten und vor allem der Topographie gelten für die<br />
mit konventioneller ZO CRS <strong>Stack</strong> Software bestimmten pseudo Wellenfeldattribute vollkommen<br />
andere Wertebereiche, als für die wahren Wellenfeldattribute gelten würden. Nur wenn dies bei<br />
der Wahl geeigneter Suchgrenzen berücksichtigt wird können diese Parameter korrekt bestimmt<br />
werden. Hinzukommt, dass im Falle des konventionellen CRS <strong>Stack</strong>s die gleichen Suchgrenzen<br />
für die gesamte Messoberfläche gelten, was bei einer ebenen Oberfläche natürlich sinnvoll ist.<br />
Hier führt dies jedoch dazu, dass die Suchbereiche der einzelnen Parameter sehr groß gewählt<br />
werden müssen, um den verschiedenen Oberflächeneigenschaften aller Stapelmittelpunkte entlang<br />
der Messoberfläche Rechnung zu tragen. Die Folge ist eine starke Erhöhung des für die<br />
Suche benötigten Rechenaufwands und der damit verbundenen Kosten.<br />
Mit der hier abgeleiteten Laufzeitformel werden im Gegensatz dazu direkt die wahren Wellenfeldattribute<br />
bestimmt, die, mit Ausnahme von β 0 unabhängig von den Eigenschaften der<br />
Messoberflächen sind. Somit gelten für K N und K NIP die selben Grenzen wie im Falle des konventionellen<br />
2D CRS <strong>Stack</strong>s (siehe Mann, 2002). Beim Abtauchwinkel β 0 muss beachtet werden,<br />
dass dieser sich auf die Tangente zur Oberfläche im Abtauchpunkt bezieht, und daher vom<br />
Dip der Messoberfläche abhängt. Dies spielt bei einer Ebenen Oberfläche keine Rolle, muss<br />
hier aber berücksichtigt werden. Leider hängt die Normal-Moveout Geschwindigkeit, die bei<br />
der heute größtenteils verwendeten pragmatischen Suchstrategie (Mann, 2002) als erster Parameter<br />
bestimmt wird, neben β 0 und K NIP , auch von der Oberflächenkrümmung und dem oberflächennahen<br />
Geschwindigkeitsgradienten ab, was die Wahl geeigneter Suchgrenzen erschwert.<br />
Zur Lösung dieses Problems wird ein Verfahren hergeleitet, mit dem es möglich ist aus den v NMO<br />
Grenzen, die für eine fiktive ebene Messoberfläche ohne oberflächennahen Geschwindigkeitsgradienten<br />
gelten würden, neue, den tatsächlichen Akquisitionsbedingungen angepasste Grenzen,<br />
zu berechnen. Diese variieren dann je nach Auftauchpunkt des betrachteten Zentralstrahls. Auf<br />
ähnliche Weise lassen sich die, bei der Verwendung konventioneller Software, zusätzlich notwendigen<br />
Grenzen für die dort bestimmten “pseudo” Attribute, aus den Grenzen, die für die wahren<br />
Wellenfeldattribute gelten würden, berechnen.<br />
Redatuming<br />
Stapelt man nun einen Datensatz, der auf einer gekrümmten Messoberfläche mit oberflächennahem<br />
Geschwindigkeitsgradienten, bestimmt wurde, so beziehen sich die erhaltenen<br />
Stapel- und Attributsektionen natürlich auf diese Oberfläche. Dies hat zur Folge, dass z.B. ebene
Reflektoren in der so simulierten ZO-Sektion, der Topographie der Messoberfläche entsprechend,<br />
gekrümmt abgebildet werden. Dasselbe gilt auch für die Attributsektionen. Um eine spätere Interpretation<br />
und Weiterverarbeitung dieser Sektionen zu erleichtern ist es nahe liegend, den Einfluss<br />
der Oberflächentopographie zu beseitigen, indem man die ermittelten Wellenfeldattribute<br />
und ZO Laufzeiten auf eine fiktive horizontale Messoberfläche bezieht. Dieser Vorgang, wird<br />
dadurch stark erleichtert, dass beim CRS <strong>Stack</strong> die Abtauchwinkel der Zentralstrahlen ermittelt<br />
werden. Es bietet sich an, die fiktive Messoberfläche oberhalb der wahren Akquisitionsfläche<br />
zu platzieren, da dann die ebenfalls benötigte Geschwindigkeit, innerhalb der so entstandenen<br />
fiktiven Schicht zwischen wahrer und fiktiver Messoberfläche, frei gewählt werden kann. Für<br />
den Fall einer gekrümmten Messoberfläche ohne oberflächennahen Geschwindigkeitsgradienten<br />
kann die Geschwindigkeit im fiktiven Layer gleich der oberflächennahen Geschwindigkeit<br />
v 0 gewählt werden, die beim CRS <strong>Stack</strong> als bekannt vorausgesetzt wird. Somit muss keine<br />
Refraktion der Zentralstrahlen und N- und NIP-Wellenfronten an der wahren Messoberfläche<br />
berücksichtigt werden. Liegt ein oberflächennaher Geschwindigkeitsgradient vor, so müssen neben<br />
dem Transmissions Gesetz, noch das Refraktionsgesetz (Hubral and Krey, 1980) und das<br />
Snellius’sche Gesetz berücksichtigt werden, da dann die fiktive Geschwindigkeit natürlich nicht<br />
gleich der (variablen) oberflächennahen Geschwindigkeit gewählt werden kann.<br />
Einführung eines globalen Koordinatensystems<br />
Für die schon erwähnte Herleitung der CRS Laufzeitformeln, wird ein lokales kartesisches Koordinatensystem<br />
verwendet, dessen x-Achse tangential zur Messoberfläche im Auftauchpunkt des<br />
Zentralstrahls verläuft und dessen Ursprung im Auftauchpunkt liegt. Im allgemeinen gilt für jeden<br />
Auftauchpunkt ein anderes lokales Koordinatensystem, in welches die jeweiligen 2D Quellund<br />
Empfängerkoordinaten transformiert werden müssen um die in den Laufzeitformeln benutzten<br />
1D Offset und Midpoint Koordinaten zu berechnen. Dies kann vermieden werden, indem<br />
man in den Laufzeitformeln die lokalen Offset und Midpoint Koordinaten durch globale ersetzt,<br />
die in einem für alle Auftauchpunkte geltenden Koordinatensystem gemessen werden. Hierdurch<br />
erscheint der Dipwinkel der Messoberfläche und die x-Koordinate des Auftauchpunktes explizit<br />
in den Laufzeitformeln.<br />
Mit Hilfe der auf diese Weise modifizierten Laufzeitformeln lassen sich die Relationen zwischen<br />
den wahren Wellenfeldattributen und den mit konventioneller ZO CRS <strong>Stack</strong> Software,<br />
ohne Berücksichtigung der lokalen Koordinaten (d.h. des Dipwinkels) bestimmten, “pseudo”<br />
Attributen ableiten. Dadurch kann diese Software, mit den genannten Einschränkungen, fast<br />
ohne Abänderung des Codes auch für Daten, die auf einer gekrümmten Oberfläche mit oberflächennahem<br />
Geschwindigkeitsgradienten gemessen wurden, verwendet werden. Die einzig<br />
notwendige Änderung ist die, dass für den pseudo Abtauchwinkel, bestimmt unter Verwendung<br />
globaler Koordinaten, auch komplexe Werte berücksichtigt werden müssen, was bei der konventionellen<br />
CRS <strong>Stack</strong> Software natürlich nicht vorgesehen ist.<br />
IX
X<br />
Diskussion der Ergebnisse anhand eines synthetischen Datenbeispiels<br />
Datensatz<br />
Der hier verwendete synthetische Datensatz wurde von ENI/Agip erzeugt und an Pedro Chira<br />
und mich weitergegeben, um daran die Anwendung der in (Chira and Hubral, 2001) vorgestellten<br />
CRS Laufzeitformel für gekrümmte Messoberflächen zu untersuchen. Das zur Erzeugung dieses<br />
Datensatzes verwendete Modell besteht aus vier homogenen, durch ebene und horizontale Reflektoren<br />
voneinander getrennte, Schichten. Die Messoberfläche besitzt eine starke ausgeprägte<br />
Topographie. Ein oberflächennaher Geschwindigkeitsgradient wurde bei der Generierung des<br />
Datensatzes nicht berücksichtigt, da dieser erst später in die Laufzeitformeln integriert worden<br />
ist (Chira et al., 2001).<br />
Standard Processing mit Hilfe statischer Korrekturen<br />
Zusätzlich zum Datensatz bekamen wir Stapel- und Attributsektionen, welche ENI/Agip erzeugt<br />
hatte, indem zuerst mit Hilfe statischer Korrekturen näherungsweise eine ebene Messoberfläche<br />
simuliert, und anschließend der herkömmliche CRS <strong>Stack</strong> durchgeführt worden war. Diese Ergebnisse<br />
erwiesen sich allerdings nicht als optimal, da die Wirkung der Korrekturen auf die<br />
zu bestimmenden Laufzeitparameter bei der Wahl geeigneter Suchgrenzen nicht berücksichtigt<br />
worden war. Dies wird am Beispiel der NMO Geschwindigkeit untersucht und durch erneutes<br />
Stapeln mit besser geeigneten NMO Geschwindigkeitsgrenzen demonstriert.<br />
Bestimmung der Oberflächenattribute<br />
Die in den hergeleiteten Laufzeitgleichungen verwendeten Eigenschaften der Messoberfläche<br />
im Auftauchpunkt des Zentralstrahls sind die Krümmung, der Dip (explizit oder implizit), die<br />
oberflächennahe Geschwindigkeit und deren Gradient. Die letzteren beiden Größen müssen im<br />
Feld bestimmt werden, beim Dip und der Krümmung ist dies im Allgemeinen nicht sinnvoll.<br />
Diese werden im Nachhinein aus den gemessenen Quell- und Empfängerkoordinaten berechnet.<br />
Hierbei muss beachtet werden, wofür diese Werte später verwendet werden sollen, nämlich zur<br />
Beschreibung der Messoberfläche im Bereich der betrachteten Stapelapertur durch eine Parabel.<br />
Schwanken Oberflächenkrümmung und Dip innerhalb der Stapelapertur, wie es in der Natur<br />
meistens der Fall ist, muss ein gemittelter Wert verwendet werden. Gemittelt bedeutet hier, dass<br />
Krümmung und Steigung der Parabel bestimmt werden müssen, die alle zur Stapelung beitragenden<br />
Quell- und Empfängerpunkte am besten repräsentiert. Die z-Koordinate des Auftauchpunkte,<br />
die genauso wie Krümmung und Dip nicht lokal bestimmt werden sollte, und zu diesen<br />
Werten konsistent sein muss, wird ebenfalls auf diese Weise ermittelt. Sind die Abweichungen<br />
der Quell- und Empfängerpunkte zu dieser Parabel zu groß, um noch als vernachlässigbar zu gelten,<br />
können statische Korrekturen, wie sie in Sektion 3.2 beschrieben werden, angewandt werden<br />
um Quell- und Empfängerpunkte zu simulieren, die auf der Parabel liegen. Zur Erweiterung der<br />
bestehenden 2D ZO CRS <strong>Stack</strong> Implementation, wurde nach ausführlichen Tests, die mit Hilfe<br />
des Computer Algebra Systems MAPLE durchgeführt worden waren, ein C++ Code geschrie-
en, der die benötigten Oberflächeneigenschaften, für beliebige Messoberflächen bestimmt. Mit<br />
diesem Programm wurden dann aus den Quell- und Empfängerkoordinaten des oben genannten<br />
Datensatzes die gesuchten Messoberflächeneigenschaften berechnet.<br />
Vorwärts modellierte Ergebnisse<br />
Aufgrund der Einfachheit des betrachteten Untergrundmodells, können die Wellenfeldattribute<br />
K N , K NIP und β 0 für beliebige Zentralstrahlen berechnet werden, ohne dass eine spezielle<br />
Raytracing-Software verwendet werden muss. Zusammen mit den zuvor ermittelten Messoberflächenattributen,<br />
können dann sowohl die Normal Moveout Geschwindigkeit, als auch die<br />
“pseudo” Wellenfeldattribute berechnet werden, die hier aus der Anwendung der herkömmliche<br />
2D ZO CRS <strong>Stack</strong> Implementation resultieren würden. Zudem lassen sich mit Hilfe der Oberflächenattribute<br />
Suchparametergrenzen, die für eine ebene Messoberfläche sinnvoll gewesen<br />
wären, in Grenzen für die aktuelle Messoberfläche umrechnen. Leider war es zeitlich nicht<br />
möglich die herkömmlichen 2D ZO CRS <strong>Stack</strong> Implementation vollständig auf den Datensatz<br />
anzuwenden und die so erhaltenen Ergebnisse für die pseudo Attribute, mit den vorwärts berechneten<br />
zu vergleichen.<br />
Ausblick<br />
Da es der zeitliche Rahmen dieser Diplomarbeit nicht erlaubt hat, neben den notwendigen theoretischen<br />
Betrachtungen, auch die praktische Implementation der gewonnenen Ergebnisse abzuschließen,<br />
kann ich am Ende dieser Arbeit diesbezüglich nur einige Vorergebnisse präsentieren.<br />
Wie schon erwähnt, werden bei der Durchführung der ZO CRS Stapelung nicht alle drei Laufzeitparameter<br />
gleichzeitig bestimmt. Als ersten Schritt, wird mit Hilfe einer datengesteuerten<br />
CMP Stapelung eine vorläufige ZO Sektion und die zugehörige NMO Geschwindigkeitssektion<br />
erzeugt. Die Ergebnisse dieses so genannten automatic CMP <strong>Stack</strong>s, angewandt auf den hier<br />
verwendeten Datensatz, werden in diesem Abschnitt diskutiert. Aufgrund der Einfachheit des,<br />
dem Datensatz zugrunde liegenden Untergrundmodells, dürfte sich jedoch die so erzeugte CMP<br />
Stapelsektion von der zu erwartende CRS Stapelsektion nicht grundlegend unterscheiden. Aus<br />
der so gewonnene NMO Geschwindigkeitssektion wurden die v NMO -Werte der einzelnen Reflektoren<br />
extrahiert. Diese stimmen im Rahmen der zu erwartenden Abweichungen mit den in der<br />
letzten Sektion vorwärts berechneten Ergebnissen überein.<br />
Bei beiden Sektionen wurde zudem ein Redatuming durchgeführt, um eine ebene Messoberfläche<br />
zu simulieren. Dies ermöglicht den direkten Vergleich zu den in Sektion 3.2 gezeigten Ergebnissen,<br />
die mit Hilfe, vor dem Stapeln durchgeführter, statischer Korrekturen erzeugt wurden.<br />
XI
XII
Contents<br />
1 Preface 1<br />
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1<br />
1.2 Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />
2 <strong>The</strong>ory 5<br />
2.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />
2.1.1 Zero-order ray theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5<br />
2.1.2 Paraxial ray theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6<br />
2.1.3 <strong>Surface</strong>-to-surface propagator matrix . . . . . . . . . . . . . . . . . . . 7<br />
2.2 Fundamental traveltime expressions . . . . . . . . . . . . . . . . . . . . . . . . 10<br />
2.2.1 Traveltime of a paraxial ray . . . . . . . . . . . . . . . . . . . . . . . . 12<br />
2.2.2 Determination of the coefficients . . . . . . . . . . . . . . . . . . . . . . 15<br />
2.2.3 Acquisition topography and near-surface velocity gradient . . . . . . . . 17<br />
2.3 Zero-offset situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20<br />
2.3.1 Relationship between (near)surface and wavefield attributes . . . . . . . 26<br />
2.3.2 Normal-moveout (NMO) and root-mean-square (RMS) velocities . . . . 28<br />
2.4 Search-range estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33<br />
2.4.1 Search range of K N and K ∗ N<br />
. . . . . . . . . . . . . . . . . . . . . . . . . 33
XIV CONTENTS<br />
2.4.2 Search range of β 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34<br />
2.4.3 Search range of the NMO velocity . . . . . . . . . . . . . . . . . . . . . 34<br />
2.5 Redatuming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38<br />
2.5.1 Mapping of X 0 and t 0 to the common-datum surface . . . . . . . . . . . 40<br />
2.5.2 Mapping of K N and K NIP to the common-datum surface . . . . . . . . . 41<br />
2.5.3 Mapping of v NMO to the common-datum surface . . . . . . . . . . . . . 42<br />
2.6 Global coordinate system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43<br />
2.6.1 <strong>The</strong> inhomogeneity factor E 0 in global coordinates . . . . . . . . . . . . 46<br />
2.6.2 <strong>The</strong> NMO and RMS velocities in global coordinates . . . . . . . . . . . 46<br />
2.6.3 <strong>The</strong> search range of K ∗ N , β ∗ 0 , and v NMO<br />
in global coordinates . . . . . . . 47<br />
2.6.4 Redatuming in global coordinates . . . . . . . . . . . . . . . . . . . . . 48<br />
3 Synthetic Data Example 51<br />
3.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51<br />
3.2 Standard processing using elevation-statics . . . . . . . . . . . . . . . . . . . . . 53<br />
3.2.1 Results of the standard processing . . . . . . . . . . . . . . . . . . . . . 56<br />
3.3 Determination of the (near)surface attributes and elevation in X 0 . . . . . . . . . 63<br />
3.3.1 Determination of α 0 , K 0 , and xz by fitting parabolas . . . . . . . . . . . . 64<br />
3.3.2 Determination of α 0 , K 0 , and xz by fitting circles . . . . . . . . . . . . . 65<br />
3.4 Forward calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68<br />
3.4.1 Forward calculated take-off angle β 0 and its search limits . . . . . . . . . 70<br />
3.4.2 Forward calculated RMS and NMO velocities . . . . . . . . . . . . . . . 72<br />
3.4.3 Forward calculated slowness p NMO and its search limits. . . . . . . . . . 74<br />
3.4.4 Forward calculated values of K N and K NIP . . . . . . . . . . . . . . . . . 76<br />
3.4.5 Pseudo attributes β ∗ 0 , K∗ N , and K∗ NIP<br />
and their search limits. . . . . . . . . 77
CONTENTS XV<br />
3.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80<br />
3.5.1 Automatic CMP stack plus redatuming . . . . . . . . . . . . . . . . . . 80<br />
3.5.2 Comparison with the predicted NMO velocities . . . . . . . . . . . . . . 88<br />
4 Summary 93<br />
A <strong>The</strong> scalar Hamilton’s equation 95<br />
B Used hard- and software 97<br />
C Acknowledgment/Danksagung 99<br />
List of figures 100<br />
Bibliography 105
XVI CONTENTS
Chapter 1<br />
Preface<br />
1.1 Introduction<br />
Seismic methods have a wide range of application. Among important applications in civil engineering<br />
and groundwater search, the oil exploration is of particular importance. Today, it is<br />
standard for oil companies to rely on seismic interpretation for selecting the sites for exploratory<br />
oil wells because thereby the likelihood of a successful venture can be highly improved. <strong>The</strong><br />
basic technique of reflection seismics consists of generating seismic waves and measuring the<br />
time required for the waves to travel from the source located on the surface downwards to the<br />
reflector in the subsurface and back to the surface where a series of receivers are positioned. <strong>The</strong><br />
receivers are usually disposed along a straight line directed toward the source. <strong>The</strong> traveltime<br />
of reflected waves depends on the elastic properties of the subsurface as well as on the position,<br />
orientation, and curvature of the reflector. Thus, it is possible to deduce information about the<br />
subsurface from the observed arrival times. In general, before an acquired seismic dataset can<br />
be interpreted it is subject to many processing steps. According to Yilmaz (1987), there is a<br />
well-established sequence for standard seismic processing. <strong>The</strong> three principal processes — deconvolution,<br />
stacking, and migration — make up the foundation of routine processing. In this<br />
work, I address the second step of the processing chain, namely the stacking procedure. <strong>The</strong><br />
so-called stacked section gives the interpreter a first image of the investigated area and serves<br />
as input to the subsequent post-stack migration. By moving source and receiver arrays along<br />
a straight line, called the seismic line, a three-dimensional multi-coverage reflection dataset is<br />
acquired. This dataset depends on shot and receiver location along the seismic line as well as<br />
on the recording time. <strong>The</strong> subsequent processing of the three-dimensional dataset is applied to<br />
get a two-dimensional image of the subsurface. For the processing, the data are conventionally<br />
sorted with respect to the midpoint position xm of shot and receiver and half-offset h, i.e., the
2 Preface<br />
half distance between shot and receiver. Thus, the multi-coverage reflection dataset is given in<br />
the xm − h −t space, where t corresponds to the recording time.<br />
Unfortunately, the dataset contains not only signals, i.e., any events on the seismic record from<br />
which we wish to obtain information about the subsurface, but also noise. Noise is often divided<br />
into coherent and incoherent noise. As most processing schemes are using primary reflections<br />
only, multiple reflections belong to the class of coherent noise. Incoherent noise or random noise<br />
is not predictable, i.e., one cannot say what a trace will be from the knowledge of nearby traces.<br />
Random noise in real data may be, for instance, due to traffic, industry, or wind shaking trees.<br />
<strong>The</strong> goal of stacking is to enhance the signal and to suppress the noise by summing up correlated<br />
events in the multi-coverage data. Zero-offset (ZO) stacking operators approximate the actual<br />
reflections in the xm − h − t space in the vicinity of a ZO point. This point is associated with a<br />
hypothetical experiment where source and receiver are coincident. <strong>The</strong> summation result along<br />
the ZO stacking operator is assigned to the respective ZO point. Doing the same for all points of<br />
the ZO gather yields the ZO stack section. Well-known conventional ZO stacking methods are<br />
the common-midpoint (CMP) stack and the normal-moveout/dip-moveout (NMO/DMO) process.<br />
Within the last five years, the <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> (CRS) stack has established as a<br />
promising alternative to the seismic reflection imaging methods, used so far. Originally designed<br />
to generate a 2D ZO stack-section (Höcht, 1998; Müller, 1999; Mann, 2002), the CRS stack was<br />
successfully extended to 3D (Höcht, 2002) and to finite-offset (FO) (Zhang et al., 2001a; Bergler,<br />
2001). <strong>The</strong> FO CRS stack operator approximates the actual reflections in the vicinity of an arbitrary<br />
point in the xm − h − t space. A special case of the FO CRS stack is the common-offset<br />
(CO) CRS stack, which is performed analogously to ZO stacking but for a point in a CO gather.<br />
<strong>The</strong>refore, the CO stacking operator approximates the reflection event in the vicinity of a point<br />
with a fixed offset. Summing up correlated events along the summation operator and assigning<br />
the result to the respective CO point for all points in a chosen CO section yields the CO stack<br />
section. Of course, also any other stacked section (consisting of arbitrary points in the xm − h −t<br />
space) can be generated this way, using the FO CRS stack operator.<br />
Conventional stacking methods, like, e.g., NMO/DMO stack, are based on simple 1D velocitymodel<br />
assumptions and use one-parametric traveltime-moveout formulas that are applied to<br />
common-midpoint data only. <strong>The</strong> CRS stack makes full use of the multi-coverage seismicreflection<br />
data and provides additional traveltime parameters. <strong>The</strong>se parameters are very useful<br />
for the extraction of further attributes of the seismic medium or for an inversion of a meaningful<br />
subsurface velocity model. Another important feature of the CRS stack method is that an a priori<br />
macro-velocity model is not required. For this reason this method is referred to the macro-model<br />
independent methods, which also include the Polystack method (de Bazelaire, 1988; de Bazelaire<br />
and Viallix, 1994) and the Multifocusing method (Gelchinsky et al., 1997). Various aspects of<br />
macro-model-independent reflection imaging methods are discussed in Hubral (1999). As practical<br />
experience has shown, these new methods are particularly successful for seismic land-data.
1.2 Structure of the thesis 3<br />
However, land data suffer in many cases from complex near-surface conditions like laterally<br />
changing near-surface velocities and undulating topography. For this reason, the existing CRS<br />
method was generalized to handle such situations (Chira et al., 2001). Until then all discussions<br />
and derivations in this regard had involved a planar measurement surface. However, the<br />
assumption of a planar measurement surface is, by no means, a requirement for the validity of<br />
the surface-to-surface propagator matrix formalism (Bortfeld, 1989), which is the basis of the<br />
derivation of the CRS traveltime-moveout formulas. It is to be mentioned that an extension of the<br />
Multifocusing method, designed to include the topographic features of the measurement surface,<br />
has been also recently proposed in Gurevich et al. (2001).<br />
1.2 Structure of the thesis<br />
<strong>The</strong> thesis is divided into two parts:<br />
<strong>The</strong>ory (Chapter 2):<br />
After giving a short review of the wave-theoretical foundations, on which the theory presented<br />
within this thesis is based, a generalized finite-offset CRS traveltime formula that considers the<br />
topography of the measurement surface as well as the near-surface velocity gradient, is derived.<br />
This formula is reduced to the zero-offset case to which the further discussions are devoted.<br />
<strong>The</strong> main purpose of these discussions will be to provide the theoretical background that is<br />
necessary to extent the current implementation of the 2D ZO CRS stack, designed for a planar<br />
measurement surface, to the more general case of considering a curved measurement surface and<br />
its near-surface velocity gradient.<br />
Synthetic Data Example (Chapter 3):<br />
In the second part of this thesis, the application of the 2D ZO CRS stack to data measured on<br />
a curved measurement surface is discussed by means of a synthetic dataset. <strong>The</strong> near-surface<br />
velocity gradient, was not yet included into the theory, when the dataset was made. Thus this<br />
point is not considered. However, this is no severe restriction, as it is reasonable to study the<br />
effect of the topography separately, at first. For comparison, standard processing using static<br />
corrections is applied to the dataset and briefly analyzed. After a brief general discussion on, how<br />
the considered characteristic properties of the measurement surface and its <strong>under</strong>lying top-layer<br />
are determined, these considerations are applied to the synthetic dataset. <strong>The</strong> results, obtained<br />
this way, serve together with forward modeled attributes of the seismic medium to point out<br />
important aspects, related to the application of the CRS stack to data measured on a curved<br />
surface. Unfortunately it was not possible to finish the extension of existing 2D CRS stack<br />
software within the narrow time-frame of a diploma thesis. However, some provisional results<br />
are presented in the outlook.
4 Preface
Chapter 2<br />
<strong>The</strong>ory<br />
2.1 Basics<br />
<strong>The</strong> purpose of this section is to give a short review of the ray-theoretical foundations, <strong>under</strong>lying<br />
this thesis. This is done with the intention, to point out the main results as well as the made<br />
assumptions, as far as it is necessary for the development and <strong>under</strong>standing of the formulas,<br />
used in the latter. For a detailed treatment I refer, e.g., to Červen´y (2001).<br />
2.1.1 Zero-order ray theory<br />
“<strong>The</strong> propagation of seismic body waves in complex, laterally varying 3D layered structures is<br />
a considerably complicated process” (Vlastislav Červen´y).<br />
Getting an exact description would require to solve the elastodynamic equations for this very<br />
general case with its multitude of degrees of freedom. However, analytical solutions of these<br />
equations are not known and - even if there would be a solution - not applicable to practical<br />
problems. Thus the most common approaches to investigate the seismic wave-field in complex<br />
media are based either on the direct numerical solution of the elastodynamic equations or on<br />
approximate asymptotic solutions of these equation valid only for high frequencies. One of<br />
the latter is the so-called ray method which is today highly developed and widely used. In<br />
the ray method the asymptotic high-frequency solution of the elastodynamic equations for each<br />
elementary body wave can be sought in the form of a so called ray series (Babich, 1956; Karal and<br />
Keller, 1959). In the frequency domain this is a series in inverse powers of the circular frequency<br />
ω. That is the reason why the ray method is often called the ray series method, or the asymptotic<br />
ray theory . In most practical applications in seismology and seismics, only the leading term of
6 <strong>The</strong>ory<br />
the ray series which is of the order ω 0 is considered. This leads to the zero order ray theory which<br />
is the <strong>under</strong>lying concept of all the formulas derived and used within the scope of this thesis.<br />
<strong>The</strong> main results of this method are the eikonal- and the transport equation. Solving the eikonal<br />
equation results in the so-called ray tracing system which determines all kinematic aspects of<br />
ray propagation. <strong>The</strong> solution of the transport equation describes all dynamic properties of the<br />
wave-field. <strong>The</strong> latter are not considered within this theses.<br />
<strong>The</strong> validity conditions of the zero-order ray theory are an extensively discussed topic, e.g., by<br />
Ben-Menachem and Beydoun (1985), Kravtsov and Orlov (1990) or Červen´y (2001). Many<br />
investigations on this subject were made in the past but nevertheless there are only heuristic<br />
criteria to determine if zero-order ray theory is applicable for a particular earth model or not. One<br />
of the most commonly used conditions is that the Fourier spectrum ˇf[ω] of the source wavelet<br />
f[t] is required to effectively vanish for frequencies<br />
ω < ω 0 = v(ˆr)/l 0 ,<br />
where l 0 is the length scale of the inhomogeneities in the medium and v(ˆr) is the wave velocity of<br />
the medium. In the following a 2D earth model is assumed that consists of isotropic, laterally and<br />
vertically inhomogeneous layers separated by continuous and smooth first-order discontinuities<br />
of almost arbitrary shape which fulfills the above criteria and should be close enough to real<br />
conditions to provide practical useful results.<br />
2.1.2 Paraxial ray theory<br />
<strong>The</strong> term “paraxial” has his seeds in optics were it represents the vicinity of the axis of the optical<br />
system. In our case it denotes the vicinity of the so-called central ray. Paraxial ray theory is an<br />
extension of the standard ray method with the purpose to describe approximatively the behavior<br />
of paraxial rays in the near vicinity of a central ray, which is assumed to be known. This is<br />
done by using the paraxial assumption saying that the ray tracing system of a particular ray is<br />
approximatively valid also in the close vicinity of this ray. <strong>The</strong> resulting paraxial ray tracing<br />
system is also called dynamic ray tracing system because it provides even dynamic information,<br />
which is very useful, e.g., in true amplitude imaging (Schleicher et al., 2002). Solving the latter<br />
ray-tracing system for arbitrary initial conditions leads to the ray-centered propagator matrix<br />
Π and to the surface-to-surface propagator matrix T (Bortfeld, 1989; Červen´y, 2001). This<br />
matrices describe the traveltime moveout of any arbitrary ray in the vicinity of a central ray<br />
in terms of quantities that refer to the central ray only. <strong>The</strong> traveltime along the paraxial ray,<br />
obtained in this way, is correct up to the second-order in q,which is the location of the paraxial<br />
ray, expressed in the ray-centered coordinate system of the central ray (see, e.g., Červen´y, 2001).
2.1 Basics 7<br />
2.1.3 <strong>Surface</strong>-to-surface propagator matrix<br />
In seismics we generally have measurement configurations were rays emanate from sources that<br />
are located on a surface and impinge at receivers that are located on another surface. According<br />
to Bortfeld these surfaces are called the anterior surface and the posterior surface. In the 2D<br />
case, on which the further discussions are focused, these surfaces reduce to lines.<br />
According to Figure 2.1, we introduce two local Cartesian coordinate systems. <strong>The</strong> first with<br />
x-axis tangent to the anterior surface at the source location S and the second analog with x-axis<br />
tangent to the posterior surface at the receiver location G. All quantities measured in the first<br />
coordinate system are denoted by the subscript S and all quantities measured in the second coordinate<br />
system by the subscript G. <strong>The</strong> central ray SG is at S (anterior surface) and G (posterior<br />
surface) completely described by its respective location and ray slowness vectors. An analog<br />
description holds for the paraxial ray SG at S and G. <strong>The</strong> location and slowness vectors of the<br />
central ray at S and G are denoted by x S and p S , and x G and p G , respectively. Analogously, the<br />
paraxial ray SG at S is described by x S and p S , and at G by x G and p G .<br />
Knowing the two surfaces and their near-surface velocities v S and v G it is possible to reduce<br />
the 2D location and slowness vectors to scalar values. <strong>The</strong> 2D vectors can later be uniquely<br />
reconstructed from their scalar description. In case of source and receiver locations x S ,x S ,x G ,x G<br />
and the slowness vectors p S ,p G of the central ray this is done by a simple projection in z-direction<br />
onto the x-axis. In case of the slowness vectors of the paraxial ray p S and p G we have to perform<br />
a projection cascade consisting of two steps (see Figure 2.2). First the slowness vector, e.g., p S<br />
is projected onto the line tangent to the measurement curve at S. Later the resulting projection<br />
vector has to be projected onto the coordinate axis tangent in S. An analog description holds for<br />
p G , the slowness vector of the paraxial ray at G.<br />
If we assume the central ray SG to be known, we can approximately calculate any paraxial ray<br />
SG using the paraxial ray theory. <strong>The</strong> parameters describing the paraxial ray with respect to the<br />
known central ray are its distance to the central ray and the deviation of its slowness vector from<br />
the slowness vector of the central ray. Here, paraxial ray theory implies that the values of these<br />
parameters at the anterior surface are linearly dependent on those at the posterior surface. This<br />
can be written as (see, e.g., Bortfeld, 1989; Hubral et al., 1992; Schleicher et al., 1993)<br />
where<br />
Δx G = AΔx S + BΔp S , (2.1a)<br />
Δp G = C Δx S + DΔp S , (2.1b)<br />
Δx S = x S − x S and Δx G = x G − x G (2.2a)
8 <strong>The</strong>ory<br />
z<br />
O<br />
z<br />
x<br />
p S<br />
£¡£¡£ ¤¡¤<br />
¤¡¤ £¡£¡£<br />
x S<br />
¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥ ¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦ �¡��<br />
¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦ ¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥<br />
¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥ ¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦<br />
¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦ ¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥<br />
¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥ ¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦<br />
¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦¡¦ ¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥¡¥<br />
G<br />
�¡�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�<br />
G ��<br />
�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�¡�<br />
�¡�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�<br />
�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�¡�<br />
�¡�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�<br />
�¡�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�<br />
p<br />
�¡�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�<br />
�¡�¡�¡�¡�¡� �¡�¡�¡�¡�¡�¡�<br />
S<br />
central ray<br />
x S<br />
§¡§¡§¡§¡§¡§¡§¡§ ¨¡¨¡¨¡¨¡¨¡¨¡¨<br />
�¡��<br />
¨¡¨¡¨¡¨¡¨¡¨¡¨ §¡§¡§¡§¡§¡§¡§¡§<br />
§¡§¡§¡§¡§¡§¡§¡§ ¨¡¨¡¨¡¨¡¨¡¨¡¨<br />
¨¡¨¡¨¡¨¡¨¡¨¡¨ §¡§¡§¡§¡§¡§¡§¡§<br />
p S<br />
§¡§¡§¡§¡§¡§¡§¡§ ¨¡¨¡¨¡¨¡¨¡¨¡¨<br />
¨¡¨¡¨¡¨¡¨¡¨¡¨ §¡§¡§¡§¡§¡§¡§¡§<br />
§¡§¡§¡§¡§¡§¡§¡§ ¨¡¨¡¨¡¨¡¨¡¨¡¨<br />
§¡§¡§¡§¡§¡§¡§¡§ ¨¡¨¡¨¡¨¡¨¡¨¡¨<br />
paraxial ray<br />
¢¡¢ ¡ ¡<br />
©¡© �<br />
� ©¡©<br />
©¡© �<br />
©¡© �<br />
p � ©¡©<br />
©¡© �<br />
G ��<br />
� ©¡©<br />
O x<br />
� �¡�<br />
� �¡�<br />
G<br />
x<br />
G<br />
G<br />
©¡© �<br />
� ©¡©<br />
S<br />
x<br />
anterior<br />
surface<br />
posterior<br />
surface<br />
Figure 2.1: Sketch of a two-dimensional inhomogeneous and isotropic medium. <strong>The</strong> central ray<br />
(depicted in green) passes through this medium starting on the anterior surface at S and ending<br />
on the posterior surface at G. <strong>The</strong> paraxial ray (depicted in red) is in close vicinity of the central<br />
ray. <strong>The</strong> quantities describing the central ray at the anterior surface are shown in green, those<br />
describing the paraxial ray in red.
2.1 Basics 9<br />
����<br />
O<br />
z<br />
����<br />
S<br />
p S<br />
��������������������� ��������������������� ��<br />
��������������������� p S<br />
���������������������<br />
S<br />
��������������������� ���������������������<br />
��������������������� ���������������������<br />
paraxial ray<br />
� ���<br />
Figure 2.2: Construction of the ray slowness vector projection: <strong>The</strong> ray slowness vector is firstly<br />
projected onto the tangent to the anterior surface at S. <strong>The</strong> resulting vector p S,T is then projected<br />
onto the tangent to the anterior surface at S,which coincides with the x-axis of the coordinate<br />
system [x,z].<br />
are the differences of the projected coordinates of the central and paraxial ray at the anterior and<br />
posterior surface, and<br />
p<br />
S,T<br />
Δp S = p S − p S and Δp G = p G − p G (2.2b)<br />
are the differences of the respective projected ray slowness values.<br />
Equations (2.1) can also be written in vector and matrix form by<br />
� � � �<br />
ΔxG ΔxS<br />
= T . (2.3)<br />
ΔpG ΔpS T is the so-called surface-to-surface ray propagator matrix (Hubral et al., 1992; Schleicher et al.,<br />
1993) and reads<br />
� �<br />
A B<br />
T = . (2.4)<br />
C D<br />
Matrix T has the following important properties:<br />
1. <strong>The</strong> Symplecticity<br />
x
10 <strong>The</strong>ory<br />
This property states that the inverse of the propagator matrix T can be written as<br />
Consequently holds<br />
T −1 =<br />
� A B<br />
C D<br />
�−1<br />
=<br />
� D −B<br />
−C A<br />
This implies that the T has only three independent values.<br />
�<br />
. (2.5)<br />
AD − BC = 1 . (2.6)<br />
2. <strong>The</strong> Chain Rule<br />
<strong>The</strong> chain rule states that we can introduce an arbitrary point M along the central ray SG<br />
to decompose the propagator matrix T (G,S) in the following way.<br />
T (G,S) = T (G,M)T(M,S). (2.7)<br />
T (M,S) denotes the propagator matrix of the first ray segment and T (G,M) the propagator<br />
matrix of the second one.<br />
3. <strong>The</strong> Reverse Ray Property<br />
A hypothetical ray, traveling from the geophone location G to the source at S on the same<br />
path as the ordinary ray SG but in the opposite direction is called reverse ray. <strong>The</strong> propagator<br />
matrix of this reverse ray T ∗ is related to the propagator matrix T of the ordinary ray<br />
by the expression<br />
T ∗ �<br />
A∗ B∗ =<br />
C ∗ D ∗<br />
�−1<br />
=<br />
� D B<br />
C A<br />
�<br />
. (2.8)<br />
By comparing equations (2.8) and (2.5) we can see that the inverse propagator matrix T −1<br />
and the propagator matrix T ∗ of the reverse ray are not identical.<br />
2.2 Fundamental traveltime expressions<br />
As the surface-to-surface propagator matrix T is valid for arbitrary rays, not only the situation<br />
as shown in Figure 2.1 but also a measurement geometry as depicted in Figure 2.3 can be<br />
considered, which is the usual measurement geometry in reflection seismics. In the following<br />
the surface-to-surface propagator matrix T is used to derive in a very convenient way the<br />
traveltime moveout of any paraxial ray. This requires only quantities that refer to the central
2.2 Fundamental traveltime expressions 11<br />
measurement surface<br />
S<br />
Δx S<br />
z S<br />
S<br />
x<br />
S<br />
central ray<br />
β S β G<br />
R<br />
R<br />
G<br />
Δ x G<br />
G<br />
paraxial ray<br />
reflector<br />
Figure 2.3: Sketch of a 2D model with a curved measurement surface. <strong>The</strong> central ray SRG and<br />
a paraxial ray S R G in its vicinity are shown. <strong>The</strong> initial point S is the origin of the local axis x S<br />
and the end point G is the origin of the local axis x G . Both axes are tangent to the measurement<br />
surface at the source S and receiver G, respectively.<br />
z G<br />
x<br />
G
12 <strong>The</strong>ory<br />
S<br />
1<br />
� ���<br />
S<br />
S<br />
paraxial ray<br />
��� �����<br />
central ray<br />
��� ���<br />
Figure 2.4: <strong>The</strong> paraxial ray from S to G in the vicinity of the central ray from S to G.<br />
ray. For a finite-offset central ray in 2D as shown in Figure 2.3, these quantities are the take-off<br />
angles of the central ray at S and G and the three independent elements of the propagator<br />
matrix T . From now on, we choose the origin of our local coordinate systems, without loss of<br />
generality, to be located in the source and receiver points, respectively, according to Figure 2.3.<br />
As mentioned before Zhang et al. (2001b) and Bergler (2001) have extended the standard<br />
zero-offset (ZO) CRS stack traveltime formula to finite-offset (FO). However, they continued<br />
to restrict the discussion to a planar measurement surface. This is not necessary, because the<br />
surface-to-surface propagator matrix formalism does not require a planar measurement surface<br />
to be valid. We will now following the lines of Chira et al. (2001) and Schleicher et al. (2002),<br />
re-derive and extend their results to the more general case of considering a curved measurement<br />
surface.<br />
2.2.1 Traveltime of a paraxial ray<br />
In this section we discuss how to derive the traveltime moveout of an arbitrary paraxial ray with<br />
respect to the known central ray. We are not looking for a representation of the exact traveltime<br />
moveout but for an asymptotic one that is valid up to the second order of the dislocation between<br />
the source and receiver points of the paraxial and the central ray. In the following the word “exact”<br />
is used synonymous to “exact within the scope of the zero-order ray theory”.<br />
G<br />
1<br />
�<br />
G<br />
G
2.2 Fundamental traveltime expressions 13<br />
We assume that we can always find a wavefront that pertains to both rays, i.e., paraxial and<br />
central ray belong to the same ray family and fulfill the eikonal equation with the same initial<br />
conditions. Only in this situation, the description of the paraxial ray in terms that refer to the<br />
central ray makes sense. This situation is depicted in Figure 2.4. In contrast to the illustration<br />
these traveltime differences shall be infinitesimal small. Considering infinitesimal small traveltime<br />
differences is fully sufficient, as we are only looking for a second-order approximation of<br />
the traveltime moveout. According to Figure 2.4 the difference between the traveltime from S to<br />
G, t(S,G), and the traveltime from S to G, t(S,G), can be expressed as<br />
dt = t(S,G) −t(S,G) = dt G − dt S , (2.9)<br />
where dt S is the traveltime from S 1 to S and dt G is the traveltime from G 1 to G.<br />
<strong>The</strong> change of the traveltime due to a infinitesimal perturbation of the paraxial, e.g., source point,<br />
d(x S − x S1 ) is given by<br />
dt S = p S · d(x S − x S1 ) , (2.10a)<br />
if p S = p S1 is assumed. Please note the analogy to the well known scalar equation v = ds/dt.<br />
This is equal to<br />
dt S = p S · d(x S − x S ) . (2.10b)<br />
because p S is parallel to d(x S − x S1 ).<br />
Certainly relations (2.10) hold also for the paraxial receiver point G. Inserting this into (2.9)<br />
leads to Hamilton’s equation for two point ray-tracing, which reads<br />
dt = p G · d(x G − x G ) − p S · d(x S − x S ), (2.11)<br />
and which is basically nothing else than an alternative mathematical formulation of Fermat’s<br />
principle, saying that the first derivative of the traveltime in the direction vertical to the ray<br />
vanishes.<br />
Our aim is to obtain a second-order approximation of the traveltime of the paraxial ray, therefore<br />
it is sufficient to consider only linear terms of Δx S and Δx G within the Hamilton’s equation. This<br />
enables us, according to Appendix A, to reduce the vectors in equation (2.11) to scalar values,<br />
because the z-components of the dot products are already of second or higher order in Δx S and<br />
Δx G , respectively.<br />
dt = p G d(Δx G ) − p S d(Δx S ) . (2.12)<br />
Now we can benefit from the surface-to-surface propagator matrix T which we introduced in the<br />
last section. If we insert equations (2.2) into equations (2.1), we get,<br />
x G − x G = Ax S − x S + B p S − p S , (2.13a)<br />
p G − p G = C x S − x S + D p S − p S . (2.13b)
14 <strong>The</strong>ory<br />
Solving the first of the two equations above for p S and the second for p G , whereas we substitute<br />
p S instantly, leads to<br />
pS = pS + B −1 ΔxG − B −1 AΔxS (2.14a)<br />
pG = pG +C ΔxS + DB −1 ΔxG − DB −1 AΔxS . (2.14b)<br />
Finally we have to insert equations (2.14a) and (2.14b) into the scalar representation of Hamilton’s<br />
equation (2.12) and integrate. Considering the symplecticity of the T matrix, we obtain the<br />
traveltime formula<br />
tpar(ΔxS ,ΔxG ) =tSG + pG ΔxG − pS ΔxS − ΔxS B −1 ΔxG + 1<br />
2 ΔxS B−1 AΔxS + 1<br />
2 Δx′ G DB−1 Δx ′ G ,<br />
(2.15)<br />
where t SG is the two-way traveltime of the central ray SG.<br />
This traveltime representation is known as the parabolic traveltime. By squaring this equation<br />
and retaining only the terms up to the second order in the dislocations Δx S and Δx G we obtain<br />
the hyperbolic traveltime equation (Ursin, 1982).<br />
t 2 hyp (ΔxS ,ΔxG ) = � �2 tSG + pG ΔxG − pS ΔxS �<br />
+ 2tSG −ΔxS B −1 ΔxG + 1<br />
2 ΔxS B−1AΔxS + 1<br />
2 ΔxGDB−1 �<br />
ΔxG .<br />
(2.16)<br />
Ursin suggested, after systematic investigations that the hyperbolic traveltime formula is a better<br />
approximation to the real traveltime response than the parabolic one. This was later verified by<br />
the work of Höcht (1998), Müller (1999), Jäger (1999), and Bergler (2001).<br />
<strong>The</strong> general representation of the second-order Taylor expansion of the traveltime t(Δx S ,Δx G )<br />
reads<br />
tpar(Δx S ,Δx G ) = t SG + a Δx S + b Δx G + c Δx S Δx G + d (Δx S ) 2 + e (Δx G ) 2 , (2.17)<br />
with the five coefficients {a,b,c,d,e}.<br />
We can also derive a hyperbolic representation of this traveltime formula by squaring and retaining<br />
only terms up to the second order in Δx S and Δx G . We get<br />
t 2 hyp (ΔxS ,ΔxG ) = � �2 �<br />
tSG + a ΔxS + b ΔxG + 2 tSG c ΔxSΔxG + d (ΔxS ) 2 + e (ΔxG ) 2� . (2.18)<br />
Of course, these two traveltime equations do not require any assumptions concerning the subsurface<br />
media or the shape of the acquisition surface. Only in order to relate the general parameters<br />
{a,b,c,d,e} to physical quantities, assumptions respective subsurface structure and measurement<br />
surface shape are necessary (see Section 2.2.3).
2.2 Fundamental traveltime expressions 15<br />
If we compare these two equations with equations (2.15) and (2.16) and introduce the two angles<br />
β S and β G , we can relate {a,b,c,d,e} to the three independent values of the propagator matrix T<br />
and β S and β G . It results<br />
a = − sinβ S /v S , (2.19a)<br />
b =sinβ G /v G , (2.19b)<br />
c = − 1/B , (2.19c)<br />
d =A/2B , (2.19d)<br />
e =D/2B , (2.19e)<br />
where β S is the take-off angle of the central ray in S and β G the respective emergence angle in G.<br />
Both are measured with regard to the surface normal (see Figure 2.3). <strong>The</strong>y are defined as<br />
β S = sin −1 (v S p S · t S ) and β G = sin −1 (v G p G · t G ) , (2.20)<br />
where p S , t S and p G , t G are the slowness and surface-tangent vectors at S and G, respectively.<br />
Equations (2.19 c,e,d) show that it is mandatory to require B not to be equal zero, i.e., the endpoint<br />
of the central ray must not be a caustic point.<br />
2.2.2 Determination of the coefficients<br />
In order to use one of the two traveltime formulas stated above for the purpose of a FO CRS<br />
stack, it is necessary to determine those values for the five parameters {a,b,c,d,e} that parameterize<br />
that traveltime surface that approximates the real reflection response, best. This is done by<br />
means of a CRS stack-based coherency analysis which is directly applied to the multi-coverage<br />
dataset (Höcht, 1998; Müller, 1999; Mann, 2002). <strong>The</strong>re, those values for {a,b,c,d,e} are determined<br />
which yield the highest coherency value. <strong>The</strong> traveltime t SG of the considered central<br />
ray and the two near-surface velocities v S and v G are assumed to be known. Due to the fact<br />
that a simultaneous search for five parameters would demand an immense computational effort,<br />
Bergler (2001) and Mann (2002) presented strategies for the FO and the ZO CRS stack which<br />
allow to split the search into separate one and two parametric searches, respectively. <strong>The</strong>se<br />
searches are performed within gathers that are related to special seismic configurations, i.e., the<br />
common-midpoint (CMP) gather, common-offset (CO) gather, the common-shot (CS) gather and<br />
the common-receiver (CR) gather. Within these special gathers the traveltime does not depend on<br />
all parameters simultaneously. Consequently it is possible to obtain all parameters by performing<br />
a convenient sequence of one and two parametric searches. However, these parameters are only<br />
an approximation of the ideal parameter set that would result from a simultaneous search.<br />
To apply these strategies to the more general case of a curved measurement surface, we have
16 <strong>The</strong>ory<br />
to modify the conventional CMP and CO measurement configurations, as described for a planar<br />
measurement surface, slightly. This is necessary, because for a curved measurement surface it<br />
is not reasonable to consider a CMP configuration, where the physical midpoint between source<br />
and receiver (not to confuse with the 1D midpoint coordinate introduced in Section 2.3) is fixed.<br />
<strong>The</strong> same holds for the CO configuration, where the 2D dislocation (offset) between source and<br />
receiver is constant. (However, it is a matter of interpretation. If midpoint and offset are <strong>under</strong>stood<br />
as 1D midpoint and offset coordinates, according to their definition in Section 2.3 (see<br />
equations (2.39)), then the CMP and CO configurations coincide with their extensions, defined<br />
in the following.)<br />
According to Chira et al. (2001) we introduce two measurement configurations which are natural<br />
extensions of the CMP and CO configurations and are just as useful for the estimation of the<br />
searched parameters. <strong>The</strong>se are :<br />
<strong>The</strong> Odd-dislocation (OD) gather: This configuration is related to the CMP gather. Paraxial<br />
source point S and receiver point G are dislocated by the same amount in opposite directions<br />
with respect to the source and receiver points S and G of the central ray. <strong>The</strong> OD condition and<br />
the corresponding OD traveltime moveout are<br />
(OD) Δx = Δx G = −Δx S , t 2 OD = [t SG + (b − a) Δx]2 + 2t SG (d + e − c) (Δx) 2 . (2.21)<br />
<strong>The</strong> Even-dislocation (ED) gather: This configuration is related to the CO situation. Paraxial<br />
source point S and receiver point G are dislocated by equal amounts in the same direction with<br />
respect to the source and receiver points S and G of the central ray. <strong>The</strong> ED condition and the<br />
corresponding ED traveltime are<br />
(ED) Δx = Δx G = Δx S , t 2 ED = [t SG + (a + b) Δx]2 + 2t SG (c + d + e) (Δx) 2 . (2.22)<br />
In addition to the OD and the ED gather we have to consider the well known common-shot (CS)<br />
and the common-receiver (CR) gather. Here the conditions and traveltime formulas are<br />
(CS) Δx = Δx G , Δx S = 0 , t 2 CS = [t SG + b Δx]2 + 2t SG e (Δx) 2 . (2.23)<br />
(CR) Δx = Δx S , Δx G = 0 , t 2 CR = [t SG + a Δx]2 + 2t SG d (Δx) 2 . (2.24)<br />
Comparing the traveltime formulas in the different gathers reveals that they are all of the same<br />
structure, i.e., one-variable quadratic polynomials with two independent parameters, which read<br />
t 2 (Δx) = (t SG + m Δx) 2 + 2t SG n (Δx) 2 . (2.25)<br />
<strong>The</strong> estimation of these two parameters in at least three specific gather via coherency analysis<br />
is sufficient to determine all five traveltime parameters {a,b,c,d,e}. This can be done either<br />
using two one-parametric searches, where first the quadratic coefficient n is assumed to be zero,
2.2 Fundamental traveltime expressions 17<br />
or performing a two-parametric search which is more accurate but also more time consuming<br />
(Bergler, 2001). <strong>The</strong> relationship between the parameters m and n of the different gathers and<br />
the attributes {a,b,c,d,e} can be derived simply from the comparison of equation (2.25) with<br />
equations (2.21), (2.22), (2.23), and (2.24). This leads to<br />
a = m CR , b = m CS = m OD + m CR , c = n ED − n CS − n CR ,<br />
d = n CR , e = n CS .<br />
where, e.g., m CS is the linear coefficient and n CS is the quadratic coefficient in the CS gather.<br />
(2.26)<br />
We can express the elements of the propagator matrix T in a similar way when we insert equations<br />
(2.26) into (2.19) and make use of the symplecticity relation.<br />
2 nCR 1<br />
A =<br />
, B =<br />
,<br />
nCS + nCR − nED nCS + nCR − nED 4 n CR n CS<br />
2 nCS C =<br />
+ nED − n<br />
nCS + nCR − n CS − nCR , D =<br />
.<br />
ED<br />
nCS + nCR − nED 2.2.3 Acquisition topography and near-surface velocity gradient<br />
(2.27)<br />
So far we made no assumptions concerning the acquisition topography. With respect to the subsurface<br />
structure the only assumption was that the near-surface velocity at S and G is known. Of<br />
course, the described paraxial vicinity is respectively small if the subsurface structure is unfavorable.<br />
Traveltime formulas (2.15) and (2.16) are fully sufficient to describe the second-order<br />
traveltime moveout of any paraxial ray by means of the three independent elements of matrix<br />
T and the two take-off angles β S and β G . <strong>The</strong> next step is to relate the parameters {a,b,c,d,e}<br />
and thus also the elements of T to so-called wavefield attributes, physical properties referred<br />
to the central ray that are theoretically measurable at the measurement surface. <strong>The</strong>se are, as<br />
mentioned before, very useful, e.g., for subsequent inversion or Fresnel-zone and geometrical<br />
spreading estimation. To do this, previously known information about acquisition surface and<br />
<strong>under</strong>lying top-layer has to be considered explicitly within the traveltime formulas, in order to<br />
separate this information from the searched-for wavefield attributes. To give an example: At<br />
the moment we are already able to determine two physical properties of the central ray from<br />
the parameters {a,b,c,d,e}, using equations (2.19). <strong>The</strong>se are the take-off angles β S and β G ,<br />
which still include the influence of the topography, as they are measured against the normal of<br />
the surface tangent in S and G, respectively. In the case of β S and β G the topography dependency<br />
does not conceal the physical meaning, thus it is not absolutely necessary to remove it. But it<br />
can be easily removed by introducing the dip angles α S and α G of the surface tangents in S and
18 <strong>The</strong>ory<br />
G and defining alternative take-off angles, now measured against the horizontal. We do this later<br />
in the zero-offset case (see Section 2.3), where we introduce the global ZO take-off angle β g<br />
0<br />
according to Figure 2.6 and equation (2.52) in order to split the take-off angle measured against<br />
the horizontal, which is an attribute of the central ray, from the surface dip, which is an attribute<br />
of the measurement surface.<br />
In oil exploration, it is usual to drill shallow holes along the seismic line to get information<br />
which can be used for static corrections. Consequently the near-surface velocity gradient is often<br />
known. In the following we do not only assume to know the acquisition topography and<br />
near-surface velocity, but also the near-surface velocity gradient. <strong>The</strong> propagator matrix T obviously<br />
incorporates, besides the overall properties of the propagating medium and the acquisition<br />
geometry, also these characteristics. More specifically, it will be seen to depend on the near<br />
surface velocity and its gradient, as well as on the curvature of the measurement surface at the<br />
source and receiver points of the central ray. Following Červen´y (2001), Chapter 4, we assume<br />
the measurement surface, in the vicinity of the source and receiver points of the central ray to be<br />
representable by parabolas. <strong>The</strong>se read<br />
zS = − 1<br />
2 KS x2S and zG = − 1<br />
2 KG x2G , (2.28)<br />
whereas the respective local coordinate systems have their origins in S and G (see 2.3) and<br />
the curvature of the surface in this points is denoted by K S and K G . To be in conformance<br />
with Červen´y we choose the negative sign in the equations above. This means that the surface<br />
curvature is defined in analogy to the curvature of an upcoming wavefront. Due to the fact that<br />
Červen´y treats the general 3D case and we are only looking for 2D solutions we assume the<br />
y-coordinate axis always to point into the drawing plane, according to the right hand rule. In<br />
contrast to Červen´y who chooses the surface normals and consequently the z-axes always in a<br />
way that keeps the angles β S and β G acute and positive, we choose both surface normals to point<br />
upwards which leads to acute angles that can be also negative (see (2.20)).<br />
Up to this point the T matrix can be seen as a black box formulation. To determine the dependency<br />
of its elements {A,B,C,D} on the properties of the measurement surface and on the<br />
near-surface velocities at S and G, it is necessary to relate this matrix to a corresponding “intrinsic”<br />
propagator matrix. Such an intrinsic matrix can be readily selected as the 2D, in-plane,<br />
ray-centered propagator matrix Π which is extensively discussed in Červen´y (2001). It reads<br />
Π(G,S) =<br />
� Q1 Q 2<br />
P 1 P 2<br />
�<br />
, with Q1P2 − Q2P1 = 1 . (2.29)<br />
If the acquisition surface in the vicinity of S and G can be represented according to equations<br />
(2.28) and if there are only first order velocity variations in the vicinity of S and G, then the matrix<br />
Π is related to the surface-to-surface propagator matrix T in the following way (see Červen´y
2.2 Fundamental traveltime expressions 19<br />
(2001) equation (4.4.90) with a different notation):<br />
T (G,S) = Y(G) Π(G,S) Y −1 (S) . (2.30)<br />
<strong>The</strong> two matrices Y(G) and Y −1 (S) that constitute the relationship depend only on properties<br />
of the measurement surface and of the top-layer in the vicinity of S and G (see Červen´y (2001),<br />
equation (4.13.21) with a different notation).<br />
with<br />
� �<br />
1/cosβG 0<br />
Y(G) =<br />
, (2.31)<br />
(EG − cosβG KG /vG )/cosβG cosβG Y −1 � �<br />
cosβS 0<br />
(S) = −<br />
, (2.32)<br />
− (ES + cosβS KS /vS )/cosβS 1/cosβS det(Y(G)) = det(Y −1 (S)) = 1 . (2.33)<br />
E S and E G are the so-called inhomogeneity factors and account for the first-order velocity variations<br />
in the vicinity of the source and receiver points of the central ray, respectively. <strong>The</strong>y are<br />
given by (see Červen´y, 2001, equation (4.13.20) with a different notation) and read<br />
E S = − sinβ S<br />
v 2 S<br />
E G = − sinβ G<br />
v 2 G<br />
[(1 + cos 2 β S )(∂x S v) S + cosβ S sinβ S (∂z S v) S ] , (2.34a)<br />
[(1 + cos 2 β G )(∂x G v) G − cosβ G sinβ G (∂z G v) G ] . (2.34b)<br />
Here, (∂x S v) S and (∂z S v) S denote the 2D in-plane components of the medium velocity gradient<br />
(∇v) at the source point S. An analog meaning holds for (∂x G v) G and (∂z G v) G , thus<br />
(∇v) S =<br />
� ∂x S v<br />
∂z S v<br />
�<br />
| S and (∇v) G =<br />
� ∂x G v<br />
∂z G v<br />
�<br />
| G . (2.35)<br />
Note that the two inhomogeneity factors E S and E G are no pure attributes of the top-layer in<br />
the vicinity of S and G, because they depend not only on the near-surface velocity gradients<br />
(∇v) S and (∇v) G , but also on the take-off angles β S and β G . In the following v S ,v G ,K S ,K G and<br />
(∇v) S ,(∇v) G are called (near)surface attributes. Equation (2.30) enables us to express the components<br />
{A,B,C,D} of the surface-to-surface propagator matrix T , in terms of their counterpart<br />
components {P 1 ,Q 1 ,P 2 ,Q 2 } of the ray-centered propagator matrix Π using characteristic quantities<br />
of the measurement surface and the top-layer in the vicinity of the points S and G. <strong>The</strong>se<br />
quantities are the near-surface velocities v S and v G the near-surface velocity gradients (∇v) S<br />
and (∇v) G and the surface curvatures K S and K G . To do this we have only to insert equations<br />
(2.31), (2.32), (2.34a) and (2.34b) into equation (2.29). Subsequently we can relate the quadratic
20 <strong>The</strong>ory<br />
coefficients c,d and e of the traveltime equations (2.15) and (2.16) to the elements of the surfaceto-surface<br />
propagator matrix, {A,B,C,D}, using equations (2.19). We remember at this point<br />
that the linear coefficients a and b depend only on the take-off and emergence angles (of the<br />
central ray) β S and β G and are implicitly related to the dip angle of the measurement surface at S<br />
and G.<br />
a = − sinβ S /v S , (2.36a)<br />
b =sinβ G /v G , (2.36b)<br />
c = − 1/B = cosβ S cosβ G /Q 2 , (2.36c)<br />
d =1/2 A/B = 1/2 ((Q 1 /Q 2 ) cos 2 β S − (cosβ S /v S )K S − E S ), (2.36d)<br />
e =1/2 D/B = 1/2 ((P 2 /Q 2 ) cos 2 β G − (cosβ G /v G )K G + E G ). (2.36e)<br />
Under the assumption that the (near)surface attributes {v S ,v G ,,K S ,K G ,(∇v) S ,(∇v) G } are known,<br />
this relation between the different sets of traveltime parameters gives us the opportunity to determine<br />
the values of the T or Π matrix. <strong>The</strong>se can be computed from the values {a,b,c,d,e},<br />
which previously have to be obtained via coherency analysis or directly by searching these values<br />
in a coherency analysis which uses a traveltime formula, where {a,b,c,d,e} are substituted<br />
according to equations (2.36). Most probably, these elements of the propagator matrices T or Π<br />
will become of utmost importance for the inversion of velocity models in the near future. First<br />
results in the zero-offset case, as reported in Biloti et al. (2001), are very encouraging. Despite<br />
all the benefit we can obtain from the five traveltime parameters for inversion problems, the<br />
main purpose of the <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> stack method remains the stacking. But also<br />
for the stacking it is important to relate the searched-for parameters {a,b,c,d,e} to physically<br />
meaningful quantities in order to estimate suitable limits in which a parameter is searched.<br />
2.3 Zero-offset situation<br />
In the last section, we have derived the very general expression (2.15) for the traveltime moveout<br />
of a ray that is paraxial to an arbitrary central ray. This expression is called the parabolic<br />
traveltime according to the mathematical form of this equation. From this equation we deduced<br />
the hyperbolic traveltime expression (2.16), which was shown to be more successful in the practical<br />
application of the CRS stack, by Höcht (1998), Müller (1999), Jäger (1999) for zero-offset<br />
and by Bergler (2001) for common-offset. By means of this formulas we are able to create a<br />
stacked section with an arbitrary offset (FO CRS stack), using central rays with arbitrary shot<br />
and receiver dislocation. In the following we want to reduce our formulas to the simpler case of<br />
a central ray that impinges perpendicularly at the reflector. This particular central ray has a coincident<br />
source and receiver point S = G and consequently an offset that is equal to zero (ZO CRS
2.3 Zero-offset situation 21<br />
Figure 2.5: ZO situation: <strong>The</strong> central ray shown in green hits the reflector perpendicularly at<br />
the normal incidence point (NIP). Consequently, the down-going and up-coming ray paths are<br />
identical. <strong>The</strong> wavefronts of the NIP-wave depicted in light-blue and the Normal-wave depicted<br />
in blue travel along the central ray emerging at the coincident source and receiver point X 0<br />
with the wavefront curvatures K NIP and K N , respectively. For the illustration, the wavefronts<br />
are approximated by circular segments with the corresponding curvature of the wavefronts at<br />
the central ray. <strong>The</strong> measurement surface depicted in brown is a smooth curve with negative<br />
curvature K 0 in X 0 .
22 <strong>The</strong>ory<br />
stack). This situation is shown in Figure 2.5. For the sake of simplicity and for consistency with<br />
former publications we denote the coincident source and receiver point of a central ray by X 0 . In<br />
order to specify the location of the coincident source and receiver point pairs of different central<br />
rays, a global coordinate system is used. In this global coordinate system, the x-coordinate of X 0<br />
is denoted by x 0 and the z-coordinate by z 0 . <strong>The</strong> origin of the global coordinate system can be<br />
chosen arbitrarily, but the origin of the local coordinate system is always X 0 . In the following X 0<br />
is called the central point, because it is the emergence point of the central ray to which the CRS<br />
traveltime surface t(m,h)| (x0 ,t0 ) is referred, whereas t0 is traveltime of the (zero-offset) central<br />
ray. Please note that we can simulate the zero-offset traveltime for any point in the vicinity of X 0 ,<br />
but these points must not be confused with the central point X 0 even if they are also coincident<br />
source and receiver points. All attributes that refer to the central point X 0 are indicated in the<br />
following with the index 0, to signify the ZO case.<br />
Considering this particular geometry leads to a strong simplification of the problem, because<br />
the up-going and the down-going ray paths coincide. Only the directions of propagation are<br />
opposite. This has two important consequences:<br />
1. As we have now only one point which is both, source and receiver, the take-off angle β S<br />
and the emergence angle β G cannot have different absolute values. But due to the fact that,<br />
per definition (equation (2.20)), the sign of the angle between a ray and a surface depends<br />
on the direction of ray propagation, we have the following relation.<br />
β S = − β G = β 0 . (2.37)<br />
This means that for a ZO central ray the take-off angle β S and the emergence angle β G<br />
reduce to one take-off angle which we will call β 0 .<br />
2. As the two ray paths as well as the source and receiver points coincide, also the surfaceto-surface<br />
propagator matrices of the ray and its respective reverse ray have to coincide.<br />
Using the property of the reverse ray propagator matrix T ∗ (2.8) stated in Section 2.1.3 we<br />
can write<br />
T ZO =<br />
Consequently holds in the ZO case<br />
� A B<br />
C D<br />
�<br />
=<br />
� D B<br />
C A<br />
�<br />
∗<br />
= T ZO . (2.38a)<br />
A = D . (2.38b)<br />
Thus, remembering the symplecticity relation (2.5), the ZO propagator matrix T ZO has<br />
only two independent values.<br />
<strong>The</strong> FO CRS stack traveltime formulas were dependent on five attributes: two angles and three<br />
independent values of the propagator matrix T or Π, respectively. In the ZO case this dependency
2.3 Zero-offset situation 23<br />
is reduced to one angle and two matrix elements. If we introduce midpoint and half-offset coordinates,<br />
instead of the source and receiver dislocations Δx S and Δx G , according to the relations<br />
m = Δx S + Δx G<br />
2<br />
, and h = Δx G − Δx S<br />
2<br />
, (2.39)<br />
we can express the parabolic (2.15) and hyperbolic (2.16) FO traveltime equations, now reduced<br />
to ZO, using only three parameters. This reads<br />
t par<br />
ZO = t 0 + σ 1 m + σ 2 m2 + σ 3 h 2 ,<br />
t hyp<br />
ZO = (t 0 + σ 1 m)2 + 2t 0 (σ 2 m 2 + σ 3 h 2 ) .<br />
(2.40)<br />
In view of our results for the FO case, equations (2.36), we readily find the relations that hold<br />
for the three coefficients {σ 1 ,σ 2 ,σ 3 } in the ZO case. We have only to replace β S and β G with<br />
β 0 , according to relation (2.37), and v S and v G with v 0 and substitute the matrix element D by A<br />
according to (2.38), within equations (2.19) and subsequently insert these into the parabolic and<br />
hyperbolic traveltime equations (2.17) and (2.18). Doing this we find the relations<br />
σ 1 = 2sinβ 0 /v 0 , σ 2 = (A − 1)/B and σ 3 = (A + 1)/B . (2.41)<br />
To find now the relation between the traveltime coefficients {σ1 ,σ2 ,σ3 }, and the (near)surface<br />
attributes {v0 ,K0 ,(∇v) 0 } at the coincident source and receiver point X0 we can follow the same<br />
strategy as in the FO case upon the use of the “intrinsic” ray-centered propagator matrix ΠZO � �<br />
Q1 Q<br />
ΠZO =<br />
2 , (2.42)<br />
P 1 P 2<br />
and reduce the expressions (2.36), which we derived by means of results from Červen´y (2001),<br />
to the ZO case like we did above. This, together with equations (2.41), leads after some straightforward<br />
algebraic calculations to the searched-for relationship between the traveltime parameters<br />
{σ 1 ,σ 2 ,σ 3 } and the surface characteristics {v 0 ,K 0 ,(∇v) 0 } in S 0 = G 0 ,<br />
where<br />
σ 1 = 2sinβ 0 /v 0 ,<br />
σ 2 = ((Q 1 + 1)/Q 2 ) cos 2 β 0 − (cosβ 0 /v 0 )K 0 − E 0 ,<br />
σ 3 = ((Q 1 − 1)/Q 2 ) cos 2 β 0 − (cosβ 0 /v 0 )K 0 − E 0 ,<br />
E0 = − sinβ0 v2 �<br />
(1 + cos<br />
0<br />
2 �<br />
β0 )(∂xv) 0 + cosβ0 sinβ0 (∂zv) 0<br />
= E S (S = S 0 )<br />
= −E G (G = G 0 ) .<br />
(2.43)<br />
(2.44)
24 <strong>The</strong>ory<br />
For the zero-offset situation it is possible to express the elements of the ray-centered propagator<br />
matrix Π by means of two wavefield attributes. <strong>The</strong>se are the wavefront curvatures of two<br />
conceptual eigenwaves (see Figure 2.5) which were firstly introduced in Hubral (1983). For the<br />
wavefront curvatures the sign convention according to Hubral and Krey (1980) is used. When a<br />
wavefront lags behind its tangent plane then the wavefront curvature is defined as positive. If the<br />
wavefront is ahead of its tangent plane the wavefront curvature is negative.<br />
1. <strong>The</strong> Normal-Incidence-Point-wave (NIP-wave) is related to a ray that originates from a<br />
point source at the coincident source and receiver point X 0 , hits the reflector perpendicular<br />
and travels back to its origin, using the same path, but in opposite direction. If we assume<br />
now a wave traveling along this ray with half the true medium velocity and starting at the<br />
reflector, its wavefront would reach X 0 with the curvature K NIP . Consequently a downward<br />
propagating wavefront with K(X 0 ) = −K NIP would shrink to a point reaching the reflector<br />
and would be reflected “into itself” and reach X 0 again with the curvature K NIP .<br />
2. <strong>The</strong> Normal-wave (N-wave) originates as a wavefront that coincides with the reflector in<br />
the vicinity of the reflection point NIP and propagates upwards reaching the surface at X 0<br />
with the curvature K N . If we propagated this wavefront back to the reflector, it would also<br />
be reflected “into itself” and reach X 0 again with the curvature K N .<br />
According to Hubral (1983) the elements of the ray-centered propagator matrix Π are related to<br />
these eigenwave curvatures as<br />
Π ZO =<br />
−v 0<br />
K NIP − K N<br />
� (KN + K NIP )/v 0<br />
2K N K NIP /v 2 0<br />
If we insert this into equation (2.43) we obtain<br />
2<br />
(K N + K NIP )/v 0<br />
σ 2 = (cos 2 β 0 /v 0 )K N − (cosβ 0 /v 0 )K 0 − E 0 ,<br />
σ 3 = (cos 2 β 0 /v 0 )K NIP − (cosβ 0 /v 0 )K 0 − E 0 .<br />
�<br />
. (2.45)<br />
(2.46)<br />
Note, in analogy to the FO case, the measurement surface in the vicinity of X 0 is assumed to be<br />
representable, in the local coordinate system of X 0 , by the parabola<br />
where K 0 is the measurement surface curvature at X 0 .<br />
z = − 1<br />
2 K 0 x2 , (2.47)
2.3 Zero-offset situation 25<br />
Finally we can rewrite our traveltime expressions for zero-offset, using only parameters that are<br />
related to physical properties that refer to the central ray and can be measured in the vicinity of<br />
the coincident source and receiver point X 0 . We get<br />
tpar(m,h) = t0 + 2 sinβ0 m +<br />
v0 1 �<br />
KN cos<br />
v0 2 �<br />
2<br />
β0 − K0 cosβ0 − v0 E0 m (2.48a)<br />
+ 1 �<br />
KNIP cos<br />
v0 2 �<br />
2<br />
β0 − K0 cosβ0 − vo E0 h , and<br />
t 2 �<br />
hyp (m,h) = t0 + 2 sinβ �2 0<br />
m +<br />
v0 2 t �<br />
0<br />
KN cos<br />
v0 2 �<br />
2<br />
β0 − K0 cosβ0 − v0 E0 m (2.48b)<br />
+ 2 t �<br />
0<br />
KNIP cos<br />
v0 2 �<br />
2<br />
β0 − K0 cosβ0 − v0 E0 h .<br />
Under the assumption of a slowly varying or even constant near-surface velocity, the inhomogeneity<br />
factor E 0 that accounts for the gradient of the medium velocity at the source and receiver<br />
points virtually vanishes. This leads to a simplified version of equation (2.48b), which was previously<br />
presented by Chira and Hubral (2001). For E 0 = 0 this reads<br />
tpar(m,h) = t0 + 2 sinβ0 m +<br />
v0 1 �<br />
KN cos<br />
v0 2 � 2<br />
β0 − K0 cosβ0 m<br />
+ 1 �<br />
KNIP cos<br />
v0 2 � 2<br />
β0 − K0 cosβ0 h , and<br />
t 2 hyp (m,h) =<br />
�<br />
t0 + 2 sinβ �2 0<br />
m +<br />
v0 2 t �<br />
0<br />
KN cos<br />
v0 2 � 2<br />
β0 − K0 cosβ0 m<br />
+ 2 t �<br />
0<br />
KNIP cos<br />
v0 2 � 2<br />
β0 − K0 cosβ0 h .<br />
(2.49a)<br />
(2.49b)<br />
For comparison we can reduce this equations to the conventional 2D ZO CRS stack traveltime<br />
formulas, valid for a planar measurement surface (Höcht, 1998; Müller, 1999; Jäger, 1999).<br />
<strong>The</strong>se can be obtained by setting K 0 = 0 in the formulas above and reads<br />
tpar(m,h) = t0 + 2 sinβ0 m +<br />
v0 1 �<br />
KN cos<br />
v0 2 � 2<br />
β0 m<br />
+ 1 �<br />
KNIP cos<br />
v0 2 �<br />
2<br />
β0 h , and<br />
t 2 �<br />
hyp (m,h) = t0 + 2 sinβ �2 0<br />
m +<br />
v0 2 t �<br />
0<br />
KN cos<br />
v0 2 �<br />
2<br />
β0 m<br />
+ 2 t �<br />
0<br />
KNIP cos<br />
v0 2 �<br />
2<br />
β0 h .<br />
(2.50a)<br />
(2.50b)<br />
Of course it is also possible to consider the case of a planar measurement surface having a near<br />
velocity gradient. This case is committed to the reader.
26 <strong>The</strong>ory<br />
Please note that, in practice, the traveltime formulas stated above are very inconvenient to use,<br />
because for different locations of X 0 different local coordinate systems have to be used. In Section<br />
(2.6) we introduce slightly modified traveltime formulas referring to the global coordinate system<br />
in which the source and receiver coordinates are usually measured. In general, these traveltime<br />
formulas are much more convenient to implement, because the source and receiver coordinates<br />
do not have to be transferred to different local coordinate systems.<br />
2.3.1 Relationship between (near)surface and wavefield attributes<br />
If we assume a curved acquisition surface, where the near-surface velocity gradient (∇v) 0 and the<br />
curvature K 0 is known in every central point X 0 , then we can approximate the traveltime response<br />
of an arbitrary reflector segment up to the second order, by using either the parabolic (2.48a) or<br />
the hyperbolic (2.48b) traveltime equation. This means that we have to find either the parabolic<br />
or the hyperbolic traveltime surface that fits best to the reflection response of the considered (but<br />
unknown) reflector segment. Doing this, e.g., by means of a coherency analysis, we obtain values<br />
of the three wavefield attributes {β 0 ,K N ,K NIP } that parameterize our traveltime formula optimal,<br />
and have in the context of the made approximations their defined physical meaning. However it is<br />
not mandatory to use these set of equations to obtain a stacked ZO section from data measured on<br />
a curved measurement surface with near surface velocity gradient. Equations (2.49) or (2.50) are<br />
also valid to represent a second-order approximation of the considered traveltime response. <strong>The</strong>y<br />
have the same mathematical form as equations (2.48) - they are of second order (parabolic and<br />
hyperbolic, respectively) and have three free parameters. In other words, equations (2.49) and<br />
(2.50) describe the same surfaces in the [t,m,h]-space as equations (2.48), but with other values<br />
for the parameters {β 0 ,K N ,K NIP }. Of course these values are not the real values of the physical<br />
wavefield attributes {β 0 ,K N ,K NIP }. <strong>The</strong>y implicitly contain the influence of the topography and<br />
of the near-surface velocity gradient, which is not explicitly considered within the traveltime<br />
formulas. We will denote this apparent wavefield attributes with a star in the following and call<br />
them pseudo attributes. If we know the (near)surface attributes {K 0 ,v 0 ,(∇v) 0 } and the dip angle<br />
of the surface in X 0 , i.e., α 0 (see Figure 2.6), we can convert these pseudo attributes to their<br />
true counterparts. <strong>The</strong> searched-for relationship can be found by comparing the coefficients of<br />
formulas (2.48) and (2.50). It makes no difference if we compare the parabolic or the hyperbolic<br />
equations.<br />
Doing this we find:<br />
β 0 = β ∗ 0 , (2.51a)<br />
K N = K ∗ N + K 0 cos(β ∗ 0 ) + E 0 v 0<br />
cos 2 (β ∗ 0 )<br />
, (2.51b)
2.3 Zero-offset situation 27<br />
central ray<br />
Figure 2.6: <strong>The</strong> relationship between the take-off angles of the normal ray, β0 and β g and the<br />
0<br />
dip angle α0 for a curved measurement surface. Note that β0 is measured in the local and β g<br />
0 in<br />
the global coordinate system. <strong>The</strong> angles are defined in the mathematical positive direction of<br />
rotation (counterclockwise). Consequently β0 has a negative value in the figure above.<br />
Please note: For this figure the origin of the global coordinate system is chosen to coincide with<br />
X0 , which is also the origin of the local coordinate system. Of course, in general this is not the<br />
case.
28 <strong>The</strong>ory<br />
and<br />
K NIP = K ∗ NIP + K 0 cos(β ∗ 0 ) + E 0 v 0<br />
cos 2 (β ∗ 0 )<br />
. (2.51c)<br />
We observe that the obtained take-off angle β 0 remains the same, irrespective of whether we use<br />
equations (2.49), (2.50) or equations (2.48). <strong>The</strong> reason for this is that the linear term in our<br />
traveltime formulas depends only on the angle between the tangent to the measurement surface<br />
in X 0 and the tangent to the considered reflector in the normal incidence point (NIP). This angle<br />
is, within the range of the <strong>under</strong>lying assumptions, equal to β 0 . Thus, it makes no difference to<br />
the linear term if the surface is planar or not.<br />
<strong>The</strong>re is one point that should be emphasized. If we use the plain topography formula (2.50) and<br />
do not know the real topography, we cannot interpret the obtained take-off angle β 0 , because it<br />
is measured against the unknown surface normal in X 0 . Only if we know the dip angle α 0 we are<br />
able to interpret β 0 geometrically. In some cases it can be useful to transfer β 0 from the local to<br />
a surface independent global coordinate system (see Figure 2.6), according to the relation<br />
β g<br />
0 = β0 + α0 . (2.52)<br />
Note that here and in the following the global coordinate system is defined as:<br />
x-axis parallel to the horizontal and z-axis parallel to the depth direction.<br />
2.3.2 Normal-moveout (NMO) and root-mean-square (RMS) velocities<br />
In the following discussion we will no longer consider the parabolic traveltime formulas, because<br />
they are not used in the current implementations of the CRS stack (see Höcht, 1998; Müller,<br />
1999; Bergler, 2001). Nevertheless, similar considerations would also hold in this case.<br />
NMO velocity<br />
To look at the odd-dislocation (OD) gather 2.2.2, which is closely related to the CMP gather, we<br />
have to insert the condition m = 0 into the traveltime formula (2.48b). We get<br />
t 2 (hyp,OD) = t2 0 + 2 t 0<br />
v 0<br />
�<br />
KNIP cos 2 �<br />
2<br />
β0 − K0 cosβ0 − v0 E0 h . (2.53)<br />
In analogy to the NMO velocity defined for a planar measurement surface (Shah, 1973), we can<br />
introduce the NMO velocity for a curved measurement surface, having a near-surface velocity
2.3 Zero-offset situation 29<br />
gradient, by rewriting equation (2.53) in the following way:<br />
t 2 (hyp,OD) (h) = t2 0<br />
v 2 NMO =<br />
4h2<br />
+<br />
v2 ,with<br />
NMO<br />
(2.54a)<br />
2v<br />
� 0<br />
t0 KNIP cos2 � .<br />
β0 − K0 cosβ0 − v0 E0 (2.54b)<br />
According to this definition the terms NMO velocity and stacking velocity can be used synonymously.<br />
This definition is common within the context of the CRS stack. It defines the NMO<br />
velocity as that velocity that specifies those hyperbolas in the CMP gather which yield the highest<br />
coherence and thus the best stacking result. Please note that this definition of the NMO<br />
velocity is a little bit different from the classical definition as that velocity which is best to reduce<br />
the quasi-hyperbolic taveltimes in the CMP gather to a horizontal line. We would get a<br />
slightly different expression for the NMO velocity if we would consider the parabolic instead<br />
of the hyperbolic traveltime formula. E.g., the parabolic NMO velocity does not depend on t 0 .<br />
Please note that even for a planar acquisition topography with constant near-surface velocity,<br />
v NMO can have imaginary values if K NIP < 0. However, this case is much more probable for a<br />
curved measurement surface with near-surface velocity gradient, as it can be seen by looking at<br />
equation (2.54b).<br />
It is evident that to obtain the NMO velocity for a curved measurement surface without nearsurface<br />
velocity gradient one has only to set E0 = 0 in the equation above. Similar if the measurement<br />
surface is planar, one has to set K0 = 0. An important question is: what NMO velocity<br />
do we find if we use, e.g., the traveltime formula valid for a planar measurement surface without<br />
near surface velocity gradient to stack data that was measured on a curved surface with<br />
(∇v) 0 �=�0? <strong>The</strong> answer can easily be found by looking at equation (2.54a). (4/v2 NMO ) is always<br />
the coefficient of the h 2 term of the considered traveltime formula. Irrespective of the used traveltime<br />
formula ((2.48b), (2.49c) or (2.50c)), this coefficient has to have the same value, if the<br />
resulting traveltime surface shall fit to the respective reflection response in the data. Thus the<br />
found NMO velocity has also the same value. However, the subsequently obtained wavefield<br />
attribute K NIP depends on, whether the real curvature and velocity gradient of the measurement<br />
surface is considered within equation (2.54b) or not. This can also be demonstrated by inserting<br />
equations (2.51a) and (2.51c) into equation (2.54b).<br />
Finally we want to derive the relation between the actually measured NMO velocity, which<br />
strongly depends on the surface curvature and the near-surface velocity gradient, and the NMO<br />
velocity which we would obtain if the measurement surface would be horizontal and without a<br />
near surface velocity gradient. Please note that the term “horizontal” always includes flatness. If<br />
we imagine to have a planar dipping measurement surface, tangent to the real curved measure-
30 <strong>The</strong>ory<br />
ment surface in X 0 and with E 0 = 0, we would measure the NMO velocity<br />
v 2 NMO,P =<br />
2v 0<br />
t 0 K NIP cos 2 β 0<br />
, (2.55)<br />
according to equation (2.54b). Similarly on a fictitious planar and horizontal measurement surface<br />
through X 0 with E 0 = 0 the obtained NMO velocity reads<br />
v 2 NMO,H =<br />
2v0 t0KNIP cos2 β g . (2.56)<br />
0<br />
K NIP would have the same value as K NIP found for the planar but dipping measurement surface<br />
because the dip of the measurement surface is considered within the moveout formula by spec-<br />
ifying the respective take off angle β0 , according to its definition (2.20), relative to the local<br />
coordinate system. For the relation between β0 and β g see Figure 2.6 and equation (2.52). <strong>The</strong><br />
0<br />
NMO velocities that would be measured in this two hypothetical experiments differ, because<br />
they are no pure subsurface attributes like KNIP and KN but depend on the surface on which they<br />
are measured. <strong>The</strong>se two hypothetical experiments lead to the same KNIP that we would obtain<br />
by applying equation (2.48b) to the data measured on the real curved measurement surface with<br />
near velocity gradient. Thus, we can solve equation (2.56) for KNIP and insert into the equation<br />
for vNMO (2.54b) to obtain the relationship between the measured NMO velocity and its corresponding<br />
value vNMO,H that would have been measured on a fictitious horizontal surface through<br />
X0 without near-surface velocity gradient. We find<br />
and vice versa,<br />
v 2 NMO =<br />
t0 v 2 NMO,H =<br />
2v 0<br />
�<br />
2v0 cos2 β0 t0v2 NMO,H cos2 β g − K0 cosβ0 − vo E0 0<br />
2v 2 NMO v 0 cos2 β 0<br />
� , (2.57a)<br />
cos2 β g<br />
0 (v2 NMOK0 cosβ0t0 + v2 NMOv0E0t 0 + 2v . (2.57b)<br />
0 )<br />
v NMO,H can be seen as the NMO velocity that would be obtained after applying a perfect static<br />
correction to pre-stack data that was measured on a surface which is in the vicinity of X 0 , representable<br />
according to equation (2.47), and which has only first order velocity variations near<br />
X 0 .<br />
Zero-dip NMO velocity<br />
For a planar measurement surface, without near-surface velocity gradient, it is possible to separate<br />
the influence of the overburden from the influence of the take off angle β 0 , within equation
2.3 Zero-offset situation 31<br />
(2.54) by introducing the zero-dip NMO velocity,<br />
and rewriting equation (2.54a) in the following way:<br />
v 2 NMO,ZD = 2v0 = v<br />
t0KNIP 2 NMO cos2 β0 , (2.58)<br />
t 2 (hyp,CMP) (h) = t2 0 + 4h2 cos 2 β 0<br />
v 2 NMO,ZD<br />
= t 2 0<br />
(2.59a)<br />
4h2<br />
+<br />
v2 −<br />
NMO,ZD<br />
4h2 sin2 β0 v2 . (2.59b)<br />
NMO,ZD<br />
<strong>The</strong> second h 2 -term, which depends on the reflector dip, can be removed in a velocity independent<br />
way by applying the so-called Gardener Dip Moveout (DMO) procedure described in<br />
(Gardener et al., 1990). Afterwards the zero-dip NMO velocity can be determined on the conventional<br />
way. In the early days of reflection seismics all reflecting layers were assumed to<br />
be horizontal and thus it was the zero-dip NMO velocity, which appeared in the first moveout<br />
formulas.<br />
Solving the left side of equation (2.58) for K NIP results<br />
K NIP =<br />
2v 0<br />
t 0 v 2 NMO,ZD<br />
. (2.60)<br />
Inserting this in equation (2.54b) leads to the useful relation between the zero-dip NMO velocity<br />
which would have been found on a fictitious planar measurement surface through X 0 without<br />
near-surface velocity gradient and the actual NMO velocity found on the real measurement surface<br />
described by K 0 , v 0 and (∇v) 0 . It reads<br />
Vice versa<br />
v 2 NMO =<br />
t0 2v 0<br />
�<br />
2v0 cos2 β0 t0v2 − K0 cosβ0 − v0E0 NMO,ZD<br />
2v 2 NMO v 0 cos2 β 0<br />
� . (2.61a)<br />
v 2 NMO,ZD =<br />
v2 NMOK0 cosβ0t0 + v2 NMOv0E0t 0 + 2v . (2.61b)<br />
0<br />
Note that we would have, in general, a different fictitious planar measurement surface for every<br />
single coincident source and receiver position, whereas only the elevation, not the dip, has an<br />
influence on the zero-dip NMO velocity. <strong>The</strong> same holds for the RMS velocity.
32 <strong>The</strong>ory<br />
RMS velocity<br />
<strong>The</strong> RMS velocity constitutes the average wave velocity, above a certain level, within a medium<br />
composed of parallel and planar homogeneous layers and can be represented as (Dix, 1955)<br />
v 2 RMS,i = v2 1 Δt 1 + v2 2 Δt 2 + ... + v2 i Δt i<br />
t 0,i<br />
i = 1,...,N , (2.62)<br />
where vi is the velocity in layer i, t0,i is the two way traveltime of the central ray reflected at layer<br />
i and Δti = t0,i − t0,i−1 is the two way traveltime within layer i. <strong>The</strong> interval velocity of layer i<br />
can be computed using the formula<br />
�<br />
v2 RMS,i t0,i − v<br />
vi =<br />
2 RMS,i−1 t �1/2 0,i−1<br />
. (2.63)<br />
t 0,i −t 0,i−1<br />
Provided that we have a medium with parallel and planar homogeneous layers, and offsets which<br />
are small compared to depth, we can express approximatively the traveltime within a CMP gather<br />
for a planar measurement surface without near-surface velocity gradient, using the RMS velocity.<br />
This reads<br />
t(h) 2 = t 2 0 + 4cos2 β0 v2 h<br />
RMS<br />
2 . (2.64)<br />
<strong>The</strong> comparison with equations (2.54), using K0 and E0 equal zero, leads to the relation between<br />
vRMS and KNIP , i.e.,<br />
KNIP = 2v0 t0v2 . (2.65)<br />
RMS<br />
It is easy to see that the zero-dip NMO velocity coincides with the (approximative) RMS velocity<br />
determined according to equation (2.64), if the conditions <strong>under</strong> which the latter is defined are<br />
met. In the context of this approximation, the RMS velocity can be seen as a special case of the<br />
more general zero-dip NMO velocity.<br />
Inserting equation (2.65) into equation (2.54b) yields the relation between the actual measured<br />
NMO velocity and the RMS that would be obtained according to equation (2.64) after removing<br />
the inhomogeneity of the top layer and the curvature of the measurement surface. We find<br />
and vice versa,<br />
v 2 NMO =<br />
t0 2v 0<br />
�<br />
2v0 cos2 β0 t0v2 − K0 cosβ0 − v0E0 RMS<br />
2v 2 NMO v 0 cos2 β 0<br />
� , (2.66a)<br />
v 2 RMS =<br />
v2 NMOK0 cosβ0t0 + v2 NMOv0E0t 0 + 2v . (2.66b)<br />
0<br />
Note that these equations can only be used, if the subsurface can be assumed, to consist of planar,<br />
homogeneous and parallel layers. Otherwise it makes no sense to use the RMS velocity. For more<br />
general media one has to replace the RMS velocity with the zero-dip NMO velocity.
2.4 Search-range estimation 33<br />
2.4 Search-range estimation<br />
In the implementation of the CRS stack, it is very important for the efficiency of the coherency<br />
based search that the search ranges of the different search parameters are defined as narrow as<br />
possible. A variable measurement surface curvature, dip, and near-surface velocity gradient lead<br />
here to a considerable complication, because some of the found parameters strongly depend on<br />
these values, namely the NMO velocity, the take-off angle β0 , and the pseudo attributes K∗ N<br />
and K∗ NIP . In order to solve this problem, it is useful to relate the actual search limits to those<br />
limits that would hold if the real measurement surface would have been replaced by a horizontal<br />
measurement surface without near-surface velocity gradient. How to chose the latter search limits<br />
is well known from the conventional 2D CRS stack (see Mann, 2002). To derive these relations,<br />
between the actual search limits and those reference limits is the purpose of this section.<br />
, respectively, are no search<br />
In the current implementations of the CRS stack KNIP and and K∗ NIP<br />
parameters. <strong>The</strong>y are computed from the obtained values of vNMO and β0 , according to equation<br />
(2.54b), whereas K∗ NIP results if K0 and E0 are neglected. Thus, we will not consider KNIP and<br />
K∗ NIP as search parameters in the following, but similar derivations as for KN and K∗ N would hold<br />
in this case, too.<br />
2.4.1 Search range of K N and K ∗ N<br />
If traveltime formula (2.48b) or (2.48a) is used and the (near)surface attributes are considered,<br />
than we search for the real wavefield attribute K N , which has of course the same search range<br />
as in case of a conventional 2D CRS stack with horizontal measurement surface and without<br />
near-surface velocity gradient. But if we use a traveltime formula that does not consider the<br />
(near)surface attributes (or not all of them), we have to transfer the search range that holds for<br />
K N in order to get a appropriate search range for K ∗ N . If we solve equation (2.51b) for K∗ N and<br />
insert at the righthand-side the search limits of K N , we get<br />
K ∗max<br />
N = Kmax N cos2 (β0 ) − K0cos(β0 ) − E0v0 cos2 , (2.67a)<br />
(β0 )<br />
K ∗min<br />
N = Kmin N cos2 (β0 ) − K0cos(β0 ) − E0v0 cos2 . (2.67b)<br />
(β0 )<br />
Please note that the use of these relations demands that β 0 is already determined. In all current<br />
implementations of the 2D ZO CRS stack K N is the last parameter that is searched-for. Otherwise<br />
a procedure similar to the v NMO search-range determination, presented in 2.4.3, would be<br />
necessary.
34 <strong>The</strong>ory<br />
Figure 2.7: This figure shows the search range [−60◦ ,60◦ ] for the take-off angle β0 (black) and<br />
its global counterpart β g (red). In order to search within the same physical limits, the two search<br />
0<br />
ranges have to coincide physically. This can be achieved by subtracting α0 from the search limits<br />
that were used for β0 . Note that |β0 | must not get bigger than 90◦ . To keep the figure simple, the<br />
z-axes are not displayed.<br />
2.4.2 Search range of β 0<br />
For the conventional CRS stack, a planar measurement surface is generally assumed to be perpendicular<br />
to the depth direction and it is not distinguished between β0 and β g.<br />
According to<br />
0<br />
Mann (2002) the search limits of β0 = β g can be defined on the basis of the maximal CMP event<br />
0<br />
dip that should be considered and can depend on t0 .<br />
How this limits for β0 = β g<br />
0 can be related to those limits for β0 that hold in case of a curved measurement<br />
surface with near surface velocity gradient, is shown in Figure 2.7. <strong>The</strong>re the search<br />
limits for β 0 are obtained by subtracting the dip-angle α 0 from the search limits that hold for β g<br />
0 .<br />
Of course, it has to be considered that it makes no sense to allow values for β 0 that are bigger<br />
than 90◦ or smaller than −90◦ . Accordingly, we obtain for the search limits β max<br />
0<br />
2.4.3 Search range of the NMO velocity<br />
and β min<br />
0<br />
β max<br />
�<br />
0 = min β g,max<br />
− α 0 0 , π<br />
�<br />
, (2.68a)<br />
2<br />
β min<br />
�<br />
0 = max β g,min<br />
− α 0 0 , − π<br />
�<br />
. (2.68b)<br />
2<br />
Usually the coefficient of the h 2 -term of the traveltime formula, and thus the NMO velocity, is<br />
the first parameter which is determined during the application of the CRS stack. This search is<br />
performed in the CMP and OD gather, respectively. For the conventional CRS stack a multitude<br />
of experience exists how the search limits of the NMO velocity can be chosen with respect to an<br />
assumed subsurface model. Consequently zero-dip NMO velocity limits (or limits for K NIP ) can
2.4 Search-range estimation 35<br />
easily be derived from the v NMO limits and the β 0 limits by means of equation (2.58). <strong>The</strong>se zerodip<br />
NMO velocity limits that hold for the conventional CRS stack, shall serve in the following<br />
as a general, acquisition independent, basis, from which the v NMO search limits that hold for the<br />
real measurement surface, can be derived. Of course the zero-dip NMO velocity limits can be<br />
replaced by RMS velocity limits, if the conditions <strong>under</strong> which the RMS velocity is defined are<br />
met.<br />
To find a relation between the limits of the actually measured NMO velocity and the zero-dip<br />
NMO velocity range mentioned above, we have to analyze equation (2.61a). <strong>The</strong> first observation<br />
is that one has to know the take off angle β 0 to relate the measured NMO velocities to the<br />
respective zero-dip NMO velocity values. This leads to a severe problem, because, as stated<br />
above, the NMO velocity is the first parameter that is determined and thus the angle β 0 is still<br />
unknown at this stage. Unfortunately it is hardly possible to permute the order in which the<br />
parameters are determined. To find a solution, we have to look at the situation more closely.<br />
Actually we are not searching the limiting values of vNMO , but the limits of the coefficient of the<br />
h2-term of the traveltime, which depends, according to equation (2.54a), on v−2 . If we call the<br />
NMO<br />
inverse of vNMO NMO slowness pNMO , then we have to find the global extrema of the function<br />
p 2 NMO = 2v 0 cos2 β 0 − K 0 t 0 cosβ 0 v 2 NMO,ZD − v 0 t 0 E 0 v2 NMO,ZD<br />
2v 0 v 2 NMO,ZD<br />
, (2.69)<br />
which is the inverse of equation (2.61a). At this point, we have to remember that the inhomogeneity<br />
factor E 0 is no pure (near)surface attribute, but depends on the take-off angle β 0 . If we<br />
substitute E 0 in equation (2.69), according to equation (2.44), we get<br />
p 2 NMO = 2v 0 cos2 β 0 − K 0 t 0 cosβ 0 v 2 NMO,ZD<br />
+<br />
This function of v 2 NMO,ZD and β 0<br />
extrema reads<br />
t 0 sinβ 0<br />
2v 0 v 2 NMO,ZO<br />
(2.70)<br />
�<br />
(1 + cos 2 �<br />
β0 )(∂xv0 ) + cosβ0 sinβ0 (∂zv0 )<br />
. (2.71)<br />
2v 2 0<br />
describes a surface in the 3D space, and the condition for the<br />
∂ p 2 NMO<br />
∂(v 2 NMO,ZD<br />
) = 0 and<br />
∂ p 2 NMO<br />
∂(β 0 )<br />
= 0 . (2.72)<br />
<strong>The</strong> first of the two conditions above, can be evaluated without problems. If we take the derivative<br />
of pNMO with respect to v2 NMO,ZD , we obtain<br />
∂ p2 NMO<br />
∂(v2 NMO,ZD ) = − cos2 β0 v2 . (2.73)<br />
NMO,ZO
36 <strong>The</strong>ory<br />
values and<br />
Thus p2 NMO is a monotone function of v2 NMO,ZD . It decreases for positive v2 NMO,ZD<br />
increases for negative v2 NMO,ZD values. Consequently the extrema lie at the borders of the p2 NMO<br />
surface, at<br />
p 2 NMO = p2 NMO<br />
�<br />
v2 �min NMO,ZD<br />
��v �min<br />
�<br />
2<br />
NMO,ZD ,β0<br />
�<br />
v2 �max NMO,ZD<br />
and p 2 NMO = p2 NMO<br />
��v �max<br />
�<br />
2<br />
NMO,ZD ,β0 , (2.74)<br />
where<br />
and<br />
are the pre-estimated limiting values of the squared<br />
�<br />
zero-dip NMO velocity. Depending on the assumed subsurface model v2 �min NMO,ZD can be<br />
both positive and negative.<br />
<strong>The</strong> first derivative of p 2 NMO with respect to β 0 reads<br />
∂ p 2 NMO<br />
∂β 0<br />
= sinβ 0 (K 0 t 0 v2 NMO,ZD − 4v 0 cosβ 0 )<br />
2v 2 0 v2 NMO,ZD<br />
+ 3t 0 v2 NMO,ZD cos3 β 0 (∂xv) 0 −t 0 v 2 NMO,ZD cosβ 0 (∂xv) 0<br />
2v 2 0 v2 NMO,ZD<br />
+ 3t 0 v2 NMO,ZD cos2 β 0 sinβ 0 (∂zv) 0 −t 0 v 2 NMO,ZD sinβ 0 (∂zv) 0<br />
2v 2 0 v2 NMO,ZD<br />
.<br />
(2.75)<br />
Now the next step would be to set the righthand side of this equation equal zero, and to solve the<br />
resulting equation for β 0 , but this is analytically hardly possible. <strong>The</strong> computer algebra system<br />
MAPLE is not able to find a solution and it is most probably that even if a solution would be<br />
found, it would be difficult to apply, due to a multitude of different cases which would be to consider,<br />
depending on the signs and possibly also on the relative values of K 0 ,v 2 NMO,ZD ,(∂xv) 0 and<br />
(∂zv) 0 . If we assume the near surface velocity to be constant and accordingly E 0 = 0, the extremal<br />
values of p NMO are much easier to find. <strong>The</strong>refore, we will discuss two different strategies, one<br />
for E 0 = 0 and one for E 0 �= 0.<br />
Strategy 1 (E0 = 0): If we set E0 = 0 within equation (2.69), the first derivative of p2 NMO (E0 = 0)<br />
with respect to β0 reads<br />
∂ p 2 NMO<br />
∂β 0<br />
= sinβ 0 (K 0 t 0 v2 NMO,ZD − 4v 0 cosβ 0 )<br />
2v 0 v 2 NMO,ZD<br />
. (2.76)<br />
If we set this equation to be equal zero, solve for β0 , and consider that the extrema have to lie at<br />
surface we find,<br />
the v min,max<br />
NMO,ZD borders of the p2 NMO<br />
β (1)<br />
0<br />
= 0 and cosβ (2)<br />
0 =<br />
� K0 t 0 (v min,max<br />
NMO,ZD )2<br />
4v 0<br />
�<br />
=<br />
K0 2Kmin,max , (2.77)<br />
NIP
2.4 Search-range estimation 37<br />
�<br />
v2 �min,max NMO,ZD<br />
were we have used equation (2.60) to express<br />
It holds<br />
K max 2v0 NIP =<br />
t0 (vmin NMO,ZD )2 and K min 2v0 NIP =<br />
t0 (vmax .<br />
)2<br />
NMO,ZD<br />
(2.78)<br />
in terms of Kmin,max. NIP<br />
At this point, we have to consider the limits of the search range of the take-off angle β 0 , derived<br />
in the last section. <strong>The</strong>se are<br />
−π/2 ≤ β min<br />
0<br />
≤ β 0<br />
≤ β max<br />
0<br />
Due to this limitation, the second solution for β0 , i.e., β (2) , is only valid for,<br />
0<br />
K0 2Kmin,max ≥ min<br />
NIP<br />
� cosβ min<br />
≤ π/2 . (2.79)<br />
0 ,cosβ max<br />
0<br />
� . (2.80)<br />
If this condition is met, we get possible candidates for the global extrema, which read<br />
C (21) = p2 NMO ((v2 NMO,ZD )min ,β (2)<br />
0 ) , C(22) = p2 NMO ((v2 NMO,ZD )max ,β (2)<br />
0<br />
) . (2.81)<br />
Otherwise, if β (2) lies outside of the considered range, possible extrema which correspond to this<br />
0<br />
solution lie at the respective edges of the p2 NMO (v2 NMO,ZD ,β0 ) surface. <strong>The</strong>se are<br />
C (a) = p2 NMO ((v2 NMO,ZD )min ,β min<br />
0 ) , C (b) = p2 NMO ((v2 NMO,ZD )min ,β max<br />
0 ) , and<br />
C (c) = p2 NMO ((v2 NMO,ZD )max ,β min<br />
0 ) , C (d) = p2 NMO ((v2 NMO,ZD )max ,β max<br />
0 ) ,<br />
respectively.<br />
<strong>The</strong> first solution β (1) is always valid and yields<br />
0<br />
C (11) = p2 NMO ((v2 NMO,ZD )min ,β (1)<br />
0 ) , C(12) = p2 NMO ((v2 NMO,ZD )max ,β (1)<br />
0<br />
(2.82)<br />
) . (2.83)<br />
To make things clearer, will discuss an example of how, according to our derivations, the maximum<br />
and minimum values of the NMO velocity are obtained.<br />
If, e.g.,<br />
K0 2Kmin ≥ min<br />
NIP<br />
� cosβ min<br />
0 ,cosβ max<br />
�<br />
0 , and<br />
K0 2Kmax < min<br />
NIP<br />
� cosβ min<br />
0 ,cosβ max<br />
�<br />
0 , (2.84)<br />
then we have the possible extrema {C (11) ,C (12) ,C (a) ,C (b) ,C (22) }.<br />
<strong>The</strong> extremum C (21) is not valid, because the value of β (2)<br />
0<br />
outside the considered range. Consequently C (21) has to be replaced by C (a) and C (b) .<br />
, which results for KNIP = Kmax NIP is
38 <strong>The</strong>ory<br />
In our example the limiting values for v 2 NMO<br />
� �max 2<br />
vNMO =<br />
� �<br />
2 min<br />
vNMO =<br />
1<br />
� �<br />
p2 min =<br />
NMO<br />
1<br />
� �<br />
p2 max =<br />
NMO<br />
would be<br />
Of course an analogous procedure is valid in any other case, too.<br />
1<br />
min{C (1) ,C (2) ,C (a) ,C (b) ,C (4) and (2.85)<br />
}<br />
1<br />
max{C (1) ,C (2) ,C (a) ,C (b) ,C (4) . (2.86)<br />
}<br />
For the sake of simplicity and brevity I abstained from completely analyzing the second derivatives<br />
of p2 NMO in order to determine in which case an extremum is a maximum or minimum.<br />
Doing this, one has to consider many different cases depending on the sign of K0 , Kmin,max and<br />
NIP<br />
(K0 − 2Kmin,max), and in view of the practical application it is faster and much more convenient<br />
NIP<br />
to perform the strategy stated above, which is to compute first all possible candidates for the<br />
extrema and to decide afterwards which their maximum and minimum value is.<br />
Note, as mentioned before, all above derivations hold also for the RMS velocity, which is nothing<br />
more than the zero-dip NMO velocity for a special subsurface structure. Thus, if the subsurface<br />
can be assumed to be constituted of homogeneous layers, divided by planar and parallel reflectors,<br />
RMS velocity limits can be used instead of zero-dip NMO velocity limits, simply by<br />
substituting vNMO,ZD by vRMS in the equations above.<br />
Strategy 2 (E 0 �= 0): If the near-surface velocity gradient can not be assumed to be zero, it is<br />
hardly possible to derive analytically the extremal values of the NMO velocity. In this case it is<br />
proposed to determine the v NMO search limits, assuming E 0 = 0 and to enlarge the search range<br />
subsequently by a certain value that accounts for the unknown influence of E 0 . In general, the<br />
influence of E 0 should be small, compared to the influence of K 0 , so that the search range has not<br />
to be extended considerably. To study this point more closely lies unfortunately outside the time<br />
frame of this thesis.<br />
2.5 Redatuming<br />
<strong>The</strong> final goal of the CRS stack is to provide a time domain image of the subsurface structure,<br />
i.e., the ZO section, and the also very important attribute sections, which are the β 0 section, the<br />
K N section, the K NIP section, and the v NMO section. It is evident that due to the fact that we<br />
are only interested in the subsurface structure, we do not want these sections to depend on the<br />
characteristics of the measurement surface and the potentially inhomogeneous top layer. If the<br />
acquisition topography meets the required conditions, then the traveltime equations (2.48) enable<br />
us to determine the (near)surface attribute independent wavefield attributes K N and K NIP and the<br />
only surface dip dependent take-off angle β 0 . According to Figure 2.6 and equation (2.52), β 0 can
2.5 Redatuming 39<br />
Figure 2.8: To remove the influence of the acquisition surface topography from the obtained ZO<br />
and attribute sections, we simulate a situation where all central rays end on the same horizontal<br />
measurement surface, e.g., the common-datum surface. This is shown exemplarily, for one central<br />
ray. In order to keep the figure simple, only the situation (∇v) 0 = 0 is displayed, were we<br />
have no refraction at the measurement surface, because we can choose v f = v 0 .
40 <strong>The</strong>ory<br />
easily be transferred to the surface- dip independent take-off angle β g,<br />
which is measured with<br />
0<br />
respect to the horizontal. Also, the NMO velocity values should be corrected to those values that<br />
would be obtained on a horizontal measurement surface without near-surface velocity gradient.<br />
This can be done by means of equation (2.57b). Alternatively, it is possible to compute the<br />
NMO velocity measured on a fictitious surface through X0 with arbitrary (near)surface attributes,<br />
from the value of KNIP . This is done by inserting KNIP into equation (2.54b), together with the<br />
(near)surface attributes of the fictitious measurement surface.<br />
<strong>The</strong> application of these corrections to the determined wavefield attributes, leads to attribute<br />
sections, which represent those values of the respective attribute which refer to a horizontal<br />
measurement surface, without near-surface velocity gradient. However, it has to be taken into<br />
account that, in general, the elevation of these fictitious horizontal measurement surfaces is different<br />
for different central points X0 . This means for the attribute sections Ai (x0 ,t0 ) and also<br />
for the ZO section AZO (x0 ,t0 ) that in general every x0 is related to a different elevation z0 . <strong>The</strong><br />
consequence is that, e.g., for the case of a measurement surface with a sinusoidal shape and a<br />
subsurface constituted of homogeneous layers separated by horizontal reflectors, we would find<br />
sinusoidal images of the reflectors in the ZO section. <strong>The</strong> same observation would hold for the<br />
attribute sections. This can be seen very well in the synthetic data example, presented in Section<br />
3.5. To remove this undesired topography effect from the ZO and attribute sections, we introduce<br />
a fictitious horizontal measurement surface to which all attributes and traveltimes are related. In<br />
other words, we simulate a situation in which all central rays start and end at the same horizonal<br />
measurement surface. Thus, we have to transfer the zero-offset traveltimes and also the attributes<br />
to those values, which would be measured on this common-datum surface. Such a procedure is<br />
called redatuming. In the following, all values that pertain to this virtual measurement surface<br />
are denoted with a prime. <strong>The</strong> key information for this procedure is the knowledge of the takeoff<br />
angle β0 , which is provided by the CRS stack. If the common-datum surface is assumed to<br />
be above the actual topography, it is possible to choose an arbitrary velocity v f for the fictitious<br />
layer between topography and new datum. If the near-surface velocity gradient is zero, then the<br />
most convenient choice is to set v f equal v0 , because this avoids that the topography has to be<br />
considered as a additional reflector. For (∇v) 0 �= 0 Snell’s Law has to be considered to derive<br />
β ′ 0 , the fictitious take-off angle at the fictitious coincident source and receiver point X ′ 0 . Knowing<br />
the take-off angles and the wave velocity within the fictitious layer, it is not difficult to forward<br />
propagate the Normal- and NIP-wave fronts up to the common-datum surface.<br />
2.5.1 Mapping of X 0 and t 0 to the common-datum surface<br />
To map the coincident source and receiver point X0 of a central ray from the original measurement<br />
surface to its corresponding location X ′ 0 at the common-datum surface, one has to transfer its
2.5 Redatuming 41<br />
coordinates x0 and z0 to their new values x ′ 0 and z′ 0 . Of course, z′ 0 is given by the elevation of the<br />
common-datum surface. To transfer x0 we have to know the emergence angle of the central ray<br />
after being refracted at the measurement surface. We will denote this angle as β f . According to<br />
0<br />
Snell’s Law, we have the relation,<br />
v0 =<br />
v f<br />
sinβ0 , (2.87)<br />
sinβ f<br />
which leads to<br />
β f<br />
0<br />
� �<br />
v f<br />
= arcsin sinβ<br />
v 0 . (2.88)<br />
0<br />
In analogy to equation (2.52), the take-off angle measured at the common-datum surface β ′ 0 is<br />
given by<br />
β ′ f<br />
0 = β0 + α0 . (2.89)<br />
If we denote the vertical distance between X0 and the common-datum surface as Δz, we can<br />
transfer now x0 to x ′ 0 . This reads<br />
x ′ 0 = x0 + Δztanβ ′ 0 . (2.90)<br />
For the new two-way traveltime of the central ray t ′ 0 , we get, after simple trigonometric considerations,<br />
t ′ 0 = t0 + 2Δz<br />
v f cosβ ′ . (2.91)<br />
0<br />
2.5.2 Mapping of K N and K NIP to the common-datum surface<br />
In order to transfer the values of the wavefield attributes K N ,K NIP to those values which would<br />
be measured at the common-datum surface we have to use the refraction law (Hubral and Krey,<br />
1980) that gives us the curvature of the N- and NIP-wave, respectively, after passing the mea-<br />
surement surface. It results<br />
K f<br />
N,NIP = KN,NIPv f cos2 β0 v0 cos2 β f +<br />
0<br />
K0 cos2 β f<br />
0<br />
� v f<br />
vo<br />
cosβ0 − cosβ f<br />
�<br />
0<br />
, (2.92)<br />
where K f are the refracted wavefront curvatures on the upper side of the measurement sur-<br />
N,NIP<br />
face. Subsequently, we use the transmission law to propagate the wavefronts of the N- and NIPwave<br />
through the fictitious layer above the real measurement surface, up to the common-datum<br />
surface.<br />
1<br />
K ′ �<br />
=<br />
N,NIP<br />
1<br />
K f<br />
+<br />
N,NIP<br />
1<br />
2 v f t �<br />
f , (2.93)<br />
with the two-way traveltime within the fictitious layer<br />
t f = t ′ 0 −t 0<br />
2Δz<br />
=<br />
v f cosβ ′ . (2.94)<br />
0
42 <strong>The</strong>ory<br />
Inserting equations (2.92) and (2.94) into (2.93) leads to the final equations for the wavefront<br />
curvatures K ′ N and K′ NIP measured at the common-datum surface<br />
1<br />
K ′ N (x′ 0 ,t′ ⎛<br />
= ⎝<br />
o )<br />
cos2 β f<br />
�0 v f<br />
vo cosβ � +<br />
0 − cosβ f<br />
0<br />
Δz<br />
cosβ ′ ⎞<br />
⎠ ,<br />
0<br />
(2.95a)<br />
1<br />
K ′ NIP (x′ 0 ,t′ ⎛<br />
= ⎝<br />
o )<br />
KN (x0 ,t0 ) v f<br />
v cos<br />
0<br />
2 β0 + K0 cos2 β f<br />
0<br />
KNIP (x0 ,t0 ) v f<br />
v cos<br />
0<br />
2 β0 + K0 � v f<br />
vo cosβ 0<br />
� +<br />
− cosβ f<br />
0<br />
Δz<br />
2.5.3 Mapping of v NMO to the common-datum surface<br />
cosβ ′ 0<br />
⎞<br />
⎠ . (2.95b)<br />
In Section 2.3.2 we have derived equation (2.57b) to transfer the NMO velocity measured at the<br />
real curved measurement surface, with near-surface velocity gradient to that NMO velocity which<br />
would be measured at a corresponding horizontal measurement surface without near surface<br />
velocity gradient, i.e., vNMO,H . Now we want to transfer vNMO,H to v ′ NMO , which is the NMO<br />
velocity which would be found at the common-datum surface. To do this, we assume an auxiliary<br />
subsurface model, for which the RMS velocity is defined. It holds<br />
v RMS = v NMO,H cosβ g<br />
0<br />
, (2.96)<br />
according to Section 2.3.2. This is allowed because it does not change anything on the actual<br />
measured NMO velocity, whether the real model or the auxiliary model is considered. According<br />
to equation (2.62) we would obtain at the common-datum surface<br />
v ′ RMS =<br />
�<br />
�<br />
�<br />
� t0v2 RMS +t f v2 f<br />
. (2.97)<br />
t0 +t f<br />
<strong>The</strong> corresponding NMO velocity, v ′ NMO , at the common-datum surface is<br />
v ′ NMO = v′ RMS<br />
cosβ ′ . (2.98)<br />
0<br />
By inserting the equations (2.88), (2.97), (2.96) and (2.94) into the equation above we get as<br />
final result the relation between the actually measured NMO velocity and the NMO velocity that<br />
would be measured at the common-datum surface,<br />
v ′ NMO =<br />
cos(β0 + α0 )<br />
�<br />
cos arcsin<br />
�<br />
v2 t<br />
NMO,H 0v f cosβ ′ 0 +2Δz v f<br />
t0v f cosβ ′ 0 +2Δz<br />
� �<br />
v f<br />
v 0<br />
sinβ 0<br />
+ α 0<br />
� , (2.99a)
2.6 Global coordinate system 43<br />
with<br />
v 2 NMO,H =<br />
according to equation (2.57b).<br />
2v 2 NMO v 0 cos2 β 0<br />
cos2 (β0 + α0 )(v2 NMOK0 cosβ0t0 + v2 NMOv0E0t 0 + 2v , (2.99b)<br />
0 )<br />
Of course it is also possible to derive v ′ NMO from K′ NIP and β ′ 0<br />
in this case<br />
v ′ NMO =<br />
�<br />
2.6 Global coordinate system<br />
2v f<br />
t ′ 0 K′ NIP cos2 β ′ 0<br />
using equation (2.56), which reads<br />
. (2.100)<br />
Figure 2.9: <strong>The</strong> transformation of the local 1D midpoint and half-offset coordinates h and m to<br />
the respective global 1D coordinates hg and mg.<br />
<strong>The</strong> traveltime formulas that we have derived up to this point assume a Cartesian coordinate system<br />
with x-axis tangent to the measurement surface and origin in X 0 (see Figure 2.6). This is the<br />
coordinate system, where, according to (2.39), half-offset h and midpoint m are defined as 1D<br />
coordinates measured along the x-axis. But for practical application, it is very inconvenient to<br />
transfer the measured source and receiver coordinates into the specific local coordinate system<br />
of every central point, in order to determine the required offset and midpoint coordinates. Particularly,<br />
if one wants to use a conventional CRS stack software and correct the found pseudo<br />
attributes afterwards, this is not possible. To solve this problem caused by the curvature of the<br />
measurement surface is the aim of this section.
44 <strong>The</strong>ory<br />
Let us first look at the much simpler case of having a planar measurement surface (see, with a<br />
different notation, Höcht, 1998). Here it is easy to extend our traveltime formulas, which were<br />
derived for a single central point, to hold for all considered central points at the same time. To do<br />
this we define a global coordinate system, equal oriented as the local ones, whereas the location<br />
of the origin is arbitrary. <strong>The</strong>n we substitute the local midpoint coordinate m by mg − x 0 ; mg<br />
is the location of the midpoint in the global coordinate system and x 0 is the x-coordinate of the<br />
central point X 0 that is considered. <strong>The</strong> local half-offset coordinate h is substituted by hg. It<br />
results<br />
tpar(mg,hg) = t0 + 2 sinβ0 (mg − x<br />
v 0 ) +<br />
0<br />
1 �<br />
KN cos<br />
v0 2 �<br />
β0 (mg − x0 ) 2<br />
(2.101a)<br />
+ 1 �<br />
KNIP cos<br />
v0 2 �<br />
2<br />
β0 hg , and<br />
t 2 hyp (mg,hg)<br />
�<br />
= t0 + 2 sinβ �2 0<br />
(mg − x<br />
v 0 ) +<br />
0<br />
2 t �<br />
0<br />
KN cos<br />
v0 2 �<br />
β0 (mg − x0 ) 2 (2.101b)<br />
+ 2 t �<br />
0<br />
KNIP cos<br />
v0 2 �<br />
2<br />
β0 hg .<br />
It is evident that this approach can not be directly applied to the case of a curved measurement<br />
surface, because here the local coordinate systems are in general different oriented. This case<br />
requires an additional coordinate rotation to transfer m and h to their global counterparts hg and<br />
mg. According to Figure 2.9 we find the transformation<br />
h = 1<br />
hg and m =<br />
cosα0 1<br />
(mg − x<br />
cosα 0 ) . (2.102)<br />
0<br />
If we apply this coordinate transformation to the general set of traveltime equations (2.48), we<br />
get a new one which does now explicitly depend on the dip angle α 0 of the measurement surface<br />
and the global x-coordinate x 0 of the central point X 0 . This reads<br />
tpar(mg,hg) = t 0 + 2 sinβ 0<br />
t 2 hyp (mg,hg) =<br />
+<br />
+<br />
v 0 cosα 0<br />
1<br />
v 0 cos 2 α 0<br />
1<br />
v 0 cos 2 α 0<br />
�<br />
t 0 + 2 sinβ 0<br />
v 0 cosα 0<br />
+ 2 t 0<br />
v 0 cos 2 α 0<br />
+ 2 t 0<br />
v 0 cos 2 α 0<br />
(mg − x 0 ) (2.103a)<br />
�<br />
KN cos 2 �<br />
β0 − K0 cosβ0 − v0 E0 (mg − x0 ) 2<br />
�<br />
KNIP cos 2 �<br />
2<br />
β0 − K0 cosβ0 − vo E0 hg , and<br />
�2 (mg − x0 )<br />
�<br />
KN cos 2 �<br />
β0 − K0 cosβ0 − v0 E0 (mg − x0 ) 2<br />
�<br />
KNIP cos 2 �<br />
2<br />
β0 − K0 cosβ0 − v0 E0 hg .<br />
(2.103b)
2.6 Global coordinate system 45<br />
From these two equations one can deduce eight different subsets, by setting one, two, or all<br />
of the (near)surface attributes {K 0 ,α 0 ,(∇v) 0 } equal zero. Note that in the equations above,<br />
β0 is still related to the local coordinate system, but it can be easily substituted by its global<br />
according to equation (2.52), if this is desired.<br />
counterpart β g<br />
0<br />
Equal to Section 2.3.1 we can relate the pseudo wavefield attributes {β ∗ 0 ,K∗ NIP ,K∗ N }, to their real<br />
values {β0 ,KNIP ,KN } by comparing the coefficients of equations (2.103) and equations (2.50).<br />
We can use either the hyperbolic or the parabolic equations. This results the following relations<br />
∗<br />
sinβ0 = cosα0 sinβ0 , (2.104a)<br />
KN = K∗ N cos2 α0 cos2 β ∗ 0 + K0 �<br />
1 − cos2 α0 sin2 β ∗ 0 + v0E0 , and (2.104b)<br />
K NIP = K∗ NIP cos2 α 0 cos 2 β ∗ 0 + K 0<br />
1 − cos 2 α 0 sin 2 β ∗ 0<br />
�<br />
1 − cos 2 α 0 sin 2 β ∗ 0 + v 0 E 0<br />
1 − cos 2 α 0 sin 2 β ∗ 0<br />
. (2.104c)<br />
Please note that β ∗ 0 becomes complex if | sinβ0 | > 1. Of course, the current implementations of<br />
cosα0 the conventional 2D ZO CRS stack assume K0 , E0 and α0 to be zero, because they were designed<br />
for a planar measurement surface and 1D source and receiver coordinates, measured along these<br />
surface. Consequently, complex values of β ∗ 0 have to be taken into account, if a conventional<br />
implementation is applied to data measured on a curved surface. In other words, they have to be<br />
considered within the search range of β ∗ 0 . Unfortunately, this is generally not possible without<br />
changing the code, since this case is not intended, as the software was designed for a planar<br />
measurement surface only.<br />
To avoid confusion we will discuss a little bit closer, how to use the relations above. E.g., if one<br />
wants to compute the real wavefield attributes {β 0 ,K NIP ,K N } from the pseudo attributes obtained<br />
by using a traveltime formula that considers K 0 , but not E 0 and α 0 , one has to set K 0 = 0 in the<br />
equations above and insert the pseudo wavefield attributes together with the actual values of E 0<br />
and α 0 into the righthand sides. At this point, it must be stressed again that α 0 is implicitly<br />
considered within the traveltime formulas that use local midpoint and offset coordinates. Not<br />
considering α 0 means, e.g., using a conventional CRS stack software that does not transfer the<br />
source and receiver coordinates into the respective local coordinate systems, but uses global<br />
midpoint and offset coordinates.<br />
<strong>The</strong> validity of the relations above can be demonstrated by setting K 0 = 0, E 0 = 0 and α 0 = 0.<br />
Doing this pseudo and real attributes are equal as it should be the case for a traveltime formula<br />
which considers all (near)surface attributes.
46 <strong>The</strong>ory<br />
2.6.1 <strong>The</strong> inhomogeneity factor E 0 in global coordinates<br />
According to the definition of the inhomogeneity factors E S , E G (2.34), and E 0 (2.44), the velocity<br />
gradient has to be measured within the respective local coordinate system. One can imagine<br />
that in practical application, it may often be more convenient to determine the velocity gradient<br />
within the global coordinate system. To transpose the velocity gradient from local to global coordinates<br />
one has to perform a rotation of the coordinate system by the dip-angle α 0 , which leads<br />
to the relation,<br />
� �<br />
cosα0 sinα<br />
(∇v) 0 =<br />
0 (∇v) g<br />
. (2.105)<br />
0<br />
−sinα 0 cosα 0<br />
If we insert this relation into (2.44), we obtain the inhomogeneity factor E 0 in global coordinates,<br />
E 0 = − sinβ 0<br />
v 2 0<br />
+ cosβ 0 sinβ 0<br />
�<br />
�1 �<br />
2<br />
+ cos β0<br />
� �<br />
∂v<br />
cosα0 � �<br />
∂v<br />
−sinα0 ∂xg<br />
�<br />
∂xg<br />
� � � �<br />
∂v<br />
+ sinα0 0<br />
∂zg 0<br />
� � ��<br />
∂v<br />
.<br />
+ cosα0 0<br />
2.6.2 <strong>The</strong> NMO and RMS velocities in global coordinates<br />
∂zg<br />
0<br />
(2.106)<br />
In analogy to Section 2.3.2, the following definition holds for the (hyperbolic) NMO velocity in<br />
global coordinates.<br />
t 2 (hyp,OD) (hg) = t 2 0 + 4h2g v2 ,with (2.107a)<br />
NMO<br />
(v g<br />
NMO )2 =<br />
2v 0 cos 2 α 0<br />
�<br />
t0 KNIP cos2 � . (2.107b)<br />
β0 − K0 cosβ0 − v0 E0 Comparing equations (2.107) and (2.54) one can easily see, that the NMO velocity measured in<br />
global coordinates differs from its counterpart, obtained in local coordinates. We find the relation<br />
(v g<br />
NMO )2 = cos 2 α0v 2 NMO . (2.108)<br />
Inserting now equation (2.60), which does not depend on the choice of the coordinate system,<br />
leads to the useful relation between the found NMO velocity and the zero-dip NMO velocity<br />
which would be found on a fictitious planar measurement surface without near-surface velocity<br />
gradient.<br />
(v g<br />
NMO )2 =<br />
t0 2v0 cos2 α<br />
� 0<br />
2v0 cos2 β0 t0v2 − K0 cosβ0 − v0E0 NMO,ZD<br />
� , (2.109a)
2.6 Global coordinate system 47<br />
and vice versa:<br />
v 2 NMO,ZD =<br />
2(v g<br />
NMO )2 v 0 cos 2 β 0<br />
((v g<br />
NMO )2 K 0 cosβ 0 t 0 + (v g<br />
NMO )2 v 0 E 0 t 0 + 2v 0 cos 2 α 0 )<br />
. (2.109b)<br />
To obtain the respective relations for a case in which the RMS velocity is defined one has only to<br />
set vRMS = vNMO,ZD in the equations above. Equation (2.109b) provides the possibility to derive<br />
the zero-dip NMO or the RMS velocities from the vg values that are obtained by fitting the<br />
NMO<br />
respective traveltime surface to the pre-stack data. <strong>The</strong> opposite direction, equation (2.109a) is<br />
very useful to estimate the search limits of v g<br />
NMO by means of the estimated range of v RMS and<br />
v NMO,ZD , respectively, according to Section 2.4.3.<br />
A similar derivation to that in Section 2.3.2, now for global coordinates, leads to the relationship<br />
between the measured NMO velocity and its corresponding value measured on a fictitious planar<br />
and horizontal surface without near velocity gradient. It reads<br />
Vice versa, we get<br />
v 2 NMO,H =<br />
(v g<br />
NMO )2 =<br />
t0 2v0 cos2 α<br />
� 0<br />
2v0 cos2 β0 t0v2 NMO,H cos2 β g − K0 cosβ0 − vo E0 0<br />
2(v g<br />
NMO )2 v 0 cos 2 β 0<br />
cos 2 β g<br />
0 ((vg<br />
NMO )2 K 0 cosβ 0 t 0 + (v g<br />
NMO )2 v 0 E 0 t 0 + 2v 0 cos 2 α 0 )<br />
� . (2.110a)<br />
. (2.110b)<br />
Note that v NMO,H has the same value whether obtained in global or in local coordinates because<br />
the two systems coincide in case of a horizontal measurement surface. <strong>The</strong> same holds for the<br />
values of v NMO,ZD and v RMS , which do not depend on the dip of the local coordinate system and<br />
thus can be seen as measured in a horizontal 1D coordinate system that is equal to the global one.<br />
2.6.3 <strong>The</strong> search range of K∗ N , β ∗ 0 , and vNMO in global coordinates<br />
Of course the search range of the Normal-wave curvature KN remains the same, but in case of<br />
the limits of the pseudo attribute K∗ N one has to use slightly modified formulas that account for<br />
the additional option not to consider α0 . <strong>The</strong>se are given by<br />
= Kmax<br />
�<br />
N 1 − cos2 α0 sin2 β ∗ � �<br />
0 − K0 1 − cos2 α0 sin2 β ∗ 0 − v0E0 , (2.111a)<br />
and<br />
K ∗max<br />
N<br />
K ∗min<br />
N<br />
cos 2 α 0 cos 2 β ∗ 0<br />
�<br />
Kmin N 1 − cos2 α0 sin<br />
= 2 β ∗ � �<br />
0 − K0 1 − cos2 α0 sin 2 β ∗ 0 − v0E0 cos2 α0 cos2 β ∗ , (2.111b)<br />
0
48 <strong>The</strong>ory<br />
which result from solving equation (2.104b) for K ∗ N<br />
and inserting the respective limiting values.<br />
<strong>The</strong> found take-off angle β 0 has the same value in both coordinate systems. Consequently the<br />
strategy to obtain the search range for β 0 does not change. But the pseudo take off angle β ∗ 0<br />
that one obtains using global coordinates and neglecting the surface dip α 0 in equation (2.103c)<br />
differs from β 0 . In this case we have to choose other search limits. Combining equations (2.68)<br />
and equation (2.104a) yields<br />
β ∗max<br />
0 =<br />
⎛<br />
arcsin⎝<br />
sin<br />
� �<br />
min β g,max − α<br />
0 0 , π β<br />
��⎞<br />
2<br />
⎠ ,<br />
cosα0 (2.112a)<br />
∗min<br />
0 =<br />
⎛<br />
arcsin⎝<br />
sin<br />
� �<br />
max β g,min − α<br />
0 0 , − π ��⎞<br />
2<br />
⎠ .<br />
cosα0 (2.112b)<br />
As mentioned before, β ∗ 0 has, <strong>under</strong> certain circumstances, complex values. Consequently, in<br />
any case were complex β ∗ 0 are possible, the respective limiting values have to be complex, too.<br />
Such a situation is always given if the argument of the arcsin in the equations above is larger than<br />
one.<br />
In global coordinates the squared NMO slowness reads<br />
(p g<br />
NMO )2 = 2v0 cos2 β0 − K0t0 cosβ0 v2 NMO,ZD − v0t0E0v2 NMO,ZD<br />
2v0 cos2 α0v2 , (2.113)<br />
NMO,ZO<br />
according to the inverse of equation (2.109a).<br />
According to equation (2.108) holds for the relation between (p g<br />
NMO )2 and the squared NMO<br />
slowness in local coordinates p 2 NMO<br />
From this equation follows that (p g<br />
(p g<br />
NMO )2 = p2 NMO<br />
cos2 , (2.114)<br />
α0 have their extrema at the same locations,<br />
NMO )2 and p2 NMO<br />
and their extremal values differ only by the factor cos2 α0 . Thus the whole strategy to determine<br />
the NMO velocity search range remains valid even for global coordinates. Only the values at the<br />
extremal points have to be computed using equation (2.113) instead of equation (2.69).<br />
2.6.4 Redatuming in global coordinates<br />
<strong>The</strong> ZO traveltimes and the wavefield attributes {K N ,K NIP ,β 0 } are certainly the same, whether<br />
obtained in global or local coordinates. Only the NMO velocities which are measured at the
2.6 Global coordinate system 49<br />
is equal to vg , be-<br />
NMO,H<br />
cause for a horizontal measurement surface, the global and the local coordinate systems coincide.<br />
Thus all derivations in Section 2.5 can be used for global coordinates, too, with one exception.<br />
In equation (2.99a), equation (2.110b) has to be used to relate vNMO,H to the actually measured<br />
NMO velocity.<br />
curved measurement surface differ (vNMO �= vg NMO ). However, vNMO,H
50 <strong>The</strong>ory
Chapter 3<br />
Synthetic Data Example<br />
<strong>The</strong> aim of this chapter is to study the application of the 2D ZO CRS stack to data measured on<br />
a curved measurement surface by means of a synthetic data example. To do this I have used a<br />
synthetic dataset which was created by ENI (Agip) and that was relinquished to Pedro Chira and<br />
me to test his extended CRS traveltime formula (2.49c), valid for a smooth curved measurement<br />
surface (Chira and Hubral, 2001; Chira et al., 2001). In addition to the dataset, we received results<br />
which ENI obtained via standard data processing, using static corrections and the conventional<br />
CRS stack software. This approach and its results are discussed at the beginning of this chapter.<br />
3.1 Model<br />
<strong>The</strong> synthetic seismic dataset used in this chapter is based on the depth model shown in Figure<br />
3.1. This model consists of four homogeneous layers separated by three horizontal reflectors, and<br />
lying <strong>under</strong>neath a curved acquisition surface. Each layer has a different P-wave velocity. <strong>The</strong><br />
values are specified in the figure. <strong>The</strong> model does not include a near surface velocity gradient,<br />
because it was designed to study the influence of the acquisition surface topography, only. Due<br />
to the different scale of the axes, the topography looks steeper than it is, but anyway the changes<br />
in height can be compared to the changes that can be found, e.g., at the border of the Apennine<br />
Mountains. At the surface between x=10.0 km and x=15.0 km a small-scale undulation of the<br />
topography can be observed. <strong>The</strong> term small-scale refers to an approximate stacking aperture<br />
of 2 km, measured in the global coordinate system. This part of the measurement surface does<br />
not meet the assumptions, required for the validity of the CRS traveltime formulas, derived in<br />
Section 2.3. For central points that lie in this area one can hardly approximate the surface within<br />
the stacking aperture by a parabola. Of course the CRS traveltime formula is still the best second-
52 Synthetic Data Example<br />
Figure 3.1: Model, consisting of four homogeneous layers with different P-wave velocities. <strong>The</strong>se<br />
are separated by three horizontal reflectors and lie <strong>under</strong>neath a curved measurement surface.
3.2 Standard processing using elevation-statics 53<br />
order approximation of the actually measured traveltime response, but the obtained wavefield<br />
attributes K N ,K NIP and β 0 have lost their expected physical meaning to a certain extent. This<br />
area provides a good example to study the validity limits of the CRS stack procedure presented<br />
within this thesis (for results see Section 3.5).<br />
3.2 Standard processing using elevation-statics<br />
A standard method to apply the conventional CRS stack, which uses traveltime formula (2.101),<br />
also to seismic data that was recorded on a curved acquisition surface is to subtract so-called<br />
static corrections from the measured traveltimes with the aim to simulate data that corresponds<br />
to a planar measurement surface. In this case ENI approximatively related all traveltimes to<br />
sources and receivers located at the sea level (see Figure 3.1). Basically, this downward continuation<br />
of the measured traveltimes requires at least the knowledge of two parameters: <strong>The</strong> take-off<br />
angle β 0 of every central ray and the velocity of the medium between the surface and the desired<br />
virtual measurement level. <strong>The</strong> upward continuation to a common level is also possible and requires<br />
besides β 0 only the knowledge of the near-surface velocity v 0 , because the velocity within<br />
the fictitious layer above the real measurement surface can be chosen arbitrarily. Nevertheless,<br />
the downward continuation is very common in oil exploration because in addition to the pure<br />
topography influence also the effect of a slow laterally inhomogeneous top-layer can be partly<br />
compensated.<br />
To determine the necessary near-surface velocity values it is usual practice to drill a sufficient<br />
number of shallow holes along the seismic line or to use informations from the shot holes. <strong>The</strong><br />
emergence angles of the upcoming rays are normally not measured and thus unknown in real<br />
pre-stack data. However, the rays can be assumed to emerge vertical, if the velocity altogether<br />
increases strongly with depth, the considered reflectors lie deep compared to the stacking aperture,<br />
and are not too steep. Another point which justifies this assumption is that in many cases<br />
the velocity difference between the top-layer and the layer below is particularly high and they are<br />
divided by a more or less horizontal interface which causes the upcoming rays to change their<br />
direction towards the vertical. An approximatively vertical emergence of the rays at the surface<br />
justifies the so-called surface-consistency assumption which assumes the static correction time<br />
to depend only on the source and receiver location and not on the angle of emergence. Using<br />
this assumption, the traveltimes t(m,h) can be related to their corresponding values t ∗ (m,h),<br />
“measured” at the sea level, by the equation<br />
t ∗ (m,h) ≈ t(m,h) − Δt stat with Δt stat = (e S + e G )/v 0 , (3.1)<br />
where Δt stat is the static correction which has to be subtracted from the considered trace and e S<br />
and e G are the elevations of the source and the receiver above the new datum. This kind of static
54 Synthetic Data Example<br />
Z g<br />
S<br />
measurement surface<br />
S* S’ G’ G*<br />
sea level = new datum<br />
reflector<br />
Figure 3.2:<br />
Visualization of the applied static corrections. <strong>The</strong> figure shows two homogeneous layers separated<br />
by a planar dipping reflector. Please note that distance and traveltime are displayed<br />
simultaneously assuming units in which the velocity has the value one. To simulate a planar<br />
measurement surface at sea level, ENI subtracted from every trace a correction time, according<br />
to equation (3.1), which is displayed here as the distance (SS ∗ + GG ∗ ). <strong>The</strong> figure shows that<br />
this correction is, due to the non-vertical ray-path too small (blue line segment) to obtain the<br />
real traveltime t ′ (m ′ ,h ′ ) from S ′ to G ′ , whereas S ′ and G ′ are the fictitious source and receiver<br />
locations at the new datum that pertain to the ray that joins S and G. But on the other hand this<br />
correction is too big (red line segment) to yield the searched-for traveltime t ∗ (m,h) from S ∗ to<br />
G ∗ .<br />
G<br />
Xg
3.2 Standard processing using elevation-statics 55<br />
corrections are usually called surface-consistent elevation-statics.<br />
On the first sight, it seems that this correction produces traveltimes which are to large for nonzero<br />
offset, because the wave propagates obliquely through the top-layer and not vertical like the<br />
correction assumes. However, Figure 3.2 shows that it is also necessary to consider the consequence<br />
of keeping offset and midpoint fixed, which is implicitly included within the assumption<br />
of vertical emergence and equation (3.1). Accordingly, subtracting the static correction Δt stat<br />
results for non-zero offset too small instead of too large traveltimes. Thus, this kind of static correction<br />
tends to pretend NMO velocities that are higher than those NMO velocities which would<br />
really be measured at the new datum.<br />
To avoid negative traveltimes, caused by the static correction, a bulk-shift of 1400 m/s was applied<br />
to the traces before they were corrected. Here it has to be remarked that doing this is not<br />
mandatory, because negative traveltimes caused by the static correction correspond to reflection<br />
events located above the simulated measurement surface. <strong>The</strong>se can be neglected therefore.<br />
Applying a bulk-shift changes the later found NMO velocities, too. This is shown by the following<br />
equation, which holds for the NMO velocity according to equation (2.54a).<br />
v 2 NMO<br />
= 4h2<br />
t 2 −t 2 0<br />
. (3.2)<br />
Due to the squares, the denominator on the righthand side gets bigger if the same positive value is<br />
added to t and t 0 . Consequently a positive bulk-shift causes the measured NMO velocities to have<br />
smaller absolute values than those NMO velocities that would be measured without bulk-shift.<br />
For the used synthetic example one finds, after the elevation-statics and the bulk-shift were applied,<br />
NMO velocities that are smaller than those values that are provided by the known depth<br />
model. Thus in this example the v NMO lowering effect of the bulk-shift overcompensates the<br />
v NMO raising effect of the static correction. As expected, both effects decrease, with increasing<br />
distance between the reflector and X 0 . Evidently similar effects also occur for the found wavefield<br />
attributes β 0 ,K NIP , and K N .<br />
Finally, it can be said that the application of static corrections, bulk-shift and a subsequent stack<br />
with the conventional traveltime formula (2.50c) leads, <strong>under</strong> certain conditions, to a good stack<br />
result, but the found attributes {β 0 ,K N ,K NIP } loose their expected geometrical meaning, up to a<br />
certain degree. In addition it is more difficult to define appropriate search limits for the traveltime<br />
parameters. However, applying this strategy without a bulk-shift in case of a subsurface structure,<br />
which justifies the surface-consistency assumption better than the model used here, should<br />
yield satisfying results, even for the wavefield attributes.<br />
In this case, one finds for the first reflector, after bulk-shift and elevation-statics were applied,<br />
a NMO velocity of approximately 1900 m/s. According to the model it should be 2500 m/s.<br />
Consequently, if the v NMO search limits for this layer are chosen too close around 2500 m/s,
56 Synthetic Data Example<br />
assuming to know the velocity roughly, one obtains a poor or wrong stack result, as the search<br />
algorithm finds stacking surfaces which are the optimal within the chosen limits, only. Exactly<br />
this happened when ENI stacked the dataset, as it is shown in the following section.<br />
3.2.1 Results of the standard processing<br />
In this section results, obtained by the aforementioned standard processing, are displayed. For<br />
a better comparability to the results presented in Section 3.5 only the results of the so-called<br />
Automatic CMP stack are shown (see Mann, 2002). This data-driven CMP stack has the purpose<br />
to provide the NMO velocity values and a first provisional ZO section, both needed in the subsequent<br />
steps of the split traveltime-parameter search. However, the obtained CRS stack section<br />
looks very similar, due to the simpleness of the used subsurface model.<br />
<strong>The</strong> first three sections, Figures 3.3, 3.4, and 3.5, were obtained using wrong NMO velocity<br />
constraints, ignoring the influence of the elevation-statics and the bulk-shift. This is clearly<br />
visible in the coherency section, which is displayed in Figure 3.4. <strong>The</strong> overall v NMO lowering<br />
effect of the applied corrections leads to NMO velocities that are below the search range for<br />
nearly the whole first reflector and half of the second reflector. Consequently, the coherency<br />
values in this area are very low. At the right side, where the topography is higher we find a smaller<br />
v NMO lowering effect, because the static corrections are bigger and the original traveltimes larger.<br />
This can also be verified by looking at the v NMO section (Figure 3.5).<br />
For all that, the obtained ZO section (Figure 3.3), looks very satisfying, particularly because<br />
images of the reflectors are absolutely planar. However, this is caused by the simpleness of the<br />
used model, where all central rays emerge vertical at the surface. This has the consequence that<br />
the applied corrections adulterate the curvature of the hyperbolas found in the OD gather, but<br />
do not cause an error respective the new location of their apex, given by x0 ,t ∗ 0 . For an arbitrary<br />
layered model, comparable results can not be expected, because in that case the error in t ∗ 0 caused<br />
by the static correction, depends on the respective global take-off angle β g and thus changes for<br />
0<br />
every central ray.<br />
<strong>The</strong> second three sections, Figures 3.6, 3.7, and 3.8, were generated using NMO velocity limits<br />
that take the effect of the applied corrections into account. Unfortunately, the lower limit is still<br />
higher than the actual NMO velocity within those areas of the first reflector that are nearest to<br />
the surface, and where the effect of the bulk shift is highest. Nevertheless, it is evident that the<br />
results are enhanced considerably. Proper search limits should provide coherency values, close<br />
to one for all three reflectors.
3.2 Standard processing using elevation-statics 57<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
-0.1<br />
-0.2<br />
-0.3<br />
-0.4<br />
-0.5<br />
-0.6<br />
-0.7<br />
-0.8<br />
-0.9<br />
-1.0<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
4.0<br />
4.2<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.3: Automatic CMP stack section, created from the original dataset after applying static<br />
corrections and a bulk-shift. Ignoring the influence of the applied corrections to the found attributes,<br />
wrong search limits for the NMO velocity were used.
58 Synthetic Data Example<br />
0.95<br />
0.90<br />
0.85<br />
0.80<br />
0.75<br />
0.70<br />
0.65<br />
0.60<br />
0.55<br />
0.50<br />
0.45<br />
0.40<br />
0.35<br />
0.30<br />
0.25<br />
0.20<br />
0.15<br />
0.10<br />
0.05<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
4.0<br />
4.2<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.4: Coherency section of the Automatic CMP stack, created from the original dataset<br />
after applying static corrections and a bulk-shift. Due to the wrong search limits for the NMO<br />
velocity, the coherency values for the first and most of the second reflector are very low.
3.2 Standard processing using elevation-statics 59<br />
2900<br />
2880<br />
2860<br />
2840<br />
2820<br />
2800<br />
2780<br />
2760<br />
2740<br />
2720<br />
2700<br />
2680<br />
2660<br />
2640<br />
2620<br />
2600<br />
2580<br />
2560<br />
2540<br />
2520<br />
2500<br />
2480<br />
2460<br />
2440<br />
2420<br />
2400<br />
2380<br />
2360<br />
2340<br />
2320<br />
2300<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
4.0<br />
4.2<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.5: NMO velocity section, created from the original dataset after applying static corrections<br />
and a bulk-shift. Due to the wrong search limits of the NMO velocity, the found NMO<br />
velocity values for the first reflector and for the left side of the second reflector are not optimal.<br />
Please note that the found NMO velocities are generally lower than those velocities that would<br />
be expected according to the used subsurface model (Figure 3.1).
60 Synthetic Data Example<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
-0.1<br />
-0.2<br />
-0.3<br />
-0.4<br />
-0.5<br />
-0.6<br />
-0.7<br />
-0.8<br />
-0.9<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
4.0<br />
4.2<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.6: Automatic CMP stack section, created from the original dataset after applying static<br />
corrections and a bulk-shift. Considering the influence of the applied corrections to the found<br />
attributes, more appropriate v NMO search limits were used.
3.2 Standard processing using elevation-statics 61<br />
0.95<br />
0.90<br />
0.85<br />
0.80<br />
0.75<br />
0.70<br />
0.65<br />
0.60<br />
0.55<br />
0.50<br />
0.45<br />
0.40<br />
0.35<br />
0.30<br />
0.25<br />
0.20<br />
0.15<br />
0.10<br />
0.05<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
4.0<br />
4.2<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.7: Coherency section of the Automatic CMP stack, created from the original dataset<br />
after applying static corrections and a bulk-shift. Due to the better suited search limits for the<br />
NMO velocity, the coherency values are close to one for the major parts of the three reflectors.
62 Synthetic Data Example<br />
3100<br />
3000<br />
2900<br />
2800<br />
2700<br />
2600<br />
2500<br />
2400<br />
2300<br />
2200<br />
2100<br />
2000<br />
1900<br />
1800<br />
1700<br />
1600<br />
1500<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
4.0<br />
4.2<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.8: NMO velocity section, created from the original dataset after applying static corrections<br />
and a bulk-shift. Due to the better suited search limits of the NMO velocity, the found NMO<br />
velocity values are correct except of the two spots beneath the lowest points of the measurement<br />
surface. Please note that found NMO velocities are generally lower than those velocities that<br />
would be expected according to the used subsurface model (Figure 3.1).
3.3 Determination of the (near)surface attributes and elevation in X 0<br />
3.3 Determination of the (near)surface attributes and elevation<br />
in X 0<br />
In order to stack with traveltime formulas (2.48) and (2.103), respectively, or to compute the real<br />
wavefield attributes from pseudo attributes according to equations (2.51) and (2.104), it is necessary<br />
to know the values of the four (near)surface attributes {K 0 ,α 0 ,v 0 ,(∇v) 0 } in all central points<br />
X 0 . <strong>The</strong> near-surface velocity and its gradient have to be obtained directly in the field, whereas<br />
the latter can be measured either to global coordinates or to local coordinates (see Section 2.6.1).<br />
<strong>The</strong>oretically, also the curvature K 0 and the dip-angle α 0 could be measured directly, or be approximatively<br />
computed from the measured source and receiver positions, vicinal to X 0 . But this<br />
makes only sense if the measurement surface is actually smooth and representable in the vicinity<br />
of each central point according to equation (2.47). If there are little deviations from this ideal<br />
shape, then the local values of the dip and curvature can fluctuate, within a large range. Such<br />
a situation is displayed in Figure 3.11, where the dip-angle and curvature of the measurement<br />
surface (depicted in blue) show considerable changes within the considered stacking aperture.<br />
Similar situations are very common in the land-data acquisition. Thus, in practice it is rarely<br />
reasonable to work with the local values of K 0 and α 0 . Here, it makes more sense to determine<br />
those values of K 0 and α 0 that represent the locations of all sources and receivers that contribute<br />
to the stack, best (according to equation (2.47)).<br />
<strong>The</strong>re are different options how to get this averaged surface attributes. For example, one could<br />
use spline functions to smooth the measurement surface and determine the surface attributes afterwards.<br />
<strong>The</strong> problem of this approach is how to define the degree of smoothness. According<br />
to the conditions demanded in the derivation, the smoothed measurement surface has to be representable<br />
by a parabola within the stacking aperture of every point X 0 . But on the other hand,<br />
it is evident that the measurement surface must not be smoothed more than necessary. This is<br />
very difficult to achieve, especially if the respective smoothing algorithm shall work automatically,<br />
i.e., without further human interaction, for different datasets. <strong>The</strong> approach I used to get<br />
averaged surface attributes requires some computational effort, but seems to me to be the most<br />
natural way to solve this problem. For every central point X 0 a parabola (or a circle) is fitted<br />
to the source and receiver points that lie within the respective stacking aperture. Accordingly,<br />
this parabola is the best parabolic representation of the measurement surface within this area and<br />
constitutes the new averaged measurement surface in the vicinity of X 0 . If the stacking aperture<br />
depends on t 0 , it would be possible to determine values of K 0 and α 0 that depend on X 0 and t 0 ,<br />
but this would demand a very high computational effort. I propose in this case to use a average<br />
stacking aperture for every point X 0 , independent of t 0 .<br />
In addition to the values of the surface attributes K 0 and α 0 , the elevation z 0 of the central points<br />
X 0 has to be determined. Here holds the same as for K 0 or α 0 . This value could be measured<br />
directly, or be approximatively computed from the neighboring source and receiver positions.<br />
63
64 Synthetic Data Example<br />
Figure 3.9: Example for the determination of the averaged measurement surface dip α 0 by fitting<br />
a straight line (red) to the source and receiver locations within the stacking aperture (blue).<br />
<strong>The</strong> knowledge of α 0 enables to transfer the global source and receiver coordinates to the local<br />
coordinate system.<br />
However, if the measurement surface is not really smooth and representable by a parabola in the<br />
vicinity of every central point, one has to choose an averaged value, consistent with the averaged<br />
surface attributes. This value is given by the z-coordinate of the fitted parabola or circle, respectively,<br />
at xg = x 0 (see Figures 3.10, 3.11 and 3.14). If the distance between the considered source<br />
and receiver points and the best fitting parabola is to big to be assumed as negligible than it is<br />
proposed to apply static corrections, as described in Section 3.2 , to simulate source and receiver<br />
points lying on these parabola.<br />
3.3.1 Determination of α 0 , K 0 , and xz by fitting parabolas<br />
For the first tests I used the computer-algebra system MAPLE to determine K 0 , α 0 , and z 0 by<br />
fitting parabolas to the respective source and receiver locations. MAPLE provides an algorithm<br />
to fit a parabola of the form z = ax 2 + b to a set of points. This equation describes a parabola<br />
in the local coordinate system, so that I had firstly to transfer the respective source and receiver<br />
coordinates from the global to the local coordinate system. To determine the orientation of the<br />
local coordinate system I fitted a straight line to the considered surface points. To fit the parabola,<br />
MAPLE uses an iterative algorithm that minimizes the so called object function, i.e., in this case<br />
the sum of the squared distances of the single points to the parabola. <strong>The</strong> straight line can be
3.3 Determination of the (near)surface attributes and elevation in X 0<br />
Figure 3.10: Example for the determination of the averaged measurement surface curvature and<br />
elevation in X 0 by fitting a parabola (red) to the the source and receiver locations within the<br />
stacking aperture (blue). This fit is performed in the local coordinate system. Note that in the<br />
local coordinate system, the stacking aperture is not symmetrical with respect to X 0 anymore.<br />
fitted analytically in one step by solving a system of linear equation. Finally α 0 is given by the<br />
angle between this line and the x-axis of the global coordinate system, K 0 is provided by the<br />
coefficient a of the fitted parabola (see equation (2.28)) and z 0 can be computed by means of α 0<br />
and the coefficient b. <strong>The</strong>se two steps of fitting have to be performed for every central point X 0 .<br />
An example for this procedure is displayed in Figures 3.9 and 3.10.<br />
Please note that applying this strategy means to split the primary three parametric search for the<br />
best fitting arbitrarily oriented parabola, to a one parametric search for α 0 and a subsequent two<br />
parametric search for K 0 and z 0 . Doing this saves much computation time in comparison to the<br />
three parametric search. However, searching the minimum of the object function with respect<br />
to α 0 does not inevitably provide those value of α 0 that also minimizes the object function with<br />
respect to all three parameters. In this case, it works very well, but this does not hold generally<br />
and has always to be checked.<br />
<strong>The</strong> primary tests with MAPLE brought results, very close to the results that I achieved with the<br />
C++ implementation presented in the following. Thus, I have abstained from displaying them<br />
here.<br />
3.3.2 Determination of α 0 , K 0 , and xz by fitting circles<br />
As my final aim was to extend the existing 2D CRS stack implementation, I wrote a C++ program<br />
to determine K 0 , α 0 , and z 0 that could be integrated in the existing C++ code. To achieve<br />
a maximum at accuracy I decided to use the alternative strategy of fitting three parameters simultaneously.<br />
An arbitrarily oriented parabola has no unique explicit description in the global<br />
coordinate system and thus is very inconvenient to handle. For this reason, I used circles instead<br />
of parabolas. An example is shown in Figure 3.11. As a circle is a very good approximation of a<br />
parabola, in the vicinity of the apex, the error produced by fitting circles instead of parabolas is<br />
negligible. <strong>The</strong> three parameters that define a circle are the radius and the x- and z-coordinates<br />
of the midpoint. <strong>The</strong> curvature in X 0 is given by the inverse of the radius of the circle, the dip-<br />
65
66 Synthetic Data Example<br />
Figure 3.11: Example for the determination the averaged measurement surface dip, curvature,<br />
and elevation in X 0 by fitting a circle (red) to the the source and receiver locations within the<br />
stacking aperture (blue).<br />
angle is provided by its midpoint location, and the elevation z 0 is defined by its zg-coordinate at<br />
xg = x 0 .<br />
First implementation<br />
In my first C++ implementation, I performed an independent search for every central point X 0 .<br />
I did not use a minimization algorithm, like the Gradient or Newton method but tried every<br />
possible combination of parameters within a certain three dimensional search grid to find that<br />
combination that minimizes my object function. Unfortunately, this strategy required an immense<br />
computational effort. To demonstrate this I will give an example. For the initial search,<br />
it was necessary to sample the search range for every parameter in at least 40 steps, this means<br />
40 3 = 64000 possible circles. For every circle, the squared distances to, e.g., 200 source and<br />
receiver points have to be computed and summed up. This means 1.28 · 10 7 computations for<br />
one central point. My synthetic dataset had 1539 central points, thus at the end approximatively<br />
2 · 10 10 computations were necessary to determine α 0 , K 0 , and z 0 in all central points.<br />
Subsequently, every found parameter triple was optimized by an additional iterative search process<br />
that was performed within the close vicinity of the obtained values. For every iteration the<br />
search grid was refined. This was done until the achieved value of the object function changed<br />
only by a amount, smaller than a certain limit. <strong>The</strong> optimization does not demand much computation<br />
time, because for this step it is sufficient to divide the search space in, e.g., five samples<br />
per dimension. <strong>The</strong> usual number of iterations lies between one and three.<br />
<strong>The</strong> current implementation of the 2D CRS stack uses a taper function that gives the signals measured<br />
at the borders of the stacking aperture a lower weight. I have used the same taper function<br />
in my object function by weighting the squared distance of source and receiver points that lie at<br />
the border of the stacking aperture less than the distance of points from the center. <strong>The</strong> results,<br />
obtained with this implementation were very similar to the results that I obtained with MAPLE.<br />
Maybe they were slightly improved, but the run-time was five times longer and thus far beyond<br />
any acceptable limit.<br />
By the way, in principal this search process is very similar to that one applied within the 2D
3.3 Determination of the (near)surface attributes and elevation in X 0<br />
ZO CRS stack itself, where for every central point X 0 and every Z0-traveltime t 0 the hyperbolic<br />
traveltime surface has to be found which yields the highest coherency value within the pre-stack<br />
data. Usually this search for β 0 , K N , and K NIP is splited to save computation time (see, e.g.,<br />
Mann, 2002).<br />
Final implementation<br />
<strong>The</strong> crucial point of any optimization process is the previously known information that can be<br />
used to confine the search and to tailor the search algorithm to the specific problem. In this case<br />
we have two informations that are very useful to restrict the search process. <strong>The</strong>se are:<br />
1. <strong>The</strong> source and receiver points within the stacking aperture do not change considerable<br />
from one central point to the next.<br />
2. <strong>The</strong> source and receiver points are not arbitrarily distributed in space but lie on a real<br />
measurement surface which can only be rough within certain “natural” limits.<br />
According to this considerations, my final implementation determines only the first parameter<br />
triple using large search ranges. From this triple suited search ranges for the next triple are derived<br />
and so on. This works very well, because most of the source an receiver points that lie in<br />
the stacking aperture of one central point, lie also within the stacking aperture of its neighboring<br />
central points. Of course, it is very important for the stability of this approach that the search<br />
ranges are not too narrow. If the search fails in one central point because the optimal parameter<br />
triple lays outside of the search space, then it fails in general in all following points, too. To handle<br />
this problem I modified the optimization step. Due to the fact that the search ranges can be<br />
chosen much smaller, since the information from the previous search is used, it is not necessary<br />
anymore to refine the search grid considerable within the optimization. Now it is possible to use<br />
the iterations that were needed before solely to refine the search grid, also to move the search<br />
grid within the parameter space towards the searched minimum, if this lays outside. <strong>The</strong> modified<br />
optimization procedure serves both, to refine the initial result and to make the search more<br />
flexible and stable. I will give an 1D example to explain the functionality of this procedure. If the<br />
minimum of the object function is given for the parameter value 10, but the initial search range<br />
of this parameter is from 1 to 7, then the search will yield the value 7. For the first optimization<br />
step the new search range is, e.g., 5 to 9 with a slightly refined grid. Accordingly the new result<br />
would be 9. For the second optimization step the grid is refined again and the respective search<br />
range is 8 to 10. This leads to the result 10. <strong>The</strong> third search step with a range from 9.5 to<br />
10.5 achieves no considerable improvement, thus the search is finished. Note that this procedure<br />
demands that the object function is well behaved, i.e., has no further local minima in the vicinity<br />
of the global minima.<br />
This modified search algorithm leads to a drastic reduction of the run-time. Since this imple-<br />
67
68 Synthetic Data Example<br />
Figure 3.12: <strong>The</strong> measurement surface curvature K 0 determined by fitting circles to all source<br />
and receiver locations within the respective stacking aperture. This figure shows the periodic<br />
small scale undulation of the measurement surface very clearly.<br />
mentation provides much more options to adjust the algorithm to the used topography data it is<br />
possible to achieves the same results as with the former implementation in 1% of the time.<br />
3.4 Forward calculation<br />
In this section, forward calculated properties of the subsurface model, displayed in Figure 3.1,<br />
were used to study by means of a practical example the validity of the search-range determination,<br />
derived in Chapter 2. Due to the simplicity of the model it is not necessary to use a<br />
ray-tracing software to obtain the wavefield attributes K N , K NIP , and β 0 . In Chapter 2 all equations<br />
were derived, needed to compute these values from the known properties of the subsurface<br />
model and the measurement surface. Knowing the wavefield attributes {K N ,K NIP ,β 0 } and the<br />
surface attributes {K 0 ,α 0 ,z 0 }, determined in the last section, it is possible to estimate the measured<br />
NMO velocity and every particular set of pseudo attributes, both in global and in local<br />
coordinates. In addition, one can transfer the search limits that would hold for a fictitious horizontal<br />
measurement surface to appropriate limits that hold for the actual surface. For the sake<br />
of brevity, the discussion will be focused upon the case of applying a conventional 2D ZO CRS
3.4 Forward calculation 69<br />
Figure 3.13: <strong>The</strong> measurement surface dip α 0 determined by fitting circles to all source and<br />
receiver locations within the respective stacking aperture.<br />
Figure 3.14: Comparison between the real measurement surface, provided by the source and<br />
receiver locations (red) and that surface that is given by the determined locations of the central<br />
points X 0 (x 0 ,z 0 ) (black).
70 Synthetic Data Example<br />
Figure 3.15: <strong>The</strong> forward calculated take-off angle β 0 (red) and its search limits (black). Due to<br />
the simplicity of the model the same take-off angle holds for all central rays that impinge at a<br />
certain central point X 0 .<br />
stack to the synthetic dataset, used in this chapter. As mentioned before, the conventional CRS<br />
stack, designed for a planar measurement surface, uses global coordinates, and does not consider<br />
α 0 , K 0 , and E 0 .<br />
3.4.1 Forward calculated take-off angle β 0 and its search limits<br />
<strong>The</strong> three reflectors of the subsurface model are horizontal, thus all central rays propagate in<br />
vertical direction and their global take-off angles β g are zero. For this very simple situation, the<br />
local take-off angles β 0 can be easily computed according to equation (2.52). Please note, in<br />
general every central ray has its specific take-off angle, but in this particular situation, the same<br />
β 0 holds for all central rays that impinge at the same central point. Here it is possible to plot β 0<br />
over x 0 without considering a specific reflector.<br />
In the following the search range of β g<br />
0 is assumed to be [−80◦ ,80 ◦ ]. Based on this, the search<br />
limits of β0 can be computed according to equation (2.68). In practice it can be reasonable to<br />
choose the β g<br />
0 search range in dependence of t0 . This case is not considered here. <strong>The</strong> resulting<br />
values of β0 and the respective search limits are displayed in Figure 3.15.<br />
0
3.4 Forward calculation 71<br />
Figure 3.16: <strong>The</strong> sine of the forward calculated pseudo take-off angle β ∗ 0 (red) and its search<br />
limits (black). Here the sine of β ∗ 0 is used for the plot, because <strong>under</strong> certain circumstances<br />
sinβ ∗ 0 becomes larger than one, and consequently β ∗ 0 complex. Due to the simplicity of the<br />
model the same pseudo take-off angle holds for all central rays that impinge at a certain central<br />
point X0 .<br />
Figure 3.17: <strong>The</strong> RMS velocities calculated for the model shown in Figure 3.1. <strong>The</strong> RMS velocities<br />
for the first layer are depicted in red those that corresponds to the bottom of the second layer<br />
in blue and those that correspond to the bottom of the third layer in green.
72 Synthetic Data Example<br />
Figure 3.18: <strong>The</strong> forward calculated NMO velocity for the first reflector. Negative values correspond<br />
to imaginary velocities, according to convention (3.4).<br />
3.4.2 Forward calculated RMS and NMO velocities<br />
<strong>The</strong> RMS velocities that hold for the three reflectors of the subsurface model (Figure 3.1), are<br />
computed according to equation (2.62). <strong>The</strong> results are shown in Figure 3.17, where the RMS<br />
velocities that correspond to the first reflector and the whole first layer are shown in red, those<br />
that correspond to the second reflector in blue and those that correspond to the third reflector in<br />
green. <strong>The</strong>se colors are also used in the following figures, to signify the considered reflector. It<br />
is evident that the RMS velocity of the first layer is constant due to the fact that it is the average<br />
velocity within a homogeneous layer. However, the RMS velocity at all points below the first<br />
reflector depends on the vertical extension and of the overlaying layers.<br />
After the RMS velocities are determined, the NMO velocities that are expected to be obtained<br />
applying the conventional CRS stack can be computed by means of equation (2.109a). Please<br />
note that it is necessary to use the ”averaged”X 0 elevation, i.e., z 0 , for the computation of the<br />
values of t 0 . <strong>The</strong> results are shown in the Figures 3.18, 3.19, and 3.20. <strong>The</strong> NMO velocities that<br />
are expected to be obtained applying the conventional CRS stack using local coordinates are in<br />
principle very similar, and can be derived according to equation (2.108). As mentioned before,<br />
the NMO velocity is imaginary if<br />
K NIP cosβ 0 − K 0 < 0, (3.3)
3.4 Forward calculation 73<br />
Figure 3.19: <strong>The</strong> forward calculated NMO velocity for the second reflector. Negative values<br />
correspond to imaginary velocities, according to convention (3.4).<br />
Figure 3.20: <strong>The</strong> forward calculated NMO velocity for the third reflector. Negative values correspond<br />
to imaginary velocities, according to convention (3.4).
74 Synthetic Data Example<br />
Figure 3.21: <strong>The</strong> forward calculated NMO slowness for the first reflector (red) and its search<br />
limits (black).<br />
according to equation (2.109a) with E0 equal zero. <strong>The</strong>refore, the so-called signed square root<br />
was used to visualize imaginary NMO velocities in the plots. This reads<br />
vNMO = sign � v 2 �<br />
NMO<br />
�<br />
|v2 | . (3.4)<br />
NMO<br />
Consequently imaginary values are displayed by negative values in the Figures 3.18, 3.19, and<br />
3.20. Please note that the analytical function given by equation (2.109a) provides of course infinite<br />
values for the squared NMO velocity, at any location where the denominator changes it<br />
sign. However, the displayed values of the NMO velocity correspond to discrete points at the<br />
measurement surface, where in general the denominator of equation (2.109a) is not exactly zero.<br />
K NIP is always positive for the assumed subsurface model. Thus the NMO velocity only becomes<br />
imaginary, if the surface curvatures K 0 is positive and larger than the respective value of<br />
K NIP cosβ 0 .<br />
3.4.3 Forward calculated slowness p NMO and its search limits.<br />
As mentioned before, the first parameter, searched within the current implementations of the 2D<br />
ZO CRS stack is the squared NMO slowness p2 NMO . In Section 2.4.3 a strategy was derived, to<br />
transfer the search limits that would hold for a planar measurement surface to limits that account
3.4 Forward calculation 75<br />
Figure 3.22: <strong>The</strong> forward calculated NMO slowness for the second reflector (blue) and its search<br />
limits (black).<br />
Figure 3.23: <strong>The</strong> forward calculated NMO slowness for the third reflector (green) and its search<br />
limits (black).
76 Synthetic Data Example<br />
Figure 3.24: <strong>The</strong> forward calculated values of the NIP-wave curvature K NIP . <strong>The</strong> values that<br />
correspond to the first reflector are depicted in red, those for the second reflector in blue and<br />
those for the third reflector in green<br />
for the actual measurement surface, if E0 = 0. In order to treat the case of applying a conventional<br />
CRS stack, global coordinates have to be considered according to Section 2.6.3. To study this<br />
are assumed:<br />
strategy by means of the used dataset, the following search limits for v RMS and β g<br />
0<br />
vmin RMS (x0 ,t0 ) = vRMS (x0 ,t0 ) − 500m/s ,<br />
g,min<br />
β = −80<br />
0<br />
◦ , (3.5a)<br />
vmax RMS (x0 ,t0 ) = vRMS (x0 ,t0 ) + 500m/s ,<br />
g,max<br />
β = +80 ◦ , (3.5b)<br />
wheres, without loss of generality, v RMS limits instead of v NMO,ZD limits are used in this case.<br />
In the plots, p NMO instead of p 2 NMO is displayed, in order to simplify the comparison to v NMO .<br />
<strong>The</strong>refore the signed square root was used, according to its definition in equation (3.4). <strong>The</strong> results<br />
are displayed in the Figures 3.21, 3.22, and 3.23. One can see that a remarkable search-range<br />
reduction can be achieved, if the influence of the measurement surface is considered, particularly<br />
in surface areas, with large and positive K 0 values.<br />
3.4.4 Forward calculated values of K N and K NIP .<br />
It is evident that the Normal-wave curvature K N is zero for the first reflector. Looking at the<br />
refraction and transmission laws, equations (2.92) and (2.93), reveals that K N is zero for the<br />
0
3.4 Forward calculation 77<br />
Figure 3.25: <strong>The</strong> forward calculated values of K∗ N<br />
fact that the reflectors are planar, both, the limiting values and the obtained values of K ∗ N<br />
same for all three layers.<br />
(red) and its search limits (black). Due to the<br />
are the<br />
second and third reflector, too. In other words, a planar wavefront does not change its curvature,<br />
passing through planar reflectors. Consequently it can be abstained from displaying the values<br />
of K N here.<br />
For the calculation of K NIP one could apply the refraction and transmission laws, too. However,<br />
this is not necessary for this simple case, were the obtained values of the RMS velocity can be<br />
used to calculate K NIP according to equation (2.65). <strong>The</strong> results are displayed in Figure 3.24.<br />
3.4.5 Pseudo attributes β ∗ 0 , K∗ N , and K∗ NIP and their search limits.<br />
In case of the pseudo take-off angle β ∗ 0 it was mentioned before that a conventional CRS stack<br />
software, has generally to consider complex values of β ∗ 0 within the search range, since β ∗ 0 is<br />
complex, if sinβ ∗ 0 is larger than one (see equation (2.104a)). Usually, this requires a slightly<br />
modification within the code of the used software.<br />
To avoid complex values in the plot, the sine of β ∗ 0 and its search limits are displayed in Figure<br />
3.16. <strong>The</strong> latter correspond to the search range of β g<br />
0 , i.e., [−80◦ ,80◦ ]. It can be seen that the<br />
pseudo take-off angles that would be obtained for this dataset by a conventional CRS stack have<br />
no complex values. This is due to the small take-off angles β0 (Figure 3.15). However the black
78 Synthetic Data Example<br />
Figure 3.26: <strong>The</strong> forward calculated values of K ∗ NIP<br />
sponding search limits (black).<br />
Figure 3.27: <strong>The</strong> forward calculated values of K ∗ NIP<br />
responding search limits (black).<br />
for the first reflector (red) and the corre-<br />
for the second reflector (blue) and the cor
3.4 Forward calculation 79<br />
Figure 3.28: <strong>The</strong> forward calculated values of K ∗ NIP<br />
sponding search limits (black).<br />
for the third reflector (green) and the corre-<br />
lines show that generally it would have been possible to obtain complex values for the pseudo<br />
take-off angles β ∗ 0 on this measurement surface.<br />
were chosen:<br />
To display the respective search range of K ∗ N the following search limits for K N<br />
K min<br />
N = − 1<br />
1000m<br />
and Kmax N = 1<br />
. (3.6)<br />
1000m<br />
On the basis of this values, the corresponding limits of K∗ N can be determined, according to equations<br />
(2.111). <strong>The</strong> values of K∗ N that would be obtained with the conventional CRS stack are<br />
computed by means of equation (2.104b). Due to the fact that the reflectors are planar, the values<br />
of K∗ N , are the same for each of the three layers. Of course, the same holds for the search limits.<br />
<strong>The</strong> results are depicted in Figure (3.25).<br />
Starting from the forward modeled values of KNIP , those values of K∗ NIP can be derived which<br />
are expected to be obtained with the conventional CRS stack. This is done by means of equation<br />
(2.104c). KNIP is with exception of confliction dip situations (see Mann, 2002) no search parameter<br />
in the current implementations of the 2D ZO CRS stack. Nevertheless, those limits for<br />
K∗ NIP that correspond to the used vRMS limits are displayed, too. <strong>The</strong>se are computed according<br />
to equation (2.111), where the subscript N has to be substituted by the subscript NIP. <strong>The</strong> results<br />
for the different layers are displayed in Figures 3.26, 3.27, and 3.28.
80 Synthetic Data Example<br />
3.5 Outlook<br />
As mentioned before, the final objective of my work presented in this thesis is to extent the existing<br />
2D ZO CRS stack software for planar measurement surfaces, such as to handle data measured<br />
on a curved surface and to consider the near surface velocity gradient. In addition, the obtained<br />
results shall be referred to a fictitious planar measurement surface to simplify interpretation and<br />
further processing. Unfortunately, it was not possible to finish theses extensions within the temporal<br />
range of this thesis. <strong>The</strong> 2D ZO CRS stack software for planar measurement surfaces is<br />
highly developed and accordingly very complex. Every single change or extension of the code<br />
has to be considered carefully and thoroughly tested.<br />
In this section I show results that were obtained, performing the first step of the CRS stack procedure<br />
(see Mann, 2002), i.e., the Automatic CMP stack, using global coordinates for half-offset<br />
and midpoint. As mentioned before, these data-driven CMP stack has the purpose to provide the<br />
NMO velocity values (here vg ) and a first provisional ZO section, both needed in the subse-<br />
NMO<br />
quent steps of the split traveltime-parameter search. <strong>The</strong> strategy to determine appropriate search<br />
limits for the NMO velocity, depending on the dip and curvature of the measurement surface of<br />
every central point (see Section 2.4.3 and 2.6.3) was not jet implemented. Thus a sufficiently<br />
large global search range for the NMO velocity and a coarsely sampled search grid were used.<br />
3.5.1 Automatic CMP stack plus redatuming<br />
As expected, the time domain images of the horizontal reflectors are, due to the influence of<br />
the topography, not planar, but curved (see Figure 3.29). <strong>The</strong> same holds for the v g<br />
NMO section,<br />
represented in Figure 3.31. To remove these curvature, a planar measurement surface at sea-level<br />
was simulated by mapping the central points and the respective traveltimes t 0 to this surface,<br />
according to the derivations made in in Section 2.5.1. Please note, the fictitious measurement<br />
surface was chosen at sea level to simplify the comparison to the results obtained via static corrections<br />
and a conventional CRS stack (see Section 3.2). In general, the velocity of the first layer<br />
is unknown and the fictitious measurement surface must be chosen above the real measurement<br />
surface. <strong>The</strong> resulting ZO section is shown in Figure 3.34, the vg section in Figure 3.31. It<br />
NMO<br />
is clearly observable that also the obtained values of the NMO velocity are strongly influenced<br />
by the acquisition topography. Below the highest points of the topography, the denominator of<br />
equation (2.54b) is negative and one finds imaginary values for the NMO velocity. <strong>The</strong>se are displayed<br />
using convention (3.4). Consequently, the simulation of a planar measurement surface is<br />
not complete, as it would also be necessary to transfer the values of vg to those values which<br />
NMO<br />
would be obtained at the fictitious measurement surface at sea level. An eye-catching property<br />
of these results, are the vertical strips, most clearly visible in the coherency section depicted in
3.5 Outlook 81<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
-0.1<br />
-0.2<br />
-0.3<br />
-0.4<br />
-0.5<br />
-0.6<br />
-0.7<br />
-0.8<br />
-0.9<br />
Time [s]<br />
0.8<br />
1.0<br />
1.2<br />
1.4<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.29: Automatic CMP stack section. Due to the influence of the acquisition topography<br />
to the measured traveltimes, the time domain images of the reflectors are curved. <strong>The</strong> effect of<br />
a, from left to right increasing, periodic small-scale topography undulation that does not satisfy<br />
the smoothness requirements, can be observed.
82 Synthetic Data Example<br />
0.95<br />
0.90<br />
0.85<br />
0.80<br />
0.75<br />
0.70<br />
0.65<br />
0.60<br />
0.55<br />
0.50<br />
0.45<br />
0.40<br />
0.35<br />
0.30<br />
0.25<br />
0.20<br />
0.15<br />
0.10<br />
0.05<br />
Time [s]<br />
0.8<br />
1.0<br />
1.2<br />
1.4<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.30: Coherency section of the automatic CMP stack. Due to the influence of the acquisition<br />
topography to the measured traveltimes, the time domain images of the reflectors are<br />
curved. <strong>The</strong> effect of a, from left to right increasing, periodic small-scale topography undulation<br />
that does not satisfy the smoothness requirements, can be observed most clearly in this section.
3.5 Outlook 83<br />
4000<br />
3500<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
-500<br />
-1000<br />
-1500<br />
-2000<br />
-2500<br />
-3000<br />
Time [s]<br />
0.8<br />
1.0<br />
1.2<br />
1.4<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
3.6<br />
3.8<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.31: NMO velocity section. Negative values correspond to imaginary velocities, according<br />
to convention (3.4). Due to the influence of the acquisition topography to the measured<br />
traveltimes, the time domain images of the reflectors are curved. <strong>The</strong> found NMO velocities are<br />
strongly influenced by the shape of the measurement surface. <strong>The</strong> effect of a, from left to right<br />
increasing, periodic small-scale topography undulation that does not satisfy the smoothness requirements<br />
can be observed.
84 Synthetic Data Example<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
-1500 -1000 -500<br />
Offset [m]<br />
0 500 1000 1500<br />
Figure 3.32: CMP gather (x 0 = 10.8 km). <strong>The</strong> periodic small-scale undulation of the measurement<br />
surface is clearly visible in the data. Please note the curvature of the traveltime functions<br />
that corresponds to imaginary NMO velocity values.
3.5 Outlook 85<br />
Time [s]<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
3.0<br />
3.2<br />
3.4<br />
-1500 -1000 -500<br />
Offset [m]<br />
0 500 1000 1500<br />
Figure 3.33: CMP gather (x 0 = 10.7 km). <strong>The</strong> periodic small-scale undulation of the measurement<br />
surface is hardly visible, because this CMP is located at a zero-crossing of the periodic<br />
undulation. Here, due to the horizontal reflectors, the traveltime variation caused by the undulation<br />
has the same absolute value on the right and on the left branch of the ray path, but the<br />
opposite sign, and thus the effect of the undulation to the whole traveltime is equal zero for all<br />
CMP experiments in this gather.
86 Synthetic Data Example<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
-0.1<br />
-0.2<br />
-0.3<br />
-0.4<br />
-0.5<br />
-0.6<br />
-0.7<br />
-0.8<br />
-0.9<br />
Time [s]<br />
0.8<br />
1.0<br />
1.2<br />
1.4<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.34: Automatic CMP stack section after redatuming. After transferring the ZO traveltimes<br />
to those values which would be measured on a horizontal measurement surface at sea<br />
level, the reflectors are planar and horizontal. However, the effect of the small-scale topography<br />
undulation cannot be removed by this procedure.
3.5 Outlook 87<br />
4000<br />
3500<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
-500<br />
-1000<br />
-1500<br />
-2000<br />
-2500<br />
-3000<br />
Time [s]<br />
0.8<br />
1.0<br />
1.2<br />
1.4<br />
1.6<br />
1.8<br />
2.0<br />
2.2<br />
2.4<br />
2.6<br />
2.8<br />
Global x-coordinate [km]<br />
0 2 4 6 8 10 12 14<br />
Figure 3.35: NMO velocity section after redatuming. After transferring the ZO traveltimes to<br />
those values which would be measured on a horizontal measurement surface at sea level, the<br />
reflectors are planar and horizontal. However, the values of the NMO velocity are not transferred<br />
to those values that would be measured at this fictitious measurement surface. <strong>The</strong> black lines<br />
denote the picked ZO traveltimes that refer to the reflectors. Of course, the effect of the smallscale<br />
topography undulation remains visible.
88 Synthetic Data Example<br />
Figure 3.30. <strong>The</strong>y are caused by a, from left to right increasing, periodic small-scale topography<br />
undulation (see Figure 3.1) that does not satisfy the smoothness requirements of the used traveltime<br />
formula. Of course, it is still the best second order approximation of the real traveltime, but<br />
the approximation is much worse than it would be for a measurement surface representable by a<br />
second order function. In addition the obtained wave-field attributes lose their physical meaning<br />
to a certain extent. High coherency values are obtained only at those central points, where the<br />
small-scale undulation crosses zero. This is caused by the periodicity of the undulation, and can<br />
be verified by, looking at the single CMP gathers at x 0 = 10.7 km and x 0 = 10.8 km, pictured in<br />
Figures 3.33 and 3.32. This strips should be hardly visible in the final CRS stack sections, because<br />
there, the data is stacked along surfaces and the effect of the undulation should be smeared.<br />
Nevertheless, I propose to remove such small-scale undulations as far as possible by static corrections<br />
before the CRS stack is performed. For very complex top-surface topography I would<br />
like to refer to an alternative approach, presented in Zhang et al. (2002), which enables to handle<br />
an arbitrary surface topography by considering the elevation of every single source and receiver<br />
point explicitly within a modified CRS traveltime formula. <strong>The</strong> implementation of this approach<br />
is still in work, but first results are very promising.<br />
3.5.2 Comparison with the predicted NMO velocities<br />
In order to compare the obtained values for the NMO velocity with those values, predicted in<br />
Section 3.4.2, the NMO velocities for the three reflectors were extracted from the v g<br />
NMO section.<br />
This was done on the basis of picked ZO traveltimes, represented as black lines in Figure 3.35.<br />
<strong>The</strong> extracted NMO velocities of the three reflectors are depicted in Figures 3.36, 3.37, and<br />
3.38. Due to the influence of the small-scale topography undulation, the picked NMO velocities<br />
fluctuate considerably in the affected region. In addition, one has no usable values for the NMO<br />
velocity in the vicinity of the left and right boarders of the acquisition surface, as there are not<br />
enough traces within the CMP gathers. Nevertheless, an overall consistency to the predicted<br />
values is observable. However, the results for the first reflector match only qualitatively, because<br />
it was not considered that the used CRS stack implementation determines the stacking aperture<br />
depending on t 0 . Contrary to this, the largest possible stacking aperture, containing all existing<br />
traces within the CMP gather, was used for the determination of the surface attributes α 0 and<br />
K 0 and for the central-point elevation z 0 . Thus the apertures actually used for the first reflector<br />
were smaller than the aperture used for the prediction. <strong>The</strong>refore, the absolute values of the<br />
determined curvature and dip were too small with respect to the averaged values within the<br />
real stacking aperture. For the second and the third reflector the maximum stacking aperture<br />
was used, as expected in the prediction. Here the predicted and the actually determined values<br />
match also quantitatively. This shows how important it is to determine the values of the dip and<br />
particularly of the curvature consistently with the later used stacking procedure.
3.5 Outlook 89<br />
VNMOg [m/s]<br />
x10<br />
7<br />
4<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
-1<br />
-2<br />
-3<br />
-4<br />
-5<br />
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15<br />
Xg [km]<br />
Figure 3.36: Extracted NMO velocity for the first layer. Negative values correspond to imaginary<br />
velocities, according to convention (3.4). To extract these velocity values, picked ZO traveltimes<br />
were used, which are represented in Figure 3.35 by black lines.
90 Synthetic Data Example<br />
VNMOg [m/s]<br />
x10<br />
3<br />
4<br />
2<br />
1<br />
0<br />
-1<br />
-2<br />
-3<br />
-4<br />
-5<br />
-6<br />
-7<br />
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15<br />
Xg [km]<br />
Figure 3.37: Extracted NMO velocity for the second layer. Negative values correspond to imaginary<br />
velocities, according to convention (3.4). To extract these velocity values, picked ZO traveltimes<br />
were used, which are represented in Figure 3.35 by black lines.
3.5 Outlook 91<br />
VNMOg [m/s]<br />
x10<br />
5<br />
4<br />
4<br />
3<br />
2<br />
1<br />
0<br />
-1<br />
-2<br />
-3<br />
-4<br />
-5<br />
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15<br />
Xg [km]<br />
Figure 3.38: Extracted NMO velocity for the third layer. Negative values correspond to imaginary<br />
velocities, according to convention (3.4). To extract these velocity values, picked ZO traveltimes<br />
were used, which are represented in Figure 3.35 by black lines.
92 Synthetic Data Example
Chapter 4<br />
Summary<br />
<strong>The</strong> presented extension of the 2D ZO CRS stack enables to stack seismic data, considering<br />
the average dip and curvature of the measurement surface within the stacking aperture and the<br />
near-surface velocity gradient at the emergence point of the central ray. As in case of a planar<br />
measurement surface, three wavefield attributes are determined that provide important informations<br />
about the investigated subsurface structure. <strong>The</strong>se can find applications in a number of<br />
kinematic and dynamic modeling, inversion and stacking problems. <strong>The</strong> stack remains purely<br />
data-driven, i.e., independent of the 2D laterally inhomogeneous velocity model. Only the nearsurface<br />
velocity and its gradient in the vicinity of the coincident source and receiver points are<br />
required.<br />
To simplify subsequent interpretation and further processing, it is reasonable to perform a redatuming<br />
process that relates the obtained results to a planar measurement surface, with constant<br />
near-surface velocity. All relations needed for that purpose were derived and discussed within<br />
this thesis.<br />
<strong>The</strong>oretically, it is possible to use the conventional CRS stacking software, assuming a planar<br />
measurement surface with constant near-surface velocity, even though the data is measured on<br />
a curved surface having a near-surface velocity gradient. It was shown that the found pseudo<br />
wavefield attributes can be corrected afterwards. All necessary relations were derived, and the<br />
advantages and disadvantages of this approach were extensively discussed.<br />
A modified formulation of the CRS traveltime formula, for data measured on a curved surface<br />
having a near-surface velocity gradient, presented in Chira et al. (2001), was derived. This<br />
formulation is better suited for the practical application, as the used midpoint and half-offset<br />
coordinates are referred to a global coordinate system, instead of the local coordinate system of<br />
the considered central point.
94 Summary<br />
Very important for an efficient implementation of the CRS stack formalism are the search limits<br />
of the traveltime parameters. It was shown, that the found NMO velocity and also the pseudo<br />
wavefield attributes, determined by conventional software, are strongly affected by near-surface<br />
velocity variations and particularly by the surface topography. For this reason, strategies were<br />
derived to calculate appropriate search limits for the actual measurement surface from those<br />
values that would be valid for a planar measurement surface with constant near-surface velocity.<br />
<strong>The</strong> validity of the theoretical considerations mentioned above was studied by means of a synthetic<br />
dataset, whereby the near-surface velocity gradient was not considered. In addition, the<br />
main steps on the way to a satisfying implementation of the presented theory, on the basis of the<br />
existing 2D CRS stack software, were made. Unfortunately, is was not possible to finish this<br />
part completely within the temporal range of this diploma thesis. <strong>The</strong>refore final results of the<br />
presented CRS stack extension could not be shown.<br />
For comparison, results obtained by ENI/Agip via standard processing using static corrections<br />
and the conventional CRS stack software were discussed. Afterwards, these results were enhanced<br />
by re-processing the dataset using better suited NMO velocity limits that take the effect<br />
of the applied corrections into account. Advantages and disadvantages of this approach in comparison<br />
to the presented CRS stack extension were pointed out.<br />
A fast and stable C++ code to determine the surface attributes α 0 , K 0 , and z 0 of an arbitrarily<br />
shaped measurement surface was written. This code was tested by means of the synthetic dataset,<br />
and included into the existing 2D CRS stack software. <strong>The</strong> used algorithm was described in detail<br />
and the obtained results were displayed.<br />
On the basis of the determined surface attributes and the forward modeled wavefield attributes<br />
β 0 , K N , and K NIP the validity of the proposed search range estimation was demonstrated and<br />
other important aspects related to the application of the CRS stack to data measured on a curved<br />
surface were pointed out.<br />
Finally, the first step of the CRS stack procedure, the Automatic CMP stack, was performed.<br />
In addition, a redatuming process was applied to the resulting NMO velocity and CMP stack<br />
sections. <strong>The</strong> obtained NMO velocities were compared to those values that were predicted on<br />
the basis of the forward modeled wavefield attributes and the determined surface attributes. It<br />
was shown that they were consistent within the expected range.
Appendix A<br />
<strong>The</strong> scalar Hamilton’s equation<br />
According to Section (2.2.1), equation (2.11), the Hamilton’s equation for two-point ray tracing<br />
reads<br />
dt = p G · d(x G − x G ) − p S · d(x S − x S ). (A.1)<br />
Let us look at the infinitesimal source dislocation d(x S − x S ) and the slowness vector of the<br />
paraxial ray at the anterior surface p S , keeping in mind that the following derivations are also<br />
valid for the corresponding values d(x G − x G ) and p G at the posterior surface. If we assume the<br />
anterior surface in the vicinity of S to be representable by the smooth analytical function f (x),<br />
we can express d(x S − x S ), in linear approximation, as<br />
d(x S − x S ) =<br />
�<br />
Δx S ,<br />
∂ f<br />
∂x (S) Δx S<br />
� T<br />
, (A.2)<br />
where Δx S , the x-coordinate of d(x S − x S ), results from the projection of d(x S − x S ) onto the<br />
tangent of the surface in S, i.e., the x-axis of the local coordinate system at the anterior surface.<br />
This means that we assume the infinitesimal source dislocation d(x S − x S ) to be tangent to the<br />
measurement surface in S, as we only consider the first derivative of f (x) at S. Consequently we<br />
can express the dot product at the righthand side of equation (A.1) as<br />
p S · d(x S − x S ) = p S,T · d(x S − x S ) , (A.3)<br />
with p S,T being the projection of p S onto the tangent to the surface in S (see Figure (2.2)).<br />
In the same way as above, we can express the slowness of the paraxial ray at the anterior surface,<br />
i.e., pS , in linear approximation, as<br />
�<br />
∂ f<br />
pS,T = pS ,<br />
∂x (S) p �T S , (A.4)
96 <strong>The</strong> scalar Hamilton’s equation<br />
where p S , the x-coordinate of p S,T , results from the projection of p S,T onto the tangent to the<br />
surface in S, according to Figure 2.2.<br />
Due to our choice of the local coordinate system, f (x) is tangent to the x-axis at origin S and<br />
consequently, the first derivative of f (x) has only first or higher order terms. This has the consequence<br />
that the z-components of the dot products on the righthand side of equation (A.1) have<br />
only terms that are of second order or higher in Δx S and Δx G , respectively (see equations (A.2),<br />
(A.3), and (A.4)). <strong>The</strong>se terms can be neglected within the derivation of a traveltime formula<br />
which shall only be valid up to the second order of Δx S and Δx G , respectively, because they get<br />
third and higher order when we finally integrate the Hamilton’s equation to obtain an expression<br />
for the traveltime. Accordingly, we can reduce in this case the vector representation of the<br />
Hamilton’s equation (A.1) to a scalar expression. This reads<br />
dt = p G d(Δx G ) − p S d(Δx S ). (A.5)
Appendix B<br />
Used hard- and software<br />
<strong>The</strong> computations were done on a dual-processor Linux PC with (S.u.S.E. Linux 6.1), on<br />
HEWLEDTT PACKARD workstations 9000 with HP-UX 10.20 and on a SILICON GRAPH-<br />
ICS ORIGIN 3200 (6 processors) with IRIX 6.5.<br />
For various analytical calculations and for visualization of data, I used Maple V Release 5.1<br />
(Waterloo Maple) and MATLAB 6.1 Release 12.1.<br />
To visualize the various stack and attribute sections I used the Seismic Unix package (Center of<br />
Wave Phenomena at Colorado School of Mines).<br />
This thesis was written on a PC (S.u.S.E. Linux 6.1) using the freely available word processing<br />
package TEX, the macro package LATEX 2 ε , and several extensions.<br />
<strong>The</strong> bibliography was generated with BIBTEX.<br />
Figures were mainly constructed with Xfig 3.2 (rev2) and Corel Draw 8.0.
98 Used hard- and software
Appendix C<br />
Acknowledgment/Danksagung<br />
An erster Stelle möchte ich mich bei meiner Familie für die Unterstützung während meines<br />
Studiums bedanken.<br />
Herrn Prof. Dr. Peter Hubral danke ich für die Übernahme des Hauptreferats und sein stets<br />
freundliches Interesse an meiner Arbeit. Er hat mir sehr viel Förderung zuteil werden lassen<br />
und zugleich alle Freiheiten gegeben.<br />
Herrn Prof. Dr. Friedemann Wenzel danke ich für die Übernahme des Korreferats.<br />
Steffen Bergler, Dr. German Höcht und besonders Jürgen Mann möchte ich für ihre vielfältige<br />
Hilfe in allen den CRS-<strong>Stack</strong> betreffenden aber auch allgemein computerspezifischen Fragen,<br />
sowie für viele nützliche und interessante Diskussionen danken.<br />
Großen Dank möchte ich Prof. Dr. Jörg Schleicher aussprechen, der sich während seines Aufenthaltes<br />
in Karlsruhe viel Zeit genommen hat um mir bei der Lösung einiger wichtiger theoretischer<br />
Fragen zu helfen.<br />
Danken möchte ich Prof. Dr. Martin Tygel, der in der Zeit, in der er hier in Karlsruhe zu Gast war,<br />
meiner Arbeit viel Interresse entgegen gebracht hat und mir einige gute Anregungen gegeben hat.<br />
Meinem Zimmerkollegen Pedro Chira möchte ich für seine Unterstützung und freundschaftliche<br />
Zusammenarbeit danken.<br />
Yonghai Zhang danke ich für seine freundliche und kompetente Hilfe, vor allem in Fragen der<br />
Wellentheorie.<br />
Ingo Koglin möchte ich besonders für die Unterstützung beim Picken und Extrahieren der NMO-<br />
Geschwindigkeiten danken.
100 Acknowledgment/Danksagung<br />
Für das Korrekturlesen meiner Arbeit möchte ich mich bedanken bei: Jürgen Mann, Steffen<br />
Bergler, Ingo Koglin, Eric Duveneck und Prof. Dr. Jörg Schleicher.<br />
Zu guter Letzt möchte ich auch noch allen anderen Mitarbeiterinnen und Mitarbeitern der Universität<br />
Karlsruhe danken, deren Hilfe ich im Laufe meines Studiums in Anspruch nehmen durfte.
Abbildungsverzeichnis<br />
2.1 Sketch of a two-dimensional inhomogeneous and isotropic medium. . . . . . . . 8<br />
2.2 Construction of the ray slowness vector projection. . . . . . . . . . . . . . . . . 9<br />
2.3 Sketch of a 2D model with a curved measurement surface. . . . . . . . . . . . . 11<br />
2.4 <strong>The</strong> paraxial ray from S to G in the vicinity of the central ray from S to G. . . . . 12<br />
2.5 ZO situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21<br />
2.6 <strong>The</strong> relationship between the take-off angles of the normal ray, β 0 and β g<br />
0 and<br />
the dip angle α 0 for a curved measurement surface. . . . . . . . . . . . . . . . . 27<br />
2.7 <strong>The</strong> search range of β 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34<br />
2.8 Redatuming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39<br />
2.9 <strong>The</strong> transformation of the local 1D coordinates h and m to the global 1D coordinates<br />
hg and mg. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43<br />
3.1 Model, consisting of four homogeneous layers with different P-wave velocities.<br />
<strong>The</strong>se are separated by three horizontal reflectors and lie <strong>under</strong>neath a curved<br />
measurement surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52<br />
3.2 Visualization of the applied static corrections . . . . . . . . . . . . . . . . . . . 54<br />
3.3 Standard processing, using wrong v NMO search limits: Automatic CMP stack<br />
section. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57<br />
3.4 Standard processing, using wrong v NMO search limits: Coherency section of the<br />
Automatic CMP stack. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
102 ABBILDUNGSVERZEICHNIS<br />
3.5 Standard processing, using wrong v NMO search limits: NMO velocity section. . . 59<br />
3.6 Standard processing, using better suited v NMO search limits: Automatic CMP<br />
stack section. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60<br />
3.7 Standard processing, using better suited v NMO search limits: Coherency section<br />
of the Automatic CMP stack. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61<br />
3.8 Standard processing, using better suited v NMO search limits: NMO velocity section. 62<br />
3.9 Example for the determination of the averaged measurement surface dip α 0 by<br />
fitting a straight line to the source and receiver locations within the stacking<br />
aperture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64<br />
3.10 Example for the determination of the averaged measurement surface curvature<br />
and elevation in X 0 by fitting a parabola to the the source and receiver locations<br />
within the stacking aperture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65<br />
3.11 Example for the determination the averaged measurement surface dip, curvature,<br />
and elevation in X 0 by fitting a circle to the the source and receiver locations<br />
within the stacking aperture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66<br />
3.12 <strong>The</strong> determined measurement surface curvature K 0 . . . . . . . . . . . . . . . . . 68<br />
3.13 <strong>The</strong> determined measurement surface dip α 0 . . . . . . . . . . . . . . . . . . . . 69<br />
3.14 Comparison between the real measurement surface, provided by the source and<br />
receiver locations and that surface that is given by the determined locations of<br />
the central points X 0 (x 0 ,z 0 ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69<br />
3.15 <strong>The</strong> forward calculated take-off angle β 0 and its search limits. . . . . . . . . . . 70<br />
3.16 <strong>The</strong> sine of the forward calculated pseudo take-off angle β ∗ 0<br />
and its search range. 71<br />
3.17 <strong>The</strong> forward calculated RMS velocities. . . . . . . . . . . . . . . . . . . . . . . 71<br />
3.18 <strong>The</strong> forward calculated NMO velocity for the first reflector. . . . . . . . . . . . . 72<br />
3.19 <strong>The</strong> forward calculated NMO velocity for the second reflector. . . . . . . . . . . 73<br />
3.20 <strong>The</strong> forward calculated NMO velocity for the third reflector. . . . . . . . . . . . 73<br />
3.21 <strong>The</strong> forward calculated NMO slowness for the first reflector and its search limits. 74
ABBILDUNGSVERZEICHNIS 103<br />
3.22 <strong>The</strong> forward calculated NMO slowness for the second reflector and its search<br />
limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75<br />
3.23 <strong>The</strong> forward calculated NMO slowness for the third reflector and its search limits. 75<br />
3.24 <strong>The</strong> forward calculated values of K NIP . . . . . . . . . . . . . . . . . . . . . . . . 76<br />
3.25 <strong>The</strong> forward calculated values of K ∗ N<br />
3.26 <strong>The</strong> forward calculated values of K ∗ NIP<br />
and its search limits. . . . . . . . . . . . . . 77<br />
for the first reflector and the corresponding<br />
search limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78<br />
for the second reflector and the corresponding<br />
search limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78<br />
3.27 <strong>The</strong> forward calculated values of K ∗ NIP<br />
for the third reflector and the corresponding<br />
search limits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79<br />
3.28 <strong>The</strong> forward calculated values of K ∗ NIP<br />
3.29 Automatic CMP stack section. . . . . . . . . . . . . . . . . . . . . . . . . . . . 81<br />
3.30 Coherency section of the automatic CMP stack. . . . . . . . . . . . . . . . . . . 82<br />
3.31 NMO velocity section. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83<br />
3.32 CMP gather (x 0 = 10.8 km). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84<br />
3.33 CMP gather (x 0 = 10.7 km). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85<br />
3.34 Automatic CMP stack section after redatuming. . . . . . . . . . . . . . . . . . . 86<br />
3.35 NMO velocity section after redatuming. . . . . . . . . . . . . . . . . . . . . . . 87<br />
3.36 Extracted NMO velocity for the first layer. . . . . . . . . . . . . . . . . . . . . . 89<br />
3.37 Extracted NMO velocity for the second layer. . . . . . . . . . . . . . . . . . . . 90<br />
3.38 Extracted NMO velocity for the third layer. . . . . . . . . . . . . . . . . . . . . 91
104 ABBILDUNGSVERZEICHNIS
Literaturverzeichnis<br />
Babich, V. (1956). Ray method of the computation of the intensity of wave fronts (in russian).<br />
Doklady Akad. Nauk SSSR, 110:355–357.<br />
Ben-Menachem, I. and Beydoun, W. B. (1985). Range of validity of seismic ray and beam<br />
methods in general inhomogeneous media - general theory. Geophys. Jrnl. Royal Astron. Soc.,<br />
82:207–234.<br />
Bergler, S. (2001). <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> stack for common offset - theory and application.<br />
Master’s thesis, Universität Karlsruhe.<br />
Bergler, S., Höecht, G., Zhang, Y., and Hubral, P. (2001). <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> stack for<br />
common offset: Practical aspects. In Extended Abstracts. 63th Mtg. Eur. Assoc. Expl Geophys..<br />
Session:P076.<br />
Biloti, R., Santos, L., and Tygel, M. (2001). Macro-velocity model inversion from traveltime<br />
attributes. In Expanded Abstracts. 71th Annual International SEG Meeting, San Antonio.<br />
Bortfeld, R. (1989). Geometrical ray theory: Rays and traveltimes in seismic systems (secondorder<br />
approximations of the traveltime). Geophysics, 48(3):1342–349.<br />
Červen´y, V. (2001). Seismic Ray <strong>The</strong>ory. Cambridge Univ. Press, New York, 2001 (IBSN 0-521-<br />
36671-2).<br />
Chira, P. and Hubral, P. (2001). Moveout formulas for a curved 2D measurement surface and<br />
near-zero-offset primary reflection. To be published in Geophysics.<br />
Chira, P., Tygel, M., Zhang, Y., and Hubral, P. (2001). Analytic CRS stack formula for a 2D curved<br />
measurement surface and finite-offset reflections. Journal of Seismic Exploration, 10(1-<br />
3):245–262.<br />
de Bazelaire, E. (1988). Normal moveout revisited - inhomogeneous media and curved interfaces.<br />
Geophysics, 53(2):143–157.
106 LITERATURVERZEICHNIS<br />
de Bazelaire, E. and Viallix, J. R. (1994). Normal moveout in focus. In Extended Abstracts, 54th<br />
Mtg. Eur. Assoc. Expl Geophys., pages 174–175.<br />
Dix, C. H. (1955). Seismic velocities from surface measurements. Geophysics, 20(1):68–86.<br />
Gardener, G., Wang, S., Pan, N., and Zhang, Z. (1990). Dip moveout and prestack imaging. In<br />
Extended Abstracts, 18th Ann. Offshore Tech. Conf., pages 75–84.<br />
Gelchinsky, B., Berkovitch, A., and Keydar, S. (1997). Multifocusing homeomorphic imaging:<br />
Parts I und II. Special Course on Homeomorphic Imaging. Wave Inversion Technology Consortium,<br />
Geophysical Institute, Karlsruhe University.<br />
Gurevich, B., Keydar, S., and Landa, E. (2001). Multifocusing imaging over an irregular topography.<br />
Geophysics, 67(2):639–643.<br />
Höcht, G. (1998). <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> stack. Master’s thesis, Universität Karlsruhe.<br />
Höcht, G. (2002). Traveltime approximations for 2D and 3D media and kinematic wavefield<br />
attributes. PhD thesis, Universität Karlsruhe.<br />
Hubral, P. (1983). Computing true amplitude reflections in a laterally inhomogeneous earth.<br />
Geophysics, 48(8):1051–1062.<br />
Hubral, P., editor (1999). Special Issue on Macro-Model Independent Seismic <strong>Reflection</strong> Imaging,<br />
volume 42 of Journal of Applied Geophysics, Amsterdam. Elsevier.<br />
Hubral, P. and Krey, T. (1980). Interval velocities from seismic reflection traveltime measurements.<br />
Soc. Expl. Geophys.<br />
Hubral, P., Schleicher, J., and Tygel, M. (1992). Three–dimensional paraxial ray properties, Part<br />
I: Basic relations. Journal of Seismic Exploration, 1:265–279.<br />
Jäger, R. (1999). <strong>The</strong> <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> stack - theory and application. Master’s<br />
thesis, Universität Karlsruhe.<br />
Karal, F. and Keller, J. (1959). Elastic wave propagation in homogeneous and inhomogeneous<br />
media. J. Acoust. Soc. Am., 31:694–705.<br />
Kravtsov, Y. A. and Orlov, Y. I. (1990). Geometrical Optics of Inhomogeneous Media. Series on<br />
Wave Phenomena. Springer-Verlag, Berlin.<br />
Mann, J. (2002). Extentions and applications of the <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> stack method.<br />
PhD thesis, Universität Karlsruhe.
LITERATURVERZEICHNIS 107<br />
Müller, T. (1999). <strong>The</strong> <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> stack method - seismic imaging without<br />
explicit knowledge of the velocity model. PhD thesis, Universität Karlsruhe.<br />
Schleicher, J., Tygel, M., and Hubral, P. (1993). Parabolic and hyperbolic paraxial two-point<br />
traveltimes in 3D media. Geophys. Prosp., 41(4):495–514.<br />
Schleicher, J., Tygel, M., and Hubral, P. (2002). Seismic true-amplitude reflection imaging.<br />
Preprint of SEG Monograph.<br />
Shah, P. M. (1973). Use of wavefront curvature to relate seismic data with subsurface parameters.<br />
Geophysics, 38(5):812–825.<br />
Ursin, B. (1982). Quadratic wavefront and traveltime approximations in inhomogeneous layered<br />
media with curved interfaces. Geophysics, 47(7):1012–1021.<br />
Yilmaz, O. (1987). Seismic data processing. Investigations in Geophysics. Soc. Expl. Geophys.<br />
(Doherty, S.M., Ed.).<br />
Zhang, Y., Bergler, S., and Hubral, P. (2001a). <strong>Common</strong>-<strong>Reflection</strong>-<strong>Surface</strong> (CRS) stack for<br />
common offset. Geophysical Prospecting, 49(6):709–718.<br />
Zhang, Y., Bergler, S., Tygel, M., and Hubral, P. (2001b). Model-independent traveltime attributes<br />
for 2D, finite-offset multi-coverage reflections. In PAGEOPH, in press.<br />
Zhang, Y., Höcht, G., and Hubral, P. (2002). 2D and 3D ZO CRS stack for a complex top-surface<br />
topography. In Extended Abstracts, 64th EAGE Conference, page P166.